xref: /openbmc/linux/drivers/net/ethernet/ibm/ibmvnic.c (revision e7f127b2)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /**************************************************************************/
3 /*                                                                        */
4 /*  IBM System i and System p Virtual NIC Device Driver                   */
5 /*  Copyright (C) 2014 IBM Corp.                                          */
6 /*  Santiago Leon (santi_leon@yahoo.com)                                  */
7 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
8 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
9 /*                                                                        */
10 /*                                                                        */
11 /* This module contains the implementation of a virtual ethernet device   */
12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
15 /*									   */
16 /* Messages are passed between the VNIC driver and the VNIC server using  */
17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
18 /* issue and receive commands that initiate communication with the server */
19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
20 /* are used by the driver to notify the server that a packet is           */
21 /* ready for transmission or that a buffer has been added to receive a    */
22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
23 /* driver that a packet transmission has been completed or that a packet  */
24 /* has been received and placed in a waiting buffer.                      */
25 /*                                                                        */
26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
28 /* or receive has been completed, the VNIC driver is required to use      */
29 /* "long term mapping". This entails that large, continuous DMA mapped    */
30 /* buffers are allocated on driver initialization and these buffers are   */
31 /* then continuously reused to pass skbs to and from the VNIC server.     */
32 /*                                                                        */
33 /**************************************************************************/
34 
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/types.h>
38 #include <linux/errno.h>
39 #include <linux/completion.h>
40 #include <linux/ioport.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kernel.h>
43 #include <linux/netdevice.h>
44 #include <linux/etherdevice.h>
45 #include <linux/skbuff.h>
46 #include <linux/init.h>
47 #include <linux/delay.h>
48 #include <linux/mm.h>
49 #include <linux/ethtool.h>
50 #include <linux/proc_fs.h>
51 #include <linux/if_arp.h>
52 #include <linux/in.h>
53 #include <linux/ip.h>
54 #include <linux/ipv6.h>
55 #include <linux/irq.h>
56 #include <linux/kthread.h>
57 #include <linux/seq_file.h>
58 #include <linux/interrupt.h>
59 #include <net/net_namespace.h>
60 #include <asm/hvcall.h>
61 #include <linux/atomic.h>
62 #include <asm/vio.h>
63 #include <asm/iommu.h>
64 #include <linux/uaccess.h>
65 #include <asm/firmware.h>
66 #include <linux/workqueue.h>
67 #include <linux/if_vlan.h>
68 #include <linux/utsname.h>
69 
70 #include "ibmvnic.h"
71 
72 static const char ibmvnic_driver_name[] = "ibmvnic";
73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
74 
75 MODULE_AUTHOR("Santiago Leon");
76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
77 MODULE_LICENSE("GPL");
78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
79 
80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
81 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
82 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
83 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
84 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
85 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
86 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
87 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
88 static int enable_scrq_irq(struct ibmvnic_adapter *,
89 			   struct ibmvnic_sub_crq_queue *);
90 static int disable_scrq_irq(struct ibmvnic_adapter *,
91 			    struct ibmvnic_sub_crq_queue *);
92 static int pending_scrq(struct ibmvnic_adapter *,
93 			struct ibmvnic_sub_crq_queue *);
94 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
95 					struct ibmvnic_sub_crq_queue *);
96 static int ibmvnic_poll(struct napi_struct *napi, int data);
97 static void send_query_map(struct ibmvnic_adapter *adapter);
98 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
99 static int send_request_unmap(struct ibmvnic_adapter *, u8);
100 static int send_login(struct ibmvnic_adapter *adapter);
101 static void send_query_cap(struct ibmvnic_adapter *adapter);
102 static int init_sub_crqs(struct ibmvnic_adapter *);
103 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
104 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
105 static void release_crq_queue(struct ibmvnic_adapter *);
106 static int __ibmvnic_set_mac(struct net_device *, u8 *);
107 static int init_crq_queue(struct ibmvnic_adapter *adapter);
108 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
109 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
110 					 struct ibmvnic_sub_crq_queue *tx_scrq);
111 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
112 				struct ibmvnic_long_term_buff *ltb);
113 
114 struct ibmvnic_stat {
115 	char name[ETH_GSTRING_LEN];
116 	int offset;
117 };
118 
119 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
120 			     offsetof(struct ibmvnic_statistics, stat))
121 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
122 
123 static const struct ibmvnic_stat ibmvnic_stats[] = {
124 	{"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
125 	{"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
126 	{"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
127 	{"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
128 	{"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
129 	{"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
130 	{"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
131 	{"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
132 	{"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
133 	{"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
134 	{"align_errors", IBMVNIC_STAT_OFF(align_errors)},
135 	{"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
136 	{"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
137 	{"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
138 	{"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
139 	{"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
140 	{"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
141 	{"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
142 	{"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
143 	{"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
144 	{"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
145 	{"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
146 };
147 
148 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
149 {
150 	union ibmvnic_crq crq;
151 
152 	memset(&crq, 0, sizeof(crq));
153 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
154 	crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
155 
156 	return ibmvnic_send_crq(adapter, &crq);
157 }
158 
159 static int send_version_xchg(struct ibmvnic_adapter *adapter)
160 {
161 	union ibmvnic_crq crq;
162 
163 	memset(&crq, 0, sizeof(crq));
164 	crq.version_exchange.first = IBMVNIC_CRQ_CMD;
165 	crq.version_exchange.cmd = VERSION_EXCHANGE;
166 	crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
167 
168 	return ibmvnic_send_crq(adapter, &crq);
169 }
170 
171 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
172 			  unsigned long length, unsigned long *number,
173 			  unsigned long *irq)
174 {
175 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
176 	long rc;
177 
178 	rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
179 	*number = retbuf[0];
180 	*irq = retbuf[1];
181 
182 	return rc;
183 }
184 
185 /**
186  * ibmvnic_wait_for_completion - Check device state and wait for completion
187  * @adapter: private device data
188  * @comp_done: completion structure to wait for
189  * @timeout: time to wait in milliseconds
190  *
191  * Wait for a completion signal or until the timeout limit is reached
192  * while checking that the device is still active.
193  */
194 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
195 				       struct completion *comp_done,
196 				       unsigned long timeout)
197 {
198 	struct net_device *netdev;
199 	unsigned long div_timeout;
200 	u8 retry;
201 
202 	netdev = adapter->netdev;
203 	retry = 5;
204 	div_timeout = msecs_to_jiffies(timeout / retry);
205 	while (true) {
206 		if (!adapter->crq.active) {
207 			netdev_err(netdev, "Device down!\n");
208 			return -ENODEV;
209 		}
210 		if (!retry--)
211 			break;
212 		if (wait_for_completion_timeout(comp_done, div_timeout))
213 			return 0;
214 	}
215 	netdev_err(netdev, "Operation timed out.\n");
216 	return -ETIMEDOUT;
217 }
218 
219 /**
220  * reuse_ltb() - Check if a long term buffer can be reused
221  * @ltb:  The long term buffer to be checked
222  * @size: The size of the long term buffer.
223  *
224  * An LTB can be reused unless its size has changed.
225  *
226  * Return: Return true if the LTB can be reused, false otherwise.
227  */
228 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
229 {
230 	return (ltb->buff && ltb->size == size);
231 }
232 
233 /**
234  * alloc_long_term_buff() - Allocate a long term buffer (LTB)
235  *
236  * @adapter: ibmvnic adapter associated to the LTB
237  * @ltb:     container object for the LTB
238  * @size:    size of the LTB
239  *
240  * Allocate an LTB of the specified size and notify VIOS.
241  *
242  * If the given @ltb already has the correct size, reuse it. Otherwise if
243  * its non-NULL, free it. Then allocate a new one of the correct size.
244  * Notify the VIOS either way since we may now be working with a new VIOS.
245  *
246  * Allocating larger chunks of memory during resets, specially LPM or under
247  * low memory situations can cause resets to fail/timeout and for LPAR to
248  * lose connectivity. So hold onto the LTB even if we fail to communicate
249  * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
250  *
251  * Return: 0 if we were able to allocate the LTB and notify the VIOS and
252  *	   a negative value otherwise.
253  */
254 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
255 				struct ibmvnic_long_term_buff *ltb, int size)
256 {
257 	struct device *dev = &adapter->vdev->dev;
258 	int rc;
259 
260 	if (!reuse_ltb(ltb, size)) {
261 		dev_dbg(dev,
262 			"LTB size changed from 0x%llx to 0x%x, reallocating\n",
263 			 ltb->size, size);
264 		free_long_term_buff(adapter, ltb);
265 	}
266 
267 	if (ltb->buff) {
268 		dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
269 			ltb->map_id, ltb->size);
270 	} else {
271 		ltb->buff = dma_alloc_coherent(dev, size, &ltb->addr,
272 					       GFP_KERNEL);
273 		if (!ltb->buff) {
274 			dev_err(dev, "Couldn't alloc long term buffer\n");
275 			return -ENOMEM;
276 		}
277 		ltb->size = size;
278 
279 		ltb->map_id = find_first_zero_bit(adapter->map_ids,
280 						  MAX_MAP_ID);
281 		bitmap_set(adapter->map_ids, ltb->map_id, 1);
282 
283 		dev_dbg(dev,
284 			"Allocated new LTB [map %d, size 0x%llx]\n",
285 			 ltb->map_id, ltb->size);
286 	}
287 
288 	/* Ensure ltb is zeroed - specially when reusing it. */
289 	memset(ltb->buff, 0, ltb->size);
290 
291 	mutex_lock(&adapter->fw_lock);
292 	adapter->fw_done_rc = 0;
293 	reinit_completion(&adapter->fw_done);
294 
295 	rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
296 	if (rc) {
297 		dev_err(dev, "send_request_map failed, rc = %d\n", rc);
298 		goto out;
299 	}
300 
301 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
302 	if (rc) {
303 		dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
304 			rc);
305 		goto out;
306 	}
307 
308 	if (adapter->fw_done_rc) {
309 		dev_err(dev, "Couldn't map LTB, rc = %d\n",
310 			adapter->fw_done_rc);
311 		rc = -EIO;
312 		goto out;
313 	}
314 	rc = 0;
315 out:
316 	/* don't free LTB on communication error - see function header */
317 	mutex_unlock(&adapter->fw_lock);
318 	return rc;
319 }
320 
321 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
322 				struct ibmvnic_long_term_buff *ltb)
323 {
324 	struct device *dev = &adapter->vdev->dev;
325 
326 	if (!ltb->buff)
327 		return;
328 
329 	/* VIOS automatically unmaps the long term buffer at remote
330 	 * end for the following resets:
331 	 * FAILOVER, MOBILITY, TIMEOUT.
332 	 */
333 	if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
334 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
335 	    adapter->reset_reason != VNIC_RESET_TIMEOUT)
336 		send_request_unmap(adapter, ltb->map_id);
337 
338 	dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
339 
340 	ltb->buff = NULL;
341 	/* mark this map_id free */
342 	bitmap_clear(adapter->map_ids, ltb->map_id, 1);
343 	ltb->map_id = 0;
344 }
345 
346 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
347 {
348 	int i;
349 
350 	for (i = 0; i < adapter->num_active_rx_pools; i++)
351 		adapter->rx_pool[i].active = 0;
352 }
353 
354 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
355 			      struct ibmvnic_rx_pool *pool)
356 {
357 	int count = pool->size - atomic_read(&pool->available);
358 	u64 handle = adapter->rx_scrq[pool->index]->handle;
359 	struct device *dev = &adapter->vdev->dev;
360 	struct ibmvnic_ind_xmit_queue *ind_bufp;
361 	struct ibmvnic_sub_crq_queue *rx_scrq;
362 	union sub_crq *sub_crq;
363 	int buffers_added = 0;
364 	unsigned long lpar_rc;
365 	struct sk_buff *skb;
366 	unsigned int offset;
367 	dma_addr_t dma_addr;
368 	unsigned char *dst;
369 	int shift = 0;
370 	int index;
371 	int i;
372 
373 	if (!pool->active)
374 		return;
375 
376 	rx_scrq = adapter->rx_scrq[pool->index];
377 	ind_bufp = &rx_scrq->ind_buf;
378 
379 	/* netdev_skb_alloc() could have failed after we saved a few skbs
380 	 * in the indir_buf and we would not have sent them to VIOS yet.
381 	 * To account for them, start the loop at ind_bufp->index rather
382 	 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
383 	 * be 0.
384 	 */
385 	for (i = ind_bufp->index; i < count; ++i) {
386 		index = pool->free_map[pool->next_free];
387 
388 		/* We maybe reusing the skb from earlier resets. Allocate
389 		 * only if necessary. But since the LTB may have changed
390 		 * during reset (see init_rx_pools()), update LTB below
391 		 * even if reusing skb.
392 		 */
393 		skb = pool->rx_buff[index].skb;
394 		if (!skb) {
395 			skb = netdev_alloc_skb(adapter->netdev,
396 					       pool->buff_size);
397 			if (!skb) {
398 				dev_err(dev, "Couldn't replenish rx buff\n");
399 				adapter->replenish_no_mem++;
400 				break;
401 			}
402 		}
403 
404 		pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
405 		pool->next_free = (pool->next_free + 1) % pool->size;
406 
407 		/* Copy the skb to the long term mapped DMA buffer */
408 		offset = index * pool->buff_size;
409 		dst = pool->long_term_buff.buff + offset;
410 		memset(dst, 0, pool->buff_size);
411 		dma_addr = pool->long_term_buff.addr + offset;
412 
413 		/* add the skb to an rx_buff in the pool */
414 		pool->rx_buff[index].data = dst;
415 		pool->rx_buff[index].dma = dma_addr;
416 		pool->rx_buff[index].skb = skb;
417 		pool->rx_buff[index].pool_index = pool->index;
418 		pool->rx_buff[index].size = pool->buff_size;
419 
420 		/* queue the rx_buff for the next send_subcrq_indirect */
421 		sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
422 		memset(sub_crq, 0, sizeof(*sub_crq));
423 		sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
424 		sub_crq->rx_add.correlator =
425 		    cpu_to_be64((u64)&pool->rx_buff[index]);
426 		sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
427 		sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
428 
429 		/* The length field of the sCRQ is defined to be 24 bits so the
430 		 * buffer size needs to be left shifted by a byte before it is
431 		 * converted to big endian to prevent the last byte from being
432 		 * truncated.
433 		 */
434 #ifdef __LITTLE_ENDIAN__
435 		shift = 8;
436 #endif
437 		sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
438 
439 		/* if send_subcrq_indirect queue is full, flush to VIOS */
440 		if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
441 		    i == count - 1) {
442 			lpar_rc =
443 				send_subcrq_indirect(adapter, handle,
444 						     (u64)ind_bufp->indir_dma,
445 						     (u64)ind_bufp->index);
446 			if (lpar_rc != H_SUCCESS)
447 				goto failure;
448 			buffers_added += ind_bufp->index;
449 			adapter->replenish_add_buff_success += ind_bufp->index;
450 			ind_bufp->index = 0;
451 		}
452 	}
453 	atomic_add(buffers_added, &pool->available);
454 	return;
455 
456 failure:
457 	if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
458 		dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
459 	for (i = ind_bufp->index - 1; i >= 0; --i) {
460 		struct ibmvnic_rx_buff *rx_buff;
461 
462 		pool->next_free = pool->next_free == 0 ?
463 				  pool->size - 1 : pool->next_free - 1;
464 		sub_crq = &ind_bufp->indir_arr[i];
465 		rx_buff = (struct ibmvnic_rx_buff *)
466 				be64_to_cpu(sub_crq->rx_add.correlator);
467 		index = (int)(rx_buff - pool->rx_buff);
468 		pool->free_map[pool->next_free] = index;
469 		dev_kfree_skb_any(pool->rx_buff[index].skb);
470 		pool->rx_buff[index].skb = NULL;
471 	}
472 	adapter->replenish_add_buff_failure += ind_bufp->index;
473 	atomic_add(buffers_added, &pool->available);
474 	ind_bufp->index = 0;
475 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
476 		/* Disable buffer pool replenishment and report carrier off if
477 		 * queue is closed or pending failover.
478 		 * Firmware guarantees that a signal will be sent to the
479 		 * driver, triggering a reset.
480 		 */
481 		deactivate_rx_pools(adapter);
482 		netif_carrier_off(adapter->netdev);
483 	}
484 }
485 
486 static void replenish_pools(struct ibmvnic_adapter *adapter)
487 {
488 	int i;
489 
490 	adapter->replenish_task_cycles++;
491 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
492 		if (adapter->rx_pool[i].active)
493 			replenish_rx_pool(adapter, &adapter->rx_pool[i]);
494 	}
495 
496 	netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
497 }
498 
499 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
500 {
501 	kfree(adapter->tx_stats_buffers);
502 	kfree(adapter->rx_stats_buffers);
503 	adapter->tx_stats_buffers = NULL;
504 	adapter->rx_stats_buffers = NULL;
505 }
506 
507 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
508 {
509 	adapter->tx_stats_buffers =
510 				kcalloc(IBMVNIC_MAX_QUEUES,
511 					sizeof(struct ibmvnic_tx_queue_stats),
512 					GFP_KERNEL);
513 	if (!adapter->tx_stats_buffers)
514 		return -ENOMEM;
515 
516 	adapter->rx_stats_buffers =
517 				kcalloc(IBMVNIC_MAX_QUEUES,
518 					sizeof(struct ibmvnic_rx_queue_stats),
519 					GFP_KERNEL);
520 	if (!adapter->rx_stats_buffers)
521 		return -ENOMEM;
522 
523 	return 0;
524 }
525 
526 static void release_stats_token(struct ibmvnic_adapter *adapter)
527 {
528 	struct device *dev = &adapter->vdev->dev;
529 
530 	if (!adapter->stats_token)
531 		return;
532 
533 	dma_unmap_single(dev, adapter->stats_token,
534 			 sizeof(struct ibmvnic_statistics),
535 			 DMA_FROM_DEVICE);
536 	adapter->stats_token = 0;
537 }
538 
539 static int init_stats_token(struct ibmvnic_adapter *adapter)
540 {
541 	struct device *dev = &adapter->vdev->dev;
542 	dma_addr_t stok;
543 	int rc;
544 
545 	stok = dma_map_single(dev, &adapter->stats,
546 			      sizeof(struct ibmvnic_statistics),
547 			      DMA_FROM_DEVICE);
548 	rc = dma_mapping_error(dev, stok);
549 	if (rc) {
550 		dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
551 		return rc;
552 	}
553 
554 	adapter->stats_token = stok;
555 	netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
556 	return 0;
557 }
558 
559 /**
560  * release_rx_pools() - Release any rx pools attached to @adapter.
561  * @adapter: ibmvnic adapter
562  *
563  * Safe to call this multiple times - even if no pools are attached.
564  */
565 static void release_rx_pools(struct ibmvnic_adapter *adapter)
566 {
567 	struct ibmvnic_rx_pool *rx_pool;
568 	int i, j;
569 
570 	if (!adapter->rx_pool)
571 		return;
572 
573 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
574 		rx_pool = &adapter->rx_pool[i];
575 
576 		netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
577 
578 		kfree(rx_pool->free_map);
579 
580 		free_long_term_buff(adapter, &rx_pool->long_term_buff);
581 
582 		if (!rx_pool->rx_buff)
583 			continue;
584 
585 		for (j = 0; j < rx_pool->size; j++) {
586 			if (rx_pool->rx_buff[j].skb) {
587 				dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
588 				rx_pool->rx_buff[j].skb = NULL;
589 			}
590 		}
591 
592 		kfree(rx_pool->rx_buff);
593 	}
594 
595 	kfree(adapter->rx_pool);
596 	adapter->rx_pool = NULL;
597 	adapter->num_active_rx_pools = 0;
598 	adapter->prev_rx_pool_size = 0;
599 }
600 
601 /**
602  * reuse_rx_pools() - Check if the existing rx pools can be reused.
603  * @adapter: ibmvnic adapter
604  *
605  * Check if the existing rx pools in the adapter can be reused. The
606  * pools can be reused if the pool parameters (number of pools,
607  * number of buffers in the pool and size of each buffer) have not
608  * changed.
609  *
610  * NOTE: This assumes that all pools have the same number of buffers
611  *       which is the case currently. If that changes, we must fix this.
612  *
613  * Return: true if the rx pools can be reused, false otherwise.
614  */
615 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
616 {
617 	u64 old_num_pools, new_num_pools;
618 	u64 old_pool_size, new_pool_size;
619 	u64 old_buff_size, new_buff_size;
620 
621 	if (!adapter->rx_pool)
622 		return false;
623 
624 	old_num_pools = adapter->num_active_rx_pools;
625 	new_num_pools = adapter->req_rx_queues;
626 
627 	old_pool_size = adapter->prev_rx_pool_size;
628 	new_pool_size = adapter->req_rx_add_entries_per_subcrq;
629 
630 	old_buff_size = adapter->prev_rx_buf_sz;
631 	new_buff_size = adapter->cur_rx_buf_sz;
632 
633 	if (old_buff_size != new_buff_size ||
634 	    old_num_pools != new_num_pools ||
635 	    old_pool_size != new_pool_size)
636 		return false;
637 
638 	return true;
639 }
640 
641 /**
642  * init_rx_pools(): Initialize the set of receiver pools in the adapter.
643  * @netdev: net device associated with the vnic interface
644  *
645  * Initialize the set of receiver pools in the ibmvnic adapter associated
646  * with the net_device @netdev. If possible, reuse the existing rx pools.
647  * Otherwise free any existing pools and  allocate a new set of pools
648  * before initializing them.
649  *
650  * Return: 0 on success and negative value on error.
651  */
652 static int init_rx_pools(struct net_device *netdev)
653 {
654 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
655 	struct device *dev = &adapter->vdev->dev;
656 	struct ibmvnic_rx_pool *rx_pool;
657 	u64 num_pools;
658 	u64 pool_size;		/* # of buffers in one pool */
659 	u64 buff_size;
660 	int i, j, rc;
661 
662 	pool_size = adapter->req_rx_add_entries_per_subcrq;
663 	num_pools = adapter->req_rx_queues;
664 	buff_size = adapter->cur_rx_buf_sz;
665 
666 	if (reuse_rx_pools(adapter)) {
667 		dev_dbg(dev, "Reusing rx pools\n");
668 		goto update_ltb;
669 	}
670 
671 	/* Allocate/populate the pools. */
672 	release_rx_pools(adapter);
673 
674 	adapter->rx_pool = kcalloc(num_pools,
675 				   sizeof(struct ibmvnic_rx_pool),
676 				   GFP_KERNEL);
677 	if (!adapter->rx_pool) {
678 		dev_err(dev, "Failed to allocate rx pools\n");
679 		return -ENOMEM;
680 	}
681 
682 	/* Set num_active_rx_pools early. If we fail below after partial
683 	 * allocation, release_rx_pools() will know how many to look for.
684 	 */
685 	adapter->num_active_rx_pools = num_pools;
686 
687 	for (i = 0; i < num_pools; i++) {
688 		rx_pool = &adapter->rx_pool[i];
689 
690 		netdev_dbg(adapter->netdev,
691 			   "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
692 			   i, pool_size, buff_size);
693 
694 		rx_pool->size = pool_size;
695 		rx_pool->index = i;
696 		rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
697 
698 		rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
699 					    GFP_KERNEL);
700 		if (!rx_pool->free_map) {
701 			dev_err(dev, "Couldn't alloc free_map %d\n", i);
702 			rc = -ENOMEM;
703 			goto out_release;
704 		}
705 
706 		rx_pool->rx_buff = kcalloc(rx_pool->size,
707 					   sizeof(struct ibmvnic_rx_buff),
708 					   GFP_KERNEL);
709 		if (!rx_pool->rx_buff) {
710 			dev_err(dev, "Couldn't alloc rx buffers\n");
711 			rc = -ENOMEM;
712 			goto out_release;
713 		}
714 	}
715 
716 	adapter->prev_rx_pool_size = pool_size;
717 	adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
718 
719 update_ltb:
720 	for (i = 0; i < num_pools; i++) {
721 		rx_pool = &adapter->rx_pool[i];
722 		dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
723 			i, rx_pool->size, rx_pool->buff_size);
724 
725 		rc = alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
726 					  rx_pool->size * rx_pool->buff_size);
727 		if (rc)
728 			goto out;
729 
730 		for (j = 0; j < rx_pool->size; ++j) {
731 			struct ibmvnic_rx_buff *rx_buff;
732 
733 			rx_pool->free_map[j] = j;
734 
735 			/* NOTE: Don't clear rx_buff->skb here - will leak
736 			 * memory! replenish_rx_pool() will reuse skbs or
737 			 * allocate as necessary.
738 			 */
739 			rx_buff = &rx_pool->rx_buff[j];
740 			rx_buff->dma = 0;
741 			rx_buff->data = 0;
742 			rx_buff->size = 0;
743 			rx_buff->pool_index = 0;
744 		}
745 
746 		/* Mark pool "empty" so replenish_rx_pools() will
747 		 * update the LTB info for each buffer
748 		 */
749 		atomic_set(&rx_pool->available, 0);
750 		rx_pool->next_alloc = 0;
751 		rx_pool->next_free = 0;
752 		/* replenish_rx_pool() may have called deactivate_rx_pools()
753 		 * on failover. Ensure pool is active now.
754 		 */
755 		rx_pool->active = 1;
756 	}
757 	return 0;
758 out_release:
759 	release_rx_pools(adapter);
760 out:
761 	/* We failed to allocate one or more LTBs or map them on the VIOS.
762 	 * Hold onto the pools and any LTBs that we did allocate/map.
763 	 */
764 	return rc;
765 }
766 
767 static void release_vpd_data(struct ibmvnic_adapter *adapter)
768 {
769 	if (!adapter->vpd)
770 		return;
771 
772 	kfree(adapter->vpd->buff);
773 	kfree(adapter->vpd);
774 
775 	adapter->vpd = NULL;
776 }
777 
778 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
779 				struct ibmvnic_tx_pool *tx_pool)
780 {
781 	kfree(tx_pool->tx_buff);
782 	kfree(tx_pool->free_map);
783 	free_long_term_buff(adapter, &tx_pool->long_term_buff);
784 }
785 
786 /**
787  * release_tx_pools() - Release any tx pools attached to @adapter.
788  * @adapter: ibmvnic adapter
789  *
790  * Safe to call this multiple times - even if no pools are attached.
791  */
792 static void release_tx_pools(struct ibmvnic_adapter *adapter)
793 {
794 	int i;
795 
796 	/* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
797 	 * both NULL or both non-NULL. So we only need to check one.
798 	 */
799 	if (!adapter->tx_pool)
800 		return;
801 
802 	for (i = 0; i < adapter->num_active_tx_pools; i++) {
803 		release_one_tx_pool(adapter, &adapter->tx_pool[i]);
804 		release_one_tx_pool(adapter, &adapter->tso_pool[i]);
805 	}
806 
807 	kfree(adapter->tx_pool);
808 	adapter->tx_pool = NULL;
809 	kfree(adapter->tso_pool);
810 	adapter->tso_pool = NULL;
811 	adapter->num_active_tx_pools = 0;
812 	adapter->prev_tx_pool_size = 0;
813 }
814 
815 static int init_one_tx_pool(struct net_device *netdev,
816 			    struct ibmvnic_tx_pool *tx_pool,
817 			    int pool_size, int buf_size)
818 {
819 	int i;
820 
821 	tx_pool->tx_buff = kcalloc(pool_size,
822 				   sizeof(struct ibmvnic_tx_buff),
823 				   GFP_KERNEL);
824 	if (!tx_pool->tx_buff)
825 		return -ENOMEM;
826 
827 	tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
828 	if (!tx_pool->free_map) {
829 		kfree(tx_pool->tx_buff);
830 		tx_pool->tx_buff = NULL;
831 		return -ENOMEM;
832 	}
833 
834 	for (i = 0; i < pool_size; i++)
835 		tx_pool->free_map[i] = i;
836 
837 	tx_pool->consumer_index = 0;
838 	tx_pool->producer_index = 0;
839 	tx_pool->num_buffers = pool_size;
840 	tx_pool->buf_size = buf_size;
841 
842 	return 0;
843 }
844 
845 /**
846  * reuse_tx_pools() - Check if the existing tx pools can be reused.
847  * @adapter: ibmvnic adapter
848  *
849  * Check if the existing tx pools in the adapter can be reused. The
850  * pools can be reused if the pool parameters (number of pools,
851  * number of buffers in the pool and mtu) have not changed.
852  *
853  * NOTE: This assumes that all pools have the same number of buffers
854  *       which is the case currently. If that changes, we must fix this.
855  *
856  * Return: true if the tx pools can be reused, false otherwise.
857  */
858 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
859 {
860 	u64 old_num_pools, new_num_pools;
861 	u64 old_pool_size, new_pool_size;
862 	u64 old_mtu, new_mtu;
863 
864 	if (!adapter->tx_pool)
865 		return false;
866 
867 	old_num_pools = adapter->num_active_tx_pools;
868 	new_num_pools = adapter->num_active_tx_scrqs;
869 	old_pool_size = adapter->prev_tx_pool_size;
870 	new_pool_size = adapter->req_tx_entries_per_subcrq;
871 	old_mtu = adapter->prev_mtu;
872 	new_mtu = adapter->req_mtu;
873 
874 	if (old_mtu != new_mtu ||
875 	    old_num_pools != new_num_pools ||
876 	    old_pool_size != new_pool_size)
877 		return false;
878 
879 	return true;
880 }
881 
882 /**
883  * init_tx_pools(): Initialize the set of transmit pools in the adapter.
884  * @netdev: net device associated with the vnic interface
885  *
886  * Initialize the set of transmit pools in the ibmvnic adapter associated
887  * with the net_device @netdev. If possible, reuse the existing tx pools.
888  * Otherwise free any existing pools and  allocate a new set of pools
889  * before initializing them.
890  *
891  * Return: 0 on success and negative value on error.
892  */
893 static int init_tx_pools(struct net_device *netdev)
894 {
895 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
896 	struct device *dev = &adapter->vdev->dev;
897 	int num_pools;
898 	u64 pool_size;		/* # of buffers in pool */
899 	u64 buff_size;
900 	int i, j, rc;
901 
902 	num_pools = adapter->req_tx_queues;
903 
904 	/* We must notify the VIOS about the LTB on all resets - but we only
905 	 * need to alloc/populate pools if either the number of buffers or
906 	 * size of each buffer in the pool has changed.
907 	 */
908 	if (reuse_tx_pools(adapter)) {
909 		netdev_dbg(netdev, "Reusing tx pools\n");
910 		goto update_ltb;
911 	}
912 
913 	/* Allocate/populate the pools. */
914 	release_tx_pools(adapter);
915 
916 	pool_size = adapter->req_tx_entries_per_subcrq;
917 	num_pools = adapter->num_active_tx_scrqs;
918 
919 	adapter->tx_pool = kcalloc(num_pools,
920 				   sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
921 	if (!adapter->tx_pool)
922 		return -ENOMEM;
923 
924 	adapter->tso_pool = kcalloc(num_pools,
925 				    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
926 	/* To simplify release_tx_pools() ensure that ->tx_pool and
927 	 * ->tso_pool are either both NULL or both non-NULL.
928 	 */
929 	if (!adapter->tso_pool) {
930 		kfree(adapter->tx_pool);
931 		adapter->tx_pool = NULL;
932 		return -ENOMEM;
933 	}
934 
935 	/* Set num_active_tx_pools early. If we fail below after partial
936 	 * allocation, release_tx_pools() will know how many to look for.
937 	 */
938 	adapter->num_active_tx_pools = num_pools;
939 
940 	buff_size = adapter->req_mtu + VLAN_HLEN;
941 	buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
942 
943 	for (i = 0; i < num_pools; i++) {
944 		dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
945 			i, adapter->req_tx_entries_per_subcrq, buff_size);
946 
947 		rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
948 				      pool_size, buff_size);
949 		if (rc)
950 			goto out_release;
951 
952 		rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
953 				      IBMVNIC_TSO_BUFS,
954 				      IBMVNIC_TSO_BUF_SZ);
955 		if (rc)
956 			goto out_release;
957 	}
958 
959 	adapter->prev_tx_pool_size = pool_size;
960 	adapter->prev_mtu = adapter->req_mtu;
961 
962 update_ltb:
963 	/* NOTE: All tx_pools have the same number of buffers (which is
964 	 *       same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
965 	 *       buffers (see calls init_one_tx_pool() for these).
966 	 *       For consistency, we use tx_pool->num_buffers and
967 	 *       tso_pool->num_buffers below.
968 	 */
969 	rc = -1;
970 	for (i = 0; i < num_pools; i++) {
971 		struct ibmvnic_tx_pool *tso_pool;
972 		struct ibmvnic_tx_pool *tx_pool;
973 		u32 ltb_size;
974 
975 		tx_pool = &adapter->tx_pool[i];
976 		ltb_size = tx_pool->num_buffers * tx_pool->buf_size;
977 		if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
978 					 ltb_size))
979 			goto out;
980 
981 		dev_dbg(dev, "Updated LTB for tx pool %d [%p, %d, %d]\n",
982 			i, tx_pool->long_term_buff.buff,
983 			tx_pool->num_buffers, tx_pool->buf_size);
984 
985 		tx_pool->consumer_index = 0;
986 		tx_pool->producer_index = 0;
987 
988 		for (j = 0; j < tx_pool->num_buffers; j++)
989 			tx_pool->free_map[j] = j;
990 
991 		tso_pool = &adapter->tso_pool[i];
992 		ltb_size = tso_pool->num_buffers * tso_pool->buf_size;
993 		if (alloc_long_term_buff(adapter, &tso_pool->long_term_buff,
994 					 ltb_size))
995 			goto out;
996 
997 		dev_dbg(dev, "Updated LTB for tso pool %d [%p, %d, %d]\n",
998 			i, tso_pool->long_term_buff.buff,
999 			tso_pool->num_buffers, tso_pool->buf_size);
1000 
1001 		tso_pool->consumer_index = 0;
1002 		tso_pool->producer_index = 0;
1003 
1004 		for (j = 0; j < tso_pool->num_buffers; j++)
1005 			tso_pool->free_map[j] = j;
1006 	}
1007 
1008 	return 0;
1009 out_release:
1010 	release_tx_pools(adapter);
1011 out:
1012 	/* We failed to allocate one or more LTBs or map them on the VIOS.
1013 	 * Hold onto the pools and any LTBs that we did allocate/map.
1014 	 */
1015 	return rc;
1016 }
1017 
1018 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1019 {
1020 	int i;
1021 
1022 	if (adapter->napi_enabled)
1023 		return;
1024 
1025 	for (i = 0; i < adapter->req_rx_queues; i++)
1026 		napi_enable(&adapter->napi[i]);
1027 
1028 	adapter->napi_enabled = true;
1029 }
1030 
1031 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1032 {
1033 	int i;
1034 
1035 	if (!adapter->napi_enabled)
1036 		return;
1037 
1038 	for (i = 0; i < adapter->req_rx_queues; i++) {
1039 		netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1040 		napi_disable(&adapter->napi[i]);
1041 	}
1042 
1043 	adapter->napi_enabled = false;
1044 }
1045 
1046 static int init_napi(struct ibmvnic_adapter *adapter)
1047 {
1048 	int i;
1049 
1050 	adapter->napi = kcalloc(adapter->req_rx_queues,
1051 				sizeof(struct napi_struct), GFP_KERNEL);
1052 	if (!adapter->napi)
1053 		return -ENOMEM;
1054 
1055 	for (i = 0; i < adapter->req_rx_queues; i++) {
1056 		netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1057 		netif_napi_add(adapter->netdev, &adapter->napi[i],
1058 			       ibmvnic_poll, NAPI_POLL_WEIGHT);
1059 	}
1060 
1061 	adapter->num_active_rx_napi = adapter->req_rx_queues;
1062 	return 0;
1063 }
1064 
1065 static void release_napi(struct ibmvnic_adapter *adapter)
1066 {
1067 	int i;
1068 
1069 	if (!adapter->napi)
1070 		return;
1071 
1072 	for (i = 0; i < adapter->num_active_rx_napi; i++) {
1073 		netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1074 		netif_napi_del(&adapter->napi[i]);
1075 	}
1076 
1077 	kfree(adapter->napi);
1078 	adapter->napi = NULL;
1079 	adapter->num_active_rx_napi = 0;
1080 	adapter->napi_enabled = false;
1081 }
1082 
1083 static const char *adapter_state_to_string(enum vnic_state state)
1084 {
1085 	switch (state) {
1086 	case VNIC_PROBING:
1087 		return "PROBING";
1088 	case VNIC_PROBED:
1089 		return "PROBED";
1090 	case VNIC_OPENING:
1091 		return "OPENING";
1092 	case VNIC_OPEN:
1093 		return "OPEN";
1094 	case VNIC_CLOSING:
1095 		return "CLOSING";
1096 	case VNIC_CLOSED:
1097 		return "CLOSED";
1098 	case VNIC_REMOVING:
1099 		return "REMOVING";
1100 	case VNIC_REMOVED:
1101 		return "REMOVED";
1102 	case VNIC_DOWN:
1103 		return "DOWN";
1104 	}
1105 	return "UNKNOWN";
1106 }
1107 
1108 static int ibmvnic_login(struct net_device *netdev)
1109 {
1110 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1111 	unsigned long timeout = msecs_to_jiffies(20000);
1112 	int retry_count = 0;
1113 	int retries = 10;
1114 	bool retry;
1115 	int rc;
1116 
1117 	do {
1118 		retry = false;
1119 		if (retry_count > retries) {
1120 			netdev_warn(netdev, "Login attempts exceeded\n");
1121 			return -EACCES;
1122 		}
1123 
1124 		adapter->init_done_rc = 0;
1125 		reinit_completion(&adapter->init_done);
1126 		rc = send_login(adapter);
1127 		if (rc)
1128 			return rc;
1129 
1130 		if (!wait_for_completion_timeout(&adapter->init_done,
1131 						 timeout)) {
1132 			netdev_warn(netdev, "Login timed out, retrying...\n");
1133 			retry = true;
1134 			adapter->init_done_rc = 0;
1135 			retry_count++;
1136 			continue;
1137 		}
1138 
1139 		if (adapter->init_done_rc == ABORTED) {
1140 			netdev_warn(netdev, "Login aborted, retrying...\n");
1141 			retry = true;
1142 			adapter->init_done_rc = 0;
1143 			retry_count++;
1144 			/* FW or device may be busy, so
1145 			 * wait a bit before retrying login
1146 			 */
1147 			msleep(500);
1148 		} else if (adapter->init_done_rc == PARTIALSUCCESS) {
1149 			retry_count++;
1150 			release_sub_crqs(adapter, 1);
1151 
1152 			retry = true;
1153 			netdev_dbg(netdev,
1154 				   "Received partial success, retrying...\n");
1155 			adapter->init_done_rc = 0;
1156 			reinit_completion(&adapter->init_done);
1157 			send_query_cap(adapter);
1158 			if (!wait_for_completion_timeout(&adapter->init_done,
1159 							 timeout)) {
1160 				netdev_warn(netdev,
1161 					    "Capabilities query timed out\n");
1162 				return -ETIMEDOUT;
1163 			}
1164 
1165 			rc = init_sub_crqs(adapter);
1166 			if (rc) {
1167 				netdev_warn(netdev,
1168 					    "SCRQ initialization failed\n");
1169 				return rc;
1170 			}
1171 
1172 			rc = init_sub_crq_irqs(adapter);
1173 			if (rc) {
1174 				netdev_warn(netdev,
1175 					    "SCRQ irq initialization failed\n");
1176 				return rc;
1177 			}
1178 		} else if (adapter->init_done_rc) {
1179 			netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
1180 				    adapter->init_done_rc);
1181 			return -EIO;
1182 		}
1183 	} while (retry);
1184 
1185 	__ibmvnic_set_mac(netdev, adapter->mac_addr);
1186 
1187 	netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1188 	return 0;
1189 }
1190 
1191 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1192 {
1193 	kfree(adapter->login_buf);
1194 	adapter->login_buf = NULL;
1195 }
1196 
1197 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1198 {
1199 	kfree(adapter->login_rsp_buf);
1200 	adapter->login_rsp_buf = NULL;
1201 }
1202 
1203 static void release_resources(struct ibmvnic_adapter *adapter)
1204 {
1205 	release_vpd_data(adapter);
1206 
1207 	release_napi(adapter);
1208 	release_login_buffer(adapter);
1209 	release_login_rsp_buffer(adapter);
1210 }
1211 
1212 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1213 {
1214 	struct net_device *netdev = adapter->netdev;
1215 	unsigned long timeout = msecs_to_jiffies(20000);
1216 	union ibmvnic_crq crq;
1217 	bool resend;
1218 	int rc;
1219 
1220 	netdev_dbg(netdev, "setting link state %d\n", link_state);
1221 
1222 	memset(&crq, 0, sizeof(crq));
1223 	crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1224 	crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1225 	crq.logical_link_state.link_state = link_state;
1226 
1227 	do {
1228 		resend = false;
1229 
1230 		reinit_completion(&adapter->init_done);
1231 		rc = ibmvnic_send_crq(adapter, &crq);
1232 		if (rc) {
1233 			netdev_err(netdev, "Failed to set link state\n");
1234 			return rc;
1235 		}
1236 
1237 		if (!wait_for_completion_timeout(&adapter->init_done,
1238 						 timeout)) {
1239 			netdev_err(netdev, "timeout setting link state\n");
1240 			return -ETIMEDOUT;
1241 		}
1242 
1243 		if (adapter->init_done_rc == PARTIALSUCCESS) {
1244 			/* Partuial success, delay and re-send */
1245 			mdelay(1000);
1246 			resend = true;
1247 		} else if (adapter->init_done_rc) {
1248 			netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1249 				    adapter->init_done_rc);
1250 			return adapter->init_done_rc;
1251 		}
1252 	} while (resend);
1253 
1254 	return 0;
1255 }
1256 
1257 static int set_real_num_queues(struct net_device *netdev)
1258 {
1259 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1260 	int rc;
1261 
1262 	netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1263 		   adapter->req_tx_queues, adapter->req_rx_queues);
1264 
1265 	rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1266 	if (rc) {
1267 		netdev_err(netdev, "failed to set the number of tx queues\n");
1268 		return rc;
1269 	}
1270 
1271 	rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1272 	if (rc)
1273 		netdev_err(netdev, "failed to set the number of rx queues\n");
1274 
1275 	return rc;
1276 }
1277 
1278 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1279 {
1280 	struct device *dev = &adapter->vdev->dev;
1281 	union ibmvnic_crq crq;
1282 	int len = 0;
1283 	int rc;
1284 
1285 	if (adapter->vpd->buff)
1286 		len = adapter->vpd->len;
1287 
1288 	mutex_lock(&adapter->fw_lock);
1289 	adapter->fw_done_rc = 0;
1290 	reinit_completion(&adapter->fw_done);
1291 
1292 	crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1293 	crq.get_vpd_size.cmd = GET_VPD_SIZE;
1294 	rc = ibmvnic_send_crq(adapter, &crq);
1295 	if (rc) {
1296 		mutex_unlock(&adapter->fw_lock);
1297 		return rc;
1298 	}
1299 
1300 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1301 	if (rc) {
1302 		dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1303 		mutex_unlock(&adapter->fw_lock);
1304 		return rc;
1305 	}
1306 	mutex_unlock(&adapter->fw_lock);
1307 
1308 	if (!adapter->vpd->len)
1309 		return -ENODATA;
1310 
1311 	if (!adapter->vpd->buff)
1312 		adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1313 	else if (adapter->vpd->len != len)
1314 		adapter->vpd->buff =
1315 			krealloc(adapter->vpd->buff,
1316 				 adapter->vpd->len, GFP_KERNEL);
1317 
1318 	if (!adapter->vpd->buff) {
1319 		dev_err(dev, "Could allocate VPD buffer\n");
1320 		return -ENOMEM;
1321 	}
1322 
1323 	adapter->vpd->dma_addr =
1324 		dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1325 			       DMA_FROM_DEVICE);
1326 	if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1327 		dev_err(dev, "Could not map VPD buffer\n");
1328 		kfree(adapter->vpd->buff);
1329 		adapter->vpd->buff = NULL;
1330 		return -ENOMEM;
1331 	}
1332 
1333 	mutex_lock(&adapter->fw_lock);
1334 	adapter->fw_done_rc = 0;
1335 	reinit_completion(&adapter->fw_done);
1336 
1337 	crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1338 	crq.get_vpd.cmd = GET_VPD;
1339 	crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1340 	crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1341 	rc = ibmvnic_send_crq(adapter, &crq);
1342 	if (rc) {
1343 		kfree(adapter->vpd->buff);
1344 		adapter->vpd->buff = NULL;
1345 		mutex_unlock(&adapter->fw_lock);
1346 		return rc;
1347 	}
1348 
1349 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1350 	if (rc) {
1351 		dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1352 		kfree(adapter->vpd->buff);
1353 		adapter->vpd->buff = NULL;
1354 		mutex_unlock(&adapter->fw_lock);
1355 		return rc;
1356 	}
1357 
1358 	mutex_unlock(&adapter->fw_lock);
1359 	return 0;
1360 }
1361 
1362 static int init_resources(struct ibmvnic_adapter *adapter)
1363 {
1364 	struct net_device *netdev = adapter->netdev;
1365 	int rc;
1366 
1367 	rc = set_real_num_queues(netdev);
1368 	if (rc)
1369 		return rc;
1370 
1371 	adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1372 	if (!adapter->vpd)
1373 		return -ENOMEM;
1374 
1375 	/* Vital Product Data (VPD) */
1376 	rc = ibmvnic_get_vpd(adapter);
1377 	if (rc) {
1378 		netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1379 		return rc;
1380 	}
1381 
1382 	rc = init_napi(adapter);
1383 	if (rc)
1384 		return rc;
1385 
1386 	send_query_map(adapter);
1387 
1388 	rc = init_rx_pools(netdev);
1389 	if (rc)
1390 		return rc;
1391 
1392 	rc = init_tx_pools(netdev);
1393 	return rc;
1394 }
1395 
1396 static int __ibmvnic_open(struct net_device *netdev)
1397 {
1398 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1399 	enum vnic_state prev_state = adapter->state;
1400 	int i, rc;
1401 
1402 	adapter->state = VNIC_OPENING;
1403 	replenish_pools(adapter);
1404 	ibmvnic_napi_enable(adapter);
1405 
1406 	/* We're ready to receive frames, enable the sub-crq interrupts and
1407 	 * set the logical link state to up
1408 	 */
1409 	for (i = 0; i < adapter->req_rx_queues; i++) {
1410 		netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1411 		if (prev_state == VNIC_CLOSED)
1412 			enable_irq(adapter->rx_scrq[i]->irq);
1413 		enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1414 	}
1415 
1416 	for (i = 0; i < adapter->req_tx_queues; i++) {
1417 		netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1418 		if (prev_state == VNIC_CLOSED)
1419 			enable_irq(adapter->tx_scrq[i]->irq);
1420 		enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1421 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1422 	}
1423 
1424 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1425 	if (rc) {
1426 		ibmvnic_napi_disable(adapter);
1427 		release_resources(adapter);
1428 		return rc;
1429 	}
1430 
1431 	netif_tx_start_all_queues(netdev);
1432 
1433 	if (prev_state == VNIC_CLOSED) {
1434 		for (i = 0; i < adapter->req_rx_queues; i++)
1435 			napi_schedule(&adapter->napi[i]);
1436 	}
1437 
1438 	adapter->state = VNIC_OPEN;
1439 	return rc;
1440 }
1441 
1442 static int ibmvnic_open(struct net_device *netdev)
1443 {
1444 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1445 	int rc;
1446 
1447 	ASSERT_RTNL();
1448 
1449 	/* If device failover is pending or we are about to reset, just set
1450 	 * device state and return. Device operation will be handled by reset
1451 	 * routine.
1452 	 *
1453 	 * It should be safe to overwrite the adapter->state here. Since
1454 	 * we hold the rtnl, either the reset has not actually started or
1455 	 * the rtnl got dropped during the set_link_state() in do_reset().
1456 	 * In the former case, no one else is changing the state (again we
1457 	 * have the rtnl) and in the latter case, do_reset() will detect and
1458 	 * honor our setting below.
1459 	 */
1460 	if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1461 		netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1462 			   adapter_state_to_string(adapter->state),
1463 			   adapter->failover_pending);
1464 		adapter->state = VNIC_OPEN;
1465 		rc = 0;
1466 		goto out;
1467 	}
1468 
1469 	if (adapter->state != VNIC_CLOSED) {
1470 		rc = ibmvnic_login(netdev);
1471 		if (rc)
1472 			goto out;
1473 
1474 		rc = init_resources(adapter);
1475 		if (rc) {
1476 			netdev_err(netdev, "failed to initialize resources\n");
1477 			release_resources(adapter);
1478 			release_rx_pools(adapter);
1479 			release_tx_pools(adapter);
1480 			goto out;
1481 		}
1482 	}
1483 
1484 	rc = __ibmvnic_open(netdev);
1485 
1486 out:
1487 	/* If open failed and there is a pending failover or in-progress reset,
1488 	 * set device state and return. Device operation will be handled by
1489 	 * reset routine. See also comments above regarding rtnl.
1490 	 */
1491 	if (rc &&
1492 	    (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1493 		adapter->state = VNIC_OPEN;
1494 		rc = 0;
1495 	}
1496 	return rc;
1497 }
1498 
1499 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1500 {
1501 	struct ibmvnic_rx_pool *rx_pool;
1502 	struct ibmvnic_rx_buff *rx_buff;
1503 	u64 rx_entries;
1504 	int rx_scrqs;
1505 	int i, j;
1506 
1507 	if (!adapter->rx_pool)
1508 		return;
1509 
1510 	rx_scrqs = adapter->num_active_rx_pools;
1511 	rx_entries = adapter->req_rx_add_entries_per_subcrq;
1512 
1513 	/* Free any remaining skbs in the rx buffer pools */
1514 	for (i = 0; i < rx_scrqs; i++) {
1515 		rx_pool = &adapter->rx_pool[i];
1516 		if (!rx_pool || !rx_pool->rx_buff)
1517 			continue;
1518 
1519 		netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1520 		for (j = 0; j < rx_entries; j++) {
1521 			rx_buff = &rx_pool->rx_buff[j];
1522 			if (rx_buff && rx_buff->skb) {
1523 				dev_kfree_skb_any(rx_buff->skb);
1524 				rx_buff->skb = NULL;
1525 			}
1526 		}
1527 	}
1528 }
1529 
1530 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1531 			      struct ibmvnic_tx_pool *tx_pool)
1532 {
1533 	struct ibmvnic_tx_buff *tx_buff;
1534 	u64 tx_entries;
1535 	int i;
1536 
1537 	if (!tx_pool || !tx_pool->tx_buff)
1538 		return;
1539 
1540 	tx_entries = tx_pool->num_buffers;
1541 
1542 	for (i = 0; i < tx_entries; i++) {
1543 		tx_buff = &tx_pool->tx_buff[i];
1544 		if (tx_buff && tx_buff->skb) {
1545 			dev_kfree_skb_any(tx_buff->skb);
1546 			tx_buff->skb = NULL;
1547 		}
1548 	}
1549 }
1550 
1551 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1552 {
1553 	int tx_scrqs;
1554 	int i;
1555 
1556 	if (!adapter->tx_pool || !adapter->tso_pool)
1557 		return;
1558 
1559 	tx_scrqs = adapter->num_active_tx_pools;
1560 
1561 	/* Free any remaining skbs in the tx buffer pools */
1562 	for (i = 0; i < tx_scrqs; i++) {
1563 		netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1564 		clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1565 		clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1566 	}
1567 }
1568 
1569 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1570 {
1571 	struct net_device *netdev = adapter->netdev;
1572 	int i;
1573 
1574 	if (adapter->tx_scrq) {
1575 		for (i = 0; i < adapter->req_tx_queues; i++)
1576 			if (adapter->tx_scrq[i]->irq) {
1577 				netdev_dbg(netdev,
1578 					   "Disabling tx_scrq[%d] irq\n", i);
1579 				disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1580 				disable_irq(adapter->tx_scrq[i]->irq);
1581 			}
1582 	}
1583 
1584 	if (adapter->rx_scrq) {
1585 		for (i = 0; i < adapter->req_rx_queues; i++) {
1586 			if (adapter->rx_scrq[i]->irq) {
1587 				netdev_dbg(netdev,
1588 					   "Disabling rx_scrq[%d] irq\n", i);
1589 				disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1590 				disable_irq(adapter->rx_scrq[i]->irq);
1591 			}
1592 		}
1593 	}
1594 }
1595 
1596 static void ibmvnic_cleanup(struct net_device *netdev)
1597 {
1598 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1599 
1600 	/* ensure that transmissions are stopped if called by do_reset */
1601 	if (test_bit(0, &adapter->resetting))
1602 		netif_tx_disable(netdev);
1603 	else
1604 		netif_tx_stop_all_queues(netdev);
1605 
1606 	ibmvnic_napi_disable(adapter);
1607 	ibmvnic_disable_irqs(adapter);
1608 }
1609 
1610 static int __ibmvnic_close(struct net_device *netdev)
1611 {
1612 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1613 	int rc = 0;
1614 
1615 	adapter->state = VNIC_CLOSING;
1616 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1617 	adapter->state = VNIC_CLOSED;
1618 	return rc;
1619 }
1620 
1621 static int ibmvnic_close(struct net_device *netdev)
1622 {
1623 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1624 	int rc;
1625 
1626 	netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
1627 		   adapter_state_to_string(adapter->state),
1628 		   adapter->failover_pending,
1629 		   adapter->force_reset_recovery);
1630 
1631 	/* If device failover is pending, just set device state and return.
1632 	 * Device operation will be handled by reset routine.
1633 	 */
1634 	if (adapter->failover_pending) {
1635 		adapter->state = VNIC_CLOSED;
1636 		return 0;
1637 	}
1638 
1639 	rc = __ibmvnic_close(netdev);
1640 	ibmvnic_cleanup(netdev);
1641 	clean_rx_pools(adapter);
1642 	clean_tx_pools(adapter);
1643 
1644 	return rc;
1645 }
1646 
1647 /**
1648  * build_hdr_data - creates L2/L3/L4 header data buffer
1649  * @hdr_field: bitfield determining needed headers
1650  * @skb: socket buffer
1651  * @hdr_len: array of header lengths
1652  * @hdr_data: buffer to write the header to
1653  *
1654  * Reads hdr_field to determine which headers are needed by firmware.
1655  * Builds a buffer containing these headers.  Saves individual header
1656  * lengths and total buffer length to be used to build descriptors.
1657  */
1658 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1659 			  int *hdr_len, u8 *hdr_data)
1660 {
1661 	int len = 0;
1662 	u8 *hdr;
1663 
1664 	if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1665 		hdr_len[0] = sizeof(struct vlan_ethhdr);
1666 	else
1667 		hdr_len[0] = sizeof(struct ethhdr);
1668 
1669 	if (skb->protocol == htons(ETH_P_IP)) {
1670 		hdr_len[1] = ip_hdr(skb)->ihl * 4;
1671 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1672 			hdr_len[2] = tcp_hdrlen(skb);
1673 		else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1674 			hdr_len[2] = sizeof(struct udphdr);
1675 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1676 		hdr_len[1] = sizeof(struct ipv6hdr);
1677 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1678 			hdr_len[2] = tcp_hdrlen(skb);
1679 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1680 			hdr_len[2] = sizeof(struct udphdr);
1681 	} else if (skb->protocol == htons(ETH_P_ARP)) {
1682 		hdr_len[1] = arp_hdr_len(skb->dev);
1683 		hdr_len[2] = 0;
1684 	}
1685 
1686 	memset(hdr_data, 0, 120);
1687 	if ((hdr_field >> 6) & 1) {
1688 		hdr = skb_mac_header(skb);
1689 		memcpy(hdr_data, hdr, hdr_len[0]);
1690 		len += hdr_len[0];
1691 	}
1692 
1693 	if ((hdr_field >> 5) & 1) {
1694 		hdr = skb_network_header(skb);
1695 		memcpy(hdr_data + len, hdr, hdr_len[1]);
1696 		len += hdr_len[1];
1697 	}
1698 
1699 	if ((hdr_field >> 4) & 1) {
1700 		hdr = skb_transport_header(skb);
1701 		memcpy(hdr_data + len, hdr, hdr_len[2]);
1702 		len += hdr_len[2];
1703 	}
1704 	return len;
1705 }
1706 
1707 /**
1708  * create_hdr_descs - create header and header extension descriptors
1709  * @hdr_field: bitfield determining needed headers
1710  * @hdr_data: buffer containing header data
1711  * @len: length of data buffer
1712  * @hdr_len: array of individual header lengths
1713  * @scrq_arr: descriptor array
1714  *
1715  * Creates header and, if needed, header extension descriptors and
1716  * places them in a descriptor array, scrq_arr
1717  */
1718 
1719 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1720 			    union sub_crq *scrq_arr)
1721 {
1722 	union sub_crq hdr_desc;
1723 	int tmp_len = len;
1724 	int num_descs = 0;
1725 	u8 *data, *cur;
1726 	int tmp;
1727 
1728 	while (tmp_len > 0) {
1729 		cur = hdr_data + len - tmp_len;
1730 
1731 		memset(&hdr_desc, 0, sizeof(hdr_desc));
1732 		if (cur != hdr_data) {
1733 			data = hdr_desc.hdr_ext.data;
1734 			tmp = tmp_len > 29 ? 29 : tmp_len;
1735 			hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1736 			hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1737 			hdr_desc.hdr_ext.len = tmp;
1738 		} else {
1739 			data = hdr_desc.hdr.data;
1740 			tmp = tmp_len > 24 ? 24 : tmp_len;
1741 			hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1742 			hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1743 			hdr_desc.hdr.len = tmp;
1744 			hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1745 			hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1746 			hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1747 			hdr_desc.hdr.flag = hdr_field << 1;
1748 		}
1749 		memcpy(data, cur, tmp);
1750 		tmp_len -= tmp;
1751 		*scrq_arr = hdr_desc;
1752 		scrq_arr++;
1753 		num_descs++;
1754 	}
1755 
1756 	return num_descs;
1757 }
1758 
1759 /**
1760  * build_hdr_descs_arr - build a header descriptor array
1761  * @skb: tx socket buffer
1762  * @indir_arr: indirect array
1763  * @num_entries: number of descriptors to be sent
1764  * @hdr_field: bit field determining which headers will be sent
1765  *
1766  * This function will build a TX descriptor array with applicable
1767  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1768  */
1769 
1770 static void build_hdr_descs_arr(struct sk_buff *skb,
1771 				union sub_crq *indir_arr,
1772 				int *num_entries, u8 hdr_field)
1773 {
1774 	int hdr_len[3] = {0, 0, 0};
1775 	u8 hdr_data[140] = {0};
1776 	int tot_len;
1777 
1778 	tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1779 				 hdr_data);
1780 	*num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1781 					 indir_arr + 1);
1782 }
1783 
1784 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1785 				    struct net_device *netdev)
1786 {
1787 	/* For some backing devices, mishandling of small packets
1788 	 * can result in a loss of connection or TX stall. Device
1789 	 * architects recommend that no packet should be smaller
1790 	 * than the minimum MTU value provided to the driver, so
1791 	 * pad any packets to that length
1792 	 */
1793 	if (skb->len < netdev->min_mtu)
1794 		return skb_put_padto(skb, netdev->min_mtu);
1795 
1796 	return 0;
1797 }
1798 
1799 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1800 					 struct ibmvnic_sub_crq_queue *tx_scrq)
1801 {
1802 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1803 	struct ibmvnic_tx_buff *tx_buff;
1804 	struct ibmvnic_tx_pool *tx_pool;
1805 	union sub_crq tx_scrq_entry;
1806 	int queue_num;
1807 	int entries;
1808 	int index;
1809 	int i;
1810 
1811 	ind_bufp = &tx_scrq->ind_buf;
1812 	entries = (u64)ind_bufp->index;
1813 	queue_num = tx_scrq->pool_index;
1814 
1815 	for (i = entries - 1; i >= 0; --i) {
1816 		tx_scrq_entry = ind_bufp->indir_arr[i];
1817 		if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1818 			continue;
1819 		index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1820 		if (index & IBMVNIC_TSO_POOL_MASK) {
1821 			tx_pool = &adapter->tso_pool[queue_num];
1822 			index &= ~IBMVNIC_TSO_POOL_MASK;
1823 		} else {
1824 			tx_pool = &adapter->tx_pool[queue_num];
1825 		}
1826 		tx_pool->free_map[tx_pool->consumer_index] = index;
1827 		tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1828 					  tx_pool->num_buffers - 1 :
1829 					  tx_pool->consumer_index - 1;
1830 		tx_buff = &tx_pool->tx_buff[index];
1831 		adapter->netdev->stats.tx_packets--;
1832 		adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1833 		adapter->tx_stats_buffers[queue_num].packets--;
1834 		adapter->tx_stats_buffers[queue_num].bytes -=
1835 						tx_buff->skb->len;
1836 		dev_kfree_skb_any(tx_buff->skb);
1837 		tx_buff->skb = NULL;
1838 		adapter->netdev->stats.tx_dropped++;
1839 	}
1840 	ind_bufp->index = 0;
1841 	if (atomic_sub_return(entries, &tx_scrq->used) <=
1842 	    (adapter->req_tx_entries_per_subcrq / 2) &&
1843 	    __netif_subqueue_stopped(adapter->netdev, queue_num) &&
1844 	    !test_bit(0, &adapter->resetting)) {
1845 		netif_wake_subqueue(adapter->netdev, queue_num);
1846 		netdev_dbg(adapter->netdev, "Started queue %d\n",
1847 			   queue_num);
1848 	}
1849 }
1850 
1851 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1852 				 struct ibmvnic_sub_crq_queue *tx_scrq)
1853 {
1854 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1855 	u64 dma_addr;
1856 	u64 entries;
1857 	u64 handle;
1858 	int rc;
1859 
1860 	ind_bufp = &tx_scrq->ind_buf;
1861 	dma_addr = (u64)ind_bufp->indir_dma;
1862 	entries = (u64)ind_bufp->index;
1863 	handle = tx_scrq->handle;
1864 
1865 	if (!entries)
1866 		return 0;
1867 	rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1868 	if (rc)
1869 		ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1870 	else
1871 		ind_bufp->index = 0;
1872 	return 0;
1873 }
1874 
1875 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1876 {
1877 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1878 	int queue_num = skb_get_queue_mapping(skb);
1879 	u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1880 	struct device *dev = &adapter->vdev->dev;
1881 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1882 	struct ibmvnic_tx_buff *tx_buff = NULL;
1883 	struct ibmvnic_sub_crq_queue *tx_scrq;
1884 	struct ibmvnic_tx_pool *tx_pool;
1885 	unsigned int tx_send_failed = 0;
1886 	netdev_tx_t ret = NETDEV_TX_OK;
1887 	unsigned int tx_map_failed = 0;
1888 	union sub_crq indir_arr[16];
1889 	unsigned int tx_dropped = 0;
1890 	unsigned int tx_packets = 0;
1891 	unsigned int tx_bytes = 0;
1892 	dma_addr_t data_dma_addr;
1893 	struct netdev_queue *txq;
1894 	unsigned long lpar_rc;
1895 	union sub_crq tx_crq;
1896 	unsigned int offset;
1897 	int num_entries = 1;
1898 	unsigned char *dst;
1899 	int index = 0;
1900 	u8 proto = 0;
1901 
1902 	tx_scrq = adapter->tx_scrq[queue_num];
1903 	txq = netdev_get_tx_queue(netdev, queue_num);
1904 	ind_bufp = &tx_scrq->ind_buf;
1905 
1906 	if (test_bit(0, &adapter->resetting)) {
1907 		dev_kfree_skb_any(skb);
1908 
1909 		tx_send_failed++;
1910 		tx_dropped++;
1911 		ret = NETDEV_TX_OK;
1912 		goto out;
1913 	}
1914 
1915 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
1916 		tx_dropped++;
1917 		tx_send_failed++;
1918 		ret = NETDEV_TX_OK;
1919 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1920 		goto out;
1921 	}
1922 	if (skb_is_gso(skb))
1923 		tx_pool = &adapter->tso_pool[queue_num];
1924 	else
1925 		tx_pool = &adapter->tx_pool[queue_num];
1926 
1927 	index = tx_pool->free_map[tx_pool->consumer_index];
1928 
1929 	if (index == IBMVNIC_INVALID_MAP) {
1930 		dev_kfree_skb_any(skb);
1931 		tx_send_failed++;
1932 		tx_dropped++;
1933 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1934 		ret = NETDEV_TX_OK;
1935 		goto out;
1936 	}
1937 
1938 	tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1939 
1940 	offset = index * tx_pool->buf_size;
1941 	dst = tx_pool->long_term_buff.buff + offset;
1942 	memset(dst, 0, tx_pool->buf_size);
1943 	data_dma_addr = tx_pool->long_term_buff.addr + offset;
1944 
1945 	if (skb_shinfo(skb)->nr_frags) {
1946 		int cur, i;
1947 
1948 		/* Copy the head */
1949 		skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1950 		cur = skb_headlen(skb);
1951 
1952 		/* Copy the frags */
1953 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1954 			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1955 
1956 			memcpy(dst + cur, skb_frag_address(frag),
1957 			       skb_frag_size(frag));
1958 			cur += skb_frag_size(frag);
1959 		}
1960 	} else {
1961 		skb_copy_from_linear_data(skb, dst, skb->len);
1962 	}
1963 
1964 	/* post changes to long_term_buff *dst before VIOS accessing it */
1965 	dma_wmb();
1966 
1967 	tx_pool->consumer_index =
1968 	    (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1969 
1970 	tx_buff = &tx_pool->tx_buff[index];
1971 	tx_buff->skb = skb;
1972 	tx_buff->index = index;
1973 	tx_buff->pool_index = queue_num;
1974 
1975 	memset(&tx_crq, 0, sizeof(tx_crq));
1976 	tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1977 	tx_crq.v1.type = IBMVNIC_TX_DESC;
1978 	tx_crq.v1.n_crq_elem = 1;
1979 	tx_crq.v1.n_sge = 1;
1980 	tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1981 
1982 	if (skb_is_gso(skb))
1983 		tx_crq.v1.correlator =
1984 			cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1985 	else
1986 		tx_crq.v1.correlator = cpu_to_be32(index);
1987 	tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1988 	tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1989 	tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1990 
1991 	if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1992 		tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1993 		tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1994 	}
1995 
1996 	if (skb->protocol == htons(ETH_P_IP)) {
1997 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1998 		proto = ip_hdr(skb)->protocol;
1999 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2000 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2001 		proto = ipv6_hdr(skb)->nexthdr;
2002 	}
2003 
2004 	if (proto == IPPROTO_TCP)
2005 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2006 	else if (proto == IPPROTO_UDP)
2007 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2008 
2009 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2010 		tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2011 		hdrs += 2;
2012 	}
2013 	if (skb_is_gso(skb)) {
2014 		tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2015 		tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2016 		hdrs += 2;
2017 	}
2018 
2019 	if ((*hdrs >> 7) & 1)
2020 		build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2021 
2022 	tx_crq.v1.n_crq_elem = num_entries;
2023 	tx_buff->num_entries = num_entries;
2024 	/* flush buffer if current entry can not fit */
2025 	if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2026 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2027 		if (lpar_rc != H_SUCCESS)
2028 			goto tx_flush_err;
2029 	}
2030 
2031 	indir_arr[0] = tx_crq;
2032 	memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2033 	       num_entries * sizeof(struct ibmvnic_generic_scrq));
2034 	ind_bufp->index += num_entries;
2035 	if (__netdev_tx_sent_queue(txq, skb->len,
2036 				   netdev_xmit_more() &&
2037 				   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2038 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2039 		if (lpar_rc != H_SUCCESS)
2040 			goto tx_err;
2041 	}
2042 
2043 	if (atomic_add_return(num_entries, &tx_scrq->used)
2044 					>= adapter->req_tx_entries_per_subcrq) {
2045 		netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2046 		netif_stop_subqueue(netdev, queue_num);
2047 	}
2048 
2049 	tx_packets++;
2050 	tx_bytes += skb->len;
2051 	txq_trans_cond_update(txq);
2052 	ret = NETDEV_TX_OK;
2053 	goto out;
2054 
2055 tx_flush_err:
2056 	dev_kfree_skb_any(skb);
2057 	tx_buff->skb = NULL;
2058 	tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2059 				  tx_pool->num_buffers - 1 :
2060 				  tx_pool->consumer_index - 1;
2061 	tx_dropped++;
2062 tx_err:
2063 	if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2064 		dev_err_ratelimited(dev, "tx: send failed\n");
2065 
2066 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2067 		/* Disable TX and report carrier off if queue is closed
2068 		 * or pending failover.
2069 		 * Firmware guarantees that a signal will be sent to the
2070 		 * driver, triggering a reset or some other action.
2071 		 */
2072 		netif_tx_stop_all_queues(netdev);
2073 		netif_carrier_off(netdev);
2074 	}
2075 out:
2076 	netdev->stats.tx_dropped += tx_dropped;
2077 	netdev->stats.tx_bytes += tx_bytes;
2078 	netdev->stats.tx_packets += tx_packets;
2079 	adapter->tx_send_failed += tx_send_failed;
2080 	adapter->tx_map_failed += tx_map_failed;
2081 	adapter->tx_stats_buffers[queue_num].packets += tx_packets;
2082 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2083 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2084 
2085 	return ret;
2086 }
2087 
2088 static void ibmvnic_set_multi(struct net_device *netdev)
2089 {
2090 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2091 	struct netdev_hw_addr *ha;
2092 	union ibmvnic_crq crq;
2093 
2094 	memset(&crq, 0, sizeof(crq));
2095 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
2096 	crq.request_capability.cmd = REQUEST_CAPABILITY;
2097 
2098 	if (netdev->flags & IFF_PROMISC) {
2099 		if (!adapter->promisc_supported)
2100 			return;
2101 	} else {
2102 		if (netdev->flags & IFF_ALLMULTI) {
2103 			/* Accept all multicast */
2104 			memset(&crq, 0, sizeof(crq));
2105 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2106 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2107 			crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2108 			ibmvnic_send_crq(adapter, &crq);
2109 		} else if (netdev_mc_empty(netdev)) {
2110 			/* Reject all multicast */
2111 			memset(&crq, 0, sizeof(crq));
2112 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2113 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2114 			crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2115 			ibmvnic_send_crq(adapter, &crq);
2116 		} else {
2117 			/* Accept one or more multicast(s) */
2118 			netdev_for_each_mc_addr(ha, netdev) {
2119 				memset(&crq, 0, sizeof(crq));
2120 				crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2121 				crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2122 				crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2123 				ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2124 						ha->addr);
2125 				ibmvnic_send_crq(adapter, &crq);
2126 			}
2127 		}
2128 	}
2129 }
2130 
2131 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2132 {
2133 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2134 	union ibmvnic_crq crq;
2135 	int rc;
2136 
2137 	if (!is_valid_ether_addr(dev_addr)) {
2138 		rc = -EADDRNOTAVAIL;
2139 		goto err;
2140 	}
2141 
2142 	memset(&crq, 0, sizeof(crq));
2143 	crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2144 	crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2145 	ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2146 
2147 	mutex_lock(&adapter->fw_lock);
2148 	adapter->fw_done_rc = 0;
2149 	reinit_completion(&adapter->fw_done);
2150 
2151 	rc = ibmvnic_send_crq(adapter, &crq);
2152 	if (rc) {
2153 		rc = -EIO;
2154 		mutex_unlock(&adapter->fw_lock);
2155 		goto err;
2156 	}
2157 
2158 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2159 	/* netdev->dev_addr is changed in handle_change_mac_rsp function */
2160 	if (rc || adapter->fw_done_rc) {
2161 		rc = -EIO;
2162 		mutex_unlock(&adapter->fw_lock);
2163 		goto err;
2164 	}
2165 	mutex_unlock(&adapter->fw_lock);
2166 	return 0;
2167 err:
2168 	ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2169 	return rc;
2170 }
2171 
2172 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2173 {
2174 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2175 	struct sockaddr *addr = p;
2176 	int rc;
2177 
2178 	rc = 0;
2179 	if (!is_valid_ether_addr(addr->sa_data))
2180 		return -EADDRNOTAVAIL;
2181 
2182 	ether_addr_copy(adapter->mac_addr, addr->sa_data);
2183 	if (adapter->state != VNIC_PROBED)
2184 		rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2185 
2186 	return rc;
2187 }
2188 
2189 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2190 {
2191 	switch (reason) {
2192 	case VNIC_RESET_FAILOVER:
2193 		return "FAILOVER";
2194 	case VNIC_RESET_MOBILITY:
2195 		return "MOBILITY";
2196 	case VNIC_RESET_FATAL:
2197 		return "FATAL";
2198 	case VNIC_RESET_NON_FATAL:
2199 		return "NON_FATAL";
2200 	case VNIC_RESET_TIMEOUT:
2201 		return "TIMEOUT";
2202 	case VNIC_RESET_CHANGE_PARAM:
2203 		return "CHANGE_PARAM";
2204 	case VNIC_RESET_PASSIVE_INIT:
2205 		return "PASSIVE_INIT";
2206 	}
2207 	return "UNKNOWN";
2208 }
2209 
2210 /*
2211  * do_reset returns zero if we are able to keep processing reset events, or
2212  * non-zero if we hit a fatal error and must halt.
2213  */
2214 static int do_reset(struct ibmvnic_adapter *adapter,
2215 		    struct ibmvnic_rwi *rwi, u32 reset_state)
2216 {
2217 	struct net_device *netdev = adapter->netdev;
2218 	u64 old_num_rx_queues, old_num_tx_queues;
2219 	u64 old_num_rx_slots, old_num_tx_slots;
2220 	int rc;
2221 
2222 	netdev_dbg(adapter->netdev,
2223 		   "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2224 		   adapter_state_to_string(adapter->state),
2225 		   adapter->failover_pending,
2226 		   reset_reason_to_string(rwi->reset_reason),
2227 		   adapter_state_to_string(reset_state));
2228 
2229 	adapter->reset_reason = rwi->reset_reason;
2230 	/* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2231 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2232 		rtnl_lock();
2233 
2234 	/* Now that we have the rtnl lock, clear any pending failover.
2235 	 * This will ensure ibmvnic_open() has either completed or will
2236 	 * block until failover is complete.
2237 	 */
2238 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2239 		adapter->failover_pending = false;
2240 
2241 	/* read the state and check (again) after getting rtnl */
2242 	reset_state = adapter->state;
2243 
2244 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2245 		rc = -EBUSY;
2246 		goto out;
2247 	}
2248 
2249 	netif_carrier_off(netdev);
2250 
2251 	old_num_rx_queues = adapter->req_rx_queues;
2252 	old_num_tx_queues = adapter->req_tx_queues;
2253 	old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2254 	old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2255 
2256 	ibmvnic_cleanup(netdev);
2257 
2258 	if (reset_state == VNIC_OPEN &&
2259 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
2260 	    adapter->reset_reason != VNIC_RESET_FAILOVER) {
2261 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2262 			rc = __ibmvnic_close(netdev);
2263 			if (rc)
2264 				goto out;
2265 		} else {
2266 			adapter->state = VNIC_CLOSING;
2267 
2268 			/* Release the RTNL lock before link state change and
2269 			 * re-acquire after the link state change to allow
2270 			 * linkwatch_event to grab the RTNL lock and run during
2271 			 * a reset.
2272 			 */
2273 			rtnl_unlock();
2274 			rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2275 			rtnl_lock();
2276 			if (rc)
2277 				goto out;
2278 
2279 			if (adapter->state == VNIC_OPEN) {
2280 				/* When we dropped rtnl, ibmvnic_open() got
2281 				 * it and noticed that we are resetting and
2282 				 * set the adapter state to OPEN. Update our
2283 				 * new "target" state, and resume the reset
2284 				 * from VNIC_CLOSING state.
2285 				 */
2286 				netdev_dbg(netdev,
2287 					   "Open changed state from %s, updating.\n",
2288 					   adapter_state_to_string(reset_state));
2289 				reset_state = VNIC_OPEN;
2290 				adapter->state = VNIC_CLOSING;
2291 			}
2292 
2293 			if (adapter->state != VNIC_CLOSING) {
2294 				/* If someone else changed the adapter state
2295 				 * when we dropped the rtnl, fail the reset
2296 				 */
2297 				rc = -EAGAIN;
2298 				goto out;
2299 			}
2300 			adapter->state = VNIC_CLOSED;
2301 		}
2302 	}
2303 
2304 	if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2305 		release_resources(adapter);
2306 		release_sub_crqs(adapter, 1);
2307 		release_crq_queue(adapter);
2308 	}
2309 
2310 	if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2311 		/* remove the closed state so when we call open it appears
2312 		 * we are coming from the probed state.
2313 		 */
2314 		adapter->state = VNIC_PROBED;
2315 
2316 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2317 			rc = init_crq_queue(adapter);
2318 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2319 			rc = ibmvnic_reenable_crq_queue(adapter);
2320 			release_sub_crqs(adapter, 1);
2321 		} else {
2322 			rc = ibmvnic_reset_crq(adapter);
2323 			if (rc == H_CLOSED || rc == H_SUCCESS) {
2324 				rc = vio_enable_interrupts(adapter->vdev);
2325 				if (rc)
2326 					netdev_err(adapter->netdev,
2327 						   "Reset failed to enable interrupts. rc=%d\n",
2328 						   rc);
2329 			}
2330 		}
2331 
2332 		if (rc) {
2333 			netdev_err(adapter->netdev,
2334 				   "Reset couldn't initialize crq. rc=%d\n", rc);
2335 			goto out;
2336 		}
2337 
2338 		rc = ibmvnic_reset_init(adapter, true);
2339 		if (rc)
2340 			goto out;
2341 
2342 		/* If the adapter was in PROBE or DOWN state prior to the reset,
2343 		 * exit here.
2344 		 */
2345 		if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2346 			rc = 0;
2347 			goto out;
2348 		}
2349 
2350 		rc = ibmvnic_login(netdev);
2351 		if (rc)
2352 			goto out;
2353 
2354 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2355 			rc = init_resources(adapter);
2356 			if (rc)
2357 				goto out;
2358 		} else if (adapter->req_rx_queues != old_num_rx_queues ||
2359 		    adapter->req_tx_queues != old_num_tx_queues ||
2360 		    adapter->req_rx_add_entries_per_subcrq !=
2361 		    old_num_rx_slots ||
2362 		    adapter->req_tx_entries_per_subcrq !=
2363 		    old_num_tx_slots ||
2364 		    !adapter->rx_pool ||
2365 		    !adapter->tso_pool ||
2366 		    !adapter->tx_pool) {
2367 			release_napi(adapter);
2368 			release_vpd_data(adapter);
2369 
2370 			rc = init_resources(adapter);
2371 			if (rc)
2372 				goto out;
2373 
2374 		} else {
2375 			rc = init_tx_pools(netdev);
2376 			if (rc) {
2377 				netdev_dbg(netdev,
2378 					   "init tx pools failed (%d)\n",
2379 					   rc);
2380 				goto out;
2381 			}
2382 
2383 			rc = init_rx_pools(netdev);
2384 			if (rc) {
2385 				netdev_dbg(netdev,
2386 					   "init rx pools failed (%d)\n",
2387 					   rc);
2388 				goto out;
2389 			}
2390 		}
2391 		ibmvnic_disable_irqs(adapter);
2392 	}
2393 	adapter->state = VNIC_CLOSED;
2394 
2395 	if (reset_state == VNIC_CLOSED) {
2396 		rc = 0;
2397 		goto out;
2398 	}
2399 
2400 	rc = __ibmvnic_open(netdev);
2401 	if (rc) {
2402 		rc = IBMVNIC_OPEN_FAILED;
2403 		goto out;
2404 	}
2405 
2406 	/* refresh device's multicast list */
2407 	ibmvnic_set_multi(netdev);
2408 
2409 	if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2410 	    adapter->reset_reason == VNIC_RESET_MOBILITY)
2411 		__netdev_notify_peers(netdev);
2412 
2413 	rc = 0;
2414 
2415 out:
2416 	/* restore the adapter state if reset failed */
2417 	if (rc)
2418 		adapter->state = reset_state;
2419 	/* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2420 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2421 		rtnl_unlock();
2422 
2423 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2424 		   adapter_state_to_string(adapter->state),
2425 		   adapter->failover_pending, rc);
2426 	return rc;
2427 }
2428 
2429 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2430 			 struct ibmvnic_rwi *rwi, u32 reset_state)
2431 {
2432 	struct net_device *netdev = adapter->netdev;
2433 	int rc;
2434 
2435 	netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2436 		   reset_reason_to_string(rwi->reset_reason));
2437 
2438 	/* read the state and check (again) after getting rtnl */
2439 	reset_state = adapter->state;
2440 
2441 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2442 		rc = -EBUSY;
2443 		goto out;
2444 	}
2445 
2446 	netif_carrier_off(netdev);
2447 	adapter->reset_reason = rwi->reset_reason;
2448 
2449 	ibmvnic_cleanup(netdev);
2450 	release_resources(adapter);
2451 	release_sub_crqs(adapter, 0);
2452 	release_crq_queue(adapter);
2453 
2454 	/* remove the closed state so when we call open it appears
2455 	 * we are coming from the probed state.
2456 	 */
2457 	adapter->state = VNIC_PROBED;
2458 
2459 	reinit_completion(&adapter->init_done);
2460 	rc = init_crq_queue(adapter);
2461 	if (rc) {
2462 		netdev_err(adapter->netdev,
2463 			   "Couldn't initialize crq. rc=%d\n", rc);
2464 		goto out;
2465 	}
2466 
2467 	rc = ibmvnic_reset_init(adapter, false);
2468 	if (rc)
2469 		goto out;
2470 
2471 	/* If the adapter was in PROBE or DOWN state prior to the reset,
2472 	 * exit here.
2473 	 */
2474 	if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
2475 		goto out;
2476 
2477 	rc = ibmvnic_login(netdev);
2478 	if (rc)
2479 		goto out;
2480 
2481 	rc = init_resources(adapter);
2482 	if (rc)
2483 		goto out;
2484 
2485 	ibmvnic_disable_irqs(adapter);
2486 	adapter->state = VNIC_CLOSED;
2487 
2488 	if (reset_state == VNIC_CLOSED)
2489 		goto out;
2490 
2491 	rc = __ibmvnic_open(netdev);
2492 	if (rc) {
2493 		rc = IBMVNIC_OPEN_FAILED;
2494 		goto out;
2495 	}
2496 
2497 	__netdev_notify_peers(netdev);
2498 out:
2499 	/* restore adapter state if reset failed */
2500 	if (rc)
2501 		adapter->state = reset_state;
2502 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
2503 		   adapter_state_to_string(adapter->state),
2504 		   adapter->failover_pending, rc);
2505 	return rc;
2506 }
2507 
2508 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2509 {
2510 	struct ibmvnic_rwi *rwi;
2511 	unsigned long flags;
2512 
2513 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2514 
2515 	if (!list_empty(&adapter->rwi_list)) {
2516 		rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2517 				       list);
2518 		list_del(&rwi->list);
2519 	} else {
2520 		rwi = NULL;
2521 	}
2522 
2523 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2524 	return rwi;
2525 }
2526 
2527 /**
2528  * do_passive_init - complete probing when partner device is detected.
2529  * @adapter: ibmvnic_adapter struct
2530  *
2531  * If the ibmvnic device does not have a partner device to communicate with at boot
2532  * and that partner device comes online at a later time, this function is called
2533  * to complete the initialization process of ibmvnic device.
2534  * Caller is expected to hold rtnl_lock().
2535  *
2536  * Returns non-zero if sub-CRQs are not initialized properly leaving the device
2537  * in the down state.
2538  * Returns 0 upon success and the device is in PROBED state.
2539  */
2540 
2541 static int do_passive_init(struct ibmvnic_adapter *adapter)
2542 {
2543 	unsigned long timeout = msecs_to_jiffies(30000);
2544 	struct net_device *netdev = adapter->netdev;
2545 	struct device *dev = &adapter->vdev->dev;
2546 	int rc;
2547 
2548 	netdev_dbg(netdev, "Partner device found, probing.\n");
2549 
2550 	adapter->state = VNIC_PROBING;
2551 	reinit_completion(&adapter->init_done);
2552 	adapter->init_done_rc = 0;
2553 	adapter->crq.active = true;
2554 
2555 	rc = send_crq_init_complete(adapter);
2556 	if (rc)
2557 		goto out;
2558 
2559 	rc = send_version_xchg(adapter);
2560 	if (rc)
2561 		netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
2562 
2563 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
2564 		dev_err(dev, "Initialization sequence timed out\n");
2565 		rc = -ETIMEDOUT;
2566 		goto out;
2567 	}
2568 
2569 	rc = init_sub_crqs(adapter);
2570 	if (rc) {
2571 		dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
2572 		goto out;
2573 	}
2574 
2575 	rc = init_sub_crq_irqs(adapter);
2576 	if (rc) {
2577 		dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
2578 		goto init_failed;
2579 	}
2580 
2581 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
2582 	netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
2583 	netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
2584 
2585 	adapter->state = VNIC_PROBED;
2586 	netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
2587 
2588 	return 0;
2589 
2590 init_failed:
2591 	release_sub_crqs(adapter, 1);
2592 out:
2593 	adapter->state = VNIC_DOWN;
2594 	return rc;
2595 }
2596 
2597 static void __ibmvnic_reset(struct work_struct *work)
2598 {
2599 	struct ibmvnic_adapter *adapter;
2600 	bool saved_state = false;
2601 	struct ibmvnic_rwi *tmprwi;
2602 	struct ibmvnic_rwi *rwi;
2603 	unsigned long flags;
2604 	u32 reset_state;
2605 	int rc = 0;
2606 
2607 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2608 
2609 	if (test_and_set_bit_lock(0, &adapter->resetting)) {
2610 		queue_delayed_work(system_long_wq,
2611 				   &adapter->ibmvnic_delayed_reset,
2612 				   IBMVNIC_RESET_DELAY);
2613 		return;
2614 	}
2615 
2616 	rwi = get_next_rwi(adapter);
2617 	while (rwi) {
2618 		spin_lock_irqsave(&adapter->state_lock, flags);
2619 
2620 		if (adapter->state == VNIC_REMOVING ||
2621 		    adapter->state == VNIC_REMOVED) {
2622 			spin_unlock_irqrestore(&adapter->state_lock, flags);
2623 			kfree(rwi);
2624 			rc = EBUSY;
2625 			break;
2626 		}
2627 
2628 		if (!saved_state) {
2629 			reset_state = adapter->state;
2630 			saved_state = true;
2631 		}
2632 		spin_unlock_irqrestore(&adapter->state_lock, flags);
2633 
2634 		if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
2635 			rtnl_lock();
2636 			rc = do_passive_init(adapter);
2637 			rtnl_unlock();
2638 			if (!rc)
2639 				netif_carrier_on(adapter->netdev);
2640 		} else if (adapter->force_reset_recovery) {
2641 			/* Since we are doing a hard reset now, clear the
2642 			 * failover_pending flag so we don't ignore any
2643 			 * future MOBILITY or other resets.
2644 			 */
2645 			adapter->failover_pending = false;
2646 
2647 			/* Transport event occurred during previous reset */
2648 			if (adapter->wait_for_reset) {
2649 				/* Previous was CHANGE_PARAM; caller locked */
2650 				adapter->force_reset_recovery = false;
2651 				rc = do_hard_reset(adapter, rwi, reset_state);
2652 			} else {
2653 				rtnl_lock();
2654 				adapter->force_reset_recovery = false;
2655 				rc = do_hard_reset(adapter, rwi, reset_state);
2656 				rtnl_unlock();
2657 			}
2658 			if (rc) {
2659 				/* give backing device time to settle down */
2660 				netdev_dbg(adapter->netdev,
2661 					   "[S:%s] Hard reset failed, waiting 60 secs\n",
2662 					   adapter_state_to_string(adapter->state));
2663 				set_current_state(TASK_UNINTERRUPTIBLE);
2664 				schedule_timeout(60 * HZ);
2665 			}
2666 		} else {
2667 			rc = do_reset(adapter, rwi, reset_state);
2668 		}
2669 		tmprwi = rwi;
2670 		adapter->last_reset_time = jiffies;
2671 
2672 		if (rc)
2673 			netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2674 
2675 		rwi = get_next_rwi(adapter);
2676 
2677 		/*
2678 		 * If there is another reset queued, free the previous rwi
2679 		 * and process the new reset even if previous reset failed
2680 		 * (the previous reset could have failed because of a fail
2681 		 * over for instance, so process the fail over).
2682 		 *
2683 		 * If there are no resets queued and the previous reset failed,
2684 		 * the adapter would be in an undefined state. So retry the
2685 		 * previous reset as a hard reset.
2686 		 */
2687 		if (rwi)
2688 			kfree(tmprwi);
2689 		else if (rc)
2690 			rwi = tmprwi;
2691 
2692 		if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2693 			    rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
2694 			adapter->force_reset_recovery = true;
2695 	}
2696 
2697 	if (adapter->wait_for_reset) {
2698 		adapter->reset_done_rc = rc;
2699 		complete(&adapter->reset_done);
2700 	}
2701 
2702 	clear_bit_unlock(0, &adapter->resetting);
2703 
2704 	netdev_dbg(adapter->netdev,
2705 		   "[S:%s FRR:%d WFR:%d] Done processing resets\n",
2706 		   adapter_state_to_string(adapter->state),
2707 		   adapter->force_reset_recovery,
2708 		   adapter->wait_for_reset);
2709 }
2710 
2711 static void __ibmvnic_delayed_reset(struct work_struct *work)
2712 {
2713 	struct ibmvnic_adapter *adapter;
2714 
2715 	adapter = container_of(work, struct ibmvnic_adapter,
2716 			       ibmvnic_delayed_reset.work);
2717 	__ibmvnic_reset(&adapter->ibmvnic_reset);
2718 }
2719 
2720 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2721 			 enum ibmvnic_reset_reason reason)
2722 {
2723 	struct list_head *entry, *tmp_entry;
2724 	struct ibmvnic_rwi *rwi, *tmp;
2725 	struct net_device *netdev = adapter->netdev;
2726 	unsigned long flags;
2727 	int ret;
2728 
2729 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2730 
2731 	/* If failover is pending don't schedule any other reset.
2732 	 * Instead let the failover complete. If there is already a
2733 	 * a failover reset scheduled, we will detect and drop the
2734 	 * duplicate reset when walking the ->rwi_list below.
2735 	 */
2736 	if (adapter->state == VNIC_REMOVING ||
2737 	    adapter->state == VNIC_REMOVED ||
2738 	    (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2739 		ret = EBUSY;
2740 		netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2741 		goto err;
2742 	}
2743 
2744 	if (adapter->state == VNIC_PROBING) {
2745 		netdev_warn(netdev, "Adapter reset during probe\n");
2746 		adapter->init_done_rc = -EAGAIN;
2747 		ret = EAGAIN;
2748 		goto err;
2749 	}
2750 
2751 	list_for_each_entry(tmp, &adapter->rwi_list, list) {
2752 		if (tmp->reset_reason == reason) {
2753 			netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
2754 				   reset_reason_to_string(reason));
2755 			ret = EBUSY;
2756 			goto err;
2757 		}
2758 	}
2759 
2760 	rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2761 	if (!rwi) {
2762 		ret = ENOMEM;
2763 		goto err;
2764 	}
2765 	/* if we just received a transport event,
2766 	 * flush reset queue and process this reset
2767 	 */
2768 	if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2769 		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2770 			list_del(entry);
2771 	}
2772 	rwi->reset_reason = reason;
2773 	list_add_tail(&rwi->list, &adapter->rwi_list);
2774 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
2775 		   reset_reason_to_string(reason));
2776 	queue_work(system_long_wq, &adapter->ibmvnic_reset);
2777 
2778 	ret = 0;
2779 err:
2780 	/* ibmvnic_close() below can block, so drop the lock first */
2781 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2782 
2783 	if (ret == ENOMEM)
2784 		ibmvnic_close(netdev);
2785 
2786 	return -ret;
2787 }
2788 
2789 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2790 {
2791 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2792 
2793 	if (test_bit(0, &adapter->resetting)) {
2794 		netdev_err(adapter->netdev,
2795 			   "Adapter is resetting, skip timeout reset\n");
2796 		return;
2797 	}
2798 	/* No queuing up reset until at least 5 seconds (default watchdog val)
2799 	 * after last reset
2800 	 */
2801 	if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2802 		netdev_dbg(dev, "Not yet time to tx timeout.\n");
2803 		return;
2804 	}
2805 	ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2806 }
2807 
2808 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2809 				  struct ibmvnic_rx_buff *rx_buff)
2810 {
2811 	struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2812 
2813 	rx_buff->skb = NULL;
2814 
2815 	pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2816 	pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2817 
2818 	atomic_dec(&pool->available);
2819 }
2820 
2821 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2822 {
2823 	struct ibmvnic_sub_crq_queue *rx_scrq;
2824 	struct ibmvnic_adapter *adapter;
2825 	struct net_device *netdev;
2826 	int frames_processed;
2827 	int scrq_num;
2828 
2829 	netdev = napi->dev;
2830 	adapter = netdev_priv(netdev);
2831 	scrq_num = (int)(napi - adapter->napi);
2832 	frames_processed = 0;
2833 	rx_scrq = adapter->rx_scrq[scrq_num];
2834 
2835 restart_poll:
2836 	while (frames_processed < budget) {
2837 		struct sk_buff *skb;
2838 		struct ibmvnic_rx_buff *rx_buff;
2839 		union sub_crq *next;
2840 		u32 length;
2841 		u16 offset;
2842 		u8 flags = 0;
2843 
2844 		if (unlikely(test_bit(0, &adapter->resetting) &&
2845 			     adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2846 			enable_scrq_irq(adapter, rx_scrq);
2847 			napi_complete_done(napi, frames_processed);
2848 			return frames_processed;
2849 		}
2850 
2851 		if (!pending_scrq(adapter, rx_scrq))
2852 			break;
2853 		next = ibmvnic_next_scrq(adapter, rx_scrq);
2854 		rx_buff = (struct ibmvnic_rx_buff *)
2855 			  be64_to_cpu(next->rx_comp.correlator);
2856 		/* do error checking */
2857 		if (next->rx_comp.rc) {
2858 			netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2859 				   be16_to_cpu(next->rx_comp.rc));
2860 			/* free the entry */
2861 			next->rx_comp.first = 0;
2862 			dev_kfree_skb_any(rx_buff->skb);
2863 			remove_buff_from_pool(adapter, rx_buff);
2864 			continue;
2865 		} else if (!rx_buff->skb) {
2866 			/* free the entry */
2867 			next->rx_comp.first = 0;
2868 			remove_buff_from_pool(adapter, rx_buff);
2869 			continue;
2870 		}
2871 
2872 		length = be32_to_cpu(next->rx_comp.len);
2873 		offset = be16_to_cpu(next->rx_comp.off_frame_data);
2874 		flags = next->rx_comp.flags;
2875 		skb = rx_buff->skb;
2876 		/* load long_term_buff before copying to skb */
2877 		dma_rmb();
2878 		skb_copy_to_linear_data(skb, rx_buff->data + offset,
2879 					length);
2880 
2881 		/* VLAN Header has been stripped by the system firmware and
2882 		 * needs to be inserted by the driver
2883 		 */
2884 		if (adapter->rx_vlan_header_insertion &&
2885 		    (flags & IBMVNIC_VLAN_STRIPPED))
2886 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2887 					       ntohs(next->rx_comp.vlan_tci));
2888 
2889 		/* free the entry */
2890 		next->rx_comp.first = 0;
2891 		remove_buff_from_pool(adapter, rx_buff);
2892 
2893 		skb_put(skb, length);
2894 		skb->protocol = eth_type_trans(skb, netdev);
2895 		skb_record_rx_queue(skb, scrq_num);
2896 
2897 		if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2898 		    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2899 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2900 		}
2901 
2902 		length = skb->len;
2903 		napi_gro_receive(napi, skb); /* send it up */
2904 		netdev->stats.rx_packets++;
2905 		netdev->stats.rx_bytes += length;
2906 		adapter->rx_stats_buffers[scrq_num].packets++;
2907 		adapter->rx_stats_buffers[scrq_num].bytes += length;
2908 		frames_processed++;
2909 	}
2910 
2911 	if (adapter->state != VNIC_CLOSING &&
2912 	    ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2913 	      adapter->req_rx_add_entries_per_subcrq / 2) ||
2914 	      frames_processed < budget))
2915 		replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2916 	if (frames_processed < budget) {
2917 		if (napi_complete_done(napi, frames_processed)) {
2918 			enable_scrq_irq(adapter, rx_scrq);
2919 			if (pending_scrq(adapter, rx_scrq)) {
2920 				if (napi_reschedule(napi)) {
2921 					disable_scrq_irq(adapter, rx_scrq);
2922 					goto restart_poll;
2923 				}
2924 			}
2925 		}
2926 	}
2927 	return frames_processed;
2928 }
2929 
2930 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2931 {
2932 	int rc, ret;
2933 
2934 	adapter->fallback.mtu = adapter->req_mtu;
2935 	adapter->fallback.rx_queues = adapter->req_rx_queues;
2936 	adapter->fallback.tx_queues = adapter->req_tx_queues;
2937 	adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2938 	adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2939 
2940 	reinit_completion(&adapter->reset_done);
2941 	adapter->wait_for_reset = true;
2942 	rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2943 
2944 	if (rc) {
2945 		ret = rc;
2946 		goto out;
2947 	}
2948 	rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2949 	if (rc) {
2950 		ret = -ENODEV;
2951 		goto out;
2952 	}
2953 
2954 	ret = 0;
2955 	if (adapter->reset_done_rc) {
2956 		ret = -EIO;
2957 		adapter->desired.mtu = adapter->fallback.mtu;
2958 		adapter->desired.rx_queues = adapter->fallback.rx_queues;
2959 		adapter->desired.tx_queues = adapter->fallback.tx_queues;
2960 		adapter->desired.rx_entries = adapter->fallback.rx_entries;
2961 		adapter->desired.tx_entries = adapter->fallback.tx_entries;
2962 
2963 		reinit_completion(&adapter->reset_done);
2964 		adapter->wait_for_reset = true;
2965 		rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2966 		if (rc) {
2967 			ret = rc;
2968 			goto out;
2969 		}
2970 		rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2971 						 60000);
2972 		if (rc) {
2973 			ret = -ENODEV;
2974 			goto out;
2975 		}
2976 	}
2977 out:
2978 	adapter->wait_for_reset = false;
2979 
2980 	return ret;
2981 }
2982 
2983 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2984 {
2985 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2986 
2987 	adapter->desired.mtu = new_mtu + ETH_HLEN;
2988 
2989 	return wait_for_reset(adapter);
2990 }
2991 
2992 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2993 						struct net_device *dev,
2994 						netdev_features_t features)
2995 {
2996 	/* Some backing hardware adapters can not
2997 	 * handle packets with a MSS less than 224
2998 	 * or with only one segment.
2999 	 */
3000 	if (skb_is_gso(skb)) {
3001 		if (skb_shinfo(skb)->gso_size < 224 ||
3002 		    skb_shinfo(skb)->gso_segs == 1)
3003 			features &= ~NETIF_F_GSO_MASK;
3004 	}
3005 
3006 	return features;
3007 }
3008 
3009 static const struct net_device_ops ibmvnic_netdev_ops = {
3010 	.ndo_open		= ibmvnic_open,
3011 	.ndo_stop		= ibmvnic_close,
3012 	.ndo_start_xmit		= ibmvnic_xmit,
3013 	.ndo_set_rx_mode	= ibmvnic_set_multi,
3014 	.ndo_set_mac_address	= ibmvnic_set_mac,
3015 	.ndo_validate_addr	= eth_validate_addr,
3016 	.ndo_tx_timeout		= ibmvnic_tx_timeout,
3017 	.ndo_change_mtu		= ibmvnic_change_mtu,
3018 	.ndo_features_check     = ibmvnic_features_check,
3019 };
3020 
3021 /* ethtool functions */
3022 
3023 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3024 				      struct ethtool_link_ksettings *cmd)
3025 {
3026 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3027 	int rc;
3028 
3029 	rc = send_query_phys_parms(adapter);
3030 	if (rc) {
3031 		adapter->speed = SPEED_UNKNOWN;
3032 		adapter->duplex = DUPLEX_UNKNOWN;
3033 	}
3034 	cmd->base.speed = adapter->speed;
3035 	cmd->base.duplex = adapter->duplex;
3036 	cmd->base.port = PORT_FIBRE;
3037 	cmd->base.phy_address = 0;
3038 	cmd->base.autoneg = AUTONEG_ENABLE;
3039 
3040 	return 0;
3041 }
3042 
3043 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3044 				struct ethtool_drvinfo *info)
3045 {
3046 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3047 
3048 	strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3049 	strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3050 	strscpy(info->fw_version, adapter->fw_version,
3051 		sizeof(info->fw_version));
3052 }
3053 
3054 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3055 {
3056 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3057 
3058 	return adapter->msg_enable;
3059 }
3060 
3061 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3062 {
3063 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3064 
3065 	adapter->msg_enable = data;
3066 }
3067 
3068 static u32 ibmvnic_get_link(struct net_device *netdev)
3069 {
3070 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3071 
3072 	/* Don't need to send a query because we request a logical link up at
3073 	 * init and then we wait for link state indications
3074 	 */
3075 	return adapter->logical_link_state;
3076 }
3077 
3078 static void ibmvnic_get_ringparam(struct net_device *netdev,
3079 				  struct ethtool_ringparam *ring,
3080 				  struct kernel_ethtool_ringparam *kernel_ring,
3081 				  struct netlink_ext_ack *extack)
3082 {
3083 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3084 
3085 	if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
3086 		ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3087 		ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3088 	} else {
3089 		ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
3090 		ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
3091 	}
3092 	ring->rx_mini_max_pending = 0;
3093 	ring->rx_jumbo_max_pending = 0;
3094 	ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3095 	ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3096 	ring->rx_mini_pending = 0;
3097 	ring->rx_jumbo_pending = 0;
3098 }
3099 
3100 static int ibmvnic_set_ringparam(struct net_device *netdev,
3101 				 struct ethtool_ringparam *ring,
3102 				 struct kernel_ethtool_ringparam *kernel_ring,
3103 				 struct netlink_ext_ack *extack)
3104 {
3105 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3106 	int ret;
3107 
3108 	ret = 0;
3109 	adapter->desired.rx_entries = ring->rx_pending;
3110 	adapter->desired.tx_entries = ring->tx_pending;
3111 
3112 	ret = wait_for_reset(adapter);
3113 
3114 	if (!ret &&
3115 	    (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
3116 	     adapter->req_tx_entries_per_subcrq != ring->tx_pending))
3117 		netdev_info(netdev,
3118 			    "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
3119 			    ring->rx_pending, ring->tx_pending,
3120 			    adapter->req_rx_add_entries_per_subcrq,
3121 			    adapter->req_tx_entries_per_subcrq);
3122 	return ret;
3123 }
3124 
3125 static void ibmvnic_get_channels(struct net_device *netdev,
3126 				 struct ethtool_channels *channels)
3127 {
3128 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3129 
3130 	if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
3131 		channels->max_rx = adapter->max_rx_queues;
3132 		channels->max_tx = adapter->max_tx_queues;
3133 	} else {
3134 		channels->max_rx = IBMVNIC_MAX_QUEUES;
3135 		channels->max_tx = IBMVNIC_MAX_QUEUES;
3136 	}
3137 
3138 	channels->max_other = 0;
3139 	channels->max_combined = 0;
3140 	channels->rx_count = adapter->req_rx_queues;
3141 	channels->tx_count = adapter->req_tx_queues;
3142 	channels->other_count = 0;
3143 	channels->combined_count = 0;
3144 }
3145 
3146 static int ibmvnic_set_channels(struct net_device *netdev,
3147 				struct ethtool_channels *channels)
3148 {
3149 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3150 	int ret;
3151 
3152 	ret = 0;
3153 	adapter->desired.rx_queues = channels->rx_count;
3154 	adapter->desired.tx_queues = channels->tx_count;
3155 
3156 	ret = wait_for_reset(adapter);
3157 
3158 	if (!ret &&
3159 	    (adapter->req_rx_queues != channels->rx_count ||
3160 	     adapter->req_tx_queues != channels->tx_count))
3161 		netdev_info(netdev,
3162 			    "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
3163 			    channels->rx_count, channels->tx_count,
3164 			    adapter->req_rx_queues, adapter->req_tx_queues);
3165 	return ret;
3166 }
3167 
3168 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3169 {
3170 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3171 	int i;
3172 
3173 	switch (stringset) {
3174 	case ETH_SS_STATS:
3175 		for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
3176 				i++, data += ETH_GSTRING_LEN)
3177 			memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3178 
3179 		for (i = 0; i < adapter->req_tx_queues; i++) {
3180 			snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3181 			data += ETH_GSTRING_LEN;
3182 
3183 			snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3184 			data += ETH_GSTRING_LEN;
3185 
3186 			snprintf(data, ETH_GSTRING_LEN,
3187 				 "tx%d_dropped_packets", i);
3188 			data += ETH_GSTRING_LEN;
3189 		}
3190 
3191 		for (i = 0; i < adapter->req_rx_queues; i++) {
3192 			snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3193 			data += ETH_GSTRING_LEN;
3194 
3195 			snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3196 			data += ETH_GSTRING_LEN;
3197 
3198 			snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3199 			data += ETH_GSTRING_LEN;
3200 		}
3201 		break;
3202 
3203 	case ETH_SS_PRIV_FLAGS:
3204 		for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
3205 			strcpy(data + i * ETH_GSTRING_LEN,
3206 			       ibmvnic_priv_flags[i]);
3207 		break;
3208 	default:
3209 		return;
3210 	}
3211 }
3212 
3213 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3214 {
3215 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3216 
3217 	switch (sset) {
3218 	case ETH_SS_STATS:
3219 		return ARRAY_SIZE(ibmvnic_stats) +
3220 		       adapter->req_tx_queues * NUM_TX_STATS +
3221 		       adapter->req_rx_queues * NUM_RX_STATS;
3222 	case ETH_SS_PRIV_FLAGS:
3223 		return ARRAY_SIZE(ibmvnic_priv_flags);
3224 	default:
3225 		return -EOPNOTSUPP;
3226 	}
3227 }
3228 
3229 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3230 				      struct ethtool_stats *stats, u64 *data)
3231 {
3232 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3233 	union ibmvnic_crq crq;
3234 	int i, j;
3235 	int rc;
3236 
3237 	memset(&crq, 0, sizeof(crq));
3238 	crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3239 	crq.request_statistics.cmd = REQUEST_STATISTICS;
3240 	crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3241 	crq.request_statistics.len =
3242 	    cpu_to_be32(sizeof(struct ibmvnic_statistics));
3243 
3244 	/* Wait for data to be written */
3245 	reinit_completion(&adapter->stats_done);
3246 	rc = ibmvnic_send_crq(adapter, &crq);
3247 	if (rc)
3248 		return;
3249 	rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3250 	if (rc)
3251 		return;
3252 
3253 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3254 		data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3255 				      (adapter, ibmvnic_stats[i].offset));
3256 
3257 	for (j = 0; j < adapter->req_tx_queues; j++) {
3258 		data[i] = adapter->tx_stats_buffers[j].packets;
3259 		i++;
3260 		data[i] = adapter->tx_stats_buffers[j].bytes;
3261 		i++;
3262 		data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3263 		i++;
3264 	}
3265 
3266 	for (j = 0; j < adapter->req_rx_queues; j++) {
3267 		data[i] = adapter->rx_stats_buffers[j].packets;
3268 		i++;
3269 		data[i] = adapter->rx_stats_buffers[j].bytes;
3270 		i++;
3271 		data[i] = adapter->rx_stats_buffers[j].interrupts;
3272 		i++;
3273 	}
3274 }
3275 
3276 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
3277 {
3278 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3279 
3280 	return adapter->priv_flags;
3281 }
3282 
3283 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
3284 {
3285 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3286 	bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
3287 
3288 	if (which_maxes)
3289 		adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
3290 	else
3291 		adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
3292 
3293 	return 0;
3294 }
3295 
3296 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3297 	.get_drvinfo		= ibmvnic_get_drvinfo,
3298 	.get_msglevel		= ibmvnic_get_msglevel,
3299 	.set_msglevel		= ibmvnic_set_msglevel,
3300 	.get_link		= ibmvnic_get_link,
3301 	.get_ringparam		= ibmvnic_get_ringparam,
3302 	.set_ringparam		= ibmvnic_set_ringparam,
3303 	.get_channels		= ibmvnic_get_channels,
3304 	.set_channels		= ibmvnic_set_channels,
3305 	.get_strings            = ibmvnic_get_strings,
3306 	.get_sset_count         = ibmvnic_get_sset_count,
3307 	.get_ethtool_stats	= ibmvnic_get_ethtool_stats,
3308 	.get_link_ksettings	= ibmvnic_get_link_ksettings,
3309 	.get_priv_flags		= ibmvnic_get_priv_flags,
3310 	.set_priv_flags		= ibmvnic_set_priv_flags,
3311 };
3312 
3313 /* Routines for managing CRQs/sCRQs  */
3314 
3315 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3316 				   struct ibmvnic_sub_crq_queue *scrq)
3317 {
3318 	int rc;
3319 
3320 	if (!scrq) {
3321 		netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3322 		return -EINVAL;
3323 	}
3324 
3325 	if (scrq->irq) {
3326 		free_irq(scrq->irq, scrq);
3327 		irq_dispose_mapping(scrq->irq);
3328 		scrq->irq = 0;
3329 	}
3330 
3331 	if (scrq->msgs) {
3332 		memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3333 		atomic_set(&scrq->used, 0);
3334 		scrq->cur = 0;
3335 		scrq->ind_buf.index = 0;
3336 	} else {
3337 		netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3338 		return -EINVAL;
3339 	}
3340 
3341 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3342 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3343 	return rc;
3344 }
3345 
3346 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3347 {
3348 	int i, rc;
3349 
3350 	if (!adapter->tx_scrq || !adapter->rx_scrq)
3351 		return -EINVAL;
3352 
3353 	for (i = 0; i < adapter->req_tx_queues; i++) {
3354 		netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3355 		rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3356 		if (rc)
3357 			return rc;
3358 	}
3359 
3360 	for (i = 0; i < adapter->req_rx_queues; i++) {
3361 		netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3362 		rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3363 		if (rc)
3364 			return rc;
3365 	}
3366 
3367 	return rc;
3368 }
3369 
3370 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3371 				  struct ibmvnic_sub_crq_queue *scrq,
3372 				  bool do_h_free)
3373 {
3374 	struct device *dev = &adapter->vdev->dev;
3375 	long rc;
3376 
3377 	netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3378 
3379 	if (do_h_free) {
3380 		/* Close the sub-crqs */
3381 		do {
3382 			rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3383 						adapter->vdev->unit_address,
3384 						scrq->crq_num);
3385 		} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3386 
3387 		if (rc) {
3388 			netdev_err(adapter->netdev,
3389 				   "Failed to release sub-CRQ %16lx, rc = %ld\n",
3390 				   scrq->crq_num, rc);
3391 		}
3392 	}
3393 
3394 	dma_free_coherent(dev,
3395 			  IBMVNIC_IND_ARR_SZ,
3396 			  scrq->ind_buf.indir_arr,
3397 			  scrq->ind_buf.indir_dma);
3398 
3399 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3400 			 DMA_BIDIRECTIONAL);
3401 	free_pages((unsigned long)scrq->msgs, 2);
3402 	kfree(scrq);
3403 }
3404 
3405 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3406 							*adapter)
3407 {
3408 	struct device *dev = &adapter->vdev->dev;
3409 	struct ibmvnic_sub_crq_queue *scrq;
3410 	int rc;
3411 
3412 	scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3413 	if (!scrq)
3414 		return NULL;
3415 
3416 	scrq->msgs =
3417 		(union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3418 	if (!scrq->msgs) {
3419 		dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3420 		goto zero_page_failed;
3421 	}
3422 
3423 	scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3424 					 DMA_BIDIRECTIONAL);
3425 	if (dma_mapping_error(dev, scrq->msg_token)) {
3426 		dev_warn(dev, "Couldn't map crq queue messages page\n");
3427 		goto map_failed;
3428 	}
3429 
3430 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3431 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3432 
3433 	if (rc == H_RESOURCE)
3434 		rc = ibmvnic_reset_crq(adapter);
3435 
3436 	if (rc == H_CLOSED) {
3437 		dev_warn(dev, "Partner adapter not ready, waiting.\n");
3438 	} else if (rc) {
3439 		dev_warn(dev, "Error %d registering sub-crq\n", rc);
3440 		goto reg_failed;
3441 	}
3442 
3443 	scrq->adapter = adapter;
3444 	scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3445 	scrq->ind_buf.index = 0;
3446 
3447 	scrq->ind_buf.indir_arr =
3448 		dma_alloc_coherent(dev,
3449 				   IBMVNIC_IND_ARR_SZ,
3450 				   &scrq->ind_buf.indir_dma,
3451 				   GFP_KERNEL);
3452 
3453 	if (!scrq->ind_buf.indir_arr)
3454 		goto indir_failed;
3455 
3456 	spin_lock_init(&scrq->lock);
3457 
3458 	netdev_dbg(adapter->netdev,
3459 		   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3460 		   scrq->crq_num, scrq->hw_irq, scrq->irq);
3461 
3462 	return scrq;
3463 
3464 indir_failed:
3465 	do {
3466 		rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3467 					adapter->vdev->unit_address,
3468 					scrq->crq_num);
3469 	} while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3470 reg_failed:
3471 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3472 			 DMA_BIDIRECTIONAL);
3473 map_failed:
3474 	free_pages((unsigned long)scrq->msgs, 2);
3475 zero_page_failed:
3476 	kfree(scrq);
3477 
3478 	return NULL;
3479 }
3480 
3481 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3482 {
3483 	int i;
3484 
3485 	if (adapter->tx_scrq) {
3486 		for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3487 			if (!adapter->tx_scrq[i])
3488 				continue;
3489 
3490 			netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3491 				   i);
3492 			ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
3493 			if (adapter->tx_scrq[i]->irq) {
3494 				free_irq(adapter->tx_scrq[i]->irq,
3495 					 adapter->tx_scrq[i]);
3496 				irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3497 				adapter->tx_scrq[i]->irq = 0;
3498 			}
3499 
3500 			release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3501 					      do_h_free);
3502 		}
3503 
3504 		kfree(adapter->tx_scrq);
3505 		adapter->tx_scrq = NULL;
3506 		adapter->num_active_tx_scrqs = 0;
3507 	}
3508 
3509 	if (adapter->rx_scrq) {
3510 		for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3511 			if (!adapter->rx_scrq[i])
3512 				continue;
3513 
3514 			netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3515 				   i);
3516 			if (adapter->rx_scrq[i]->irq) {
3517 				free_irq(adapter->rx_scrq[i]->irq,
3518 					 adapter->rx_scrq[i]);
3519 				irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3520 				adapter->rx_scrq[i]->irq = 0;
3521 			}
3522 
3523 			release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3524 					      do_h_free);
3525 		}
3526 
3527 		kfree(adapter->rx_scrq);
3528 		adapter->rx_scrq = NULL;
3529 		adapter->num_active_rx_scrqs = 0;
3530 	}
3531 }
3532 
3533 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3534 			    struct ibmvnic_sub_crq_queue *scrq)
3535 {
3536 	struct device *dev = &adapter->vdev->dev;
3537 	unsigned long rc;
3538 
3539 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3540 				H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3541 	if (rc)
3542 		dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3543 			scrq->hw_irq, rc);
3544 	return rc;
3545 }
3546 
3547 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3548 			   struct ibmvnic_sub_crq_queue *scrq)
3549 {
3550 	struct device *dev = &adapter->vdev->dev;
3551 	unsigned long rc;
3552 
3553 	if (scrq->hw_irq > 0x100000000ULL) {
3554 		dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3555 		return 1;
3556 	}
3557 
3558 	if (test_bit(0, &adapter->resetting) &&
3559 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
3560 		u64 val = (0xff000000) | scrq->hw_irq;
3561 
3562 		rc = plpar_hcall_norets(H_EOI, val);
3563 		/* H_EOI would fail with rc = H_FUNCTION when running
3564 		 * in XIVE mode which is expected, but not an error.
3565 		 */
3566 		if (rc && (rc != H_FUNCTION))
3567 			dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3568 				val, rc);
3569 	}
3570 
3571 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3572 				H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3573 	if (rc)
3574 		dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3575 			scrq->hw_irq, rc);
3576 	return rc;
3577 }
3578 
3579 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3580 			       struct ibmvnic_sub_crq_queue *scrq)
3581 {
3582 	struct device *dev = &adapter->vdev->dev;
3583 	struct ibmvnic_tx_pool *tx_pool;
3584 	struct ibmvnic_tx_buff *txbuff;
3585 	struct netdev_queue *txq;
3586 	union sub_crq *next;
3587 	int index;
3588 	int i;
3589 
3590 restart_loop:
3591 	while (pending_scrq(adapter, scrq)) {
3592 		unsigned int pool = scrq->pool_index;
3593 		int num_entries = 0;
3594 		int total_bytes = 0;
3595 		int num_packets = 0;
3596 
3597 		next = ibmvnic_next_scrq(adapter, scrq);
3598 		for (i = 0; i < next->tx_comp.num_comps; i++) {
3599 			index = be32_to_cpu(next->tx_comp.correlators[i]);
3600 			if (index & IBMVNIC_TSO_POOL_MASK) {
3601 				tx_pool = &adapter->tso_pool[pool];
3602 				index &= ~IBMVNIC_TSO_POOL_MASK;
3603 			} else {
3604 				tx_pool = &adapter->tx_pool[pool];
3605 			}
3606 
3607 			txbuff = &tx_pool->tx_buff[index];
3608 			num_packets++;
3609 			num_entries += txbuff->num_entries;
3610 			if (txbuff->skb) {
3611 				total_bytes += txbuff->skb->len;
3612 				if (next->tx_comp.rcs[i]) {
3613 					dev_err(dev, "tx error %x\n",
3614 						next->tx_comp.rcs[i]);
3615 					dev_kfree_skb_irq(txbuff->skb);
3616 				} else {
3617 					dev_consume_skb_irq(txbuff->skb);
3618 				}
3619 				txbuff->skb = NULL;
3620 			} else {
3621 				netdev_warn(adapter->netdev,
3622 					    "TX completion received with NULL socket buffer\n");
3623 			}
3624 			tx_pool->free_map[tx_pool->producer_index] = index;
3625 			tx_pool->producer_index =
3626 				(tx_pool->producer_index + 1) %
3627 					tx_pool->num_buffers;
3628 		}
3629 		/* remove tx_comp scrq*/
3630 		next->tx_comp.first = 0;
3631 
3632 		txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3633 		netdev_tx_completed_queue(txq, num_packets, total_bytes);
3634 
3635 		if (atomic_sub_return(num_entries, &scrq->used) <=
3636 		    (adapter->req_tx_entries_per_subcrq / 2) &&
3637 		    __netif_subqueue_stopped(adapter->netdev,
3638 					     scrq->pool_index)) {
3639 			netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3640 			netdev_dbg(adapter->netdev, "Started queue %d\n",
3641 				   scrq->pool_index);
3642 		}
3643 	}
3644 
3645 	enable_scrq_irq(adapter, scrq);
3646 
3647 	if (pending_scrq(adapter, scrq)) {
3648 		disable_scrq_irq(adapter, scrq);
3649 		goto restart_loop;
3650 	}
3651 
3652 	return 0;
3653 }
3654 
3655 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3656 {
3657 	struct ibmvnic_sub_crq_queue *scrq = instance;
3658 	struct ibmvnic_adapter *adapter = scrq->adapter;
3659 
3660 	disable_scrq_irq(adapter, scrq);
3661 	ibmvnic_complete_tx(adapter, scrq);
3662 
3663 	return IRQ_HANDLED;
3664 }
3665 
3666 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3667 {
3668 	struct ibmvnic_sub_crq_queue *scrq = instance;
3669 	struct ibmvnic_adapter *adapter = scrq->adapter;
3670 
3671 	/* When booting a kdump kernel we can hit pending interrupts
3672 	 * prior to completing driver initialization.
3673 	 */
3674 	if (unlikely(adapter->state != VNIC_OPEN))
3675 		return IRQ_NONE;
3676 
3677 	adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3678 
3679 	if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3680 		disable_scrq_irq(adapter, scrq);
3681 		__napi_schedule(&adapter->napi[scrq->scrq_num]);
3682 	}
3683 
3684 	return IRQ_HANDLED;
3685 }
3686 
3687 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3688 {
3689 	struct device *dev = &adapter->vdev->dev;
3690 	struct ibmvnic_sub_crq_queue *scrq;
3691 	int i = 0, j = 0;
3692 	int rc = 0;
3693 
3694 	for (i = 0; i < adapter->req_tx_queues; i++) {
3695 		netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3696 			   i);
3697 		scrq = adapter->tx_scrq[i];
3698 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3699 
3700 		if (!scrq->irq) {
3701 			rc = -EINVAL;
3702 			dev_err(dev, "Error mapping irq\n");
3703 			goto req_tx_irq_failed;
3704 		}
3705 
3706 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3707 			 adapter->vdev->unit_address, i);
3708 		rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3709 				 0, scrq->name, scrq);
3710 
3711 		if (rc) {
3712 			dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3713 				scrq->irq, rc);
3714 			irq_dispose_mapping(scrq->irq);
3715 			goto req_tx_irq_failed;
3716 		}
3717 	}
3718 
3719 	for (i = 0; i < adapter->req_rx_queues; i++) {
3720 		netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3721 			   i);
3722 		scrq = adapter->rx_scrq[i];
3723 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3724 		if (!scrq->irq) {
3725 			rc = -EINVAL;
3726 			dev_err(dev, "Error mapping irq\n");
3727 			goto req_rx_irq_failed;
3728 		}
3729 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3730 			 adapter->vdev->unit_address, i);
3731 		rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3732 				 0, scrq->name, scrq);
3733 		if (rc) {
3734 			dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3735 				scrq->irq, rc);
3736 			irq_dispose_mapping(scrq->irq);
3737 			goto req_rx_irq_failed;
3738 		}
3739 	}
3740 	return rc;
3741 
3742 req_rx_irq_failed:
3743 	for (j = 0; j < i; j++) {
3744 		free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3745 		irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3746 	}
3747 	i = adapter->req_tx_queues;
3748 req_tx_irq_failed:
3749 	for (j = 0; j < i; j++) {
3750 		free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3751 		irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3752 	}
3753 	release_sub_crqs(adapter, 1);
3754 	return rc;
3755 }
3756 
3757 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3758 {
3759 	struct device *dev = &adapter->vdev->dev;
3760 	struct ibmvnic_sub_crq_queue **allqueues;
3761 	int registered_queues = 0;
3762 	int total_queues;
3763 	int more = 0;
3764 	int i;
3765 
3766 	total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3767 
3768 	allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3769 	if (!allqueues)
3770 		return -ENOMEM;
3771 
3772 	for (i = 0; i < total_queues; i++) {
3773 		allqueues[i] = init_sub_crq_queue(adapter);
3774 		if (!allqueues[i]) {
3775 			dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3776 			break;
3777 		}
3778 		registered_queues++;
3779 	}
3780 
3781 	/* Make sure we were able to register the minimum number of queues */
3782 	if (registered_queues <
3783 	    adapter->min_tx_queues + adapter->min_rx_queues) {
3784 		dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3785 		goto tx_failed;
3786 	}
3787 
3788 	/* Distribute the failed allocated queues*/
3789 	for (i = 0; i < total_queues - registered_queues + more ; i++) {
3790 		netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3791 		switch (i % 3) {
3792 		case 0:
3793 			if (adapter->req_rx_queues > adapter->min_rx_queues)
3794 				adapter->req_rx_queues--;
3795 			else
3796 				more++;
3797 			break;
3798 		case 1:
3799 			if (adapter->req_tx_queues > adapter->min_tx_queues)
3800 				adapter->req_tx_queues--;
3801 			else
3802 				more++;
3803 			break;
3804 		}
3805 	}
3806 
3807 	adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3808 				   sizeof(*adapter->tx_scrq), GFP_KERNEL);
3809 	if (!adapter->tx_scrq)
3810 		goto tx_failed;
3811 
3812 	for (i = 0; i < adapter->req_tx_queues; i++) {
3813 		adapter->tx_scrq[i] = allqueues[i];
3814 		adapter->tx_scrq[i]->pool_index = i;
3815 		adapter->num_active_tx_scrqs++;
3816 	}
3817 
3818 	adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3819 				   sizeof(*adapter->rx_scrq), GFP_KERNEL);
3820 	if (!adapter->rx_scrq)
3821 		goto rx_failed;
3822 
3823 	for (i = 0; i < adapter->req_rx_queues; i++) {
3824 		adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3825 		adapter->rx_scrq[i]->scrq_num = i;
3826 		adapter->num_active_rx_scrqs++;
3827 	}
3828 
3829 	kfree(allqueues);
3830 	return 0;
3831 
3832 rx_failed:
3833 	kfree(adapter->tx_scrq);
3834 	adapter->tx_scrq = NULL;
3835 tx_failed:
3836 	for (i = 0; i < registered_queues; i++)
3837 		release_sub_crq_queue(adapter, allqueues[i], 1);
3838 	kfree(allqueues);
3839 	return -ENOMEM;
3840 }
3841 
3842 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3843 {
3844 	struct device *dev = &adapter->vdev->dev;
3845 	union ibmvnic_crq crq;
3846 	int max_entries;
3847 
3848 	if (!retry) {
3849 		/* Sub-CRQ entries are 32 byte long */
3850 		int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3851 
3852 		if (adapter->min_tx_entries_per_subcrq > entries_page ||
3853 		    adapter->min_rx_add_entries_per_subcrq > entries_page) {
3854 			dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3855 			return;
3856 		}
3857 
3858 		if (adapter->desired.mtu)
3859 			adapter->req_mtu = adapter->desired.mtu;
3860 		else
3861 			adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3862 
3863 		if (!adapter->desired.tx_entries)
3864 			adapter->desired.tx_entries =
3865 					adapter->max_tx_entries_per_subcrq;
3866 		if (!adapter->desired.rx_entries)
3867 			adapter->desired.rx_entries =
3868 					adapter->max_rx_add_entries_per_subcrq;
3869 
3870 		max_entries = IBMVNIC_MAX_LTB_SIZE /
3871 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3872 
3873 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3874 			adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3875 			adapter->desired.tx_entries = max_entries;
3876 		}
3877 
3878 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3879 			adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3880 			adapter->desired.rx_entries = max_entries;
3881 		}
3882 
3883 		if (adapter->desired.tx_entries)
3884 			adapter->req_tx_entries_per_subcrq =
3885 					adapter->desired.tx_entries;
3886 		else
3887 			adapter->req_tx_entries_per_subcrq =
3888 					adapter->max_tx_entries_per_subcrq;
3889 
3890 		if (adapter->desired.rx_entries)
3891 			adapter->req_rx_add_entries_per_subcrq =
3892 					adapter->desired.rx_entries;
3893 		else
3894 			adapter->req_rx_add_entries_per_subcrq =
3895 					adapter->max_rx_add_entries_per_subcrq;
3896 
3897 		if (adapter->desired.tx_queues)
3898 			adapter->req_tx_queues =
3899 					adapter->desired.tx_queues;
3900 		else
3901 			adapter->req_tx_queues =
3902 					adapter->opt_tx_comp_sub_queues;
3903 
3904 		if (adapter->desired.rx_queues)
3905 			adapter->req_rx_queues =
3906 					adapter->desired.rx_queues;
3907 		else
3908 			adapter->req_rx_queues =
3909 					adapter->opt_rx_comp_queues;
3910 
3911 		adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3912 	}
3913 
3914 	memset(&crq, 0, sizeof(crq));
3915 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
3916 	crq.request_capability.cmd = REQUEST_CAPABILITY;
3917 
3918 	crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3919 	crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3920 	atomic_inc(&adapter->running_cap_crqs);
3921 	ibmvnic_send_crq(adapter, &crq);
3922 
3923 	crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3924 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3925 	atomic_inc(&adapter->running_cap_crqs);
3926 	ibmvnic_send_crq(adapter, &crq);
3927 
3928 	crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3929 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3930 	atomic_inc(&adapter->running_cap_crqs);
3931 	ibmvnic_send_crq(adapter, &crq);
3932 
3933 	crq.request_capability.capability =
3934 	    cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3935 	crq.request_capability.number =
3936 	    cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3937 	atomic_inc(&adapter->running_cap_crqs);
3938 	ibmvnic_send_crq(adapter, &crq);
3939 
3940 	crq.request_capability.capability =
3941 	    cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3942 	crq.request_capability.number =
3943 	    cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3944 	atomic_inc(&adapter->running_cap_crqs);
3945 	ibmvnic_send_crq(adapter, &crq);
3946 
3947 	crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3948 	crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3949 	atomic_inc(&adapter->running_cap_crqs);
3950 	ibmvnic_send_crq(adapter, &crq);
3951 
3952 	if (adapter->netdev->flags & IFF_PROMISC) {
3953 		if (adapter->promisc_supported) {
3954 			crq.request_capability.capability =
3955 			    cpu_to_be16(PROMISC_REQUESTED);
3956 			crq.request_capability.number = cpu_to_be64(1);
3957 			atomic_inc(&adapter->running_cap_crqs);
3958 			ibmvnic_send_crq(adapter, &crq);
3959 		}
3960 	} else {
3961 		crq.request_capability.capability =
3962 		    cpu_to_be16(PROMISC_REQUESTED);
3963 		crq.request_capability.number = cpu_to_be64(0);
3964 		atomic_inc(&adapter->running_cap_crqs);
3965 		ibmvnic_send_crq(adapter, &crq);
3966 	}
3967 }
3968 
3969 static int pending_scrq(struct ibmvnic_adapter *adapter,
3970 			struct ibmvnic_sub_crq_queue *scrq)
3971 {
3972 	union sub_crq *entry = &scrq->msgs[scrq->cur];
3973 	int rc;
3974 
3975 	rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3976 
3977 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
3978 	 * contents of the SCRQ descriptor
3979 	 */
3980 	dma_rmb();
3981 
3982 	return rc;
3983 }
3984 
3985 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3986 					struct ibmvnic_sub_crq_queue *scrq)
3987 {
3988 	union sub_crq *entry;
3989 	unsigned long flags;
3990 
3991 	spin_lock_irqsave(&scrq->lock, flags);
3992 	entry = &scrq->msgs[scrq->cur];
3993 	if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3994 		if (++scrq->cur == scrq->size)
3995 			scrq->cur = 0;
3996 	} else {
3997 		entry = NULL;
3998 	}
3999 	spin_unlock_irqrestore(&scrq->lock, flags);
4000 
4001 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4002 	 * contents of the SCRQ descriptor
4003 	 */
4004 	dma_rmb();
4005 
4006 	return entry;
4007 }
4008 
4009 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4010 {
4011 	struct ibmvnic_crq_queue *queue = &adapter->crq;
4012 	union ibmvnic_crq *crq;
4013 
4014 	crq = &queue->msgs[queue->cur];
4015 	if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4016 		if (++queue->cur == queue->size)
4017 			queue->cur = 0;
4018 	} else {
4019 		crq = NULL;
4020 	}
4021 
4022 	return crq;
4023 }
4024 
4025 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4026 {
4027 	switch (rc) {
4028 	case H_PARAMETER:
4029 		dev_warn_ratelimited(dev,
4030 				     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4031 				     func, rc);
4032 		break;
4033 	case H_CLOSED:
4034 		dev_warn_ratelimited(dev,
4035 				     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4036 				     func, rc);
4037 		break;
4038 	default:
4039 		dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4040 		break;
4041 	}
4042 }
4043 
4044 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4045 				u64 remote_handle, u64 ioba, u64 num_entries)
4046 {
4047 	unsigned int ua = adapter->vdev->unit_address;
4048 	struct device *dev = &adapter->vdev->dev;
4049 	int rc;
4050 
4051 	/* Make sure the hypervisor sees the complete request */
4052 	dma_wmb();
4053 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4054 				cpu_to_be64(remote_handle),
4055 				ioba, num_entries);
4056 
4057 	if (rc)
4058 		print_subcrq_error(dev, rc, __func__);
4059 
4060 	return rc;
4061 }
4062 
4063 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4064 			    union ibmvnic_crq *crq)
4065 {
4066 	unsigned int ua = adapter->vdev->unit_address;
4067 	struct device *dev = &adapter->vdev->dev;
4068 	u64 *u64_crq = (u64 *)crq;
4069 	int rc;
4070 
4071 	netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4072 		   (unsigned long)cpu_to_be64(u64_crq[0]),
4073 		   (unsigned long)cpu_to_be64(u64_crq[1]));
4074 
4075 	if (!adapter->crq.active &&
4076 	    crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4077 		dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4078 		return -EINVAL;
4079 	}
4080 
4081 	/* Make sure the hypervisor sees the complete request */
4082 	dma_wmb();
4083 
4084 	rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4085 				cpu_to_be64(u64_crq[0]),
4086 				cpu_to_be64(u64_crq[1]));
4087 
4088 	if (rc) {
4089 		if (rc == H_CLOSED) {
4090 			dev_warn(dev, "CRQ Queue closed\n");
4091 			/* do not reset, report the fail, wait for passive init from server */
4092 		}
4093 
4094 		dev_warn(dev, "Send error (rc=%d)\n", rc);
4095 	}
4096 
4097 	return rc;
4098 }
4099 
4100 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4101 {
4102 	struct device *dev = &adapter->vdev->dev;
4103 	union ibmvnic_crq crq;
4104 	int retries = 100;
4105 	int rc;
4106 
4107 	memset(&crq, 0, sizeof(crq));
4108 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4109 	crq.generic.cmd = IBMVNIC_CRQ_INIT;
4110 	netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4111 
4112 	do {
4113 		rc = ibmvnic_send_crq(adapter, &crq);
4114 		if (rc != H_CLOSED)
4115 			break;
4116 		retries--;
4117 		msleep(50);
4118 
4119 	} while (retries > 0);
4120 
4121 	if (rc) {
4122 		dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4123 		return rc;
4124 	}
4125 
4126 	return 0;
4127 }
4128 
4129 struct vnic_login_client_data {
4130 	u8	type;
4131 	__be16	len;
4132 	char	name[];
4133 } __packed;
4134 
4135 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4136 {
4137 	int len;
4138 
4139 	/* Calculate the amount of buffer space needed for the
4140 	 * vnic client data in the login buffer. There are four entries,
4141 	 * OS name, LPAR name, device name, and a null last entry.
4142 	 */
4143 	len = 4 * sizeof(struct vnic_login_client_data);
4144 	len += 6; /* "Linux" plus NULL */
4145 	len += strlen(utsname()->nodename) + 1;
4146 	len += strlen(adapter->netdev->name) + 1;
4147 
4148 	return len;
4149 }
4150 
4151 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4152 				 struct vnic_login_client_data *vlcd)
4153 {
4154 	const char *os_name = "Linux";
4155 	int len;
4156 
4157 	/* Type 1 - LPAR OS */
4158 	vlcd->type = 1;
4159 	len = strlen(os_name) + 1;
4160 	vlcd->len = cpu_to_be16(len);
4161 	strscpy(vlcd->name, os_name, len);
4162 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4163 
4164 	/* Type 2 - LPAR name */
4165 	vlcd->type = 2;
4166 	len = strlen(utsname()->nodename) + 1;
4167 	vlcd->len = cpu_to_be16(len);
4168 	strscpy(vlcd->name, utsname()->nodename, len);
4169 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4170 
4171 	/* Type 3 - device name */
4172 	vlcd->type = 3;
4173 	len = strlen(adapter->netdev->name) + 1;
4174 	vlcd->len = cpu_to_be16(len);
4175 	strscpy(vlcd->name, adapter->netdev->name, len);
4176 }
4177 
4178 static int send_login(struct ibmvnic_adapter *adapter)
4179 {
4180 	struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4181 	struct ibmvnic_login_buffer *login_buffer;
4182 	struct device *dev = &adapter->vdev->dev;
4183 	struct vnic_login_client_data *vlcd;
4184 	dma_addr_t rsp_buffer_token;
4185 	dma_addr_t buffer_token;
4186 	size_t rsp_buffer_size;
4187 	union ibmvnic_crq crq;
4188 	int client_data_len;
4189 	size_t buffer_size;
4190 	__be64 *tx_list_p;
4191 	__be64 *rx_list_p;
4192 	int rc;
4193 	int i;
4194 
4195 	if (!adapter->tx_scrq || !adapter->rx_scrq) {
4196 		netdev_err(adapter->netdev,
4197 			   "RX or TX queues are not allocated, device login failed\n");
4198 		return -ENOMEM;
4199 	}
4200 
4201 	release_login_buffer(adapter);
4202 	release_login_rsp_buffer(adapter);
4203 
4204 	client_data_len = vnic_client_data_len(adapter);
4205 
4206 	buffer_size =
4207 	    sizeof(struct ibmvnic_login_buffer) +
4208 	    sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4209 	    client_data_len;
4210 
4211 	login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4212 	if (!login_buffer)
4213 		goto buf_alloc_failed;
4214 
4215 	buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4216 				      DMA_TO_DEVICE);
4217 	if (dma_mapping_error(dev, buffer_token)) {
4218 		dev_err(dev, "Couldn't map login buffer\n");
4219 		goto buf_map_failed;
4220 	}
4221 
4222 	rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4223 			  sizeof(u64) * adapter->req_tx_queues +
4224 			  sizeof(u64) * adapter->req_rx_queues +
4225 			  sizeof(u64) * adapter->req_rx_queues +
4226 			  sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4227 
4228 	login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4229 	if (!login_rsp_buffer)
4230 		goto buf_rsp_alloc_failed;
4231 
4232 	rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4233 					  rsp_buffer_size, DMA_FROM_DEVICE);
4234 	if (dma_mapping_error(dev, rsp_buffer_token)) {
4235 		dev_err(dev, "Couldn't map login rsp buffer\n");
4236 		goto buf_rsp_map_failed;
4237 	}
4238 
4239 	adapter->login_buf = login_buffer;
4240 	adapter->login_buf_token = buffer_token;
4241 	adapter->login_buf_sz = buffer_size;
4242 	adapter->login_rsp_buf = login_rsp_buffer;
4243 	adapter->login_rsp_buf_token = rsp_buffer_token;
4244 	adapter->login_rsp_buf_sz = rsp_buffer_size;
4245 
4246 	login_buffer->len = cpu_to_be32(buffer_size);
4247 	login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4248 	login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4249 	login_buffer->off_txcomp_subcrqs =
4250 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4251 	login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4252 	login_buffer->off_rxcomp_subcrqs =
4253 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4254 			sizeof(u64) * adapter->req_tx_queues);
4255 	login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4256 	login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4257 
4258 	tx_list_p = (__be64 *)((char *)login_buffer +
4259 				      sizeof(struct ibmvnic_login_buffer));
4260 	rx_list_p = (__be64 *)((char *)login_buffer +
4261 				      sizeof(struct ibmvnic_login_buffer) +
4262 				      sizeof(u64) * adapter->req_tx_queues);
4263 
4264 	for (i = 0; i < adapter->req_tx_queues; i++) {
4265 		if (adapter->tx_scrq[i]) {
4266 			tx_list_p[i] =
4267 				cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4268 		}
4269 	}
4270 
4271 	for (i = 0; i < adapter->req_rx_queues; i++) {
4272 		if (adapter->rx_scrq[i]) {
4273 			rx_list_p[i] =
4274 				cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4275 		}
4276 	}
4277 
4278 	/* Insert vNIC login client data */
4279 	vlcd = (struct vnic_login_client_data *)
4280 		((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4281 	login_buffer->client_data_offset =
4282 			cpu_to_be32((char *)vlcd - (char *)login_buffer);
4283 	login_buffer->client_data_len = cpu_to_be32(client_data_len);
4284 
4285 	vnic_add_client_data(adapter, vlcd);
4286 
4287 	netdev_dbg(adapter->netdev, "Login Buffer:\n");
4288 	for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4289 		netdev_dbg(adapter->netdev, "%016lx\n",
4290 			   ((unsigned long *)(adapter->login_buf))[i]);
4291 	}
4292 
4293 	memset(&crq, 0, sizeof(crq));
4294 	crq.login.first = IBMVNIC_CRQ_CMD;
4295 	crq.login.cmd = LOGIN;
4296 	crq.login.ioba = cpu_to_be32(buffer_token);
4297 	crq.login.len = cpu_to_be32(buffer_size);
4298 
4299 	adapter->login_pending = true;
4300 	rc = ibmvnic_send_crq(adapter, &crq);
4301 	if (rc) {
4302 		adapter->login_pending = false;
4303 		netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4304 		goto buf_rsp_map_failed;
4305 	}
4306 
4307 	return 0;
4308 
4309 buf_rsp_map_failed:
4310 	kfree(login_rsp_buffer);
4311 	adapter->login_rsp_buf = NULL;
4312 buf_rsp_alloc_failed:
4313 	dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4314 buf_map_failed:
4315 	kfree(login_buffer);
4316 	adapter->login_buf = NULL;
4317 buf_alloc_failed:
4318 	return -ENOMEM;
4319 }
4320 
4321 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4322 			    u32 len, u8 map_id)
4323 {
4324 	union ibmvnic_crq crq;
4325 
4326 	memset(&crq, 0, sizeof(crq));
4327 	crq.request_map.first = IBMVNIC_CRQ_CMD;
4328 	crq.request_map.cmd = REQUEST_MAP;
4329 	crq.request_map.map_id = map_id;
4330 	crq.request_map.ioba = cpu_to_be32(addr);
4331 	crq.request_map.len = cpu_to_be32(len);
4332 	return ibmvnic_send_crq(adapter, &crq);
4333 }
4334 
4335 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4336 {
4337 	union ibmvnic_crq crq;
4338 
4339 	memset(&crq, 0, sizeof(crq));
4340 	crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4341 	crq.request_unmap.cmd = REQUEST_UNMAP;
4342 	crq.request_unmap.map_id = map_id;
4343 	return ibmvnic_send_crq(adapter, &crq);
4344 }
4345 
4346 static void send_query_map(struct ibmvnic_adapter *adapter)
4347 {
4348 	union ibmvnic_crq crq;
4349 
4350 	memset(&crq, 0, sizeof(crq));
4351 	crq.query_map.first = IBMVNIC_CRQ_CMD;
4352 	crq.query_map.cmd = QUERY_MAP;
4353 	ibmvnic_send_crq(adapter, &crq);
4354 }
4355 
4356 /* Send a series of CRQs requesting various capabilities of the VNIC server */
4357 static void send_query_cap(struct ibmvnic_adapter *adapter)
4358 {
4359 	union ibmvnic_crq crq;
4360 
4361 	atomic_set(&adapter->running_cap_crqs, 0);
4362 	memset(&crq, 0, sizeof(crq));
4363 	crq.query_capability.first = IBMVNIC_CRQ_CMD;
4364 	crq.query_capability.cmd = QUERY_CAPABILITY;
4365 
4366 	crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4367 	atomic_inc(&adapter->running_cap_crqs);
4368 	ibmvnic_send_crq(adapter, &crq);
4369 
4370 	crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4371 	atomic_inc(&adapter->running_cap_crqs);
4372 	ibmvnic_send_crq(adapter, &crq);
4373 
4374 	crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4375 	atomic_inc(&adapter->running_cap_crqs);
4376 	ibmvnic_send_crq(adapter, &crq);
4377 
4378 	crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4379 	atomic_inc(&adapter->running_cap_crqs);
4380 	ibmvnic_send_crq(adapter, &crq);
4381 
4382 	crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4383 	atomic_inc(&adapter->running_cap_crqs);
4384 	ibmvnic_send_crq(adapter, &crq);
4385 
4386 	crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4387 	atomic_inc(&adapter->running_cap_crqs);
4388 	ibmvnic_send_crq(adapter, &crq);
4389 
4390 	crq.query_capability.capability =
4391 	    cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4392 	atomic_inc(&adapter->running_cap_crqs);
4393 	ibmvnic_send_crq(adapter, &crq);
4394 
4395 	crq.query_capability.capability =
4396 	    cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4397 	atomic_inc(&adapter->running_cap_crqs);
4398 	ibmvnic_send_crq(adapter, &crq);
4399 
4400 	crq.query_capability.capability =
4401 	    cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4402 	atomic_inc(&adapter->running_cap_crqs);
4403 	ibmvnic_send_crq(adapter, &crq);
4404 
4405 	crq.query_capability.capability =
4406 	    cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4407 	atomic_inc(&adapter->running_cap_crqs);
4408 	ibmvnic_send_crq(adapter, &crq);
4409 
4410 	crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4411 	atomic_inc(&adapter->running_cap_crqs);
4412 	ibmvnic_send_crq(adapter, &crq);
4413 
4414 	crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4415 	atomic_inc(&adapter->running_cap_crqs);
4416 	ibmvnic_send_crq(adapter, &crq);
4417 
4418 	crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4419 	atomic_inc(&adapter->running_cap_crqs);
4420 	ibmvnic_send_crq(adapter, &crq);
4421 
4422 	crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4423 	atomic_inc(&adapter->running_cap_crqs);
4424 	ibmvnic_send_crq(adapter, &crq);
4425 
4426 	crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4427 	atomic_inc(&adapter->running_cap_crqs);
4428 	ibmvnic_send_crq(adapter, &crq);
4429 
4430 	crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4431 	atomic_inc(&adapter->running_cap_crqs);
4432 	ibmvnic_send_crq(adapter, &crq);
4433 
4434 	crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4435 	atomic_inc(&adapter->running_cap_crqs);
4436 	ibmvnic_send_crq(adapter, &crq);
4437 
4438 	crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4439 	atomic_inc(&adapter->running_cap_crqs);
4440 	ibmvnic_send_crq(adapter, &crq);
4441 
4442 	crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4443 	atomic_inc(&adapter->running_cap_crqs);
4444 	ibmvnic_send_crq(adapter, &crq);
4445 
4446 	crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4447 	atomic_inc(&adapter->running_cap_crqs);
4448 	ibmvnic_send_crq(adapter, &crq);
4449 
4450 	crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4451 	atomic_inc(&adapter->running_cap_crqs);
4452 	ibmvnic_send_crq(adapter, &crq);
4453 
4454 	crq.query_capability.capability =
4455 			cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4456 	atomic_inc(&adapter->running_cap_crqs);
4457 	ibmvnic_send_crq(adapter, &crq);
4458 
4459 	crq.query_capability.capability =
4460 			cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4461 	atomic_inc(&adapter->running_cap_crqs);
4462 	ibmvnic_send_crq(adapter, &crq);
4463 
4464 	crq.query_capability.capability =
4465 			cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4466 	atomic_inc(&adapter->running_cap_crqs);
4467 	ibmvnic_send_crq(adapter, &crq);
4468 
4469 	crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4470 	atomic_inc(&adapter->running_cap_crqs);
4471 	ibmvnic_send_crq(adapter, &crq);
4472 }
4473 
4474 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4475 {
4476 	int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4477 	struct device *dev = &adapter->vdev->dev;
4478 	union ibmvnic_crq crq;
4479 
4480 	adapter->ip_offload_tok =
4481 		dma_map_single(dev,
4482 			       &adapter->ip_offload_buf,
4483 			       buf_sz,
4484 			       DMA_FROM_DEVICE);
4485 
4486 	if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4487 		if (!firmware_has_feature(FW_FEATURE_CMO))
4488 			dev_err(dev, "Couldn't map offload buffer\n");
4489 		return;
4490 	}
4491 
4492 	memset(&crq, 0, sizeof(crq));
4493 	crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4494 	crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4495 	crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4496 	crq.query_ip_offload.ioba =
4497 	    cpu_to_be32(adapter->ip_offload_tok);
4498 
4499 	ibmvnic_send_crq(adapter, &crq);
4500 }
4501 
4502 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4503 {
4504 	struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4505 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4506 	struct device *dev = &adapter->vdev->dev;
4507 	netdev_features_t old_hw_features = 0;
4508 	union ibmvnic_crq crq;
4509 
4510 	adapter->ip_offload_ctrl_tok =
4511 		dma_map_single(dev,
4512 			       ctrl_buf,
4513 			       sizeof(adapter->ip_offload_ctrl),
4514 			       DMA_TO_DEVICE);
4515 
4516 	if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4517 		dev_err(dev, "Couldn't map ip offload control buffer\n");
4518 		return;
4519 	}
4520 
4521 	ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4522 	ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4523 	ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4524 	ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4525 	ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4526 	ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4527 	ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4528 	ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4529 	ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4530 	ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4531 
4532 	/* large_rx disabled for now, additional features needed */
4533 	ctrl_buf->large_rx_ipv4 = 0;
4534 	ctrl_buf->large_rx_ipv6 = 0;
4535 
4536 	if (adapter->state != VNIC_PROBING) {
4537 		old_hw_features = adapter->netdev->hw_features;
4538 		adapter->netdev->hw_features = 0;
4539 	}
4540 
4541 	adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4542 
4543 	if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4544 		adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4545 
4546 	if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4547 		adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4548 
4549 	if ((adapter->netdev->features &
4550 	    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4551 		adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4552 
4553 	if (buf->large_tx_ipv4)
4554 		adapter->netdev->hw_features |= NETIF_F_TSO;
4555 	if (buf->large_tx_ipv6)
4556 		adapter->netdev->hw_features |= NETIF_F_TSO6;
4557 
4558 	if (adapter->state == VNIC_PROBING) {
4559 		adapter->netdev->features |= adapter->netdev->hw_features;
4560 	} else if (old_hw_features != adapter->netdev->hw_features) {
4561 		netdev_features_t tmp = 0;
4562 
4563 		/* disable features no longer supported */
4564 		adapter->netdev->features &= adapter->netdev->hw_features;
4565 		/* turn on features now supported if previously enabled */
4566 		tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4567 			adapter->netdev->hw_features;
4568 		adapter->netdev->features |=
4569 				tmp & adapter->netdev->wanted_features;
4570 	}
4571 
4572 	memset(&crq, 0, sizeof(crq));
4573 	crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4574 	crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4575 	crq.control_ip_offload.len =
4576 	    cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4577 	crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4578 	ibmvnic_send_crq(adapter, &crq);
4579 }
4580 
4581 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4582 				struct ibmvnic_adapter *adapter)
4583 {
4584 	struct device *dev = &adapter->vdev->dev;
4585 
4586 	if (crq->get_vpd_size_rsp.rc.code) {
4587 		dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4588 			crq->get_vpd_size_rsp.rc.code);
4589 		complete(&adapter->fw_done);
4590 		return;
4591 	}
4592 
4593 	adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4594 	complete(&adapter->fw_done);
4595 }
4596 
4597 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4598 			   struct ibmvnic_adapter *adapter)
4599 {
4600 	struct device *dev = &adapter->vdev->dev;
4601 	unsigned char *substr = NULL;
4602 	u8 fw_level_len = 0;
4603 
4604 	memset(adapter->fw_version, 0, 32);
4605 
4606 	dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4607 			 DMA_FROM_DEVICE);
4608 
4609 	if (crq->get_vpd_rsp.rc.code) {
4610 		dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4611 			crq->get_vpd_rsp.rc.code);
4612 		goto complete;
4613 	}
4614 
4615 	/* get the position of the firmware version info
4616 	 * located after the ASCII 'RM' substring in the buffer
4617 	 */
4618 	substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4619 	if (!substr) {
4620 		dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4621 		goto complete;
4622 	}
4623 
4624 	/* get length of firmware level ASCII substring */
4625 	if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4626 		fw_level_len = *(substr + 2);
4627 	} else {
4628 		dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4629 		goto complete;
4630 	}
4631 
4632 	/* copy firmware version string from vpd into adapter */
4633 	if ((substr + 3 + fw_level_len) <
4634 	    (adapter->vpd->buff + adapter->vpd->len)) {
4635 		strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4636 	} else {
4637 		dev_info(dev, "FW substr extrapolated VPD buff\n");
4638 	}
4639 
4640 complete:
4641 	if (adapter->fw_version[0] == '\0')
4642 		strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
4643 	complete(&adapter->fw_done);
4644 }
4645 
4646 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4647 {
4648 	struct device *dev = &adapter->vdev->dev;
4649 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4650 	int i;
4651 
4652 	dma_unmap_single(dev, adapter->ip_offload_tok,
4653 			 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4654 
4655 	netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4656 	for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4657 		netdev_dbg(adapter->netdev, "%016lx\n",
4658 			   ((unsigned long *)(buf))[i]);
4659 
4660 	netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4661 	netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4662 	netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4663 		   buf->tcp_ipv4_chksum);
4664 	netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4665 		   buf->tcp_ipv6_chksum);
4666 	netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4667 		   buf->udp_ipv4_chksum);
4668 	netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4669 		   buf->udp_ipv6_chksum);
4670 	netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4671 		   buf->large_tx_ipv4);
4672 	netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4673 		   buf->large_tx_ipv6);
4674 	netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4675 		   buf->large_rx_ipv4);
4676 	netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4677 		   buf->large_rx_ipv6);
4678 	netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4679 		   buf->max_ipv4_header_size);
4680 	netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4681 		   buf->max_ipv6_header_size);
4682 	netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4683 		   buf->max_tcp_header_size);
4684 	netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4685 		   buf->max_udp_header_size);
4686 	netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4687 		   buf->max_large_tx_size);
4688 	netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4689 		   buf->max_large_rx_size);
4690 	netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4691 		   buf->ipv6_extension_header);
4692 	netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4693 		   buf->tcp_pseudosum_req);
4694 	netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4695 		   buf->num_ipv6_ext_headers);
4696 	netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4697 		   buf->off_ipv6_ext_headers);
4698 
4699 	send_control_ip_offload(adapter);
4700 }
4701 
4702 static const char *ibmvnic_fw_err_cause(u16 cause)
4703 {
4704 	switch (cause) {
4705 	case ADAPTER_PROBLEM:
4706 		return "adapter problem";
4707 	case BUS_PROBLEM:
4708 		return "bus problem";
4709 	case FW_PROBLEM:
4710 		return "firmware problem";
4711 	case DD_PROBLEM:
4712 		return "device driver problem";
4713 	case EEH_RECOVERY:
4714 		return "EEH recovery";
4715 	case FW_UPDATED:
4716 		return "firmware updated";
4717 	case LOW_MEMORY:
4718 		return "low Memory";
4719 	default:
4720 		return "unknown";
4721 	}
4722 }
4723 
4724 static void handle_error_indication(union ibmvnic_crq *crq,
4725 				    struct ibmvnic_adapter *adapter)
4726 {
4727 	struct device *dev = &adapter->vdev->dev;
4728 	u16 cause;
4729 
4730 	cause = be16_to_cpu(crq->error_indication.error_cause);
4731 
4732 	dev_warn_ratelimited(dev,
4733 			     "Firmware reports %serror, cause: %s. Starting recovery...\n",
4734 			     crq->error_indication.flags
4735 				& IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4736 			     ibmvnic_fw_err_cause(cause));
4737 
4738 	if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4739 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4740 	else
4741 		ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4742 }
4743 
4744 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4745 				 struct ibmvnic_adapter *adapter)
4746 {
4747 	struct net_device *netdev = adapter->netdev;
4748 	struct device *dev = &adapter->vdev->dev;
4749 	long rc;
4750 
4751 	rc = crq->change_mac_addr_rsp.rc.code;
4752 	if (rc) {
4753 		dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4754 		goto out;
4755 	}
4756 	/* crq->change_mac_addr.mac_addr is the requested one
4757 	 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4758 	 */
4759 	eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
4760 	ether_addr_copy(adapter->mac_addr,
4761 			&crq->change_mac_addr_rsp.mac_addr[0]);
4762 out:
4763 	complete(&adapter->fw_done);
4764 	return rc;
4765 }
4766 
4767 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4768 				   struct ibmvnic_adapter *adapter)
4769 {
4770 	struct device *dev = &adapter->vdev->dev;
4771 	u64 *req_value;
4772 	char *name;
4773 
4774 	atomic_dec(&adapter->running_cap_crqs);
4775 	switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4776 	case REQ_TX_QUEUES:
4777 		req_value = &adapter->req_tx_queues;
4778 		name = "tx";
4779 		break;
4780 	case REQ_RX_QUEUES:
4781 		req_value = &adapter->req_rx_queues;
4782 		name = "rx";
4783 		break;
4784 	case REQ_RX_ADD_QUEUES:
4785 		req_value = &adapter->req_rx_add_queues;
4786 		name = "rx_add";
4787 		break;
4788 	case REQ_TX_ENTRIES_PER_SUBCRQ:
4789 		req_value = &adapter->req_tx_entries_per_subcrq;
4790 		name = "tx_entries_per_subcrq";
4791 		break;
4792 	case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4793 		req_value = &adapter->req_rx_add_entries_per_subcrq;
4794 		name = "rx_add_entries_per_subcrq";
4795 		break;
4796 	case REQ_MTU:
4797 		req_value = &adapter->req_mtu;
4798 		name = "mtu";
4799 		break;
4800 	case PROMISC_REQUESTED:
4801 		req_value = &adapter->promisc;
4802 		name = "promisc";
4803 		break;
4804 	default:
4805 		dev_err(dev, "Got invalid cap request rsp %d\n",
4806 			crq->request_capability.capability);
4807 		return;
4808 	}
4809 
4810 	switch (crq->request_capability_rsp.rc.code) {
4811 	case SUCCESS:
4812 		break;
4813 	case PARTIALSUCCESS:
4814 		dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4815 			 *req_value,
4816 			 (long)be64_to_cpu(crq->request_capability_rsp.number),
4817 			 name);
4818 
4819 		if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4820 		    REQ_MTU) {
4821 			pr_err("mtu of %llu is not supported. Reverting.\n",
4822 			       *req_value);
4823 			*req_value = adapter->fallback.mtu;
4824 		} else {
4825 			*req_value =
4826 				be64_to_cpu(crq->request_capability_rsp.number);
4827 		}
4828 
4829 		send_request_cap(adapter, 1);
4830 		return;
4831 	default:
4832 		dev_err(dev, "Error %d in request cap rsp\n",
4833 			crq->request_capability_rsp.rc.code);
4834 		return;
4835 	}
4836 
4837 	/* Done receiving requested capabilities, query IP offload support */
4838 	if (atomic_read(&adapter->running_cap_crqs) == 0) {
4839 		adapter->wait_capability = false;
4840 		send_query_ip_offload(adapter);
4841 	}
4842 }
4843 
4844 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4845 			    struct ibmvnic_adapter *adapter)
4846 {
4847 	struct device *dev = &adapter->vdev->dev;
4848 	struct net_device *netdev = adapter->netdev;
4849 	struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4850 	struct ibmvnic_login_buffer *login = adapter->login_buf;
4851 	u64 *tx_handle_array;
4852 	u64 *rx_handle_array;
4853 	int num_tx_pools;
4854 	int num_rx_pools;
4855 	u64 *size_array;
4856 	int i;
4857 
4858 	/* CHECK: Test/set of login_pending does not need to be atomic
4859 	 * because only ibmvnic_tasklet tests/clears this.
4860 	 */
4861 	if (!adapter->login_pending) {
4862 		netdev_warn(netdev, "Ignoring unexpected login response\n");
4863 		return 0;
4864 	}
4865 	adapter->login_pending = false;
4866 
4867 	dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4868 			 DMA_TO_DEVICE);
4869 	dma_unmap_single(dev, adapter->login_rsp_buf_token,
4870 			 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4871 
4872 	/* If the number of queues requested can't be allocated by the
4873 	 * server, the login response will return with code 1. We will need
4874 	 * to resend the login buffer with fewer queues requested.
4875 	 */
4876 	if (login_rsp_crq->generic.rc.code) {
4877 		adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4878 		complete(&adapter->init_done);
4879 		return 0;
4880 	}
4881 
4882 	if (adapter->failover_pending) {
4883 		adapter->init_done_rc = -EAGAIN;
4884 		netdev_dbg(netdev, "Failover pending, ignoring login response\n");
4885 		complete(&adapter->init_done);
4886 		/* login response buffer will be released on reset */
4887 		return 0;
4888 	}
4889 
4890 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
4891 
4892 	netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4893 	for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4894 		netdev_dbg(adapter->netdev, "%016lx\n",
4895 			   ((unsigned long *)(adapter->login_rsp_buf))[i]);
4896 	}
4897 
4898 	/* Sanity checks */
4899 	if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4900 	    (be32_to_cpu(login->num_rxcomp_subcrqs) *
4901 	     adapter->req_rx_add_queues !=
4902 	     be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4903 		dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4904 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4905 		return -EIO;
4906 	}
4907 	size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4908 		be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4909 	/* variable buffer sizes are not supported, so just read the
4910 	 * first entry.
4911 	 */
4912 	adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4913 
4914 	num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4915 	num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4916 
4917 	tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4918 				  be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4919 	rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4920 				  be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4921 
4922 	for (i = 0; i < num_tx_pools; i++)
4923 		adapter->tx_scrq[i]->handle = tx_handle_array[i];
4924 
4925 	for (i = 0; i < num_rx_pools; i++)
4926 		adapter->rx_scrq[i]->handle = rx_handle_array[i];
4927 
4928 	adapter->num_active_tx_scrqs = num_tx_pools;
4929 	adapter->num_active_rx_scrqs = num_rx_pools;
4930 	release_login_rsp_buffer(adapter);
4931 	release_login_buffer(adapter);
4932 	complete(&adapter->init_done);
4933 
4934 	return 0;
4935 }
4936 
4937 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4938 				     struct ibmvnic_adapter *adapter)
4939 {
4940 	struct device *dev = &adapter->vdev->dev;
4941 	long rc;
4942 
4943 	rc = crq->request_unmap_rsp.rc.code;
4944 	if (rc)
4945 		dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4946 }
4947 
4948 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4949 				 struct ibmvnic_adapter *adapter)
4950 {
4951 	struct net_device *netdev = adapter->netdev;
4952 	struct device *dev = &adapter->vdev->dev;
4953 	long rc;
4954 
4955 	rc = crq->query_map_rsp.rc.code;
4956 	if (rc) {
4957 		dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4958 		return;
4959 	}
4960 	netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
4961 		   crq->query_map_rsp.page_size,
4962 		   __be32_to_cpu(crq->query_map_rsp.tot_pages),
4963 		   __be32_to_cpu(crq->query_map_rsp.free_pages));
4964 }
4965 
4966 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4967 				 struct ibmvnic_adapter *adapter)
4968 {
4969 	struct net_device *netdev = adapter->netdev;
4970 	struct device *dev = &adapter->vdev->dev;
4971 	long rc;
4972 
4973 	atomic_dec(&adapter->running_cap_crqs);
4974 	netdev_dbg(netdev, "Outstanding queries: %d\n",
4975 		   atomic_read(&adapter->running_cap_crqs));
4976 	rc = crq->query_capability.rc.code;
4977 	if (rc) {
4978 		dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4979 		goto out;
4980 	}
4981 
4982 	switch (be16_to_cpu(crq->query_capability.capability)) {
4983 	case MIN_TX_QUEUES:
4984 		adapter->min_tx_queues =
4985 		    be64_to_cpu(crq->query_capability.number);
4986 		netdev_dbg(netdev, "min_tx_queues = %lld\n",
4987 			   adapter->min_tx_queues);
4988 		break;
4989 	case MIN_RX_QUEUES:
4990 		adapter->min_rx_queues =
4991 		    be64_to_cpu(crq->query_capability.number);
4992 		netdev_dbg(netdev, "min_rx_queues = %lld\n",
4993 			   adapter->min_rx_queues);
4994 		break;
4995 	case MIN_RX_ADD_QUEUES:
4996 		adapter->min_rx_add_queues =
4997 		    be64_to_cpu(crq->query_capability.number);
4998 		netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4999 			   adapter->min_rx_add_queues);
5000 		break;
5001 	case MAX_TX_QUEUES:
5002 		adapter->max_tx_queues =
5003 		    be64_to_cpu(crq->query_capability.number);
5004 		netdev_dbg(netdev, "max_tx_queues = %lld\n",
5005 			   adapter->max_tx_queues);
5006 		break;
5007 	case MAX_RX_QUEUES:
5008 		adapter->max_rx_queues =
5009 		    be64_to_cpu(crq->query_capability.number);
5010 		netdev_dbg(netdev, "max_rx_queues = %lld\n",
5011 			   adapter->max_rx_queues);
5012 		break;
5013 	case MAX_RX_ADD_QUEUES:
5014 		adapter->max_rx_add_queues =
5015 		    be64_to_cpu(crq->query_capability.number);
5016 		netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5017 			   adapter->max_rx_add_queues);
5018 		break;
5019 	case MIN_TX_ENTRIES_PER_SUBCRQ:
5020 		adapter->min_tx_entries_per_subcrq =
5021 		    be64_to_cpu(crq->query_capability.number);
5022 		netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5023 			   adapter->min_tx_entries_per_subcrq);
5024 		break;
5025 	case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5026 		adapter->min_rx_add_entries_per_subcrq =
5027 		    be64_to_cpu(crq->query_capability.number);
5028 		netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5029 			   adapter->min_rx_add_entries_per_subcrq);
5030 		break;
5031 	case MAX_TX_ENTRIES_PER_SUBCRQ:
5032 		adapter->max_tx_entries_per_subcrq =
5033 		    be64_to_cpu(crq->query_capability.number);
5034 		netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5035 			   adapter->max_tx_entries_per_subcrq);
5036 		break;
5037 	case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5038 		adapter->max_rx_add_entries_per_subcrq =
5039 		    be64_to_cpu(crq->query_capability.number);
5040 		netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5041 			   adapter->max_rx_add_entries_per_subcrq);
5042 		break;
5043 	case TCP_IP_OFFLOAD:
5044 		adapter->tcp_ip_offload =
5045 		    be64_to_cpu(crq->query_capability.number);
5046 		netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5047 			   adapter->tcp_ip_offload);
5048 		break;
5049 	case PROMISC_SUPPORTED:
5050 		adapter->promisc_supported =
5051 		    be64_to_cpu(crq->query_capability.number);
5052 		netdev_dbg(netdev, "promisc_supported = %lld\n",
5053 			   adapter->promisc_supported);
5054 		break;
5055 	case MIN_MTU:
5056 		adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5057 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5058 		netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5059 		break;
5060 	case MAX_MTU:
5061 		adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5062 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5063 		netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5064 		break;
5065 	case MAX_MULTICAST_FILTERS:
5066 		adapter->max_multicast_filters =
5067 		    be64_to_cpu(crq->query_capability.number);
5068 		netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5069 			   adapter->max_multicast_filters);
5070 		break;
5071 	case VLAN_HEADER_INSERTION:
5072 		adapter->vlan_header_insertion =
5073 		    be64_to_cpu(crq->query_capability.number);
5074 		if (adapter->vlan_header_insertion)
5075 			netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5076 		netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5077 			   adapter->vlan_header_insertion);
5078 		break;
5079 	case RX_VLAN_HEADER_INSERTION:
5080 		adapter->rx_vlan_header_insertion =
5081 		    be64_to_cpu(crq->query_capability.number);
5082 		netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5083 			   adapter->rx_vlan_header_insertion);
5084 		break;
5085 	case MAX_TX_SG_ENTRIES:
5086 		adapter->max_tx_sg_entries =
5087 		    be64_to_cpu(crq->query_capability.number);
5088 		netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5089 			   adapter->max_tx_sg_entries);
5090 		break;
5091 	case RX_SG_SUPPORTED:
5092 		adapter->rx_sg_supported =
5093 		    be64_to_cpu(crq->query_capability.number);
5094 		netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5095 			   adapter->rx_sg_supported);
5096 		break;
5097 	case OPT_TX_COMP_SUB_QUEUES:
5098 		adapter->opt_tx_comp_sub_queues =
5099 		    be64_to_cpu(crq->query_capability.number);
5100 		netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5101 			   adapter->opt_tx_comp_sub_queues);
5102 		break;
5103 	case OPT_RX_COMP_QUEUES:
5104 		adapter->opt_rx_comp_queues =
5105 		    be64_to_cpu(crq->query_capability.number);
5106 		netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5107 			   adapter->opt_rx_comp_queues);
5108 		break;
5109 	case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5110 		adapter->opt_rx_bufadd_q_per_rx_comp_q =
5111 		    be64_to_cpu(crq->query_capability.number);
5112 		netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5113 			   adapter->opt_rx_bufadd_q_per_rx_comp_q);
5114 		break;
5115 	case OPT_TX_ENTRIES_PER_SUBCRQ:
5116 		adapter->opt_tx_entries_per_subcrq =
5117 		    be64_to_cpu(crq->query_capability.number);
5118 		netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5119 			   adapter->opt_tx_entries_per_subcrq);
5120 		break;
5121 	case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5122 		adapter->opt_rxba_entries_per_subcrq =
5123 		    be64_to_cpu(crq->query_capability.number);
5124 		netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5125 			   adapter->opt_rxba_entries_per_subcrq);
5126 		break;
5127 	case TX_RX_DESC_REQ:
5128 		adapter->tx_rx_desc_req = crq->query_capability.number;
5129 		netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5130 			   adapter->tx_rx_desc_req);
5131 		break;
5132 
5133 	default:
5134 		netdev_err(netdev, "Got invalid cap rsp %d\n",
5135 			   crq->query_capability.capability);
5136 	}
5137 
5138 out:
5139 	if (atomic_read(&adapter->running_cap_crqs) == 0) {
5140 		adapter->wait_capability = false;
5141 		send_request_cap(adapter, 0);
5142 	}
5143 }
5144 
5145 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5146 {
5147 	union ibmvnic_crq crq;
5148 	int rc;
5149 
5150 	memset(&crq, 0, sizeof(crq));
5151 	crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5152 	crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5153 
5154 	mutex_lock(&adapter->fw_lock);
5155 	adapter->fw_done_rc = 0;
5156 	reinit_completion(&adapter->fw_done);
5157 
5158 	rc = ibmvnic_send_crq(adapter, &crq);
5159 	if (rc) {
5160 		mutex_unlock(&adapter->fw_lock);
5161 		return rc;
5162 	}
5163 
5164 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5165 	if (rc) {
5166 		mutex_unlock(&adapter->fw_lock);
5167 		return rc;
5168 	}
5169 
5170 	mutex_unlock(&adapter->fw_lock);
5171 	return adapter->fw_done_rc ? -EIO : 0;
5172 }
5173 
5174 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5175 				       struct ibmvnic_adapter *adapter)
5176 {
5177 	struct net_device *netdev = adapter->netdev;
5178 	int rc;
5179 	__be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5180 
5181 	rc = crq->query_phys_parms_rsp.rc.code;
5182 	if (rc) {
5183 		netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5184 		return rc;
5185 	}
5186 	switch (rspeed) {
5187 	case IBMVNIC_10MBPS:
5188 		adapter->speed = SPEED_10;
5189 		break;
5190 	case IBMVNIC_100MBPS:
5191 		adapter->speed = SPEED_100;
5192 		break;
5193 	case IBMVNIC_1GBPS:
5194 		adapter->speed = SPEED_1000;
5195 		break;
5196 	case IBMVNIC_10GBPS:
5197 		adapter->speed = SPEED_10000;
5198 		break;
5199 	case IBMVNIC_25GBPS:
5200 		adapter->speed = SPEED_25000;
5201 		break;
5202 	case IBMVNIC_40GBPS:
5203 		adapter->speed = SPEED_40000;
5204 		break;
5205 	case IBMVNIC_50GBPS:
5206 		adapter->speed = SPEED_50000;
5207 		break;
5208 	case IBMVNIC_100GBPS:
5209 		adapter->speed = SPEED_100000;
5210 		break;
5211 	case IBMVNIC_200GBPS:
5212 		adapter->speed = SPEED_200000;
5213 		break;
5214 	default:
5215 		if (netif_carrier_ok(netdev))
5216 			netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5217 		adapter->speed = SPEED_UNKNOWN;
5218 	}
5219 	if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5220 		adapter->duplex = DUPLEX_FULL;
5221 	else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5222 		adapter->duplex = DUPLEX_HALF;
5223 	else
5224 		adapter->duplex = DUPLEX_UNKNOWN;
5225 
5226 	return rc;
5227 }
5228 
5229 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5230 			       struct ibmvnic_adapter *adapter)
5231 {
5232 	struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5233 	struct net_device *netdev = adapter->netdev;
5234 	struct device *dev = &adapter->vdev->dev;
5235 	u64 *u64_crq = (u64 *)crq;
5236 	long rc;
5237 
5238 	netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5239 		   (unsigned long)cpu_to_be64(u64_crq[0]),
5240 		   (unsigned long)cpu_to_be64(u64_crq[1]));
5241 	switch (gen_crq->first) {
5242 	case IBMVNIC_CRQ_INIT_RSP:
5243 		switch (gen_crq->cmd) {
5244 		case IBMVNIC_CRQ_INIT:
5245 			dev_info(dev, "Partner initialized\n");
5246 			adapter->from_passive_init = true;
5247 			/* Discard any stale login responses from prev reset.
5248 			 * CHECK: should we clear even on INIT_COMPLETE?
5249 			 */
5250 			adapter->login_pending = false;
5251 
5252 			if (adapter->state == VNIC_DOWN)
5253 				rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5254 			else
5255 				rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5256 
5257 			if (rc && rc != -EBUSY) {
5258 				/* We were unable to schedule the failover
5259 				 * reset either because the adapter was still
5260 				 * probing (eg: during kexec) or we could not
5261 				 * allocate memory. Clear the failover_pending
5262 				 * flag since no one else will. We ignore
5263 				 * EBUSY because it means either FAILOVER reset
5264 				 * is already scheduled or the adapter is
5265 				 * being removed.
5266 				 */
5267 				netdev_err(netdev,
5268 					   "Error %ld scheduling failover reset\n",
5269 					   rc);
5270 				adapter->failover_pending = false;
5271 			}
5272 
5273 			if (!completion_done(&adapter->init_done)) {
5274 				complete(&adapter->init_done);
5275 				if (!adapter->init_done_rc)
5276 					adapter->init_done_rc = -EAGAIN;
5277 			}
5278 
5279 			break;
5280 		case IBMVNIC_CRQ_INIT_COMPLETE:
5281 			dev_info(dev, "Partner initialization complete\n");
5282 			adapter->crq.active = true;
5283 			send_version_xchg(adapter);
5284 			break;
5285 		default:
5286 			dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5287 		}
5288 		return;
5289 	case IBMVNIC_CRQ_XPORT_EVENT:
5290 		netif_carrier_off(netdev);
5291 		adapter->crq.active = false;
5292 		/* terminate any thread waiting for a response
5293 		 * from the device
5294 		 */
5295 		if (!completion_done(&adapter->fw_done)) {
5296 			adapter->fw_done_rc = -EIO;
5297 			complete(&adapter->fw_done);
5298 		}
5299 		if (!completion_done(&adapter->stats_done))
5300 			complete(&adapter->stats_done);
5301 		if (test_bit(0, &adapter->resetting))
5302 			adapter->force_reset_recovery = true;
5303 		if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5304 			dev_info(dev, "Migrated, re-enabling adapter\n");
5305 			ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5306 		} else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5307 			dev_info(dev, "Backing device failover detected\n");
5308 			adapter->failover_pending = true;
5309 		} else {
5310 			/* The adapter lost the connection */
5311 			dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
5312 				gen_crq->cmd);
5313 			ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5314 		}
5315 		return;
5316 	case IBMVNIC_CRQ_CMD_RSP:
5317 		break;
5318 	default:
5319 		dev_err(dev, "Got an invalid msg type 0x%02x\n",
5320 			gen_crq->first);
5321 		return;
5322 	}
5323 
5324 	switch (gen_crq->cmd) {
5325 	case VERSION_EXCHANGE_RSP:
5326 		rc = crq->version_exchange_rsp.rc.code;
5327 		if (rc) {
5328 			dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
5329 			break;
5330 		}
5331 		ibmvnic_version =
5332 			    be16_to_cpu(crq->version_exchange_rsp.version);
5333 		dev_info(dev, "Partner protocol version is %d\n",
5334 			 ibmvnic_version);
5335 		send_query_cap(adapter);
5336 		break;
5337 	case QUERY_CAPABILITY_RSP:
5338 		handle_query_cap_rsp(crq, adapter);
5339 		break;
5340 	case QUERY_MAP_RSP:
5341 		handle_query_map_rsp(crq, adapter);
5342 		break;
5343 	case REQUEST_MAP_RSP:
5344 		adapter->fw_done_rc = crq->request_map_rsp.rc.code;
5345 		complete(&adapter->fw_done);
5346 		break;
5347 	case REQUEST_UNMAP_RSP:
5348 		handle_request_unmap_rsp(crq, adapter);
5349 		break;
5350 	case REQUEST_CAPABILITY_RSP:
5351 		handle_request_cap_rsp(crq, adapter);
5352 		break;
5353 	case LOGIN_RSP:
5354 		netdev_dbg(netdev, "Got Login Response\n");
5355 		handle_login_rsp(crq, adapter);
5356 		break;
5357 	case LOGICAL_LINK_STATE_RSP:
5358 		netdev_dbg(netdev,
5359 			   "Got Logical Link State Response, state: %d rc: %d\n",
5360 			   crq->logical_link_state_rsp.link_state,
5361 			   crq->logical_link_state_rsp.rc.code);
5362 		adapter->logical_link_state =
5363 		    crq->logical_link_state_rsp.link_state;
5364 		adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5365 		complete(&adapter->init_done);
5366 		break;
5367 	case LINK_STATE_INDICATION:
5368 		netdev_dbg(netdev, "Got Logical Link State Indication\n");
5369 		adapter->phys_link_state =
5370 		    crq->link_state_indication.phys_link_state;
5371 		adapter->logical_link_state =
5372 		    crq->link_state_indication.logical_link_state;
5373 		if (adapter->phys_link_state && adapter->logical_link_state)
5374 			netif_carrier_on(netdev);
5375 		else
5376 			netif_carrier_off(netdev);
5377 		break;
5378 	case CHANGE_MAC_ADDR_RSP:
5379 		netdev_dbg(netdev, "Got MAC address change Response\n");
5380 		adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
5381 		break;
5382 	case ERROR_INDICATION:
5383 		netdev_dbg(netdev, "Got Error Indication\n");
5384 		handle_error_indication(crq, adapter);
5385 		break;
5386 	case REQUEST_STATISTICS_RSP:
5387 		netdev_dbg(netdev, "Got Statistics Response\n");
5388 		complete(&adapter->stats_done);
5389 		break;
5390 	case QUERY_IP_OFFLOAD_RSP:
5391 		netdev_dbg(netdev, "Got Query IP offload Response\n");
5392 		handle_query_ip_offload_rsp(adapter);
5393 		break;
5394 	case MULTICAST_CTRL_RSP:
5395 		netdev_dbg(netdev, "Got multicast control Response\n");
5396 		break;
5397 	case CONTROL_IP_OFFLOAD_RSP:
5398 		netdev_dbg(netdev, "Got Control IP offload Response\n");
5399 		dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5400 				 sizeof(adapter->ip_offload_ctrl),
5401 				 DMA_TO_DEVICE);
5402 		complete(&adapter->init_done);
5403 		break;
5404 	case COLLECT_FW_TRACE_RSP:
5405 		netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5406 		complete(&adapter->fw_done);
5407 		break;
5408 	case GET_VPD_SIZE_RSP:
5409 		handle_vpd_size_rsp(crq, adapter);
5410 		break;
5411 	case GET_VPD_RSP:
5412 		handle_vpd_rsp(crq, adapter);
5413 		break;
5414 	case QUERY_PHYS_PARMS_RSP:
5415 		adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5416 		complete(&adapter->fw_done);
5417 		break;
5418 	default:
5419 		netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5420 			   gen_crq->cmd);
5421 	}
5422 }
5423 
5424 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5425 {
5426 	struct ibmvnic_adapter *adapter = instance;
5427 
5428 	tasklet_schedule(&adapter->tasklet);
5429 	return IRQ_HANDLED;
5430 }
5431 
5432 static void ibmvnic_tasklet(struct tasklet_struct *t)
5433 {
5434 	struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5435 	struct ibmvnic_crq_queue *queue = &adapter->crq;
5436 	union ibmvnic_crq *crq;
5437 	unsigned long flags;
5438 	bool done = false;
5439 
5440 	spin_lock_irqsave(&queue->lock, flags);
5441 	while (!done) {
5442 		/* Pull all the valid messages off the CRQ */
5443 		while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5444 			/* This barrier makes sure ibmvnic_next_crq()'s
5445 			 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5446 			 * before ibmvnic_handle_crq()'s
5447 			 * switch(gen_crq->first) and switch(gen_crq->cmd).
5448 			 */
5449 			dma_rmb();
5450 			ibmvnic_handle_crq(crq, adapter);
5451 			crq->generic.first = 0;
5452 		}
5453 
5454 		/* remain in tasklet until all
5455 		 * capabilities responses are received
5456 		 */
5457 		if (!adapter->wait_capability)
5458 			done = true;
5459 	}
5460 	/* if capabilities CRQ's were sent in this tasklet, the following
5461 	 * tasklet must wait until all responses are received
5462 	 */
5463 	if (atomic_read(&adapter->running_cap_crqs) != 0)
5464 		adapter->wait_capability = true;
5465 	spin_unlock_irqrestore(&queue->lock, flags);
5466 }
5467 
5468 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5469 {
5470 	struct vio_dev *vdev = adapter->vdev;
5471 	int rc;
5472 
5473 	do {
5474 		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5475 	} while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5476 
5477 	if (rc)
5478 		dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5479 
5480 	return rc;
5481 }
5482 
5483 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5484 {
5485 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5486 	struct device *dev = &adapter->vdev->dev;
5487 	struct vio_dev *vdev = adapter->vdev;
5488 	int rc;
5489 
5490 	/* Close the CRQ */
5491 	do {
5492 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5493 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5494 
5495 	/* Clean out the queue */
5496 	if (!crq->msgs)
5497 		return -EINVAL;
5498 
5499 	memset(crq->msgs, 0, PAGE_SIZE);
5500 	crq->cur = 0;
5501 	crq->active = false;
5502 
5503 	/* And re-open it again */
5504 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5505 				crq->msg_token, PAGE_SIZE);
5506 
5507 	if (rc == H_CLOSED)
5508 		/* Adapter is good, but other end is not ready */
5509 		dev_warn(dev, "Partner adapter not ready\n");
5510 	else if (rc != 0)
5511 		dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5512 
5513 	return rc;
5514 }
5515 
5516 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5517 {
5518 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5519 	struct vio_dev *vdev = adapter->vdev;
5520 	long rc;
5521 
5522 	if (!crq->msgs)
5523 		return;
5524 
5525 	netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5526 	free_irq(vdev->irq, adapter);
5527 	tasklet_kill(&adapter->tasklet);
5528 	do {
5529 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5530 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5531 
5532 	dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5533 			 DMA_BIDIRECTIONAL);
5534 	free_page((unsigned long)crq->msgs);
5535 	crq->msgs = NULL;
5536 	crq->active = false;
5537 }
5538 
5539 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5540 {
5541 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5542 	struct device *dev = &adapter->vdev->dev;
5543 	struct vio_dev *vdev = adapter->vdev;
5544 	int rc, retrc = -ENOMEM;
5545 
5546 	if (crq->msgs)
5547 		return 0;
5548 
5549 	crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5550 	/* Should we allocate more than one page? */
5551 
5552 	if (!crq->msgs)
5553 		return -ENOMEM;
5554 
5555 	crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5556 	crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5557 					DMA_BIDIRECTIONAL);
5558 	if (dma_mapping_error(dev, crq->msg_token))
5559 		goto map_failed;
5560 
5561 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5562 				crq->msg_token, PAGE_SIZE);
5563 
5564 	if (rc == H_RESOURCE)
5565 		/* maybe kexecing and resource is busy. try a reset */
5566 		rc = ibmvnic_reset_crq(adapter);
5567 	retrc = rc;
5568 
5569 	if (rc == H_CLOSED) {
5570 		dev_warn(dev, "Partner adapter not ready\n");
5571 	} else if (rc) {
5572 		dev_warn(dev, "Error %d opening adapter\n", rc);
5573 		goto reg_crq_failed;
5574 	}
5575 
5576 	retrc = 0;
5577 
5578 	tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5579 
5580 	netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5581 	snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5582 		 adapter->vdev->unit_address);
5583 	rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5584 	if (rc) {
5585 		dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5586 			vdev->irq, rc);
5587 		goto req_irq_failed;
5588 	}
5589 
5590 	rc = vio_enable_interrupts(vdev);
5591 	if (rc) {
5592 		dev_err(dev, "Error %d enabling interrupts\n", rc);
5593 		goto req_irq_failed;
5594 	}
5595 
5596 	crq->cur = 0;
5597 	spin_lock_init(&crq->lock);
5598 
5599 	/* process any CRQs that were queued before we enabled interrupts */
5600 	tasklet_schedule(&adapter->tasklet);
5601 
5602 	return retrc;
5603 
5604 req_irq_failed:
5605 	tasklet_kill(&adapter->tasklet);
5606 	do {
5607 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5608 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5609 reg_crq_failed:
5610 	dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5611 map_failed:
5612 	free_page((unsigned long)crq->msgs);
5613 	crq->msgs = NULL;
5614 	return retrc;
5615 }
5616 
5617 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5618 {
5619 	struct device *dev = &adapter->vdev->dev;
5620 	unsigned long timeout = msecs_to_jiffies(20000);
5621 	u64 old_num_rx_queues = adapter->req_rx_queues;
5622 	u64 old_num_tx_queues = adapter->req_tx_queues;
5623 	int rc;
5624 
5625 	adapter->from_passive_init = false;
5626 
5627 	if (reset)
5628 		reinit_completion(&adapter->init_done);
5629 
5630 	adapter->init_done_rc = 0;
5631 	rc = ibmvnic_send_crq_init(adapter);
5632 	if (rc) {
5633 		dev_err(dev, "Send crq init failed with error %d\n", rc);
5634 		return rc;
5635 	}
5636 
5637 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5638 		dev_err(dev, "Initialization sequence timed out\n");
5639 		return -ETIMEDOUT;
5640 	}
5641 
5642 	if (adapter->init_done_rc) {
5643 		release_crq_queue(adapter);
5644 		return adapter->init_done_rc;
5645 	}
5646 
5647 	if (adapter->from_passive_init) {
5648 		adapter->state = VNIC_OPEN;
5649 		adapter->from_passive_init = false;
5650 		return -EINVAL;
5651 	}
5652 
5653 	if (reset &&
5654 	    test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5655 	    adapter->reset_reason != VNIC_RESET_MOBILITY) {
5656 		if (adapter->req_rx_queues != old_num_rx_queues ||
5657 		    adapter->req_tx_queues != old_num_tx_queues) {
5658 			release_sub_crqs(adapter, 0);
5659 			rc = init_sub_crqs(adapter);
5660 		} else {
5661 			rc = reset_sub_crq_queues(adapter);
5662 		}
5663 	} else {
5664 		rc = init_sub_crqs(adapter);
5665 	}
5666 
5667 	if (rc) {
5668 		dev_err(dev, "Initialization of sub crqs failed\n");
5669 		release_crq_queue(adapter);
5670 		return rc;
5671 	}
5672 
5673 	rc = init_sub_crq_irqs(adapter);
5674 	if (rc) {
5675 		dev_err(dev, "Failed to initialize sub crq irqs\n");
5676 		release_crq_queue(adapter);
5677 	}
5678 
5679 	return rc;
5680 }
5681 
5682 static struct device_attribute dev_attr_failover;
5683 
5684 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5685 {
5686 	struct ibmvnic_adapter *adapter;
5687 	struct net_device *netdev;
5688 	unsigned char *mac_addr_p;
5689 	bool init_success;
5690 	int rc;
5691 
5692 	dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5693 		dev->unit_address);
5694 
5695 	mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5696 							VETH_MAC_ADDR, NULL);
5697 	if (!mac_addr_p) {
5698 		dev_err(&dev->dev,
5699 			"(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5700 			__FILE__, __LINE__);
5701 		return 0;
5702 	}
5703 
5704 	netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5705 				   IBMVNIC_MAX_QUEUES);
5706 	if (!netdev)
5707 		return -ENOMEM;
5708 
5709 	adapter = netdev_priv(netdev);
5710 	adapter->state = VNIC_PROBING;
5711 	dev_set_drvdata(&dev->dev, netdev);
5712 	adapter->vdev = dev;
5713 	adapter->netdev = netdev;
5714 	adapter->login_pending = false;
5715 	memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
5716 	/* map_ids start at 1, so ensure map_id 0 is always "in-use" */
5717 	bitmap_set(adapter->map_ids, 0, 1);
5718 
5719 	ether_addr_copy(adapter->mac_addr, mac_addr_p);
5720 	eth_hw_addr_set(netdev, adapter->mac_addr);
5721 	netdev->irq = dev->irq;
5722 	netdev->netdev_ops = &ibmvnic_netdev_ops;
5723 	netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5724 	SET_NETDEV_DEV(netdev, &dev->dev);
5725 
5726 	INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5727 	INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5728 			  __ibmvnic_delayed_reset);
5729 	INIT_LIST_HEAD(&adapter->rwi_list);
5730 	spin_lock_init(&adapter->rwi_lock);
5731 	spin_lock_init(&adapter->state_lock);
5732 	mutex_init(&adapter->fw_lock);
5733 	init_completion(&adapter->init_done);
5734 	init_completion(&adapter->fw_done);
5735 	init_completion(&adapter->reset_done);
5736 	init_completion(&adapter->stats_done);
5737 	clear_bit(0, &adapter->resetting);
5738 	adapter->prev_rx_buf_sz = 0;
5739 	adapter->prev_mtu = 0;
5740 
5741 	init_success = false;
5742 	do {
5743 		rc = init_crq_queue(adapter);
5744 		if (rc) {
5745 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5746 				rc);
5747 			goto ibmvnic_init_fail;
5748 		}
5749 
5750 		rc = ibmvnic_reset_init(adapter, false);
5751 	} while (rc == -EAGAIN);
5752 
5753 	/* We are ignoring the error from ibmvnic_reset_init() assuming that the
5754 	 * partner is not ready. CRQ is not active. When the partner becomes
5755 	 * ready, we will do the passive init reset.
5756 	 */
5757 
5758 	if (!rc)
5759 		init_success = true;
5760 
5761 	rc = init_stats_buffers(adapter);
5762 	if (rc)
5763 		goto ibmvnic_init_fail;
5764 
5765 	rc = init_stats_token(adapter);
5766 	if (rc)
5767 		goto ibmvnic_stats_fail;
5768 
5769 	rc = device_create_file(&dev->dev, &dev_attr_failover);
5770 	if (rc)
5771 		goto ibmvnic_dev_file_err;
5772 
5773 	netif_carrier_off(netdev);
5774 	rc = register_netdev(netdev);
5775 	if (rc) {
5776 		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5777 		goto ibmvnic_register_fail;
5778 	}
5779 	dev_info(&dev->dev, "ibmvnic registered\n");
5780 
5781 	if (init_success) {
5782 		adapter->state = VNIC_PROBED;
5783 		netdev->mtu = adapter->req_mtu - ETH_HLEN;
5784 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5785 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5786 	} else {
5787 		adapter->state = VNIC_DOWN;
5788 	}
5789 
5790 	adapter->wait_for_reset = false;
5791 	adapter->last_reset_time = jiffies;
5792 	return 0;
5793 
5794 ibmvnic_register_fail:
5795 	device_remove_file(&dev->dev, &dev_attr_failover);
5796 
5797 ibmvnic_dev_file_err:
5798 	release_stats_token(adapter);
5799 
5800 ibmvnic_stats_fail:
5801 	release_stats_buffers(adapter);
5802 
5803 ibmvnic_init_fail:
5804 	release_sub_crqs(adapter, 1);
5805 	release_crq_queue(adapter);
5806 	mutex_destroy(&adapter->fw_lock);
5807 	free_netdev(netdev);
5808 
5809 	return rc;
5810 }
5811 
5812 static void ibmvnic_remove(struct vio_dev *dev)
5813 {
5814 	struct net_device *netdev = dev_get_drvdata(&dev->dev);
5815 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5816 	unsigned long flags;
5817 
5818 	spin_lock_irqsave(&adapter->state_lock, flags);
5819 
5820 	/* If ibmvnic_reset() is scheduling a reset, wait for it to
5821 	 * finish. Then, set the state to REMOVING to prevent it from
5822 	 * scheduling any more work and to have reset functions ignore
5823 	 * any resets that have already been scheduled. Drop the lock
5824 	 * after setting state, so __ibmvnic_reset() which is called
5825 	 * from the flush_work() below, can make progress.
5826 	 */
5827 	spin_lock(&adapter->rwi_lock);
5828 	adapter->state = VNIC_REMOVING;
5829 	spin_unlock(&adapter->rwi_lock);
5830 
5831 	spin_unlock_irqrestore(&adapter->state_lock, flags);
5832 
5833 	flush_work(&adapter->ibmvnic_reset);
5834 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5835 
5836 	rtnl_lock();
5837 	unregister_netdevice(netdev);
5838 
5839 	release_resources(adapter);
5840 	release_rx_pools(adapter);
5841 	release_tx_pools(adapter);
5842 	release_sub_crqs(adapter, 1);
5843 	release_crq_queue(adapter);
5844 
5845 	release_stats_token(adapter);
5846 	release_stats_buffers(adapter);
5847 
5848 	adapter->state = VNIC_REMOVED;
5849 
5850 	rtnl_unlock();
5851 	mutex_destroy(&adapter->fw_lock);
5852 	device_remove_file(&dev->dev, &dev_attr_failover);
5853 	free_netdev(netdev);
5854 	dev_set_drvdata(&dev->dev, NULL);
5855 }
5856 
5857 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5858 			      const char *buf, size_t count)
5859 {
5860 	struct net_device *netdev = dev_get_drvdata(dev);
5861 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5862 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5863 	__be64 session_token;
5864 	long rc;
5865 
5866 	if (!sysfs_streq(buf, "1"))
5867 		return -EINVAL;
5868 
5869 	rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5870 			 H_GET_SESSION_TOKEN, 0, 0, 0);
5871 	if (rc) {
5872 		netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5873 			   rc);
5874 		goto last_resort;
5875 	}
5876 
5877 	session_token = (__be64)retbuf[0];
5878 	netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5879 		   be64_to_cpu(session_token));
5880 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5881 				H_SESSION_ERR_DETECTED, session_token, 0, 0);
5882 	if (rc)
5883 		netdev_err(netdev,
5884 			   "H_VIOCTL initiated failover failed, rc %ld\n",
5885 			   rc);
5886 
5887 last_resort:
5888 	netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
5889 	ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5890 
5891 	return count;
5892 }
5893 static DEVICE_ATTR_WO(failover);
5894 
5895 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5896 {
5897 	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5898 	struct ibmvnic_adapter *adapter;
5899 	struct iommu_table *tbl;
5900 	unsigned long ret = 0;
5901 	int i;
5902 
5903 	tbl = get_iommu_table_base(&vdev->dev);
5904 
5905 	/* netdev inits at probe time along with the structures we need below*/
5906 	if (!netdev)
5907 		return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5908 
5909 	adapter = netdev_priv(netdev);
5910 
5911 	ret += PAGE_SIZE; /* the crq message queue */
5912 	ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5913 
5914 	for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5915 		ret += 4 * PAGE_SIZE; /* the scrq message queue */
5916 
5917 	for (i = 0; i < adapter->num_active_rx_pools; i++)
5918 		ret += adapter->rx_pool[i].size *
5919 		    IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5920 
5921 	return ret;
5922 }
5923 
5924 static int ibmvnic_resume(struct device *dev)
5925 {
5926 	struct net_device *netdev = dev_get_drvdata(dev);
5927 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5928 
5929 	if (adapter->state != VNIC_OPEN)
5930 		return 0;
5931 
5932 	tasklet_schedule(&adapter->tasklet);
5933 
5934 	return 0;
5935 }
5936 
5937 static const struct vio_device_id ibmvnic_device_table[] = {
5938 	{"network", "IBM,vnic"},
5939 	{"", "" }
5940 };
5941 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5942 
5943 static const struct dev_pm_ops ibmvnic_pm_ops = {
5944 	.resume = ibmvnic_resume
5945 };
5946 
5947 static struct vio_driver ibmvnic_driver = {
5948 	.id_table       = ibmvnic_device_table,
5949 	.probe          = ibmvnic_probe,
5950 	.remove         = ibmvnic_remove,
5951 	.get_desired_dma = ibmvnic_get_desired_dma,
5952 	.name		= ibmvnic_driver_name,
5953 	.pm		= &ibmvnic_pm_ops,
5954 };
5955 
5956 /* module functions */
5957 static int __init ibmvnic_module_init(void)
5958 {
5959 	pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5960 		IBMVNIC_DRIVER_VERSION);
5961 
5962 	return vio_register_driver(&ibmvnic_driver);
5963 }
5964 
5965 static void __exit ibmvnic_module_exit(void)
5966 {
5967 	vio_unregister_driver(&ibmvnic_driver);
5968 }
5969 
5970 module_init(ibmvnic_module_init);
5971 module_exit(ibmvnic_module_exit);
5972