xref: /linux/drivers/net/ethernet/ibm/ibmvnic.c (revision 69bfec7548f4c1595bac0e3ddfc0458a5af31f4c)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /**************************************************************************/
3 /*                                                                        */
4 /*  IBM System i and System p Virtual NIC Device Driver                   */
5 /*  Copyright (C) 2014 IBM Corp.                                          */
6 /*  Santiago Leon (santi_leon@yahoo.com)                                  */
7 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
8 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
9 /*                                                                        */
10 /*                                                                        */
11 /* This module contains the implementation of a virtual ethernet device   */
12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
15 /*									   */
16 /* Messages are passed between the VNIC driver and the VNIC server using  */
17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
18 /* issue and receive commands that initiate communication with the server */
19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
20 /* are used by the driver to notify the server that a packet is           */
21 /* ready for transmission or that a buffer has been added to receive a    */
22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
23 /* driver that a packet transmission has been completed or that a packet  */
24 /* has been received and placed in a waiting buffer.                      */
25 /*                                                                        */
26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
28 /* or receive has been completed, the VNIC driver is required to use      */
29 /* "long term mapping". This entails that large, continuous DMA mapped    */
30 /* buffers are allocated on driver initialization and these buffers are   */
31 /* then continuously reused to pass skbs to and from the VNIC server.     */
32 /*                                                                        */
33 /**************************************************************************/
34 
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/types.h>
38 #include <linux/errno.h>
39 #include <linux/completion.h>
40 #include <linux/ioport.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kernel.h>
43 #include <linux/netdevice.h>
44 #include <linux/etherdevice.h>
45 #include <linux/skbuff.h>
46 #include <linux/init.h>
47 #include <linux/delay.h>
48 #include <linux/mm.h>
49 #include <linux/ethtool.h>
50 #include <linux/proc_fs.h>
51 #include <linux/if_arp.h>
52 #include <linux/in.h>
53 #include <linux/ip.h>
54 #include <linux/ipv6.h>
55 #include <linux/irq.h>
56 #include <linux/irqdomain.h>
57 #include <linux/kthread.h>
58 #include <linux/seq_file.h>
59 #include <linux/interrupt.h>
60 #include <net/net_namespace.h>
61 #include <asm/hvcall.h>
62 #include <linux/atomic.h>
63 #include <asm/vio.h>
64 #include <asm/xive.h>
65 #include <asm/iommu.h>
66 #include <linux/uaccess.h>
67 #include <asm/firmware.h>
68 #include <linux/workqueue.h>
69 #include <linux/if_vlan.h>
70 #include <linux/utsname.h>
71 #include <linux/cpu.h>
72 
73 #include "ibmvnic.h"
74 
75 static const char ibmvnic_driver_name[] = "ibmvnic";
76 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
77 
78 MODULE_AUTHOR("Santiago Leon");
79 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
80 MODULE_LICENSE("GPL");
81 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
82 
83 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
84 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
85 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
86 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
87 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
88 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
89 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
90 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
91 static int enable_scrq_irq(struct ibmvnic_adapter *,
92 			   struct ibmvnic_sub_crq_queue *);
93 static int disable_scrq_irq(struct ibmvnic_adapter *,
94 			    struct ibmvnic_sub_crq_queue *);
95 static int pending_scrq(struct ibmvnic_adapter *,
96 			struct ibmvnic_sub_crq_queue *);
97 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
98 					struct ibmvnic_sub_crq_queue *);
99 static int ibmvnic_poll(struct napi_struct *napi, int data);
100 static void send_query_map(struct ibmvnic_adapter *adapter);
101 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
102 static int send_request_unmap(struct ibmvnic_adapter *, u8);
103 static int send_login(struct ibmvnic_adapter *adapter);
104 static void send_query_cap(struct ibmvnic_adapter *adapter);
105 static int init_sub_crqs(struct ibmvnic_adapter *);
106 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
107 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
108 static void release_crq_queue(struct ibmvnic_adapter *);
109 static int __ibmvnic_set_mac(struct net_device *, u8 *);
110 static int init_crq_queue(struct ibmvnic_adapter *adapter);
111 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
112 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
113 					 struct ibmvnic_sub_crq_queue *tx_scrq);
114 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
115 				struct ibmvnic_long_term_buff *ltb);
116 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter);
117 
118 struct ibmvnic_stat {
119 	char name[ETH_GSTRING_LEN];
120 	int offset;
121 };
122 
123 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
124 			     offsetof(struct ibmvnic_statistics, stat))
125 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
126 
127 static const struct ibmvnic_stat ibmvnic_stats[] = {
128 	{"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
129 	{"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
130 	{"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
131 	{"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
132 	{"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
133 	{"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
134 	{"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
135 	{"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
136 	{"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
137 	{"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
138 	{"align_errors", IBMVNIC_STAT_OFF(align_errors)},
139 	{"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
140 	{"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
141 	{"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
142 	{"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
143 	{"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
144 	{"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
145 	{"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
146 	{"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
147 	{"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
148 	{"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
149 	{"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
150 };
151 
152 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
153 {
154 	union ibmvnic_crq crq;
155 
156 	memset(&crq, 0, sizeof(crq));
157 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
158 	crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
159 
160 	return ibmvnic_send_crq(adapter, &crq);
161 }
162 
163 static int send_version_xchg(struct ibmvnic_adapter *adapter)
164 {
165 	union ibmvnic_crq crq;
166 
167 	memset(&crq, 0, sizeof(crq));
168 	crq.version_exchange.first = IBMVNIC_CRQ_CMD;
169 	crq.version_exchange.cmd = VERSION_EXCHANGE;
170 	crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
171 
172 	return ibmvnic_send_crq(adapter, &crq);
173 }
174 
175 static void ibmvnic_clean_queue_affinity(struct ibmvnic_adapter *adapter,
176 					 struct ibmvnic_sub_crq_queue *queue)
177 {
178 	if (!(queue && queue->irq))
179 		return;
180 
181 	cpumask_clear(queue->affinity_mask);
182 
183 	if (irq_set_affinity_and_hint(queue->irq, NULL))
184 		netdev_warn(adapter->netdev,
185 			    "%s: Clear affinity failed, queue addr = %p, IRQ = %d\n",
186 			    __func__, queue, queue->irq);
187 }
188 
189 static void ibmvnic_clean_affinity(struct ibmvnic_adapter *adapter)
190 {
191 	struct ibmvnic_sub_crq_queue **rxqs;
192 	struct ibmvnic_sub_crq_queue **txqs;
193 	int num_rxqs, num_txqs;
194 	int rc, i;
195 
196 	rc = 0;
197 	rxqs = adapter->rx_scrq;
198 	txqs = adapter->tx_scrq;
199 	num_txqs = adapter->num_active_tx_scrqs;
200 	num_rxqs = adapter->num_active_rx_scrqs;
201 
202 	netdev_dbg(adapter->netdev, "%s: Cleaning irq affinity hints", __func__);
203 	if (txqs) {
204 		for (i = 0; i < num_txqs; i++)
205 			ibmvnic_clean_queue_affinity(adapter, txqs[i]);
206 	}
207 	if (rxqs) {
208 		for (i = 0; i < num_rxqs; i++)
209 			ibmvnic_clean_queue_affinity(adapter, rxqs[i]);
210 	}
211 }
212 
213 static int ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue *queue,
214 				      unsigned int *cpu, int *stragglers,
215 				      int stride)
216 {
217 	cpumask_var_t mask;
218 	int i;
219 	int rc = 0;
220 
221 	if (!(queue && queue->irq))
222 		return rc;
223 
224 	/* cpumask_var_t is either a pointer or array, allocation works here */
225 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
226 		return -ENOMEM;
227 
228 	/* while we have extra cpu give one extra to this irq */
229 	if (*stragglers) {
230 		stride++;
231 		(*stragglers)--;
232 	}
233 	/* atomic write is safer than writing bit by bit directly */
234 	for (i = 0; i < stride; i++) {
235 		cpumask_set_cpu(*cpu, mask);
236 		*cpu = cpumask_next_wrap(*cpu, cpu_online_mask,
237 					 nr_cpu_ids, false);
238 	}
239 	/* set queue affinity mask */
240 	cpumask_copy(queue->affinity_mask, mask);
241 	rc = irq_set_affinity_and_hint(queue->irq, queue->affinity_mask);
242 	free_cpumask_var(mask);
243 
244 	return rc;
245 }
246 
247 /* assumes cpu read lock is held */
248 static void ibmvnic_set_affinity(struct ibmvnic_adapter *adapter)
249 {
250 	struct ibmvnic_sub_crq_queue **rxqs = adapter->rx_scrq;
251 	struct ibmvnic_sub_crq_queue **txqs = adapter->tx_scrq;
252 	struct ibmvnic_sub_crq_queue *queue;
253 	int num_rxqs = adapter->num_active_rx_scrqs, i_rxqs = 0;
254 	int num_txqs = adapter->num_active_tx_scrqs, i_txqs = 0;
255 	int total_queues, stride, stragglers, i;
256 	unsigned int num_cpu, cpu;
257 	bool is_rx_queue;
258 	int rc = 0;
259 
260 	netdev_dbg(adapter->netdev, "%s: Setting irq affinity hints", __func__);
261 	if (!(adapter->rx_scrq && adapter->tx_scrq)) {
262 		netdev_warn(adapter->netdev,
263 			    "%s: Set affinity failed, queues not allocated\n",
264 			    __func__);
265 		return;
266 	}
267 
268 	total_queues = num_rxqs + num_txqs;
269 	num_cpu = num_online_cpus();
270 	/* number of cpu's assigned per irq */
271 	stride = max_t(int, num_cpu / total_queues, 1);
272 	/* number of leftover cpu's */
273 	stragglers = num_cpu >= total_queues ? num_cpu % total_queues : 0;
274 	/* next available cpu to assign irq to */
275 	cpu = cpumask_next(-1, cpu_online_mask);
276 
277 	for (i = 0; i < total_queues; i++) {
278 		is_rx_queue = false;
279 		/* balance core load by alternating rx and tx assignments
280 		 * ex: TX0 -> RX0 -> TX1 -> RX1 etc.
281 		 */
282 		if ((i % 2 == 1 && i_rxqs < num_rxqs) || i_txqs == num_txqs) {
283 			queue = rxqs[i_rxqs++];
284 			is_rx_queue = true;
285 		} else {
286 			queue = txqs[i_txqs++];
287 		}
288 
289 		rc = ibmvnic_set_queue_affinity(queue, &cpu, &stragglers,
290 						stride);
291 		if (rc)
292 			goto out;
293 
294 		if (!queue || is_rx_queue)
295 			continue;
296 
297 		rc = __netif_set_xps_queue(adapter->netdev,
298 					   cpumask_bits(queue->affinity_mask),
299 					   i_txqs - 1, XPS_CPUS);
300 		if (rc)
301 			netdev_warn(adapter->netdev, "%s: Set XPS on queue %d failed, rc = %d.\n",
302 				    __func__, i_txqs - 1, rc);
303 	}
304 
305 out:
306 	if (rc) {
307 		netdev_warn(adapter->netdev,
308 			    "%s: Set affinity failed, queue addr = %p, IRQ = %d, rc = %d.\n",
309 			    __func__, queue, queue->irq, rc);
310 		ibmvnic_clean_affinity(adapter);
311 	}
312 }
313 
314 static int ibmvnic_cpu_online(unsigned int cpu, struct hlist_node *node)
315 {
316 	struct ibmvnic_adapter *adapter;
317 
318 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
319 	ibmvnic_set_affinity(adapter);
320 	return 0;
321 }
322 
323 static int ibmvnic_cpu_dead(unsigned int cpu, struct hlist_node *node)
324 {
325 	struct ibmvnic_adapter *adapter;
326 
327 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node_dead);
328 	ibmvnic_set_affinity(adapter);
329 	return 0;
330 }
331 
332 static int ibmvnic_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
333 {
334 	struct ibmvnic_adapter *adapter;
335 
336 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
337 	ibmvnic_clean_affinity(adapter);
338 	return 0;
339 }
340 
341 static enum cpuhp_state ibmvnic_online;
342 
343 static int ibmvnic_cpu_notif_add(struct ibmvnic_adapter *adapter)
344 {
345 	int ret;
346 
347 	ret = cpuhp_state_add_instance_nocalls(ibmvnic_online, &adapter->node);
348 	if (ret)
349 		return ret;
350 	ret = cpuhp_state_add_instance_nocalls(CPUHP_IBMVNIC_DEAD,
351 					       &adapter->node_dead);
352 	if (!ret)
353 		return ret;
354 	cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node);
355 	return ret;
356 }
357 
358 static void ibmvnic_cpu_notif_remove(struct ibmvnic_adapter *adapter)
359 {
360 	cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node);
361 	cpuhp_state_remove_instance_nocalls(CPUHP_IBMVNIC_DEAD,
362 					    &adapter->node_dead);
363 }
364 
365 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
366 			  unsigned long length, unsigned long *number,
367 			  unsigned long *irq)
368 {
369 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
370 	long rc;
371 
372 	rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
373 	*number = retbuf[0];
374 	*irq = retbuf[1];
375 
376 	return rc;
377 }
378 
379 /**
380  * ibmvnic_wait_for_completion - Check device state and wait for completion
381  * @adapter: private device data
382  * @comp_done: completion structure to wait for
383  * @timeout: time to wait in milliseconds
384  *
385  * Wait for a completion signal or until the timeout limit is reached
386  * while checking that the device is still active.
387  */
388 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
389 				       struct completion *comp_done,
390 				       unsigned long timeout)
391 {
392 	struct net_device *netdev;
393 	unsigned long div_timeout;
394 	u8 retry;
395 
396 	netdev = adapter->netdev;
397 	retry = 5;
398 	div_timeout = msecs_to_jiffies(timeout / retry);
399 	while (true) {
400 		if (!adapter->crq.active) {
401 			netdev_err(netdev, "Device down!\n");
402 			return -ENODEV;
403 		}
404 		if (!retry--)
405 			break;
406 		if (wait_for_completion_timeout(comp_done, div_timeout))
407 			return 0;
408 	}
409 	netdev_err(netdev, "Operation timed out.\n");
410 	return -ETIMEDOUT;
411 }
412 
413 /**
414  * reuse_ltb() - Check if a long term buffer can be reused
415  * @ltb:  The long term buffer to be checked
416  * @size: The size of the long term buffer.
417  *
418  * An LTB can be reused unless its size has changed.
419  *
420  * Return: Return true if the LTB can be reused, false otherwise.
421  */
422 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
423 {
424 	return (ltb->buff && ltb->size == size);
425 }
426 
427 /**
428  * alloc_long_term_buff() - Allocate a long term buffer (LTB)
429  *
430  * @adapter: ibmvnic adapter associated to the LTB
431  * @ltb:     container object for the LTB
432  * @size:    size of the LTB
433  *
434  * Allocate an LTB of the specified size and notify VIOS.
435  *
436  * If the given @ltb already has the correct size, reuse it. Otherwise if
437  * its non-NULL, free it. Then allocate a new one of the correct size.
438  * Notify the VIOS either way since we may now be working with a new VIOS.
439  *
440  * Allocating larger chunks of memory during resets, specially LPM or under
441  * low memory situations can cause resets to fail/timeout and for LPAR to
442  * lose connectivity. So hold onto the LTB even if we fail to communicate
443  * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
444  *
445  * Return: 0 if we were able to allocate the LTB and notify the VIOS and
446  *	   a negative value otherwise.
447  */
448 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
449 				struct ibmvnic_long_term_buff *ltb, int size)
450 {
451 	struct device *dev = &adapter->vdev->dev;
452 	u64 prev = 0;
453 	int rc;
454 
455 	if (!reuse_ltb(ltb, size)) {
456 		dev_dbg(dev,
457 			"LTB size changed from 0x%llx to 0x%x, reallocating\n",
458 			 ltb->size, size);
459 		prev = ltb->size;
460 		free_long_term_buff(adapter, ltb);
461 	}
462 
463 	if (ltb->buff) {
464 		dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
465 			ltb->map_id, ltb->size);
466 	} else {
467 		ltb->buff = dma_alloc_coherent(dev, size, &ltb->addr,
468 					       GFP_KERNEL);
469 		if (!ltb->buff) {
470 			dev_err(dev, "Couldn't alloc long term buffer\n");
471 			return -ENOMEM;
472 		}
473 		ltb->size = size;
474 
475 		ltb->map_id = find_first_zero_bit(adapter->map_ids,
476 						  MAX_MAP_ID);
477 		bitmap_set(adapter->map_ids, ltb->map_id, 1);
478 
479 		dev_dbg(dev,
480 			"Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n",
481 			 ltb->map_id, ltb->size, prev);
482 	}
483 
484 	/* Ensure ltb is zeroed - specially when reusing it. */
485 	memset(ltb->buff, 0, ltb->size);
486 
487 	mutex_lock(&adapter->fw_lock);
488 	adapter->fw_done_rc = 0;
489 	reinit_completion(&adapter->fw_done);
490 
491 	rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
492 	if (rc) {
493 		dev_err(dev, "send_request_map failed, rc = %d\n", rc);
494 		goto out;
495 	}
496 
497 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
498 	if (rc) {
499 		dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
500 			rc);
501 		goto out;
502 	}
503 
504 	if (adapter->fw_done_rc) {
505 		dev_err(dev, "Couldn't map LTB, rc = %d\n",
506 			adapter->fw_done_rc);
507 		rc = -EIO;
508 		goto out;
509 	}
510 	rc = 0;
511 out:
512 	/* don't free LTB on communication error - see function header */
513 	mutex_unlock(&adapter->fw_lock);
514 	return rc;
515 }
516 
517 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
518 				struct ibmvnic_long_term_buff *ltb)
519 {
520 	struct device *dev = &adapter->vdev->dev;
521 
522 	if (!ltb->buff)
523 		return;
524 
525 	/* VIOS automatically unmaps the long term buffer at remote
526 	 * end for the following resets:
527 	 * FAILOVER, MOBILITY, TIMEOUT.
528 	 */
529 	if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
530 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
531 	    adapter->reset_reason != VNIC_RESET_TIMEOUT)
532 		send_request_unmap(adapter, ltb->map_id);
533 
534 	dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
535 
536 	ltb->buff = NULL;
537 	/* mark this map_id free */
538 	bitmap_clear(adapter->map_ids, ltb->map_id, 1);
539 	ltb->map_id = 0;
540 }
541 
542 /**
543  * free_ltb_set - free the given set of long term buffers (LTBS)
544  * @adapter: The ibmvnic adapter containing this ltb set
545  * @ltb_set: The ltb_set to be freed
546  *
547  * Free the set of LTBs in the given set.
548  */
549 
550 static void free_ltb_set(struct ibmvnic_adapter *adapter,
551 			 struct ibmvnic_ltb_set *ltb_set)
552 {
553 	int i;
554 
555 	for (i = 0; i < ltb_set->num_ltbs; i++)
556 		free_long_term_buff(adapter, &ltb_set->ltbs[i]);
557 
558 	kfree(ltb_set->ltbs);
559 	ltb_set->ltbs = NULL;
560 	ltb_set->num_ltbs = 0;
561 }
562 
563 /**
564  * alloc_ltb_set() - Allocate a set of long term buffers (LTBs)
565  *
566  * @adapter: ibmvnic adapter associated to the LTB
567  * @ltb_set: container object for the set of LTBs
568  * @num_buffs: Number of buffers in the LTB
569  * @buff_size: Size of each buffer in the LTB
570  *
571  * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size
572  * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the
573  * new set of LTBs have fewer LTBs than the old set, free the excess LTBs.
574  * If new set needs more than in old set, allocate the remaining ones.
575  * Try and reuse as many LTBs as possible and avoid reallocation.
576  *
577  * Any changes to this allocation strategy must be reflected in
578  * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb().
579  */
580 static int alloc_ltb_set(struct ibmvnic_adapter *adapter,
581 			 struct ibmvnic_ltb_set *ltb_set, int num_buffs,
582 			 int buff_size)
583 {
584 	struct device *dev = &adapter->vdev->dev;
585 	struct ibmvnic_ltb_set old_set;
586 	struct ibmvnic_ltb_set new_set;
587 	int rem_size;
588 	int tot_size;		/* size of all ltbs */
589 	int ltb_size;		/* size of one ltb */
590 	int nltbs;
591 	int rc;
592 	int n;
593 	int i;
594 
595 	dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs,
596 		buff_size);
597 
598 	ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size);
599 	tot_size = num_buffs * buff_size;
600 
601 	if (ltb_size > tot_size)
602 		ltb_size = tot_size;
603 
604 	nltbs = tot_size / ltb_size;
605 	if (tot_size % ltb_size)
606 		nltbs++;
607 
608 	old_set = *ltb_set;
609 
610 	if (old_set.num_ltbs == nltbs) {
611 		new_set = old_set;
612 	} else {
613 		int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff);
614 
615 		new_set.ltbs = kzalloc(tmp, GFP_KERNEL);
616 		if (!new_set.ltbs)
617 			return -ENOMEM;
618 
619 		new_set.num_ltbs = nltbs;
620 
621 		/* Free any excess ltbs in old set */
622 		for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++)
623 			free_long_term_buff(adapter, &old_set.ltbs[i]);
624 
625 		/* Copy remaining ltbs to new set. All LTBs except the
626 		 * last one are of the same size. alloc_long_term_buff()
627 		 * will realloc if the size changes.
628 		 */
629 		n = min(old_set.num_ltbs, new_set.num_ltbs);
630 		for (i = 0; i < n; i++)
631 			new_set.ltbs[i] = old_set.ltbs[i];
632 
633 		/* Any additional ltbs in new set will have NULL ltbs for
634 		 * now and will be allocated in alloc_long_term_buff().
635 		 */
636 
637 		/* We no longer need the old_set so free it. Note that we
638 		 * may have reused some ltbs from old set and freed excess
639 		 * ltbs above. So we only need to free the container now
640 		 * not the LTBs themselves. (i.e. dont free_ltb_set()!)
641 		 */
642 		kfree(old_set.ltbs);
643 		old_set.ltbs = NULL;
644 		old_set.num_ltbs = 0;
645 
646 		/* Install the new set. If allocations fail below, we will
647 		 * retry later and know what size LTBs we need.
648 		 */
649 		*ltb_set = new_set;
650 	}
651 
652 	i = 0;
653 	rem_size = tot_size;
654 	while (rem_size) {
655 		if (ltb_size > rem_size)
656 			ltb_size = rem_size;
657 
658 		rem_size -= ltb_size;
659 
660 		rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size);
661 		if (rc)
662 			goto out;
663 		i++;
664 	}
665 
666 	WARN_ON(i != new_set.num_ltbs);
667 
668 	return 0;
669 out:
670 	/* We may have allocated one/more LTBs before failing and we
671 	 * want to try and reuse on next reset. So don't free ltb set.
672 	 */
673 	return rc;
674 }
675 
676 /**
677  * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB.
678  * @rxpool: The receive buffer pool containing buffer
679  * @bufidx: Index of buffer in rxpool
680  * @ltbp: (Output) pointer to the long term buffer containing the buffer
681  * @offset: (Output) offset of buffer in the LTB from @ltbp
682  *
683  * Map the given buffer identified by [rxpool, bufidx] to an LTB in the
684  * pool and its corresponding offset. Assume for now that each LTB is of
685  * different size but could possibly be optimized based on the allocation
686  * strategy in alloc_ltb_set().
687  */
688 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool,
689 				  unsigned int bufidx,
690 				  struct ibmvnic_long_term_buff **ltbp,
691 				  unsigned int *offset)
692 {
693 	struct ibmvnic_long_term_buff *ltb;
694 	int nbufs;	/* # of buffers in one ltb */
695 	int i;
696 
697 	WARN_ON(bufidx >= rxpool->size);
698 
699 	for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) {
700 		ltb = &rxpool->ltb_set.ltbs[i];
701 		nbufs = ltb->size / rxpool->buff_size;
702 		if (bufidx < nbufs)
703 			break;
704 		bufidx -= nbufs;
705 	}
706 
707 	*ltbp = ltb;
708 	*offset = bufidx * rxpool->buff_size;
709 }
710 
711 /**
712  * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB.
713  * @txpool: The transmit buffer pool containing buffer
714  * @bufidx: Index of buffer in txpool
715  * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer
716  * @offset: (Output) offset of buffer in the LTB from @ltbp
717  *
718  * Map the given buffer identified by [txpool, bufidx] to an LTB in the
719  * pool and its corresponding offset.
720  */
721 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool,
722 				  unsigned int bufidx,
723 				  struct ibmvnic_long_term_buff **ltbp,
724 				  unsigned int *offset)
725 {
726 	struct ibmvnic_long_term_buff *ltb;
727 	int nbufs;	/* # of buffers in one ltb */
728 	int i;
729 
730 	WARN_ON_ONCE(bufidx >= txpool->num_buffers);
731 
732 	for (i = 0; i < txpool->ltb_set.num_ltbs; i++) {
733 		ltb = &txpool->ltb_set.ltbs[i];
734 		nbufs = ltb->size / txpool->buf_size;
735 		if (bufidx < nbufs)
736 			break;
737 		bufidx -= nbufs;
738 	}
739 
740 	*ltbp = ltb;
741 	*offset = bufidx * txpool->buf_size;
742 }
743 
744 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
745 {
746 	int i;
747 
748 	for (i = 0; i < adapter->num_active_rx_pools; i++)
749 		adapter->rx_pool[i].active = 0;
750 }
751 
752 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
753 			      struct ibmvnic_rx_pool *pool)
754 {
755 	int count = pool->size - atomic_read(&pool->available);
756 	u64 handle = adapter->rx_scrq[pool->index]->handle;
757 	struct device *dev = &adapter->vdev->dev;
758 	struct ibmvnic_ind_xmit_queue *ind_bufp;
759 	struct ibmvnic_sub_crq_queue *rx_scrq;
760 	struct ibmvnic_long_term_buff *ltb;
761 	union sub_crq *sub_crq;
762 	int buffers_added = 0;
763 	unsigned long lpar_rc;
764 	struct sk_buff *skb;
765 	unsigned int offset;
766 	dma_addr_t dma_addr;
767 	unsigned char *dst;
768 	int shift = 0;
769 	int bufidx;
770 	int i;
771 
772 	if (!pool->active)
773 		return;
774 
775 	rx_scrq = adapter->rx_scrq[pool->index];
776 	ind_bufp = &rx_scrq->ind_buf;
777 
778 	/* netdev_skb_alloc() could have failed after we saved a few skbs
779 	 * in the indir_buf and we would not have sent them to VIOS yet.
780 	 * To account for them, start the loop at ind_bufp->index rather
781 	 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
782 	 * be 0.
783 	 */
784 	for (i = ind_bufp->index; i < count; ++i) {
785 		bufidx = pool->free_map[pool->next_free];
786 
787 		/* We maybe reusing the skb from earlier resets. Allocate
788 		 * only if necessary. But since the LTB may have changed
789 		 * during reset (see init_rx_pools()), update LTB below
790 		 * even if reusing skb.
791 		 */
792 		skb = pool->rx_buff[bufidx].skb;
793 		if (!skb) {
794 			skb = netdev_alloc_skb(adapter->netdev,
795 					       pool->buff_size);
796 			if (!skb) {
797 				dev_err(dev, "Couldn't replenish rx buff\n");
798 				adapter->replenish_no_mem++;
799 				break;
800 			}
801 		}
802 
803 		pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
804 		pool->next_free = (pool->next_free + 1) % pool->size;
805 
806 		/* Copy the skb to the long term mapped DMA buffer */
807 		map_rxpool_buf_to_ltb(pool, bufidx, &ltb, &offset);
808 		dst = ltb->buff + offset;
809 		memset(dst, 0, pool->buff_size);
810 		dma_addr = ltb->addr + offset;
811 
812 		/* add the skb to an rx_buff in the pool */
813 		pool->rx_buff[bufidx].data = dst;
814 		pool->rx_buff[bufidx].dma = dma_addr;
815 		pool->rx_buff[bufidx].skb = skb;
816 		pool->rx_buff[bufidx].pool_index = pool->index;
817 		pool->rx_buff[bufidx].size = pool->buff_size;
818 
819 		/* queue the rx_buff for the next send_subcrq_indirect */
820 		sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
821 		memset(sub_crq, 0, sizeof(*sub_crq));
822 		sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
823 		sub_crq->rx_add.correlator =
824 		    cpu_to_be64((u64)&pool->rx_buff[bufidx]);
825 		sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
826 		sub_crq->rx_add.map_id = ltb->map_id;
827 
828 		/* The length field of the sCRQ is defined to be 24 bits so the
829 		 * buffer size needs to be left shifted by a byte before it is
830 		 * converted to big endian to prevent the last byte from being
831 		 * truncated.
832 		 */
833 #ifdef __LITTLE_ENDIAN__
834 		shift = 8;
835 #endif
836 		sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
837 
838 		/* if send_subcrq_indirect queue is full, flush to VIOS */
839 		if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
840 		    i == count - 1) {
841 			lpar_rc =
842 				send_subcrq_indirect(adapter, handle,
843 						     (u64)ind_bufp->indir_dma,
844 						     (u64)ind_bufp->index);
845 			if (lpar_rc != H_SUCCESS)
846 				goto failure;
847 			buffers_added += ind_bufp->index;
848 			adapter->replenish_add_buff_success += ind_bufp->index;
849 			ind_bufp->index = 0;
850 		}
851 	}
852 	atomic_add(buffers_added, &pool->available);
853 	return;
854 
855 failure:
856 	if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
857 		dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
858 	for (i = ind_bufp->index - 1; i >= 0; --i) {
859 		struct ibmvnic_rx_buff *rx_buff;
860 
861 		pool->next_free = pool->next_free == 0 ?
862 				  pool->size - 1 : pool->next_free - 1;
863 		sub_crq = &ind_bufp->indir_arr[i];
864 		rx_buff = (struct ibmvnic_rx_buff *)
865 				be64_to_cpu(sub_crq->rx_add.correlator);
866 		bufidx = (int)(rx_buff - pool->rx_buff);
867 		pool->free_map[pool->next_free] = bufidx;
868 		dev_kfree_skb_any(pool->rx_buff[bufidx].skb);
869 		pool->rx_buff[bufidx].skb = NULL;
870 	}
871 	adapter->replenish_add_buff_failure += ind_bufp->index;
872 	atomic_add(buffers_added, &pool->available);
873 	ind_bufp->index = 0;
874 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
875 		/* Disable buffer pool replenishment and report carrier off if
876 		 * queue is closed or pending failover.
877 		 * Firmware guarantees that a signal will be sent to the
878 		 * driver, triggering a reset.
879 		 */
880 		deactivate_rx_pools(adapter);
881 		netif_carrier_off(adapter->netdev);
882 	}
883 }
884 
885 static void replenish_pools(struct ibmvnic_adapter *adapter)
886 {
887 	int i;
888 
889 	adapter->replenish_task_cycles++;
890 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
891 		if (adapter->rx_pool[i].active)
892 			replenish_rx_pool(adapter, &adapter->rx_pool[i]);
893 	}
894 
895 	netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
896 }
897 
898 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
899 {
900 	kfree(adapter->tx_stats_buffers);
901 	kfree(adapter->rx_stats_buffers);
902 	adapter->tx_stats_buffers = NULL;
903 	adapter->rx_stats_buffers = NULL;
904 }
905 
906 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
907 {
908 	adapter->tx_stats_buffers =
909 				kcalloc(IBMVNIC_MAX_QUEUES,
910 					sizeof(struct ibmvnic_tx_queue_stats),
911 					GFP_KERNEL);
912 	if (!adapter->tx_stats_buffers)
913 		return -ENOMEM;
914 
915 	adapter->rx_stats_buffers =
916 				kcalloc(IBMVNIC_MAX_QUEUES,
917 					sizeof(struct ibmvnic_rx_queue_stats),
918 					GFP_KERNEL);
919 	if (!adapter->rx_stats_buffers)
920 		return -ENOMEM;
921 
922 	return 0;
923 }
924 
925 static void release_stats_token(struct ibmvnic_adapter *adapter)
926 {
927 	struct device *dev = &adapter->vdev->dev;
928 
929 	if (!adapter->stats_token)
930 		return;
931 
932 	dma_unmap_single(dev, adapter->stats_token,
933 			 sizeof(struct ibmvnic_statistics),
934 			 DMA_FROM_DEVICE);
935 	adapter->stats_token = 0;
936 }
937 
938 static int init_stats_token(struct ibmvnic_adapter *adapter)
939 {
940 	struct device *dev = &adapter->vdev->dev;
941 	dma_addr_t stok;
942 	int rc;
943 
944 	stok = dma_map_single(dev, &adapter->stats,
945 			      sizeof(struct ibmvnic_statistics),
946 			      DMA_FROM_DEVICE);
947 	rc = dma_mapping_error(dev, stok);
948 	if (rc) {
949 		dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
950 		return rc;
951 	}
952 
953 	adapter->stats_token = stok;
954 	netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
955 	return 0;
956 }
957 
958 /**
959  * release_rx_pools() - Release any rx pools attached to @adapter.
960  * @adapter: ibmvnic adapter
961  *
962  * Safe to call this multiple times - even if no pools are attached.
963  */
964 static void release_rx_pools(struct ibmvnic_adapter *adapter)
965 {
966 	struct ibmvnic_rx_pool *rx_pool;
967 	int i, j;
968 
969 	if (!adapter->rx_pool)
970 		return;
971 
972 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
973 		rx_pool = &adapter->rx_pool[i];
974 
975 		netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
976 
977 		kfree(rx_pool->free_map);
978 
979 		free_ltb_set(adapter, &rx_pool->ltb_set);
980 
981 		if (!rx_pool->rx_buff)
982 			continue;
983 
984 		for (j = 0; j < rx_pool->size; j++) {
985 			if (rx_pool->rx_buff[j].skb) {
986 				dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
987 				rx_pool->rx_buff[j].skb = NULL;
988 			}
989 		}
990 
991 		kfree(rx_pool->rx_buff);
992 	}
993 
994 	kfree(adapter->rx_pool);
995 	adapter->rx_pool = NULL;
996 	adapter->num_active_rx_pools = 0;
997 	adapter->prev_rx_pool_size = 0;
998 }
999 
1000 /**
1001  * reuse_rx_pools() - Check if the existing rx pools can be reused.
1002  * @adapter: ibmvnic adapter
1003  *
1004  * Check if the existing rx pools in the adapter can be reused. The
1005  * pools can be reused if the pool parameters (number of pools,
1006  * number of buffers in the pool and size of each buffer) have not
1007  * changed.
1008  *
1009  * NOTE: This assumes that all pools have the same number of buffers
1010  *       which is the case currently. If that changes, we must fix this.
1011  *
1012  * Return: true if the rx pools can be reused, false otherwise.
1013  */
1014 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
1015 {
1016 	u64 old_num_pools, new_num_pools;
1017 	u64 old_pool_size, new_pool_size;
1018 	u64 old_buff_size, new_buff_size;
1019 
1020 	if (!adapter->rx_pool)
1021 		return false;
1022 
1023 	old_num_pools = adapter->num_active_rx_pools;
1024 	new_num_pools = adapter->req_rx_queues;
1025 
1026 	old_pool_size = adapter->prev_rx_pool_size;
1027 	new_pool_size = adapter->req_rx_add_entries_per_subcrq;
1028 
1029 	old_buff_size = adapter->prev_rx_buf_sz;
1030 	new_buff_size = adapter->cur_rx_buf_sz;
1031 
1032 	if (old_buff_size != new_buff_size ||
1033 	    old_num_pools != new_num_pools ||
1034 	    old_pool_size != new_pool_size)
1035 		return false;
1036 
1037 	return true;
1038 }
1039 
1040 /**
1041  * init_rx_pools(): Initialize the set of receiver pools in the adapter.
1042  * @netdev: net device associated with the vnic interface
1043  *
1044  * Initialize the set of receiver pools in the ibmvnic adapter associated
1045  * with the net_device @netdev. If possible, reuse the existing rx pools.
1046  * Otherwise free any existing pools and  allocate a new set of pools
1047  * before initializing them.
1048  *
1049  * Return: 0 on success and negative value on error.
1050  */
1051 static int init_rx_pools(struct net_device *netdev)
1052 {
1053 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1054 	struct device *dev = &adapter->vdev->dev;
1055 	struct ibmvnic_rx_pool *rx_pool;
1056 	u64 num_pools;
1057 	u64 pool_size;		/* # of buffers in one pool */
1058 	u64 buff_size;
1059 	int i, j, rc;
1060 
1061 	pool_size = adapter->req_rx_add_entries_per_subcrq;
1062 	num_pools = adapter->req_rx_queues;
1063 	buff_size = adapter->cur_rx_buf_sz;
1064 
1065 	if (reuse_rx_pools(adapter)) {
1066 		dev_dbg(dev, "Reusing rx pools\n");
1067 		goto update_ltb;
1068 	}
1069 
1070 	/* Allocate/populate the pools. */
1071 	release_rx_pools(adapter);
1072 
1073 	adapter->rx_pool = kcalloc(num_pools,
1074 				   sizeof(struct ibmvnic_rx_pool),
1075 				   GFP_KERNEL);
1076 	if (!adapter->rx_pool) {
1077 		dev_err(dev, "Failed to allocate rx pools\n");
1078 		return -ENOMEM;
1079 	}
1080 
1081 	/* Set num_active_rx_pools early. If we fail below after partial
1082 	 * allocation, release_rx_pools() will know how many to look for.
1083 	 */
1084 	adapter->num_active_rx_pools = num_pools;
1085 
1086 	for (i = 0; i < num_pools; i++) {
1087 		rx_pool = &adapter->rx_pool[i];
1088 
1089 		netdev_dbg(adapter->netdev,
1090 			   "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
1091 			   i, pool_size, buff_size);
1092 
1093 		rx_pool->size = pool_size;
1094 		rx_pool->index = i;
1095 		rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1096 
1097 		rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
1098 					    GFP_KERNEL);
1099 		if (!rx_pool->free_map) {
1100 			dev_err(dev, "Couldn't alloc free_map %d\n", i);
1101 			rc = -ENOMEM;
1102 			goto out_release;
1103 		}
1104 
1105 		rx_pool->rx_buff = kcalloc(rx_pool->size,
1106 					   sizeof(struct ibmvnic_rx_buff),
1107 					   GFP_KERNEL);
1108 		if (!rx_pool->rx_buff) {
1109 			dev_err(dev, "Couldn't alloc rx buffers\n");
1110 			rc = -ENOMEM;
1111 			goto out_release;
1112 		}
1113 	}
1114 
1115 	adapter->prev_rx_pool_size = pool_size;
1116 	adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
1117 
1118 update_ltb:
1119 	for (i = 0; i < num_pools; i++) {
1120 		rx_pool = &adapter->rx_pool[i];
1121 		dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
1122 			i, rx_pool->size, rx_pool->buff_size);
1123 
1124 		rc = alloc_ltb_set(adapter, &rx_pool->ltb_set,
1125 				   rx_pool->size, rx_pool->buff_size);
1126 		if (rc)
1127 			goto out;
1128 
1129 		for (j = 0; j < rx_pool->size; ++j) {
1130 			struct ibmvnic_rx_buff *rx_buff;
1131 
1132 			rx_pool->free_map[j] = j;
1133 
1134 			/* NOTE: Don't clear rx_buff->skb here - will leak
1135 			 * memory! replenish_rx_pool() will reuse skbs or
1136 			 * allocate as necessary.
1137 			 */
1138 			rx_buff = &rx_pool->rx_buff[j];
1139 			rx_buff->dma = 0;
1140 			rx_buff->data = 0;
1141 			rx_buff->size = 0;
1142 			rx_buff->pool_index = 0;
1143 		}
1144 
1145 		/* Mark pool "empty" so replenish_rx_pools() will
1146 		 * update the LTB info for each buffer
1147 		 */
1148 		atomic_set(&rx_pool->available, 0);
1149 		rx_pool->next_alloc = 0;
1150 		rx_pool->next_free = 0;
1151 		/* replenish_rx_pool() may have called deactivate_rx_pools()
1152 		 * on failover. Ensure pool is active now.
1153 		 */
1154 		rx_pool->active = 1;
1155 	}
1156 	return 0;
1157 out_release:
1158 	release_rx_pools(adapter);
1159 out:
1160 	/* We failed to allocate one or more LTBs or map them on the VIOS.
1161 	 * Hold onto the pools and any LTBs that we did allocate/map.
1162 	 */
1163 	return rc;
1164 }
1165 
1166 static void release_vpd_data(struct ibmvnic_adapter *adapter)
1167 {
1168 	if (!adapter->vpd)
1169 		return;
1170 
1171 	kfree(adapter->vpd->buff);
1172 	kfree(adapter->vpd);
1173 
1174 	adapter->vpd = NULL;
1175 }
1176 
1177 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
1178 				struct ibmvnic_tx_pool *tx_pool)
1179 {
1180 	kfree(tx_pool->tx_buff);
1181 	kfree(tx_pool->free_map);
1182 	free_ltb_set(adapter, &tx_pool->ltb_set);
1183 }
1184 
1185 /**
1186  * release_tx_pools() - Release any tx pools attached to @adapter.
1187  * @adapter: ibmvnic adapter
1188  *
1189  * Safe to call this multiple times - even if no pools are attached.
1190  */
1191 static void release_tx_pools(struct ibmvnic_adapter *adapter)
1192 {
1193 	int i;
1194 
1195 	/* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
1196 	 * both NULL or both non-NULL. So we only need to check one.
1197 	 */
1198 	if (!adapter->tx_pool)
1199 		return;
1200 
1201 	for (i = 0; i < adapter->num_active_tx_pools; i++) {
1202 		release_one_tx_pool(adapter, &adapter->tx_pool[i]);
1203 		release_one_tx_pool(adapter, &adapter->tso_pool[i]);
1204 	}
1205 
1206 	kfree(adapter->tx_pool);
1207 	adapter->tx_pool = NULL;
1208 	kfree(adapter->tso_pool);
1209 	adapter->tso_pool = NULL;
1210 	adapter->num_active_tx_pools = 0;
1211 	adapter->prev_tx_pool_size = 0;
1212 }
1213 
1214 static int init_one_tx_pool(struct net_device *netdev,
1215 			    struct ibmvnic_tx_pool *tx_pool,
1216 			    int pool_size, int buf_size)
1217 {
1218 	int i;
1219 
1220 	tx_pool->tx_buff = kcalloc(pool_size,
1221 				   sizeof(struct ibmvnic_tx_buff),
1222 				   GFP_KERNEL);
1223 	if (!tx_pool->tx_buff)
1224 		return -ENOMEM;
1225 
1226 	tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
1227 	if (!tx_pool->free_map) {
1228 		kfree(tx_pool->tx_buff);
1229 		tx_pool->tx_buff = NULL;
1230 		return -ENOMEM;
1231 	}
1232 
1233 	for (i = 0; i < pool_size; i++)
1234 		tx_pool->free_map[i] = i;
1235 
1236 	tx_pool->consumer_index = 0;
1237 	tx_pool->producer_index = 0;
1238 	tx_pool->num_buffers = pool_size;
1239 	tx_pool->buf_size = buf_size;
1240 
1241 	return 0;
1242 }
1243 
1244 /**
1245  * reuse_tx_pools() - Check if the existing tx pools can be reused.
1246  * @adapter: ibmvnic adapter
1247  *
1248  * Check if the existing tx pools in the adapter can be reused. The
1249  * pools can be reused if the pool parameters (number of pools,
1250  * number of buffers in the pool and mtu) have not changed.
1251  *
1252  * NOTE: This assumes that all pools have the same number of buffers
1253  *       which is the case currently. If that changes, we must fix this.
1254  *
1255  * Return: true if the tx pools can be reused, false otherwise.
1256  */
1257 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
1258 {
1259 	u64 old_num_pools, new_num_pools;
1260 	u64 old_pool_size, new_pool_size;
1261 	u64 old_mtu, new_mtu;
1262 
1263 	if (!adapter->tx_pool)
1264 		return false;
1265 
1266 	old_num_pools = adapter->num_active_tx_pools;
1267 	new_num_pools = adapter->num_active_tx_scrqs;
1268 	old_pool_size = adapter->prev_tx_pool_size;
1269 	new_pool_size = adapter->req_tx_entries_per_subcrq;
1270 	old_mtu = adapter->prev_mtu;
1271 	new_mtu = adapter->req_mtu;
1272 
1273 	if (old_mtu != new_mtu ||
1274 	    old_num_pools != new_num_pools ||
1275 	    old_pool_size != new_pool_size)
1276 		return false;
1277 
1278 	return true;
1279 }
1280 
1281 /**
1282  * init_tx_pools(): Initialize the set of transmit pools in the adapter.
1283  * @netdev: net device associated with the vnic interface
1284  *
1285  * Initialize the set of transmit pools in the ibmvnic adapter associated
1286  * with the net_device @netdev. If possible, reuse the existing tx pools.
1287  * Otherwise free any existing pools and  allocate a new set of pools
1288  * before initializing them.
1289  *
1290  * Return: 0 on success and negative value on error.
1291  */
1292 static int init_tx_pools(struct net_device *netdev)
1293 {
1294 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1295 	struct device *dev = &adapter->vdev->dev;
1296 	int num_pools;
1297 	u64 pool_size;		/* # of buffers in pool */
1298 	u64 buff_size;
1299 	int i, j, rc;
1300 
1301 	num_pools = adapter->req_tx_queues;
1302 
1303 	/* We must notify the VIOS about the LTB on all resets - but we only
1304 	 * need to alloc/populate pools if either the number of buffers or
1305 	 * size of each buffer in the pool has changed.
1306 	 */
1307 	if (reuse_tx_pools(adapter)) {
1308 		netdev_dbg(netdev, "Reusing tx pools\n");
1309 		goto update_ltb;
1310 	}
1311 
1312 	/* Allocate/populate the pools. */
1313 	release_tx_pools(adapter);
1314 
1315 	pool_size = adapter->req_tx_entries_per_subcrq;
1316 	num_pools = adapter->num_active_tx_scrqs;
1317 
1318 	adapter->tx_pool = kcalloc(num_pools,
1319 				   sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1320 	if (!adapter->tx_pool)
1321 		return -ENOMEM;
1322 
1323 	adapter->tso_pool = kcalloc(num_pools,
1324 				    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1325 	/* To simplify release_tx_pools() ensure that ->tx_pool and
1326 	 * ->tso_pool are either both NULL or both non-NULL.
1327 	 */
1328 	if (!adapter->tso_pool) {
1329 		kfree(adapter->tx_pool);
1330 		adapter->tx_pool = NULL;
1331 		return -ENOMEM;
1332 	}
1333 
1334 	/* Set num_active_tx_pools early. If we fail below after partial
1335 	 * allocation, release_tx_pools() will know how many to look for.
1336 	 */
1337 	adapter->num_active_tx_pools = num_pools;
1338 
1339 	buff_size = adapter->req_mtu + VLAN_HLEN;
1340 	buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1341 
1342 	for (i = 0; i < num_pools; i++) {
1343 		dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
1344 			i, adapter->req_tx_entries_per_subcrq, buff_size);
1345 
1346 		rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
1347 				      pool_size, buff_size);
1348 		if (rc)
1349 			goto out_release;
1350 
1351 		rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
1352 				      IBMVNIC_TSO_BUFS,
1353 				      IBMVNIC_TSO_BUF_SZ);
1354 		if (rc)
1355 			goto out_release;
1356 	}
1357 
1358 	adapter->prev_tx_pool_size = pool_size;
1359 	adapter->prev_mtu = adapter->req_mtu;
1360 
1361 update_ltb:
1362 	/* NOTE: All tx_pools have the same number of buffers (which is
1363 	 *       same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
1364 	 *       buffers (see calls init_one_tx_pool() for these).
1365 	 *       For consistency, we use tx_pool->num_buffers and
1366 	 *       tso_pool->num_buffers below.
1367 	 */
1368 	rc = -1;
1369 	for (i = 0; i < num_pools; i++) {
1370 		struct ibmvnic_tx_pool *tso_pool;
1371 		struct ibmvnic_tx_pool *tx_pool;
1372 
1373 		tx_pool = &adapter->tx_pool[i];
1374 
1375 		dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n",
1376 			i, tx_pool->num_buffers, tx_pool->buf_size);
1377 
1378 		rc = alloc_ltb_set(adapter, &tx_pool->ltb_set,
1379 				   tx_pool->num_buffers, tx_pool->buf_size);
1380 		if (rc)
1381 			goto out;
1382 
1383 		tx_pool->consumer_index = 0;
1384 		tx_pool->producer_index = 0;
1385 
1386 		for (j = 0; j < tx_pool->num_buffers; j++)
1387 			tx_pool->free_map[j] = j;
1388 
1389 		tso_pool = &adapter->tso_pool[i];
1390 
1391 		dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n",
1392 			i, tso_pool->num_buffers, tso_pool->buf_size);
1393 
1394 		rc = alloc_ltb_set(adapter, &tso_pool->ltb_set,
1395 				   tso_pool->num_buffers, tso_pool->buf_size);
1396 		if (rc)
1397 			goto out;
1398 
1399 		tso_pool->consumer_index = 0;
1400 		tso_pool->producer_index = 0;
1401 
1402 		for (j = 0; j < tso_pool->num_buffers; j++)
1403 			tso_pool->free_map[j] = j;
1404 	}
1405 
1406 	return 0;
1407 out_release:
1408 	release_tx_pools(adapter);
1409 out:
1410 	/* We failed to allocate one or more LTBs or map them on the VIOS.
1411 	 * Hold onto the pools and any LTBs that we did allocate/map.
1412 	 */
1413 	return rc;
1414 }
1415 
1416 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1417 {
1418 	int i;
1419 
1420 	if (adapter->napi_enabled)
1421 		return;
1422 
1423 	for (i = 0; i < adapter->req_rx_queues; i++)
1424 		napi_enable(&adapter->napi[i]);
1425 
1426 	adapter->napi_enabled = true;
1427 }
1428 
1429 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1430 {
1431 	int i;
1432 
1433 	if (!adapter->napi_enabled)
1434 		return;
1435 
1436 	for (i = 0; i < adapter->req_rx_queues; i++) {
1437 		netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1438 		napi_disable(&adapter->napi[i]);
1439 	}
1440 
1441 	adapter->napi_enabled = false;
1442 }
1443 
1444 static int init_napi(struct ibmvnic_adapter *adapter)
1445 {
1446 	int i;
1447 
1448 	adapter->napi = kcalloc(adapter->req_rx_queues,
1449 				sizeof(struct napi_struct), GFP_KERNEL);
1450 	if (!adapter->napi)
1451 		return -ENOMEM;
1452 
1453 	for (i = 0; i < adapter->req_rx_queues; i++) {
1454 		netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1455 		netif_napi_add(adapter->netdev, &adapter->napi[i],
1456 			       ibmvnic_poll);
1457 	}
1458 
1459 	adapter->num_active_rx_napi = adapter->req_rx_queues;
1460 	return 0;
1461 }
1462 
1463 static void release_napi(struct ibmvnic_adapter *adapter)
1464 {
1465 	int i;
1466 
1467 	if (!adapter->napi)
1468 		return;
1469 
1470 	for (i = 0; i < adapter->num_active_rx_napi; i++) {
1471 		netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1472 		netif_napi_del(&adapter->napi[i]);
1473 	}
1474 
1475 	kfree(adapter->napi);
1476 	adapter->napi = NULL;
1477 	adapter->num_active_rx_napi = 0;
1478 	adapter->napi_enabled = false;
1479 }
1480 
1481 static const char *adapter_state_to_string(enum vnic_state state)
1482 {
1483 	switch (state) {
1484 	case VNIC_PROBING:
1485 		return "PROBING";
1486 	case VNIC_PROBED:
1487 		return "PROBED";
1488 	case VNIC_OPENING:
1489 		return "OPENING";
1490 	case VNIC_OPEN:
1491 		return "OPEN";
1492 	case VNIC_CLOSING:
1493 		return "CLOSING";
1494 	case VNIC_CLOSED:
1495 		return "CLOSED";
1496 	case VNIC_REMOVING:
1497 		return "REMOVING";
1498 	case VNIC_REMOVED:
1499 		return "REMOVED";
1500 	case VNIC_DOWN:
1501 		return "DOWN";
1502 	}
1503 	return "UNKNOWN";
1504 }
1505 
1506 static int ibmvnic_login(struct net_device *netdev)
1507 {
1508 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1509 	unsigned long timeout = msecs_to_jiffies(20000);
1510 	int retry_count = 0;
1511 	int retries = 10;
1512 	bool retry;
1513 	int rc;
1514 
1515 	do {
1516 		retry = false;
1517 		if (retry_count > retries) {
1518 			netdev_warn(netdev, "Login attempts exceeded\n");
1519 			return -EACCES;
1520 		}
1521 
1522 		adapter->init_done_rc = 0;
1523 		reinit_completion(&adapter->init_done);
1524 		rc = send_login(adapter);
1525 		if (rc)
1526 			return rc;
1527 
1528 		if (!wait_for_completion_timeout(&adapter->init_done,
1529 						 timeout)) {
1530 			netdev_warn(netdev, "Login timed out, retrying...\n");
1531 			retry = true;
1532 			adapter->init_done_rc = 0;
1533 			retry_count++;
1534 			continue;
1535 		}
1536 
1537 		if (adapter->init_done_rc == ABORTED) {
1538 			netdev_warn(netdev, "Login aborted, retrying...\n");
1539 			retry = true;
1540 			adapter->init_done_rc = 0;
1541 			retry_count++;
1542 			/* FW or device may be busy, so
1543 			 * wait a bit before retrying login
1544 			 */
1545 			msleep(500);
1546 		} else if (adapter->init_done_rc == PARTIALSUCCESS) {
1547 			retry_count++;
1548 			release_sub_crqs(adapter, 1);
1549 
1550 			retry = true;
1551 			netdev_dbg(netdev,
1552 				   "Received partial success, retrying...\n");
1553 			adapter->init_done_rc = 0;
1554 			reinit_completion(&adapter->init_done);
1555 			send_query_cap(adapter);
1556 			if (!wait_for_completion_timeout(&adapter->init_done,
1557 							 timeout)) {
1558 				netdev_warn(netdev,
1559 					    "Capabilities query timed out\n");
1560 				return -ETIMEDOUT;
1561 			}
1562 
1563 			rc = init_sub_crqs(adapter);
1564 			if (rc) {
1565 				netdev_warn(netdev,
1566 					    "SCRQ initialization failed\n");
1567 				return rc;
1568 			}
1569 
1570 			rc = init_sub_crq_irqs(adapter);
1571 			if (rc) {
1572 				netdev_warn(netdev,
1573 					    "SCRQ irq initialization failed\n");
1574 				return rc;
1575 			}
1576 		} else if (adapter->init_done_rc) {
1577 			netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
1578 				    adapter->init_done_rc);
1579 			return -EIO;
1580 		}
1581 	} while (retry);
1582 
1583 	__ibmvnic_set_mac(netdev, adapter->mac_addr);
1584 
1585 	netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1586 	return 0;
1587 }
1588 
1589 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1590 {
1591 	kfree(adapter->login_buf);
1592 	adapter->login_buf = NULL;
1593 }
1594 
1595 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1596 {
1597 	kfree(adapter->login_rsp_buf);
1598 	adapter->login_rsp_buf = NULL;
1599 }
1600 
1601 static void release_resources(struct ibmvnic_adapter *adapter)
1602 {
1603 	release_vpd_data(adapter);
1604 
1605 	release_napi(adapter);
1606 	release_login_buffer(adapter);
1607 	release_login_rsp_buffer(adapter);
1608 }
1609 
1610 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1611 {
1612 	struct net_device *netdev = adapter->netdev;
1613 	unsigned long timeout = msecs_to_jiffies(20000);
1614 	union ibmvnic_crq crq;
1615 	bool resend;
1616 	int rc;
1617 
1618 	netdev_dbg(netdev, "setting link state %d\n", link_state);
1619 
1620 	memset(&crq, 0, sizeof(crq));
1621 	crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1622 	crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1623 	crq.logical_link_state.link_state = link_state;
1624 
1625 	do {
1626 		resend = false;
1627 
1628 		reinit_completion(&adapter->init_done);
1629 		rc = ibmvnic_send_crq(adapter, &crq);
1630 		if (rc) {
1631 			netdev_err(netdev, "Failed to set link state\n");
1632 			return rc;
1633 		}
1634 
1635 		if (!wait_for_completion_timeout(&adapter->init_done,
1636 						 timeout)) {
1637 			netdev_err(netdev, "timeout setting link state\n");
1638 			return -ETIMEDOUT;
1639 		}
1640 
1641 		if (adapter->init_done_rc == PARTIALSUCCESS) {
1642 			/* Partuial success, delay and re-send */
1643 			mdelay(1000);
1644 			resend = true;
1645 		} else if (adapter->init_done_rc) {
1646 			netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1647 				    adapter->init_done_rc);
1648 			return adapter->init_done_rc;
1649 		}
1650 	} while (resend);
1651 
1652 	return 0;
1653 }
1654 
1655 static int set_real_num_queues(struct net_device *netdev)
1656 {
1657 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1658 	int rc;
1659 
1660 	netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1661 		   adapter->req_tx_queues, adapter->req_rx_queues);
1662 
1663 	rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1664 	if (rc) {
1665 		netdev_err(netdev, "failed to set the number of tx queues\n");
1666 		return rc;
1667 	}
1668 
1669 	rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1670 	if (rc)
1671 		netdev_err(netdev, "failed to set the number of rx queues\n");
1672 
1673 	return rc;
1674 }
1675 
1676 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1677 {
1678 	struct device *dev = &adapter->vdev->dev;
1679 	union ibmvnic_crq crq;
1680 	int len = 0;
1681 	int rc;
1682 
1683 	if (adapter->vpd->buff)
1684 		len = adapter->vpd->len;
1685 
1686 	mutex_lock(&adapter->fw_lock);
1687 	adapter->fw_done_rc = 0;
1688 	reinit_completion(&adapter->fw_done);
1689 
1690 	crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1691 	crq.get_vpd_size.cmd = GET_VPD_SIZE;
1692 	rc = ibmvnic_send_crq(adapter, &crq);
1693 	if (rc) {
1694 		mutex_unlock(&adapter->fw_lock);
1695 		return rc;
1696 	}
1697 
1698 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1699 	if (rc) {
1700 		dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1701 		mutex_unlock(&adapter->fw_lock);
1702 		return rc;
1703 	}
1704 	mutex_unlock(&adapter->fw_lock);
1705 
1706 	if (!adapter->vpd->len)
1707 		return -ENODATA;
1708 
1709 	if (!adapter->vpd->buff)
1710 		adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1711 	else if (adapter->vpd->len != len)
1712 		adapter->vpd->buff =
1713 			krealloc(adapter->vpd->buff,
1714 				 adapter->vpd->len, GFP_KERNEL);
1715 
1716 	if (!adapter->vpd->buff) {
1717 		dev_err(dev, "Could allocate VPD buffer\n");
1718 		return -ENOMEM;
1719 	}
1720 
1721 	adapter->vpd->dma_addr =
1722 		dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1723 			       DMA_FROM_DEVICE);
1724 	if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1725 		dev_err(dev, "Could not map VPD buffer\n");
1726 		kfree(adapter->vpd->buff);
1727 		adapter->vpd->buff = NULL;
1728 		return -ENOMEM;
1729 	}
1730 
1731 	mutex_lock(&adapter->fw_lock);
1732 	adapter->fw_done_rc = 0;
1733 	reinit_completion(&adapter->fw_done);
1734 
1735 	crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1736 	crq.get_vpd.cmd = GET_VPD;
1737 	crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1738 	crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1739 	rc = ibmvnic_send_crq(adapter, &crq);
1740 	if (rc) {
1741 		kfree(adapter->vpd->buff);
1742 		adapter->vpd->buff = NULL;
1743 		mutex_unlock(&adapter->fw_lock);
1744 		return rc;
1745 	}
1746 
1747 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1748 	if (rc) {
1749 		dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1750 		kfree(adapter->vpd->buff);
1751 		adapter->vpd->buff = NULL;
1752 		mutex_unlock(&adapter->fw_lock);
1753 		return rc;
1754 	}
1755 
1756 	mutex_unlock(&adapter->fw_lock);
1757 	return 0;
1758 }
1759 
1760 static int init_resources(struct ibmvnic_adapter *adapter)
1761 {
1762 	struct net_device *netdev = adapter->netdev;
1763 	int rc;
1764 
1765 	rc = set_real_num_queues(netdev);
1766 	if (rc)
1767 		return rc;
1768 
1769 	adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1770 	if (!adapter->vpd)
1771 		return -ENOMEM;
1772 
1773 	/* Vital Product Data (VPD) */
1774 	rc = ibmvnic_get_vpd(adapter);
1775 	if (rc) {
1776 		netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1777 		return rc;
1778 	}
1779 
1780 	rc = init_napi(adapter);
1781 	if (rc)
1782 		return rc;
1783 
1784 	send_query_map(adapter);
1785 
1786 	rc = init_rx_pools(netdev);
1787 	if (rc)
1788 		return rc;
1789 
1790 	rc = init_tx_pools(netdev);
1791 	return rc;
1792 }
1793 
1794 static int __ibmvnic_open(struct net_device *netdev)
1795 {
1796 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1797 	enum vnic_state prev_state = adapter->state;
1798 	int i, rc;
1799 
1800 	adapter->state = VNIC_OPENING;
1801 	replenish_pools(adapter);
1802 	ibmvnic_napi_enable(adapter);
1803 
1804 	/* We're ready to receive frames, enable the sub-crq interrupts and
1805 	 * set the logical link state to up
1806 	 */
1807 	for (i = 0; i < adapter->req_rx_queues; i++) {
1808 		netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1809 		if (prev_state == VNIC_CLOSED)
1810 			enable_irq(adapter->rx_scrq[i]->irq);
1811 		enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1812 	}
1813 
1814 	for (i = 0; i < adapter->req_tx_queues; i++) {
1815 		netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1816 		if (prev_state == VNIC_CLOSED)
1817 			enable_irq(adapter->tx_scrq[i]->irq);
1818 		enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1819 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1820 	}
1821 
1822 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1823 	if (rc) {
1824 		ibmvnic_napi_disable(adapter);
1825 		ibmvnic_disable_irqs(adapter);
1826 		return rc;
1827 	}
1828 
1829 	adapter->tx_queues_active = true;
1830 
1831 	/* Since queues were stopped until now, there shouldn't be any
1832 	 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we
1833 	 * don't need the synchronize_rcu()? Leaving it for consistency
1834 	 * with setting ->tx_queues_active = false.
1835 	 */
1836 	synchronize_rcu();
1837 
1838 	netif_tx_start_all_queues(netdev);
1839 
1840 	if (prev_state == VNIC_CLOSED) {
1841 		for (i = 0; i < adapter->req_rx_queues; i++)
1842 			napi_schedule(&adapter->napi[i]);
1843 	}
1844 
1845 	adapter->state = VNIC_OPEN;
1846 	return rc;
1847 }
1848 
1849 static int ibmvnic_open(struct net_device *netdev)
1850 {
1851 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1852 	int rc;
1853 
1854 	ASSERT_RTNL();
1855 
1856 	/* If device failover is pending or we are about to reset, just set
1857 	 * device state and return. Device operation will be handled by reset
1858 	 * routine.
1859 	 *
1860 	 * It should be safe to overwrite the adapter->state here. Since
1861 	 * we hold the rtnl, either the reset has not actually started or
1862 	 * the rtnl got dropped during the set_link_state() in do_reset().
1863 	 * In the former case, no one else is changing the state (again we
1864 	 * have the rtnl) and in the latter case, do_reset() will detect and
1865 	 * honor our setting below.
1866 	 */
1867 	if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1868 		netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1869 			   adapter_state_to_string(adapter->state),
1870 			   adapter->failover_pending);
1871 		adapter->state = VNIC_OPEN;
1872 		rc = 0;
1873 		goto out;
1874 	}
1875 
1876 	if (adapter->state != VNIC_CLOSED) {
1877 		rc = ibmvnic_login(netdev);
1878 		if (rc)
1879 			goto out;
1880 
1881 		rc = init_resources(adapter);
1882 		if (rc) {
1883 			netdev_err(netdev, "failed to initialize resources\n");
1884 			goto out;
1885 		}
1886 	}
1887 
1888 	rc = __ibmvnic_open(netdev);
1889 
1890 out:
1891 	/* If open failed and there is a pending failover or in-progress reset,
1892 	 * set device state and return. Device operation will be handled by
1893 	 * reset routine. See also comments above regarding rtnl.
1894 	 */
1895 	if (rc &&
1896 	    (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1897 		adapter->state = VNIC_OPEN;
1898 		rc = 0;
1899 	}
1900 
1901 	if (rc) {
1902 		release_resources(adapter);
1903 		release_rx_pools(adapter);
1904 		release_tx_pools(adapter);
1905 	}
1906 
1907 	return rc;
1908 }
1909 
1910 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1911 {
1912 	struct ibmvnic_rx_pool *rx_pool;
1913 	struct ibmvnic_rx_buff *rx_buff;
1914 	u64 rx_entries;
1915 	int rx_scrqs;
1916 	int i, j;
1917 
1918 	if (!adapter->rx_pool)
1919 		return;
1920 
1921 	rx_scrqs = adapter->num_active_rx_pools;
1922 	rx_entries = adapter->req_rx_add_entries_per_subcrq;
1923 
1924 	/* Free any remaining skbs in the rx buffer pools */
1925 	for (i = 0; i < rx_scrqs; i++) {
1926 		rx_pool = &adapter->rx_pool[i];
1927 		if (!rx_pool || !rx_pool->rx_buff)
1928 			continue;
1929 
1930 		netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1931 		for (j = 0; j < rx_entries; j++) {
1932 			rx_buff = &rx_pool->rx_buff[j];
1933 			if (rx_buff && rx_buff->skb) {
1934 				dev_kfree_skb_any(rx_buff->skb);
1935 				rx_buff->skb = NULL;
1936 			}
1937 		}
1938 	}
1939 }
1940 
1941 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1942 			      struct ibmvnic_tx_pool *tx_pool)
1943 {
1944 	struct ibmvnic_tx_buff *tx_buff;
1945 	u64 tx_entries;
1946 	int i;
1947 
1948 	if (!tx_pool || !tx_pool->tx_buff)
1949 		return;
1950 
1951 	tx_entries = tx_pool->num_buffers;
1952 
1953 	for (i = 0; i < tx_entries; i++) {
1954 		tx_buff = &tx_pool->tx_buff[i];
1955 		if (tx_buff && tx_buff->skb) {
1956 			dev_kfree_skb_any(tx_buff->skb);
1957 			tx_buff->skb = NULL;
1958 		}
1959 	}
1960 }
1961 
1962 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1963 {
1964 	int tx_scrqs;
1965 	int i;
1966 
1967 	if (!adapter->tx_pool || !adapter->tso_pool)
1968 		return;
1969 
1970 	tx_scrqs = adapter->num_active_tx_pools;
1971 
1972 	/* Free any remaining skbs in the tx buffer pools */
1973 	for (i = 0; i < tx_scrqs; i++) {
1974 		netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1975 		clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1976 		clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1977 	}
1978 }
1979 
1980 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1981 {
1982 	struct net_device *netdev = adapter->netdev;
1983 	int i;
1984 
1985 	if (adapter->tx_scrq) {
1986 		for (i = 0; i < adapter->req_tx_queues; i++)
1987 			if (adapter->tx_scrq[i]->irq) {
1988 				netdev_dbg(netdev,
1989 					   "Disabling tx_scrq[%d] irq\n", i);
1990 				disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1991 				disable_irq(adapter->tx_scrq[i]->irq);
1992 			}
1993 	}
1994 
1995 	if (adapter->rx_scrq) {
1996 		for (i = 0; i < adapter->req_rx_queues; i++) {
1997 			if (adapter->rx_scrq[i]->irq) {
1998 				netdev_dbg(netdev,
1999 					   "Disabling rx_scrq[%d] irq\n", i);
2000 				disable_scrq_irq(adapter, adapter->rx_scrq[i]);
2001 				disable_irq(adapter->rx_scrq[i]->irq);
2002 			}
2003 		}
2004 	}
2005 }
2006 
2007 static void ibmvnic_cleanup(struct net_device *netdev)
2008 {
2009 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2010 
2011 	/* ensure that transmissions are stopped if called by do_reset */
2012 
2013 	adapter->tx_queues_active = false;
2014 
2015 	/* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active
2016 	 * update so they don't restart a queue after we stop it below.
2017 	 */
2018 	synchronize_rcu();
2019 
2020 	if (test_bit(0, &adapter->resetting))
2021 		netif_tx_disable(netdev);
2022 	else
2023 		netif_tx_stop_all_queues(netdev);
2024 
2025 	ibmvnic_napi_disable(adapter);
2026 	ibmvnic_disable_irqs(adapter);
2027 }
2028 
2029 static int __ibmvnic_close(struct net_device *netdev)
2030 {
2031 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2032 	int rc = 0;
2033 
2034 	adapter->state = VNIC_CLOSING;
2035 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2036 	adapter->state = VNIC_CLOSED;
2037 	return rc;
2038 }
2039 
2040 static int ibmvnic_close(struct net_device *netdev)
2041 {
2042 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2043 	int rc;
2044 
2045 	netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
2046 		   adapter_state_to_string(adapter->state),
2047 		   adapter->failover_pending,
2048 		   adapter->force_reset_recovery);
2049 
2050 	/* If device failover is pending, just set device state and return.
2051 	 * Device operation will be handled by reset routine.
2052 	 */
2053 	if (adapter->failover_pending) {
2054 		adapter->state = VNIC_CLOSED;
2055 		return 0;
2056 	}
2057 
2058 	rc = __ibmvnic_close(netdev);
2059 	ibmvnic_cleanup(netdev);
2060 	clean_rx_pools(adapter);
2061 	clean_tx_pools(adapter);
2062 
2063 	return rc;
2064 }
2065 
2066 /**
2067  * build_hdr_data - creates L2/L3/L4 header data buffer
2068  * @hdr_field: bitfield determining needed headers
2069  * @skb: socket buffer
2070  * @hdr_len: array of header lengths
2071  * @hdr_data: buffer to write the header to
2072  *
2073  * Reads hdr_field to determine which headers are needed by firmware.
2074  * Builds a buffer containing these headers.  Saves individual header
2075  * lengths and total buffer length to be used to build descriptors.
2076  */
2077 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
2078 			  int *hdr_len, u8 *hdr_data)
2079 {
2080 	int len = 0;
2081 	u8 *hdr;
2082 
2083 	if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
2084 		hdr_len[0] = sizeof(struct vlan_ethhdr);
2085 	else
2086 		hdr_len[0] = sizeof(struct ethhdr);
2087 
2088 	if (skb->protocol == htons(ETH_P_IP)) {
2089 		hdr_len[1] = ip_hdr(skb)->ihl * 4;
2090 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2091 			hdr_len[2] = tcp_hdrlen(skb);
2092 		else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
2093 			hdr_len[2] = sizeof(struct udphdr);
2094 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2095 		hdr_len[1] = sizeof(struct ipv6hdr);
2096 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2097 			hdr_len[2] = tcp_hdrlen(skb);
2098 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
2099 			hdr_len[2] = sizeof(struct udphdr);
2100 	} else if (skb->protocol == htons(ETH_P_ARP)) {
2101 		hdr_len[1] = arp_hdr_len(skb->dev);
2102 		hdr_len[2] = 0;
2103 	}
2104 
2105 	memset(hdr_data, 0, 120);
2106 	if ((hdr_field >> 6) & 1) {
2107 		hdr = skb_mac_header(skb);
2108 		memcpy(hdr_data, hdr, hdr_len[0]);
2109 		len += hdr_len[0];
2110 	}
2111 
2112 	if ((hdr_field >> 5) & 1) {
2113 		hdr = skb_network_header(skb);
2114 		memcpy(hdr_data + len, hdr, hdr_len[1]);
2115 		len += hdr_len[1];
2116 	}
2117 
2118 	if ((hdr_field >> 4) & 1) {
2119 		hdr = skb_transport_header(skb);
2120 		memcpy(hdr_data + len, hdr, hdr_len[2]);
2121 		len += hdr_len[2];
2122 	}
2123 	return len;
2124 }
2125 
2126 /**
2127  * create_hdr_descs - create header and header extension descriptors
2128  * @hdr_field: bitfield determining needed headers
2129  * @hdr_data: buffer containing header data
2130  * @len: length of data buffer
2131  * @hdr_len: array of individual header lengths
2132  * @scrq_arr: descriptor array
2133  *
2134  * Creates header and, if needed, header extension descriptors and
2135  * places them in a descriptor array, scrq_arr
2136  */
2137 
2138 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
2139 			    union sub_crq *scrq_arr)
2140 {
2141 	union sub_crq hdr_desc;
2142 	int tmp_len = len;
2143 	int num_descs = 0;
2144 	u8 *data, *cur;
2145 	int tmp;
2146 
2147 	while (tmp_len > 0) {
2148 		cur = hdr_data + len - tmp_len;
2149 
2150 		memset(&hdr_desc, 0, sizeof(hdr_desc));
2151 		if (cur != hdr_data) {
2152 			data = hdr_desc.hdr_ext.data;
2153 			tmp = tmp_len > 29 ? 29 : tmp_len;
2154 			hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
2155 			hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
2156 			hdr_desc.hdr_ext.len = tmp;
2157 		} else {
2158 			data = hdr_desc.hdr.data;
2159 			tmp = tmp_len > 24 ? 24 : tmp_len;
2160 			hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
2161 			hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
2162 			hdr_desc.hdr.len = tmp;
2163 			hdr_desc.hdr.l2_len = (u8)hdr_len[0];
2164 			hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
2165 			hdr_desc.hdr.l4_len = (u8)hdr_len[2];
2166 			hdr_desc.hdr.flag = hdr_field << 1;
2167 		}
2168 		memcpy(data, cur, tmp);
2169 		tmp_len -= tmp;
2170 		*scrq_arr = hdr_desc;
2171 		scrq_arr++;
2172 		num_descs++;
2173 	}
2174 
2175 	return num_descs;
2176 }
2177 
2178 /**
2179  * build_hdr_descs_arr - build a header descriptor array
2180  * @skb: tx socket buffer
2181  * @indir_arr: indirect array
2182  * @num_entries: number of descriptors to be sent
2183  * @hdr_field: bit field determining which headers will be sent
2184  *
2185  * This function will build a TX descriptor array with applicable
2186  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
2187  */
2188 
2189 static void build_hdr_descs_arr(struct sk_buff *skb,
2190 				union sub_crq *indir_arr,
2191 				int *num_entries, u8 hdr_field)
2192 {
2193 	int hdr_len[3] = {0, 0, 0};
2194 	u8 hdr_data[140] = {0};
2195 	int tot_len;
2196 
2197 	tot_len = build_hdr_data(hdr_field, skb, hdr_len,
2198 				 hdr_data);
2199 	*num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
2200 					 indir_arr + 1);
2201 }
2202 
2203 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
2204 				    struct net_device *netdev)
2205 {
2206 	/* For some backing devices, mishandling of small packets
2207 	 * can result in a loss of connection or TX stall. Device
2208 	 * architects recommend that no packet should be smaller
2209 	 * than the minimum MTU value provided to the driver, so
2210 	 * pad any packets to that length
2211 	 */
2212 	if (skb->len < netdev->min_mtu)
2213 		return skb_put_padto(skb, netdev->min_mtu);
2214 
2215 	return 0;
2216 }
2217 
2218 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
2219 					 struct ibmvnic_sub_crq_queue *tx_scrq)
2220 {
2221 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2222 	struct ibmvnic_tx_buff *tx_buff;
2223 	struct ibmvnic_tx_pool *tx_pool;
2224 	union sub_crq tx_scrq_entry;
2225 	int queue_num;
2226 	int entries;
2227 	int index;
2228 	int i;
2229 
2230 	ind_bufp = &tx_scrq->ind_buf;
2231 	entries = (u64)ind_bufp->index;
2232 	queue_num = tx_scrq->pool_index;
2233 
2234 	for (i = entries - 1; i >= 0; --i) {
2235 		tx_scrq_entry = ind_bufp->indir_arr[i];
2236 		if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
2237 			continue;
2238 		index = be32_to_cpu(tx_scrq_entry.v1.correlator);
2239 		if (index & IBMVNIC_TSO_POOL_MASK) {
2240 			tx_pool = &adapter->tso_pool[queue_num];
2241 			index &= ~IBMVNIC_TSO_POOL_MASK;
2242 		} else {
2243 			tx_pool = &adapter->tx_pool[queue_num];
2244 		}
2245 		tx_pool->free_map[tx_pool->consumer_index] = index;
2246 		tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2247 					  tx_pool->num_buffers - 1 :
2248 					  tx_pool->consumer_index - 1;
2249 		tx_buff = &tx_pool->tx_buff[index];
2250 		adapter->netdev->stats.tx_packets--;
2251 		adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
2252 		adapter->tx_stats_buffers[queue_num].packets--;
2253 		adapter->tx_stats_buffers[queue_num].bytes -=
2254 						tx_buff->skb->len;
2255 		dev_kfree_skb_any(tx_buff->skb);
2256 		tx_buff->skb = NULL;
2257 		adapter->netdev->stats.tx_dropped++;
2258 	}
2259 
2260 	ind_bufp->index = 0;
2261 
2262 	if (atomic_sub_return(entries, &tx_scrq->used) <=
2263 	    (adapter->req_tx_entries_per_subcrq / 2) &&
2264 	    __netif_subqueue_stopped(adapter->netdev, queue_num)) {
2265 		rcu_read_lock();
2266 
2267 		if (adapter->tx_queues_active) {
2268 			netif_wake_subqueue(adapter->netdev, queue_num);
2269 			netdev_dbg(adapter->netdev, "Started queue %d\n",
2270 				   queue_num);
2271 		}
2272 
2273 		rcu_read_unlock();
2274 	}
2275 }
2276 
2277 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
2278 				 struct ibmvnic_sub_crq_queue *tx_scrq)
2279 {
2280 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2281 	u64 dma_addr;
2282 	u64 entries;
2283 	u64 handle;
2284 	int rc;
2285 
2286 	ind_bufp = &tx_scrq->ind_buf;
2287 	dma_addr = (u64)ind_bufp->indir_dma;
2288 	entries = (u64)ind_bufp->index;
2289 	handle = tx_scrq->handle;
2290 
2291 	if (!entries)
2292 		return 0;
2293 	rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
2294 	if (rc)
2295 		ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
2296 	else
2297 		ind_bufp->index = 0;
2298 	return 0;
2299 }
2300 
2301 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
2302 {
2303 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2304 	int queue_num = skb_get_queue_mapping(skb);
2305 	u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
2306 	struct device *dev = &adapter->vdev->dev;
2307 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2308 	struct ibmvnic_tx_buff *tx_buff = NULL;
2309 	struct ibmvnic_sub_crq_queue *tx_scrq;
2310 	struct ibmvnic_long_term_buff *ltb;
2311 	struct ibmvnic_tx_pool *tx_pool;
2312 	unsigned int tx_send_failed = 0;
2313 	netdev_tx_t ret = NETDEV_TX_OK;
2314 	unsigned int tx_map_failed = 0;
2315 	union sub_crq indir_arr[16];
2316 	unsigned int tx_dropped = 0;
2317 	unsigned int tx_packets = 0;
2318 	unsigned int tx_bytes = 0;
2319 	dma_addr_t data_dma_addr;
2320 	struct netdev_queue *txq;
2321 	unsigned long lpar_rc;
2322 	union sub_crq tx_crq;
2323 	unsigned int offset;
2324 	int num_entries = 1;
2325 	unsigned char *dst;
2326 	int bufidx = 0;
2327 	u8 proto = 0;
2328 
2329 	/* If a reset is in progress, drop the packet since
2330 	 * the scrqs may get torn down. Otherwise use the
2331 	 * rcu to ensure reset waits for us to complete.
2332 	 */
2333 	rcu_read_lock();
2334 	if (!adapter->tx_queues_active) {
2335 		dev_kfree_skb_any(skb);
2336 
2337 		tx_send_failed++;
2338 		tx_dropped++;
2339 		ret = NETDEV_TX_OK;
2340 		goto out;
2341 	}
2342 
2343 	tx_scrq = adapter->tx_scrq[queue_num];
2344 	txq = netdev_get_tx_queue(netdev, queue_num);
2345 	ind_bufp = &tx_scrq->ind_buf;
2346 
2347 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
2348 		tx_dropped++;
2349 		tx_send_failed++;
2350 		ret = NETDEV_TX_OK;
2351 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2352 		goto out;
2353 	}
2354 
2355 	if (skb_is_gso(skb))
2356 		tx_pool = &adapter->tso_pool[queue_num];
2357 	else
2358 		tx_pool = &adapter->tx_pool[queue_num];
2359 
2360 	bufidx = tx_pool->free_map[tx_pool->consumer_index];
2361 
2362 	if (bufidx == IBMVNIC_INVALID_MAP) {
2363 		dev_kfree_skb_any(skb);
2364 		tx_send_failed++;
2365 		tx_dropped++;
2366 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2367 		ret = NETDEV_TX_OK;
2368 		goto out;
2369 	}
2370 
2371 	tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
2372 
2373 	map_txpool_buf_to_ltb(tx_pool, bufidx, &ltb, &offset);
2374 
2375 	dst = ltb->buff + offset;
2376 	memset(dst, 0, tx_pool->buf_size);
2377 	data_dma_addr = ltb->addr + offset;
2378 
2379 	if (skb_shinfo(skb)->nr_frags) {
2380 		int cur, i;
2381 
2382 		/* Copy the head */
2383 		skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
2384 		cur = skb_headlen(skb);
2385 
2386 		/* Copy the frags */
2387 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2388 			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2389 
2390 			memcpy(dst + cur, skb_frag_address(frag),
2391 			       skb_frag_size(frag));
2392 			cur += skb_frag_size(frag);
2393 		}
2394 	} else {
2395 		skb_copy_from_linear_data(skb, dst, skb->len);
2396 	}
2397 
2398 	/* post changes to long_term_buff *dst before VIOS accessing it */
2399 	dma_wmb();
2400 
2401 	tx_pool->consumer_index =
2402 	    (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
2403 
2404 	tx_buff = &tx_pool->tx_buff[bufidx];
2405 	tx_buff->skb = skb;
2406 	tx_buff->index = bufidx;
2407 	tx_buff->pool_index = queue_num;
2408 
2409 	memset(&tx_crq, 0, sizeof(tx_crq));
2410 	tx_crq.v1.first = IBMVNIC_CRQ_CMD;
2411 	tx_crq.v1.type = IBMVNIC_TX_DESC;
2412 	tx_crq.v1.n_crq_elem = 1;
2413 	tx_crq.v1.n_sge = 1;
2414 	tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
2415 
2416 	if (skb_is_gso(skb))
2417 		tx_crq.v1.correlator =
2418 			cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK);
2419 	else
2420 		tx_crq.v1.correlator = cpu_to_be32(bufidx);
2421 	tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id);
2422 	tx_crq.v1.sge_len = cpu_to_be32(skb->len);
2423 	tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
2424 
2425 	if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
2426 		tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
2427 		tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2428 	}
2429 
2430 	if (skb->protocol == htons(ETH_P_IP)) {
2431 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2432 		proto = ip_hdr(skb)->protocol;
2433 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2434 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2435 		proto = ipv6_hdr(skb)->nexthdr;
2436 	}
2437 
2438 	if (proto == IPPROTO_TCP)
2439 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2440 	else if (proto == IPPROTO_UDP)
2441 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2442 
2443 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2444 		tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2445 		hdrs += 2;
2446 	}
2447 	if (skb_is_gso(skb)) {
2448 		tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2449 		tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2450 		hdrs += 2;
2451 	}
2452 
2453 	if ((*hdrs >> 7) & 1)
2454 		build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2455 
2456 	tx_crq.v1.n_crq_elem = num_entries;
2457 	tx_buff->num_entries = num_entries;
2458 	/* flush buffer if current entry can not fit */
2459 	if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2460 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2461 		if (lpar_rc != H_SUCCESS)
2462 			goto tx_flush_err;
2463 	}
2464 
2465 	indir_arr[0] = tx_crq;
2466 	memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2467 	       num_entries * sizeof(struct ibmvnic_generic_scrq));
2468 	ind_bufp->index += num_entries;
2469 	if (__netdev_tx_sent_queue(txq, skb->len,
2470 				   netdev_xmit_more() &&
2471 				   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2472 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2473 		if (lpar_rc != H_SUCCESS)
2474 			goto tx_err;
2475 	}
2476 
2477 	if (atomic_add_return(num_entries, &tx_scrq->used)
2478 					>= adapter->req_tx_entries_per_subcrq) {
2479 		netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2480 		netif_stop_subqueue(netdev, queue_num);
2481 	}
2482 
2483 	tx_packets++;
2484 	tx_bytes += skb->len;
2485 	txq_trans_cond_update(txq);
2486 	ret = NETDEV_TX_OK;
2487 	goto out;
2488 
2489 tx_flush_err:
2490 	dev_kfree_skb_any(skb);
2491 	tx_buff->skb = NULL;
2492 	tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2493 				  tx_pool->num_buffers - 1 :
2494 				  tx_pool->consumer_index - 1;
2495 	tx_dropped++;
2496 tx_err:
2497 	if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2498 		dev_err_ratelimited(dev, "tx: send failed\n");
2499 
2500 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2501 		/* Disable TX and report carrier off if queue is closed
2502 		 * or pending failover.
2503 		 * Firmware guarantees that a signal will be sent to the
2504 		 * driver, triggering a reset or some other action.
2505 		 */
2506 		netif_tx_stop_all_queues(netdev);
2507 		netif_carrier_off(netdev);
2508 	}
2509 out:
2510 	rcu_read_unlock();
2511 	netdev->stats.tx_dropped += tx_dropped;
2512 	netdev->stats.tx_bytes += tx_bytes;
2513 	netdev->stats.tx_packets += tx_packets;
2514 	adapter->tx_send_failed += tx_send_failed;
2515 	adapter->tx_map_failed += tx_map_failed;
2516 	adapter->tx_stats_buffers[queue_num].packets += tx_packets;
2517 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2518 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2519 
2520 	return ret;
2521 }
2522 
2523 static void ibmvnic_set_multi(struct net_device *netdev)
2524 {
2525 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2526 	struct netdev_hw_addr *ha;
2527 	union ibmvnic_crq crq;
2528 
2529 	memset(&crq, 0, sizeof(crq));
2530 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
2531 	crq.request_capability.cmd = REQUEST_CAPABILITY;
2532 
2533 	if (netdev->flags & IFF_PROMISC) {
2534 		if (!adapter->promisc_supported)
2535 			return;
2536 	} else {
2537 		if (netdev->flags & IFF_ALLMULTI) {
2538 			/* Accept all multicast */
2539 			memset(&crq, 0, sizeof(crq));
2540 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2541 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2542 			crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2543 			ibmvnic_send_crq(adapter, &crq);
2544 		} else if (netdev_mc_empty(netdev)) {
2545 			/* Reject all multicast */
2546 			memset(&crq, 0, sizeof(crq));
2547 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2548 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2549 			crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2550 			ibmvnic_send_crq(adapter, &crq);
2551 		} else {
2552 			/* Accept one or more multicast(s) */
2553 			netdev_for_each_mc_addr(ha, netdev) {
2554 				memset(&crq, 0, sizeof(crq));
2555 				crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2556 				crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2557 				crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2558 				ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2559 						ha->addr);
2560 				ibmvnic_send_crq(adapter, &crq);
2561 			}
2562 		}
2563 	}
2564 }
2565 
2566 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2567 {
2568 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2569 	union ibmvnic_crq crq;
2570 	int rc;
2571 
2572 	if (!is_valid_ether_addr(dev_addr)) {
2573 		rc = -EADDRNOTAVAIL;
2574 		goto err;
2575 	}
2576 
2577 	memset(&crq, 0, sizeof(crq));
2578 	crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2579 	crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2580 	ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2581 
2582 	mutex_lock(&adapter->fw_lock);
2583 	adapter->fw_done_rc = 0;
2584 	reinit_completion(&adapter->fw_done);
2585 
2586 	rc = ibmvnic_send_crq(adapter, &crq);
2587 	if (rc) {
2588 		rc = -EIO;
2589 		mutex_unlock(&adapter->fw_lock);
2590 		goto err;
2591 	}
2592 
2593 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2594 	/* netdev->dev_addr is changed in handle_change_mac_rsp function */
2595 	if (rc || adapter->fw_done_rc) {
2596 		rc = -EIO;
2597 		mutex_unlock(&adapter->fw_lock);
2598 		goto err;
2599 	}
2600 	mutex_unlock(&adapter->fw_lock);
2601 	return 0;
2602 err:
2603 	ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2604 	return rc;
2605 }
2606 
2607 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2608 {
2609 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2610 	struct sockaddr *addr = p;
2611 	int rc;
2612 
2613 	rc = 0;
2614 	if (!is_valid_ether_addr(addr->sa_data))
2615 		return -EADDRNOTAVAIL;
2616 
2617 	ether_addr_copy(adapter->mac_addr, addr->sa_data);
2618 	if (adapter->state != VNIC_PROBED)
2619 		rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2620 
2621 	return rc;
2622 }
2623 
2624 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2625 {
2626 	switch (reason) {
2627 	case VNIC_RESET_FAILOVER:
2628 		return "FAILOVER";
2629 	case VNIC_RESET_MOBILITY:
2630 		return "MOBILITY";
2631 	case VNIC_RESET_FATAL:
2632 		return "FATAL";
2633 	case VNIC_RESET_NON_FATAL:
2634 		return "NON_FATAL";
2635 	case VNIC_RESET_TIMEOUT:
2636 		return "TIMEOUT";
2637 	case VNIC_RESET_CHANGE_PARAM:
2638 		return "CHANGE_PARAM";
2639 	case VNIC_RESET_PASSIVE_INIT:
2640 		return "PASSIVE_INIT";
2641 	}
2642 	return "UNKNOWN";
2643 }
2644 
2645 /*
2646  * Initialize the init_done completion and return code values. We
2647  * can get a transport event just after registering the CRQ and the
2648  * tasklet will use this to communicate the transport event. To ensure
2649  * we don't miss the notification/error, initialize these _before_
2650  * regisering the CRQ.
2651  */
2652 static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
2653 {
2654 	reinit_completion(&adapter->init_done);
2655 	adapter->init_done_rc = 0;
2656 }
2657 
2658 /*
2659  * do_reset returns zero if we are able to keep processing reset events, or
2660  * non-zero if we hit a fatal error and must halt.
2661  */
2662 static int do_reset(struct ibmvnic_adapter *adapter,
2663 		    struct ibmvnic_rwi *rwi, u32 reset_state)
2664 {
2665 	struct net_device *netdev = adapter->netdev;
2666 	u64 old_num_rx_queues, old_num_tx_queues;
2667 	u64 old_num_rx_slots, old_num_tx_slots;
2668 	int rc;
2669 
2670 	netdev_dbg(adapter->netdev,
2671 		   "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2672 		   adapter_state_to_string(adapter->state),
2673 		   adapter->failover_pending,
2674 		   reset_reason_to_string(rwi->reset_reason),
2675 		   adapter_state_to_string(reset_state));
2676 
2677 	adapter->reset_reason = rwi->reset_reason;
2678 	/* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2679 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2680 		rtnl_lock();
2681 
2682 	/* Now that we have the rtnl lock, clear any pending failover.
2683 	 * This will ensure ibmvnic_open() has either completed or will
2684 	 * block until failover is complete.
2685 	 */
2686 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2687 		adapter->failover_pending = false;
2688 
2689 	/* read the state and check (again) after getting rtnl */
2690 	reset_state = adapter->state;
2691 
2692 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2693 		rc = -EBUSY;
2694 		goto out;
2695 	}
2696 
2697 	netif_carrier_off(netdev);
2698 
2699 	old_num_rx_queues = adapter->req_rx_queues;
2700 	old_num_tx_queues = adapter->req_tx_queues;
2701 	old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2702 	old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2703 
2704 	ibmvnic_cleanup(netdev);
2705 
2706 	if (reset_state == VNIC_OPEN &&
2707 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
2708 	    adapter->reset_reason != VNIC_RESET_FAILOVER) {
2709 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2710 			rc = __ibmvnic_close(netdev);
2711 			if (rc)
2712 				goto out;
2713 		} else {
2714 			adapter->state = VNIC_CLOSING;
2715 
2716 			/* Release the RTNL lock before link state change and
2717 			 * re-acquire after the link state change to allow
2718 			 * linkwatch_event to grab the RTNL lock and run during
2719 			 * a reset.
2720 			 */
2721 			rtnl_unlock();
2722 			rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2723 			rtnl_lock();
2724 			if (rc)
2725 				goto out;
2726 
2727 			if (adapter->state == VNIC_OPEN) {
2728 				/* When we dropped rtnl, ibmvnic_open() got
2729 				 * it and noticed that we are resetting and
2730 				 * set the adapter state to OPEN. Update our
2731 				 * new "target" state, and resume the reset
2732 				 * from VNIC_CLOSING state.
2733 				 */
2734 				netdev_dbg(netdev,
2735 					   "Open changed state from %s, updating.\n",
2736 					   adapter_state_to_string(reset_state));
2737 				reset_state = VNIC_OPEN;
2738 				adapter->state = VNIC_CLOSING;
2739 			}
2740 
2741 			if (adapter->state != VNIC_CLOSING) {
2742 				/* If someone else changed the adapter state
2743 				 * when we dropped the rtnl, fail the reset
2744 				 */
2745 				rc = -EAGAIN;
2746 				goto out;
2747 			}
2748 			adapter->state = VNIC_CLOSED;
2749 		}
2750 	}
2751 
2752 	if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2753 		release_resources(adapter);
2754 		release_sub_crqs(adapter, 1);
2755 		release_crq_queue(adapter);
2756 	}
2757 
2758 	if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2759 		/* remove the closed state so when we call open it appears
2760 		 * we are coming from the probed state.
2761 		 */
2762 		adapter->state = VNIC_PROBED;
2763 
2764 		reinit_init_done(adapter);
2765 
2766 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2767 			rc = init_crq_queue(adapter);
2768 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2769 			rc = ibmvnic_reenable_crq_queue(adapter);
2770 			release_sub_crqs(adapter, 1);
2771 		} else {
2772 			rc = ibmvnic_reset_crq(adapter);
2773 			if (rc == H_CLOSED || rc == H_SUCCESS) {
2774 				rc = vio_enable_interrupts(adapter->vdev);
2775 				if (rc)
2776 					netdev_err(adapter->netdev,
2777 						   "Reset failed to enable interrupts. rc=%d\n",
2778 						   rc);
2779 			}
2780 		}
2781 
2782 		if (rc) {
2783 			netdev_err(adapter->netdev,
2784 				   "Reset couldn't initialize crq. rc=%d\n", rc);
2785 			goto out;
2786 		}
2787 
2788 		rc = ibmvnic_reset_init(adapter, true);
2789 		if (rc)
2790 			goto out;
2791 
2792 		/* If the adapter was in PROBE or DOWN state prior to the reset,
2793 		 * exit here.
2794 		 */
2795 		if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2796 			rc = 0;
2797 			goto out;
2798 		}
2799 
2800 		rc = ibmvnic_login(netdev);
2801 		if (rc)
2802 			goto out;
2803 
2804 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2805 			rc = init_resources(adapter);
2806 			if (rc)
2807 				goto out;
2808 		} else if (adapter->req_rx_queues != old_num_rx_queues ||
2809 		    adapter->req_tx_queues != old_num_tx_queues ||
2810 		    adapter->req_rx_add_entries_per_subcrq !=
2811 		    old_num_rx_slots ||
2812 		    adapter->req_tx_entries_per_subcrq !=
2813 		    old_num_tx_slots ||
2814 		    !adapter->rx_pool ||
2815 		    !adapter->tso_pool ||
2816 		    !adapter->tx_pool) {
2817 			release_napi(adapter);
2818 			release_vpd_data(adapter);
2819 
2820 			rc = init_resources(adapter);
2821 			if (rc)
2822 				goto out;
2823 
2824 		} else {
2825 			rc = init_tx_pools(netdev);
2826 			if (rc) {
2827 				netdev_dbg(netdev,
2828 					   "init tx pools failed (%d)\n",
2829 					   rc);
2830 				goto out;
2831 			}
2832 
2833 			rc = init_rx_pools(netdev);
2834 			if (rc) {
2835 				netdev_dbg(netdev,
2836 					   "init rx pools failed (%d)\n",
2837 					   rc);
2838 				goto out;
2839 			}
2840 		}
2841 		ibmvnic_disable_irqs(adapter);
2842 	}
2843 	adapter->state = VNIC_CLOSED;
2844 
2845 	if (reset_state == VNIC_CLOSED) {
2846 		rc = 0;
2847 		goto out;
2848 	}
2849 
2850 	rc = __ibmvnic_open(netdev);
2851 	if (rc) {
2852 		rc = IBMVNIC_OPEN_FAILED;
2853 		goto out;
2854 	}
2855 
2856 	/* refresh device's multicast list */
2857 	ibmvnic_set_multi(netdev);
2858 
2859 	if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2860 	    adapter->reset_reason == VNIC_RESET_MOBILITY)
2861 		__netdev_notify_peers(netdev);
2862 
2863 	rc = 0;
2864 
2865 out:
2866 	/* restore the adapter state if reset failed */
2867 	if (rc)
2868 		adapter->state = reset_state;
2869 	/* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2870 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2871 		rtnl_unlock();
2872 
2873 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2874 		   adapter_state_to_string(adapter->state),
2875 		   adapter->failover_pending, rc);
2876 	return rc;
2877 }
2878 
2879 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2880 			 struct ibmvnic_rwi *rwi, u32 reset_state)
2881 {
2882 	struct net_device *netdev = adapter->netdev;
2883 	int rc;
2884 
2885 	netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2886 		   reset_reason_to_string(rwi->reset_reason));
2887 
2888 	/* read the state and check (again) after getting rtnl */
2889 	reset_state = adapter->state;
2890 
2891 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2892 		rc = -EBUSY;
2893 		goto out;
2894 	}
2895 
2896 	netif_carrier_off(netdev);
2897 	adapter->reset_reason = rwi->reset_reason;
2898 
2899 	ibmvnic_cleanup(netdev);
2900 	release_resources(adapter);
2901 	release_sub_crqs(adapter, 0);
2902 	release_crq_queue(adapter);
2903 
2904 	/* remove the closed state so when we call open it appears
2905 	 * we are coming from the probed state.
2906 	 */
2907 	adapter->state = VNIC_PROBED;
2908 
2909 	reinit_init_done(adapter);
2910 
2911 	rc = init_crq_queue(adapter);
2912 	if (rc) {
2913 		netdev_err(adapter->netdev,
2914 			   "Couldn't initialize crq. rc=%d\n", rc);
2915 		goto out;
2916 	}
2917 
2918 	rc = ibmvnic_reset_init(adapter, false);
2919 	if (rc)
2920 		goto out;
2921 
2922 	/* If the adapter was in PROBE or DOWN state prior to the reset,
2923 	 * exit here.
2924 	 */
2925 	if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
2926 		goto out;
2927 
2928 	rc = ibmvnic_login(netdev);
2929 	if (rc)
2930 		goto out;
2931 
2932 	rc = init_resources(adapter);
2933 	if (rc)
2934 		goto out;
2935 
2936 	ibmvnic_disable_irqs(adapter);
2937 	adapter->state = VNIC_CLOSED;
2938 
2939 	if (reset_state == VNIC_CLOSED)
2940 		goto out;
2941 
2942 	rc = __ibmvnic_open(netdev);
2943 	if (rc) {
2944 		rc = IBMVNIC_OPEN_FAILED;
2945 		goto out;
2946 	}
2947 
2948 	__netdev_notify_peers(netdev);
2949 out:
2950 	/* restore adapter state if reset failed */
2951 	if (rc)
2952 		adapter->state = reset_state;
2953 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
2954 		   adapter_state_to_string(adapter->state),
2955 		   adapter->failover_pending, rc);
2956 	return rc;
2957 }
2958 
2959 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2960 {
2961 	struct ibmvnic_rwi *rwi;
2962 	unsigned long flags;
2963 
2964 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2965 
2966 	if (!list_empty(&adapter->rwi_list)) {
2967 		rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2968 				       list);
2969 		list_del(&rwi->list);
2970 	} else {
2971 		rwi = NULL;
2972 	}
2973 
2974 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2975 	return rwi;
2976 }
2977 
2978 /**
2979  * do_passive_init - complete probing when partner device is detected.
2980  * @adapter: ibmvnic_adapter struct
2981  *
2982  * If the ibmvnic device does not have a partner device to communicate with at boot
2983  * and that partner device comes online at a later time, this function is called
2984  * to complete the initialization process of ibmvnic device.
2985  * Caller is expected to hold rtnl_lock().
2986  *
2987  * Returns non-zero if sub-CRQs are not initialized properly leaving the device
2988  * in the down state.
2989  * Returns 0 upon success and the device is in PROBED state.
2990  */
2991 
2992 static int do_passive_init(struct ibmvnic_adapter *adapter)
2993 {
2994 	unsigned long timeout = msecs_to_jiffies(30000);
2995 	struct net_device *netdev = adapter->netdev;
2996 	struct device *dev = &adapter->vdev->dev;
2997 	int rc;
2998 
2999 	netdev_dbg(netdev, "Partner device found, probing.\n");
3000 
3001 	adapter->state = VNIC_PROBING;
3002 	reinit_completion(&adapter->init_done);
3003 	adapter->init_done_rc = 0;
3004 	adapter->crq.active = true;
3005 
3006 	rc = send_crq_init_complete(adapter);
3007 	if (rc)
3008 		goto out;
3009 
3010 	rc = send_version_xchg(adapter);
3011 	if (rc)
3012 		netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
3013 
3014 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
3015 		dev_err(dev, "Initialization sequence timed out\n");
3016 		rc = -ETIMEDOUT;
3017 		goto out;
3018 	}
3019 
3020 	rc = init_sub_crqs(adapter);
3021 	if (rc) {
3022 		dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
3023 		goto out;
3024 	}
3025 
3026 	rc = init_sub_crq_irqs(adapter);
3027 	if (rc) {
3028 		dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
3029 		goto init_failed;
3030 	}
3031 
3032 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
3033 	netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
3034 	netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
3035 
3036 	adapter->state = VNIC_PROBED;
3037 	netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
3038 
3039 	return 0;
3040 
3041 init_failed:
3042 	release_sub_crqs(adapter, 1);
3043 out:
3044 	adapter->state = VNIC_DOWN;
3045 	return rc;
3046 }
3047 
3048 static void __ibmvnic_reset(struct work_struct *work)
3049 {
3050 	struct ibmvnic_adapter *adapter;
3051 	unsigned int timeout = 5000;
3052 	struct ibmvnic_rwi *tmprwi;
3053 	bool saved_state = false;
3054 	struct ibmvnic_rwi *rwi;
3055 	unsigned long flags;
3056 	struct device *dev;
3057 	bool need_reset;
3058 	int num_fails = 0;
3059 	u32 reset_state;
3060 	int rc = 0;
3061 
3062 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
3063 		dev = &adapter->vdev->dev;
3064 
3065 	/* Wait for ibmvnic_probe() to complete. If probe is taking too long
3066 	 * or if another reset is in progress, defer work for now. If probe
3067 	 * eventually fails it will flush and terminate our work.
3068 	 *
3069 	 * Three possibilities here:
3070 	 * 1. Adpater being removed  - just return
3071 	 * 2. Timed out on probe or another reset in progress - delay the work
3072 	 * 3. Completed probe - perform any resets in queue
3073 	 */
3074 	if (adapter->state == VNIC_PROBING &&
3075 	    !wait_for_completion_timeout(&adapter->probe_done, timeout)) {
3076 		dev_err(dev, "Reset thread timed out on probe");
3077 		queue_delayed_work(system_long_wq,
3078 				   &adapter->ibmvnic_delayed_reset,
3079 				   IBMVNIC_RESET_DELAY);
3080 		return;
3081 	}
3082 
3083 	/* adapter is done with probe (i.e state is never VNIC_PROBING now) */
3084 	if (adapter->state == VNIC_REMOVING)
3085 		return;
3086 
3087 	/* ->rwi_list is stable now (no one else is removing entries) */
3088 
3089 	/* ibmvnic_probe() may have purged the reset queue after we were
3090 	 * scheduled to process a reset so there maybe no resets to process.
3091 	 * Before setting the ->resetting bit though, we have to make sure
3092 	 * that there is infact a reset to process. Otherwise we may race
3093 	 * with ibmvnic_open() and end up leaving the vnic down:
3094 	 *
3095 	 *	__ibmvnic_reset()	    ibmvnic_open()
3096 	 *	-----------------	    --------------
3097 	 *
3098 	 *  set ->resetting bit
3099 	 *  				find ->resetting bit is set
3100 	 *  				set ->state to IBMVNIC_OPEN (i.e
3101 	 *  				assume reset will open device)
3102 	 *  				return
3103 	 *  find reset queue empty
3104 	 *  return
3105 	 *
3106 	 *  	Neither performed vnic login/open and vnic stays down
3107 	 *
3108 	 * If we hold the lock and conditionally set the bit, either we
3109 	 * or ibmvnic_open() will complete the open.
3110 	 */
3111 	need_reset = false;
3112 	spin_lock(&adapter->rwi_lock);
3113 	if (!list_empty(&adapter->rwi_list)) {
3114 		if (test_and_set_bit_lock(0, &adapter->resetting)) {
3115 			queue_delayed_work(system_long_wq,
3116 					   &adapter->ibmvnic_delayed_reset,
3117 					   IBMVNIC_RESET_DELAY);
3118 		} else {
3119 			need_reset = true;
3120 		}
3121 	}
3122 	spin_unlock(&adapter->rwi_lock);
3123 
3124 	if (!need_reset)
3125 		return;
3126 
3127 	rwi = get_next_rwi(adapter);
3128 	while (rwi) {
3129 		spin_lock_irqsave(&adapter->state_lock, flags);
3130 
3131 		if (adapter->state == VNIC_REMOVING ||
3132 		    adapter->state == VNIC_REMOVED) {
3133 			spin_unlock_irqrestore(&adapter->state_lock, flags);
3134 			kfree(rwi);
3135 			rc = EBUSY;
3136 			break;
3137 		}
3138 
3139 		if (!saved_state) {
3140 			reset_state = adapter->state;
3141 			saved_state = true;
3142 		}
3143 		spin_unlock_irqrestore(&adapter->state_lock, flags);
3144 
3145 		if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
3146 			rtnl_lock();
3147 			rc = do_passive_init(adapter);
3148 			rtnl_unlock();
3149 			if (!rc)
3150 				netif_carrier_on(adapter->netdev);
3151 		} else if (adapter->force_reset_recovery) {
3152 			/* Since we are doing a hard reset now, clear the
3153 			 * failover_pending flag so we don't ignore any
3154 			 * future MOBILITY or other resets.
3155 			 */
3156 			adapter->failover_pending = false;
3157 
3158 			/* Transport event occurred during previous reset */
3159 			if (adapter->wait_for_reset) {
3160 				/* Previous was CHANGE_PARAM; caller locked */
3161 				adapter->force_reset_recovery = false;
3162 				rc = do_hard_reset(adapter, rwi, reset_state);
3163 			} else {
3164 				rtnl_lock();
3165 				adapter->force_reset_recovery = false;
3166 				rc = do_hard_reset(adapter, rwi, reset_state);
3167 				rtnl_unlock();
3168 			}
3169 			if (rc)
3170 				num_fails++;
3171 			else
3172 				num_fails = 0;
3173 
3174 			/* If auto-priority-failover is enabled we can get
3175 			 * back to back failovers during resets, resulting
3176 			 * in at least two failed resets (from high-priority
3177 			 * backing device to low-priority one and then back)
3178 			 * If resets continue to fail beyond that, give the
3179 			 * adapter some time to settle down before retrying.
3180 			 */
3181 			if (num_fails >= 3) {
3182 				netdev_dbg(adapter->netdev,
3183 					   "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
3184 					   adapter_state_to_string(adapter->state),
3185 					   num_fails);
3186 				set_current_state(TASK_UNINTERRUPTIBLE);
3187 				schedule_timeout(60 * HZ);
3188 			}
3189 		} else {
3190 			rc = do_reset(adapter, rwi, reset_state);
3191 		}
3192 		tmprwi = rwi;
3193 		adapter->last_reset_time = jiffies;
3194 
3195 		if (rc)
3196 			netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
3197 
3198 		rwi = get_next_rwi(adapter);
3199 
3200 		/*
3201 		 * If there are no resets queued and the previous reset failed,
3202 		 * the adapter would be in an undefined state. So retry the
3203 		 * previous reset as a hard reset.
3204 		 *
3205 		 * Else, free the previous rwi and, if there is another reset
3206 		 * queued, process the new reset even if previous reset failed
3207 		 * (the previous reset could have failed because of a fail
3208 		 * over for instance, so process the fail over).
3209 		 */
3210 		if (!rwi && rc)
3211 			rwi = tmprwi;
3212 		else
3213 			kfree(tmprwi);
3214 
3215 		if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
3216 			    rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
3217 			adapter->force_reset_recovery = true;
3218 	}
3219 
3220 	if (adapter->wait_for_reset) {
3221 		adapter->reset_done_rc = rc;
3222 		complete(&adapter->reset_done);
3223 	}
3224 
3225 	clear_bit_unlock(0, &adapter->resetting);
3226 
3227 	netdev_dbg(adapter->netdev,
3228 		   "[S:%s FRR:%d WFR:%d] Done processing resets\n",
3229 		   adapter_state_to_string(adapter->state),
3230 		   adapter->force_reset_recovery,
3231 		   adapter->wait_for_reset);
3232 }
3233 
3234 static void __ibmvnic_delayed_reset(struct work_struct *work)
3235 {
3236 	struct ibmvnic_adapter *adapter;
3237 
3238 	adapter = container_of(work, struct ibmvnic_adapter,
3239 			       ibmvnic_delayed_reset.work);
3240 	__ibmvnic_reset(&adapter->ibmvnic_reset);
3241 }
3242 
3243 static void flush_reset_queue(struct ibmvnic_adapter *adapter)
3244 {
3245 	struct list_head *entry, *tmp_entry;
3246 
3247 	if (!list_empty(&adapter->rwi_list)) {
3248 		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
3249 			list_del(entry);
3250 			kfree(list_entry(entry, struct ibmvnic_rwi, list));
3251 		}
3252 	}
3253 }
3254 
3255 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
3256 			 enum ibmvnic_reset_reason reason)
3257 {
3258 	struct net_device *netdev = adapter->netdev;
3259 	struct ibmvnic_rwi *rwi, *tmp;
3260 	unsigned long flags;
3261 	int ret;
3262 
3263 	spin_lock_irqsave(&adapter->rwi_lock, flags);
3264 
3265 	/* If failover is pending don't schedule any other reset.
3266 	 * Instead let the failover complete. If there is already a
3267 	 * a failover reset scheduled, we will detect and drop the
3268 	 * duplicate reset when walking the ->rwi_list below.
3269 	 */
3270 	if (adapter->state == VNIC_REMOVING ||
3271 	    adapter->state == VNIC_REMOVED ||
3272 	    (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
3273 		ret = EBUSY;
3274 		netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
3275 		goto err;
3276 	}
3277 
3278 	list_for_each_entry(tmp, &adapter->rwi_list, list) {
3279 		if (tmp->reset_reason == reason) {
3280 			netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
3281 				   reset_reason_to_string(reason));
3282 			ret = EBUSY;
3283 			goto err;
3284 		}
3285 	}
3286 
3287 	rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
3288 	if (!rwi) {
3289 		ret = ENOMEM;
3290 		goto err;
3291 	}
3292 	/* if we just received a transport event,
3293 	 * flush reset queue and process this reset
3294 	 */
3295 	if (adapter->force_reset_recovery)
3296 		flush_reset_queue(adapter);
3297 
3298 	rwi->reset_reason = reason;
3299 	list_add_tail(&rwi->list, &adapter->rwi_list);
3300 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
3301 		   reset_reason_to_string(reason));
3302 	queue_work(system_long_wq, &adapter->ibmvnic_reset);
3303 
3304 	ret = 0;
3305 err:
3306 	/* ibmvnic_close() below can block, so drop the lock first */
3307 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3308 
3309 	if (ret == ENOMEM)
3310 		ibmvnic_close(netdev);
3311 
3312 	return -ret;
3313 }
3314 
3315 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
3316 {
3317 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3318 
3319 	if (test_bit(0, &adapter->resetting)) {
3320 		netdev_err(adapter->netdev,
3321 			   "Adapter is resetting, skip timeout reset\n");
3322 		return;
3323 	}
3324 	/* No queuing up reset until at least 5 seconds (default watchdog val)
3325 	 * after last reset
3326 	 */
3327 	if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
3328 		netdev_dbg(dev, "Not yet time to tx timeout.\n");
3329 		return;
3330 	}
3331 	ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
3332 }
3333 
3334 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
3335 				  struct ibmvnic_rx_buff *rx_buff)
3336 {
3337 	struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
3338 
3339 	rx_buff->skb = NULL;
3340 
3341 	pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
3342 	pool->next_alloc = (pool->next_alloc + 1) % pool->size;
3343 
3344 	atomic_dec(&pool->available);
3345 }
3346 
3347 static int ibmvnic_poll(struct napi_struct *napi, int budget)
3348 {
3349 	struct ibmvnic_sub_crq_queue *rx_scrq;
3350 	struct ibmvnic_adapter *adapter;
3351 	struct net_device *netdev;
3352 	int frames_processed;
3353 	int scrq_num;
3354 
3355 	netdev = napi->dev;
3356 	adapter = netdev_priv(netdev);
3357 	scrq_num = (int)(napi - adapter->napi);
3358 	frames_processed = 0;
3359 	rx_scrq = adapter->rx_scrq[scrq_num];
3360 
3361 restart_poll:
3362 	while (frames_processed < budget) {
3363 		struct sk_buff *skb;
3364 		struct ibmvnic_rx_buff *rx_buff;
3365 		union sub_crq *next;
3366 		u32 length;
3367 		u16 offset;
3368 		u8 flags = 0;
3369 
3370 		if (unlikely(test_bit(0, &adapter->resetting) &&
3371 			     adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
3372 			enable_scrq_irq(adapter, rx_scrq);
3373 			napi_complete_done(napi, frames_processed);
3374 			return frames_processed;
3375 		}
3376 
3377 		if (!pending_scrq(adapter, rx_scrq))
3378 			break;
3379 		next = ibmvnic_next_scrq(adapter, rx_scrq);
3380 		rx_buff = (struct ibmvnic_rx_buff *)
3381 			  be64_to_cpu(next->rx_comp.correlator);
3382 		/* do error checking */
3383 		if (next->rx_comp.rc) {
3384 			netdev_dbg(netdev, "rx buffer returned with rc %x\n",
3385 				   be16_to_cpu(next->rx_comp.rc));
3386 			/* free the entry */
3387 			next->rx_comp.first = 0;
3388 			dev_kfree_skb_any(rx_buff->skb);
3389 			remove_buff_from_pool(adapter, rx_buff);
3390 			continue;
3391 		} else if (!rx_buff->skb) {
3392 			/* free the entry */
3393 			next->rx_comp.first = 0;
3394 			remove_buff_from_pool(adapter, rx_buff);
3395 			continue;
3396 		}
3397 
3398 		length = be32_to_cpu(next->rx_comp.len);
3399 		offset = be16_to_cpu(next->rx_comp.off_frame_data);
3400 		flags = next->rx_comp.flags;
3401 		skb = rx_buff->skb;
3402 		/* load long_term_buff before copying to skb */
3403 		dma_rmb();
3404 		skb_copy_to_linear_data(skb, rx_buff->data + offset,
3405 					length);
3406 
3407 		/* VLAN Header has been stripped by the system firmware and
3408 		 * needs to be inserted by the driver
3409 		 */
3410 		if (adapter->rx_vlan_header_insertion &&
3411 		    (flags & IBMVNIC_VLAN_STRIPPED))
3412 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3413 					       ntohs(next->rx_comp.vlan_tci));
3414 
3415 		/* free the entry */
3416 		next->rx_comp.first = 0;
3417 		remove_buff_from_pool(adapter, rx_buff);
3418 
3419 		skb_put(skb, length);
3420 		skb->protocol = eth_type_trans(skb, netdev);
3421 		skb_record_rx_queue(skb, scrq_num);
3422 
3423 		if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
3424 		    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
3425 			skb->ip_summed = CHECKSUM_UNNECESSARY;
3426 		}
3427 
3428 		length = skb->len;
3429 		napi_gro_receive(napi, skb); /* send it up */
3430 		netdev->stats.rx_packets++;
3431 		netdev->stats.rx_bytes += length;
3432 		adapter->rx_stats_buffers[scrq_num].packets++;
3433 		adapter->rx_stats_buffers[scrq_num].bytes += length;
3434 		frames_processed++;
3435 	}
3436 
3437 	if (adapter->state != VNIC_CLOSING &&
3438 	    ((atomic_read(&adapter->rx_pool[scrq_num].available) <
3439 	      adapter->req_rx_add_entries_per_subcrq / 2) ||
3440 	      frames_processed < budget))
3441 		replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
3442 	if (frames_processed < budget) {
3443 		if (napi_complete_done(napi, frames_processed)) {
3444 			enable_scrq_irq(adapter, rx_scrq);
3445 			if (pending_scrq(adapter, rx_scrq)) {
3446 				if (napi_reschedule(napi)) {
3447 					disable_scrq_irq(adapter, rx_scrq);
3448 					goto restart_poll;
3449 				}
3450 			}
3451 		}
3452 	}
3453 	return frames_processed;
3454 }
3455 
3456 static int wait_for_reset(struct ibmvnic_adapter *adapter)
3457 {
3458 	int rc, ret;
3459 
3460 	adapter->fallback.mtu = adapter->req_mtu;
3461 	adapter->fallback.rx_queues = adapter->req_rx_queues;
3462 	adapter->fallback.tx_queues = adapter->req_tx_queues;
3463 	adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
3464 	adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
3465 
3466 	reinit_completion(&adapter->reset_done);
3467 	adapter->wait_for_reset = true;
3468 	rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3469 
3470 	if (rc) {
3471 		ret = rc;
3472 		goto out;
3473 	}
3474 	rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
3475 	if (rc) {
3476 		ret = -ENODEV;
3477 		goto out;
3478 	}
3479 
3480 	ret = 0;
3481 	if (adapter->reset_done_rc) {
3482 		ret = -EIO;
3483 		adapter->desired.mtu = adapter->fallback.mtu;
3484 		adapter->desired.rx_queues = adapter->fallback.rx_queues;
3485 		adapter->desired.tx_queues = adapter->fallback.tx_queues;
3486 		adapter->desired.rx_entries = adapter->fallback.rx_entries;
3487 		adapter->desired.tx_entries = adapter->fallback.tx_entries;
3488 
3489 		reinit_completion(&adapter->reset_done);
3490 		adapter->wait_for_reset = true;
3491 		rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3492 		if (rc) {
3493 			ret = rc;
3494 			goto out;
3495 		}
3496 		rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
3497 						 60000);
3498 		if (rc) {
3499 			ret = -ENODEV;
3500 			goto out;
3501 		}
3502 	}
3503 out:
3504 	adapter->wait_for_reset = false;
3505 
3506 	return ret;
3507 }
3508 
3509 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
3510 {
3511 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3512 
3513 	adapter->desired.mtu = new_mtu + ETH_HLEN;
3514 
3515 	return wait_for_reset(adapter);
3516 }
3517 
3518 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3519 						struct net_device *dev,
3520 						netdev_features_t features)
3521 {
3522 	/* Some backing hardware adapters can not
3523 	 * handle packets with a MSS less than 224
3524 	 * or with only one segment.
3525 	 */
3526 	if (skb_is_gso(skb)) {
3527 		if (skb_shinfo(skb)->gso_size < 224 ||
3528 		    skb_shinfo(skb)->gso_segs == 1)
3529 			features &= ~NETIF_F_GSO_MASK;
3530 	}
3531 
3532 	return features;
3533 }
3534 
3535 static const struct net_device_ops ibmvnic_netdev_ops = {
3536 	.ndo_open		= ibmvnic_open,
3537 	.ndo_stop		= ibmvnic_close,
3538 	.ndo_start_xmit		= ibmvnic_xmit,
3539 	.ndo_set_rx_mode	= ibmvnic_set_multi,
3540 	.ndo_set_mac_address	= ibmvnic_set_mac,
3541 	.ndo_validate_addr	= eth_validate_addr,
3542 	.ndo_tx_timeout		= ibmvnic_tx_timeout,
3543 	.ndo_change_mtu		= ibmvnic_change_mtu,
3544 	.ndo_features_check     = ibmvnic_features_check,
3545 };
3546 
3547 /* ethtool functions */
3548 
3549 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3550 				      struct ethtool_link_ksettings *cmd)
3551 {
3552 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3553 	int rc;
3554 
3555 	rc = send_query_phys_parms(adapter);
3556 	if (rc) {
3557 		adapter->speed = SPEED_UNKNOWN;
3558 		adapter->duplex = DUPLEX_UNKNOWN;
3559 	}
3560 	cmd->base.speed = adapter->speed;
3561 	cmd->base.duplex = adapter->duplex;
3562 	cmd->base.port = PORT_FIBRE;
3563 	cmd->base.phy_address = 0;
3564 	cmd->base.autoneg = AUTONEG_ENABLE;
3565 
3566 	return 0;
3567 }
3568 
3569 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3570 				struct ethtool_drvinfo *info)
3571 {
3572 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3573 
3574 	strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3575 	strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3576 	strscpy(info->fw_version, adapter->fw_version,
3577 		sizeof(info->fw_version));
3578 }
3579 
3580 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3581 {
3582 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3583 
3584 	return adapter->msg_enable;
3585 }
3586 
3587 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3588 {
3589 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3590 
3591 	adapter->msg_enable = data;
3592 }
3593 
3594 static u32 ibmvnic_get_link(struct net_device *netdev)
3595 {
3596 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3597 
3598 	/* Don't need to send a query because we request a logical link up at
3599 	 * init and then we wait for link state indications
3600 	 */
3601 	return adapter->logical_link_state;
3602 }
3603 
3604 static void ibmvnic_get_ringparam(struct net_device *netdev,
3605 				  struct ethtool_ringparam *ring,
3606 				  struct kernel_ethtool_ringparam *kernel_ring,
3607 				  struct netlink_ext_ack *extack)
3608 {
3609 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3610 
3611 	ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3612 	ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3613 	ring->rx_mini_max_pending = 0;
3614 	ring->rx_jumbo_max_pending = 0;
3615 	ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3616 	ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3617 	ring->rx_mini_pending = 0;
3618 	ring->rx_jumbo_pending = 0;
3619 }
3620 
3621 static int ibmvnic_set_ringparam(struct net_device *netdev,
3622 				 struct ethtool_ringparam *ring,
3623 				 struct kernel_ethtool_ringparam *kernel_ring,
3624 				 struct netlink_ext_ack *extack)
3625 {
3626 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3627 
3628 	if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
3629 	    ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
3630 		netdev_err(netdev, "Invalid request.\n");
3631 		netdev_err(netdev, "Max tx buffers = %llu\n",
3632 			   adapter->max_rx_add_entries_per_subcrq);
3633 		netdev_err(netdev, "Max rx buffers = %llu\n",
3634 			   adapter->max_tx_entries_per_subcrq);
3635 		return -EINVAL;
3636 	}
3637 
3638 	adapter->desired.rx_entries = ring->rx_pending;
3639 	adapter->desired.tx_entries = ring->tx_pending;
3640 
3641 	return wait_for_reset(adapter);
3642 }
3643 
3644 static void ibmvnic_get_channels(struct net_device *netdev,
3645 				 struct ethtool_channels *channels)
3646 {
3647 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3648 
3649 	channels->max_rx = adapter->max_rx_queues;
3650 	channels->max_tx = adapter->max_tx_queues;
3651 	channels->max_other = 0;
3652 	channels->max_combined = 0;
3653 	channels->rx_count = adapter->req_rx_queues;
3654 	channels->tx_count = adapter->req_tx_queues;
3655 	channels->other_count = 0;
3656 	channels->combined_count = 0;
3657 }
3658 
3659 static int ibmvnic_set_channels(struct net_device *netdev,
3660 				struct ethtool_channels *channels)
3661 {
3662 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3663 
3664 	adapter->desired.rx_queues = channels->rx_count;
3665 	adapter->desired.tx_queues = channels->tx_count;
3666 
3667 	return wait_for_reset(adapter);
3668 }
3669 
3670 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3671 {
3672 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3673 	int i;
3674 
3675 	if (stringset != ETH_SS_STATS)
3676 		return;
3677 
3678 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
3679 		memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3680 
3681 	for (i = 0; i < adapter->req_tx_queues; i++) {
3682 		snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3683 		data += ETH_GSTRING_LEN;
3684 
3685 		snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3686 		data += ETH_GSTRING_LEN;
3687 
3688 		snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
3689 		data += ETH_GSTRING_LEN;
3690 	}
3691 
3692 	for (i = 0; i < adapter->req_rx_queues; i++) {
3693 		snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3694 		data += ETH_GSTRING_LEN;
3695 
3696 		snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3697 		data += ETH_GSTRING_LEN;
3698 
3699 		snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3700 		data += ETH_GSTRING_LEN;
3701 	}
3702 }
3703 
3704 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3705 {
3706 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3707 
3708 	switch (sset) {
3709 	case ETH_SS_STATS:
3710 		return ARRAY_SIZE(ibmvnic_stats) +
3711 		       adapter->req_tx_queues * NUM_TX_STATS +
3712 		       adapter->req_rx_queues * NUM_RX_STATS;
3713 	default:
3714 		return -EOPNOTSUPP;
3715 	}
3716 }
3717 
3718 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3719 				      struct ethtool_stats *stats, u64 *data)
3720 {
3721 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3722 	union ibmvnic_crq crq;
3723 	int i, j;
3724 	int rc;
3725 
3726 	memset(&crq, 0, sizeof(crq));
3727 	crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3728 	crq.request_statistics.cmd = REQUEST_STATISTICS;
3729 	crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3730 	crq.request_statistics.len =
3731 	    cpu_to_be32(sizeof(struct ibmvnic_statistics));
3732 
3733 	/* Wait for data to be written */
3734 	reinit_completion(&adapter->stats_done);
3735 	rc = ibmvnic_send_crq(adapter, &crq);
3736 	if (rc)
3737 		return;
3738 	rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3739 	if (rc)
3740 		return;
3741 
3742 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3743 		data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3744 				      (adapter, ibmvnic_stats[i].offset));
3745 
3746 	for (j = 0; j < adapter->req_tx_queues; j++) {
3747 		data[i] = adapter->tx_stats_buffers[j].packets;
3748 		i++;
3749 		data[i] = adapter->tx_stats_buffers[j].bytes;
3750 		i++;
3751 		data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3752 		i++;
3753 	}
3754 
3755 	for (j = 0; j < adapter->req_rx_queues; j++) {
3756 		data[i] = adapter->rx_stats_buffers[j].packets;
3757 		i++;
3758 		data[i] = adapter->rx_stats_buffers[j].bytes;
3759 		i++;
3760 		data[i] = adapter->rx_stats_buffers[j].interrupts;
3761 		i++;
3762 	}
3763 }
3764 
3765 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3766 	.get_drvinfo		= ibmvnic_get_drvinfo,
3767 	.get_msglevel		= ibmvnic_get_msglevel,
3768 	.set_msglevel		= ibmvnic_set_msglevel,
3769 	.get_link		= ibmvnic_get_link,
3770 	.get_ringparam		= ibmvnic_get_ringparam,
3771 	.set_ringparam		= ibmvnic_set_ringparam,
3772 	.get_channels		= ibmvnic_get_channels,
3773 	.set_channels		= ibmvnic_set_channels,
3774 	.get_strings            = ibmvnic_get_strings,
3775 	.get_sset_count         = ibmvnic_get_sset_count,
3776 	.get_ethtool_stats	= ibmvnic_get_ethtool_stats,
3777 	.get_link_ksettings	= ibmvnic_get_link_ksettings,
3778 };
3779 
3780 /* Routines for managing CRQs/sCRQs  */
3781 
3782 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3783 				   struct ibmvnic_sub_crq_queue *scrq)
3784 {
3785 	int rc;
3786 
3787 	if (!scrq) {
3788 		netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3789 		return -EINVAL;
3790 	}
3791 
3792 	if (scrq->irq) {
3793 		free_irq(scrq->irq, scrq);
3794 		irq_dispose_mapping(scrq->irq);
3795 		scrq->irq = 0;
3796 	}
3797 
3798 	if (scrq->msgs) {
3799 		memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3800 		atomic_set(&scrq->used, 0);
3801 		scrq->cur = 0;
3802 		scrq->ind_buf.index = 0;
3803 	} else {
3804 		netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3805 		return -EINVAL;
3806 	}
3807 
3808 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3809 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3810 	return rc;
3811 }
3812 
3813 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3814 {
3815 	int i, rc;
3816 
3817 	if (!adapter->tx_scrq || !adapter->rx_scrq)
3818 		return -EINVAL;
3819 
3820 	ibmvnic_clean_affinity(adapter);
3821 
3822 	for (i = 0; i < adapter->req_tx_queues; i++) {
3823 		netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3824 		rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3825 		if (rc)
3826 			return rc;
3827 	}
3828 
3829 	for (i = 0; i < adapter->req_rx_queues; i++) {
3830 		netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3831 		rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3832 		if (rc)
3833 			return rc;
3834 	}
3835 
3836 	return rc;
3837 }
3838 
3839 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3840 				  struct ibmvnic_sub_crq_queue *scrq,
3841 				  bool do_h_free)
3842 {
3843 	struct device *dev = &adapter->vdev->dev;
3844 	long rc;
3845 
3846 	netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3847 
3848 	if (do_h_free) {
3849 		/* Close the sub-crqs */
3850 		do {
3851 			rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3852 						adapter->vdev->unit_address,
3853 						scrq->crq_num);
3854 		} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3855 
3856 		if (rc) {
3857 			netdev_err(adapter->netdev,
3858 				   "Failed to release sub-CRQ %16lx, rc = %ld\n",
3859 				   scrq->crq_num, rc);
3860 		}
3861 	}
3862 
3863 	dma_free_coherent(dev,
3864 			  IBMVNIC_IND_ARR_SZ,
3865 			  scrq->ind_buf.indir_arr,
3866 			  scrq->ind_buf.indir_dma);
3867 
3868 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3869 			 DMA_BIDIRECTIONAL);
3870 	free_pages((unsigned long)scrq->msgs, 2);
3871 	free_cpumask_var(scrq->affinity_mask);
3872 	kfree(scrq);
3873 }
3874 
3875 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3876 							*adapter)
3877 {
3878 	struct device *dev = &adapter->vdev->dev;
3879 	struct ibmvnic_sub_crq_queue *scrq;
3880 	int rc;
3881 
3882 	scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3883 	if (!scrq)
3884 		return NULL;
3885 
3886 	scrq->msgs =
3887 		(union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3888 	if (!scrq->msgs) {
3889 		dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3890 		goto zero_page_failed;
3891 	}
3892 	if (!zalloc_cpumask_var(&scrq->affinity_mask, GFP_KERNEL))
3893 		goto cpumask_alloc_failed;
3894 
3895 	scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3896 					 DMA_BIDIRECTIONAL);
3897 	if (dma_mapping_error(dev, scrq->msg_token)) {
3898 		dev_warn(dev, "Couldn't map crq queue messages page\n");
3899 		goto map_failed;
3900 	}
3901 
3902 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3903 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3904 
3905 	if (rc == H_RESOURCE)
3906 		rc = ibmvnic_reset_crq(adapter);
3907 
3908 	if (rc == H_CLOSED) {
3909 		dev_warn(dev, "Partner adapter not ready, waiting.\n");
3910 	} else if (rc) {
3911 		dev_warn(dev, "Error %d registering sub-crq\n", rc);
3912 		goto reg_failed;
3913 	}
3914 
3915 	scrq->adapter = adapter;
3916 	scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3917 	scrq->ind_buf.index = 0;
3918 
3919 	scrq->ind_buf.indir_arr =
3920 		dma_alloc_coherent(dev,
3921 				   IBMVNIC_IND_ARR_SZ,
3922 				   &scrq->ind_buf.indir_dma,
3923 				   GFP_KERNEL);
3924 
3925 	if (!scrq->ind_buf.indir_arr)
3926 		goto indir_failed;
3927 
3928 	spin_lock_init(&scrq->lock);
3929 
3930 	netdev_dbg(adapter->netdev,
3931 		   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3932 		   scrq->crq_num, scrq->hw_irq, scrq->irq);
3933 
3934 	return scrq;
3935 
3936 indir_failed:
3937 	do {
3938 		rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3939 					adapter->vdev->unit_address,
3940 					scrq->crq_num);
3941 	} while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3942 reg_failed:
3943 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3944 			 DMA_BIDIRECTIONAL);
3945 map_failed:
3946 	free_cpumask_var(scrq->affinity_mask);
3947 cpumask_alloc_failed:
3948 	free_pages((unsigned long)scrq->msgs, 2);
3949 zero_page_failed:
3950 	kfree(scrq);
3951 
3952 	return NULL;
3953 }
3954 
3955 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3956 {
3957 	int i;
3958 
3959 	ibmvnic_clean_affinity(adapter);
3960 	if (adapter->tx_scrq) {
3961 		for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3962 			if (!adapter->tx_scrq[i])
3963 				continue;
3964 
3965 			netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3966 				   i);
3967 			ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
3968 			if (adapter->tx_scrq[i]->irq) {
3969 				free_irq(adapter->tx_scrq[i]->irq,
3970 					 adapter->tx_scrq[i]);
3971 				irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3972 				adapter->tx_scrq[i]->irq = 0;
3973 			}
3974 
3975 			release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3976 					      do_h_free);
3977 		}
3978 
3979 		kfree(adapter->tx_scrq);
3980 		adapter->tx_scrq = NULL;
3981 		adapter->num_active_tx_scrqs = 0;
3982 	}
3983 
3984 	if (adapter->rx_scrq) {
3985 		for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3986 			if (!adapter->rx_scrq[i])
3987 				continue;
3988 
3989 			netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3990 				   i);
3991 			if (adapter->rx_scrq[i]->irq) {
3992 				free_irq(adapter->rx_scrq[i]->irq,
3993 					 adapter->rx_scrq[i]);
3994 				irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3995 				adapter->rx_scrq[i]->irq = 0;
3996 			}
3997 
3998 			release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3999 					      do_h_free);
4000 		}
4001 
4002 		kfree(adapter->rx_scrq);
4003 		adapter->rx_scrq = NULL;
4004 		adapter->num_active_rx_scrqs = 0;
4005 	}
4006 }
4007 
4008 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
4009 			    struct ibmvnic_sub_crq_queue *scrq)
4010 {
4011 	struct device *dev = &adapter->vdev->dev;
4012 	unsigned long rc;
4013 
4014 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4015 				H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4016 	if (rc)
4017 		dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
4018 			scrq->hw_irq, rc);
4019 	return rc;
4020 }
4021 
4022 /* We can not use the IRQ chip EOI handler because that has the
4023  * unintended effect of changing the interrupt priority.
4024  */
4025 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq)
4026 {
4027 	u64 val = 0xff000000 | scrq->hw_irq;
4028 	unsigned long rc;
4029 
4030 	rc = plpar_hcall_norets(H_EOI, val);
4031 	if (rc)
4032 		dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc);
4033 }
4034 
4035 /* Due to a firmware bug, the hypervisor can send an interrupt to a
4036  * transmit or receive queue just prior to a partition migration.
4037  * Force an EOI after migration.
4038  */
4039 static void ibmvnic_clear_pending_interrupt(struct device *dev,
4040 					    struct ibmvnic_sub_crq_queue *scrq)
4041 {
4042 	if (!xive_enabled())
4043 		ibmvnic_xics_eoi(dev, scrq);
4044 }
4045 
4046 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
4047 			   struct ibmvnic_sub_crq_queue *scrq)
4048 {
4049 	struct device *dev = &adapter->vdev->dev;
4050 	unsigned long rc;
4051 
4052 	if (scrq->hw_irq > 0x100000000ULL) {
4053 		dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
4054 		return 1;
4055 	}
4056 
4057 	if (test_bit(0, &adapter->resetting) &&
4058 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
4059 		ibmvnic_clear_pending_interrupt(dev, scrq);
4060 	}
4061 
4062 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4063 				H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4064 	if (rc)
4065 		dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
4066 			scrq->hw_irq, rc);
4067 	return rc;
4068 }
4069 
4070 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
4071 			       struct ibmvnic_sub_crq_queue *scrq)
4072 {
4073 	struct device *dev = &adapter->vdev->dev;
4074 	struct ibmvnic_tx_pool *tx_pool;
4075 	struct ibmvnic_tx_buff *txbuff;
4076 	struct netdev_queue *txq;
4077 	union sub_crq *next;
4078 	int index;
4079 	int i;
4080 
4081 restart_loop:
4082 	while (pending_scrq(adapter, scrq)) {
4083 		unsigned int pool = scrq->pool_index;
4084 		int num_entries = 0;
4085 		int total_bytes = 0;
4086 		int num_packets = 0;
4087 
4088 		next = ibmvnic_next_scrq(adapter, scrq);
4089 		for (i = 0; i < next->tx_comp.num_comps; i++) {
4090 			index = be32_to_cpu(next->tx_comp.correlators[i]);
4091 			if (index & IBMVNIC_TSO_POOL_MASK) {
4092 				tx_pool = &adapter->tso_pool[pool];
4093 				index &= ~IBMVNIC_TSO_POOL_MASK;
4094 			} else {
4095 				tx_pool = &adapter->tx_pool[pool];
4096 			}
4097 
4098 			txbuff = &tx_pool->tx_buff[index];
4099 			num_packets++;
4100 			num_entries += txbuff->num_entries;
4101 			if (txbuff->skb) {
4102 				total_bytes += txbuff->skb->len;
4103 				if (next->tx_comp.rcs[i]) {
4104 					dev_err(dev, "tx error %x\n",
4105 						next->tx_comp.rcs[i]);
4106 					dev_kfree_skb_irq(txbuff->skb);
4107 				} else {
4108 					dev_consume_skb_irq(txbuff->skb);
4109 				}
4110 				txbuff->skb = NULL;
4111 			} else {
4112 				netdev_warn(adapter->netdev,
4113 					    "TX completion received with NULL socket buffer\n");
4114 			}
4115 			tx_pool->free_map[tx_pool->producer_index] = index;
4116 			tx_pool->producer_index =
4117 				(tx_pool->producer_index + 1) %
4118 					tx_pool->num_buffers;
4119 		}
4120 		/* remove tx_comp scrq*/
4121 		next->tx_comp.first = 0;
4122 
4123 		txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
4124 		netdev_tx_completed_queue(txq, num_packets, total_bytes);
4125 
4126 		if (atomic_sub_return(num_entries, &scrq->used) <=
4127 		    (adapter->req_tx_entries_per_subcrq / 2) &&
4128 		    __netif_subqueue_stopped(adapter->netdev,
4129 					     scrq->pool_index)) {
4130 			rcu_read_lock();
4131 			if (adapter->tx_queues_active) {
4132 				netif_wake_subqueue(adapter->netdev,
4133 						    scrq->pool_index);
4134 				netdev_dbg(adapter->netdev,
4135 					   "Started queue %d\n",
4136 					   scrq->pool_index);
4137 			}
4138 			rcu_read_unlock();
4139 		}
4140 	}
4141 
4142 	enable_scrq_irq(adapter, scrq);
4143 
4144 	if (pending_scrq(adapter, scrq)) {
4145 		disable_scrq_irq(adapter, scrq);
4146 		goto restart_loop;
4147 	}
4148 
4149 	return 0;
4150 }
4151 
4152 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
4153 {
4154 	struct ibmvnic_sub_crq_queue *scrq = instance;
4155 	struct ibmvnic_adapter *adapter = scrq->adapter;
4156 
4157 	disable_scrq_irq(adapter, scrq);
4158 	ibmvnic_complete_tx(adapter, scrq);
4159 
4160 	return IRQ_HANDLED;
4161 }
4162 
4163 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
4164 {
4165 	struct ibmvnic_sub_crq_queue *scrq = instance;
4166 	struct ibmvnic_adapter *adapter = scrq->adapter;
4167 
4168 	/* When booting a kdump kernel we can hit pending interrupts
4169 	 * prior to completing driver initialization.
4170 	 */
4171 	if (unlikely(adapter->state != VNIC_OPEN))
4172 		return IRQ_NONE;
4173 
4174 	adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
4175 
4176 	if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
4177 		disable_scrq_irq(adapter, scrq);
4178 		__napi_schedule(&adapter->napi[scrq->scrq_num]);
4179 	}
4180 
4181 	return IRQ_HANDLED;
4182 }
4183 
4184 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
4185 {
4186 	struct device *dev = &adapter->vdev->dev;
4187 	struct ibmvnic_sub_crq_queue *scrq;
4188 	int i = 0, j = 0;
4189 	int rc = 0;
4190 
4191 	for (i = 0; i < adapter->req_tx_queues; i++) {
4192 		netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
4193 			   i);
4194 		scrq = adapter->tx_scrq[i];
4195 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4196 
4197 		if (!scrq->irq) {
4198 			rc = -EINVAL;
4199 			dev_err(dev, "Error mapping irq\n");
4200 			goto req_tx_irq_failed;
4201 		}
4202 
4203 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
4204 			 adapter->vdev->unit_address, i);
4205 		rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
4206 				 0, scrq->name, scrq);
4207 
4208 		if (rc) {
4209 			dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
4210 				scrq->irq, rc);
4211 			irq_dispose_mapping(scrq->irq);
4212 			goto req_tx_irq_failed;
4213 		}
4214 	}
4215 
4216 	for (i = 0; i < adapter->req_rx_queues; i++) {
4217 		netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
4218 			   i);
4219 		scrq = adapter->rx_scrq[i];
4220 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4221 		if (!scrq->irq) {
4222 			rc = -EINVAL;
4223 			dev_err(dev, "Error mapping irq\n");
4224 			goto req_rx_irq_failed;
4225 		}
4226 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
4227 			 adapter->vdev->unit_address, i);
4228 		rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
4229 				 0, scrq->name, scrq);
4230 		if (rc) {
4231 			dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
4232 				scrq->irq, rc);
4233 			irq_dispose_mapping(scrq->irq);
4234 			goto req_rx_irq_failed;
4235 		}
4236 	}
4237 
4238 	cpus_read_lock();
4239 	ibmvnic_set_affinity(adapter);
4240 	cpus_read_unlock();
4241 
4242 	return rc;
4243 
4244 req_rx_irq_failed:
4245 	for (j = 0; j < i; j++) {
4246 		free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
4247 		irq_dispose_mapping(adapter->rx_scrq[j]->irq);
4248 	}
4249 	i = adapter->req_tx_queues;
4250 req_tx_irq_failed:
4251 	for (j = 0; j < i; j++) {
4252 		free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
4253 		irq_dispose_mapping(adapter->tx_scrq[j]->irq);
4254 	}
4255 	release_sub_crqs(adapter, 1);
4256 	return rc;
4257 }
4258 
4259 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
4260 {
4261 	struct device *dev = &adapter->vdev->dev;
4262 	struct ibmvnic_sub_crq_queue **allqueues;
4263 	int registered_queues = 0;
4264 	int total_queues;
4265 	int more = 0;
4266 	int i;
4267 
4268 	total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
4269 
4270 	allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
4271 	if (!allqueues)
4272 		return -ENOMEM;
4273 
4274 	for (i = 0; i < total_queues; i++) {
4275 		allqueues[i] = init_sub_crq_queue(adapter);
4276 		if (!allqueues[i]) {
4277 			dev_warn(dev, "Couldn't allocate all sub-crqs\n");
4278 			break;
4279 		}
4280 		registered_queues++;
4281 	}
4282 
4283 	/* Make sure we were able to register the minimum number of queues */
4284 	if (registered_queues <
4285 	    adapter->min_tx_queues + adapter->min_rx_queues) {
4286 		dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
4287 		goto tx_failed;
4288 	}
4289 
4290 	/* Distribute the failed allocated queues*/
4291 	for (i = 0; i < total_queues - registered_queues + more ; i++) {
4292 		netdev_dbg(adapter->netdev, "Reducing number of queues\n");
4293 		switch (i % 3) {
4294 		case 0:
4295 			if (adapter->req_rx_queues > adapter->min_rx_queues)
4296 				adapter->req_rx_queues--;
4297 			else
4298 				more++;
4299 			break;
4300 		case 1:
4301 			if (adapter->req_tx_queues > adapter->min_tx_queues)
4302 				adapter->req_tx_queues--;
4303 			else
4304 				more++;
4305 			break;
4306 		}
4307 	}
4308 
4309 	adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
4310 				   sizeof(*adapter->tx_scrq), GFP_KERNEL);
4311 	if (!adapter->tx_scrq)
4312 		goto tx_failed;
4313 
4314 	for (i = 0; i < adapter->req_tx_queues; i++) {
4315 		adapter->tx_scrq[i] = allqueues[i];
4316 		adapter->tx_scrq[i]->pool_index = i;
4317 		adapter->num_active_tx_scrqs++;
4318 	}
4319 
4320 	adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
4321 				   sizeof(*adapter->rx_scrq), GFP_KERNEL);
4322 	if (!adapter->rx_scrq)
4323 		goto rx_failed;
4324 
4325 	for (i = 0; i < adapter->req_rx_queues; i++) {
4326 		adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
4327 		adapter->rx_scrq[i]->scrq_num = i;
4328 		adapter->num_active_rx_scrqs++;
4329 	}
4330 
4331 	kfree(allqueues);
4332 	return 0;
4333 
4334 rx_failed:
4335 	kfree(adapter->tx_scrq);
4336 	adapter->tx_scrq = NULL;
4337 tx_failed:
4338 	for (i = 0; i < registered_queues; i++)
4339 		release_sub_crq_queue(adapter, allqueues[i], 1);
4340 	kfree(allqueues);
4341 	return -ENOMEM;
4342 }
4343 
4344 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
4345 {
4346 	struct device *dev = &adapter->vdev->dev;
4347 	union ibmvnic_crq crq;
4348 	int max_entries;
4349 	int cap_reqs;
4350 
4351 	/* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
4352 	 * the PROMISC flag). Initialize this count upfront. When the tasklet
4353 	 * receives a response to all of these, it will send the next protocol
4354 	 * message (QUERY_IP_OFFLOAD).
4355 	 */
4356 	if (!(adapter->netdev->flags & IFF_PROMISC) ||
4357 	    adapter->promisc_supported)
4358 		cap_reqs = 7;
4359 	else
4360 		cap_reqs = 6;
4361 
4362 	if (!retry) {
4363 		/* Sub-CRQ entries are 32 byte long */
4364 		int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
4365 
4366 		atomic_set(&adapter->running_cap_crqs, cap_reqs);
4367 
4368 		if (adapter->min_tx_entries_per_subcrq > entries_page ||
4369 		    adapter->min_rx_add_entries_per_subcrq > entries_page) {
4370 			dev_err(dev, "Fatal, invalid entries per sub-crq\n");
4371 			return;
4372 		}
4373 
4374 		if (adapter->desired.mtu)
4375 			adapter->req_mtu = adapter->desired.mtu;
4376 		else
4377 			adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
4378 
4379 		if (!adapter->desired.tx_entries)
4380 			adapter->desired.tx_entries =
4381 					adapter->max_tx_entries_per_subcrq;
4382 		if (!adapter->desired.rx_entries)
4383 			adapter->desired.rx_entries =
4384 					adapter->max_rx_add_entries_per_subcrq;
4385 
4386 		max_entries = IBMVNIC_LTB_SET_SIZE /
4387 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
4388 
4389 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4390 			adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) {
4391 			adapter->desired.tx_entries = max_entries;
4392 		}
4393 
4394 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4395 			adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) {
4396 			adapter->desired.rx_entries = max_entries;
4397 		}
4398 
4399 		if (adapter->desired.tx_entries)
4400 			adapter->req_tx_entries_per_subcrq =
4401 					adapter->desired.tx_entries;
4402 		else
4403 			adapter->req_tx_entries_per_subcrq =
4404 					adapter->max_tx_entries_per_subcrq;
4405 
4406 		if (adapter->desired.rx_entries)
4407 			adapter->req_rx_add_entries_per_subcrq =
4408 					adapter->desired.rx_entries;
4409 		else
4410 			adapter->req_rx_add_entries_per_subcrq =
4411 					adapter->max_rx_add_entries_per_subcrq;
4412 
4413 		if (adapter->desired.tx_queues)
4414 			adapter->req_tx_queues =
4415 					adapter->desired.tx_queues;
4416 		else
4417 			adapter->req_tx_queues =
4418 					adapter->opt_tx_comp_sub_queues;
4419 
4420 		if (adapter->desired.rx_queues)
4421 			adapter->req_rx_queues =
4422 					adapter->desired.rx_queues;
4423 		else
4424 			adapter->req_rx_queues =
4425 					adapter->opt_rx_comp_queues;
4426 
4427 		adapter->req_rx_add_queues = adapter->max_rx_add_queues;
4428 	} else {
4429 		atomic_add(cap_reqs, &adapter->running_cap_crqs);
4430 	}
4431 	memset(&crq, 0, sizeof(crq));
4432 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
4433 	crq.request_capability.cmd = REQUEST_CAPABILITY;
4434 
4435 	crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
4436 	crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
4437 	cap_reqs--;
4438 	ibmvnic_send_crq(adapter, &crq);
4439 
4440 	crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
4441 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
4442 	cap_reqs--;
4443 	ibmvnic_send_crq(adapter, &crq);
4444 
4445 	crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
4446 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
4447 	cap_reqs--;
4448 	ibmvnic_send_crq(adapter, &crq);
4449 
4450 	crq.request_capability.capability =
4451 	    cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
4452 	crq.request_capability.number =
4453 	    cpu_to_be64(adapter->req_tx_entries_per_subcrq);
4454 	cap_reqs--;
4455 	ibmvnic_send_crq(adapter, &crq);
4456 
4457 	crq.request_capability.capability =
4458 	    cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
4459 	crq.request_capability.number =
4460 	    cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
4461 	cap_reqs--;
4462 	ibmvnic_send_crq(adapter, &crq);
4463 
4464 	crq.request_capability.capability = cpu_to_be16(REQ_MTU);
4465 	crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
4466 	cap_reqs--;
4467 	ibmvnic_send_crq(adapter, &crq);
4468 
4469 	if (adapter->netdev->flags & IFF_PROMISC) {
4470 		if (adapter->promisc_supported) {
4471 			crq.request_capability.capability =
4472 			    cpu_to_be16(PROMISC_REQUESTED);
4473 			crq.request_capability.number = cpu_to_be64(1);
4474 			cap_reqs--;
4475 			ibmvnic_send_crq(adapter, &crq);
4476 		}
4477 	} else {
4478 		crq.request_capability.capability =
4479 		    cpu_to_be16(PROMISC_REQUESTED);
4480 		crq.request_capability.number = cpu_to_be64(0);
4481 		cap_reqs--;
4482 		ibmvnic_send_crq(adapter, &crq);
4483 	}
4484 
4485 	/* Keep at end to catch any discrepancy between expected and actual
4486 	 * CRQs sent.
4487 	 */
4488 	WARN_ON(cap_reqs != 0);
4489 }
4490 
4491 static int pending_scrq(struct ibmvnic_adapter *adapter,
4492 			struct ibmvnic_sub_crq_queue *scrq)
4493 {
4494 	union sub_crq *entry = &scrq->msgs[scrq->cur];
4495 	int rc;
4496 
4497 	rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4498 
4499 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4500 	 * contents of the SCRQ descriptor
4501 	 */
4502 	dma_rmb();
4503 
4504 	return rc;
4505 }
4506 
4507 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4508 					struct ibmvnic_sub_crq_queue *scrq)
4509 {
4510 	union sub_crq *entry;
4511 	unsigned long flags;
4512 
4513 	spin_lock_irqsave(&scrq->lock, flags);
4514 	entry = &scrq->msgs[scrq->cur];
4515 	if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4516 		if (++scrq->cur == scrq->size)
4517 			scrq->cur = 0;
4518 	} else {
4519 		entry = NULL;
4520 	}
4521 	spin_unlock_irqrestore(&scrq->lock, flags);
4522 
4523 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4524 	 * contents of the SCRQ descriptor
4525 	 */
4526 	dma_rmb();
4527 
4528 	return entry;
4529 }
4530 
4531 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4532 {
4533 	struct ibmvnic_crq_queue *queue = &adapter->crq;
4534 	union ibmvnic_crq *crq;
4535 
4536 	crq = &queue->msgs[queue->cur];
4537 	if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4538 		if (++queue->cur == queue->size)
4539 			queue->cur = 0;
4540 	} else {
4541 		crq = NULL;
4542 	}
4543 
4544 	return crq;
4545 }
4546 
4547 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4548 {
4549 	switch (rc) {
4550 	case H_PARAMETER:
4551 		dev_warn_ratelimited(dev,
4552 				     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4553 				     func, rc);
4554 		break;
4555 	case H_CLOSED:
4556 		dev_warn_ratelimited(dev,
4557 				     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4558 				     func, rc);
4559 		break;
4560 	default:
4561 		dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4562 		break;
4563 	}
4564 }
4565 
4566 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4567 				u64 remote_handle, u64 ioba, u64 num_entries)
4568 {
4569 	unsigned int ua = adapter->vdev->unit_address;
4570 	struct device *dev = &adapter->vdev->dev;
4571 	int rc;
4572 
4573 	/* Make sure the hypervisor sees the complete request */
4574 	dma_wmb();
4575 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4576 				cpu_to_be64(remote_handle),
4577 				ioba, num_entries);
4578 
4579 	if (rc)
4580 		print_subcrq_error(dev, rc, __func__);
4581 
4582 	return rc;
4583 }
4584 
4585 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4586 			    union ibmvnic_crq *crq)
4587 {
4588 	unsigned int ua = adapter->vdev->unit_address;
4589 	struct device *dev = &adapter->vdev->dev;
4590 	u64 *u64_crq = (u64 *)crq;
4591 	int rc;
4592 
4593 	netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4594 		   (unsigned long)cpu_to_be64(u64_crq[0]),
4595 		   (unsigned long)cpu_to_be64(u64_crq[1]));
4596 
4597 	if (!adapter->crq.active &&
4598 	    crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4599 		dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4600 		return -EINVAL;
4601 	}
4602 
4603 	/* Make sure the hypervisor sees the complete request */
4604 	dma_wmb();
4605 
4606 	rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4607 				cpu_to_be64(u64_crq[0]),
4608 				cpu_to_be64(u64_crq[1]));
4609 
4610 	if (rc) {
4611 		if (rc == H_CLOSED) {
4612 			dev_warn(dev, "CRQ Queue closed\n");
4613 			/* do not reset, report the fail, wait for passive init from server */
4614 		}
4615 
4616 		dev_warn(dev, "Send error (rc=%d)\n", rc);
4617 	}
4618 
4619 	return rc;
4620 }
4621 
4622 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4623 {
4624 	struct device *dev = &adapter->vdev->dev;
4625 	union ibmvnic_crq crq;
4626 	int retries = 100;
4627 	int rc;
4628 
4629 	memset(&crq, 0, sizeof(crq));
4630 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4631 	crq.generic.cmd = IBMVNIC_CRQ_INIT;
4632 	netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4633 
4634 	do {
4635 		rc = ibmvnic_send_crq(adapter, &crq);
4636 		if (rc != H_CLOSED)
4637 			break;
4638 		retries--;
4639 		msleep(50);
4640 
4641 	} while (retries > 0);
4642 
4643 	if (rc) {
4644 		dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4645 		return rc;
4646 	}
4647 
4648 	return 0;
4649 }
4650 
4651 struct vnic_login_client_data {
4652 	u8	type;
4653 	__be16	len;
4654 	char	name[];
4655 } __packed;
4656 
4657 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4658 {
4659 	int len;
4660 
4661 	/* Calculate the amount of buffer space needed for the
4662 	 * vnic client data in the login buffer. There are four entries,
4663 	 * OS name, LPAR name, device name, and a null last entry.
4664 	 */
4665 	len = 4 * sizeof(struct vnic_login_client_data);
4666 	len += 6; /* "Linux" plus NULL */
4667 	len += strlen(utsname()->nodename) + 1;
4668 	len += strlen(adapter->netdev->name) + 1;
4669 
4670 	return len;
4671 }
4672 
4673 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4674 				 struct vnic_login_client_data *vlcd)
4675 {
4676 	const char *os_name = "Linux";
4677 	int len;
4678 
4679 	/* Type 1 - LPAR OS */
4680 	vlcd->type = 1;
4681 	len = strlen(os_name) + 1;
4682 	vlcd->len = cpu_to_be16(len);
4683 	strscpy(vlcd->name, os_name, len);
4684 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4685 
4686 	/* Type 2 - LPAR name */
4687 	vlcd->type = 2;
4688 	len = strlen(utsname()->nodename) + 1;
4689 	vlcd->len = cpu_to_be16(len);
4690 	strscpy(vlcd->name, utsname()->nodename, len);
4691 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4692 
4693 	/* Type 3 - device name */
4694 	vlcd->type = 3;
4695 	len = strlen(adapter->netdev->name) + 1;
4696 	vlcd->len = cpu_to_be16(len);
4697 	strscpy(vlcd->name, adapter->netdev->name, len);
4698 }
4699 
4700 static int send_login(struct ibmvnic_adapter *adapter)
4701 {
4702 	struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4703 	struct ibmvnic_login_buffer *login_buffer;
4704 	struct device *dev = &adapter->vdev->dev;
4705 	struct vnic_login_client_data *vlcd;
4706 	dma_addr_t rsp_buffer_token;
4707 	dma_addr_t buffer_token;
4708 	size_t rsp_buffer_size;
4709 	union ibmvnic_crq crq;
4710 	int client_data_len;
4711 	size_t buffer_size;
4712 	__be64 *tx_list_p;
4713 	__be64 *rx_list_p;
4714 	int rc;
4715 	int i;
4716 
4717 	if (!adapter->tx_scrq || !adapter->rx_scrq) {
4718 		netdev_err(adapter->netdev,
4719 			   "RX or TX queues are not allocated, device login failed\n");
4720 		return -ENOMEM;
4721 	}
4722 
4723 	release_login_buffer(adapter);
4724 	release_login_rsp_buffer(adapter);
4725 
4726 	client_data_len = vnic_client_data_len(adapter);
4727 
4728 	buffer_size =
4729 	    sizeof(struct ibmvnic_login_buffer) +
4730 	    sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4731 	    client_data_len;
4732 
4733 	login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4734 	if (!login_buffer)
4735 		goto buf_alloc_failed;
4736 
4737 	buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4738 				      DMA_TO_DEVICE);
4739 	if (dma_mapping_error(dev, buffer_token)) {
4740 		dev_err(dev, "Couldn't map login buffer\n");
4741 		goto buf_map_failed;
4742 	}
4743 
4744 	rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4745 			  sizeof(u64) * adapter->req_tx_queues +
4746 			  sizeof(u64) * adapter->req_rx_queues +
4747 			  sizeof(u64) * adapter->req_rx_queues +
4748 			  sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4749 
4750 	login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4751 	if (!login_rsp_buffer)
4752 		goto buf_rsp_alloc_failed;
4753 
4754 	rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4755 					  rsp_buffer_size, DMA_FROM_DEVICE);
4756 	if (dma_mapping_error(dev, rsp_buffer_token)) {
4757 		dev_err(dev, "Couldn't map login rsp buffer\n");
4758 		goto buf_rsp_map_failed;
4759 	}
4760 
4761 	adapter->login_buf = login_buffer;
4762 	adapter->login_buf_token = buffer_token;
4763 	adapter->login_buf_sz = buffer_size;
4764 	adapter->login_rsp_buf = login_rsp_buffer;
4765 	adapter->login_rsp_buf_token = rsp_buffer_token;
4766 	adapter->login_rsp_buf_sz = rsp_buffer_size;
4767 
4768 	login_buffer->len = cpu_to_be32(buffer_size);
4769 	login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4770 	login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4771 	login_buffer->off_txcomp_subcrqs =
4772 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4773 	login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4774 	login_buffer->off_rxcomp_subcrqs =
4775 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4776 			sizeof(u64) * adapter->req_tx_queues);
4777 	login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4778 	login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4779 
4780 	tx_list_p = (__be64 *)((char *)login_buffer +
4781 				      sizeof(struct ibmvnic_login_buffer));
4782 	rx_list_p = (__be64 *)((char *)login_buffer +
4783 				      sizeof(struct ibmvnic_login_buffer) +
4784 				      sizeof(u64) * adapter->req_tx_queues);
4785 
4786 	for (i = 0; i < adapter->req_tx_queues; i++) {
4787 		if (adapter->tx_scrq[i]) {
4788 			tx_list_p[i] =
4789 				cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4790 		}
4791 	}
4792 
4793 	for (i = 0; i < adapter->req_rx_queues; i++) {
4794 		if (adapter->rx_scrq[i]) {
4795 			rx_list_p[i] =
4796 				cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4797 		}
4798 	}
4799 
4800 	/* Insert vNIC login client data */
4801 	vlcd = (struct vnic_login_client_data *)
4802 		((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4803 	login_buffer->client_data_offset =
4804 			cpu_to_be32((char *)vlcd - (char *)login_buffer);
4805 	login_buffer->client_data_len = cpu_to_be32(client_data_len);
4806 
4807 	vnic_add_client_data(adapter, vlcd);
4808 
4809 	netdev_dbg(adapter->netdev, "Login Buffer:\n");
4810 	for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4811 		netdev_dbg(adapter->netdev, "%016lx\n",
4812 			   ((unsigned long *)(adapter->login_buf))[i]);
4813 	}
4814 
4815 	memset(&crq, 0, sizeof(crq));
4816 	crq.login.first = IBMVNIC_CRQ_CMD;
4817 	crq.login.cmd = LOGIN;
4818 	crq.login.ioba = cpu_to_be32(buffer_token);
4819 	crq.login.len = cpu_to_be32(buffer_size);
4820 
4821 	adapter->login_pending = true;
4822 	rc = ibmvnic_send_crq(adapter, &crq);
4823 	if (rc) {
4824 		adapter->login_pending = false;
4825 		netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4826 		goto buf_rsp_map_failed;
4827 	}
4828 
4829 	return 0;
4830 
4831 buf_rsp_map_failed:
4832 	kfree(login_rsp_buffer);
4833 	adapter->login_rsp_buf = NULL;
4834 buf_rsp_alloc_failed:
4835 	dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4836 buf_map_failed:
4837 	kfree(login_buffer);
4838 	adapter->login_buf = NULL;
4839 buf_alloc_failed:
4840 	return -ENOMEM;
4841 }
4842 
4843 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4844 			    u32 len, u8 map_id)
4845 {
4846 	union ibmvnic_crq crq;
4847 
4848 	memset(&crq, 0, sizeof(crq));
4849 	crq.request_map.first = IBMVNIC_CRQ_CMD;
4850 	crq.request_map.cmd = REQUEST_MAP;
4851 	crq.request_map.map_id = map_id;
4852 	crq.request_map.ioba = cpu_to_be32(addr);
4853 	crq.request_map.len = cpu_to_be32(len);
4854 	return ibmvnic_send_crq(adapter, &crq);
4855 }
4856 
4857 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4858 {
4859 	union ibmvnic_crq crq;
4860 
4861 	memset(&crq, 0, sizeof(crq));
4862 	crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4863 	crq.request_unmap.cmd = REQUEST_UNMAP;
4864 	crq.request_unmap.map_id = map_id;
4865 	return ibmvnic_send_crq(adapter, &crq);
4866 }
4867 
4868 static void send_query_map(struct ibmvnic_adapter *adapter)
4869 {
4870 	union ibmvnic_crq crq;
4871 
4872 	memset(&crq, 0, sizeof(crq));
4873 	crq.query_map.first = IBMVNIC_CRQ_CMD;
4874 	crq.query_map.cmd = QUERY_MAP;
4875 	ibmvnic_send_crq(adapter, &crq);
4876 }
4877 
4878 /* Send a series of CRQs requesting various capabilities of the VNIC server */
4879 static void send_query_cap(struct ibmvnic_adapter *adapter)
4880 {
4881 	union ibmvnic_crq crq;
4882 	int cap_reqs;
4883 
4884 	/* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
4885 	 * upfront. When the tasklet receives a response to all of these, it
4886 	 * can send out the next protocol messaage (REQUEST_CAPABILITY).
4887 	 */
4888 	cap_reqs = 25;
4889 
4890 	atomic_set(&adapter->running_cap_crqs, cap_reqs);
4891 
4892 	memset(&crq, 0, sizeof(crq));
4893 	crq.query_capability.first = IBMVNIC_CRQ_CMD;
4894 	crq.query_capability.cmd = QUERY_CAPABILITY;
4895 
4896 	crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4897 	ibmvnic_send_crq(adapter, &crq);
4898 	cap_reqs--;
4899 
4900 	crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4901 	ibmvnic_send_crq(adapter, &crq);
4902 	cap_reqs--;
4903 
4904 	crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4905 	ibmvnic_send_crq(adapter, &crq);
4906 	cap_reqs--;
4907 
4908 	crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4909 	ibmvnic_send_crq(adapter, &crq);
4910 	cap_reqs--;
4911 
4912 	crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4913 	ibmvnic_send_crq(adapter, &crq);
4914 	cap_reqs--;
4915 
4916 	crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4917 	ibmvnic_send_crq(adapter, &crq);
4918 	cap_reqs--;
4919 
4920 	crq.query_capability.capability =
4921 	    cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4922 	ibmvnic_send_crq(adapter, &crq);
4923 	cap_reqs--;
4924 
4925 	crq.query_capability.capability =
4926 	    cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4927 	ibmvnic_send_crq(adapter, &crq);
4928 	cap_reqs--;
4929 
4930 	crq.query_capability.capability =
4931 	    cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4932 	ibmvnic_send_crq(adapter, &crq);
4933 	cap_reqs--;
4934 
4935 	crq.query_capability.capability =
4936 	    cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4937 	ibmvnic_send_crq(adapter, &crq);
4938 	cap_reqs--;
4939 
4940 	crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4941 	ibmvnic_send_crq(adapter, &crq);
4942 	cap_reqs--;
4943 
4944 	crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4945 	ibmvnic_send_crq(adapter, &crq);
4946 	cap_reqs--;
4947 
4948 	crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4949 	ibmvnic_send_crq(adapter, &crq);
4950 	cap_reqs--;
4951 
4952 	crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4953 	ibmvnic_send_crq(adapter, &crq);
4954 	cap_reqs--;
4955 
4956 	crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4957 	ibmvnic_send_crq(adapter, &crq);
4958 	cap_reqs--;
4959 
4960 	crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4961 	ibmvnic_send_crq(adapter, &crq);
4962 	cap_reqs--;
4963 
4964 	crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4965 	ibmvnic_send_crq(adapter, &crq);
4966 	cap_reqs--;
4967 
4968 	crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4969 	ibmvnic_send_crq(adapter, &crq);
4970 	cap_reqs--;
4971 
4972 	crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4973 	ibmvnic_send_crq(adapter, &crq);
4974 	cap_reqs--;
4975 
4976 	crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4977 	ibmvnic_send_crq(adapter, &crq);
4978 	cap_reqs--;
4979 
4980 	crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4981 	ibmvnic_send_crq(adapter, &crq);
4982 	cap_reqs--;
4983 
4984 	crq.query_capability.capability =
4985 			cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4986 	ibmvnic_send_crq(adapter, &crq);
4987 	cap_reqs--;
4988 
4989 	crq.query_capability.capability =
4990 			cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4991 	ibmvnic_send_crq(adapter, &crq);
4992 	cap_reqs--;
4993 
4994 	crq.query_capability.capability =
4995 			cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4996 	ibmvnic_send_crq(adapter, &crq);
4997 	cap_reqs--;
4998 
4999 	crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
5000 
5001 	ibmvnic_send_crq(adapter, &crq);
5002 	cap_reqs--;
5003 
5004 	/* Keep at end to catch any discrepancy between expected and actual
5005 	 * CRQs sent.
5006 	 */
5007 	WARN_ON(cap_reqs != 0);
5008 }
5009 
5010 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
5011 {
5012 	int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
5013 	struct device *dev = &adapter->vdev->dev;
5014 	union ibmvnic_crq crq;
5015 
5016 	adapter->ip_offload_tok =
5017 		dma_map_single(dev,
5018 			       &adapter->ip_offload_buf,
5019 			       buf_sz,
5020 			       DMA_FROM_DEVICE);
5021 
5022 	if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
5023 		if (!firmware_has_feature(FW_FEATURE_CMO))
5024 			dev_err(dev, "Couldn't map offload buffer\n");
5025 		return;
5026 	}
5027 
5028 	memset(&crq, 0, sizeof(crq));
5029 	crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
5030 	crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
5031 	crq.query_ip_offload.len = cpu_to_be32(buf_sz);
5032 	crq.query_ip_offload.ioba =
5033 	    cpu_to_be32(adapter->ip_offload_tok);
5034 
5035 	ibmvnic_send_crq(adapter, &crq);
5036 }
5037 
5038 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
5039 {
5040 	struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
5041 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5042 	struct device *dev = &adapter->vdev->dev;
5043 	netdev_features_t old_hw_features = 0;
5044 	union ibmvnic_crq crq;
5045 
5046 	adapter->ip_offload_ctrl_tok =
5047 		dma_map_single(dev,
5048 			       ctrl_buf,
5049 			       sizeof(adapter->ip_offload_ctrl),
5050 			       DMA_TO_DEVICE);
5051 
5052 	if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
5053 		dev_err(dev, "Couldn't map ip offload control buffer\n");
5054 		return;
5055 	}
5056 
5057 	ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5058 	ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
5059 	ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
5060 	ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
5061 	ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
5062 	ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
5063 	ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
5064 	ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
5065 	ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
5066 	ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
5067 
5068 	/* large_rx disabled for now, additional features needed */
5069 	ctrl_buf->large_rx_ipv4 = 0;
5070 	ctrl_buf->large_rx_ipv6 = 0;
5071 
5072 	if (adapter->state != VNIC_PROBING) {
5073 		old_hw_features = adapter->netdev->hw_features;
5074 		adapter->netdev->hw_features = 0;
5075 	}
5076 
5077 	adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
5078 
5079 	if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
5080 		adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
5081 
5082 	if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
5083 		adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
5084 
5085 	if ((adapter->netdev->features &
5086 	    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
5087 		adapter->netdev->hw_features |= NETIF_F_RXCSUM;
5088 
5089 	if (buf->large_tx_ipv4)
5090 		adapter->netdev->hw_features |= NETIF_F_TSO;
5091 	if (buf->large_tx_ipv6)
5092 		adapter->netdev->hw_features |= NETIF_F_TSO6;
5093 
5094 	if (adapter->state == VNIC_PROBING) {
5095 		adapter->netdev->features |= adapter->netdev->hw_features;
5096 	} else if (old_hw_features != adapter->netdev->hw_features) {
5097 		netdev_features_t tmp = 0;
5098 
5099 		/* disable features no longer supported */
5100 		adapter->netdev->features &= adapter->netdev->hw_features;
5101 		/* turn on features now supported if previously enabled */
5102 		tmp = (old_hw_features ^ adapter->netdev->hw_features) &
5103 			adapter->netdev->hw_features;
5104 		adapter->netdev->features |=
5105 				tmp & adapter->netdev->wanted_features;
5106 	}
5107 
5108 	memset(&crq, 0, sizeof(crq));
5109 	crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
5110 	crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
5111 	crq.control_ip_offload.len =
5112 	    cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5113 	crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
5114 	ibmvnic_send_crq(adapter, &crq);
5115 }
5116 
5117 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
5118 				struct ibmvnic_adapter *adapter)
5119 {
5120 	struct device *dev = &adapter->vdev->dev;
5121 
5122 	if (crq->get_vpd_size_rsp.rc.code) {
5123 		dev_err(dev, "Error retrieving VPD size, rc=%x\n",
5124 			crq->get_vpd_size_rsp.rc.code);
5125 		complete(&adapter->fw_done);
5126 		return;
5127 	}
5128 
5129 	adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
5130 	complete(&adapter->fw_done);
5131 }
5132 
5133 static void handle_vpd_rsp(union ibmvnic_crq *crq,
5134 			   struct ibmvnic_adapter *adapter)
5135 {
5136 	struct device *dev = &adapter->vdev->dev;
5137 	unsigned char *substr = NULL;
5138 	u8 fw_level_len = 0;
5139 
5140 	memset(adapter->fw_version, 0, 32);
5141 
5142 	dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
5143 			 DMA_FROM_DEVICE);
5144 
5145 	if (crq->get_vpd_rsp.rc.code) {
5146 		dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
5147 			crq->get_vpd_rsp.rc.code);
5148 		goto complete;
5149 	}
5150 
5151 	/* get the position of the firmware version info
5152 	 * located after the ASCII 'RM' substring in the buffer
5153 	 */
5154 	substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
5155 	if (!substr) {
5156 		dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
5157 		goto complete;
5158 	}
5159 
5160 	/* get length of firmware level ASCII substring */
5161 	if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
5162 		fw_level_len = *(substr + 2);
5163 	} else {
5164 		dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
5165 		goto complete;
5166 	}
5167 
5168 	/* copy firmware version string from vpd into adapter */
5169 	if ((substr + 3 + fw_level_len) <
5170 	    (adapter->vpd->buff + adapter->vpd->len)) {
5171 		strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
5172 	} else {
5173 		dev_info(dev, "FW substr extrapolated VPD buff\n");
5174 	}
5175 
5176 complete:
5177 	if (adapter->fw_version[0] == '\0')
5178 		strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
5179 	complete(&adapter->fw_done);
5180 }
5181 
5182 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
5183 {
5184 	struct device *dev = &adapter->vdev->dev;
5185 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5186 	int i;
5187 
5188 	dma_unmap_single(dev, adapter->ip_offload_tok,
5189 			 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
5190 
5191 	netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
5192 	for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
5193 		netdev_dbg(adapter->netdev, "%016lx\n",
5194 			   ((unsigned long *)(buf))[i]);
5195 
5196 	netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
5197 	netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
5198 	netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
5199 		   buf->tcp_ipv4_chksum);
5200 	netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
5201 		   buf->tcp_ipv6_chksum);
5202 	netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
5203 		   buf->udp_ipv4_chksum);
5204 	netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
5205 		   buf->udp_ipv6_chksum);
5206 	netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
5207 		   buf->large_tx_ipv4);
5208 	netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
5209 		   buf->large_tx_ipv6);
5210 	netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
5211 		   buf->large_rx_ipv4);
5212 	netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
5213 		   buf->large_rx_ipv6);
5214 	netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
5215 		   buf->max_ipv4_header_size);
5216 	netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
5217 		   buf->max_ipv6_header_size);
5218 	netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
5219 		   buf->max_tcp_header_size);
5220 	netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
5221 		   buf->max_udp_header_size);
5222 	netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
5223 		   buf->max_large_tx_size);
5224 	netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
5225 		   buf->max_large_rx_size);
5226 	netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
5227 		   buf->ipv6_extension_header);
5228 	netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
5229 		   buf->tcp_pseudosum_req);
5230 	netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
5231 		   buf->num_ipv6_ext_headers);
5232 	netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
5233 		   buf->off_ipv6_ext_headers);
5234 
5235 	send_control_ip_offload(adapter);
5236 }
5237 
5238 static const char *ibmvnic_fw_err_cause(u16 cause)
5239 {
5240 	switch (cause) {
5241 	case ADAPTER_PROBLEM:
5242 		return "adapter problem";
5243 	case BUS_PROBLEM:
5244 		return "bus problem";
5245 	case FW_PROBLEM:
5246 		return "firmware problem";
5247 	case DD_PROBLEM:
5248 		return "device driver problem";
5249 	case EEH_RECOVERY:
5250 		return "EEH recovery";
5251 	case FW_UPDATED:
5252 		return "firmware updated";
5253 	case LOW_MEMORY:
5254 		return "low Memory";
5255 	default:
5256 		return "unknown";
5257 	}
5258 }
5259 
5260 static void handle_error_indication(union ibmvnic_crq *crq,
5261 				    struct ibmvnic_adapter *adapter)
5262 {
5263 	struct device *dev = &adapter->vdev->dev;
5264 	u16 cause;
5265 
5266 	cause = be16_to_cpu(crq->error_indication.error_cause);
5267 
5268 	dev_warn_ratelimited(dev,
5269 			     "Firmware reports %serror, cause: %s. Starting recovery...\n",
5270 			     crq->error_indication.flags
5271 				& IBMVNIC_FATAL_ERROR ? "FATAL " : "",
5272 			     ibmvnic_fw_err_cause(cause));
5273 
5274 	if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
5275 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5276 	else
5277 		ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
5278 }
5279 
5280 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
5281 				 struct ibmvnic_adapter *adapter)
5282 {
5283 	struct net_device *netdev = adapter->netdev;
5284 	struct device *dev = &adapter->vdev->dev;
5285 	long rc;
5286 
5287 	rc = crq->change_mac_addr_rsp.rc.code;
5288 	if (rc) {
5289 		dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
5290 		goto out;
5291 	}
5292 	/* crq->change_mac_addr.mac_addr is the requested one
5293 	 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
5294 	 */
5295 	eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
5296 	ether_addr_copy(adapter->mac_addr,
5297 			&crq->change_mac_addr_rsp.mac_addr[0]);
5298 out:
5299 	complete(&adapter->fw_done);
5300 	return rc;
5301 }
5302 
5303 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
5304 				   struct ibmvnic_adapter *adapter)
5305 {
5306 	struct device *dev = &adapter->vdev->dev;
5307 	u64 *req_value;
5308 	char *name;
5309 
5310 	atomic_dec(&adapter->running_cap_crqs);
5311 	netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
5312 		   atomic_read(&adapter->running_cap_crqs));
5313 	switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
5314 	case REQ_TX_QUEUES:
5315 		req_value = &adapter->req_tx_queues;
5316 		name = "tx";
5317 		break;
5318 	case REQ_RX_QUEUES:
5319 		req_value = &adapter->req_rx_queues;
5320 		name = "rx";
5321 		break;
5322 	case REQ_RX_ADD_QUEUES:
5323 		req_value = &adapter->req_rx_add_queues;
5324 		name = "rx_add";
5325 		break;
5326 	case REQ_TX_ENTRIES_PER_SUBCRQ:
5327 		req_value = &adapter->req_tx_entries_per_subcrq;
5328 		name = "tx_entries_per_subcrq";
5329 		break;
5330 	case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
5331 		req_value = &adapter->req_rx_add_entries_per_subcrq;
5332 		name = "rx_add_entries_per_subcrq";
5333 		break;
5334 	case REQ_MTU:
5335 		req_value = &adapter->req_mtu;
5336 		name = "mtu";
5337 		break;
5338 	case PROMISC_REQUESTED:
5339 		req_value = &adapter->promisc;
5340 		name = "promisc";
5341 		break;
5342 	default:
5343 		dev_err(dev, "Got invalid cap request rsp %d\n",
5344 			crq->request_capability.capability);
5345 		return;
5346 	}
5347 
5348 	switch (crq->request_capability_rsp.rc.code) {
5349 	case SUCCESS:
5350 		break;
5351 	case PARTIALSUCCESS:
5352 		dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
5353 			 *req_value,
5354 			 (long)be64_to_cpu(crq->request_capability_rsp.number),
5355 			 name);
5356 
5357 		if (be16_to_cpu(crq->request_capability_rsp.capability) ==
5358 		    REQ_MTU) {
5359 			pr_err("mtu of %llu is not supported. Reverting.\n",
5360 			       *req_value);
5361 			*req_value = adapter->fallback.mtu;
5362 		} else {
5363 			*req_value =
5364 				be64_to_cpu(crq->request_capability_rsp.number);
5365 		}
5366 
5367 		send_request_cap(adapter, 1);
5368 		return;
5369 	default:
5370 		dev_err(dev, "Error %d in request cap rsp\n",
5371 			crq->request_capability_rsp.rc.code);
5372 		return;
5373 	}
5374 
5375 	/* Done receiving requested capabilities, query IP offload support */
5376 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5377 		send_query_ip_offload(adapter);
5378 }
5379 
5380 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
5381 			    struct ibmvnic_adapter *adapter)
5382 {
5383 	struct device *dev = &adapter->vdev->dev;
5384 	struct net_device *netdev = adapter->netdev;
5385 	struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
5386 	struct ibmvnic_login_buffer *login = adapter->login_buf;
5387 	u64 *tx_handle_array;
5388 	u64 *rx_handle_array;
5389 	int num_tx_pools;
5390 	int num_rx_pools;
5391 	u64 *size_array;
5392 	int i;
5393 
5394 	/* CHECK: Test/set of login_pending does not need to be atomic
5395 	 * because only ibmvnic_tasklet tests/clears this.
5396 	 */
5397 	if (!adapter->login_pending) {
5398 		netdev_warn(netdev, "Ignoring unexpected login response\n");
5399 		return 0;
5400 	}
5401 	adapter->login_pending = false;
5402 
5403 	dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
5404 			 DMA_TO_DEVICE);
5405 	dma_unmap_single(dev, adapter->login_rsp_buf_token,
5406 			 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
5407 
5408 	/* If the number of queues requested can't be allocated by the
5409 	 * server, the login response will return with code 1. We will need
5410 	 * to resend the login buffer with fewer queues requested.
5411 	 */
5412 	if (login_rsp_crq->generic.rc.code) {
5413 		adapter->init_done_rc = login_rsp_crq->generic.rc.code;
5414 		complete(&adapter->init_done);
5415 		return 0;
5416 	}
5417 
5418 	if (adapter->failover_pending) {
5419 		adapter->init_done_rc = -EAGAIN;
5420 		netdev_dbg(netdev, "Failover pending, ignoring login response\n");
5421 		complete(&adapter->init_done);
5422 		/* login response buffer will be released on reset */
5423 		return 0;
5424 	}
5425 
5426 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
5427 
5428 	netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
5429 	for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
5430 		netdev_dbg(adapter->netdev, "%016lx\n",
5431 			   ((unsigned long *)(adapter->login_rsp_buf))[i]);
5432 	}
5433 
5434 	/* Sanity checks */
5435 	if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
5436 	    (be32_to_cpu(login->num_rxcomp_subcrqs) *
5437 	     adapter->req_rx_add_queues !=
5438 	     be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
5439 		dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
5440 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5441 		return -EIO;
5442 	}
5443 	size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5444 		be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
5445 	/* variable buffer sizes are not supported, so just read the
5446 	 * first entry.
5447 	 */
5448 	adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
5449 
5450 	num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
5451 	num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
5452 
5453 	tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5454 				  be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
5455 	rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5456 				  be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
5457 
5458 	for (i = 0; i < num_tx_pools; i++)
5459 		adapter->tx_scrq[i]->handle = tx_handle_array[i];
5460 
5461 	for (i = 0; i < num_rx_pools; i++)
5462 		adapter->rx_scrq[i]->handle = rx_handle_array[i];
5463 
5464 	adapter->num_active_tx_scrqs = num_tx_pools;
5465 	adapter->num_active_rx_scrqs = num_rx_pools;
5466 	release_login_rsp_buffer(adapter);
5467 	release_login_buffer(adapter);
5468 	complete(&adapter->init_done);
5469 
5470 	return 0;
5471 }
5472 
5473 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
5474 				     struct ibmvnic_adapter *adapter)
5475 {
5476 	struct device *dev = &adapter->vdev->dev;
5477 	long rc;
5478 
5479 	rc = crq->request_unmap_rsp.rc.code;
5480 	if (rc)
5481 		dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
5482 }
5483 
5484 static void handle_query_map_rsp(union ibmvnic_crq *crq,
5485 				 struct ibmvnic_adapter *adapter)
5486 {
5487 	struct net_device *netdev = adapter->netdev;
5488 	struct device *dev = &adapter->vdev->dev;
5489 	long rc;
5490 
5491 	rc = crq->query_map_rsp.rc.code;
5492 	if (rc) {
5493 		dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5494 		return;
5495 	}
5496 	netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5497 		   crq->query_map_rsp.page_size,
5498 		   __be32_to_cpu(crq->query_map_rsp.tot_pages),
5499 		   __be32_to_cpu(crq->query_map_rsp.free_pages));
5500 }
5501 
5502 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5503 				 struct ibmvnic_adapter *adapter)
5504 {
5505 	struct net_device *netdev = adapter->netdev;
5506 	struct device *dev = &adapter->vdev->dev;
5507 	long rc;
5508 
5509 	atomic_dec(&adapter->running_cap_crqs);
5510 	netdev_dbg(netdev, "Outstanding queries: %d\n",
5511 		   atomic_read(&adapter->running_cap_crqs));
5512 	rc = crq->query_capability.rc.code;
5513 	if (rc) {
5514 		dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5515 		goto out;
5516 	}
5517 
5518 	switch (be16_to_cpu(crq->query_capability.capability)) {
5519 	case MIN_TX_QUEUES:
5520 		adapter->min_tx_queues =
5521 		    be64_to_cpu(crq->query_capability.number);
5522 		netdev_dbg(netdev, "min_tx_queues = %lld\n",
5523 			   adapter->min_tx_queues);
5524 		break;
5525 	case MIN_RX_QUEUES:
5526 		adapter->min_rx_queues =
5527 		    be64_to_cpu(crq->query_capability.number);
5528 		netdev_dbg(netdev, "min_rx_queues = %lld\n",
5529 			   adapter->min_rx_queues);
5530 		break;
5531 	case MIN_RX_ADD_QUEUES:
5532 		adapter->min_rx_add_queues =
5533 		    be64_to_cpu(crq->query_capability.number);
5534 		netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5535 			   adapter->min_rx_add_queues);
5536 		break;
5537 	case MAX_TX_QUEUES:
5538 		adapter->max_tx_queues =
5539 		    be64_to_cpu(crq->query_capability.number);
5540 		netdev_dbg(netdev, "max_tx_queues = %lld\n",
5541 			   adapter->max_tx_queues);
5542 		break;
5543 	case MAX_RX_QUEUES:
5544 		adapter->max_rx_queues =
5545 		    be64_to_cpu(crq->query_capability.number);
5546 		netdev_dbg(netdev, "max_rx_queues = %lld\n",
5547 			   adapter->max_rx_queues);
5548 		break;
5549 	case MAX_RX_ADD_QUEUES:
5550 		adapter->max_rx_add_queues =
5551 		    be64_to_cpu(crq->query_capability.number);
5552 		netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5553 			   adapter->max_rx_add_queues);
5554 		break;
5555 	case MIN_TX_ENTRIES_PER_SUBCRQ:
5556 		adapter->min_tx_entries_per_subcrq =
5557 		    be64_to_cpu(crq->query_capability.number);
5558 		netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5559 			   adapter->min_tx_entries_per_subcrq);
5560 		break;
5561 	case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5562 		adapter->min_rx_add_entries_per_subcrq =
5563 		    be64_to_cpu(crq->query_capability.number);
5564 		netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5565 			   adapter->min_rx_add_entries_per_subcrq);
5566 		break;
5567 	case MAX_TX_ENTRIES_PER_SUBCRQ:
5568 		adapter->max_tx_entries_per_subcrq =
5569 		    be64_to_cpu(crq->query_capability.number);
5570 		netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5571 			   adapter->max_tx_entries_per_subcrq);
5572 		break;
5573 	case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5574 		adapter->max_rx_add_entries_per_subcrq =
5575 		    be64_to_cpu(crq->query_capability.number);
5576 		netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5577 			   adapter->max_rx_add_entries_per_subcrq);
5578 		break;
5579 	case TCP_IP_OFFLOAD:
5580 		adapter->tcp_ip_offload =
5581 		    be64_to_cpu(crq->query_capability.number);
5582 		netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5583 			   adapter->tcp_ip_offload);
5584 		break;
5585 	case PROMISC_SUPPORTED:
5586 		adapter->promisc_supported =
5587 		    be64_to_cpu(crq->query_capability.number);
5588 		netdev_dbg(netdev, "promisc_supported = %lld\n",
5589 			   adapter->promisc_supported);
5590 		break;
5591 	case MIN_MTU:
5592 		adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5593 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5594 		netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5595 		break;
5596 	case MAX_MTU:
5597 		adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5598 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5599 		netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5600 		break;
5601 	case MAX_MULTICAST_FILTERS:
5602 		adapter->max_multicast_filters =
5603 		    be64_to_cpu(crq->query_capability.number);
5604 		netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5605 			   adapter->max_multicast_filters);
5606 		break;
5607 	case VLAN_HEADER_INSERTION:
5608 		adapter->vlan_header_insertion =
5609 		    be64_to_cpu(crq->query_capability.number);
5610 		if (adapter->vlan_header_insertion)
5611 			netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5612 		netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5613 			   adapter->vlan_header_insertion);
5614 		break;
5615 	case RX_VLAN_HEADER_INSERTION:
5616 		adapter->rx_vlan_header_insertion =
5617 		    be64_to_cpu(crq->query_capability.number);
5618 		netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5619 			   adapter->rx_vlan_header_insertion);
5620 		break;
5621 	case MAX_TX_SG_ENTRIES:
5622 		adapter->max_tx_sg_entries =
5623 		    be64_to_cpu(crq->query_capability.number);
5624 		netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5625 			   adapter->max_tx_sg_entries);
5626 		break;
5627 	case RX_SG_SUPPORTED:
5628 		adapter->rx_sg_supported =
5629 		    be64_to_cpu(crq->query_capability.number);
5630 		netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5631 			   adapter->rx_sg_supported);
5632 		break;
5633 	case OPT_TX_COMP_SUB_QUEUES:
5634 		adapter->opt_tx_comp_sub_queues =
5635 		    be64_to_cpu(crq->query_capability.number);
5636 		netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5637 			   adapter->opt_tx_comp_sub_queues);
5638 		break;
5639 	case OPT_RX_COMP_QUEUES:
5640 		adapter->opt_rx_comp_queues =
5641 		    be64_to_cpu(crq->query_capability.number);
5642 		netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5643 			   adapter->opt_rx_comp_queues);
5644 		break;
5645 	case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5646 		adapter->opt_rx_bufadd_q_per_rx_comp_q =
5647 		    be64_to_cpu(crq->query_capability.number);
5648 		netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5649 			   adapter->opt_rx_bufadd_q_per_rx_comp_q);
5650 		break;
5651 	case OPT_TX_ENTRIES_PER_SUBCRQ:
5652 		adapter->opt_tx_entries_per_subcrq =
5653 		    be64_to_cpu(crq->query_capability.number);
5654 		netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5655 			   adapter->opt_tx_entries_per_subcrq);
5656 		break;
5657 	case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5658 		adapter->opt_rxba_entries_per_subcrq =
5659 		    be64_to_cpu(crq->query_capability.number);
5660 		netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5661 			   adapter->opt_rxba_entries_per_subcrq);
5662 		break;
5663 	case TX_RX_DESC_REQ:
5664 		adapter->tx_rx_desc_req = crq->query_capability.number;
5665 		netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5666 			   adapter->tx_rx_desc_req);
5667 		break;
5668 
5669 	default:
5670 		netdev_err(netdev, "Got invalid cap rsp %d\n",
5671 			   crq->query_capability.capability);
5672 	}
5673 
5674 out:
5675 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5676 		send_request_cap(adapter, 0);
5677 }
5678 
5679 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5680 {
5681 	union ibmvnic_crq crq;
5682 	int rc;
5683 
5684 	memset(&crq, 0, sizeof(crq));
5685 	crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5686 	crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5687 
5688 	mutex_lock(&adapter->fw_lock);
5689 	adapter->fw_done_rc = 0;
5690 	reinit_completion(&adapter->fw_done);
5691 
5692 	rc = ibmvnic_send_crq(adapter, &crq);
5693 	if (rc) {
5694 		mutex_unlock(&adapter->fw_lock);
5695 		return rc;
5696 	}
5697 
5698 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5699 	if (rc) {
5700 		mutex_unlock(&adapter->fw_lock);
5701 		return rc;
5702 	}
5703 
5704 	mutex_unlock(&adapter->fw_lock);
5705 	return adapter->fw_done_rc ? -EIO : 0;
5706 }
5707 
5708 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5709 				       struct ibmvnic_adapter *adapter)
5710 {
5711 	struct net_device *netdev = adapter->netdev;
5712 	int rc;
5713 	__be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5714 
5715 	rc = crq->query_phys_parms_rsp.rc.code;
5716 	if (rc) {
5717 		netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5718 		return rc;
5719 	}
5720 	switch (rspeed) {
5721 	case IBMVNIC_10MBPS:
5722 		adapter->speed = SPEED_10;
5723 		break;
5724 	case IBMVNIC_100MBPS:
5725 		adapter->speed = SPEED_100;
5726 		break;
5727 	case IBMVNIC_1GBPS:
5728 		adapter->speed = SPEED_1000;
5729 		break;
5730 	case IBMVNIC_10GBPS:
5731 		adapter->speed = SPEED_10000;
5732 		break;
5733 	case IBMVNIC_25GBPS:
5734 		adapter->speed = SPEED_25000;
5735 		break;
5736 	case IBMVNIC_40GBPS:
5737 		adapter->speed = SPEED_40000;
5738 		break;
5739 	case IBMVNIC_50GBPS:
5740 		adapter->speed = SPEED_50000;
5741 		break;
5742 	case IBMVNIC_100GBPS:
5743 		adapter->speed = SPEED_100000;
5744 		break;
5745 	case IBMVNIC_200GBPS:
5746 		adapter->speed = SPEED_200000;
5747 		break;
5748 	default:
5749 		if (netif_carrier_ok(netdev))
5750 			netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5751 		adapter->speed = SPEED_UNKNOWN;
5752 	}
5753 	if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5754 		adapter->duplex = DUPLEX_FULL;
5755 	else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5756 		adapter->duplex = DUPLEX_HALF;
5757 	else
5758 		adapter->duplex = DUPLEX_UNKNOWN;
5759 
5760 	return rc;
5761 }
5762 
5763 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5764 			       struct ibmvnic_adapter *adapter)
5765 {
5766 	struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5767 	struct net_device *netdev = adapter->netdev;
5768 	struct device *dev = &adapter->vdev->dev;
5769 	u64 *u64_crq = (u64 *)crq;
5770 	long rc;
5771 
5772 	netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5773 		   (unsigned long)cpu_to_be64(u64_crq[0]),
5774 		   (unsigned long)cpu_to_be64(u64_crq[1]));
5775 	switch (gen_crq->first) {
5776 	case IBMVNIC_CRQ_INIT_RSP:
5777 		switch (gen_crq->cmd) {
5778 		case IBMVNIC_CRQ_INIT:
5779 			dev_info(dev, "Partner initialized\n");
5780 			adapter->from_passive_init = true;
5781 			/* Discard any stale login responses from prev reset.
5782 			 * CHECK: should we clear even on INIT_COMPLETE?
5783 			 */
5784 			adapter->login_pending = false;
5785 
5786 			if (adapter->state == VNIC_DOWN)
5787 				rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5788 			else
5789 				rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5790 
5791 			if (rc && rc != -EBUSY) {
5792 				/* We were unable to schedule the failover
5793 				 * reset either because the adapter was still
5794 				 * probing (eg: during kexec) or we could not
5795 				 * allocate memory. Clear the failover_pending
5796 				 * flag since no one else will. We ignore
5797 				 * EBUSY because it means either FAILOVER reset
5798 				 * is already scheduled or the adapter is
5799 				 * being removed.
5800 				 */
5801 				netdev_err(netdev,
5802 					   "Error %ld scheduling failover reset\n",
5803 					   rc);
5804 				adapter->failover_pending = false;
5805 			}
5806 
5807 			if (!completion_done(&adapter->init_done)) {
5808 				if (!adapter->init_done_rc)
5809 					adapter->init_done_rc = -EAGAIN;
5810 				complete(&adapter->init_done);
5811 			}
5812 
5813 			break;
5814 		case IBMVNIC_CRQ_INIT_COMPLETE:
5815 			dev_info(dev, "Partner initialization complete\n");
5816 			adapter->crq.active = true;
5817 			send_version_xchg(adapter);
5818 			break;
5819 		default:
5820 			dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5821 		}
5822 		return;
5823 	case IBMVNIC_CRQ_XPORT_EVENT:
5824 		netif_carrier_off(netdev);
5825 		adapter->crq.active = false;
5826 		/* terminate any thread waiting for a response
5827 		 * from the device
5828 		 */
5829 		if (!completion_done(&adapter->fw_done)) {
5830 			adapter->fw_done_rc = -EIO;
5831 			complete(&adapter->fw_done);
5832 		}
5833 
5834 		/* if we got here during crq-init, retry crq-init */
5835 		if (!completion_done(&adapter->init_done)) {
5836 			adapter->init_done_rc = -EAGAIN;
5837 			complete(&adapter->init_done);
5838 		}
5839 
5840 		if (!completion_done(&adapter->stats_done))
5841 			complete(&adapter->stats_done);
5842 		if (test_bit(0, &adapter->resetting))
5843 			adapter->force_reset_recovery = true;
5844 		if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5845 			dev_info(dev, "Migrated, re-enabling adapter\n");
5846 			ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5847 		} else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5848 			dev_info(dev, "Backing device failover detected\n");
5849 			adapter->failover_pending = true;
5850 		} else {
5851 			/* The adapter lost the connection */
5852 			dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
5853 				gen_crq->cmd);
5854 			ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5855 		}
5856 		return;
5857 	case IBMVNIC_CRQ_CMD_RSP:
5858 		break;
5859 	default:
5860 		dev_err(dev, "Got an invalid msg type 0x%02x\n",
5861 			gen_crq->first);
5862 		return;
5863 	}
5864 
5865 	switch (gen_crq->cmd) {
5866 	case VERSION_EXCHANGE_RSP:
5867 		rc = crq->version_exchange_rsp.rc.code;
5868 		if (rc) {
5869 			dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
5870 			break;
5871 		}
5872 		ibmvnic_version =
5873 			    be16_to_cpu(crq->version_exchange_rsp.version);
5874 		dev_info(dev, "Partner protocol version is %d\n",
5875 			 ibmvnic_version);
5876 		send_query_cap(adapter);
5877 		break;
5878 	case QUERY_CAPABILITY_RSP:
5879 		handle_query_cap_rsp(crq, adapter);
5880 		break;
5881 	case QUERY_MAP_RSP:
5882 		handle_query_map_rsp(crq, adapter);
5883 		break;
5884 	case REQUEST_MAP_RSP:
5885 		adapter->fw_done_rc = crq->request_map_rsp.rc.code;
5886 		complete(&adapter->fw_done);
5887 		break;
5888 	case REQUEST_UNMAP_RSP:
5889 		handle_request_unmap_rsp(crq, adapter);
5890 		break;
5891 	case REQUEST_CAPABILITY_RSP:
5892 		handle_request_cap_rsp(crq, adapter);
5893 		break;
5894 	case LOGIN_RSP:
5895 		netdev_dbg(netdev, "Got Login Response\n");
5896 		handle_login_rsp(crq, adapter);
5897 		break;
5898 	case LOGICAL_LINK_STATE_RSP:
5899 		netdev_dbg(netdev,
5900 			   "Got Logical Link State Response, state: %d rc: %d\n",
5901 			   crq->logical_link_state_rsp.link_state,
5902 			   crq->logical_link_state_rsp.rc.code);
5903 		adapter->logical_link_state =
5904 		    crq->logical_link_state_rsp.link_state;
5905 		adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5906 		complete(&adapter->init_done);
5907 		break;
5908 	case LINK_STATE_INDICATION:
5909 		netdev_dbg(netdev, "Got Logical Link State Indication\n");
5910 		adapter->phys_link_state =
5911 		    crq->link_state_indication.phys_link_state;
5912 		adapter->logical_link_state =
5913 		    crq->link_state_indication.logical_link_state;
5914 		if (adapter->phys_link_state && adapter->logical_link_state)
5915 			netif_carrier_on(netdev);
5916 		else
5917 			netif_carrier_off(netdev);
5918 		break;
5919 	case CHANGE_MAC_ADDR_RSP:
5920 		netdev_dbg(netdev, "Got MAC address change Response\n");
5921 		adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
5922 		break;
5923 	case ERROR_INDICATION:
5924 		netdev_dbg(netdev, "Got Error Indication\n");
5925 		handle_error_indication(crq, adapter);
5926 		break;
5927 	case REQUEST_STATISTICS_RSP:
5928 		netdev_dbg(netdev, "Got Statistics Response\n");
5929 		complete(&adapter->stats_done);
5930 		break;
5931 	case QUERY_IP_OFFLOAD_RSP:
5932 		netdev_dbg(netdev, "Got Query IP offload Response\n");
5933 		handle_query_ip_offload_rsp(adapter);
5934 		break;
5935 	case MULTICAST_CTRL_RSP:
5936 		netdev_dbg(netdev, "Got multicast control Response\n");
5937 		break;
5938 	case CONTROL_IP_OFFLOAD_RSP:
5939 		netdev_dbg(netdev, "Got Control IP offload Response\n");
5940 		dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5941 				 sizeof(adapter->ip_offload_ctrl),
5942 				 DMA_TO_DEVICE);
5943 		complete(&adapter->init_done);
5944 		break;
5945 	case COLLECT_FW_TRACE_RSP:
5946 		netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5947 		complete(&adapter->fw_done);
5948 		break;
5949 	case GET_VPD_SIZE_RSP:
5950 		handle_vpd_size_rsp(crq, adapter);
5951 		break;
5952 	case GET_VPD_RSP:
5953 		handle_vpd_rsp(crq, adapter);
5954 		break;
5955 	case QUERY_PHYS_PARMS_RSP:
5956 		adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5957 		complete(&adapter->fw_done);
5958 		break;
5959 	default:
5960 		netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5961 			   gen_crq->cmd);
5962 	}
5963 }
5964 
5965 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5966 {
5967 	struct ibmvnic_adapter *adapter = instance;
5968 
5969 	tasklet_schedule(&adapter->tasklet);
5970 	return IRQ_HANDLED;
5971 }
5972 
5973 static void ibmvnic_tasklet(struct tasklet_struct *t)
5974 {
5975 	struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5976 	struct ibmvnic_crq_queue *queue = &adapter->crq;
5977 	union ibmvnic_crq *crq;
5978 	unsigned long flags;
5979 
5980 	spin_lock_irqsave(&queue->lock, flags);
5981 
5982 	/* Pull all the valid messages off the CRQ */
5983 	while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5984 		/* This barrier makes sure ibmvnic_next_crq()'s
5985 		 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5986 		 * before ibmvnic_handle_crq()'s
5987 		 * switch(gen_crq->first) and switch(gen_crq->cmd).
5988 		 */
5989 		dma_rmb();
5990 		ibmvnic_handle_crq(crq, adapter);
5991 		crq->generic.first = 0;
5992 	}
5993 
5994 	spin_unlock_irqrestore(&queue->lock, flags);
5995 }
5996 
5997 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5998 {
5999 	struct vio_dev *vdev = adapter->vdev;
6000 	int rc;
6001 
6002 	do {
6003 		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
6004 	} while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
6005 
6006 	if (rc)
6007 		dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
6008 
6009 	return rc;
6010 }
6011 
6012 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
6013 {
6014 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6015 	struct device *dev = &adapter->vdev->dev;
6016 	struct vio_dev *vdev = adapter->vdev;
6017 	int rc;
6018 
6019 	/* Close the CRQ */
6020 	do {
6021 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6022 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6023 
6024 	/* Clean out the queue */
6025 	if (!crq->msgs)
6026 		return -EINVAL;
6027 
6028 	memset(crq->msgs, 0, PAGE_SIZE);
6029 	crq->cur = 0;
6030 	crq->active = false;
6031 
6032 	/* And re-open it again */
6033 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6034 				crq->msg_token, PAGE_SIZE);
6035 
6036 	if (rc == H_CLOSED)
6037 		/* Adapter is good, but other end is not ready */
6038 		dev_warn(dev, "Partner adapter not ready\n");
6039 	else if (rc != 0)
6040 		dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
6041 
6042 	return rc;
6043 }
6044 
6045 static void release_crq_queue(struct ibmvnic_adapter *adapter)
6046 {
6047 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6048 	struct vio_dev *vdev = adapter->vdev;
6049 	long rc;
6050 
6051 	if (!crq->msgs)
6052 		return;
6053 
6054 	netdev_dbg(adapter->netdev, "Releasing CRQ\n");
6055 	free_irq(vdev->irq, adapter);
6056 	tasklet_kill(&adapter->tasklet);
6057 	do {
6058 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6059 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6060 
6061 	dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
6062 			 DMA_BIDIRECTIONAL);
6063 	free_page((unsigned long)crq->msgs);
6064 	crq->msgs = NULL;
6065 	crq->active = false;
6066 }
6067 
6068 static int init_crq_queue(struct ibmvnic_adapter *adapter)
6069 {
6070 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6071 	struct device *dev = &adapter->vdev->dev;
6072 	struct vio_dev *vdev = adapter->vdev;
6073 	int rc, retrc = -ENOMEM;
6074 
6075 	if (crq->msgs)
6076 		return 0;
6077 
6078 	crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
6079 	/* Should we allocate more than one page? */
6080 
6081 	if (!crq->msgs)
6082 		return -ENOMEM;
6083 
6084 	crq->size = PAGE_SIZE / sizeof(*crq->msgs);
6085 	crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
6086 					DMA_BIDIRECTIONAL);
6087 	if (dma_mapping_error(dev, crq->msg_token))
6088 		goto map_failed;
6089 
6090 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6091 				crq->msg_token, PAGE_SIZE);
6092 
6093 	if (rc == H_RESOURCE)
6094 		/* maybe kexecing and resource is busy. try a reset */
6095 		rc = ibmvnic_reset_crq(adapter);
6096 	retrc = rc;
6097 
6098 	if (rc == H_CLOSED) {
6099 		dev_warn(dev, "Partner adapter not ready\n");
6100 	} else if (rc) {
6101 		dev_warn(dev, "Error %d opening adapter\n", rc);
6102 		goto reg_crq_failed;
6103 	}
6104 
6105 	retrc = 0;
6106 
6107 	tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
6108 
6109 	netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
6110 	snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
6111 		 adapter->vdev->unit_address);
6112 	rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
6113 	if (rc) {
6114 		dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
6115 			vdev->irq, rc);
6116 		goto req_irq_failed;
6117 	}
6118 
6119 	rc = vio_enable_interrupts(vdev);
6120 	if (rc) {
6121 		dev_err(dev, "Error %d enabling interrupts\n", rc);
6122 		goto req_irq_failed;
6123 	}
6124 
6125 	crq->cur = 0;
6126 	spin_lock_init(&crq->lock);
6127 
6128 	/* process any CRQs that were queued before we enabled interrupts */
6129 	tasklet_schedule(&adapter->tasklet);
6130 
6131 	return retrc;
6132 
6133 req_irq_failed:
6134 	tasklet_kill(&adapter->tasklet);
6135 	do {
6136 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6137 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6138 reg_crq_failed:
6139 	dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
6140 map_failed:
6141 	free_page((unsigned long)crq->msgs);
6142 	crq->msgs = NULL;
6143 	return retrc;
6144 }
6145 
6146 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
6147 {
6148 	struct device *dev = &adapter->vdev->dev;
6149 	unsigned long timeout = msecs_to_jiffies(20000);
6150 	u64 old_num_rx_queues = adapter->req_rx_queues;
6151 	u64 old_num_tx_queues = adapter->req_tx_queues;
6152 	int rc;
6153 
6154 	adapter->from_passive_init = false;
6155 
6156 	rc = ibmvnic_send_crq_init(adapter);
6157 	if (rc) {
6158 		dev_err(dev, "Send crq init failed with error %d\n", rc);
6159 		return rc;
6160 	}
6161 
6162 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
6163 		dev_err(dev, "Initialization sequence timed out\n");
6164 		return -ETIMEDOUT;
6165 	}
6166 
6167 	if (adapter->init_done_rc) {
6168 		release_crq_queue(adapter);
6169 		dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
6170 		return adapter->init_done_rc;
6171 	}
6172 
6173 	if (adapter->from_passive_init) {
6174 		adapter->state = VNIC_OPEN;
6175 		adapter->from_passive_init = false;
6176 		dev_err(dev, "CRQ-init failed, passive-init\n");
6177 		return -EINVAL;
6178 	}
6179 
6180 	if (reset &&
6181 	    test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
6182 	    adapter->reset_reason != VNIC_RESET_MOBILITY) {
6183 		if (adapter->req_rx_queues != old_num_rx_queues ||
6184 		    adapter->req_tx_queues != old_num_tx_queues) {
6185 			release_sub_crqs(adapter, 0);
6186 			rc = init_sub_crqs(adapter);
6187 		} else {
6188 			/* no need to reinitialize completely, but we do
6189 			 * need to clean up transmits that were in flight
6190 			 * when we processed the reset.  Failure to do so
6191 			 * will confound the upper layer, usually TCP, by
6192 			 * creating the illusion of transmits that are
6193 			 * awaiting completion.
6194 			 */
6195 			clean_tx_pools(adapter);
6196 
6197 			rc = reset_sub_crq_queues(adapter);
6198 		}
6199 	} else {
6200 		rc = init_sub_crqs(adapter);
6201 	}
6202 
6203 	if (rc) {
6204 		dev_err(dev, "Initialization of sub crqs failed\n");
6205 		release_crq_queue(adapter);
6206 		return rc;
6207 	}
6208 
6209 	rc = init_sub_crq_irqs(adapter);
6210 	if (rc) {
6211 		dev_err(dev, "Failed to initialize sub crq irqs\n");
6212 		release_crq_queue(adapter);
6213 	}
6214 
6215 	return rc;
6216 }
6217 
6218 static struct device_attribute dev_attr_failover;
6219 
6220 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
6221 {
6222 	struct ibmvnic_adapter *adapter;
6223 	struct net_device *netdev;
6224 	unsigned char *mac_addr_p;
6225 	unsigned long flags;
6226 	bool init_success;
6227 	int rc;
6228 
6229 	dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
6230 		dev->unit_address);
6231 
6232 	mac_addr_p = (unsigned char *)vio_get_attribute(dev,
6233 							VETH_MAC_ADDR, NULL);
6234 	if (!mac_addr_p) {
6235 		dev_err(&dev->dev,
6236 			"(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
6237 			__FILE__, __LINE__);
6238 		return 0;
6239 	}
6240 
6241 	netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
6242 				   IBMVNIC_MAX_QUEUES);
6243 	if (!netdev)
6244 		return -ENOMEM;
6245 
6246 	adapter = netdev_priv(netdev);
6247 	adapter->state = VNIC_PROBING;
6248 	dev_set_drvdata(&dev->dev, netdev);
6249 	adapter->vdev = dev;
6250 	adapter->netdev = netdev;
6251 	adapter->login_pending = false;
6252 	memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
6253 	/* map_ids start at 1, so ensure map_id 0 is always "in-use" */
6254 	bitmap_set(adapter->map_ids, 0, 1);
6255 
6256 	ether_addr_copy(adapter->mac_addr, mac_addr_p);
6257 	eth_hw_addr_set(netdev, adapter->mac_addr);
6258 	netdev->irq = dev->irq;
6259 	netdev->netdev_ops = &ibmvnic_netdev_ops;
6260 	netdev->ethtool_ops = &ibmvnic_ethtool_ops;
6261 	SET_NETDEV_DEV(netdev, &dev->dev);
6262 
6263 	INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
6264 	INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
6265 			  __ibmvnic_delayed_reset);
6266 	INIT_LIST_HEAD(&adapter->rwi_list);
6267 	spin_lock_init(&adapter->rwi_lock);
6268 	spin_lock_init(&adapter->state_lock);
6269 	mutex_init(&adapter->fw_lock);
6270 	init_completion(&adapter->probe_done);
6271 	init_completion(&adapter->init_done);
6272 	init_completion(&adapter->fw_done);
6273 	init_completion(&adapter->reset_done);
6274 	init_completion(&adapter->stats_done);
6275 	clear_bit(0, &adapter->resetting);
6276 	adapter->prev_rx_buf_sz = 0;
6277 	adapter->prev_mtu = 0;
6278 
6279 	init_success = false;
6280 	do {
6281 		reinit_init_done(adapter);
6282 
6283 		/* clear any failovers we got in the previous pass
6284 		 * since we are reinitializing the CRQ
6285 		 */
6286 		adapter->failover_pending = false;
6287 
6288 		/* If we had already initialized CRQ, we may have one or
6289 		 * more resets queued already. Discard those and release
6290 		 * the CRQ before initializing the CRQ again.
6291 		 */
6292 		release_crq_queue(adapter);
6293 
6294 		/* Since we are still in PROBING state, __ibmvnic_reset()
6295 		 * will not access the ->rwi_list and since we released CRQ,
6296 		 * we won't get _new_ transport events. But there maybe an
6297 		 * ongoing ibmvnic_reset() call. So serialize access to
6298 		 * rwi_list. If we win the race, ibvmnic_reset() could add
6299 		 * a reset after we purged but thats ok - we just may end
6300 		 * up with an extra reset (i.e similar to having two or more
6301 		 * resets in the queue at once).
6302 		 * CHECK.
6303 		 */
6304 		spin_lock_irqsave(&adapter->rwi_lock, flags);
6305 		flush_reset_queue(adapter);
6306 		spin_unlock_irqrestore(&adapter->rwi_lock, flags);
6307 
6308 		rc = init_crq_queue(adapter);
6309 		if (rc) {
6310 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
6311 				rc);
6312 			goto ibmvnic_init_fail;
6313 		}
6314 
6315 		rc = ibmvnic_reset_init(adapter, false);
6316 	} while (rc == -EAGAIN);
6317 
6318 	/* We are ignoring the error from ibmvnic_reset_init() assuming that the
6319 	 * partner is not ready. CRQ is not active. When the partner becomes
6320 	 * ready, we will do the passive init reset.
6321 	 */
6322 
6323 	if (!rc)
6324 		init_success = true;
6325 
6326 	rc = init_stats_buffers(adapter);
6327 	if (rc)
6328 		goto ibmvnic_init_fail;
6329 
6330 	rc = init_stats_token(adapter);
6331 	if (rc)
6332 		goto ibmvnic_stats_fail;
6333 
6334 	rc = device_create_file(&dev->dev, &dev_attr_failover);
6335 	if (rc)
6336 		goto ibmvnic_dev_file_err;
6337 
6338 	netif_carrier_off(netdev);
6339 
6340 	if (init_success) {
6341 		adapter->state = VNIC_PROBED;
6342 		netdev->mtu = adapter->req_mtu - ETH_HLEN;
6343 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
6344 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
6345 	} else {
6346 		adapter->state = VNIC_DOWN;
6347 	}
6348 
6349 	adapter->wait_for_reset = false;
6350 	adapter->last_reset_time = jiffies;
6351 
6352 	rc = register_netdev(netdev);
6353 	if (rc) {
6354 		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
6355 		goto ibmvnic_register_fail;
6356 	}
6357 	dev_info(&dev->dev, "ibmvnic registered\n");
6358 
6359 	rc = ibmvnic_cpu_notif_add(adapter);
6360 	if (rc) {
6361 		netdev_err(netdev, "Registering cpu notifier failed\n");
6362 		goto cpu_notif_add_failed;
6363 	}
6364 
6365 	complete(&adapter->probe_done);
6366 
6367 	return 0;
6368 
6369 cpu_notif_add_failed:
6370 	unregister_netdev(netdev);
6371 
6372 ibmvnic_register_fail:
6373 	device_remove_file(&dev->dev, &dev_attr_failover);
6374 
6375 ibmvnic_dev_file_err:
6376 	release_stats_token(adapter);
6377 
6378 ibmvnic_stats_fail:
6379 	release_stats_buffers(adapter);
6380 
6381 ibmvnic_init_fail:
6382 	release_sub_crqs(adapter, 1);
6383 	release_crq_queue(adapter);
6384 
6385 	/* cleanup worker thread after releasing CRQ so we don't get
6386 	 * transport events (i.e new work items for the worker thread).
6387 	 */
6388 	adapter->state = VNIC_REMOVING;
6389 	complete(&adapter->probe_done);
6390 	flush_work(&adapter->ibmvnic_reset);
6391 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6392 
6393 	flush_reset_queue(adapter);
6394 
6395 	mutex_destroy(&adapter->fw_lock);
6396 	free_netdev(netdev);
6397 
6398 	return rc;
6399 }
6400 
6401 static void ibmvnic_remove(struct vio_dev *dev)
6402 {
6403 	struct net_device *netdev = dev_get_drvdata(&dev->dev);
6404 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6405 	unsigned long flags;
6406 
6407 	spin_lock_irqsave(&adapter->state_lock, flags);
6408 
6409 	/* If ibmvnic_reset() is scheduling a reset, wait for it to
6410 	 * finish. Then, set the state to REMOVING to prevent it from
6411 	 * scheduling any more work and to have reset functions ignore
6412 	 * any resets that have already been scheduled. Drop the lock
6413 	 * after setting state, so __ibmvnic_reset() which is called
6414 	 * from the flush_work() below, can make progress.
6415 	 */
6416 	spin_lock(&adapter->rwi_lock);
6417 	adapter->state = VNIC_REMOVING;
6418 	spin_unlock(&adapter->rwi_lock);
6419 
6420 	spin_unlock_irqrestore(&adapter->state_lock, flags);
6421 
6422 	ibmvnic_cpu_notif_remove(adapter);
6423 
6424 	flush_work(&adapter->ibmvnic_reset);
6425 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6426 
6427 	rtnl_lock();
6428 	unregister_netdevice(netdev);
6429 
6430 	release_resources(adapter);
6431 	release_rx_pools(adapter);
6432 	release_tx_pools(adapter);
6433 	release_sub_crqs(adapter, 1);
6434 	release_crq_queue(adapter);
6435 
6436 	release_stats_token(adapter);
6437 	release_stats_buffers(adapter);
6438 
6439 	adapter->state = VNIC_REMOVED;
6440 
6441 	rtnl_unlock();
6442 	mutex_destroy(&adapter->fw_lock);
6443 	device_remove_file(&dev->dev, &dev_attr_failover);
6444 	free_netdev(netdev);
6445 	dev_set_drvdata(&dev->dev, NULL);
6446 }
6447 
6448 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
6449 			      const char *buf, size_t count)
6450 {
6451 	struct net_device *netdev = dev_get_drvdata(dev);
6452 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6453 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
6454 	__be64 session_token;
6455 	long rc;
6456 
6457 	if (!sysfs_streq(buf, "1"))
6458 		return -EINVAL;
6459 
6460 	rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
6461 			 H_GET_SESSION_TOKEN, 0, 0, 0);
6462 	if (rc) {
6463 		netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
6464 			   rc);
6465 		goto last_resort;
6466 	}
6467 
6468 	session_token = (__be64)retbuf[0];
6469 	netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
6470 		   be64_to_cpu(session_token));
6471 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
6472 				H_SESSION_ERR_DETECTED, session_token, 0, 0);
6473 	if (rc) {
6474 		netdev_err(netdev,
6475 			   "H_VIOCTL initiated failover failed, rc %ld\n",
6476 			   rc);
6477 		goto last_resort;
6478 	}
6479 
6480 	return count;
6481 
6482 last_resort:
6483 	netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
6484 	ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
6485 
6486 	return count;
6487 }
6488 static DEVICE_ATTR_WO(failover);
6489 
6490 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
6491 {
6492 	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
6493 	struct ibmvnic_adapter *adapter;
6494 	struct iommu_table *tbl;
6495 	unsigned long ret = 0;
6496 	int i;
6497 
6498 	tbl = get_iommu_table_base(&vdev->dev);
6499 
6500 	/* netdev inits at probe time along with the structures we need below*/
6501 	if (!netdev)
6502 		return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
6503 
6504 	adapter = netdev_priv(netdev);
6505 
6506 	ret += PAGE_SIZE; /* the crq message queue */
6507 	ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
6508 
6509 	for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
6510 		ret += 4 * PAGE_SIZE; /* the scrq message queue */
6511 
6512 	for (i = 0; i < adapter->num_active_rx_pools; i++)
6513 		ret += adapter->rx_pool[i].size *
6514 		    IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
6515 
6516 	return ret;
6517 }
6518 
6519 static int ibmvnic_resume(struct device *dev)
6520 {
6521 	struct net_device *netdev = dev_get_drvdata(dev);
6522 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6523 
6524 	if (adapter->state != VNIC_OPEN)
6525 		return 0;
6526 
6527 	tasklet_schedule(&adapter->tasklet);
6528 
6529 	return 0;
6530 }
6531 
6532 static const struct vio_device_id ibmvnic_device_table[] = {
6533 	{"network", "IBM,vnic"},
6534 	{"", "" }
6535 };
6536 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
6537 
6538 static const struct dev_pm_ops ibmvnic_pm_ops = {
6539 	.resume = ibmvnic_resume
6540 };
6541 
6542 static struct vio_driver ibmvnic_driver = {
6543 	.id_table       = ibmvnic_device_table,
6544 	.probe          = ibmvnic_probe,
6545 	.remove         = ibmvnic_remove,
6546 	.get_desired_dma = ibmvnic_get_desired_dma,
6547 	.name		= ibmvnic_driver_name,
6548 	.pm		= &ibmvnic_pm_ops,
6549 };
6550 
6551 /* module functions */
6552 static int __init ibmvnic_module_init(void)
6553 {
6554 	int ret;
6555 
6556 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "net/ibmvnic:online",
6557 				      ibmvnic_cpu_online,
6558 				      ibmvnic_cpu_down_prep);
6559 	if (ret < 0)
6560 		goto out;
6561 	ibmvnic_online = ret;
6562 	ret = cpuhp_setup_state_multi(CPUHP_IBMVNIC_DEAD, "net/ibmvnic:dead",
6563 				      NULL, ibmvnic_cpu_dead);
6564 	if (ret)
6565 		goto err_dead;
6566 
6567 	ret = vio_register_driver(&ibmvnic_driver);
6568 	if (ret)
6569 		goto err_vio_register;
6570 
6571 	pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
6572 		IBMVNIC_DRIVER_VERSION);
6573 
6574 	return 0;
6575 err_vio_register:
6576 	cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6577 err_dead:
6578 	cpuhp_remove_multi_state(ibmvnic_online);
6579 out:
6580 	return ret;
6581 }
6582 
6583 static void __exit ibmvnic_module_exit(void)
6584 {
6585 	vio_unregister_driver(&ibmvnic_driver);
6586 	cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6587 	cpuhp_remove_multi_state(ibmvnic_online);
6588 }
6589 
6590 module_init(ibmvnic_module_init);
6591 module_exit(ibmvnic_module_exit);
6592