xref: /linux/drivers/net/ethernet/ibm/ibmvnic.c (revision 55d0969c451159cff86949b38c39171cab962069)
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 int reset_sub_crq_queues(struct ibmvnic_adapter *adapter);
101 static inline void reinit_init_done(struct ibmvnic_adapter *adapter);
102 static void send_query_map(struct ibmvnic_adapter *adapter);
103 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
104 static int send_request_unmap(struct ibmvnic_adapter *, u8);
105 static int send_login(struct ibmvnic_adapter *adapter);
106 static void send_query_cap(struct ibmvnic_adapter *adapter);
107 static int init_sub_crqs(struct ibmvnic_adapter *);
108 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
109 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
110 static void release_crq_queue(struct ibmvnic_adapter *);
111 static int __ibmvnic_set_mac(struct net_device *, u8 *);
112 static int init_crq_queue(struct ibmvnic_adapter *adapter);
113 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
114 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
115 					 struct ibmvnic_sub_crq_queue *tx_scrq);
116 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
117 				struct ibmvnic_long_term_buff *ltb);
118 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter);
119 static void flush_reset_queue(struct ibmvnic_adapter *adapter);
120 static void print_subcrq_error(struct device *dev, int rc, const char *func);
121 
122 struct ibmvnic_stat {
123 	char name[ETH_GSTRING_LEN];
124 	int offset;
125 };
126 
127 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
128 			     offsetof(struct ibmvnic_statistics, stat))
129 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
130 
131 static const struct ibmvnic_stat ibmvnic_stats[] = {
132 	{"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
133 	{"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
134 	{"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
135 	{"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
136 	{"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
137 	{"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
138 	{"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
139 	{"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
140 	{"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
141 	{"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
142 	{"align_errors", IBMVNIC_STAT_OFF(align_errors)},
143 	{"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
144 	{"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
145 	{"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
146 	{"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
147 	{"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
148 	{"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
149 	{"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
150 	{"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
151 	{"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
152 	{"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
153 	{"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
154 };
155 
156 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
157 {
158 	union ibmvnic_crq crq;
159 
160 	memset(&crq, 0, sizeof(crq));
161 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
162 	crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
163 
164 	return ibmvnic_send_crq(adapter, &crq);
165 }
166 
167 static int send_version_xchg(struct ibmvnic_adapter *adapter)
168 {
169 	union ibmvnic_crq crq;
170 
171 	memset(&crq, 0, sizeof(crq));
172 	crq.version_exchange.first = IBMVNIC_CRQ_CMD;
173 	crq.version_exchange.cmd = VERSION_EXCHANGE;
174 	crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
175 
176 	return ibmvnic_send_crq(adapter, &crq);
177 }
178 
179 static void ibmvnic_clean_queue_affinity(struct ibmvnic_adapter *adapter,
180 					 struct ibmvnic_sub_crq_queue *queue)
181 {
182 	if (!(queue && queue->irq))
183 		return;
184 
185 	cpumask_clear(queue->affinity_mask);
186 
187 	if (irq_set_affinity_and_hint(queue->irq, NULL))
188 		netdev_warn(adapter->netdev,
189 			    "%s: Clear affinity failed, queue addr = %p, IRQ = %d\n",
190 			    __func__, queue, queue->irq);
191 }
192 
193 static void ibmvnic_clean_affinity(struct ibmvnic_adapter *adapter)
194 {
195 	struct ibmvnic_sub_crq_queue **rxqs;
196 	struct ibmvnic_sub_crq_queue **txqs;
197 	int num_rxqs, num_txqs;
198 	int i;
199 
200 	rxqs = adapter->rx_scrq;
201 	txqs = adapter->tx_scrq;
202 	num_txqs = adapter->num_active_tx_scrqs;
203 	num_rxqs = adapter->num_active_rx_scrqs;
204 
205 	netdev_dbg(adapter->netdev, "%s: Cleaning irq affinity hints", __func__);
206 	if (txqs) {
207 		for (i = 0; i < num_txqs; i++)
208 			ibmvnic_clean_queue_affinity(adapter, txqs[i]);
209 	}
210 	if (rxqs) {
211 		for (i = 0; i < num_rxqs; i++)
212 			ibmvnic_clean_queue_affinity(adapter, rxqs[i]);
213 	}
214 }
215 
216 static int ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue *queue,
217 				      unsigned int *cpu, int *stragglers,
218 				      int stride)
219 {
220 	cpumask_var_t mask;
221 	int i;
222 	int rc = 0;
223 
224 	if (!(queue && queue->irq))
225 		return rc;
226 
227 	/* cpumask_var_t is either a pointer or array, allocation works here */
228 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
229 		return -ENOMEM;
230 
231 	/* while we have extra cpu give one extra to this irq */
232 	if (*stragglers) {
233 		stride++;
234 		(*stragglers)--;
235 	}
236 	/* atomic write is safer than writing bit by bit directly */
237 	for (i = 0; i < stride; i++) {
238 		cpumask_set_cpu(*cpu, mask);
239 		*cpu = cpumask_next_wrap(*cpu, cpu_online_mask,
240 					 nr_cpu_ids, false);
241 	}
242 	/* set queue affinity mask */
243 	cpumask_copy(queue->affinity_mask, mask);
244 	rc = irq_set_affinity_and_hint(queue->irq, queue->affinity_mask);
245 	free_cpumask_var(mask);
246 
247 	return rc;
248 }
249 
250 /* assumes cpu read lock is held */
251 static void ibmvnic_set_affinity(struct ibmvnic_adapter *adapter)
252 {
253 	struct ibmvnic_sub_crq_queue **rxqs = adapter->rx_scrq;
254 	struct ibmvnic_sub_crq_queue **txqs = adapter->tx_scrq;
255 	struct ibmvnic_sub_crq_queue *queue;
256 	int num_rxqs = adapter->num_active_rx_scrqs, i_rxqs = 0;
257 	int num_txqs = adapter->num_active_tx_scrqs, i_txqs = 0;
258 	int total_queues, stride, stragglers, i;
259 	unsigned int num_cpu, cpu;
260 	bool is_rx_queue;
261 	int rc = 0;
262 
263 	netdev_dbg(adapter->netdev, "%s: Setting irq affinity hints", __func__);
264 	if (!(adapter->rx_scrq && adapter->tx_scrq)) {
265 		netdev_warn(adapter->netdev,
266 			    "%s: Set affinity failed, queues not allocated\n",
267 			    __func__);
268 		return;
269 	}
270 
271 	total_queues = num_rxqs + num_txqs;
272 	num_cpu = num_online_cpus();
273 	/* number of cpu's assigned per irq */
274 	stride = max_t(int, num_cpu / total_queues, 1);
275 	/* number of leftover cpu's */
276 	stragglers = num_cpu >= total_queues ? num_cpu % total_queues : 0;
277 	/* next available cpu to assign irq to */
278 	cpu = cpumask_next(-1, cpu_online_mask);
279 
280 	for (i = 0; i < total_queues; i++) {
281 		is_rx_queue = false;
282 		/* balance core load by alternating rx and tx assignments
283 		 * ex: TX0 -> RX0 -> TX1 -> RX1 etc.
284 		 */
285 		if ((i % 2 == 1 && i_rxqs < num_rxqs) || i_txqs == num_txqs) {
286 			queue = rxqs[i_rxqs++];
287 			is_rx_queue = true;
288 		} else {
289 			queue = txqs[i_txqs++];
290 		}
291 
292 		rc = ibmvnic_set_queue_affinity(queue, &cpu, &stragglers,
293 						stride);
294 		if (rc)
295 			goto out;
296 
297 		if (!queue || is_rx_queue)
298 			continue;
299 
300 		rc = __netif_set_xps_queue(adapter->netdev,
301 					   cpumask_bits(queue->affinity_mask),
302 					   i_txqs - 1, XPS_CPUS);
303 		if (rc)
304 			netdev_warn(adapter->netdev, "%s: Set XPS on queue %d failed, rc = %d.\n",
305 				    __func__, i_txqs - 1, rc);
306 	}
307 
308 out:
309 	if (rc) {
310 		netdev_warn(adapter->netdev,
311 			    "%s: Set affinity failed, queue addr = %p, IRQ = %d, rc = %d.\n",
312 			    __func__, queue, queue->irq, rc);
313 		ibmvnic_clean_affinity(adapter);
314 	}
315 }
316 
317 static int ibmvnic_cpu_online(unsigned int cpu, struct hlist_node *node)
318 {
319 	struct ibmvnic_adapter *adapter;
320 
321 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
322 	ibmvnic_set_affinity(adapter);
323 	return 0;
324 }
325 
326 static int ibmvnic_cpu_dead(unsigned int cpu, struct hlist_node *node)
327 {
328 	struct ibmvnic_adapter *adapter;
329 
330 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node_dead);
331 	ibmvnic_set_affinity(adapter);
332 	return 0;
333 }
334 
335 static int ibmvnic_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
336 {
337 	struct ibmvnic_adapter *adapter;
338 
339 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
340 	ibmvnic_clean_affinity(adapter);
341 	return 0;
342 }
343 
344 static enum cpuhp_state ibmvnic_online;
345 
346 static int ibmvnic_cpu_notif_add(struct ibmvnic_adapter *adapter)
347 {
348 	int ret;
349 
350 	ret = cpuhp_state_add_instance_nocalls(ibmvnic_online, &adapter->node);
351 	if (ret)
352 		return ret;
353 	ret = cpuhp_state_add_instance_nocalls(CPUHP_IBMVNIC_DEAD,
354 					       &adapter->node_dead);
355 	if (!ret)
356 		return ret;
357 	cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node);
358 	return ret;
359 }
360 
361 static void ibmvnic_cpu_notif_remove(struct ibmvnic_adapter *adapter)
362 {
363 	cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node);
364 	cpuhp_state_remove_instance_nocalls(CPUHP_IBMVNIC_DEAD,
365 					    &adapter->node_dead);
366 }
367 
368 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
369 			  unsigned long length, unsigned long *number,
370 			  unsigned long *irq)
371 {
372 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
373 	long rc;
374 
375 	rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
376 	*number = retbuf[0];
377 	*irq = retbuf[1];
378 
379 	return rc;
380 }
381 
382 /**
383  * ibmvnic_wait_for_completion - Check device state and wait for completion
384  * @adapter: private device data
385  * @comp_done: completion structure to wait for
386  * @timeout: time to wait in milliseconds
387  *
388  * Wait for a completion signal or until the timeout limit is reached
389  * while checking that the device is still active.
390  */
391 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
392 				       struct completion *comp_done,
393 				       unsigned long timeout)
394 {
395 	struct net_device *netdev;
396 	unsigned long div_timeout;
397 	u8 retry;
398 
399 	netdev = adapter->netdev;
400 	retry = 5;
401 	div_timeout = msecs_to_jiffies(timeout / retry);
402 	while (true) {
403 		if (!adapter->crq.active) {
404 			netdev_err(netdev, "Device down!\n");
405 			return -ENODEV;
406 		}
407 		if (!retry--)
408 			break;
409 		if (wait_for_completion_timeout(comp_done, div_timeout))
410 			return 0;
411 	}
412 	netdev_err(netdev, "Operation timed out.\n");
413 	return -ETIMEDOUT;
414 }
415 
416 /**
417  * reuse_ltb() - Check if a long term buffer can be reused
418  * @ltb:  The long term buffer to be checked
419  * @size: The size of the long term buffer.
420  *
421  * An LTB can be reused unless its size has changed.
422  *
423  * Return: Return true if the LTB can be reused, false otherwise.
424  */
425 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
426 {
427 	return (ltb->buff && ltb->size == size);
428 }
429 
430 /**
431  * alloc_long_term_buff() - Allocate a long term buffer (LTB)
432  *
433  * @adapter: ibmvnic adapter associated to the LTB
434  * @ltb:     container object for the LTB
435  * @size:    size of the LTB
436  *
437  * Allocate an LTB of the specified size and notify VIOS.
438  *
439  * If the given @ltb already has the correct size, reuse it. Otherwise if
440  * its non-NULL, free it. Then allocate a new one of the correct size.
441  * Notify the VIOS either way since we may now be working with a new VIOS.
442  *
443  * Allocating larger chunks of memory during resets, specially LPM or under
444  * low memory situations can cause resets to fail/timeout and for LPAR to
445  * lose connectivity. So hold onto the LTB even if we fail to communicate
446  * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
447  *
448  * Return: 0 if we were able to allocate the LTB and notify the VIOS and
449  *	   a negative value otherwise.
450  */
451 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
452 				struct ibmvnic_long_term_buff *ltb, int size)
453 {
454 	struct device *dev = &adapter->vdev->dev;
455 	u64 prev = 0;
456 	int rc;
457 
458 	if (!reuse_ltb(ltb, size)) {
459 		dev_dbg(dev,
460 			"LTB size changed from 0x%llx to 0x%x, reallocating\n",
461 			 ltb->size, size);
462 		prev = ltb->size;
463 		free_long_term_buff(adapter, ltb);
464 	}
465 
466 	if (ltb->buff) {
467 		dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
468 			ltb->map_id, ltb->size);
469 	} else {
470 		ltb->buff = dma_alloc_coherent(dev, size, &ltb->addr,
471 					       GFP_KERNEL);
472 		if (!ltb->buff) {
473 			dev_err(dev, "Couldn't alloc long term buffer\n");
474 			return -ENOMEM;
475 		}
476 		ltb->size = size;
477 
478 		ltb->map_id = find_first_zero_bit(adapter->map_ids,
479 						  MAX_MAP_ID);
480 		bitmap_set(adapter->map_ids, ltb->map_id, 1);
481 
482 		dev_dbg(dev,
483 			"Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n",
484 			 ltb->map_id, ltb->size, prev);
485 	}
486 
487 	/* Ensure ltb is zeroed - specially when reusing it. */
488 	memset(ltb->buff, 0, ltb->size);
489 
490 	mutex_lock(&adapter->fw_lock);
491 	adapter->fw_done_rc = 0;
492 	reinit_completion(&adapter->fw_done);
493 
494 	rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
495 	if (rc) {
496 		dev_err(dev, "send_request_map failed, rc = %d\n", rc);
497 		goto out;
498 	}
499 
500 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
501 	if (rc) {
502 		dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
503 			rc);
504 		goto out;
505 	}
506 
507 	if (adapter->fw_done_rc) {
508 		dev_err(dev, "Couldn't map LTB, rc = %d\n",
509 			adapter->fw_done_rc);
510 		rc = -EIO;
511 		goto out;
512 	}
513 	rc = 0;
514 out:
515 	/* don't free LTB on communication error - see function header */
516 	mutex_unlock(&adapter->fw_lock);
517 	return rc;
518 }
519 
520 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
521 				struct ibmvnic_long_term_buff *ltb)
522 {
523 	struct device *dev = &adapter->vdev->dev;
524 
525 	if (!ltb->buff)
526 		return;
527 
528 	/* VIOS automatically unmaps the long term buffer at remote
529 	 * end for the following resets:
530 	 * FAILOVER, MOBILITY, TIMEOUT.
531 	 */
532 	if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
533 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
534 	    adapter->reset_reason != VNIC_RESET_TIMEOUT)
535 		send_request_unmap(adapter, ltb->map_id);
536 
537 	dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
538 
539 	ltb->buff = NULL;
540 	/* mark this map_id free */
541 	bitmap_clear(adapter->map_ids, ltb->map_id, 1);
542 	ltb->map_id = 0;
543 }
544 
545 /**
546  * free_ltb_set - free the given set of long term buffers (LTBS)
547  * @adapter: The ibmvnic adapter containing this ltb set
548  * @ltb_set: The ltb_set to be freed
549  *
550  * Free the set of LTBs in the given set.
551  */
552 
553 static void free_ltb_set(struct ibmvnic_adapter *adapter,
554 			 struct ibmvnic_ltb_set *ltb_set)
555 {
556 	int i;
557 
558 	for (i = 0; i < ltb_set->num_ltbs; i++)
559 		free_long_term_buff(adapter, &ltb_set->ltbs[i]);
560 
561 	kfree(ltb_set->ltbs);
562 	ltb_set->ltbs = NULL;
563 	ltb_set->num_ltbs = 0;
564 }
565 
566 /**
567  * alloc_ltb_set() - Allocate a set of long term buffers (LTBs)
568  *
569  * @adapter: ibmvnic adapter associated to the LTB
570  * @ltb_set: container object for the set of LTBs
571  * @num_buffs: Number of buffers in the LTB
572  * @buff_size: Size of each buffer in the LTB
573  *
574  * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size
575  * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the
576  * new set of LTBs have fewer LTBs than the old set, free the excess LTBs.
577  * If new set needs more than in old set, allocate the remaining ones.
578  * Try and reuse as many LTBs as possible and avoid reallocation.
579  *
580  * Any changes to this allocation strategy must be reflected in
581  * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb().
582  */
583 static int alloc_ltb_set(struct ibmvnic_adapter *adapter,
584 			 struct ibmvnic_ltb_set *ltb_set, int num_buffs,
585 			 int buff_size)
586 {
587 	struct device *dev = &adapter->vdev->dev;
588 	struct ibmvnic_ltb_set old_set;
589 	struct ibmvnic_ltb_set new_set;
590 	int rem_size;
591 	int tot_size;		/* size of all ltbs */
592 	int ltb_size;		/* size of one ltb */
593 	int nltbs;
594 	int rc;
595 	int n;
596 	int i;
597 
598 	dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs,
599 		buff_size);
600 
601 	ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size);
602 	tot_size = num_buffs * buff_size;
603 
604 	if (ltb_size > tot_size)
605 		ltb_size = tot_size;
606 
607 	nltbs = tot_size / ltb_size;
608 	if (tot_size % ltb_size)
609 		nltbs++;
610 
611 	old_set = *ltb_set;
612 
613 	if (old_set.num_ltbs == nltbs) {
614 		new_set = old_set;
615 	} else {
616 		int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff);
617 
618 		new_set.ltbs = kzalloc(tmp, GFP_KERNEL);
619 		if (!new_set.ltbs)
620 			return -ENOMEM;
621 
622 		new_set.num_ltbs = nltbs;
623 
624 		/* Free any excess ltbs in old set */
625 		for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++)
626 			free_long_term_buff(adapter, &old_set.ltbs[i]);
627 
628 		/* Copy remaining ltbs to new set. All LTBs except the
629 		 * last one are of the same size. alloc_long_term_buff()
630 		 * will realloc if the size changes.
631 		 */
632 		n = min(old_set.num_ltbs, new_set.num_ltbs);
633 		for (i = 0; i < n; i++)
634 			new_set.ltbs[i] = old_set.ltbs[i];
635 
636 		/* Any additional ltbs in new set will have NULL ltbs for
637 		 * now and will be allocated in alloc_long_term_buff().
638 		 */
639 
640 		/* We no longer need the old_set so free it. Note that we
641 		 * may have reused some ltbs from old set and freed excess
642 		 * ltbs above. So we only need to free the container now
643 		 * not the LTBs themselves. (i.e. dont free_ltb_set()!)
644 		 */
645 		kfree(old_set.ltbs);
646 		old_set.ltbs = NULL;
647 		old_set.num_ltbs = 0;
648 
649 		/* Install the new set. If allocations fail below, we will
650 		 * retry later and know what size LTBs we need.
651 		 */
652 		*ltb_set = new_set;
653 	}
654 
655 	i = 0;
656 	rem_size = tot_size;
657 	while (rem_size) {
658 		if (ltb_size > rem_size)
659 			ltb_size = rem_size;
660 
661 		rem_size -= ltb_size;
662 
663 		rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size);
664 		if (rc)
665 			goto out;
666 		i++;
667 	}
668 
669 	WARN_ON(i != new_set.num_ltbs);
670 
671 	return 0;
672 out:
673 	/* We may have allocated one/more LTBs before failing and we
674 	 * want to try and reuse on next reset. So don't free ltb set.
675 	 */
676 	return rc;
677 }
678 
679 /**
680  * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB.
681  * @rxpool: The receive buffer pool containing buffer
682  * @bufidx: Index of buffer in rxpool
683  * @ltbp: (Output) pointer to the long term buffer containing the buffer
684  * @offset: (Output) offset of buffer in the LTB from @ltbp
685  *
686  * Map the given buffer identified by [rxpool, bufidx] to an LTB in the
687  * pool and its corresponding offset. Assume for now that each LTB is of
688  * different size but could possibly be optimized based on the allocation
689  * strategy in alloc_ltb_set().
690  */
691 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool,
692 				  unsigned int bufidx,
693 				  struct ibmvnic_long_term_buff **ltbp,
694 				  unsigned int *offset)
695 {
696 	struct ibmvnic_long_term_buff *ltb;
697 	int nbufs;	/* # of buffers in one ltb */
698 	int i;
699 
700 	WARN_ON(bufidx >= rxpool->size);
701 
702 	for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) {
703 		ltb = &rxpool->ltb_set.ltbs[i];
704 		nbufs = ltb->size / rxpool->buff_size;
705 		if (bufidx < nbufs)
706 			break;
707 		bufidx -= nbufs;
708 	}
709 
710 	*ltbp = ltb;
711 	*offset = bufidx * rxpool->buff_size;
712 }
713 
714 /**
715  * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB.
716  * @txpool: The transmit buffer pool containing buffer
717  * @bufidx: Index of buffer in txpool
718  * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer
719  * @offset: (Output) offset of buffer in the LTB from @ltbp
720  *
721  * Map the given buffer identified by [txpool, bufidx] to an LTB in the
722  * pool and its corresponding offset.
723  */
724 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool,
725 				  unsigned int bufidx,
726 				  struct ibmvnic_long_term_buff **ltbp,
727 				  unsigned int *offset)
728 {
729 	struct ibmvnic_long_term_buff *ltb;
730 	int nbufs;	/* # of buffers in one ltb */
731 	int i;
732 
733 	WARN_ON_ONCE(bufidx >= txpool->num_buffers);
734 
735 	for (i = 0; i < txpool->ltb_set.num_ltbs; i++) {
736 		ltb = &txpool->ltb_set.ltbs[i];
737 		nbufs = ltb->size / txpool->buf_size;
738 		if (bufidx < nbufs)
739 			break;
740 		bufidx -= nbufs;
741 	}
742 
743 	*ltbp = ltb;
744 	*offset = bufidx * txpool->buf_size;
745 }
746 
747 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
748 {
749 	int i;
750 
751 	for (i = 0; i < adapter->num_active_rx_pools; i++)
752 		adapter->rx_pool[i].active = 0;
753 }
754 
755 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
756 			      struct ibmvnic_rx_pool *pool)
757 {
758 	int count = pool->size - atomic_read(&pool->available);
759 	u64 handle = adapter->rx_scrq[pool->index]->handle;
760 	struct device *dev = &adapter->vdev->dev;
761 	struct ibmvnic_ind_xmit_queue *ind_bufp;
762 	struct ibmvnic_sub_crq_queue *rx_scrq;
763 	struct ibmvnic_long_term_buff *ltb;
764 	union sub_crq *sub_crq;
765 	int buffers_added = 0;
766 	unsigned long lpar_rc;
767 	struct sk_buff *skb;
768 	unsigned int offset;
769 	dma_addr_t dma_addr;
770 	unsigned char *dst;
771 	int shift = 0;
772 	int bufidx;
773 	int i;
774 
775 	if (!pool->active)
776 		return;
777 
778 	rx_scrq = adapter->rx_scrq[pool->index];
779 	ind_bufp = &rx_scrq->ind_buf;
780 
781 	/* netdev_skb_alloc() could have failed after we saved a few skbs
782 	 * in the indir_buf and we would not have sent them to VIOS yet.
783 	 * To account for them, start the loop at ind_bufp->index rather
784 	 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
785 	 * be 0.
786 	 */
787 	for (i = ind_bufp->index; i < count; ++i) {
788 		bufidx = pool->free_map[pool->next_free];
789 
790 		/* We maybe reusing the skb from earlier resets. Allocate
791 		 * only if necessary. But since the LTB may have changed
792 		 * during reset (see init_rx_pools()), update LTB below
793 		 * even if reusing skb.
794 		 */
795 		skb = pool->rx_buff[bufidx].skb;
796 		if (!skb) {
797 			skb = netdev_alloc_skb(adapter->netdev,
798 					       pool->buff_size);
799 			if (!skb) {
800 				dev_err(dev, "Couldn't replenish rx buff\n");
801 				adapter->replenish_no_mem++;
802 				break;
803 			}
804 		}
805 
806 		pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
807 		pool->next_free = (pool->next_free + 1) % pool->size;
808 
809 		/* Copy the skb to the long term mapped DMA buffer */
810 		map_rxpool_buf_to_ltb(pool, bufidx, &ltb, &offset);
811 		dst = ltb->buff + offset;
812 		memset(dst, 0, pool->buff_size);
813 		dma_addr = ltb->addr + offset;
814 
815 		/* add the skb to an rx_buff in the pool */
816 		pool->rx_buff[bufidx].data = dst;
817 		pool->rx_buff[bufidx].dma = dma_addr;
818 		pool->rx_buff[bufidx].skb = skb;
819 		pool->rx_buff[bufidx].pool_index = pool->index;
820 		pool->rx_buff[bufidx].size = pool->buff_size;
821 
822 		/* queue the rx_buff for the next send_subcrq_indirect */
823 		sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
824 		memset(sub_crq, 0, sizeof(*sub_crq));
825 		sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
826 		sub_crq->rx_add.correlator =
827 		    cpu_to_be64((u64)&pool->rx_buff[bufidx]);
828 		sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
829 		sub_crq->rx_add.map_id = ltb->map_id;
830 
831 		/* The length field of the sCRQ is defined to be 24 bits so the
832 		 * buffer size needs to be left shifted by a byte before it is
833 		 * converted to big endian to prevent the last byte from being
834 		 * truncated.
835 		 */
836 #ifdef __LITTLE_ENDIAN__
837 		shift = 8;
838 #endif
839 		sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
840 
841 		/* if send_subcrq_indirect queue is full, flush to VIOS */
842 		if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
843 		    i == count - 1) {
844 			lpar_rc =
845 				send_subcrq_indirect(adapter, handle,
846 						     (u64)ind_bufp->indir_dma,
847 						     (u64)ind_bufp->index);
848 			if (lpar_rc != H_SUCCESS)
849 				goto failure;
850 			buffers_added += ind_bufp->index;
851 			adapter->replenish_add_buff_success += ind_bufp->index;
852 			ind_bufp->index = 0;
853 		}
854 	}
855 	atomic_add(buffers_added, &pool->available);
856 	return;
857 
858 failure:
859 	if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
860 		dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
861 	for (i = ind_bufp->index - 1; i >= 0; --i) {
862 		struct ibmvnic_rx_buff *rx_buff;
863 
864 		pool->next_free = pool->next_free == 0 ?
865 				  pool->size - 1 : pool->next_free - 1;
866 		sub_crq = &ind_bufp->indir_arr[i];
867 		rx_buff = (struct ibmvnic_rx_buff *)
868 				be64_to_cpu(sub_crq->rx_add.correlator);
869 		bufidx = (int)(rx_buff - pool->rx_buff);
870 		pool->free_map[pool->next_free] = bufidx;
871 		dev_kfree_skb_any(pool->rx_buff[bufidx].skb);
872 		pool->rx_buff[bufidx].skb = NULL;
873 	}
874 	adapter->replenish_add_buff_failure += ind_bufp->index;
875 	atomic_add(buffers_added, &pool->available);
876 	ind_bufp->index = 0;
877 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
878 		/* Disable buffer pool replenishment and report carrier off if
879 		 * queue is closed or pending failover.
880 		 * Firmware guarantees that a signal will be sent to the
881 		 * driver, triggering a reset.
882 		 */
883 		deactivate_rx_pools(adapter);
884 		netif_carrier_off(adapter->netdev);
885 	}
886 }
887 
888 static void replenish_pools(struct ibmvnic_adapter *adapter)
889 {
890 	int i;
891 
892 	adapter->replenish_task_cycles++;
893 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
894 		if (adapter->rx_pool[i].active)
895 			replenish_rx_pool(adapter, &adapter->rx_pool[i]);
896 	}
897 
898 	netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
899 }
900 
901 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
902 {
903 	kfree(adapter->tx_stats_buffers);
904 	kfree(adapter->rx_stats_buffers);
905 	adapter->tx_stats_buffers = NULL;
906 	adapter->rx_stats_buffers = NULL;
907 }
908 
909 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
910 {
911 	adapter->tx_stats_buffers =
912 				kcalloc(IBMVNIC_MAX_QUEUES,
913 					sizeof(struct ibmvnic_tx_queue_stats),
914 					GFP_KERNEL);
915 	if (!adapter->tx_stats_buffers)
916 		return -ENOMEM;
917 
918 	adapter->rx_stats_buffers =
919 				kcalloc(IBMVNIC_MAX_QUEUES,
920 					sizeof(struct ibmvnic_rx_queue_stats),
921 					GFP_KERNEL);
922 	if (!adapter->rx_stats_buffers)
923 		return -ENOMEM;
924 
925 	return 0;
926 }
927 
928 static void release_stats_token(struct ibmvnic_adapter *adapter)
929 {
930 	struct device *dev = &adapter->vdev->dev;
931 
932 	if (!adapter->stats_token)
933 		return;
934 
935 	dma_unmap_single(dev, adapter->stats_token,
936 			 sizeof(struct ibmvnic_statistics),
937 			 DMA_FROM_DEVICE);
938 	adapter->stats_token = 0;
939 }
940 
941 static int init_stats_token(struct ibmvnic_adapter *adapter)
942 {
943 	struct device *dev = &adapter->vdev->dev;
944 	dma_addr_t stok;
945 	int rc;
946 
947 	stok = dma_map_single(dev, &adapter->stats,
948 			      sizeof(struct ibmvnic_statistics),
949 			      DMA_FROM_DEVICE);
950 	rc = dma_mapping_error(dev, stok);
951 	if (rc) {
952 		dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
953 		return rc;
954 	}
955 
956 	adapter->stats_token = stok;
957 	netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
958 	return 0;
959 }
960 
961 /**
962  * release_rx_pools() - Release any rx pools attached to @adapter.
963  * @adapter: ibmvnic adapter
964  *
965  * Safe to call this multiple times - even if no pools are attached.
966  */
967 static void release_rx_pools(struct ibmvnic_adapter *adapter)
968 {
969 	struct ibmvnic_rx_pool *rx_pool;
970 	int i, j;
971 
972 	if (!adapter->rx_pool)
973 		return;
974 
975 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
976 		rx_pool = &adapter->rx_pool[i];
977 
978 		netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
979 
980 		kfree(rx_pool->free_map);
981 
982 		free_ltb_set(adapter, &rx_pool->ltb_set);
983 
984 		if (!rx_pool->rx_buff)
985 			continue;
986 
987 		for (j = 0; j < rx_pool->size; j++) {
988 			if (rx_pool->rx_buff[j].skb) {
989 				dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
990 				rx_pool->rx_buff[j].skb = NULL;
991 			}
992 		}
993 
994 		kfree(rx_pool->rx_buff);
995 	}
996 
997 	kfree(adapter->rx_pool);
998 	adapter->rx_pool = NULL;
999 	adapter->num_active_rx_pools = 0;
1000 	adapter->prev_rx_pool_size = 0;
1001 }
1002 
1003 /**
1004  * reuse_rx_pools() - Check if the existing rx pools can be reused.
1005  * @adapter: ibmvnic adapter
1006  *
1007  * Check if the existing rx pools in the adapter can be reused. The
1008  * pools can be reused if the pool parameters (number of pools,
1009  * number of buffers in the pool and size of each buffer) have not
1010  * changed.
1011  *
1012  * NOTE: This assumes that all pools have the same number of buffers
1013  *       which is the case currently. If that changes, we must fix this.
1014  *
1015  * Return: true if the rx pools can be reused, false otherwise.
1016  */
1017 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
1018 {
1019 	u64 old_num_pools, new_num_pools;
1020 	u64 old_pool_size, new_pool_size;
1021 	u64 old_buff_size, new_buff_size;
1022 
1023 	if (!adapter->rx_pool)
1024 		return false;
1025 
1026 	old_num_pools = adapter->num_active_rx_pools;
1027 	new_num_pools = adapter->req_rx_queues;
1028 
1029 	old_pool_size = adapter->prev_rx_pool_size;
1030 	new_pool_size = adapter->req_rx_add_entries_per_subcrq;
1031 
1032 	old_buff_size = adapter->prev_rx_buf_sz;
1033 	new_buff_size = adapter->cur_rx_buf_sz;
1034 
1035 	if (old_buff_size != new_buff_size ||
1036 	    old_num_pools != new_num_pools ||
1037 	    old_pool_size != new_pool_size)
1038 		return false;
1039 
1040 	return true;
1041 }
1042 
1043 /**
1044  * init_rx_pools(): Initialize the set of receiver pools in the adapter.
1045  * @netdev: net device associated with the vnic interface
1046  *
1047  * Initialize the set of receiver pools in the ibmvnic adapter associated
1048  * with the net_device @netdev. If possible, reuse the existing rx pools.
1049  * Otherwise free any existing pools and  allocate a new set of pools
1050  * before initializing them.
1051  *
1052  * Return: 0 on success and negative value on error.
1053  */
1054 static int init_rx_pools(struct net_device *netdev)
1055 {
1056 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1057 	struct device *dev = &adapter->vdev->dev;
1058 	struct ibmvnic_rx_pool *rx_pool;
1059 	u64 num_pools;
1060 	u64 pool_size;		/* # of buffers in one pool */
1061 	u64 buff_size;
1062 	int i, j, rc;
1063 
1064 	pool_size = adapter->req_rx_add_entries_per_subcrq;
1065 	num_pools = adapter->req_rx_queues;
1066 	buff_size = adapter->cur_rx_buf_sz;
1067 
1068 	if (reuse_rx_pools(adapter)) {
1069 		dev_dbg(dev, "Reusing rx pools\n");
1070 		goto update_ltb;
1071 	}
1072 
1073 	/* Allocate/populate the pools. */
1074 	release_rx_pools(adapter);
1075 
1076 	adapter->rx_pool = kcalloc(num_pools,
1077 				   sizeof(struct ibmvnic_rx_pool),
1078 				   GFP_KERNEL);
1079 	if (!adapter->rx_pool) {
1080 		dev_err(dev, "Failed to allocate rx pools\n");
1081 		return -ENOMEM;
1082 	}
1083 
1084 	/* Set num_active_rx_pools early. If we fail below after partial
1085 	 * allocation, release_rx_pools() will know how many to look for.
1086 	 */
1087 	adapter->num_active_rx_pools = num_pools;
1088 
1089 	for (i = 0; i < num_pools; i++) {
1090 		rx_pool = &adapter->rx_pool[i];
1091 
1092 		netdev_dbg(adapter->netdev,
1093 			   "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
1094 			   i, pool_size, buff_size);
1095 
1096 		rx_pool->size = pool_size;
1097 		rx_pool->index = i;
1098 		rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1099 
1100 		rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
1101 					    GFP_KERNEL);
1102 		if (!rx_pool->free_map) {
1103 			dev_err(dev, "Couldn't alloc free_map %d\n", i);
1104 			rc = -ENOMEM;
1105 			goto out_release;
1106 		}
1107 
1108 		rx_pool->rx_buff = kcalloc(rx_pool->size,
1109 					   sizeof(struct ibmvnic_rx_buff),
1110 					   GFP_KERNEL);
1111 		if (!rx_pool->rx_buff) {
1112 			dev_err(dev, "Couldn't alloc rx buffers\n");
1113 			rc = -ENOMEM;
1114 			goto out_release;
1115 		}
1116 	}
1117 
1118 	adapter->prev_rx_pool_size = pool_size;
1119 	adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
1120 
1121 update_ltb:
1122 	for (i = 0; i < num_pools; i++) {
1123 		rx_pool = &adapter->rx_pool[i];
1124 		dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
1125 			i, rx_pool->size, rx_pool->buff_size);
1126 
1127 		rc = alloc_ltb_set(adapter, &rx_pool->ltb_set,
1128 				   rx_pool->size, rx_pool->buff_size);
1129 		if (rc)
1130 			goto out;
1131 
1132 		for (j = 0; j < rx_pool->size; ++j) {
1133 			struct ibmvnic_rx_buff *rx_buff;
1134 
1135 			rx_pool->free_map[j] = j;
1136 
1137 			/* NOTE: Don't clear rx_buff->skb here - will leak
1138 			 * memory! replenish_rx_pool() will reuse skbs or
1139 			 * allocate as necessary.
1140 			 */
1141 			rx_buff = &rx_pool->rx_buff[j];
1142 			rx_buff->dma = 0;
1143 			rx_buff->data = 0;
1144 			rx_buff->size = 0;
1145 			rx_buff->pool_index = 0;
1146 		}
1147 
1148 		/* Mark pool "empty" so replenish_rx_pools() will
1149 		 * update the LTB info for each buffer
1150 		 */
1151 		atomic_set(&rx_pool->available, 0);
1152 		rx_pool->next_alloc = 0;
1153 		rx_pool->next_free = 0;
1154 		/* replenish_rx_pool() may have called deactivate_rx_pools()
1155 		 * on failover. Ensure pool is active now.
1156 		 */
1157 		rx_pool->active = 1;
1158 	}
1159 	return 0;
1160 out_release:
1161 	release_rx_pools(adapter);
1162 out:
1163 	/* We failed to allocate one or more LTBs or map them on the VIOS.
1164 	 * Hold onto the pools and any LTBs that we did allocate/map.
1165 	 */
1166 	return rc;
1167 }
1168 
1169 static void release_vpd_data(struct ibmvnic_adapter *adapter)
1170 {
1171 	if (!adapter->vpd)
1172 		return;
1173 
1174 	kfree(adapter->vpd->buff);
1175 	kfree(adapter->vpd);
1176 
1177 	adapter->vpd = NULL;
1178 }
1179 
1180 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
1181 				struct ibmvnic_tx_pool *tx_pool)
1182 {
1183 	kfree(tx_pool->tx_buff);
1184 	kfree(tx_pool->free_map);
1185 	free_ltb_set(adapter, &tx_pool->ltb_set);
1186 }
1187 
1188 /**
1189  * release_tx_pools() - Release any tx pools attached to @adapter.
1190  * @adapter: ibmvnic adapter
1191  *
1192  * Safe to call this multiple times - even if no pools are attached.
1193  */
1194 static void release_tx_pools(struct ibmvnic_adapter *adapter)
1195 {
1196 	int i;
1197 
1198 	/* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
1199 	 * both NULL or both non-NULL. So we only need to check one.
1200 	 */
1201 	if (!adapter->tx_pool)
1202 		return;
1203 
1204 	for (i = 0; i < adapter->num_active_tx_pools; i++) {
1205 		release_one_tx_pool(adapter, &adapter->tx_pool[i]);
1206 		release_one_tx_pool(adapter, &adapter->tso_pool[i]);
1207 	}
1208 
1209 	kfree(adapter->tx_pool);
1210 	adapter->tx_pool = NULL;
1211 	kfree(adapter->tso_pool);
1212 	adapter->tso_pool = NULL;
1213 	adapter->num_active_tx_pools = 0;
1214 	adapter->prev_tx_pool_size = 0;
1215 }
1216 
1217 static int init_one_tx_pool(struct net_device *netdev,
1218 			    struct ibmvnic_tx_pool *tx_pool,
1219 			    int pool_size, int buf_size)
1220 {
1221 	int i;
1222 
1223 	tx_pool->tx_buff = kcalloc(pool_size,
1224 				   sizeof(struct ibmvnic_tx_buff),
1225 				   GFP_KERNEL);
1226 	if (!tx_pool->tx_buff)
1227 		return -ENOMEM;
1228 
1229 	tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
1230 	if (!tx_pool->free_map) {
1231 		kfree(tx_pool->tx_buff);
1232 		tx_pool->tx_buff = NULL;
1233 		return -ENOMEM;
1234 	}
1235 
1236 	for (i = 0; i < pool_size; i++)
1237 		tx_pool->free_map[i] = i;
1238 
1239 	tx_pool->consumer_index = 0;
1240 	tx_pool->producer_index = 0;
1241 	tx_pool->num_buffers = pool_size;
1242 	tx_pool->buf_size = buf_size;
1243 
1244 	return 0;
1245 }
1246 
1247 /**
1248  * reuse_tx_pools() - Check if the existing tx pools can be reused.
1249  * @adapter: ibmvnic adapter
1250  *
1251  * Check if the existing tx pools in the adapter can be reused. The
1252  * pools can be reused if the pool parameters (number of pools,
1253  * number of buffers in the pool and mtu) have not changed.
1254  *
1255  * NOTE: This assumes that all pools have the same number of buffers
1256  *       which is the case currently. If that changes, we must fix this.
1257  *
1258  * Return: true if the tx pools can be reused, false otherwise.
1259  */
1260 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
1261 {
1262 	u64 old_num_pools, new_num_pools;
1263 	u64 old_pool_size, new_pool_size;
1264 	u64 old_mtu, new_mtu;
1265 
1266 	if (!adapter->tx_pool)
1267 		return false;
1268 
1269 	old_num_pools = adapter->num_active_tx_pools;
1270 	new_num_pools = adapter->num_active_tx_scrqs;
1271 	old_pool_size = adapter->prev_tx_pool_size;
1272 	new_pool_size = adapter->req_tx_entries_per_subcrq;
1273 	old_mtu = adapter->prev_mtu;
1274 	new_mtu = adapter->req_mtu;
1275 
1276 	if (old_mtu != new_mtu ||
1277 	    old_num_pools != new_num_pools ||
1278 	    old_pool_size != new_pool_size)
1279 		return false;
1280 
1281 	return true;
1282 }
1283 
1284 /**
1285  * init_tx_pools(): Initialize the set of transmit pools in the adapter.
1286  * @netdev: net device associated with the vnic interface
1287  *
1288  * Initialize the set of transmit pools in the ibmvnic adapter associated
1289  * with the net_device @netdev. If possible, reuse the existing tx pools.
1290  * Otherwise free any existing pools and  allocate a new set of pools
1291  * before initializing them.
1292  *
1293  * Return: 0 on success and negative value on error.
1294  */
1295 static int init_tx_pools(struct net_device *netdev)
1296 {
1297 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1298 	struct device *dev = &adapter->vdev->dev;
1299 	int num_pools;
1300 	u64 pool_size;		/* # of buffers in pool */
1301 	u64 buff_size;
1302 	int i, j, rc;
1303 
1304 	num_pools = adapter->req_tx_queues;
1305 
1306 	/* We must notify the VIOS about the LTB on all resets - but we only
1307 	 * need to alloc/populate pools if either the number of buffers or
1308 	 * size of each buffer in the pool has changed.
1309 	 */
1310 	if (reuse_tx_pools(adapter)) {
1311 		netdev_dbg(netdev, "Reusing tx pools\n");
1312 		goto update_ltb;
1313 	}
1314 
1315 	/* Allocate/populate the pools. */
1316 	release_tx_pools(adapter);
1317 
1318 	pool_size = adapter->req_tx_entries_per_subcrq;
1319 	num_pools = adapter->num_active_tx_scrqs;
1320 
1321 	adapter->tx_pool = kcalloc(num_pools,
1322 				   sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1323 	if (!adapter->tx_pool)
1324 		return -ENOMEM;
1325 
1326 	adapter->tso_pool = kcalloc(num_pools,
1327 				    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1328 	/* To simplify release_tx_pools() ensure that ->tx_pool and
1329 	 * ->tso_pool are either both NULL or both non-NULL.
1330 	 */
1331 	if (!adapter->tso_pool) {
1332 		kfree(adapter->tx_pool);
1333 		adapter->tx_pool = NULL;
1334 		return -ENOMEM;
1335 	}
1336 
1337 	/* Set num_active_tx_pools early. If we fail below after partial
1338 	 * allocation, release_tx_pools() will know how many to look for.
1339 	 */
1340 	adapter->num_active_tx_pools = num_pools;
1341 
1342 	buff_size = adapter->req_mtu + VLAN_HLEN;
1343 	buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1344 
1345 	for (i = 0; i < num_pools; i++) {
1346 		dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
1347 			i, adapter->req_tx_entries_per_subcrq, buff_size);
1348 
1349 		rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
1350 				      pool_size, buff_size);
1351 		if (rc)
1352 			goto out_release;
1353 
1354 		rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
1355 				      IBMVNIC_TSO_BUFS,
1356 				      IBMVNIC_TSO_BUF_SZ);
1357 		if (rc)
1358 			goto out_release;
1359 	}
1360 
1361 	adapter->prev_tx_pool_size = pool_size;
1362 	adapter->prev_mtu = adapter->req_mtu;
1363 
1364 update_ltb:
1365 	/* NOTE: All tx_pools have the same number of buffers (which is
1366 	 *       same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
1367 	 *       buffers (see calls init_one_tx_pool() for these).
1368 	 *       For consistency, we use tx_pool->num_buffers and
1369 	 *       tso_pool->num_buffers below.
1370 	 */
1371 	rc = -1;
1372 	for (i = 0; i < num_pools; i++) {
1373 		struct ibmvnic_tx_pool *tso_pool;
1374 		struct ibmvnic_tx_pool *tx_pool;
1375 
1376 		tx_pool = &adapter->tx_pool[i];
1377 
1378 		dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n",
1379 			i, tx_pool->num_buffers, tx_pool->buf_size);
1380 
1381 		rc = alloc_ltb_set(adapter, &tx_pool->ltb_set,
1382 				   tx_pool->num_buffers, tx_pool->buf_size);
1383 		if (rc)
1384 			goto out;
1385 
1386 		tx_pool->consumer_index = 0;
1387 		tx_pool->producer_index = 0;
1388 
1389 		for (j = 0; j < tx_pool->num_buffers; j++)
1390 			tx_pool->free_map[j] = j;
1391 
1392 		tso_pool = &adapter->tso_pool[i];
1393 
1394 		dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n",
1395 			i, tso_pool->num_buffers, tso_pool->buf_size);
1396 
1397 		rc = alloc_ltb_set(adapter, &tso_pool->ltb_set,
1398 				   tso_pool->num_buffers, tso_pool->buf_size);
1399 		if (rc)
1400 			goto out;
1401 
1402 		tso_pool->consumer_index = 0;
1403 		tso_pool->producer_index = 0;
1404 
1405 		for (j = 0; j < tso_pool->num_buffers; j++)
1406 			tso_pool->free_map[j] = j;
1407 	}
1408 
1409 	return 0;
1410 out_release:
1411 	release_tx_pools(adapter);
1412 out:
1413 	/* We failed to allocate one or more LTBs or map them on the VIOS.
1414 	 * Hold onto the pools and any LTBs that we did allocate/map.
1415 	 */
1416 	return rc;
1417 }
1418 
1419 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1420 {
1421 	int i;
1422 
1423 	if (adapter->napi_enabled)
1424 		return;
1425 
1426 	for (i = 0; i < adapter->req_rx_queues; i++)
1427 		napi_enable(&adapter->napi[i]);
1428 
1429 	adapter->napi_enabled = true;
1430 }
1431 
1432 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1433 {
1434 	int i;
1435 
1436 	if (!adapter->napi_enabled)
1437 		return;
1438 
1439 	for (i = 0; i < adapter->req_rx_queues; i++) {
1440 		netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1441 		napi_disable(&adapter->napi[i]);
1442 	}
1443 
1444 	adapter->napi_enabled = false;
1445 }
1446 
1447 static int init_napi(struct ibmvnic_adapter *adapter)
1448 {
1449 	int i;
1450 
1451 	adapter->napi = kcalloc(adapter->req_rx_queues,
1452 				sizeof(struct napi_struct), GFP_KERNEL);
1453 	if (!adapter->napi)
1454 		return -ENOMEM;
1455 
1456 	for (i = 0; i < adapter->req_rx_queues; i++) {
1457 		netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1458 		netif_napi_add(adapter->netdev, &adapter->napi[i],
1459 			       ibmvnic_poll);
1460 	}
1461 
1462 	adapter->num_active_rx_napi = adapter->req_rx_queues;
1463 	return 0;
1464 }
1465 
1466 static void release_napi(struct ibmvnic_adapter *adapter)
1467 {
1468 	int i;
1469 
1470 	if (!adapter->napi)
1471 		return;
1472 
1473 	for (i = 0; i < adapter->num_active_rx_napi; i++) {
1474 		netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1475 		netif_napi_del(&adapter->napi[i]);
1476 	}
1477 
1478 	kfree(adapter->napi);
1479 	adapter->napi = NULL;
1480 	adapter->num_active_rx_napi = 0;
1481 	adapter->napi_enabled = false;
1482 }
1483 
1484 static const char *adapter_state_to_string(enum vnic_state state)
1485 {
1486 	switch (state) {
1487 	case VNIC_PROBING:
1488 		return "PROBING";
1489 	case VNIC_PROBED:
1490 		return "PROBED";
1491 	case VNIC_OPENING:
1492 		return "OPENING";
1493 	case VNIC_OPEN:
1494 		return "OPEN";
1495 	case VNIC_CLOSING:
1496 		return "CLOSING";
1497 	case VNIC_CLOSED:
1498 		return "CLOSED";
1499 	case VNIC_REMOVING:
1500 		return "REMOVING";
1501 	case VNIC_REMOVED:
1502 		return "REMOVED";
1503 	case VNIC_DOWN:
1504 		return "DOWN";
1505 	}
1506 	return "UNKNOWN";
1507 }
1508 
1509 static int ibmvnic_login(struct net_device *netdev)
1510 {
1511 	unsigned long flags, timeout = msecs_to_jiffies(20000);
1512 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1513 	int retry_count = 0;
1514 	int retries = 10;
1515 	bool retry;
1516 	int rc;
1517 
1518 	do {
1519 		retry = false;
1520 		if (retry_count > retries) {
1521 			netdev_warn(netdev, "Login attempts exceeded\n");
1522 			return -EACCES;
1523 		}
1524 
1525 		adapter->init_done_rc = 0;
1526 		reinit_completion(&adapter->init_done);
1527 		rc = send_login(adapter);
1528 		if (rc)
1529 			return rc;
1530 
1531 		if (!wait_for_completion_timeout(&adapter->init_done,
1532 						 timeout)) {
1533 			netdev_warn(netdev, "Login timed out\n");
1534 			adapter->login_pending = false;
1535 			goto partial_reset;
1536 		}
1537 
1538 		if (adapter->init_done_rc == ABORTED) {
1539 			netdev_warn(netdev, "Login aborted, retrying...\n");
1540 			retry = true;
1541 			adapter->init_done_rc = 0;
1542 			retry_count++;
1543 			/* FW or device may be busy, so
1544 			 * wait a bit before retrying login
1545 			 */
1546 			msleep(500);
1547 		} else if (adapter->init_done_rc == PARTIALSUCCESS) {
1548 			retry_count++;
1549 			release_sub_crqs(adapter, 1);
1550 
1551 			retry = true;
1552 			netdev_dbg(netdev,
1553 				   "Received partial success, retrying...\n");
1554 			adapter->init_done_rc = 0;
1555 			reinit_completion(&adapter->init_done);
1556 			send_query_cap(adapter);
1557 			if (!wait_for_completion_timeout(&adapter->init_done,
1558 							 timeout)) {
1559 				netdev_warn(netdev,
1560 					    "Capabilities query timed out\n");
1561 				return -ETIMEDOUT;
1562 			}
1563 
1564 			rc = init_sub_crqs(adapter);
1565 			if (rc) {
1566 				netdev_warn(netdev,
1567 					    "SCRQ initialization failed\n");
1568 				return rc;
1569 			}
1570 
1571 			rc = init_sub_crq_irqs(adapter);
1572 			if (rc) {
1573 				netdev_warn(netdev,
1574 					    "SCRQ irq initialization failed\n");
1575 				return rc;
1576 			}
1577 		/* Default/timeout error handling, reset and start fresh */
1578 		} else if (adapter->init_done_rc) {
1579 			netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
1580 				    adapter->init_done_rc);
1581 
1582 partial_reset:
1583 			/* adapter login failed, so free any CRQs or sub-CRQs
1584 			 * and register again before attempting to login again.
1585 			 * If we don't do this then the VIOS may think that
1586 			 * we are already logged in and reject any subsequent
1587 			 * attempts
1588 			 */
1589 			netdev_warn(netdev,
1590 				    "Freeing and re-registering CRQs before attempting to login again\n");
1591 			retry = true;
1592 			adapter->init_done_rc = 0;
1593 			release_sub_crqs(adapter, true);
1594 			/* Much of this is similar logic as ibmvnic_probe(),
1595 			 * we are essentially re-initializing communication
1596 			 * with the server. We really should not run any
1597 			 * resets/failovers here because this is already a form
1598 			 * of reset and we do not want parallel resets occurring
1599 			 */
1600 			do {
1601 				reinit_init_done(adapter);
1602 				/* Clear any failovers we got in the previous
1603 				 * pass since we are re-initializing the CRQ
1604 				 */
1605 				adapter->failover_pending = false;
1606 				release_crq_queue(adapter);
1607 				/* If we don't sleep here then we risk an
1608 				 * unnecessary failover event from the VIOS.
1609 				 * This is a known VIOS issue caused by a vnic
1610 				 * device freeing and registering a CRQ too
1611 				 * quickly.
1612 				 */
1613 				msleep(1500);
1614 				/* Avoid any resets, since we are currently
1615 				 * resetting.
1616 				 */
1617 				spin_lock_irqsave(&adapter->rwi_lock, flags);
1618 				flush_reset_queue(adapter);
1619 				spin_unlock_irqrestore(&adapter->rwi_lock,
1620 						       flags);
1621 
1622 				rc = init_crq_queue(adapter);
1623 				if (rc) {
1624 					netdev_err(netdev, "login recovery: init CRQ failed %d\n",
1625 						   rc);
1626 					return -EIO;
1627 				}
1628 
1629 				rc = ibmvnic_reset_init(adapter, false);
1630 				if (rc)
1631 					netdev_err(netdev, "login recovery: Reset init failed %d\n",
1632 						   rc);
1633 				/* IBMVNIC_CRQ_INIT will return EAGAIN if it
1634 				 * fails, since ibmvnic_reset_init will free
1635 				 * irq's in failure, we won't be able to receive
1636 				 * new CRQs so we need to keep trying. probe()
1637 				 * handles this similarly.
1638 				 */
1639 			} while (rc == -EAGAIN && retry_count++ < retries);
1640 		}
1641 	} while (retry);
1642 
1643 	__ibmvnic_set_mac(netdev, adapter->mac_addr);
1644 
1645 	netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1646 	return 0;
1647 }
1648 
1649 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1650 {
1651 	if (!adapter->login_buf)
1652 		return;
1653 
1654 	dma_unmap_single(&adapter->vdev->dev, adapter->login_buf_token,
1655 			 adapter->login_buf_sz, DMA_TO_DEVICE);
1656 	kfree(adapter->login_buf);
1657 	adapter->login_buf = NULL;
1658 }
1659 
1660 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1661 {
1662 	if (!adapter->login_rsp_buf)
1663 		return;
1664 
1665 	dma_unmap_single(&adapter->vdev->dev, adapter->login_rsp_buf_token,
1666 			 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
1667 	kfree(adapter->login_rsp_buf);
1668 	adapter->login_rsp_buf = NULL;
1669 }
1670 
1671 static void release_resources(struct ibmvnic_adapter *adapter)
1672 {
1673 	release_vpd_data(adapter);
1674 
1675 	release_napi(adapter);
1676 	release_login_buffer(adapter);
1677 	release_login_rsp_buffer(adapter);
1678 }
1679 
1680 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1681 {
1682 	struct net_device *netdev = adapter->netdev;
1683 	unsigned long timeout = msecs_to_jiffies(20000);
1684 	union ibmvnic_crq crq;
1685 	bool resend;
1686 	int rc;
1687 
1688 	netdev_dbg(netdev, "setting link state %d\n", link_state);
1689 
1690 	memset(&crq, 0, sizeof(crq));
1691 	crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1692 	crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1693 	crq.logical_link_state.link_state = link_state;
1694 
1695 	do {
1696 		resend = false;
1697 
1698 		reinit_completion(&adapter->init_done);
1699 		rc = ibmvnic_send_crq(adapter, &crq);
1700 		if (rc) {
1701 			netdev_err(netdev, "Failed to set link state\n");
1702 			return rc;
1703 		}
1704 
1705 		if (!wait_for_completion_timeout(&adapter->init_done,
1706 						 timeout)) {
1707 			netdev_err(netdev, "timeout setting link state\n");
1708 			return -ETIMEDOUT;
1709 		}
1710 
1711 		if (adapter->init_done_rc == PARTIALSUCCESS) {
1712 			/* Partuial success, delay and re-send */
1713 			mdelay(1000);
1714 			resend = true;
1715 		} else if (adapter->init_done_rc) {
1716 			netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1717 				    adapter->init_done_rc);
1718 			return adapter->init_done_rc;
1719 		}
1720 	} while (resend);
1721 
1722 	return 0;
1723 }
1724 
1725 static int set_real_num_queues(struct net_device *netdev)
1726 {
1727 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1728 	int rc;
1729 
1730 	netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1731 		   adapter->req_tx_queues, adapter->req_rx_queues);
1732 
1733 	rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1734 	if (rc) {
1735 		netdev_err(netdev, "failed to set the number of tx queues\n");
1736 		return rc;
1737 	}
1738 
1739 	rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1740 	if (rc)
1741 		netdev_err(netdev, "failed to set the number of rx queues\n");
1742 
1743 	return rc;
1744 }
1745 
1746 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1747 {
1748 	struct device *dev = &adapter->vdev->dev;
1749 	union ibmvnic_crq crq;
1750 	int len = 0;
1751 	int rc;
1752 
1753 	if (adapter->vpd->buff)
1754 		len = adapter->vpd->len;
1755 
1756 	mutex_lock(&adapter->fw_lock);
1757 	adapter->fw_done_rc = 0;
1758 	reinit_completion(&adapter->fw_done);
1759 
1760 	crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1761 	crq.get_vpd_size.cmd = GET_VPD_SIZE;
1762 	rc = ibmvnic_send_crq(adapter, &crq);
1763 	if (rc) {
1764 		mutex_unlock(&adapter->fw_lock);
1765 		return rc;
1766 	}
1767 
1768 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1769 	if (rc) {
1770 		dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1771 		mutex_unlock(&adapter->fw_lock);
1772 		return rc;
1773 	}
1774 	mutex_unlock(&adapter->fw_lock);
1775 
1776 	if (!adapter->vpd->len)
1777 		return -ENODATA;
1778 
1779 	if (!adapter->vpd->buff)
1780 		adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1781 	else if (adapter->vpd->len != len)
1782 		adapter->vpd->buff =
1783 			krealloc(adapter->vpd->buff,
1784 				 adapter->vpd->len, GFP_KERNEL);
1785 
1786 	if (!adapter->vpd->buff) {
1787 		dev_err(dev, "Could allocate VPD buffer\n");
1788 		return -ENOMEM;
1789 	}
1790 
1791 	adapter->vpd->dma_addr =
1792 		dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1793 			       DMA_FROM_DEVICE);
1794 	if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1795 		dev_err(dev, "Could not map VPD buffer\n");
1796 		kfree(adapter->vpd->buff);
1797 		adapter->vpd->buff = NULL;
1798 		return -ENOMEM;
1799 	}
1800 
1801 	mutex_lock(&adapter->fw_lock);
1802 	adapter->fw_done_rc = 0;
1803 	reinit_completion(&adapter->fw_done);
1804 
1805 	crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1806 	crq.get_vpd.cmd = GET_VPD;
1807 	crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1808 	crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1809 	rc = ibmvnic_send_crq(adapter, &crq);
1810 	if (rc) {
1811 		kfree(adapter->vpd->buff);
1812 		adapter->vpd->buff = NULL;
1813 		mutex_unlock(&adapter->fw_lock);
1814 		return rc;
1815 	}
1816 
1817 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1818 	if (rc) {
1819 		dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1820 		kfree(adapter->vpd->buff);
1821 		adapter->vpd->buff = NULL;
1822 		mutex_unlock(&adapter->fw_lock);
1823 		return rc;
1824 	}
1825 
1826 	mutex_unlock(&adapter->fw_lock);
1827 	return 0;
1828 }
1829 
1830 static int init_resources(struct ibmvnic_adapter *adapter)
1831 {
1832 	struct net_device *netdev = adapter->netdev;
1833 	int rc;
1834 
1835 	rc = set_real_num_queues(netdev);
1836 	if (rc)
1837 		return rc;
1838 
1839 	adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1840 	if (!adapter->vpd)
1841 		return -ENOMEM;
1842 
1843 	/* Vital Product Data (VPD) */
1844 	rc = ibmvnic_get_vpd(adapter);
1845 	if (rc) {
1846 		netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1847 		return rc;
1848 	}
1849 
1850 	rc = init_napi(adapter);
1851 	if (rc)
1852 		return rc;
1853 
1854 	send_query_map(adapter);
1855 
1856 	rc = init_rx_pools(netdev);
1857 	if (rc)
1858 		return rc;
1859 
1860 	rc = init_tx_pools(netdev);
1861 	return rc;
1862 }
1863 
1864 static int __ibmvnic_open(struct net_device *netdev)
1865 {
1866 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1867 	enum vnic_state prev_state = adapter->state;
1868 	int i, rc;
1869 
1870 	adapter->state = VNIC_OPENING;
1871 	replenish_pools(adapter);
1872 	ibmvnic_napi_enable(adapter);
1873 
1874 	/* We're ready to receive frames, enable the sub-crq interrupts and
1875 	 * set the logical link state to up
1876 	 */
1877 	for (i = 0; i < adapter->req_rx_queues; i++) {
1878 		netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1879 		if (prev_state == VNIC_CLOSED)
1880 			enable_irq(adapter->rx_scrq[i]->irq);
1881 		enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1882 	}
1883 
1884 	for (i = 0; i < adapter->req_tx_queues; i++) {
1885 		netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1886 		if (prev_state == VNIC_CLOSED)
1887 			enable_irq(adapter->tx_scrq[i]->irq);
1888 		enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1889 		/* netdev_tx_reset_queue will reset dql stats. During NON_FATAL
1890 		 * resets, don't reset the stats because there could be batched
1891 		 * skb's waiting to be sent. If we reset dql stats, we risk
1892 		 * num_completed being greater than num_queued. This will cause
1893 		 * a BUG_ON in dql_completed().
1894 		 */
1895 		if (adapter->reset_reason != VNIC_RESET_NON_FATAL)
1896 			netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1897 	}
1898 
1899 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1900 	if (rc) {
1901 		ibmvnic_napi_disable(adapter);
1902 		ibmvnic_disable_irqs(adapter);
1903 		return rc;
1904 	}
1905 
1906 	adapter->tx_queues_active = true;
1907 
1908 	/* Since queues were stopped until now, there shouldn't be any
1909 	 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we
1910 	 * don't need the synchronize_rcu()? Leaving it for consistency
1911 	 * with setting ->tx_queues_active = false.
1912 	 */
1913 	synchronize_rcu();
1914 
1915 	netif_tx_start_all_queues(netdev);
1916 
1917 	if (prev_state == VNIC_CLOSED) {
1918 		for (i = 0; i < adapter->req_rx_queues; i++)
1919 			napi_schedule(&adapter->napi[i]);
1920 	}
1921 
1922 	adapter->state = VNIC_OPEN;
1923 	return rc;
1924 }
1925 
1926 static int ibmvnic_open(struct net_device *netdev)
1927 {
1928 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1929 	int rc;
1930 
1931 	ASSERT_RTNL();
1932 
1933 	/* If device failover is pending or we are about to reset, just set
1934 	 * device state and return. Device operation will be handled by reset
1935 	 * routine.
1936 	 *
1937 	 * It should be safe to overwrite the adapter->state here. Since
1938 	 * we hold the rtnl, either the reset has not actually started or
1939 	 * the rtnl got dropped during the set_link_state() in do_reset().
1940 	 * In the former case, no one else is changing the state (again we
1941 	 * have the rtnl) and in the latter case, do_reset() will detect and
1942 	 * honor our setting below.
1943 	 */
1944 	if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1945 		netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1946 			   adapter_state_to_string(adapter->state),
1947 			   adapter->failover_pending);
1948 		adapter->state = VNIC_OPEN;
1949 		rc = 0;
1950 		goto out;
1951 	}
1952 
1953 	if (adapter->state != VNIC_CLOSED) {
1954 		rc = ibmvnic_login(netdev);
1955 		if (rc)
1956 			goto out;
1957 
1958 		rc = init_resources(adapter);
1959 		if (rc) {
1960 			netdev_err(netdev, "failed to initialize resources\n");
1961 			goto out;
1962 		}
1963 	}
1964 
1965 	rc = __ibmvnic_open(netdev);
1966 
1967 out:
1968 	/* If open failed and there is a pending failover or in-progress reset,
1969 	 * set device state and return. Device operation will be handled by
1970 	 * reset routine. See also comments above regarding rtnl.
1971 	 */
1972 	if (rc &&
1973 	    (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1974 		adapter->state = VNIC_OPEN;
1975 		rc = 0;
1976 	}
1977 
1978 	if (rc) {
1979 		release_resources(adapter);
1980 		release_rx_pools(adapter);
1981 		release_tx_pools(adapter);
1982 	}
1983 
1984 	return rc;
1985 }
1986 
1987 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1988 {
1989 	struct ibmvnic_rx_pool *rx_pool;
1990 	struct ibmvnic_rx_buff *rx_buff;
1991 	u64 rx_entries;
1992 	int rx_scrqs;
1993 	int i, j;
1994 
1995 	if (!adapter->rx_pool)
1996 		return;
1997 
1998 	rx_scrqs = adapter->num_active_rx_pools;
1999 	rx_entries = adapter->req_rx_add_entries_per_subcrq;
2000 
2001 	/* Free any remaining skbs in the rx buffer pools */
2002 	for (i = 0; i < rx_scrqs; i++) {
2003 		rx_pool = &adapter->rx_pool[i];
2004 		if (!rx_pool || !rx_pool->rx_buff)
2005 			continue;
2006 
2007 		netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
2008 		for (j = 0; j < rx_entries; j++) {
2009 			rx_buff = &rx_pool->rx_buff[j];
2010 			if (rx_buff && rx_buff->skb) {
2011 				dev_kfree_skb_any(rx_buff->skb);
2012 				rx_buff->skb = NULL;
2013 			}
2014 		}
2015 	}
2016 }
2017 
2018 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
2019 			      struct ibmvnic_tx_pool *tx_pool)
2020 {
2021 	struct ibmvnic_tx_buff *tx_buff;
2022 	u64 tx_entries;
2023 	int i;
2024 
2025 	if (!tx_pool || !tx_pool->tx_buff)
2026 		return;
2027 
2028 	tx_entries = tx_pool->num_buffers;
2029 
2030 	for (i = 0; i < tx_entries; i++) {
2031 		tx_buff = &tx_pool->tx_buff[i];
2032 		if (tx_buff && tx_buff->skb) {
2033 			dev_kfree_skb_any(tx_buff->skb);
2034 			tx_buff->skb = NULL;
2035 		}
2036 	}
2037 }
2038 
2039 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
2040 {
2041 	int tx_scrqs;
2042 	int i;
2043 
2044 	if (!adapter->tx_pool || !adapter->tso_pool)
2045 		return;
2046 
2047 	tx_scrqs = adapter->num_active_tx_pools;
2048 
2049 	/* Free any remaining skbs in the tx buffer pools */
2050 	for (i = 0; i < tx_scrqs; i++) {
2051 		netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
2052 		clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
2053 		clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
2054 	}
2055 }
2056 
2057 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
2058 {
2059 	struct net_device *netdev = adapter->netdev;
2060 	int i;
2061 
2062 	if (adapter->tx_scrq) {
2063 		for (i = 0; i < adapter->req_tx_queues; i++)
2064 			if (adapter->tx_scrq[i]->irq) {
2065 				netdev_dbg(netdev,
2066 					   "Disabling tx_scrq[%d] irq\n", i);
2067 				disable_scrq_irq(adapter, adapter->tx_scrq[i]);
2068 				disable_irq(adapter->tx_scrq[i]->irq);
2069 			}
2070 	}
2071 
2072 	if (adapter->rx_scrq) {
2073 		for (i = 0; i < adapter->req_rx_queues; i++) {
2074 			if (adapter->rx_scrq[i]->irq) {
2075 				netdev_dbg(netdev,
2076 					   "Disabling rx_scrq[%d] irq\n", i);
2077 				disable_scrq_irq(adapter, adapter->rx_scrq[i]);
2078 				disable_irq(adapter->rx_scrq[i]->irq);
2079 			}
2080 		}
2081 	}
2082 }
2083 
2084 static void ibmvnic_cleanup(struct net_device *netdev)
2085 {
2086 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2087 
2088 	/* ensure that transmissions are stopped if called by do_reset */
2089 
2090 	adapter->tx_queues_active = false;
2091 
2092 	/* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active
2093 	 * update so they don't restart a queue after we stop it below.
2094 	 */
2095 	synchronize_rcu();
2096 
2097 	if (test_bit(0, &adapter->resetting))
2098 		netif_tx_disable(netdev);
2099 	else
2100 		netif_tx_stop_all_queues(netdev);
2101 
2102 	ibmvnic_napi_disable(adapter);
2103 	ibmvnic_disable_irqs(adapter);
2104 }
2105 
2106 static int __ibmvnic_close(struct net_device *netdev)
2107 {
2108 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2109 	int rc = 0;
2110 
2111 	adapter->state = VNIC_CLOSING;
2112 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2113 	adapter->state = VNIC_CLOSED;
2114 	return rc;
2115 }
2116 
2117 static int ibmvnic_close(struct net_device *netdev)
2118 {
2119 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2120 	int rc;
2121 
2122 	netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
2123 		   adapter_state_to_string(adapter->state),
2124 		   adapter->failover_pending,
2125 		   adapter->force_reset_recovery);
2126 
2127 	/* If device failover is pending, just set device state and return.
2128 	 * Device operation will be handled by reset routine.
2129 	 */
2130 	if (adapter->failover_pending) {
2131 		adapter->state = VNIC_CLOSED;
2132 		return 0;
2133 	}
2134 
2135 	rc = __ibmvnic_close(netdev);
2136 	ibmvnic_cleanup(netdev);
2137 	clean_rx_pools(adapter);
2138 	clean_tx_pools(adapter);
2139 
2140 	return rc;
2141 }
2142 
2143 /**
2144  * get_hdr_lens - fills list of L2/L3/L4 hdr lens
2145  * @hdr_field: bitfield determining needed headers
2146  * @skb: socket buffer
2147  * @hdr_len: array of header lengths to be filled
2148  *
2149  * Reads hdr_field to determine which headers are needed by firmware.
2150  * Builds a buffer containing these headers.  Saves individual header
2151  * lengths and total buffer length to be used to build descriptors.
2152  *
2153  * Return: total len of all headers
2154  */
2155 static int get_hdr_lens(u8 hdr_field, struct sk_buff *skb,
2156 			int *hdr_len)
2157 {
2158 	int len = 0;
2159 
2160 
2161 	if ((hdr_field >> 6) & 1) {
2162 		hdr_len[0] = skb_mac_header_len(skb);
2163 		len += hdr_len[0];
2164 	}
2165 
2166 	if ((hdr_field >> 5) & 1) {
2167 		hdr_len[1] = skb_network_header_len(skb);
2168 		len += hdr_len[1];
2169 	}
2170 
2171 	if (!((hdr_field >> 4) & 1))
2172 		return len;
2173 
2174 	if (skb->protocol == htons(ETH_P_IP)) {
2175 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2176 			hdr_len[2] = tcp_hdrlen(skb);
2177 		else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
2178 			hdr_len[2] = sizeof(struct udphdr);
2179 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2180 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2181 			hdr_len[2] = tcp_hdrlen(skb);
2182 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
2183 			hdr_len[2] = sizeof(struct udphdr);
2184 	}
2185 
2186 	return len + hdr_len[2];
2187 }
2188 
2189 /**
2190  * create_hdr_descs - create header and header extension descriptors
2191  * @hdr_field: bitfield determining needed headers
2192  * @hdr_data: buffer containing header data
2193  * @len: length of data buffer
2194  * @hdr_len: array of individual header lengths
2195  * @scrq_arr: descriptor array
2196  *
2197  * Creates header and, if needed, header extension descriptors and
2198  * places them in a descriptor array, scrq_arr
2199  *
2200  * Return: Number of header descs
2201  */
2202 
2203 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
2204 			    union sub_crq *scrq_arr)
2205 {
2206 	union sub_crq *hdr_desc;
2207 	int tmp_len = len;
2208 	int num_descs = 0;
2209 	u8 *data, *cur;
2210 	int tmp;
2211 
2212 	while (tmp_len > 0) {
2213 		cur = hdr_data + len - tmp_len;
2214 
2215 		hdr_desc = &scrq_arr[num_descs];
2216 		if (num_descs) {
2217 			data = hdr_desc->hdr_ext.data;
2218 			tmp = tmp_len > 29 ? 29 : tmp_len;
2219 			hdr_desc->hdr_ext.first = IBMVNIC_CRQ_CMD;
2220 			hdr_desc->hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
2221 			hdr_desc->hdr_ext.len = tmp;
2222 		} else {
2223 			data = hdr_desc->hdr.data;
2224 			tmp = tmp_len > 24 ? 24 : tmp_len;
2225 			hdr_desc->hdr.first = IBMVNIC_CRQ_CMD;
2226 			hdr_desc->hdr.type = IBMVNIC_HDR_DESC;
2227 			hdr_desc->hdr.len = tmp;
2228 			hdr_desc->hdr.l2_len = (u8)hdr_len[0];
2229 			hdr_desc->hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
2230 			hdr_desc->hdr.l4_len = (u8)hdr_len[2];
2231 			hdr_desc->hdr.flag = hdr_field << 1;
2232 		}
2233 		memcpy(data, cur, tmp);
2234 		tmp_len -= tmp;
2235 		num_descs++;
2236 	}
2237 
2238 	return num_descs;
2239 }
2240 
2241 /**
2242  * build_hdr_descs_arr - build a header descriptor array
2243  * @skb: tx socket buffer
2244  * @indir_arr: indirect array
2245  * @num_entries: number of descriptors to be sent
2246  * @hdr_field: bit field determining which headers will be sent
2247  *
2248  * This function will build a TX descriptor array with applicable
2249  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
2250  */
2251 
2252 static void build_hdr_descs_arr(struct sk_buff *skb,
2253 				union sub_crq *indir_arr,
2254 				int *num_entries, u8 hdr_field)
2255 {
2256 	int hdr_len[3] = {0, 0, 0};
2257 	int tot_len;
2258 
2259 	tot_len = get_hdr_lens(hdr_field, skb, hdr_len);
2260 	*num_entries += create_hdr_descs(hdr_field, skb_mac_header(skb),
2261 					 tot_len, hdr_len, indir_arr + 1);
2262 }
2263 
2264 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
2265 				    struct net_device *netdev)
2266 {
2267 	/* For some backing devices, mishandling of small packets
2268 	 * can result in a loss of connection or TX stall. Device
2269 	 * architects recommend that no packet should be smaller
2270 	 * than the minimum MTU value provided to the driver, so
2271 	 * pad any packets to that length
2272 	 */
2273 	if (skb->len < netdev->min_mtu)
2274 		return skb_put_padto(skb, netdev->min_mtu);
2275 
2276 	return 0;
2277 }
2278 
2279 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
2280 					 struct ibmvnic_sub_crq_queue *tx_scrq)
2281 {
2282 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2283 	struct ibmvnic_tx_buff *tx_buff;
2284 	struct ibmvnic_tx_pool *tx_pool;
2285 	union sub_crq tx_scrq_entry;
2286 	int queue_num;
2287 	int entries;
2288 	int index;
2289 	int i;
2290 
2291 	ind_bufp = &tx_scrq->ind_buf;
2292 	entries = (u64)ind_bufp->index;
2293 	queue_num = tx_scrq->pool_index;
2294 
2295 	for (i = entries - 1; i >= 0; --i) {
2296 		tx_scrq_entry = ind_bufp->indir_arr[i];
2297 		if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
2298 			continue;
2299 		index = be32_to_cpu(tx_scrq_entry.v1.correlator);
2300 		if (index & IBMVNIC_TSO_POOL_MASK) {
2301 			tx_pool = &adapter->tso_pool[queue_num];
2302 			index &= ~IBMVNIC_TSO_POOL_MASK;
2303 		} else {
2304 			tx_pool = &adapter->tx_pool[queue_num];
2305 		}
2306 		tx_pool->free_map[tx_pool->consumer_index] = index;
2307 		tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2308 					  tx_pool->num_buffers - 1 :
2309 					  tx_pool->consumer_index - 1;
2310 		tx_buff = &tx_pool->tx_buff[index];
2311 		adapter->netdev->stats.tx_packets--;
2312 		adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
2313 		adapter->tx_stats_buffers[queue_num].packets--;
2314 		adapter->tx_stats_buffers[queue_num].bytes -=
2315 						tx_buff->skb->len;
2316 		dev_kfree_skb_any(tx_buff->skb);
2317 		tx_buff->skb = NULL;
2318 		adapter->netdev->stats.tx_dropped++;
2319 	}
2320 
2321 	ind_bufp->index = 0;
2322 
2323 	if (atomic_sub_return(entries, &tx_scrq->used) <=
2324 	    (adapter->req_tx_entries_per_subcrq / 2) &&
2325 	    __netif_subqueue_stopped(adapter->netdev, queue_num)) {
2326 		rcu_read_lock();
2327 
2328 		if (adapter->tx_queues_active) {
2329 			netif_wake_subqueue(adapter->netdev, queue_num);
2330 			netdev_dbg(adapter->netdev, "Started queue %d\n",
2331 				   queue_num);
2332 		}
2333 
2334 		rcu_read_unlock();
2335 	}
2336 }
2337 
2338 static int send_subcrq_direct(struct ibmvnic_adapter *adapter,
2339 			      u64 remote_handle, u64 *entry)
2340 {
2341 	unsigned int ua = adapter->vdev->unit_address;
2342 	struct device *dev = &adapter->vdev->dev;
2343 	int rc;
2344 
2345 	/* Make sure the hypervisor sees the complete request */
2346 	dma_wmb();
2347 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
2348 				cpu_to_be64(remote_handle),
2349 				cpu_to_be64(entry[0]), cpu_to_be64(entry[1]),
2350 				cpu_to_be64(entry[2]), cpu_to_be64(entry[3]));
2351 
2352 	if (rc)
2353 		print_subcrq_error(dev, rc, __func__);
2354 
2355 	return rc;
2356 }
2357 
2358 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
2359 				 struct ibmvnic_sub_crq_queue *tx_scrq,
2360 				 bool indirect)
2361 {
2362 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2363 	u64 dma_addr;
2364 	u64 entries;
2365 	u64 handle;
2366 	int rc;
2367 
2368 	ind_bufp = &tx_scrq->ind_buf;
2369 	dma_addr = (u64)ind_bufp->indir_dma;
2370 	entries = (u64)ind_bufp->index;
2371 	handle = tx_scrq->handle;
2372 
2373 	if (!entries)
2374 		return 0;
2375 
2376 	if (indirect)
2377 		rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
2378 	else
2379 		rc = send_subcrq_direct(adapter, handle,
2380 					(u64 *)ind_bufp->indir_arr);
2381 
2382 	if (rc)
2383 		ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
2384 	else
2385 		ind_bufp->index = 0;
2386 	return rc;
2387 }
2388 
2389 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
2390 {
2391 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2392 	int queue_num = skb_get_queue_mapping(skb);
2393 	u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
2394 	struct device *dev = &adapter->vdev->dev;
2395 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2396 	struct ibmvnic_tx_buff *tx_buff = NULL;
2397 	struct ibmvnic_sub_crq_queue *tx_scrq;
2398 	struct ibmvnic_long_term_buff *ltb;
2399 	struct ibmvnic_tx_pool *tx_pool;
2400 	unsigned int tx_send_failed = 0;
2401 	netdev_tx_t ret = NETDEV_TX_OK;
2402 	unsigned int tx_map_failed = 0;
2403 	union sub_crq indir_arr[16];
2404 	unsigned int tx_dropped = 0;
2405 	unsigned int tx_packets = 0;
2406 	unsigned int tx_bytes = 0;
2407 	dma_addr_t data_dma_addr;
2408 	struct netdev_queue *txq;
2409 	unsigned long lpar_rc;
2410 	union sub_crq tx_crq;
2411 	unsigned int offset;
2412 	bool use_scrq_send_direct = false;
2413 	int num_entries = 1;
2414 	unsigned char *dst;
2415 	int bufidx = 0;
2416 	u8 proto = 0;
2417 
2418 	/* If a reset is in progress, drop the packet since
2419 	 * the scrqs may get torn down. Otherwise use the
2420 	 * rcu to ensure reset waits for us to complete.
2421 	 */
2422 	rcu_read_lock();
2423 	if (!adapter->tx_queues_active) {
2424 		dev_kfree_skb_any(skb);
2425 
2426 		tx_send_failed++;
2427 		tx_dropped++;
2428 		ret = NETDEV_TX_OK;
2429 		goto out;
2430 	}
2431 
2432 	tx_scrq = adapter->tx_scrq[queue_num];
2433 	txq = netdev_get_tx_queue(netdev, queue_num);
2434 	ind_bufp = &tx_scrq->ind_buf;
2435 
2436 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
2437 		tx_dropped++;
2438 		tx_send_failed++;
2439 		ret = NETDEV_TX_OK;
2440 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2441 		if (lpar_rc != H_SUCCESS)
2442 			goto tx_err;
2443 		goto out;
2444 	}
2445 
2446 	if (skb_is_gso(skb))
2447 		tx_pool = &adapter->tso_pool[queue_num];
2448 	else
2449 		tx_pool = &adapter->tx_pool[queue_num];
2450 
2451 	bufidx = tx_pool->free_map[tx_pool->consumer_index];
2452 
2453 	if (bufidx == IBMVNIC_INVALID_MAP) {
2454 		dev_kfree_skb_any(skb);
2455 		tx_send_failed++;
2456 		tx_dropped++;
2457 		ret = NETDEV_TX_OK;
2458 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2459 		if (lpar_rc != H_SUCCESS)
2460 			goto tx_err;
2461 		goto out;
2462 	}
2463 
2464 	tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
2465 
2466 	map_txpool_buf_to_ltb(tx_pool, bufidx, &ltb, &offset);
2467 
2468 	dst = ltb->buff + offset;
2469 	memset(dst, 0, tx_pool->buf_size);
2470 	data_dma_addr = ltb->addr + offset;
2471 
2472 	/* if we are going to send_subcrq_direct this then we need to
2473 	 * update the checksum before copying the data into ltb. Essentially
2474 	 * these packets force disable CSO so that we can guarantee that
2475 	 * FW does not need header info and we can send direct. Also, vnic
2476 	 * server must be able to xmit standard packets without header data
2477 	 */
2478 	if (*hdrs == 0 && !skb_is_gso(skb) &&
2479 	    !ind_bufp->index && !netdev_xmit_more()) {
2480 		use_scrq_send_direct = true;
2481 		if (skb->ip_summed == CHECKSUM_PARTIAL &&
2482 		    skb_checksum_help(skb))
2483 			use_scrq_send_direct = false;
2484 	}
2485 
2486 	if (skb_shinfo(skb)->nr_frags) {
2487 		int cur, i;
2488 
2489 		/* Copy the head */
2490 		skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
2491 		cur = skb_headlen(skb);
2492 
2493 		/* Copy the frags */
2494 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2495 			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2496 
2497 			memcpy(dst + cur, skb_frag_address(frag),
2498 			       skb_frag_size(frag));
2499 			cur += skb_frag_size(frag);
2500 		}
2501 	} else {
2502 		skb_copy_from_linear_data(skb, dst, skb->len);
2503 	}
2504 
2505 	tx_pool->consumer_index =
2506 	    (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
2507 
2508 	tx_buff = &tx_pool->tx_buff[bufidx];
2509 
2510 	/* Sanity checks on our free map to make sure it points to an index
2511 	 * that is not being occupied by another skb. If skb memory is
2512 	 * not freed then we see congestion control kick in and halt tx.
2513 	 */
2514 	if (unlikely(tx_buff->skb)) {
2515 		dev_warn_ratelimited(dev, "TX free map points to untracked skb (%s %d idx=%d)\n",
2516 				     skb_is_gso(skb) ? "tso_pool" : "tx_pool",
2517 				     queue_num, bufidx);
2518 		dev_kfree_skb_any(tx_buff->skb);
2519 	}
2520 
2521 	tx_buff->skb = skb;
2522 	tx_buff->index = bufidx;
2523 	tx_buff->pool_index = queue_num;
2524 
2525 	memset(&tx_crq, 0, sizeof(tx_crq));
2526 	tx_crq.v1.first = IBMVNIC_CRQ_CMD;
2527 	tx_crq.v1.type = IBMVNIC_TX_DESC;
2528 	tx_crq.v1.n_crq_elem = 1;
2529 	tx_crq.v1.n_sge = 1;
2530 	tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
2531 
2532 	if (skb_is_gso(skb))
2533 		tx_crq.v1.correlator =
2534 			cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK);
2535 	else
2536 		tx_crq.v1.correlator = cpu_to_be32(bufidx);
2537 	tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id);
2538 	tx_crq.v1.sge_len = cpu_to_be32(skb->len);
2539 	tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
2540 
2541 	if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
2542 		tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
2543 		tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2544 	}
2545 
2546 	if (skb->protocol == htons(ETH_P_IP)) {
2547 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2548 		proto = ip_hdr(skb)->protocol;
2549 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2550 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2551 		proto = ipv6_hdr(skb)->nexthdr;
2552 	}
2553 
2554 	if (proto == IPPROTO_TCP)
2555 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2556 	else if (proto == IPPROTO_UDP)
2557 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2558 
2559 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2560 		tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2561 		hdrs += 2;
2562 	}
2563 	if (skb_is_gso(skb)) {
2564 		tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2565 		tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2566 		hdrs += 2;
2567 	} else if (use_scrq_send_direct) {
2568 		/* See above comment, CSO disabled with direct xmit */
2569 		tx_crq.v1.flags1 &= ~(IBMVNIC_TX_CHKSUM_OFFLOAD);
2570 		ind_bufp->index = 1;
2571 		tx_buff->num_entries = 1;
2572 		netdev_tx_sent_queue(txq, skb->len);
2573 		ind_bufp->indir_arr[0] = tx_crq;
2574 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, false);
2575 		if (lpar_rc != H_SUCCESS)
2576 			goto tx_err;
2577 
2578 		goto early_exit;
2579 	}
2580 
2581 	if ((*hdrs >> 7) & 1)
2582 		build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2583 
2584 	tx_crq.v1.n_crq_elem = num_entries;
2585 	tx_buff->num_entries = num_entries;
2586 	/* flush buffer if current entry can not fit */
2587 	if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2588 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2589 		if (lpar_rc != H_SUCCESS)
2590 			goto tx_flush_err;
2591 	}
2592 
2593 	indir_arr[0] = tx_crq;
2594 	memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2595 	       num_entries * sizeof(struct ibmvnic_generic_scrq));
2596 
2597 	ind_bufp->index += num_entries;
2598 	if (__netdev_tx_sent_queue(txq, skb->len,
2599 				   netdev_xmit_more() &&
2600 				   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2601 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2602 		if (lpar_rc != H_SUCCESS)
2603 			goto tx_err;
2604 	}
2605 
2606 early_exit:
2607 	if (atomic_add_return(num_entries, &tx_scrq->used)
2608 					>= adapter->req_tx_entries_per_subcrq) {
2609 		netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2610 		netif_stop_subqueue(netdev, queue_num);
2611 	}
2612 
2613 	tx_packets++;
2614 	tx_bytes += skb->len;
2615 	txq_trans_cond_update(txq);
2616 	ret = NETDEV_TX_OK;
2617 	goto out;
2618 
2619 tx_flush_err:
2620 	dev_kfree_skb_any(skb);
2621 	tx_buff->skb = NULL;
2622 	tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2623 				  tx_pool->num_buffers - 1 :
2624 				  tx_pool->consumer_index - 1;
2625 	tx_dropped++;
2626 tx_err:
2627 	if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2628 		dev_err_ratelimited(dev, "tx: send failed\n");
2629 
2630 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2631 		/* Disable TX and report carrier off if queue is closed
2632 		 * or pending failover.
2633 		 * Firmware guarantees that a signal will be sent to the
2634 		 * driver, triggering a reset or some other action.
2635 		 */
2636 		netif_tx_stop_all_queues(netdev);
2637 		netif_carrier_off(netdev);
2638 	}
2639 out:
2640 	rcu_read_unlock();
2641 	netdev->stats.tx_dropped += tx_dropped;
2642 	netdev->stats.tx_bytes += tx_bytes;
2643 	netdev->stats.tx_packets += tx_packets;
2644 	adapter->tx_send_failed += tx_send_failed;
2645 	adapter->tx_map_failed += tx_map_failed;
2646 	adapter->tx_stats_buffers[queue_num].packets += tx_packets;
2647 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2648 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2649 
2650 	return ret;
2651 }
2652 
2653 static void ibmvnic_set_multi(struct net_device *netdev)
2654 {
2655 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2656 	struct netdev_hw_addr *ha;
2657 	union ibmvnic_crq crq;
2658 
2659 	memset(&crq, 0, sizeof(crq));
2660 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
2661 	crq.request_capability.cmd = REQUEST_CAPABILITY;
2662 
2663 	if (netdev->flags & IFF_PROMISC) {
2664 		if (!adapter->promisc_supported)
2665 			return;
2666 	} else {
2667 		if (netdev->flags & IFF_ALLMULTI) {
2668 			/* Accept all multicast */
2669 			memset(&crq, 0, sizeof(crq));
2670 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2671 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2672 			crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2673 			ibmvnic_send_crq(adapter, &crq);
2674 		} else if (netdev_mc_empty(netdev)) {
2675 			/* Reject all multicast */
2676 			memset(&crq, 0, sizeof(crq));
2677 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2678 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2679 			crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2680 			ibmvnic_send_crq(adapter, &crq);
2681 		} else {
2682 			/* Accept one or more multicast(s) */
2683 			netdev_for_each_mc_addr(ha, netdev) {
2684 				memset(&crq, 0, sizeof(crq));
2685 				crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2686 				crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2687 				crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2688 				ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2689 						ha->addr);
2690 				ibmvnic_send_crq(adapter, &crq);
2691 			}
2692 		}
2693 	}
2694 }
2695 
2696 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2697 {
2698 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2699 	union ibmvnic_crq crq;
2700 	int rc;
2701 
2702 	if (!is_valid_ether_addr(dev_addr)) {
2703 		rc = -EADDRNOTAVAIL;
2704 		goto err;
2705 	}
2706 
2707 	memset(&crq, 0, sizeof(crq));
2708 	crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2709 	crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2710 	ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2711 
2712 	mutex_lock(&adapter->fw_lock);
2713 	adapter->fw_done_rc = 0;
2714 	reinit_completion(&adapter->fw_done);
2715 
2716 	rc = ibmvnic_send_crq(adapter, &crq);
2717 	if (rc) {
2718 		rc = -EIO;
2719 		mutex_unlock(&adapter->fw_lock);
2720 		goto err;
2721 	}
2722 
2723 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2724 	/* netdev->dev_addr is changed in handle_change_mac_rsp function */
2725 	if (rc || adapter->fw_done_rc) {
2726 		rc = -EIO;
2727 		mutex_unlock(&adapter->fw_lock);
2728 		goto err;
2729 	}
2730 	mutex_unlock(&adapter->fw_lock);
2731 	return 0;
2732 err:
2733 	ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2734 	return rc;
2735 }
2736 
2737 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2738 {
2739 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2740 	struct sockaddr *addr = p;
2741 	int rc;
2742 
2743 	rc = 0;
2744 	if (!is_valid_ether_addr(addr->sa_data))
2745 		return -EADDRNOTAVAIL;
2746 
2747 	ether_addr_copy(adapter->mac_addr, addr->sa_data);
2748 	if (adapter->state != VNIC_PROBED)
2749 		rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2750 
2751 	return rc;
2752 }
2753 
2754 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2755 {
2756 	switch (reason) {
2757 	case VNIC_RESET_FAILOVER:
2758 		return "FAILOVER";
2759 	case VNIC_RESET_MOBILITY:
2760 		return "MOBILITY";
2761 	case VNIC_RESET_FATAL:
2762 		return "FATAL";
2763 	case VNIC_RESET_NON_FATAL:
2764 		return "NON_FATAL";
2765 	case VNIC_RESET_TIMEOUT:
2766 		return "TIMEOUT";
2767 	case VNIC_RESET_CHANGE_PARAM:
2768 		return "CHANGE_PARAM";
2769 	case VNIC_RESET_PASSIVE_INIT:
2770 		return "PASSIVE_INIT";
2771 	}
2772 	return "UNKNOWN";
2773 }
2774 
2775 /*
2776  * Initialize the init_done completion and return code values. We
2777  * can get a transport event just after registering the CRQ and the
2778  * tasklet will use this to communicate the transport event. To ensure
2779  * we don't miss the notification/error, initialize these _before_
2780  * regisering the CRQ.
2781  */
2782 static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
2783 {
2784 	reinit_completion(&adapter->init_done);
2785 	adapter->init_done_rc = 0;
2786 }
2787 
2788 /*
2789  * do_reset returns zero if we are able to keep processing reset events, or
2790  * non-zero if we hit a fatal error and must halt.
2791  */
2792 static int do_reset(struct ibmvnic_adapter *adapter,
2793 		    struct ibmvnic_rwi *rwi, u32 reset_state)
2794 {
2795 	struct net_device *netdev = adapter->netdev;
2796 	u64 old_num_rx_queues, old_num_tx_queues;
2797 	u64 old_num_rx_slots, old_num_tx_slots;
2798 	int rc;
2799 
2800 	netdev_dbg(adapter->netdev,
2801 		   "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2802 		   adapter_state_to_string(adapter->state),
2803 		   adapter->failover_pending,
2804 		   reset_reason_to_string(rwi->reset_reason),
2805 		   adapter_state_to_string(reset_state));
2806 
2807 	adapter->reset_reason = rwi->reset_reason;
2808 	/* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2809 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2810 		rtnl_lock();
2811 
2812 	/* Now that we have the rtnl lock, clear any pending failover.
2813 	 * This will ensure ibmvnic_open() has either completed or will
2814 	 * block until failover is complete.
2815 	 */
2816 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2817 		adapter->failover_pending = false;
2818 
2819 	/* read the state and check (again) after getting rtnl */
2820 	reset_state = adapter->state;
2821 
2822 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2823 		rc = -EBUSY;
2824 		goto out;
2825 	}
2826 
2827 	netif_carrier_off(netdev);
2828 
2829 	old_num_rx_queues = adapter->req_rx_queues;
2830 	old_num_tx_queues = adapter->req_tx_queues;
2831 	old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2832 	old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2833 
2834 	ibmvnic_cleanup(netdev);
2835 
2836 	if (reset_state == VNIC_OPEN &&
2837 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
2838 	    adapter->reset_reason != VNIC_RESET_FAILOVER) {
2839 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2840 			rc = __ibmvnic_close(netdev);
2841 			if (rc)
2842 				goto out;
2843 		} else {
2844 			adapter->state = VNIC_CLOSING;
2845 
2846 			/* Release the RTNL lock before link state change and
2847 			 * re-acquire after the link state change to allow
2848 			 * linkwatch_event to grab the RTNL lock and run during
2849 			 * a reset.
2850 			 */
2851 			rtnl_unlock();
2852 			rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2853 			rtnl_lock();
2854 			if (rc)
2855 				goto out;
2856 
2857 			if (adapter->state == VNIC_OPEN) {
2858 				/* When we dropped rtnl, ibmvnic_open() got
2859 				 * it and noticed that we are resetting and
2860 				 * set the adapter state to OPEN. Update our
2861 				 * new "target" state, and resume the reset
2862 				 * from VNIC_CLOSING state.
2863 				 */
2864 				netdev_dbg(netdev,
2865 					   "Open changed state from %s, updating.\n",
2866 					   adapter_state_to_string(reset_state));
2867 				reset_state = VNIC_OPEN;
2868 				adapter->state = VNIC_CLOSING;
2869 			}
2870 
2871 			if (adapter->state != VNIC_CLOSING) {
2872 				/* If someone else changed the adapter state
2873 				 * when we dropped the rtnl, fail the reset
2874 				 */
2875 				rc = -EAGAIN;
2876 				goto out;
2877 			}
2878 			adapter->state = VNIC_CLOSED;
2879 		}
2880 	}
2881 
2882 	if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2883 		release_resources(adapter);
2884 		release_sub_crqs(adapter, 1);
2885 		release_crq_queue(adapter);
2886 	}
2887 
2888 	if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2889 		/* remove the closed state so when we call open it appears
2890 		 * we are coming from the probed state.
2891 		 */
2892 		adapter->state = VNIC_PROBED;
2893 
2894 		reinit_init_done(adapter);
2895 
2896 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2897 			rc = init_crq_queue(adapter);
2898 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2899 			rc = ibmvnic_reenable_crq_queue(adapter);
2900 			release_sub_crqs(adapter, 1);
2901 		} else {
2902 			rc = ibmvnic_reset_crq(adapter);
2903 			if (rc == H_CLOSED || rc == H_SUCCESS) {
2904 				rc = vio_enable_interrupts(adapter->vdev);
2905 				if (rc)
2906 					netdev_err(adapter->netdev,
2907 						   "Reset failed to enable interrupts. rc=%d\n",
2908 						   rc);
2909 			}
2910 		}
2911 
2912 		if (rc) {
2913 			netdev_err(adapter->netdev,
2914 				   "Reset couldn't initialize crq. rc=%d\n", rc);
2915 			goto out;
2916 		}
2917 
2918 		rc = ibmvnic_reset_init(adapter, true);
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 			rc = 0;
2927 			goto out;
2928 		}
2929 
2930 		rc = ibmvnic_login(netdev);
2931 		if (rc)
2932 			goto out;
2933 
2934 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2935 			rc = init_resources(adapter);
2936 			if (rc)
2937 				goto out;
2938 		} else if (adapter->req_rx_queues != old_num_rx_queues ||
2939 		    adapter->req_tx_queues != old_num_tx_queues ||
2940 		    adapter->req_rx_add_entries_per_subcrq !=
2941 		    old_num_rx_slots ||
2942 		    adapter->req_tx_entries_per_subcrq !=
2943 		    old_num_tx_slots ||
2944 		    !adapter->rx_pool ||
2945 		    !adapter->tso_pool ||
2946 		    !adapter->tx_pool) {
2947 			release_napi(adapter);
2948 			release_vpd_data(adapter);
2949 
2950 			rc = init_resources(adapter);
2951 			if (rc)
2952 				goto out;
2953 
2954 		} else {
2955 			rc = init_tx_pools(netdev);
2956 			if (rc) {
2957 				netdev_dbg(netdev,
2958 					   "init tx pools failed (%d)\n",
2959 					   rc);
2960 				goto out;
2961 			}
2962 
2963 			rc = init_rx_pools(netdev);
2964 			if (rc) {
2965 				netdev_dbg(netdev,
2966 					   "init rx pools failed (%d)\n",
2967 					   rc);
2968 				goto out;
2969 			}
2970 		}
2971 		ibmvnic_disable_irqs(adapter);
2972 	}
2973 	adapter->state = VNIC_CLOSED;
2974 
2975 	if (reset_state == VNIC_CLOSED) {
2976 		rc = 0;
2977 		goto out;
2978 	}
2979 
2980 	rc = __ibmvnic_open(netdev);
2981 	if (rc) {
2982 		rc = IBMVNIC_OPEN_FAILED;
2983 		goto out;
2984 	}
2985 
2986 	/* refresh device's multicast list */
2987 	ibmvnic_set_multi(netdev);
2988 
2989 	if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2990 	    adapter->reset_reason == VNIC_RESET_MOBILITY)
2991 		__netdev_notify_peers(netdev);
2992 
2993 	rc = 0;
2994 
2995 out:
2996 	/* restore the adapter state if reset failed */
2997 	if (rc)
2998 		adapter->state = reset_state;
2999 	/* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
3000 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
3001 		rtnl_unlock();
3002 
3003 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
3004 		   adapter_state_to_string(adapter->state),
3005 		   adapter->failover_pending, rc);
3006 	return rc;
3007 }
3008 
3009 static int do_hard_reset(struct ibmvnic_adapter *adapter,
3010 			 struct ibmvnic_rwi *rwi, u32 reset_state)
3011 {
3012 	struct net_device *netdev = adapter->netdev;
3013 	int rc;
3014 
3015 	netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
3016 		   reset_reason_to_string(rwi->reset_reason));
3017 
3018 	/* read the state and check (again) after getting rtnl */
3019 	reset_state = adapter->state;
3020 
3021 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
3022 		rc = -EBUSY;
3023 		goto out;
3024 	}
3025 
3026 	netif_carrier_off(netdev);
3027 	adapter->reset_reason = rwi->reset_reason;
3028 
3029 	ibmvnic_cleanup(netdev);
3030 	release_resources(adapter);
3031 	release_sub_crqs(adapter, 0);
3032 	release_crq_queue(adapter);
3033 
3034 	/* remove the closed state so when we call open it appears
3035 	 * we are coming from the probed state.
3036 	 */
3037 	adapter->state = VNIC_PROBED;
3038 
3039 	reinit_init_done(adapter);
3040 
3041 	rc = init_crq_queue(adapter);
3042 	if (rc) {
3043 		netdev_err(adapter->netdev,
3044 			   "Couldn't initialize crq. rc=%d\n", rc);
3045 		goto out;
3046 	}
3047 
3048 	rc = ibmvnic_reset_init(adapter, false);
3049 	if (rc)
3050 		goto out;
3051 
3052 	/* If the adapter was in PROBE or DOWN state prior to the reset,
3053 	 * exit here.
3054 	 */
3055 	if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
3056 		goto out;
3057 
3058 	rc = ibmvnic_login(netdev);
3059 	if (rc)
3060 		goto out;
3061 
3062 	rc = init_resources(adapter);
3063 	if (rc)
3064 		goto out;
3065 
3066 	ibmvnic_disable_irqs(adapter);
3067 	adapter->state = VNIC_CLOSED;
3068 
3069 	if (reset_state == VNIC_CLOSED)
3070 		goto out;
3071 
3072 	rc = __ibmvnic_open(netdev);
3073 	if (rc) {
3074 		rc = IBMVNIC_OPEN_FAILED;
3075 		goto out;
3076 	}
3077 
3078 	__netdev_notify_peers(netdev);
3079 out:
3080 	/* restore adapter state if reset failed */
3081 	if (rc)
3082 		adapter->state = reset_state;
3083 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
3084 		   adapter_state_to_string(adapter->state),
3085 		   adapter->failover_pending, rc);
3086 	return rc;
3087 }
3088 
3089 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
3090 {
3091 	struct ibmvnic_rwi *rwi;
3092 	unsigned long flags;
3093 
3094 	spin_lock_irqsave(&adapter->rwi_lock, flags);
3095 
3096 	if (!list_empty(&adapter->rwi_list)) {
3097 		rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
3098 				       list);
3099 		list_del(&rwi->list);
3100 	} else {
3101 		rwi = NULL;
3102 	}
3103 
3104 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3105 	return rwi;
3106 }
3107 
3108 /**
3109  * do_passive_init - complete probing when partner device is detected.
3110  * @adapter: ibmvnic_adapter struct
3111  *
3112  * If the ibmvnic device does not have a partner device to communicate with at boot
3113  * and that partner device comes online at a later time, this function is called
3114  * to complete the initialization process of ibmvnic device.
3115  * Caller is expected to hold rtnl_lock().
3116  *
3117  * Returns non-zero if sub-CRQs are not initialized properly leaving the device
3118  * in the down state.
3119  * Returns 0 upon success and the device is in PROBED state.
3120  */
3121 
3122 static int do_passive_init(struct ibmvnic_adapter *adapter)
3123 {
3124 	unsigned long timeout = msecs_to_jiffies(30000);
3125 	struct net_device *netdev = adapter->netdev;
3126 	struct device *dev = &adapter->vdev->dev;
3127 	int rc;
3128 
3129 	netdev_dbg(netdev, "Partner device found, probing.\n");
3130 
3131 	adapter->state = VNIC_PROBING;
3132 	reinit_completion(&adapter->init_done);
3133 	adapter->init_done_rc = 0;
3134 	adapter->crq.active = true;
3135 
3136 	rc = send_crq_init_complete(adapter);
3137 	if (rc)
3138 		goto out;
3139 
3140 	rc = send_version_xchg(adapter);
3141 	if (rc)
3142 		netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
3143 
3144 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
3145 		dev_err(dev, "Initialization sequence timed out\n");
3146 		rc = -ETIMEDOUT;
3147 		goto out;
3148 	}
3149 
3150 	rc = init_sub_crqs(adapter);
3151 	if (rc) {
3152 		dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
3153 		goto out;
3154 	}
3155 
3156 	rc = init_sub_crq_irqs(adapter);
3157 	if (rc) {
3158 		dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
3159 		goto init_failed;
3160 	}
3161 
3162 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
3163 	netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
3164 	netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
3165 
3166 	adapter->state = VNIC_PROBED;
3167 	netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
3168 
3169 	return 0;
3170 
3171 init_failed:
3172 	release_sub_crqs(adapter, 1);
3173 out:
3174 	adapter->state = VNIC_DOWN;
3175 	return rc;
3176 }
3177 
3178 static void __ibmvnic_reset(struct work_struct *work)
3179 {
3180 	struct ibmvnic_adapter *adapter;
3181 	unsigned int timeout = 5000;
3182 	struct ibmvnic_rwi *tmprwi;
3183 	bool saved_state = false;
3184 	struct ibmvnic_rwi *rwi;
3185 	unsigned long flags;
3186 	struct device *dev;
3187 	bool need_reset;
3188 	int num_fails = 0;
3189 	u32 reset_state;
3190 	int rc = 0;
3191 
3192 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
3193 		dev = &adapter->vdev->dev;
3194 
3195 	/* Wait for ibmvnic_probe() to complete. If probe is taking too long
3196 	 * or if another reset is in progress, defer work for now. If probe
3197 	 * eventually fails it will flush and terminate our work.
3198 	 *
3199 	 * Three possibilities here:
3200 	 * 1. Adpater being removed  - just return
3201 	 * 2. Timed out on probe or another reset in progress - delay the work
3202 	 * 3. Completed probe - perform any resets in queue
3203 	 */
3204 	if (adapter->state == VNIC_PROBING &&
3205 	    !wait_for_completion_timeout(&adapter->probe_done, timeout)) {
3206 		dev_err(dev, "Reset thread timed out on probe");
3207 		queue_delayed_work(system_long_wq,
3208 				   &adapter->ibmvnic_delayed_reset,
3209 				   IBMVNIC_RESET_DELAY);
3210 		return;
3211 	}
3212 
3213 	/* adapter is done with probe (i.e state is never VNIC_PROBING now) */
3214 	if (adapter->state == VNIC_REMOVING)
3215 		return;
3216 
3217 	/* ->rwi_list is stable now (no one else is removing entries) */
3218 
3219 	/* ibmvnic_probe() may have purged the reset queue after we were
3220 	 * scheduled to process a reset so there maybe no resets to process.
3221 	 * Before setting the ->resetting bit though, we have to make sure
3222 	 * that there is infact a reset to process. Otherwise we may race
3223 	 * with ibmvnic_open() and end up leaving the vnic down:
3224 	 *
3225 	 *	__ibmvnic_reset()	    ibmvnic_open()
3226 	 *	-----------------	    --------------
3227 	 *
3228 	 *  set ->resetting bit
3229 	 *  				find ->resetting bit is set
3230 	 *  				set ->state to IBMVNIC_OPEN (i.e
3231 	 *  				assume reset will open device)
3232 	 *  				return
3233 	 *  find reset queue empty
3234 	 *  return
3235 	 *
3236 	 *  	Neither performed vnic login/open and vnic stays down
3237 	 *
3238 	 * If we hold the lock and conditionally set the bit, either we
3239 	 * or ibmvnic_open() will complete the open.
3240 	 */
3241 	need_reset = false;
3242 	spin_lock(&adapter->rwi_lock);
3243 	if (!list_empty(&adapter->rwi_list)) {
3244 		if (test_and_set_bit_lock(0, &adapter->resetting)) {
3245 			queue_delayed_work(system_long_wq,
3246 					   &adapter->ibmvnic_delayed_reset,
3247 					   IBMVNIC_RESET_DELAY);
3248 		} else {
3249 			need_reset = true;
3250 		}
3251 	}
3252 	spin_unlock(&adapter->rwi_lock);
3253 
3254 	if (!need_reset)
3255 		return;
3256 
3257 	rwi = get_next_rwi(adapter);
3258 	while (rwi) {
3259 		spin_lock_irqsave(&adapter->state_lock, flags);
3260 
3261 		if (adapter->state == VNIC_REMOVING ||
3262 		    adapter->state == VNIC_REMOVED) {
3263 			spin_unlock_irqrestore(&adapter->state_lock, flags);
3264 			kfree(rwi);
3265 			rc = EBUSY;
3266 			break;
3267 		}
3268 
3269 		if (!saved_state) {
3270 			reset_state = adapter->state;
3271 			saved_state = true;
3272 		}
3273 		spin_unlock_irqrestore(&adapter->state_lock, flags);
3274 
3275 		if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
3276 			rtnl_lock();
3277 			rc = do_passive_init(adapter);
3278 			rtnl_unlock();
3279 			if (!rc)
3280 				netif_carrier_on(adapter->netdev);
3281 		} else if (adapter->force_reset_recovery) {
3282 			/* Since we are doing a hard reset now, clear the
3283 			 * failover_pending flag so we don't ignore any
3284 			 * future MOBILITY or other resets.
3285 			 */
3286 			adapter->failover_pending = false;
3287 
3288 			/* Transport event occurred during previous reset */
3289 			if (adapter->wait_for_reset) {
3290 				/* Previous was CHANGE_PARAM; caller locked */
3291 				adapter->force_reset_recovery = false;
3292 				rc = do_hard_reset(adapter, rwi, reset_state);
3293 			} else {
3294 				rtnl_lock();
3295 				adapter->force_reset_recovery = false;
3296 				rc = do_hard_reset(adapter, rwi, reset_state);
3297 				rtnl_unlock();
3298 			}
3299 			if (rc)
3300 				num_fails++;
3301 			else
3302 				num_fails = 0;
3303 
3304 			/* If auto-priority-failover is enabled we can get
3305 			 * back to back failovers during resets, resulting
3306 			 * in at least two failed resets (from high-priority
3307 			 * backing device to low-priority one and then back)
3308 			 * If resets continue to fail beyond that, give the
3309 			 * adapter some time to settle down before retrying.
3310 			 */
3311 			if (num_fails >= 3) {
3312 				netdev_dbg(adapter->netdev,
3313 					   "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
3314 					   adapter_state_to_string(adapter->state),
3315 					   num_fails);
3316 				set_current_state(TASK_UNINTERRUPTIBLE);
3317 				schedule_timeout(60 * HZ);
3318 			}
3319 		} else {
3320 			rc = do_reset(adapter, rwi, reset_state);
3321 		}
3322 		tmprwi = rwi;
3323 		adapter->last_reset_time = jiffies;
3324 
3325 		if (rc)
3326 			netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
3327 
3328 		rwi = get_next_rwi(adapter);
3329 
3330 		/*
3331 		 * If there are no resets queued and the previous reset failed,
3332 		 * the adapter would be in an undefined state. So retry the
3333 		 * previous reset as a hard reset.
3334 		 *
3335 		 * Else, free the previous rwi and, if there is another reset
3336 		 * queued, process the new reset even if previous reset failed
3337 		 * (the previous reset could have failed because of a fail
3338 		 * over for instance, so process the fail over).
3339 		 */
3340 		if (!rwi && rc)
3341 			rwi = tmprwi;
3342 		else
3343 			kfree(tmprwi);
3344 
3345 		if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
3346 			    rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
3347 			adapter->force_reset_recovery = true;
3348 	}
3349 
3350 	if (adapter->wait_for_reset) {
3351 		adapter->reset_done_rc = rc;
3352 		complete(&adapter->reset_done);
3353 	}
3354 
3355 	clear_bit_unlock(0, &adapter->resetting);
3356 
3357 	netdev_dbg(adapter->netdev,
3358 		   "[S:%s FRR:%d WFR:%d] Done processing resets\n",
3359 		   adapter_state_to_string(adapter->state),
3360 		   adapter->force_reset_recovery,
3361 		   adapter->wait_for_reset);
3362 }
3363 
3364 static void __ibmvnic_delayed_reset(struct work_struct *work)
3365 {
3366 	struct ibmvnic_adapter *adapter;
3367 
3368 	adapter = container_of(work, struct ibmvnic_adapter,
3369 			       ibmvnic_delayed_reset.work);
3370 	__ibmvnic_reset(&adapter->ibmvnic_reset);
3371 }
3372 
3373 static void flush_reset_queue(struct ibmvnic_adapter *adapter)
3374 {
3375 	struct list_head *entry, *tmp_entry;
3376 
3377 	if (!list_empty(&adapter->rwi_list)) {
3378 		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
3379 			list_del(entry);
3380 			kfree(list_entry(entry, struct ibmvnic_rwi, list));
3381 		}
3382 	}
3383 }
3384 
3385 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
3386 			 enum ibmvnic_reset_reason reason)
3387 {
3388 	struct net_device *netdev = adapter->netdev;
3389 	struct ibmvnic_rwi *rwi, *tmp;
3390 	unsigned long flags;
3391 	int ret;
3392 
3393 	spin_lock_irqsave(&adapter->rwi_lock, flags);
3394 
3395 	/* If failover is pending don't schedule any other reset.
3396 	 * Instead let the failover complete. If there is already a
3397 	 * a failover reset scheduled, we will detect and drop the
3398 	 * duplicate reset when walking the ->rwi_list below.
3399 	 */
3400 	if (adapter->state == VNIC_REMOVING ||
3401 	    adapter->state == VNIC_REMOVED ||
3402 	    (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
3403 		ret = EBUSY;
3404 		netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
3405 		goto err;
3406 	}
3407 
3408 	list_for_each_entry(tmp, &adapter->rwi_list, list) {
3409 		if (tmp->reset_reason == reason) {
3410 			netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
3411 				   reset_reason_to_string(reason));
3412 			ret = EBUSY;
3413 			goto err;
3414 		}
3415 	}
3416 
3417 	rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
3418 	if (!rwi) {
3419 		ret = ENOMEM;
3420 		goto err;
3421 	}
3422 	/* if we just received a transport event,
3423 	 * flush reset queue and process this reset
3424 	 */
3425 	if (adapter->force_reset_recovery)
3426 		flush_reset_queue(adapter);
3427 
3428 	rwi->reset_reason = reason;
3429 	list_add_tail(&rwi->list, &adapter->rwi_list);
3430 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
3431 		   reset_reason_to_string(reason));
3432 	queue_work(system_long_wq, &adapter->ibmvnic_reset);
3433 
3434 	ret = 0;
3435 err:
3436 	/* ibmvnic_close() below can block, so drop the lock first */
3437 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3438 
3439 	if (ret == ENOMEM)
3440 		ibmvnic_close(netdev);
3441 
3442 	return -ret;
3443 }
3444 
3445 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
3446 {
3447 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3448 
3449 	if (test_bit(0, &adapter->resetting)) {
3450 		netdev_err(adapter->netdev,
3451 			   "Adapter is resetting, skip timeout reset\n");
3452 		return;
3453 	}
3454 	/* No queuing up reset until at least 5 seconds (default watchdog val)
3455 	 * after last reset
3456 	 */
3457 	if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
3458 		netdev_dbg(dev, "Not yet time to tx timeout.\n");
3459 		return;
3460 	}
3461 	ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
3462 }
3463 
3464 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
3465 				  struct ibmvnic_rx_buff *rx_buff)
3466 {
3467 	struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
3468 
3469 	rx_buff->skb = NULL;
3470 
3471 	pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
3472 	pool->next_alloc = (pool->next_alloc + 1) % pool->size;
3473 
3474 	atomic_dec(&pool->available);
3475 }
3476 
3477 static int ibmvnic_poll(struct napi_struct *napi, int budget)
3478 {
3479 	struct ibmvnic_sub_crq_queue *rx_scrq;
3480 	struct ibmvnic_adapter *adapter;
3481 	struct net_device *netdev;
3482 	int frames_processed;
3483 	int scrq_num;
3484 
3485 	netdev = napi->dev;
3486 	adapter = netdev_priv(netdev);
3487 	scrq_num = (int)(napi - adapter->napi);
3488 	frames_processed = 0;
3489 	rx_scrq = adapter->rx_scrq[scrq_num];
3490 
3491 restart_poll:
3492 	while (frames_processed < budget) {
3493 		struct sk_buff *skb;
3494 		struct ibmvnic_rx_buff *rx_buff;
3495 		union sub_crq *next;
3496 		u32 length;
3497 		u16 offset;
3498 		u8 flags = 0;
3499 
3500 		if (unlikely(test_bit(0, &adapter->resetting) &&
3501 			     adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
3502 			enable_scrq_irq(adapter, rx_scrq);
3503 			napi_complete_done(napi, frames_processed);
3504 			return frames_processed;
3505 		}
3506 
3507 		if (!pending_scrq(adapter, rx_scrq))
3508 			break;
3509 		next = ibmvnic_next_scrq(adapter, rx_scrq);
3510 		rx_buff = (struct ibmvnic_rx_buff *)
3511 			  be64_to_cpu(next->rx_comp.correlator);
3512 		/* do error checking */
3513 		if (next->rx_comp.rc) {
3514 			netdev_dbg(netdev, "rx buffer returned with rc %x\n",
3515 				   be16_to_cpu(next->rx_comp.rc));
3516 			/* free the entry */
3517 			next->rx_comp.first = 0;
3518 			dev_kfree_skb_any(rx_buff->skb);
3519 			remove_buff_from_pool(adapter, rx_buff);
3520 			continue;
3521 		} else if (!rx_buff->skb) {
3522 			/* free the entry */
3523 			next->rx_comp.first = 0;
3524 			remove_buff_from_pool(adapter, rx_buff);
3525 			continue;
3526 		}
3527 
3528 		length = be32_to_cpu(next->rx_comp.len);
3529 		offset = be16_to_cpu(next->rx_comp.off_frame_data);
3530 		flags = next->rx_comp.flags;
3531 		skb = rx_buff->skb;
3532 		/* load long_term_buff before copying to skb */
3533 		dma_rmb();
3534 		skb_copy_to_linear_data(skb, rx_buff->data + offset,
3535 					length);
3536 
3537 		/* VLAN Header has been stripped by the system firmware and
3538 		 * needs to be inserted by the driver
3539 		 */
3540 		if (adapter->rx_vlan_header_insertion &&
3541 		    (flags & IBMVNIC_VLAN_STRIPPED))
3542 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3543 					       ntohs(next->rx_comp.vlan_tci));
3544 
3545 		/* free the entry */
3546 		next->rx_comp.first = 0;
3547 		remove_buff_from_pool(adapter, rx_buff);
3548 
3549 		skb_put(skb, length);
3550 		skb->protocol = eth_type_trans(skb, netdev);
3551 		skb_record_rx_queue(skb, scrq_num);
3552 
3553 		if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
3554 		    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
3555 			skb->ip_summed = CHECKSUM_UNNECESSARY;
3556 		}
3557 
3558 		length = skb->len;
3559 		napi_gro_receive(napi, skb); /* send it up */
3560 		netdev->stats.rx_packets++;
3561 		netdev->stats.rx_bytes += length;
3562 		adapter->rx_stats_buffers[scrq_num].packets++;
3563 		adapter->rx_stats_buffers[scrq_num].bytes += length;
3564 		frames_processed++;
3565 	}
3566 
3567 	if (adapter->state != VNIC_CLOSING &&
3568 	    (atomic_read(&adapter->rx_pool[scrq_num].available) <
3569 	      adapter->req_rx_add_entries_per_subcrq / 2))
3570 		replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
3571 	if (frames_processed < budget) {
3572 		if (napi_complete_done(napi, frames_processed)) {
3573 			enable_scrq_irq(adapter, rx_scrq);
3574 			if (pending_scrq(adapter, rx_scrq)) {
3575 				if (napi_schedule(napi)) {
3576 					disable_scrq_irq(adapter, rx_scrq);
3577 					goto restart_poll;
3578 				}
3579 			}
3580 		}
3581 	}
3582 	return frames_processed;
3583 }
3584 
3585 static int wait_for_reset(struct ibmvnic_adapter *adapter)
3586 {
3587 	int rc, ret;
3588 
3589 	adapter->fallback.mtu = adapter->req_mtu;
3590 	adapter->fallback.rx_queues = adapter->req_rx_queues;
3591 	adapter->fallback.tx_queues = adapter->req_tx_queues;
3592 	adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
3593 	adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
3594 
3595 	reinit_completion(&adapter->reset_done);
3596 	adapter->wait_for_reset = true;
3597 	rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3598 
3599 	if (rc) {
3600 		ret = rc;
3601 		goto out;
3602 	}
3603 	rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
3604 	if (rc) {
3605 		ret = -ENODEV;
3606 		goto out;
3607 	}
3608 
3609 	ret = 0;
3610 	if (adapter->reset_done_rc) {
3611 		ret = -EIO;
3612 		adapter->desired.mtu = adapter->fallback.mtu;
3613 		adapter->desired.rx_queues = adapter->fallback.rx_queues;
3614 		adapter->desired.tx_queues = adapter->fallback.tx_queues;
3615 		adapter->desired.rx_entries = adapter->fallback.rx_entries;
3616 		adapter->desired.tx_entries = adapter->fallback.tx_entries;
3617 
3618 		reinit_completion(&adapter->reset_done);
3619 		adapter->wait_for_reset = true;
3620 		rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3621 		if (rc) {
3622 			ret = rc;
3623 			goto out;
3624 		}
3625 		rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
3626 						 60000);
3627 		if (rc) {
3628 			ret = -ENODEV;
3629 			goto out;
3630 		}
3631 	}
3632 out:
3633 	adapter->wait_for_reset = false;
3634 
3635 	return ret;
3636 }
3637 
3638 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
3639 {
3640 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3641 
3642 	adapter->desired.mtu = new_mtu + ETH_HLEN;
3643 
3644 	return wait_for_reset(adapter);
3645 }
3646 
3647 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3648 						struct net_device *dev,
3649 						netdev_features_t features)
3650 {
3651 	/* Some backing hardware adapters can not
3652 	 * handle packets with a MSS less than 224
3653 	 * or with only one segment.
3654 	 */
3655 	if (skb_is_gso(skb)) {
3656 		if (skb_shinfo(skb)->gso_size < 224 ||
3657 		    skb_shinfo(skb)->gso_segs == 1)
3658 			features &= ~NETIF_F_GSO_MASK;
3659 	}
3660 
3661 	return features;
3662 }
3663 
3664 static const struct net_device_ops ibmvnic_netdev_ops = {
3665 	.ndo_open		= ibmvnic_open,
3666 	.ndo_stop		= ibmvnic_close,
3667 	.ndo_start_xmit		= ibmvnic_xmit,
3668 	.ndo_set_rx_mode	= ibmvnic_set_multi,
3669 	.ndo_set_mac_address	= ibmvnic_set_mac,
3670 	.ndo_validate_addr	= eth_validate_addr,
3671 	.ndo_tx_timeout		= ibmvnic_tx_timeout,
3672 	.ndo_change_mtu		= ibmvnic_change_mtu,
3673 	.ndo_features_check     = ibmvnic_features_check,
3674 };
3675 
3676 /* ethtool functions */
3677 
3678 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3679 				      struct ethtool_link_ksettings *cmd)
3680 {
3681 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3682 	int rc;
3683 
3684 	rc = send_query_phys_parms(adapter);
3685 	if (rc) {
3686 		adapter->speed = SPEED_UNKNOWN;
3687 		adapter->duplex = DUPLEX_UNKNOWN;
3688 	}
3689 	cmd->base.speed = adapter->speed;
3690 	cmd->base.duplex = adapter->duplex;
3691 	cmd->base.port = PORT_FIBRE;
3692 	cmd->base.phy_address = 0;
3693 	cmd->base.autoneg = AUTONEG_ENABLE;
3694 
3695 	return 0;
3696 }
3697 
3698 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3699 				struct ethtool_drvinfo *info)
3700 {
3701 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3702 
3703 	strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3704 	strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3705 	strscpy(info->fw_version, adapter->fw_version,
3706 		sizeof(info->fw_version));
3707 }
3708 
3709 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3710 {
3711 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3712 
3713 	return adapter->msg_enable;
3714 }
3715 
3716 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3717 {
3718 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3719 
3720 	adapter->msg_enable = data;
3721 }
3722 
3723 static u32 ibmvnic_get_link(struct net_device *netdev)
3724 {
3725 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3726 
3727 	/* Don't need to send a query because we request a logical link up at
3728 	 * init and then we wait for link state indications
3729 	 */
3730 	return adapter->logical_link_state;
3731 }
3732 
3733 static void ibmvnic_get_ringparam(struct net_device *netdev,
3734 				  struct ethtool_ringparam *ring,
3735 				  struct kernel_ethtool_ringparam *kernel_ring,
3736 				  struct netlink_ext_ack *extack)
3737 {
3738 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3739 
3740 	ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3741 	ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3742 	ring->rx_mini_max_pending = 0;
3743 	ring->rx_jumbo_max_pending = 0;
3744 	ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3745 	ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3746 	ring->rx_mini_pending = 0;
3747 	ring->rx_jumbo_pending = 0;
3748 }
3749 
3750 static int ibmvnic_set_ringparam(struct net_device *netdev,
3751 				 struct ethtool_ringparam *ring,
3752 				 struct kernel_ethtool_ringparam *kernel_ring,
3753 				 struct netlink_ext_ack *extack)
3754 {
3755 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3756 
3757 	if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
3758 	    ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
3759 		netdev_err(netdev, "Invalid request.\n");
3760 		netdev_err(netdev, "Max tx buffers = %llu\n",
3761 			   adapter->max_rx_add_entries_per_subcrq);
3762 		netdev_err(netdev, "Max rx buffers = %llu\n",
3763 			   adapter->max_tx_entries_per_subcrq);
3764 		return -EINVAL;
3765 	}
3766 
3767 	adapter->desired.rx_entries = ring->rx_pending;
3768 	adapter->desired.tx_entries = ring->tx_pending;
3769 
3770 	return wait_for_reset(adapter);
3771 }
3772 
3773 static void ibmvnic_get_channels(struct net_device *netdev,
3774 				 struct ethtool_channels *channels)
3775 {
3776 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3777 
3778 	channels->max_rx = adapter->max_rx_queues;
3779 	channels->max_tx = adapter->max_tx_queues;
3780 	channels->max_other = 0;
3781 	channels->max_combined = 0;
3782 	channels->rx_count = adapter->req_rx_queues;
3783 	channels->tx_count = adapter->req_tx_queues;
3784 	channels->other_count = 0;
3785 	channels->combined_count = 0;
3786 }
3787 
3788 static int ibmvnic_set_channels(struct net_device *netdev,
3789 				struct ethtool_channels *channels)
3790 {
3791 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3792 
3793 	adapter->desired.rx_queues = channels->rx_count;
3794 	adapter->desired.tx_queues = channels->tx_count;
3795 
3796 	return wait_for_reset(adapter);
3797 }
3798 
3799 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3800 {
3801 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3802 	int i;
3803 
3804 	if (stringset != ETH_SS_STATS)
3805 		return;
3806 
3807 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
3808 		memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3809 
3810 	for (i = 0; i < adapter->req_tx_queues; i++) {
3811 		snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3812 		data += ETH_GSTRING_LEN;
3813 
3814 		snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3815 		data += ETH_GSTRING_LEN;
3816 
3817 		snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
3818 		data += ETH_GSTRING_LEN;
3819 	}
3820 
3821 	for (i = 0; i < adapter->req_rx_queues; i++) {
3822 		snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3823 		data += ETH_GSTRING_LEN;
3824 
3825 		snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3826 		data += ETH_GSTRING_LEN;
3827 
3828 		snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3829 		data += ETH_GSTRING_LEN;
3830 	}
3831 }
3832 
3833 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3834 {
3835 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3836 
3837 	switch (sset) {
3838 	case ETH_SS_STATS:
3839 		return ARRAY_SIZE(ibmvnic_stats) +
3840 		       adapter->req_tx_queues * NUM_TX_STATS +
3841 		       adapter->req_rx_queues * NUM_RX_STATS;
3842 	default:
3843 		return -EOPNOTSUPP;
3844 	}
3845 }
3846 
3847 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3848 				      struct ethtool_stats *stats, u64 *data)
3849 {
3850 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3851 	union ibmvnic_crq crq;
3852 	int i, j;
3853 	int rc;
3854 
3855 	memset(&crq, 0, sizeof(crq));
3856 	crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3857 	crq.request_statistics.cmd = REQUEST_STATISTICS;
3858 	crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3859 	crq.request_statistics.len =
3860 	    cpu_to_be32(sizeof(struct ibmvnic_statistics));
3861 
3862 	/* Wait for data to be written */
3863 	reinit_completion(&adapter->stats_done);
3864 	rc = ibmvnic_send_crq(adapter, &crq);
3865 	if (rc)
3866 		return;
3867 	rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3868 	if (rc)
3869 		return;
3870 
3871 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3872 		data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3873 				      (adapter, ibmvnic_stats[i].offset));
3874 
3875 	for (j = 0; j < adapter->req_tx_queues; j++) {
3876 		data[i] = adapter->tx_stats_buffers[j].packets;
3877 		i++;
3878 		data[i] = adapter->tx_stats_buffers[j].bytes;
3879 		i++;
3880 		data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3881 		i++;
3882 	}
3883 
3884 	for (j = 0; j < adapter->req_rx_queues; j++) {
3885 		data[i] = adapter->rx_stats_buffers[j].packets;
3886 		i++;
3887 		data[i] = adapter->rx_stats_buffers[j].bytes;
3888 		i++;
3889 		data[i] = adapter->rx_stats_buffers[j].interrupts;
3890 		i++;
3891 	}
3892 }
3893 
3894 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3895 	.get_drvinfo		= ibmvnic_get_drvinfo,
3896 	.get_msglevel		= ibmvnic_get_msglevel,
3897 	.set_msglevel		= ibmvnic_set_msglevel,
3898 	.get_link		= ibmvnic_get_link,
3899 	.get_ringparam		= ibmvnic_get_ringparam,
3900 	.set_ringparam		= ibmvnic_set_ringparam,
3901 	.get_channels		= ibmvnic_get_channels,
3902 	.set_channels		= ibmvnic_set_channels,
3903 	.get_strings            = ibmvnic_get_strings,
3904 	.get_sset_count         = ibmvnic_get_sset_count,
3905 	.get_ethtool_stats	= ibmvnic_get_ethtool_stats,
3906 	.get_link_ksettings	= ibmvnic_get_link_ksettings,
3907 };
3908 
3909 /* Routines for managing CRQs/sCRQs  */
3910 
3911 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3912 				   struct ibmvnic_sub_crq_queue *scrq)
3913 {
3914 	int rc;
3915 
3916 	if (!scrq) {
3917 		netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3918 		return -EINVAL;
3919 	}
3920 
3921 	if (scrq->irq) {
3922 		free_irq(scrq->irq, scrq);
3923 		irq_dispose_mapping(scrq->irq);
3924 		scrq->irq = 0;
3925 	}
3926 
3927 	if (scrq->msgs) {
3928 		memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3929 		atomic_set(&scrq->used, 0);
3930 		scrq->cur = 0;
3931 		scrq->ind_buf.index = 0;
3932 	} else {
3933 		netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3934 		return -EINVAL;
3935 	}
3936 
3937 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3938 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3939 	return rc;
3940 }
3941 
3942 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3943 {
3944 	int i, rc;
3945 
3946 	if (!adapter->tx_scrq || !adapter->rx_scrq)
3947 		return -EINVAL;
3948 
3949 	ibmvnic_clean_affinity(adapter);
3950 
3951 	for (i = 0; i < adapter->req_tx_queues; i++) {
3952 		netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3953 		rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3954 		if (rc)
3955 			return rc;
3956 	}
3957 
3958 	for (i = 0; i < adapter->req_rx_queues; i++) {
3959 		netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3960 		rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3961 		if (rc)
3962 			return rc;
3963 	}
3964 
3965 	return rc;
3966 }
3967 
3968 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3969 				  struct ibmvnic_sub_crq_queue *scrq,
3970 				  bool do_h_free)
3971 {
3972 	struct device *dev = &adapter->vdev->dev;
3973 	long rc;
3974 
3975 	netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3976 
3977 	if (do_h_free) {
3978 		/* Close the sub-crqs */
3979 		do {
3980 			rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3981 						adapter->vdev->unit_address,
3982 						scrq->crq_num);
3983 		} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3984 
3985 		if (rc) {
3986 			netdev_err(adapter->netdev,
3987 				   "Failed to release sub-CRQ %16lx, rc = %ld\n",
3988 				   scrq->crq_num, rc);
3989 		}
3990 	}
3991 
3992 	dma_free_coherent(dev,
3993 			  IBMVNIC_IND_ARR_SZ,
3994 			  scrq->ind_buf.indir_arr,
3995 			  scrq->ind_buf.indir_dma);
3996 
3997 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3998 			 DMA_BIDIRECTIONAL);
3999 	free_pages((unsigned long)scrq->msgs, 2);
4000 	free_cpumask_var(scrq->affinity_mask);
4001 	kfree(scrq);
4002 }
4003 
4004 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
4005 							*adapter)
4006 {
4007 	struct device *dev = &adapter->vdev->dev;
4008 	struct ibmvnic_sub_crq_queue *scrq;
4009 	int rc;
4010 
4011 	scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
4012 	if (!scrq)
4013 		return NULL;
4014 
4015 	scrq->msgs =
4016 		(union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
4017 	if (!scrq->msgs) {
4018 		dev_warn(dev, "Couldn't allocate crq queue messages page\n");
4019 		goto zero_page_failed;
4020 	}
4021 	if (!zalloc_cpumask_var(&scrq->affinity_mask, GFP_KERNEL))
4022 		goto cpumask_alloc_failed;
4023 
4024 	scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
4025 					 DMA_BIDIRECTIONAL);
4026 	if (dma_mapping_error(dev, scrq->msg_token)) {
4027 		dev_warn(dev, "Couldn't map crq queue messages page\n");
4028 		goto map_failed;
4029 	}
4030 
4031 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
4032 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
4033 
4034 	if (rc == H_RESOURCE)
4035 		rc = ibmvnic_reset_crq(adapter);
4036 
4037 	if (rc == H_CLOSED) {
4038 		dev_warn(dev, "Partner adapter not ready, waiting.\n");
4039 	} else if (rc) {
4040 		dev_warn(dev, "Error %d registering sub-crq\n", rc);
4041 		goto reg_failed;
4042 	}
4043 
4044 	scrq->adapter = adapter;
4045 	scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
4046 	scrq->ind_buf.index = 0;
4047 
4048 	scrq->ind_buf.indir_arr =
4049 		dma_alloc_coherent(dev,
4050 				   IBMVNIC_IND_ARR_SZ,
4051 				   &scrq->ind_buf.indir_dma,
4052 				   GFP_KERNEL);
4053 
4054 	if (!scrq->ind_buf.indir_arr)
4055 		goto indir_failed;
4056 
4057 	spin_lock_init(&scrq->lock);
4058 
4059 	netdev_dbg(adapter->netdev,
4060 		   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
4061 		   scrq->crq_num, scrq->hw_irq, scrq->irq);
4062 
4063 	return scrq;
4064 
4065 indir_failed:
4066 	do {
4067 		rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
4068 					adapter->vdev->unit_address,
4069 					scrq->crq_num);
4070 	} while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
4071 reg_failed:
4072 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
4073 			 DMA_BIDIRECTIONAL);
4074 map_failed:
4075 	free_cpumask_var(scrq->affinity_mask);
4076 cpumask_alloc_failed:
4077 	free_pages((unsigned long)scrq->msgs, 2);
4078 zero_page_failed:
4079 	kfree(scrq);
4080 
4081 	return NULL;
4082 }
4083 
4084 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
4085 {
4086 	int i;
4087 
4088 	ibmvnic_clean_affinity(adapter);
4089 	if (adapter->tx_scrq) {
4090 		for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
4091 			if (!adapter->tx_scrq[i])
4092 				continue;
4093 
4094 			netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
4095 				   i);
4096 			ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
4097 			if (adapter->tx_scrq[i]->irq) {
4098 				free_irq(adapter->tx_scrq[i]->irq,
4099 					 adapter->tx_scrq[i]);
4100 				irq_dispose_mapping(adapter->tx_scrq[i]->irq);
4101 				adapter->tx_scrq[i]->irq = 0;
4102 			}
4103 
4104 			release_sub_crq_queue(adapter, adapter->tx_scrq[i],
4105 					      do_h_free);
4106 		}
4107 
4108 		kfree(adapter->tx_scrq);
4109 		adapter->tx_scrq = NULL;
4110 		adapter->num_active_tx_scrqs = 0;
4111 	}
4112 
4113 	/* Clean any remaining outstanding SKBs
4114 	 * we freed the irq so we won't be hearing
4115 	 * from them
4116 	 */
4117 	clean_tx_pools(adapter);
4118 
4119 	if (adapter->rx_scrq) {
4120 		for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
4121 			if (!adapter->rx_scrq[i])
4122 				continue;
4123 
4124 			netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
4125 				   i);
4126 			if (adapter->rx_scrq[i]->irq) {
4127 				free_irq(adapter->rx_scrq[i]->irq,
4128 					 adapter->rx_scrq[i]);
4129 				irq_dispose_mapping(adapter->rx_scrq[i]->irq);
4130 				adapter->rx_scrq[i]->irq = 0;
4131 			}
4132 
4133 			release_sub_crq_queue(adapter, adapter->rx_scrq[i],
4134 					      do_h_free);
4135 		}
4136 
4137 		kfree(adapter->rx_scrq);
4138 		adapter->rx_scrq = NULL;
4139 		adapter->num_active_rx_scrqs = 0;
4140 	}
4141 }
4142 
4143 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
4144 			    struct ibmvnic_sub_crq_queue *scrq)
4145 {
4146 	struct device *dev = &adapter->vdev->dev;
4147 	unsigned long rc;
4148 
4149 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4150 				H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4151 	if (rc)
4152 		dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
4153 			scrq->hw_irq, rc);
4154 	return rc;
4155 }
4156 
4157 /* We can not use the IRQ chip EOI handler because that has the
4158  * unintended effect of changing the interrupt priority.
4159  */
4160 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq)
4161 {
4162 	u64 val = 0xff000000 | scrq->hw_irq;
4163 	unsigned long rc;
4164 
4165 	rc = plpar_hcall_norets(H_EOI, val);
4166 	if (rc)
4167 		dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc);
4168 }
4169 
4170 /* Due to a firmware bug, the hypervisor can send an interrupt to a
4171  * transmit or receive queue just prior to a partition migration.
4172  * Force an EOI after migration.
4173  */
4174 static void ibmvnic_clear_pending_interrupt(struct device *dev,
4175 					    struct ibmvnic_sub_crq_queue *scrq)
4176 {
4177 	if (!xive_enabled())
4178 		ibmvnic_xics_eoi(dev, scrq);
4179 }
4180 
4181 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
4182 			   struct ibmvnic_sub_crq_queue *scrq)
4183 {
4184 	struct device *dev = &adapter->vdev->dev;
4185 	unsigned long rc;
4186 
4187 	if (scrq->hw_irq > 0x100000000ULL) {
4188 		dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
4189 		return 1;
4190 	}
4191 
4192 	if (test_bit(0, &adapter->resetting) &&
4193 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
4194 		ibmvnic_clear_pending_interrupt(dev, scrq);
4195 	}
4196 
4197 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4198 				H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4199 	if (rc)
4200 		dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
4201 			scrq->hw_irq, rc);
4202 	return rc;
4203 }
4204 
4205 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
4206 			       struct ibmvnic_sub_crq_queue *scrq)
4207 {
4208 	struct device *dev = &adapter->vdev->dev;
4209 	int num_packets = 0, total_bytes = 0;
4210 	struct ibmvnic_tx_pool *tx_pool;
4211 	struct ibmvnic_tx_buff *txbuff;
4212 	struct netdev_queue *txq;
4213 	union sub_crq *next;
4214 	int index, i;
4215 
4216 restart_loop:
4217 	while (pending_scrq(adapter, scrq)) {
4218 		unsigned int pool = scrq->pool_index;
4219 		int num_entries = 0;
4220 		next = ibmvnic_next_scrq(adapter, scrq);
4221 		for (i = 0; i < next->tx_comp.num_comps; i++) {
4222 			index = be32_to_cpu(next->tx_comp.correlators[i]);
4223 			if (index & IBMVNIC_TSO_POOL_MASK) {
4224 				tx_pool = &adapter->tso_pool[pool];
4225 				index &= ~IBMVNIC_TSO_POOL_MASK;
4226 			} else {
4227 				tx_pool = &adapter->tx_pool[pool];
4228 			}
4229 
4230 			txbuff = &tx_pool->tx_buff[index];
4231 			num_packets++;
4232 			num_entries += txbuff->num_entries;
4233 			if (txbuff->skb) {
4234 				total_bytes += txbuff->skb->len;
4235 				if (next->tx_comp.rcs[i]) {
4236 					dev_err(dev, "tx error %x\n",
4237 						next->tx_comp.rcs[i]);
4238 					dev_kfree_skb_irq(txbuff->skb);
4239 				} else {
4240 					dev_consume_skb_irq(txbuff->skb);
4241 				}
4242 				txbuff->skb = NULL;
4243 			} else {
4244 				netdev_warn(adapter->netdev,
4245 					    "TX completion received with NULL socket buffer\n");
4246 			}
4247 			tx_pool->free_map[tx_pool->producer_index] = index;
4248 			tx_pool->producer_index =
4249 				(tx_pool->producer_index + 1) %
4250 					tx_pool->num_buffers;
4251 		}
4252 		/* remove tx_comp scrq*/
4253 		next->tx_comp.first = 0;
4254 
4255 
4256 		if (atomic_sub_return(num_entries, &scrq->used) <=
4257 		    (adapter->req_tx_entries_per_subcrq / 2) &&
4258 		    __netif_subqueue_stopped(adapter->netdev,
4259 					     scrq->pool_index)) {
4260 			rcu_read_lock();
4261 			if (adapter->tx_queues_active) {
4262 				netif_wake_subqueue(adapter->netdev,
4263 						    scrq->pool_index);
4264 				netdev_dbg(adapter->netdev,
4265 					   "Started queue %d\n",
4266 					   scrq->pool_index);
4267 			}
4268 			rcu_read_unlock();
4269 		}
4270 	}
4271 
4272 	enable_scrq_irq(adapter, scrq);
4273 
4274 	if (pending_scrq(adapter, scrq)) {
4275 		disable_scrq_irq(adapter, scrq);
4276 		goto restart_loop;
4277 	}
4278 
4279 	txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
4280 	netdev_tx_completed_queue(txq, num_packets, total_bytes);
4281 
4282 	return 0;
4283 }
4284 
4285 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
4286 {
4287 	struct ibmvnic_sub_crq_queue *scrq = instance;
4288 	struct ibmvnic_adapter *adapter = scrq->adapter;
4289 
4290 	disable_scrq_irq(adapter, scrq);
4291 	ibmvnic_complete_tx(adapter, scrq);
4292 
4293 	return IRQ_HANDLED;
4294 }
4295 
4296 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
4297 {
4298 	struct ibmvnic_sub_crq_queue *scrq = instance;
4299 	struct ibmvnic_adapter *adapter = scrq->adapter;
4300 
4301 	/* When booting a kdump kernel we can hit pending interrupts
4302 	 * prior to completing driver initialization.
4303 	 */
4304 	if (unlikely(adapter->state != VNIC_OPEN))
4305 		return IRQ_NONE;
4306 
4307 	adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
4308 
4309 	if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
4310 		disable_scrq_irq(adapter, scrq);
4311 		__napi_schedule(&adapter->napi[scrq->scrq_num]);
4312 	}
4313 
4314 	return IRQ_HANDLED;
4315 }
4316 
4317 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
4318 {
4319 	struct device *dev = &adapter->vdev->dev;
4320 	struct ibmvnic_sub_crq_queue *scrq;
4321 	int i = 0, j = 0;
4322 	int rc = 0;
4323 
4324 	for (i = 0; i < adapter->req_tx_queues; i++) {
4325 		netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
4326 			   i);
4327 		scrq = adapter->tx_scrq[i];
4328 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4329 
4330 		if (!scrq->irq) {
4331 			rc = -EINVAL;
4332 			dev_err(dev, "Error mapping irq\n");
4333 			goto req_tx_irq_failed;
4334 		}
4335 
4336 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
4337 			 adapter->vdev->unit_address, i);
4338 		rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
4339 				 0, scrq->name, scrq);
4340 
4341 		if (rc) {
4342 			dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
4343 				scrq->irq, rc);
4344 			irq_dispose_mapping(scrq->irq);
4345 			goto req_tx_irq_failed;
4346 		}
4347 	}
4348 
4349 	for (i = 0; i < adapter->req_rx_queues; i++) {
4350 		netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
4351 			   i);
4352 		scrq = adapter->rx_scrq[i];
4353 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4354 		if (!scrq->irq) {
4355 			rc = -EINVAL;
4356 			dev_err(dev, "Error mapping irq\n");
4357 			goto req_rx_irq_failed;
4358 		}
4359 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
4360 			 adapter->vdev->unit_address, i);
4361 		rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
4362 				 0, scrq->name, scrq);
4363 		if (rc) {
4364 			dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
4365 				scrq->irq, rc);
4366 			irq_dispose_mapping(scrq->irq);
4367 			goto req_rx_irq_failed;
4368 		}
4369 	}
4370 
4371 	cpus_read_lock();
4372 	ibmvnic_set_affinity(adapter);
4373 	cpus_read_unlock();
4374 
4375 	return rc;
4376 
4377 req_rx_irq_failed:
4378 	for (j = 0; j < i; j++) {
4379 		free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
4380 		irq_dispose_mapping(adapter->rx_scrq[j]->irq);
4381 	}
4382 	i = adapter->req_tx_queues;
4383 req_tx_irq_failed:
4384 	for (j = 0; j < i; j++) {
4385 		free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
4386 		irq_dispose_mapping(adapter->tx_scrq[j]->irq);
4387 	}
4388 	release_sub_crqs(adapter, 1);
4389 	return rc;
4390 }
4391 
4392 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
4393 {
4394 	struct device *dev = &adapter->vdev->dev;
4395 	struct ibmvnic_sub_crq_queue **allqueues;
4396 	int registered_queues = 0;
4397 	int total_queues;
4398 	int more = 0;
4399 	int i;
4400 
4401 	total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
4402 
4403 	allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
4404 	if (!allqueues)
4405 		return -ENOMEM;
4406 
4407 	for (i = 0; i < total_queues; i++) {
4408 		allqueues[i] = init_sub_crq_queue(adapter);
4409 		if (!allqueues[i]) {
4410 			dev_warn(dev, "Couldn't allocate all sub-crqs\n");
4411 			break;
4412 		}
4413 		registered_queues++;
4414 	}
4415 
4416 	/* Make sure we were able to register the minimum number of queues */
4417 	if (registered_queues <
4418 	    adapter->min_tx_queues + adapter->min_rx_queues) {
4419 		dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
4420 		goto tx_failed;
4421 	}
4422 
4423 	/* Distribute the failed allocated queues*/
4424 	for (i = 0; i < total_queues - registered_queues + more ; i++) {
4425 		netdev_dbg(adapter->netdev, "Reducing number of queues\n");
4426 		switch (i % 3) {
4427 		case 0:
4428 			if (adapter->req_rx_queues > adapter->min_rx_queues)
4429 				adapter->req_rx_queues--;
4430 			else
4431 				more++;
4432 			break;
4433 		case 1:
4434 			if (adapter->req_tx_queues > adapter->min_tx_queues)
4435 				adapter->req_tx_queues--;
4436 			else
4437 				more++;
4438 			break;
4439 		}
4440 	}
4441 
4442 	adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
4443 				   sizeof(*adapter->tx_scrq), GFP_KERNEL);
4444 	if (!adapter->tx_scrq)
4445 		goto tx_failed;
4446 
4447 	for (i = 0; i < adapter->req_tx_queues; i++) {
4448 		adapter->tx_scrq[i] = allqueues[i];
4449 		adapter->tx_scrq[i]->pool_index = i;
4450 		adapter->num_active_tx_scrqs++;
4451 	}
4452 
4453 	adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
4454 				   sizeof(*adapter->rx_scrq), GFP_KERNEL);
4455 	if (!adapter->rx_scrq)
4456 		goto rx_failed;
4457 
4458 	for (i = 0; i < adapter->req_rx_queues; i++) {
4459 		adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
4460 		adapter->rx_scrq[i]->scrq_num = i;
4461 		adapter->num_active_rx_scrqs++;
4462 	}
4463 
4464 	kfree(allqueues);
4465 	return 0;
4466 
4467 rx_failed:
4468 	kfree(adapter->tx_scrq);
4469 	adapter->tx_scrq = NULL;
4470 tx_failed:
4471 	for (i = 0; i < registered_queues; i++)
4472 		release_sub_crq_queue(adapter, allqueues[i], 1);
4473 	kfree(allqueues);
4474 	return -ENOMEM;
4475 }
4476 
4477 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
4478 {
4479 	struct device *dev = &adapter->vdev->dev;
4480 	union ibmvnic_crq crq;
4481 	int max_entries;
4482 	int cap_reqs;
4483 
4484 	/* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
4485 	 * the PROMISC flag). Initialize this count upfront. When the tasklet
4486 	 * receives a response to all of these, it will send the next protocol
4487 	 * message (QUERY_IP_OFFLOAD).
4488 	 */
4489 	if (!(adapter->netdev->flags & IFF_PROMISC) ||
4490 	    adapter->promisc_supported)
4491 		cap_reqs = 7;
4492 	else
4493 		cap_reqs = 6;
4494 
4495 	if (!retry) {
4496 		/* Sub-CRQ entries are 32 byte long */
4497 		int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
4498 
4499 		atomic_set(&adapter->running_cap_crqs, cap_reqs);
4500 
4501 		if (adapter->min_tx_entries_per_subcrq > entries_page ||
4502 		    adapter->min_rx_add_entries_per_subcrq > entries_page) {
4503 			dev_err(dev, "Fatal, invalid entries per sub-crq\n");
4504 			return;
4505 		}
4506 
4507 		if (adapter->desired.mtu)
4508 			adapter->req_mtu = adapter->desired.mtu;
4509 		else
4510 			adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
4511 
4512 		if (!adapter->desired.tx_entries)
4513 			adapter->desired.tx_entries =
4514 					adapter->max_tx_entries_per_subcrq;
4515 		if (!adapter->desired.rx_entries)
4516 			adapter->desired.rx_entries =
4517 					adapter->max_rx_add_entries_per_subcrq;
4518 
4519 		max_entries = IBMVNIC_LTB_SET_SIZE /
4520 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
4521 
4522 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4523 			adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) {
4524 			adapter->desired.tx_entries = max_entries;
4525 		}
4526 
4527 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4528 			adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) {
4529 			adapter->desired.rx_entries = max_entries;
4530 		}
4531 
4532 		if (adapter->desired.tx_entries)
4533 			adapter->req_tx_entries_per_subcrq =
4534 					adapter->desired.tx_entries;
4535 		else
4536 			adapter->req_tx_entries_per_subcrq =
4537 					adapter->max_tx_entries_per_subcrq;
4538 
4539 		if (adapter->desired.rx_entries)
4540 			adapter->req_rx_add_entries_per_subcrq =
4541 					adapter->desired.rx_entries;
4542 		else
4543 			adapter->req_rx_add_entries_per_subcrq =
4544 					adapter->max_rx_add_entries_per_subcrq;
4545 
4546 		if (adapter->desired.tx_queues)
4547 			adapter->req_tx_queues =
4548 					adapter->desired.tx_queues;
4549 		else
4550 			adapter->req_tx_queues =
4551 					adapter->opt_tx_comp_sub_queues;
4552 
4553 		if (adapter->desired.rx_queues)
4554 			adapter->req_rx_queues =
4555 					adapter->desired.rx_queues;
4556 		else
4557 			adapter->req_rx_queues =
4558 					adapter->opt_rx_comp_queues;
4559 
4560 		adapter->req_rx_add_queues = adapter->max_rx_add_queues;
4561 	} else {
4562 		atomic_add(cap_reqs, &adapter->running_cap_crqs);
4563 	}
4564 	memset(&crq, 0, sizeof(crq));
4565 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
4566 	crq.request_capability.cmd = REQUEST_CAPABILITY;
4567 
4568 	crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
4569 	crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
4570 	cap_reqs--;
4571 	ibmvnic_send_crq(adapter, &crq);
4572 
4573 	crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
4574 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
4575 	cap_reqs--;
4576 	ibmvnic_send_crq(adapter, &crq);
4577 
4578 	crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
4579 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
4580 	cap_reqs--;
4581 	ibmvnic_send_crq(adapter, &crq);
4582 
4583 	crq.request_capability.capability =
4584 	    cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
4585 	crq.request_capability.number =
4586 	    cpu_to_be64(adapter->req_tx_entries_per_subcrq);
4587 	cap_reqs--;
4588 	ibmvnic_send_crq(adapter, &crq);
4589 
4590 	crq.request_capability.capability =
4591 	    cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
4592 	crq.request_capability.number =
4593 	    cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
4594 	cap_reqs--;
4595 	ibmvnic_send_crq(adapter, &crq);
4596 
4597 	crq.request_capability.capability = cpu_to_be16(REQ_MTU);
4598 	crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
4599 	cap_reqs--;
4600 	ibmvnic_send_crq(adapter, &crq);
4601 
4602 	if (adapter->netdev->flags & IFF_PROMISC) {
4603 		if (adapter->promisc_supported) {
4604 			crq.request_capability.capability =
4605 			    cpu_to_be16(PROMISC_REQUESTED);
4606 			crq.request_capability.number = cpu_to_be64(1);
4607 			cap_reqs--;
4608 			ibmvnic_send_crq(adapter, &crq);
4609 		}
4610 	} else {
4611 		crq.request_capability.capability =
4612 		    cpu_to_be16(PROMISC_REQUESTED);
4613 		crq.request_capability.number = cpu_to_be64(0);
4614 		cap_reqs--;
4615 		ibmvnic_send_crq(adapter, &crq);
4616 	}
4617 
4618 	/* Keep at end to catch any discrepancy between expected and actual
4619 	 * CRQs sent.
4620 	 */
4621 	WARN_ON(cap_reqs != 0);
4622 }
4623 
4624 static int pending_scrq(struct ibmvnic_adapter *adapter,
4625 			struct ibmvnic_sub_crq_queue *scrq)
4626 {
4627 	union sub_crq *entry = &scrq->msgs[scrq->cur];
4628 	int rc;
4629 
4630 	rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4631 
4632 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4633 	 * contents of the SCRQ descriptor
4634 	 */
4635 	dma_rmb();
4636 
4637 	return rc;
4638 }
4639 
4640 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4641 					struct ibmvnic_sub_crq_queue *scrq)
4642 {
4643 	union sub_crq *entry;
4644 	unsigned long flags;
4645 
4646 	spin_lock_irqsave(&scrq->lock, flags);
4647 	entry = &scrq->msgs[scrq->cur];
4648 	if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4649 		if (++scrq->cur == scrq->size)
4650 			scrq->cur = 0;
4651 	} else {
4652 		entry = NULL;
4653 	}
4654 	spin_unlock_irqrestore(&scrq->lock, flags);
4655 
4656 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4657 	 * contents of the SCRQ descriptor
4658 	 */
4659 	dma_rmb();
4660 
4661 	return entry;
4662 }
4663 
4664 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4665 {
4666 	struct ibmvnic_crq_queue *queue = &adapter->crq;
4667 	union ibmvnic_crq *crq;
4668 
4669 	crq = &queue->msgs[queue->cur];
4670 	if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4671 		if (++queue->cur == queue->size)
4672 			queue->cur = 0;
4673 	} else {
4674 		crq = NULL;
4675 	}
4676 
4677 	return crq;
4678 }
4679 
4680 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4681 {
4682 	switch (rc) {
4683 	case H_PARAMETER:
4684 		dev_warn_ratelimited(dev,
4685 				     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4686 				     func, rc);
4687 		break;
4688 	case H_CLOSED:
4689 		dev_warn_ratelimited(dev,
4690 				     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4691 				     func, rc);
4692 		break;
4693 	default:
4694 		dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4695 		break;
4696 	}
4697 }
4698 
4699 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4700 				u64 remote_handle, u64 ioba, u64 num_entries)
4701 {
4702 	unsigned int ua = adapter->vdev->unit_address;
4703 	struct device *dev = &adapter->vdev->dev;
4704 	int rc;
4705 
4706 	/* Make sure the hypervisor sees the complete request */
4707 	dma_wmb();
4708 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4709 				cpu_to_be64(remote_handle),
4710 				ioba, num_entries);
4711 
4712 	if (rc)
4713 		print_subcrq_error(dev, rc, __func__);
4714 
4715 	return rc;
4716 }
4717 
4718 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4719 			    union ibmvnic_crq *crq)
4720 {
4721 	unsigned int ua = adapter->vdev->unit_address;
4722 	struct device *dev = &adapter->vdev->dev;
4723 	u64 *u64_crq = (u64 *)crq;
4724 	int rc;
4725 
4726 	netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4727 		   (unsigned long)cpu_to_be64(u64_crq[0]),
4728 		   (unsigned long)cpu_to_be64(u64_crq[1]));
4729 
4730 	if (!adapter->crq.active &&
4731 	    crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4732 		dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4733 		return -EINVAL;
4734 	}
4735 
4736 	/* Make sure the hypervisor sees the complete request */
4737 	dma_wmb();
4738 
4739 	rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4740 				cpu_to_be64(u64_crq[0]),
4741 				cpu_to_be64(u64_crq[1]));
4742 
4743 	if (rc) {
4744 		if (rc == H_CLOSED) {
4745 			dev_warn(dev, "CRQ Queue closed\n");
4746 			/* do not reset, report the fail, wait for passive init from server */
4747 		}
4748 
4749 		dev_warn(dev, "Send error (rc=%d)\n", rc);
4750 	}
4751 
4752 	return rc;
4753 }
4754 
4755 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4756 {
4757 	struct device *dev = &adapter->vdev->dev;
4758 	union ibmvnic_crq crq;
4759 	int retries = 100;
4760 	int rc;
4761 
4762 	memset(&crq, 0, sizeof(crq));
4763 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4764 	crq.generic.cmd = IBMVNIC_CRQ_INIT;
4765 	netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4766 
4767 	do {
4768 		rc = ibmvnic_send_crq(adapter, &crq);
4769 		if (rc != H_CLOSED)
4770 			break;
4771 		retries--;
4772 		msleep(50);
4773 
4774 	} while (retries > 0);
4775 
4776 	if (rc) {
4777 		dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4778 		return rc;
4779 	}
4780 
4781 	return 0;
4782 }
4783 
4784 struct vnic_login_client_data {
4785 	u8	type;
4786 	__be16	len;
4787 	char	name[];
4788 } __packed;
4789 
4790 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4791 {
4792 	int len;
4793 
4794 	/* Calculate the amount of buffer space needed for the
4795 	 * vnic client data in the login buffer. There are four entries,
4796 	 * OS name, LPAR name, device name, and a null last entry.
4797 	 */
4798 	len = 4 * sizeof(struct vnic_login_client_data);
4799 	len += 6; /* "Linux" plus NULL */
4800 	len += strlen(utsname()->nodename) + 1;
4801 	len += strlen(adapter->netdev->name) + 1;
4802 
4803 	return len;
4804 }
4805 
4806 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4807 				 struct vnic_login_client_data *vlcd)
4808 {
4809 	const char *os_name = "Linux";
4810 	int len;
4811 
4812 	/* Type 1 - LPAR OS */
4813 	vlcd->type = 1;
4814 	len = strlen(os_name) + 1;
4815 	vlcd->len = cpu_to_be16(len);
4816 	strscpy(vlcd->name, os_name, len);
4817 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4818 
4819 	/* Type 2 - LPAR name */
4820 	vlcd->type = 2;
4821 	len = strlen(utsname()->nodename) + 1;
4822 	vlcd->len = cpu_to_be16(len);
4823 	strscpy(vlcd->name, utsname()->nodename, len);
4824 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4825 
4826 	/* Type 3 - device name */
4827 	vlcd->type = 3;
4828 	len = strlen(adapter->netdev->name) + 1;
4829 	vlcd->len = cpu_to_be16(len);
4830 	strscpy(vlcd->name, adapter->netdev->name, len);
4831 }
4832 
4833 static int send_login(struct ibmvnic_adapter *adapter)
4834 {
4835 	struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4836 	struct ibmvnic_login_buffer *login_buffer;
4837 	struct device *dev = &adapter->vdev->dev;
4838 	struct vnic_login_client_data *vlcd;
4839 	dma_addr_t rsp_buffer_token;
4840 	dma_addr_t buffer_token;
4841 	size_t rsp_buffer_size;
4842 	union ibmvnic_crq crq;
4843 	int client_data_len;
4844 	size_t buffer_size;
4845 	__be64 *tx_list_p;
4846 	__be64 *rx_list_p;
4847 	int rc;
4848 	int i;
4849 
4850 	if (!adapter->tx_scrq || !adapter->rx_scrq) {
4851 		netdev_err(adapter->netdev,
4852 			   "RX or TX queues are not allocated, device login failed\n");
4853 		return -ENOMEM;
4854 	}
4855 
4856 	release_login_buffer(adapter);
4857 	release_login_rsp_buffer(adapter);
4858 
4859 	client_data_len = vnic_client_data_len(adapter);
4860 
4861 	buffer_size =
4862 	    sizeof(struct ibmvnic_login_buffer) +
4863 	    sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4864 	    client_data_len;
4865 
4866 	login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4867 	if (!login_buffer)
4868 		goto buf_alloc_failed;
4869 
4870 	buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4871 				      DMA_TO_DEVICE);
4872 	if (dma_mapping_error(dev, buffer_token)) {
4873 		dev_err(dev, "Couldn't map login buffer\n");
4874 		goto buf_map_failed;
4875 	}
4876 
4877 	rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4878 			  sizeof(u64) * adapter->req_tx_queues +
4879 			  sizeof(u64) * adapter->req_rx_queues +
4880 			  sizeof(u64) * adapter->req_rx_queues +
4881 			  sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4882 
4883 	login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4884 	if (!login_rsp_buffer)
4885 		goto buf_rsp_alloc_failed;
4886 
4887 	rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4888 					  rsp_buffer_size, DMA_FROM_DEVICE);
4889 	if (dma_mapping_error(dev, rsp_buffer_token)) {
4890 		dev_err(dev, "Couldn't map login rsp buffer\n");
4891 		goto buf_rsp_map_failed;
4892 	}
4893 
4894 	adapter->login_buf = login_buffer;
4895 	adapter->login_buf_token = buffer_token;
4896 	adapter->login_buf_sz = buffer_size;
4897 	adapter->login_rsp_buf = login_rsp_buffer;
4898 	adapter->login_rsp_buf_token = rsp_buffer_token;
4899 	adapter->login_rsp_buf_sz = rsp_buffer_size;
4900 
4901 	login_buffer->len = cpu_to_be32(buffer_size);
4902 	login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4903 	login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4904 	login_buffer->off_txcomp_subcrqs =
4905 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4906 	login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4907 	login_buffer->off_rxcomp_subcrqs =
4908 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4909 			sizeof(u64) * adapter->req_tx_queues);
4910 	login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4911 	login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4912 
4913 	tx_list_p = (__be64 *)((char *)login_buffer +
4914 				      sizeof(struct ibmvnic_login_buffer));
4915 	rx_list_p = (__be64 *)((char *)login_buffer +
4916 				      sizeof(struct ibmvnic_login_buffer) +
4917 				      sizeof(u64) * adapter->req_tx_queues);
4918 
4919 	for (i = 0; i < adapter->req_tx_queues; i++) {
4920 		if (adapter->tx_scrq[i]) {
4921 			tx_list_p[i] =
4922 				cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4923 		}
4924 	}
4925 
4926 	for (i = 0; i < adapter->req_rx_queues; i++) {
4927 		if (adapter->rx_scrq[i]) {
4928 			rx_list_p[i] =
4929 				cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4930 		}
4931 	}
4932 
4933 	/* Insert vNIC login client data */
4934 	vlcd = (struct vnic_login_client_data *)
4935 		((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4936 	login_buffer->client_data_offset =
4937 			cpu_to_be32((char *)vlcd - (char *)login_buffer);
4938 	login_buffer->client_data_len = cpu_to_be32(client_data_len);
4939 
4940 	vnic_add_client_data(adapter, vlcd);
4941 
4942 	netdev_dbg(adapter->netdev, "Login Buffer:\n");
4943 	for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4944 		netdev_dbg(adapter->netdev, "%016lx\n",
4945 			   ((unsigned long *)(adapter->login_buf))[i]);
4946 	}
4947 
4948 	memset(&crq, 0, sizeof(crq));
4949 	crq.login.first = IBMVNIC_CRQ_CMD;
4950 	crq.login.cmd = LOGIN;
4951 	crq.login.ioba = cpu_to_be32(buffer_token);
4952 	crq.login.len = cpu_to_be32(buffer_size);
4953 
4954 	adapter->login_pending = true;
4955 	rc = ibmvnic_send_crq(adapter, &crq);
4956 	if (rc) {
4957 		adapter->login_pending = false;
4958 		netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4959 		goto buf_send_failed;
4960 	}
4961 
4962 	return 0;
4963 
4964 buf_send_failed:
4965 	dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size,
4966 			 DMA_FROM_DEVICE);
4967 buf_rsp_map_failed:
4968 	kfree(login_rsp_buffer);
4969 	adapter->login_rsp_buf = NULL;
4970 buf_rsp_alloc_failed:
4971 	dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4972 buf_map_failed:
4973 	kfree(login_buffer);
4974 	adapter->login_buf = NULL;
4975 buf_alloc_failed:
4976 	return -ENOMEM;
4977 }
4978 
4979 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4980 			    u32 len, u8 map_id)
4981 {
4982 	union ibmvnic_crq crq;
4983 
4984 	memset(&crq, 0, sizeof(crq));
4985 	crq.request_map.first = IBMVNIC_CRQ_CMD;
4986 	crq.request_map.cmd = REQUEST_MAP;
4987 	crq.request_map.map_id = map_id;
4988 	crq.request_map.ioba = cpu_to_be32(addr);
4989 	crq.request_map.len = cpu_to_be32(len);
4990 	return ibmvnic_send_crq(adapter, &crq);
4991 }
4992 
4993 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4994 {
4995 	union ibmvnic_crq crq;
4996 
4997 	memset(&crq, 0, sizeof(crq));
4998 	crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4999 	crq.request_unmap.cmd = REQUEST_UNMAP;
5000 	crq.request_unmap.map_id = map_id;
5001 	return ibmvnic_send_crq(adapter, &crq);
5002 }
5003 
5004 static void send_query_map(struct ibmvnic_adapter *adapter)
5005 {
5006 	union ibmvnic_crq crq;
5007 
5008 	memset(&crq, 0, sizeof(crq));
5009 	crq.query_map.first = IBMVNIC_CRQ_CMD;
5010 	crq.query_map.cmd = QUERY_MAP;
5011 	ibmvnic_send_crq(adapter, &crq);
5012 }
5013 
5014 /* Send a series of CRQs requesting various capabilities of the VNIC server */
5015 static void send_query_cap(struct ibmvnic_adapter *adapter)
5016 {
5017 	union ibmvnic_crq crq;
5018 	int cap_reqs;
5019 
5020 	/* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
5021 	 * upfront. When the tasklet receives a response to all of these, it
5022 	 * can send out the next protocol messaage (REQUEST_CAPABILITY).
5023 	 */
5024 	cap_reqs = 25;
5025 
5026 	atomic_set(&adapter->running_cap_crqs, cap_reqs);
5027 
5028 	memset(&crq, 0, sizeof(crq));
5029 	crq.query_capability.first = IBMVNIC_CRQ_CMD;
5030 	crq.query_capability.cmd = QUERY_CAPABILITY;
5031 
5032 	crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
5033 	ibmvnic_send_crq(adapter, &crq);
5034 	cap_reqs--;
5035 
5036 	crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
5037 	ibmvnic_send_crq(adapter, &crq);
5038 	cap_reqs--;
5039 
5040 	crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
5041 	ibmvnic_send_crq(adapter, &crq);
5042 	cap_reqs--;
5043 
5044 	crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
5045 	ibmvnic_send_crq(adapter, &crq);
5046 	cap_reqs--;
5047 
5048 	crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
5049 	ibmvnic_send_crq(adapter, &crq);
5050 	cap_reqs--;
5051 
5052 	crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
5053 	ibmvnic_send_crq(adapter, &crq);
5054 	cap_reqs--;
5055 
5056 	crq.query_capability.capability =
5057 	    cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
5058 	ibmvnic_send_crq(adapter, &crq);
5059 	cap_reqs--;
5060 
5061 	crq.query_capability.capability =
5062 	    cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
5063 	ibmvnic_send_crq(adapter, &crq);
5064 	cap_reqs--;
5065 
5066 	crq.query_capability.capability =
5067 	    cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
5068 	ibmvnic_send_crq(adapter, &crq);
5069 	cap_reqs--;
5070 
5071 	crq.query_capability.capability =
5072 	    cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
5073 	ibmvnic_send_crq(adapter, &crq);
5074 	cap_reqs--;
5075 
5076 	crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
5077 	ibmvnic_send_crq(adapter, &crq);
5078 	cap_reqs--;
5079 
5080 	crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
5081 	ibmvnic_send_crq(adapter, &crq);
5082 	cap_reqs--;
5083 
5084 	crq.query_capability.capability = cpu_to_be16(MIN_MTU);
5085 	ibmvnic_send_crq(adapter, &crq);
5086 	cap_reqs--;
5087 
5088 	crq.query_capability.capability = cpu_to_be16(MAX_MTU);
5089 	ibmvnic_send_crq(adapter, &crq);
5090 	cap_reqs--;
5091 
5092 	crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
5093 	ibmvnic_send_crq(adapter, &crq);
5094 	cap_reqs--;
5095 
5096 	crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
5097 	ibmvnic_send_crq(adapter, &crq);
5098 	cap_reqs--;
5099 
5100 	crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
5101 	ibmvnic_send_crq(adapter, &crq);
5102 	cap_reqs--;
5103 
5104 	crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
5105 	ibmvnic_send_crq(adapter, &crq);
5106 	cap_reqs--;
5107 
5108 	crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
5109 	ibmvnic_send_crq(adapter, &crq);
5110 	cap_reqs--;
5111 
5112 	crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
5113 	ibmvnic_send_crq(adapter, &crq);
5114 	cap_reqs--;
5115 
5116 	crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
5117 	ibmvnic_send_crq(adapter, &crq);
5118 	cap_reqs--;
5119 
5120 	crq.query_capability.capability =
5121 			cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
5122 	ibmvnic_send_crq(adapter, &crq);
5123 	cap_reqs--;
5124 
5125 	crq.query_capability.capability =
5126 			cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
5127 	ibmvnic_send_crq(adapter, &crq);
5128 	cap_reqs--;
5129 
5130 	crq.query_capability.capability =
5131 			cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
5132 	ibmvnic_send_crq(adapter, &crq);
5133 	cap_reqs--;
5134 
5135 	crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
5136 
5137 	ibmvnic_send_crq(adapter, &crq);
5138 	cap_reqs--;
5139 
5140 	/* Keep at end to catch any discrepancy between expected and actual
5141 	 * CRQs sent.
5142 	 */
5143 	WARN_ON(cap_reqs != 0);
5144 }
5145 
5146 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
5147 {
5148 	int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
5149 	struct device *dev = &adapter->vdev->dev;
5150 	union ibmvnic_crq crq;
5151 
5152 	adapter->ip_offload_tok =
5153 		dma_map_single(dev,
5154 			       &adapter->ip_offload_buf,
5155 			       buf_sz,
5156 			       DMA_FROM_DEVICE);
5157 
5158 	if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
5159 		if (!firmware_has_feature(FW_FEATURE_CMO))
5160 			dev_err(dev, "Couldn't map offload buffer\n");
5161 		return;
5162 	}
5163 
5164 	memset(&crq, 0, sizeof(crq));
5165 	crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
5166 	crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
5167 	crq.query_ip_offload.len = cpu_to_be32(buf_sz);
5168 	crq.query_ip_offload.ioba =
5169 	    cpu_to_be32(adapter->ip_offload_tok);
5170 
5171 	ibmvnic_send_crq(adapter, &crq);
5172 }
5173 
5174 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
5175 {
5176 	struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
5177 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5178 	struct device *dev = &adapter->vdev->dev;
5179 	netdev_features_t old_hw_features = 0;
5180 	union ibmvnic_crq crq;
5181 
5182 	adapter->ip_offload_ctrl_tok =
5183 		dma_map_single(dev,
5184 			       ctrl_buf,
5185 			       sizeof(adapter->ip_offload_ctrl),
5186 			       DMA_TO_DEVICE);
5187 
5188 	if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
5189 		dev_err(dev, "Couldn't map ip offload control buffer\n");
5190 		return;
5191 	}
5192 
5193 	ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5194 	ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
5195 	ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
5196 	ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
5197 	ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
5198 	ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
5199 	ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
5200 	ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
5201 	ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
5202 	ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
5203 
5204 	/* large_rx disabled for now, additional features needed */
5205 	ctrl_buf->large_rx_ipv4 = 0;
5206 	ctrl_buf->large_rx_ipv6 = 0;
5207 
5208 	if (adapter->state != VNIC_PROBING) {
5209 		old_hw_features = adapter->netdev->hw_features;
5210 		adapter->netdev->hw_features = 0;
5211 	}
5212 
5213 	adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
5214 
5215 	if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
5216 		adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
5217 
5218 	if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
5219 		adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
5220 
5221 	if ((adapter->netdev->features &
5222 	    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
5223 		adapter->netdev->hw_features |= NETIF_F_RXCSUM;
5224 
5225 	if (buf->large_tx_ipv4)
5226 		adapter->netdev->hw_features |= NETIF_F_TSO;
5227 	if (buf->large_tx_ipv6)
5228 		adapter->netdev->hw_features |= NETIF_F_TSO6;
5229 
5230 	if (adapter->state == VNIC_PROBING) {
5231 		adapter->netdev->features |= adapter->netdev->hw_features;
5232 	} else if (old_hw_features != adapter->netdev->hw_features) {
5233 		netdev_features_t tmp = 0;
5234 
5235 		/* disable features no longer supported */
5236 		adapter->netdev->features &= adapter->netdev->hw_features;
5237 		/* turn on features now supported if previously enabled */
5238 		tmp = (old_hw_features ^ adapter->netdev->hw_features) &
5239 			adapter->netdev->hw_features;
5240 		adapter->netdev->features |=
5241 				tmp & adapter->netdev->wanted_features;
5242 	}
5243 
5244 	memset(&crq, 0, sizeof(crq));
5245 	crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
5246 	crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
5247 	crq.control_ip_offload.len =
5248 	    cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5249 	crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
5250 	ibmvnic_send_crq(adapter, &crq);
5251 }
5252 
5253 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
5254 				struct ibmvnic_adapter *adapter)
5255 {
5256 	struct device *dev = &adapter->vdev->dev;
5257 
5258 	if (crq->get_vpd_size_rsp.rc.code) {
5259 		dev_err(dev, "Error retrieving VPD size, rc=%x\n",
5260 			crq->get_vpd_size_rsp.rc.code);
5261 		complete(&adapter->fw_done);
5262 		return;
5263 	}
5264 
5265 	adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
5266 	complete(&adapter->fw_done);
5267 }
5268 
5269 static void handle_vpd_rsp(union ibmvnic_crq *crq,
5270 			   struct ibmvnic_adapter *adapter)
5271 {
5272 	struct device *dev = &adapter->vdev->dev;
5273 	unsigned char *substr = NULL;
5274 	u8 fw_level_len = 0;
5275 
5276 	memset(adapter->fw_version, 0, 32);
5277 
5278 	dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
5279 			 DMA_FROM_DEVICE);
5280 
5281 	if (crq->get_vpd_rsp.rc.code) {
5282 		dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
5283 			crq->get_vpd_rsp.rc.code);
5284 		goto complete;
5285 	}
5286 
5287 	/* get the position of the firmware version info
5288 	 * located after the ASCII 'RM' substring in the buffer
5289 	 */
5290 	substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
5291 	if (!substr) {
5292 		dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
5293 		goto complete;
5294 	}
5295 
5296 	/* get length of firmware level ASCII substring */
5297 	if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
5298 		fw_level_len = *(substr + 2);
5299 	} else {
5300 		dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
5301 		goto complete;
5302 	}
5303 
5304 	/* copy firmware version string from vpd into adapter */
5305 	if ((substr + 3 + fw_level_len) <
5306 	    (adapter->vpd->buff + adapter->vpd->len)) {
5307 		strscpy(adapter->fw_version, substr + 3,
5308 			sizeof(adapter->fw_version));
5309 	} else {
5310 		dev_info(dev, "FW substr extrapolated VPD buff\n");
5311 	}
5312 
5313 complete:
5314 	if (adapter->fw_version[0] == '\0')
5315 		strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
5316 	complete(&adapter->fw_done);
5317 }
5318 
5319 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
5320 {
5321 	struct device *dev = &adapter->vdev->dev;
5322 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5323 	int i;
5324 
5325 	dma_unmap_single(dev, adapter->ip_offload_tok,
5326 			 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
5327 
5328 	netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
5329 	for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
5330 		netdev_dbg(adapter->netdev, "%016lx\n",
5331 			   ((unsigned long *)(buf))[i]);
5332 
5333 	netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
5334 	netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
5335 	netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
5336 		   buf->tcp_ipv4_chksum);
5337 	netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
5338 		   buf->tcp_ipv6_chksum);
5339 	netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
5340 		   buf->udp_ipv4_chksum);
5341 	netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
5342 		   buf->udp_ipv6_chksum);
5343 	netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
5344 		   buf->large_tx_ipv4);
5345 	netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
5346 		   buf->large_tx_ipv6);
5347 	netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
5348 		   buf->large_rx_ipv4);
5349 	netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
5350 		   buf->large_rx_ipv6);
5351 	netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
5352 		   buf->max_ipv4_header_size);
5353 	netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
5354 		   buf->max_ipv6_header_size);
5355 	netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
5356 		   buf->max_tcp_header_size);
5357 	netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
5358 		   buf->max_udp_header_size);
5359 	netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
5360 		   buf->max_large_tx_size);
5361 	netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
5362 		   buf->max_large_rx_size);
5363 	netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
5364 		   buf->ipv6_extension_header);
5365 	netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
5366 		   buf->tcp_pseudosum_req);
5367 	netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
5368 		   buf->num_ipv6_ext_headers);
5369 	netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
5370 		   buf->off_ipv6_ext_headers);
5371 
5372 	send_control_ip_offload(adapter);
5373 }
5374 
5375 static const char *ibmvnic_fw_err_cause(u16 cause)
5376 {
5377 	switch (cause) {
5378 	case ADAPTER_PROBLEM:
5379 		return "adapter problem";
5380 	case BUS_PROBLEM:
5381 		return "bus problem";
5382 	case FW_PROBLEM:
5383 		return "firmware problem";
5384 	case DD_PROBLEM:
5385 		return "device driver problem";
5386 	case EEH_RECOVERY:
5387 		return "EEH recovery";
5388 	case FW_UPDATED:
5389 		return "firmware updated";
5390 	case LOW_MEMORY:
5391 		return "low Memory";
5392 	default:
5393 		return "unknown";
5394 	}
5395 }
5396 
5397 static void handle_error_indication(union ibmvnic_crq *crq,
5398 				    struct ibmvnic_adapter *adapter)
5399 {
5400 	struct device *dev = &adapter->vdev->dev;
5401 	u16 cause;
5402 
5403 	cause = be16_to_cpu(crq->error_indication.error_cause);
5404 
5405 	dev_warn_ratelimited(dev,
5406 			     "Firmware reports %serror, cause: %s. Starting recovery...\n",
5407 			     crq->error_indication.flags
5408 				& IBMVNIC_FATAL_ERROR ? "FATAL " : "",
5409 			     ibmvnic_fw_err_cause(cause));
5410 
5411 	if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
5412 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5413 	else
5414 		ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
5415 }
5416 
5417 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
5418 				 struct ibmvnic_adapter *adapter)
5419 {
5420 	struct net_device *netdev = adapter->netdev;
5421 	struct device *dev = &adapter->vdev->dev;
5422 	long rc;
5423 
5424 	rc = crq->change_mac_addr_rsp.rc.code;
5425 	if (rc) {
5426 		dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
5427 		goto out;
5428 	}
5429 	/* crq->change_mac_addr.mac_addr is the requested one
5430 	 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
5431 	 */
5432 	eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
5433 	ether_addr_copy(adapter->mac_addr,
5434 			&crq->change_mac_addr_rsp.mac_addr[0]);
5435 out:
5436 	complete(&adapter->fw_done);
5437 	return rc;
5438 }
5439 
5440 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
5441 				   struct ibmvnic_adapter *adapter)
5442 {
5443 	struct device *dev = &adapter->vdev->dev;
5444 	u64 *req_value;
5445 	char *name;
5446 
5447 	atomic_dec(&adapter->running_cap_crqs);
5448 	netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
5449 		   atomic_read(&adapter->running_cap_crqs));
5450 	switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
5451 	case REQ_TX_QUEUES:
5452 		req_value = &adapter->req_tx_queues;
5453 		name = "tx";
5454 		break;
5455 	case REQ_RX_QUEUES:
5456 		req_value = &adapter->req_rx_queues;
5457 		name = "rx";
5458 		break;
5459 	case REQ_RX_ADD_QUEUES:
5460 		req_value = &adapter->req_rx_add_queues;
5461 		name = "rx_add";
5462 		break;
5463 	case REQ_TX_ENTRIES_PER_SUBCRQ:
5464 		req_value = &adapter->req_tx_entries_per_subcrq;
5465 		name = "tx_entries_per_subcrq";
5466 		break;
5467 	case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
5468 		req_value = &adapter->req_rx_add_entries_per_subcrq;
5469 		name = "rx_add_entries_per_subcrq";
5470 		break;
5471 	case REQ_MTU:
5472 		req_value = &adapter->req_mtu;
5473 		name = "mtu";
5474 		break;
5475 	case PROMISC_REQUESTED:
5476 		req_value = &adapter->promisc;
5477 		name = "promisc";
5478 		break;
5479 	default:
5480 		dev_err(dev, "Got invalid cap request rsp %d\n",
5481 			crq->request_capability.capability);
5482 		return;
5483 	}
5484 
5485 	switch (crq->request_capability_rsp.rc.code) {
5486 	case SUCCESS:
5487 		break;
5488 	case PARTIALSUCCESS:
5489 		dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
5490 			 *req_value,
5491 			 (long)be64_to_cpu(crq->request_capability_rsp.number),
5492 			 name);
5493 
5494 		if (be16_to_cpu(crq->request_capability_rsp.capability) ==
5495 		    REQ_MTU) {
5496 			pr_err("mtu of %llu is not supported. Reverting.\n",
5497 			       *req_value);
5498 			*req_value = adapter->fallback.mtu;
5499 		} else {
5500 			*req_value =
5501 				be64_to_cpu(crq->request_capability_rsp.number);
5502 		}
5503 
5504 		send_request_cap(adapter, 1);
5505 		return;
5506 	default:
5507 		dev_err(dev, "Error %d in request cap rsp\n",
5508 			crq->request_capability_rsp.rc.code);
5509 		return;
5510 	}
5511 
5512 	/* Done receiving requested capabilities, query IP offload support */
5513 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5514 		send_query_ip_offload(adapter);
5515 }
5516 
5517 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
5518 			    struct ibmvnic_adapter *adapter)
5519 {
5520 	struct device *dev = &adapter->vdev->dev;
5521 	struct net_device *netdev = adapter->netdev;
5522 	struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
5523 	struct ibmvnic_login_buffer *login = adapter->login_buf;
5524 	u64 *tx_handle_array;
5525 	u64 *rx_handle_array;
5526 	int num_tx_pools;
5527 	int num_rx_pools;
5528 	u64 *size_array;
5529 	u32 rsp_len;
5530 	int i;
5531 
5532 	/* CHECK: Test/set of login_pending does not need to be atomic
5533 	 * because only ibmvnic_tasklet tests/clears this.
5534 	 */
5535 	if (!adapter->login_pending) {
5536 		netdev_warn(netdev, "Ignoring unexpected login response\n");
5537 		return 0;
5538 	}
5539 	adapter->login_pending = false;
5540 
5541 	/* If the number of queues requested can't be allocated by the
5542 	 * server, the login response will return with code 1. We will need
5543 	 * to resend the login buffer with fewer queues requested.
5544 	 */
5545 	if (login_rsp_crq->generic.rc.code) {
5546 		adapter->init_done_rc = login_rsp_crq->generic.rc.code;
5547 		complete(&adapter->init_done);
5548 		return 0;
5549 	}
5550 
5551 	if (adapter->failover_pending) {
5552 		adapter->init_done_rc = -EAGAIN;
5553 		netdev_dbg(netdev, "Failover pending, ignoring login response\n");
5554 		complete(&adapter->init_done);
5555 		/* login response buffer will be released on reset */
5556 		return 0;
5557 	}
5558 
5559 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
5560 
5561 	netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
5562 	for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
5563 		netdev_dbg(adapter->netdev, "%016lx\n",
5564 			   ((unsigned long *)(adapter->login_rsp_buf))[i]);
5565 	}
5566 
5567 	/* Sanity checks */
5568 	if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
5569 	    (be32_to_cpu(login->num_rxcomp_subcrqs) *
5570 	     adapter->req_rx_add_queues !=
5571 	     be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
5572 		dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
5573 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5574 		return -EIO;
5575 	}
5576 
5577 	rsp_len = be32_to_cpu(login_rsp->len);
5578 	if (be32_to_cpu(login->login_rsp_len) < rsp_len ||
5579 	    rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) ||
5580 	    rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) ||
5581 	    rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) ||
5582 	    rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) {
5583 		/* This can happen if a login request times out and there are
5584 		 * 2 outstanding login requests sent, the LOGIN_RSP crq
5585 		 * could have been for the older login request. So we are
5586 		 * parsing the newer response buffer which may be incomplete
5587 		 */
5588 		dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n");
5589 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5590 		return -EIO;
5591 	}
5592 
5593 	size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5594 		be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
5595 	/* variable buffer sizes are not supported, so just read the
5596 	 * first entry.
5597 	 */
5598 	adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
5599 
5600 	num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
5601 	num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
5602 
5603 	tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5604 				  be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
5605 	rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5606 				  be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
5607 
5608 	for (i = 0; i < num_tx_pools; i++)
5609 		adapter->tx_scrq[i]->handle = tx_handle_array[i];
5610 
5611 	for (i = 0; i < num_rx_pools; i++)
5612 		adapter->rx_scrq[i]->handle = rx_handle_array[i];
5613 
5614 	adapter->num_active_tx_scrqs = num_tx_pools;
5615 	adapter->num_active_rx_scrqs = num_rx_pools;
5616 	release_login_rsp_buffer(adapter);
5617 	release_login_buffer(adapter);
5618 	complete(&adapter->init_done);
5619 
5620 	return 0;
5621 }
5622 
5623 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
5624 				     struct ibmvnic_adapter *adapter)
5625 {
5626 	struct device *dev = &adapter->vdev->dev;
5627 	long rc;
5628 
5629 	rc = crq->request_unmap_rsp.rc.code;
5630 	if (rc)
5631 		dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
5632 }
5633 
5634 static void handle_query_map_rsp(union ibmvnic_crq *crq,
5635 				 struct ibmvnic_adapter *adapter)
5636 {
5637 	struct net_device *netdev = adapter->netdev;
5638 	struct device *dev = &adapter->vdev->dev;
5639 	long rc;
5640 
5641 	rc = crq->query_map_rsp.rc.code;
5642 	if (rc) {
5643 		dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5644 		return;
5645 	}
5646 	netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5647 		   crq->query_map_rsp.page_size,
5648 		   __be32_to_cpu(crq->query_map_rsp.tot_pages),
5649 		   __be32_to_cpu(crq->query_map_rsp.free_pages));
5650 }
5651 
5652 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5653 				 struct ibmvnic_adapter *adapter)
5654 {
5655 	struct net_device *netdev = adapter->netdev;
5656 	struct device *dev = &adapter->vdev->dev;
5657 	long rc;
5658 
5659 	atomic_dec(&adapter->running_cap_crqs);
5660 	netdev_dbg(netdev, "Outstanding queries: %d\n",
5661 		   atomic_read(&adapter->running_cap_crqs));
5662 	rc = crq->query_capability.rc.code;
5663 	if (rc) {
5664 		dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5665 		goto out;
5666 	}
5667 
5668 	switch (be16_to_cpu(crq->query_capability.capability)) {
5669 	case MIN_TX_QUEUES:
5670 		adapter->min_tx_queues =
5671 		    be64_to_cpu(crq->query_capability.number);
5672 		netdev_dbg(netdev, "min_tx_queues = %lld\n",
5673 			   adapter->min_tx_queues);
5674 		break;
5675 	case MIN_RX_QUEUES:
5676 		adapter->min_rx_queues =
5677 		    be64_to_cpu(crq->query_capability.number);
5678 		netdev_dbg(netdev, "min_rx_queues = %lld\n",
5679 			   adapter->min_rx_queues);
5680 		break;
5681 	case MIN_RX_ADD_QUEUES:
5682 		adapter->min_rx_add_queues =
5683 		    be64_to_cpu(crq->query_capability.number);
5684 		netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5685 			   adapter->min_rx_add_queues);
5686 		break;
5687 	case MAX_TX_QUEUES:
5688 		adapter->max_tx_queues =
5689 		    be64_to_cpu(crq->query_capability.number);
5690 		netdev_dbg(netdev, "max_tx_queues = %lld\n",
5691 			   adapter->max_tx_queues);
5692 		break;
5693 	case MAX_RX_QUEUES:
5694 		adapter->max_rx_queues =
5695 		    be64_to_cpu(crq->query_capability.number);
5696 		netdev_dbg(netdev, "max_rx_queues = %lld\n",
5697 			   adapter->max_rx_queues);
5698 		break;
5699 	case MAX_RX_ADD_QUEUES:
5700 		adapter->max_rx_add_queues =
5701 		    be64_to_cpu(crq->query_capability.number);
5702 		netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5703 			   adapter->max_rx_add_queues);
5704 		break;
5705 	case MIN_TX_ENTRIES_PER_SUBCRQ:
5706 		adapter->min_tx_entries_per_subcrq =
5707 		    be64_to_cpu(crq->query_capability.number);
5708 		netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5709 			   adapter->min_tx_entries_per_subcrq);
5710 		break;
5711 	case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5712 		adapter->min_rx_add_entries_per_subcrq =
5713 		    be64_to_cpu(crq->query_capability.number);
5714 		netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5715 			   adapter->min_rx_add_entries_per_subcrq);
5716 		break;
5717 	case MAX_TX_ENTRIES_PER_SUBCRQ:
5718 		adapter->max_tx_entries_per_subcrq =
5719 		    be64_to_cpu(crq->query_capability.number);
5720 		netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5721 			   adapter->max_tx_entries_per_subcrq);
5722 		break;
5723 	case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5724 		adapter->max_rx_add_entries_per_subcrq =
5725 		    be64_to_cpu(crq->query_capability.number);
5726 		netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5727 			   adapter->max_rx_add_entries_per_subcrq);
5728 		break;
5729 	case TCP_IP_OFFLOAD:
5730 		adapter->tcp_ip_offload =
5731 		    be64_to_cpu(crq->query_capability.number);
5732 		netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5733 			   adapter->tcp_ip_offload);
5734 		break;
5735 	case PROMISC_SUPPORTED:
5736 		adapter->promisc_supported =
5737 		    be64_to_cpu(crq->query_capability.number);
5738 		netdev_dbg(netdev, "promisc_supported = %lld\n",
5739 			   adapter->promisc_supported);
5740 		break;
5741 	case MIN_MTU:
5742 		adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5743 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5744 		netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5745 		break;
5746 	case MAX_MTU:
5747 		adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5748 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5749 		netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5750 		break;
5751 	case MAX_MULTICAST_FILTERS:
5752 		adapter->max_multicast_filters =
5753 		    be64_to_cpu(crq->query_capability.number);
5754 		netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5755 			   adapter->max_multicast_filters);
5756 		break;
5757 	case VLAN_HEADER_INSERTION:
5758 		adapter->vlan_header_insertion =
5759 		    be64_to_cpu(crq->query_capability.number);
5760 		if (adapter->vlan_header_insertion)
5761 			netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5762 		netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5763 			   adapter->vlan_header_insertion);
5764 		break;
5765 	case RX_VLAN_HEADER_INSERTION:
5766 		adapter->rx_vlan_header_insertion =
5767 		    be64_to_cpu(crq->query_capability.number);
5768 		netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5769 			   adapter->rx_vlan_header_insertion);
5770 		break;
5771 	case MAX_TX_SG_ENTRIES:
5772 		adapter->max_tx_sg_entries =
5773 		    be64_to_cpu(crq->query_capability.number);
5774 		netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5775 			   adapter->max_tx_sg_entries);
5776 		break;
5777 	case RX_SG_SUPPORTED:
5778 		adapter->rx_sg_supported =
5779 		    be64_to_cpu(crq->query_capability.number);
5780 		netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5781 			   adapter->rx_sg_supported);
5782 		break;
5783 	case OPT_TX_COMP_SUB_QUEUES:
5784 		adapter->opt_tx_comp_sub_queues =
5785 		    be64_to_cpu(crq->query_capability.number);
5786 		netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5787 			   adapter->opt_tx_comp_sub_queues);
5788 		break;
5789 	case OPT_RX_COMP_QUEUES:
5790 		adapter->opt_rx_comp_queues =
5791 		    be64_to_cpu(crq->query_capability.number);
5792 		netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5793 			   adapter->opt_rx_comp_queues);
5794 		break;
5795 	case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5796 		adapter->opt_rx_bufadd_q_per_rx_comp_q =
5797 		    be64_to_cpu(crq->query_capability.number);
5798 		netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5799 			   adapter->opt_rx_bufadd_q_per_rx_comp_q);
5800 		break;
5801 	case OPT_TX_ENTRIES_PER_SUBCRQ:
5802 		adapter->opt_tx_entries_per_subcrq =
5803 		    be64_to_cpu(crq->query_capability.number);
5804 		netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5805 			   adapter->opt_tx_entries_per_subcrq);
5806 		break;
5807 	case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5808 		adapter->opt_rxba_entries_per_subcrq =
5809 		    be64_to_cpu(crq->query_capability.number);
5810 		netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5811 			   adapter->opt_rxba_entries_per_subcrq);
5812 		break;
5813 	case TX_RX_DESC_REQ:
5814 		adapter->tx_rx_desc_req = crq->query_capability.number;
5815 		netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5816 			   adapter->tx_rx_desc_req);
5817 		break;
5818 
5819 	default:
5820 		netdev_err(netdev, "Got invalid cap rsp %d\n",
5821 			   crq->query_capability.capability);
5822 	}
5823 
5824 out:
5825 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5826 		send_request_cap(adapter, 0);
5827 }
5828 
5829 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5830 {
5831 	union ibmvnic_crq crq;
5832 	int rc;
5833 
5834 	memset(&crq, 0, sizeof(crq));
5835 	crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5836 	crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5837 
5838 	mutex_lock(&adapter->fw_lock);
5839 	adapter->fw_done_rc = 0;
5840 	reinit_completion(&adapter->fw_done);
5841 
5842 	rc = ibmvnic_send_crq(adapter, &crq);
5843 	if (rc) {
5844 		mutex_unlock(&adapter->fw_lock);
5845 		return rc;
5846 	}
5847 
5848 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5849 	if (rc) {
5850 		mutex_unlock(&adapter->fw_lock);
5851 		return rc;
5852 	}
5853 
5854 	mutex_unlock(&adapter->fw_lock);
5855 	return adapter->fw_done_rc ? -EIO : 0;
5856 }
5857 
5858 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5859 				       struct ibmvnic_adapter *adapter)
5860 {
5861 	struct net_device *netdev = adapter->netdev;
5862 	int rc;
5863 	__be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5864 
5865 	rc = crq->query_phys_parms_rsp.rc.code;
5866 	if (rc) {
5867 		netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5868 		return rc;
5869 	}
5870 	switch (rspeed) {
5871 	case IBMVNIC_10MBPS:
5872 		adapter->speed = SPEED_10;
5873 		break;
5874 	case IBMVNIC_100MBPS:
5875 		adapter->speed = SPEED_100;
5876 		break;
5877 	case IBMVNIC_1GBPS:
5878 		adapter->speed = SPEED_1000;
5879 		break;
5880 	case IBMVNIC_10GBPS:
5881 		adapter->speed = SPEED_10000;
5882 		break;
5883 	case IBMVNIC_25GBPS:
5884 		adapter->speed = SPEED_25000;
5885 		break;
5886 	case IBMVNIC_40GBPS:
5887 		adapter->speed = SPEED_40000;
5888 		break;
5889 	case IBMVNIC_50GBPS:
5890 		adapter->speed = SPEED_50000;
5891 		break;
5892 	case IBMVNIC_100GBPS:
5893 		adapter->speed = SPEED_100000;
5894 		break;
5895 	case IBMVNIC_200GBPS:
5896 		adapter->speed = SPEED_200000;
5897 		break;
5898 	default:
5899 		if (netif_carrier_ok(netdev))
5900 			netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5901 		adapter->speed = SPEED_UNKNOWN;
5902 	}
5903 	if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5904 		adapter->duplex = DUPLEX_FULL;
5905 	else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5906 		adapter->duplex = DUPLEX_HALF;
5907 	else
5908 		adapter->duplex = DUPLEX_UNKNOWN;
5909 
5910 	return rc;
5911 }
5912 
5913 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5914 			       struct ibmvnic_adapter *adapter)
5915 {
5916 	struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5917 	struct net_device *netdev = adapter->netdev;
5918 	struct device *dev = &adapter->vdev->dev;
5919 	u64 *u64_crq = (u64 *)crq;
5920 	long rc;
5921 
5922 	netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5923 		   (unsigned long)cpu_to_be64(u64_crq[0]),
5924 		   (unsigned long)cpu_to_be64(u64_crq[1]));
5925 	switch (gen_crq->first) {
5926 	case IBMVNIC_CRQ_INIT_RSP:
5927 		switch (gen_crq->cmd) {
5928 		case IBMVNIC_CRQ_INIT:
5929 			dev_info(dev, "Partner initialized\n");
5930 			adapter->from_passive_init = true;
5931 			/* Discard any stale login responses from prev reset.
5932 			 * CHECK: should we clear even on INIT_COMPLETE?
5933 			 */
5934 			adapter->login_pending = false;
5935 
5936 			if (adapter->state == VNIC_DOWN)
5937 				rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5938 			else
5939 				rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5940 
5941 			if (rc && rc != -EBUSY) {
5942 				/* We were unable to schedule the failover
5943 				 * reset either because the adapter was still
5944 				 * probing (eg: during kexec) or we could not
5945 				 * allocate memory. Clear the failover_pending
5946 				 * flag since no one else will. We ignore
5947 				 * EBUSY because it means either FAILOVER reset
5948 				 * is already scheduled or the adapter is
5949 				 * being removed.
5950 				 */
5951 				netdev_err(netdev,
5952 					   "Error %ld scheduling failover reset\n",
5953 					   rc);
5954 				adapter->failover_pending = false;
5955 			}
5956 
5957 			if (!completion_done(&adapter->init_done)) {
5958 				if (!adapter->init_done_rc)
5959 					adapter->init_done_rc = -EAGAIN;
5960 				complete(&adapter->init_done);
5961 			}
5962 
5963 			break;
5964 		case IBMVNIC_CRQ_INIT_COMPLETE:
5965 			dev_info(dev, "Partner initialization complete\n");
5966 			adapter->crq.active = true;
5967 			send_version_xchg(adapter);
5968 			break;
5969 		default:
5970 			dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5971 		}
5972 		return;
5973 	case IBMVNIC_CRQ_XPORT_EVENT:
5974 		netif_carrier_off(netdev);
5975 		adapter->crq.active = false;
5976 		/* terminate any thread waiting for a response
5977 		 * from the device
5978 		 */
5979 		if (!completion_done(&adapter->fw_done)) {
5980 			adapter->fw_done_rc = -EIO;
5981 			complete(&adapter->fw_done);
5982 		}
5983 
5984 		/* if we got here during crq-init, retry crq-init */
5985 		if (!completion_done(&adapter->init_done)) {
5986 			adapter->init_done_rc = -EAGAIN;
5987 			complete(&adapter->init_done);
5988 		}
5989 
5990 		if (!completion_done(&adapter->stats_done))
5991 			complete(&adapter->stats_done);
5992 		if (test_bit(0, &adapter->resetting))
5993 			adapter->force_reset_recovery = true;
5994 		if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5995 			dev_info(dev, "Migrated, re-enabling adapter\n");
5996 			ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5997 		} else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5998 			dev_info(dev, "Backing device failover detected\n");
5999 			adapter->failover_pending = true;
6000 		} else {
6001 			/* The adapter lost the connection */
6002 			dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
6003 				gen_crq->cmd);
6004 			ibmvnic_reset(adapter, VNIC_RESET_FATAL);
6005 		}
6006 		return;
6007 	case IBMVNIC_CRQ_CMD_RSP:
6008 		break;
6009 	default:
6010 		dev_err(dev, "Got an invalid msg type 0x%02x\n",
6011 			gen_crq->first);
6012 		return;
6013 	}
6014 
6015 	switch (gen_crq->cmd) {
6016 	case VERSION_EXCHANGE_RSP:
6017 		rc = crq->version_exchange_rsp.rc.code;
6018 		if (rc) {
6019 			dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
6020 			break;
6021 		}
6022 		ibmvnic_version =
6023 			    be16_to_cpu(crq->version_exchange_rsp.version);
6024 		dev_info(dev, "Partner protocol version is %d\n",
6025 			 ibmvnic_version);
6026 		send_query_cap(adapter);
6027 		break;
6028 	case QUERY_CAPABILITY_RSP:
6029 		handle_query_cap_rsp(crq, adapter);
6030 		break;
6031 	case QUERY_MAP_RSP:
6032 		handle_query_map_rsp(crq, adapter);
6033 		break;
6034 	case REQUEST_MAP_RSP:
6035 		adapter->fw_done_rc = crq->request_map_rsp.rc.code;
6036 		complete(&adapter->fw_done);
6037 		break;
6038 	case REQUEST_UNMAP_RSP:
6039 		handle_request_unmap_rsp(crq, adapter);
6040 		break;
6041 	case REQUEST_CAPABILITY_RSP:
6042 		handle_request_cap_rsp(crq, adapter);
6043 		break;
6044 	case LOGIN_RSP:
6045 		netdev_dbg(netdev, "Got Login Response\n");
6046 		handle_login_rsp(crq, adapter);
6047 		break;
6048 	case LOGICAL_LINK_STATE_RSP:
6049 		netdev_dbg(netdev,
6050 			   "Got Logical Link State Response, state: %d rc: %d\n",
6051 			   crq->logical_link_state_rsp.link_state,
6052 			   crq->logical_link_state_rsp.rc.code);
6053 		adapter->logical_link_state =
6054 		    crq->logical_link_state_rsp.link_state;
6055 		adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
6056 		complete(&adapter->init_done);
6057 		break;
6058 	case LINK_STATE_INDICATION:
6059 		netdev_dbg(netdev, "Got Logical Link State Indication\n");
6060 		adapter->phys_link_state =
6061 		    crq->link_state_indication.phys_link_state;
6062 		adapter->logical_link_state =
6063 		    crq->link_state_indication.logical_link_state;
6064 		if (adapter->phys_link_state && adapter->logical_link_state)
6065 			netif_carrier_on(netdev);
6066 		else
6067 			netif_carrier_off(netdev);
6068 		break;
6069 	case CHANGE_MAC_ADDR_RSP:
6070 		netdev_dbg(netdev, "Got MAC address change Response\n");
6071 		adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
6072 		break;
6073 	case ERROR_INDICATION:
6074 		netdev_dbg(netdev, "Got Error Indication\n");
6075 		handle_error_indication(crq, adapter);
6076 		break;
6077 	case REQUEST_STATISTICS_RSP:
6078 		netdev_dbg(netdev, "Got Statistics Response\n");
6079 		complete(&adapter->stats_done);
6080 		break;
6081 	case QUERY_IP_OFFLOAD_RSP:
6082 		netdev_dbg(netdev, "Got Query IP offload Response\n");
6083 		handle_query_ip_offload_rsp(adapter);
6084 		break;
6085 	case MULTICAST_CTRL_RSP:
6086 		netdev_dbg(netdev, "Got multicast control Response\n");
6087 		break;
6088 	case CONTROL_IP_OFFLOAD_RSP:
6089 		netdev_dbg(netdev, "Got Control IP offload Response\n");
6090 		dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
6091 				 sizeof(adapter->ip_offload_ctrl),
6092 				 DMA_TO_DEVICE);
6093 		complete(&adapter->init_done);
6094 		break;
6095 	case COLLECT_FW_TRACE_RSP:
6096 		netdev_dbg(netdev, "Got Collect firmware trace Response\n");
6097 		complete(&adapter->fw_done);
6098 		break;
6099 	case GET_VPD_SIZE_RSP:
6100 		handle_vpd_size_rsp(crq, adapter);
6101 		break;
6102 	case GET_VPD_RSP:
6103 		handle_vpd_rsp(crq, adapter);
6104 		break;
6105 	case QUERY_PHYS_PARMS_RSP:
6106 		adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
6107 		complete(&adapter->fw_done);
6108 		break;
6109 	default:
6110 		netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
6111 			   gen_crq->cmd);
6112 	}
6113 }
6114 
6115 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
6116 {
6117 	struct ibmvnic_adapter *adapter = instance;
6118 
6119 	tasklet_schedule(&adapter->tasklet);
6120 	return IRQ_HANDLED;
6121 }
6122 
6123 static void ibmvnic_tasklet(struct tasklet_struct *t)
6124 {
6125 	struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
6126 	struct ibmvnic_crq_queue *queue = &adapter->crq;
6127 	union ibmvnic_crq *crq;
6128 	unsigned long flags;
6129 
6130 	spin_lock_irqsave(&queue->lock, flags);
6131 
6132 	/* Pull all the valid messages off the CRQ */
6133 	while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
6134 		/* This barrier makes sure ibmvnic_next_crq()'s
6135 		 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
6136 		 * before ibmvnic_handle_crq()'s
6137 		 * switch(gen_crq->first) and switch(gen_crq->cmd).
6138 		 */
6139 		dma_rmb();
6140 		ibmvnic_handle_crq(crq, adapter);
6141 		crq->generic.first = 0;
6142 	}
6143 
6144 	spin_unlock_irqrestore(&queue->lock, flags);
6145 }
6146 
6147 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
6148 {
6149 	struct vio_dev *vdev = adapter->vdev;
6150 	int rc;
6151 
6152 	do {
6153 		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
6154 	} while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
6155 
6156 	if (rc)
6157 		dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
6158 
6159 	return rc;
6160 }
6161 
6162 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
6163 {
6164 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6165 	struct device *dev = &adapter->vdev->dev;
6166 	struct vio_dev *vdev = adapter->vdev;
6167 	int rc;
6168 
6169 	/* Close the CRQ */
6170 	do {
6171 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6172 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6173 
6174 	/* Clean out the queue */
6175 	if (!crq->msgs)
6176 		return -EINVAL;
6177 
6178 	memset(crq->msgs, 0, PAGE_SIZE);
6179 	crq->cur = 0;
6180 	crq->active = false;
6181 
6182 	/* And re-open it again */
6183 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6184 				crq->msg_token, PAGE_SIZE);
6185 
6186 	if (rc == H_CLOSED)
6187 		/* Adapter is good, but other end is not ready */
6188 		dev_warn(dev, "Partner adapter not ready\n");
6189 	else if (rc != 0)
6190 		dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
6191 
6192 	return rc;
6193 }
6194 
6195 static void release_crq_queue(struct ibmvnic_adapter *adapter)
6196 {
6197 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6198 	struct vio_dev *vdev = adapter->vdev;
6199 	long rc;
6200 
6201 	if (!crq->msgs)
6202 		return;
6203 
6204 	netdev_dbg(adapter->netdev, "Releasing CRQ\n");
6205 	free_irq(vdev->irq, adapter);
6206 	tasklet_kill(&adapter->tasklet);
6207 	do {
6208 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6209 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6210 
6211 	dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
6212 			 DMA_BIDIRECTIONAL);
6213 	free_page((unsigned long)crq->msgs);
6214 	crq->msgs = NULL;
6215 	crq->active = false;
6216 }
6217 
6218 static int init_crq_queue(struct ibmvnic_adapter *adapter)
6219 {
6220 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6221 	struct device *dev = &adapter->vdev->dev;
6222 	struct vio_dev *vdev = adapter->vdev;
6223 	int rc, retrc = -ENOMEM;
6224 
6225 	if (crq->msgs)
6226 		return 0;
6227 
6228 	crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
6229 	/* Should we allocate more than one page? */
6230 
6231 	if (!crq->msgs)
6232 		return -ENOMEM;
6233 
6234 	crq->size = PAGE_SIZE / sizeof(*crq->msgs);
6235 	crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
6236 					DMA_BIDIRECTIONAL);
6237 	if (dma_mapping_error(dev, crq->msg_token))
6238 		goto map_failed;
6239 
6240 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6241 				crq->msg_token, PAGE_SIZE);
6242 
6243 	if (rc == H_RESOURCE)
6244 		/* maybe kexecing and resource is busy. try a reset */
6245 		rc = ibmvnic_reset_crq(adapter);
6246 	retrc = rc;
6247 
6248 	if (rc == H_CLOSED) {
6249 		dev_warn(dev, "Partner adapter not ready\n");
6250 	} else if (rc) {
6251 		dev_warn(dev, "Error %d opening adapter\n", rc);
6252 		goto reg_crq_failed;
6253 	}
6254 
6255 	retrc = 0;
6256 
6257 	tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
6258 
6259 	netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
6260 	snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
6261 		 adapter->vdev->unit_address);
6262 	rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
6263 	if (rc) {
6264 		dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
6265 			vdev->irq, rc);
6266 		goto req_irq_failed;
6267 	}
6268 
6269 	rc = vio_enable_interrupts(vdev);
6270 	if (rc) {
6271 		dev_err(dev, "Error %d enabling interrupts\n", rc);
6272 		goto req_irq_failed;
6273 	}
6274 
6275 	crq->cur = 0;
6276 	spin_lock_init(&crq->lock);
6277 
6278 	/* process any CRQs that were queued before we enabled interrupts */
6279 	tasklet_schedule(&adapter->tasklet);
6280 
6281 	return retrc;
6282 
6283 req_irq_failed:
6284 	tasklet_kill(&adapter->tasklet);
6285 	do {
6286 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6287 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6288 reg_crq_failed:
6289 	dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
6290 map_failed:
6291 	free_page((unsigned long)crq->msgs);
6292 	crq->msgs = NULL;
6293 	return retrc;
6294 }
6295 
6296 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
6297 {
6298 	struct device *dev = &adapter->vdev->dev;
6299 	unsigned long timeout = msecs_to_jiffies(20000);
6300 	u64 old_num_rx_queues = adapter->req_rx_queues;
6301 	u64 old_num_tx_queues = adapter->req_tx_queues;
6302 	int rc;
6303 
6304 	adapter->from_passive_init = false;
6305 
6306 	rc = ibmvnic_send_crq_init(adapter);
6307 	if (rc) {
6308 		dev_err(dev, "Send crq init failed with error %d\n", rc);
6309 		return rc;
6310 	}
6311 
6312 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
6313 		dev_err(dev, "Initialization sequence timed out\n");
6314 		return -ETIMEDOUT;
6315 	}
6316 
6317 	if (adapter->init_done_rc) {
6318 		release_crq_queue(adapter);
6319 		dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
6320 		return adapter->init_done_rc;
6321 	}
6322 
6323 	if (adapter->from_passive_init) {
6324 		adapter->state = VNIC_OPEN;
6325 		adapter->from_passive_init = false;
6326 		dev_err(dev, "CRQ-init failed, passive-init\n");
6327 		return -EINVAL;
6328 	}
6329 
6330 	if (reset &&
6331 	    test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
6332 	    adapter->reset_reason != VNIC_RESET_MOBILITY) {
6333 		if (adapter->req_rx_queues != old_num_rx_queues ||
6334 		    adapter->req_tx_queues != old_num_tx_queues) {
6335 			release_sub_crqs(adapter, 0);
6336 			rc = init_sub_crqs(adapter);
6337 		} else {
6338 			/* no need to reinitialize completely, but we do
6339 			 * need to clean up transmits that were in flight
6340 			 * when we processed the reset.  Failure to do so
6341 			 * will confound the upper layer, usually TCP, by
6342 			 * creating the illusion of transmits that are
6343 			 * awaiting completion.
6344 			 */
6345 			clean_tx_pools(adapter);
6346 
6347 			rc = reset_sub_crq_queues(adapter);
6348 		}
6349 	} else {
6350 		rc = init_sub_crqs(adapter);
6351 	}
6352 
6353 	if (rc) {
6354 		dev_err(dev, "Initialization of sub crqs failed\n");
6355 		release_crq_queue(adapter);
6356 		return rc;
6357 	}
6358 
6359 	rc = init_sub_crq_irqs(adapter);
6360 	if (rc) {
6361 		dev_err(dev, "Failed to initialize sub crq irqs\n");
6362 		release_crq_queue(adapter);
6363 	}
6364 
6365 	return rc;
6366 }
6367 
6368 static struct device_attribute dev_attr_failover;
6369 
6370 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
6371 {
6372 	struct ibmvnic_adapter *adapter;
6373 	struct net_device *netdev;
6374 	unsigned char *mac_addr_p;
6375 	unsigned long flags;
6376 	bool init_success;
6377 	int rc;
6378 
6379 	dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
6380 		dev->unit_address);
6381 
6382 	mac_addr_p = (unsigned char *)vio_get_attribute(dev,
6383 							VETH_MAC_ADDR, NULL);
6384 	if (!mac_addr_p) {
6385 		dev_err(&dev->dev,
6386 			"(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
6387 			__FILE__, __LINE__);
6388 		return 0;
6389 	}
6390 
6391 	netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
6392 				   IBMVNIC_MAX_QUEUES);
6393 	if (!netdev)
6394 		return -ENOMEM;
6395 
6396 	adapter = netdev_priv(netdev);
6397 	adapter->state = VNIC_PROBING;
6398 	dev_set_drvdata(&dev->dev, netdev);
6399 	adapter->vdev = dev;
6400 	adapter->netdev = netdev;
6401 	adapter->login_pending = false;
6402 	memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
6403 	/* map_ids start at 1, so ensure map_id 0 is always "in-use" */
6404 	bitmap_set(adapter->map_ids, 0, 1);
6405 
6406 	ether_addr_copy(adapter->mac_addr, mac_addr_p);
6407 	eth_hw_addr_set(netdev, adapter->mac_addr);
6408 	netdev->irq = dev->irq;
6409 	netdev->netdev_ops = &ibmvnic_netdev_ops;
6410 	netdev->ethtool_ops = &ibmvnic_ethtool_ops;
6411 	SET_NETDEV_DEV(netdev, &dev->dev);
6412 
6413 	INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
6414 	INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
6415 			  __ibmvnic_delayed_reset);
6416 	INIT_LIST_HEAD(&adapter->rwi_list);
6417 	spin_lock_init(&adapter->rwi_lock);
6418 	spin_lock_init(&adapter->state_lock);
6419 	mutex_init(&adapter->fw_lock);
6420 	init_completion(&adapter->probe_done);
6421 	init_completion(&adapter->init_done);
6422 	init_completion(&adapter->fw_done);
6423 	init_completion(&adapter->reset_done);
6424 	init_completion(&adapter->stats_done);
6425 	clear_bit(0, &adapter->resetting);
6426 	adapter->prev_rx_buf_sz = 0;
6427 	adapter->prev_mtu = 0;
6428 
6429 	init_success = false;
6430 	do {
6431 		reinit_init_done(adapter);
6432 
6433 		/* clear any failovers we got in the previous pass
6434 		 * since we are reinitializing the CRQ
6435 		 */
6436 		adapter->failover_pending = false;
6437 
6438 		/* If we had already initialized CRQ, we may have one or
6439 		 * more resets queued already. Discard those and release
6440 		 * the CRQ before initializing the CRQ again.
6441 		 */
6442 		release_crq_queue(adapter);
6443 
6444 		/* Since we are still in PROBING state, __ibmvnic_reset()
6445 		 * will not access the ->rwi_list and since we released CRQ,
6446 		 * we won't get _new_ transport events. But there maybe an
6447 		 * ongoing ibmvnic_reset() call. So serialize access to
6448 		 * rwi_list. If we win the race, ibvmnic_reset() could add
6449 		 * a reset after we purged but thats ok - we just may end
6450 		 * up with an extra reset (i.e similar to having two or more
6451 		 * resets in the queue at once).
6452 		 * CHECK.
6453 		 */
6454 		spin_lock_irqsave(&adapter->rwi_lock, flags);
6455 		flush_reset_queue(adapter);
6456 		spin_unlock_irqrestore(&adapter->rwi_lock, flags);
6457 
6458 		rc = init_crq_queue(adapter);
6459 		if (rc) {
6460 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
6461 				rc);
6462 			goto ibmvnic_init_fail;
6463 		}
6464 
6465 		rc = ibmvnic_reset_init(adapter, false);
6466 	} while (rc == -EAGAIN);
6467 
6468 	/* We are ignoring the error from ibmvnic_reset_init() assuming that the
6469 	 * partner is not ready. CRQ is not active. When the partner becomes
6470 	 * ready, we will do the passive init reset.
6471 	 */
6472 
6473 	if (!rc)
6474 		init_success = true;
6475 
6476 	rc = init_stats_buffers(adapter);
6477 	if (rc)
6478 		goto ibmvnic_init_fail;
6479 
6480 	rc = init_stats_token(adapter);
6481 	if (rc)
6482 		goto ibmvnic_stats_fail;
6483 
6484 	rc = device_create_file(&dev->dev, &dev_attr_failover);
6485 	if (rc)
6486 		goto ibmvnic_dev_file_err;
6487 
6488 	netif_carrier_off(netdev);
6489 
6490 	if (init_success) {
6491 		adapter->state = VNIC_PROBED;
6492 		netdev->mtu = adapter->req_mtu - ETH_HLEN;
6493 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
6494 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
6495 	} else {
6496 		adapter->state = VNIC_DOWN;
6497 	}
6498 
6499 	adapter->wait_for_reset = false;
6500 	adapter->last_reset_time = jiffies;
6501 
6502 	rc = register_netdev(netdev);
6503 	if (rc) {
6504 		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
6505 		goto ibmvnic_register_fail;
6506 	}
6507 	dev_info(&dev->dev, "ibmvnic registered\n");
6508 
6509 	rc = ibmvnic_cpu_notif_add(adapter);
6510 	if (rc) {
6511 		netdev_err(netdev, "Registering cpu notifier failed\n");
6512 		goto cpu_notif_add_failed;
6513 	}
6514 
6515 	complete(&adapter->probe_done);
6516 
6517 	return 0;
6518 
6519 cpu_notif_add_failed:
6520 	unregister_netdev(netdev);
6521 
6522 ibmvnic_register_fail:
6523 	device_remove_file(&dev->dev, &dev_attr_failover);
6524 
6525 ibmvnic_dev_file_err:
6526 	release_stats_token(adapter);
6527 
6528 ibmvnic_stats_fail:
6529 	release_stats_buffers(adapter);
6530 
6531 ibmvnic_init_fail:
6532 	release_sub_crqs(adapter, 1);
6533 	release_crq_queue(adapter);
6534 
6535 	/* cleanup worker thread after releasing CRQ so we don't get
6536 	 * transport events (i.e new work items for the worker thread).
6537 	 */
6538 	adapter->state = VNIC_REMOVING;
6539 	complete(&adapter->probe_done);
6540 	flush_work(&adapter->ibmvnic_reset);
6541 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6542 
6543 	flush_reset_queue(adapter);
6544 
6545 	mutex_destroy(&adapter->fw_lock);
6546 	free_netdev(netdev);
6547 
6548 	return rc;
6549 }
6550 
6551 static void ibmvnic_remove(struct vio_dev *dev)
6552 {
6553 	struct net_device *netdev = dev_get_drvdata(&dev->dev);
6554 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6555 	unsigned long flags;
6556 
6557 	spin_lock_irqsave(&adapter->state_lock, flags);
6558 
6559 	/* If ibmvnic_reset() is scheduling a reset, wait for it to
6560 	 * finish. Then, set the state to REMOVING to prevent it from
6561 	 * scheduling any more work and to have reset functions ignore
6562 	 * any resets that have already been scheduled. Drop the lock
6563 	 * after setting state, so __ibmvnic_reset() which is called
6564 	 * from the flush_work() below, can make progress.
6565 	 */
6566 	spin_lock(&adapter->rwi_lock);
6567 	adapter->state = VNIC_REMOVING;
6568 	spin_unlock(&adapter->rwi_lock);
6569 
6570 	spin_unlock_irqrestore(&adapter->state_lock, flags);
6571 
6572 	ibmvnic_cpu_notif_remove(adapter);
6573 
6574 	flush_work(&adapter->ibmvnic_reset);
6575 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6576 
6577 	rtnl_lock();
6578 	unregister_netdevice(netdev);
6579 
6580 	release_resources(adapter);
6581 	release_rx_pools(adapter);
6582 	release_tx_pools(adapter);
6583 	release_sub_crqs(adapter, 1);
6584 	release_crq_queue(adapter);
6585 
6586 	release_stats_token(adapter);
6587 	release_stats_buffers(adapter);
6588 
6589 	adapter->state = VNIC_REMOVED;
6590 
6591 	rtnl_unlock();
6592 	mutex_destroy(&adapter->fw_lock);
6593 	device_remove_file(&dev->dev, &dev_attr_failover);
6594 	free_netdev(netdev);
6595 	dev_set_drvdata(&dev->dev, NULL);
6596 }
6597 
6598 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
6599 			      const char *buf, size_t count)
6600 {
6601 	struct net_device *netdev = dev_get_drvdata(dev);
6602 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6603 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
6604 	__be64 session_token;
6605 	long rc;
6606 
6607 	if (!sysfs_streq(buf, "1"))
6608 		return -EINVAL;
6609 
6610 	rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
6611 			 H_GET_SESSION_TOKEN, 0, 0, 0);
6612 	if (rc) {
6613 		netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
6614 			   rc);
6615 		goto last_resort;
6616 	}
6617 
6618 	session_token = (__be64)retbuf[0];
6619 	netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
6620 		   be64_to_cpu(session_token));
6621 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
6622 				H_SESSION_ERR_DETECTED, session_token, 0, 0);
6623 	if (rc) {
6624 		netdev_err(netdev,
6625 			   "H_VIOCTL initiated failover failed, rc %ld\n",
6626 			   rc);
6627 		goto last_resort;
6628 	}
6629 
6630 	return count;
6631 
6632 last_resort:
6633 	netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
6634 	ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
6635 
6636 	return count;
6637 }
6638 static DEVICE_ATTR_WO(failover);
6639 
6640 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
6641 {
6642 	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
6643 	struct ibmvnic_adapter *adapter;
6644 	struct iommu_table *tbl;
6645 	unsigned long ret = 0;
6646 	int i;
6647 
6648 	tbl = get_iommu_table_base(&vdev->dev);
6649 
6650 	/* netdev inits at probe time along with the structures we need below*/
6651 	if (!netdev)
6652 		return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
6653 
6654 	adapter = netdev_priv(netdev);
6655 
6656 	ret += PAGE_SIZE; /* the crq message queue */
6657 	ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
6658 
6659 	for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
6660 		ret += 4 * PAGE_SIZE; /* the scrq message queue */
6661 
6662 	for (i = 0; i < adapter->num_active_rx_pools; i++)
6663 		ret += adapter->rx_pool[i].size *
6664 		    IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
6665 
6666 	return ret;
6667 }
6668 
6669 static int ibmvnic_resume(struct device *dev)
6670 {
6671 	struct net_device *netdev = dev_get_drvdata(dev);
6672 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6673 
6674 	if (adapter->state != VNIC_OPEN)
6675 		return 0;
6676 
6677 	tasklet_schedule(&adapter->tasklet);
6678 
6679 	return 0;
6680 }
6681 
6682 static const struct vio_device_id ibmvnic_device_table[] = {
6683 	{"network", "IBM,vnic"},
6684 	{"", "" }
6685 };
6686 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
6687 
6688 static const struct dev_pm_ops ibmvnic_pm_ops = {
6689 	.resume = ibmvnic_resume
6690 };
6691 
6692 static struct vio_driver ibmvnic_driver = {
6693 	.id_table       = ibmvnic_device_table,
6694 	.probe          = ibmvnic_probe,
6695 	.remove         = ibmvnic_remove,
6696 	.get_desired_dma = ibmvnic_get_desired_dma,
6697 	.name		= ibmvnic_driver_name,
6698 	.pm		= &ibmvnic_pm_ops,
6699 };
6700 
6701 /* module functions */
6702 static int __init ibmvnic_module_init(void)
6703 {
6704 	int ret;
6705 
6706 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "net/ibmvnic:online",
6707 				      ibmvnic_cpu_online,
6708 				      ibmvnic_cpu_down_prep);
6709 	if (ret < 0)
6710 		goto out;
6711 	ibmvnic_online = ret;
6712 	ret = cpuhp_setup_state_multi(CPUHP_IBMVNIC_DEAD, "net/ibmvnic:dead",
6713 				      NULL, ibmvnic_cpu_dead);
6714 	if (ret)
6715 		goto err_dead;
6716 
6717 	ret = vio_register_driver(&ibmvnic_driver);
6718 	if (ret)
6719 		goto err_vio_register;
6720 
6721 	pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
6722 		IBMVNIC_DRIVER_VERSION);
6723 
6724 	return 0;
6725 err_vio_register:
6726 	cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6727 err_dead:
6728 	cpuhp_remove_multi_state(ibmvnic_online);
6729 out:
6730 	return ret;
6731 }
6732 
6733 static void __exit ibmvnic_module_exit(void)
6734 {
6735 	vio_unregister_driver(&ibmvnic_driver);
6736 	cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6737 	cpuhp_remove_multi_state(ibmvnic_online);
6738 }
6739 
6740 module_init(ibmvnic_module_init);
6741 module_exit(ibmvnic_module_exit);
6742