xref: /openbmc/linux/drivers/net/ethernet/ibm/ibmvnic.c (revision 8ebc80a25f9d9bf7a8e368b266d5b740c485c362)
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 
send_crq_init_complete(struct ibmvnic_adapter * adapter)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 
send_version_xchg(struct ibmvnic_adapter * adapter)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 
ibmvnic_clean_queue_affinity(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * queue)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 
ibmvnic_clean_affinity(struct ibmvnic_adapter * adapter)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 
ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue * queue,unsigned int * cpu,int * stragglers,int stride)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 */
ibmvnic_set_affinity(struct ibmvnic_adapter * adapter)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 
ibmvnic_cpu_online(unsigned int cpu,struct hlist_node * node)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 
ibmvnic_cpu_dead(unsigned int cpu,struct hlist_node * node)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 
ibmvnic_cpu_down_prep(unsigned int cpu,struct hlist_node * node)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 
ibmvnic_cpu_notif_add(struct ibmvnic_adapter * adapter)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 
ibmvnic_cpu_notif_remove(struct ibmvnic_adapter * adapter)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 
h_reg_sub_crq(unsigned long unit_address,unsigned long token,unsigned long length,unsigned long * number,unsigned long * irq)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  */
ibmvnic_wait_for_completion(struct ibmvnic_adapter * adapter,struct completion * comp_done,unsigned long timeout)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  */
reuse_ltb(struct ibmvnic_long_term_buff * ltb,int size)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  */
alloc_long_term_buff(struct ibmvnic_adapter * adapter,struct ibmvnic_long_term_buff * ltb,int size)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 
free_long_term_buff(struct ibmvnic_adapter * adapter,struct ibmvnic_long_term_buff * ltb)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 
free_ltb_set(struct ibmvnic_adapter * adapter,struct ibmvnic_ltb_set * ltb_set)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  */
alloc_ltb_set(struct ibmvnic_adapter * adapter,struct ibmvnic_ltb_set * ltb_set,int num_buffs,int buff_size)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  */
map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool * rxpool,unsigned int bufidx,struct ibmvnic_long_term_buff ** ltbp,unsigned int * offset)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  */
map_txpool_buf_to_ltb(struct ibmvnic_tx_pool * txpool,unsigned int bufidx,struct ibmvnic_long_term_buff ** ltbp,unsigned int * offset)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 
deactivate_rx_pools(struct ibmvnic_adapter * adapter)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 
replenish_rx_pool(struct ibmvnic_adapter * adapter,struct ibmvnic_rx_pool * pool)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 
replenish_pools(struct ibmvnic_adapter * adapter)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 
release_stats_buffers(struct ibmvnic_adapter * adapter)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 
init_stats_buffers(struct ibmvnic_adapter * adapter)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 
release_stats_token(struct ibmvnic_adapter * adapter)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 
init_stats_token(struct ibmvnic_adapter * adapter)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  */
release_rx_pools(struct ibmvnic_adapter * adapter)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  */
reuse_rx_pools(struct ibmvnic_adapter * adapter)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  */
init_rx_pools(struct net_device * netdev)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 
release_vpd_data(struct ibmvnic_adapter * adapter)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 
release_one_tx_pool(struct ibmvnic_adapter * adapter,struct ibmvnic_tx_pool * tx_pool)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  */
release_tx_pools(struct ibmvnic_adapter * adapter)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 
init_one_tx_pool(struct net_device * netdev,struct ibmvnic_tx_pool * tx_pool,int pool_size,int buf_size)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  */
reuse_tx_pools(struct ibmvnic_adapter * adapter)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  */
init_tx_pools(struct net_device * netdev)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 
ibmvnic_napi_enable(struct ibmvnic_adapter * adapter)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 
ibmvnic_napi_disable(struct ibmvnic_adapter * adapter)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 
init_napi(struct ibmvnic_adapter * adapter)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 
release_napi(struct ibmvnic_adapter * adapter)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 
adapter_state_to_string(enum vnic_state state)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 
ibmvnic_login(struct net_device * netdev)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 
release_login_buffer(struct ibmvnic_adapter * adapter)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 
release_login_rsp_buffer(struct ibmvnic_adapter * adapter)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 
release_resources(struct ibmvnic_adapter * adapter)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 
set_link_state(struct ibmvnic_adapter * adapter,u8 link_state)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 
set_real_num_queues(struct net_device * netdev)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 
ibmvnic_get_vpd(struct ibmvnic_adapter * adapter)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 
init_resources(struct ibmvnic_adapter * adapter)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 
__ibmvnic_open(struct net_device * netdev)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 
ibmvnic_open(struct net_device * netdev)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 
clean_rx_pools(struct ibmvnic_adapter * adapter)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 
clean_one_tx_pool(struct ibmvnic_adapter * adapter,struct ibmvnic_tx_pool * tx_pool)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 
clean_tx_pools(struct ibmvnic_adapter * adapter)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 
ibmvnic_disable_irqs(struct ibmvnic_adapter * adapter)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 
ibmvnic_cleanup(struct net_device * netdev)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 
__ibmvnic_close(struct net_device * netdev)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 
ibmvnic_close(struct net_device * netdev)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  * build_hdr_data - creates L2/L3/L4 header data buffer
2145  * @hdr_field: bitfield determining needed headers
2146  * @skb: socket buffer
2147  * @hdr_len: array of header lengths
2148  * @hdr_data: buffer to write the header to
2149  *
2150  * Reads hdr_field to determine which headers are needed by firmware.
2151  * Builds a buffer containing these headers.  Saves individual header
2152  * lengths and total buffer length to be used to build descriptors.
2153  */
build_hdr_data(u8 hdr_field,struct sk_buff * skb,int * hdr_len,u8 * hdr_data)2154 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
2155 			  int *hdr_len, u8 *hdr_data)
2156 {
2157 	int len = 0;
2158 	u8 *hdr;
2159 
2160 	if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
2161 		hdr_len[0] = sizeof(struct vlan_ethhdr);
2162 	else
2163 		hdr_len[0] = sizeof(struct ethhdr);
2164 
2165 	if (skb->protocol == htons(ETH_P_IP)) {
2166 		hdr_len[1] = ip_hdr(skb)->ihl * 4;
2167 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2168 			hdr_len[2] = tcp_hdrlen(skb);
2169 		else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
2170 			hdr_len[2] = sizeof(struct udphdr);
2171 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2172 		hdr_len[1] = sizeof(struct ipv6hdr);
2173 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2174 			hdr_len[2] = tcp_hdrlen(skb);
2175 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
2176 			hdr_len[2] = sizeof(struct udphdr);
2177 	} else if (skb->protocol == htons(ETH_P_ARP)) {
2178 		hdr_len[1] = arp_hdr_len(skb->dev);
2179 		hdr_len[2] = 0;
2180 	}
2181 
2182 	memset(hdr_data, 0, 120);
2183 	if ((hdr_field >> 6) & 1) {
2184 		hdr = skb_mac_header(skb);
2185 		memcpy(hdr_data, hdr, hdr_len[0]);
2186 		len += hdr_len[0];
2187 	}
2188 
2189 	if ((hdr_field >> 5) & 1) {
2190 		hdr = skb_network_header(skb);
2191 		memcpy(hdr_data + len, hdr, hdr_len[1]);
2192 		len += hdr_len[1];
2193 	}
2194 
2195 	if ((hdr_field >> 4) & 1) {
2196 		hdr = skb_transport_header(skb);
2197 		memcpy(hdr_data + len, hdr, hdr_len[2]);
2198 		len += hdr_len[2];
2199 	}
2200 	return len;
2201 }
2202 
2203 /**
2204  * create_hdr_descs - create header and header extension descriptors
2205  * @hdr_field: bitfield determining needed headers
2206  * @hdr_data: buffer containing header data
2207  * @len: length of data buffer
2208  * @hdr_len: array of individual header lengths
2209  * @scrq_arr: descriptor array
2210  *
2211  * Creates header and, if needed, header extension descriptors and
2212  * places them in a descriptor array, scrq_arr
2213  */
2214 
create_hdr_descs(u8 hdr_field,u8 * hdr_data,int len,int * hdr_len,union sub_crq * scrq_arr)2215 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
2216 			    union sub_crq *scrq_arr)
2217 {
2218 	union sub_crq hdr_desc;
2219 	int tmp_len = len;
2220 	int num_descs = 0;
2221 	u8 *data, *cur;
2222 	int tmp;
2223 
2224 	while (tmp_len > 0) {
2225 		cur = hdr_data + len - tmp_len;
2226 
2227 		memset(&hdr_desc, 0, sizeof(hdr_desc));
2228 		if (cur != hdr_data) {
2229 			data = hdr_desc.hdr_ext.data;
2230 			tmp = tmp_len > 29 ? 29 : tmp_len;
2231 			hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
2232 			hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
2233 			hdr_desc.hdr_ext.len = tmp;
2234 		} else {
2235 			data = hdr_desc.hdr.data;
2236 			tmp = tmp_len > 24 ? 24 : tmp_len;
2237 			hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
2238 			hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
2239 			hdr_desc.hdr.len = tmp;
2240 			hdr_desc.hdr.l2_len = (u8)hdr_len[0];
2241 			hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
2242 			hdr_desc.hdr.l4_len = (u8)hdr_len[2];
2243 			hdr_desc.hdr.flag = hdr_field << 1;
2244 		}
2245 		memcpy(data, cur, tmp);
2246 		tmp_len -= tmp;
2247 		*scrq_arr = hdr_desc;
2248 		scrq_arr++;
2249 		num_descs++;
2250 	}
2251 
2252 	return num_descs;
2253 }
2254 
2255 /**
2256  * build_hdr_descs_arr - build a header descriptor array
2257  * @skb: tx socket buffer
2258  * @indir_arr: indirect array
2259  * @num_entries: number of descriptors to be sent
2260  * @hdr_field: bit field determining which headers will be sent
2261  *
2262  * This function will build a TX descriptor array with applicable
2263  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
2264  */
2265 
build_hdr_descs_arr(struct sk_buff * skb,union sub_crq * indir_arr,int * num_entries,u8 hdr_field)2266 static void build_hdr_descs_arr(struct sk_buff *skb,
2267 				union sub_crq *indir_arr,
2268 				int *num_entries, u8 hdr_field)
2269 {
2270 	int hdr_len[3] = {0, 0, 0};
2271 	u8 hdr_data[140] = {0};
2272 	int tot_len;
2273 
2274 	tot_len = build_hdr_data(hdr_field, skb, hdr_len,
2275 				 hdr_data);
2276 	*num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
2277 					 indir_arr + 1);
2278 }
2279 
ibmvnic_xmit_workarounds(struct sk_buff * skb,struct net_device * netdev)2280 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
2281 				    struct net_device *netdev)
2282 {
2283 	/* For some backing devices, mishandling of small packets
2284 	 * can result in a loss of connection or TX stall. Device
2285 	 * architects recommend that no packet should be smaller
2286 	 * than the minimum MTU value provided to the driver, so
2287 	 * pad any packets to that length
2288 	 */
2289 	if (skb->len < netdev->min_mtu)
2290 		return skb_put_padto(skb, netdev->min_mtu);
2291 
2292 	return 0;
2293 }
2294 
ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * tx_scrq)2295 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
2296 					 struct ibmvnic_sub_crq_queue *tx_scrq)
2297 {
2298 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2299 	struct ibmvnic_tx_buff *tx_buff;
2300 	struct ibmvnic_tx_pool *tx_pool;
2301 	union sub_crq tx_scrq_entry;
2302 	int queue_num;
2303 	int entries;
2304 	int index;
2305 	int i;
2306 
2307 	ind_bufp = &tx_scrq->ind_buf;
2308 	entries = (u64)ind_bufp->index;
2309 	queue_num = tx_scrq->pool_index;
2310 
2311 	for (i = entries - 1; i >= 0; --i) {
2312 		tx_scrq_entry = ind_bufp->indir_arr[i];
2313 		if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
2314 			continue;
2315 		index = be32_to_cpu(tx_scrq_entry.v1.correlator);
2316 		if (index & IBMVNIC_TSO_POOL_MASK) {
2317 			tx_pool = &adapter->tso_pool[queue_num];
2318 			index &= ~IBMVNIC_TSO_POOL_MASK;
2319 		} else {
2320 			tx_pool = &adapter->tx_pool[queue_num];
2321 		}
2322 		tx_pool->free_map[tx_pool->consumer_index] = index;
2323 		tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2324 					  tx_pool->num_buffers - 1 :
2325 					  tx_pool->consumer_index - 1;
2326 		tx_buff = &tx_pool->tx_buff[index];
2327 		adapter->netdev->stats.tx_packets--;
2328 		adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
2329 		adapter->tx_stats_buffers[queue_num].batched_packets--;
2330 		adapter->tx_stats_buffers[queue_num].bytes -=
2331 						tx_buff->skb->len;
2332 		dev_kfree_skb_any(tx_buff->skb);
2333 		tx_buff->skb = NULL;
2334 		adapter->netdev->stats.tx_dropped++;
2335 	}
2336 
2337 	ind_bufp->index = 0;
2338 
2339 	if (atomic_sub_return(entries, &tx_scrq->used) <=
2340 	    (adapter->req_tx_entries_per_subcrq / 2) &&
2341 	    __netif_subqueue_stopped(adapter->netdev, queue_num)) {
2342 		rcu_read_lock();
2343 
2344 		if (adapter->tx_queues_active) {
2345 			netif_wake_subqueue(adapter->netdev, queue_num);
2346 			netdev_dbg(adapter->netdev, "Started queue %d\n",
2347 				   queue_num);
2348 		}
2349 
2350 		rcu_read_unlock();
2351 	}
2352 }
2353 
send_subcrq_direct(struct ibmvnic_adapter * adapter,u64 remote_handle,u64 * entry)2354 static int send_subcrq_direct(struct ibmvnic_adapter *adapter,
2355 			      u64 remote_handle, u64 *entry)
2356 {
2357 	unsigned int ua = adapter->vdev->unit_address;
2358 	struct device *dev = &adapter->vdev->dev;
2359 	int rc;
2360 
2361 	/* Make sure the hypervisor sees the complete request */
2362 	dma_wmb();
2363 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
2364 				cpu_to_be64(remote_handle),
2365 				cpu_to_be64(entry[0]), cpu_to_be64(entry[1]),
2366 				cpu_to_be64(entry[2]), cpu_to_be64(entry[3]));
2367 
2368 	if (rc)
2369 		print_subcrq_error(dev, rc, __func__);
2370 
2371 	return rc;
2372 }
2373 
ibmvnic_tx_scrq_flush(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * tx_scrq,bool indirect)2374 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
2375 				 struct ibmvnic_sub_crq_queue *tx_scrq,
2376 				 bool indirect)
2377 {
2378 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2379 	u64 dma_addr;
2380 	u64 entries;
2381 	u64 handle;
2382 	int rc;
2383 
2384 	ind_bufp = &tx_scrq->ind_buf;
2385 	dma_addr = (u64)ind_bufp->indir_dma;
2386 	entries = (u64)ind_bufp->index;
2387 	handle = tx_scrq->handle;
2388 
2389 	if (!entries)
2390 		return 0;
2391 
2392 	if (indirect)
2393 		rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
2394 	else
2395 		rc = send_subcrq_direct(adapter, handle,
2396 					(u64 *)ind_bufp->indir_arr);
2397 
2398 	if (rc)
2399 		ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
2400 	else
2401 		ind_bufp->index = 0;
2402 	return rc;
2403 }
2404 
ibmvnic_xmit(struct sk_buff * skb,struct net_device * netdev)2405 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
2406 {
2407 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2408 	int queue_num = skb_get_queue_mapping(skb);
2409 	u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
2410 	struct device *dev = &adapter->vdev->dev;
2411 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2412 	struct ibmvnic_tx_buff *tx_buff = NULL;
2413 	struct ibmvnic_sub_crq_queue *tx_scrq;
2414 	struct ibmvnic_long_term_buff *ltb;
2415 	struct ibmvnic_tx_pool *tx_pool;
2416 	unsigned int tx_send_failed = 0;
2417 	netdev_tx_t ret = NETDEV_TX_OK;
2418 	unsigned int tx_map_failed = 0;
2419 	union sub_crq indir_arr[16];
2420 	unsigned int tx_dropped = 0;
2421 	unsigned int tx_dpackets = 0;
2422 	unsigned int tx_bpackets = 0;
2423 	unsigned int tx_bytes = 0;
2424 	dma_addr_t data_dma_addr;
2425 	struct netdev_queue *txq;
2426 	unsigned long lpar_rc;
2427 	unsigned int skblen;
2428 	union sub_crq tx_crq;
2429 	unsigned int offset;
2430 	bool use_scrq_send_direct = false;
2431 	int num_entries = 1;
2432 	unsigned char *dst;
2433 	int bufidx = 0;
2434 	u8 proto = 0;
2435 
2436 	/* If a reset is in progress, drop the packet since
2437 	 * the scrqs may get torn down. Otherwise use the
2438 	 * rcu to ensure reset waits for us to complete.
2439 	 */
2440 	rcu_read_lock();
2441 	if (!adapter->tx_queues_active) {
2442 		dev_kfree_skb_any(skb);
2443 
2444 		tx_send_failed++;
2445 		tx_dropped++;
2446 		ret = NETDEV_TX_OK;
2447 		goto out;
2448 	}
2449 
2450 	tx_scrq = adapter->tx_scrq[queue_num];
2451 	txq = netdev_get_tx_queue(netdev, queue_num);
2452 	ind_bufp = &tx_scrq->ind_buf;
2453 
2454 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
2455 		tx_dropped++;
2456 		tx_send_failed++;
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 	if (skb_is_gso(skb))
2465 		tx_pool = &adapter->tso_pool[queue_num];
2466 	else
2467 		tx_pool = &adapter->tx_pool[queue_num];
2468 
2469 	bufidx = tx_pool->free_map[tx_pool->consumer_index];
2470 
2471 	if (bufidx == IBMVNIC_INVALID_MAP) {
2472 		dev_kfree_skb_any(skb);
2473 		tx_send_failed++;
2474 		tx_dropped++;
2475 		ret = NETDEV_TX_OK;
2476 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2477 		if (lpar_rc != H_SUCCESS)
2478 			goto tx_err;
2479 		goto out;
2480 	}
2481 
2482 	tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
2483 
2484 	map_txpool_buf_to_ltb(tx_pool, bufidx, &ltb, &offset);
2485 
2486 	dst = ltb->buff + offset;
2487 	memset(dst, 0, tx_pool->buf_size);
2488 	data_dma_addr = ltb->addr + offset;
2489 
2490 	/* if we are going to send_subcrq_direct this then we need to
2491 	 * update the checksum before copying the data into ltb. Essentially
2492 	 * these packets force disable CSO so that we can guarantee that
2493 	 * FW does not need header info and we can send direct. Also, vnic
2494 	 * server must be able to xmit standard packets without header data
2495 	 */
2496 	if (*hdrs == 0 && !skb_is_gso(skb) &&
2497 	    !ind_bufp->index && !netdev_xmit_more()) {
2498 		use_scrq_send_direct = true;
2499 		if (skb->ip_summed == CHECKSUM_PARTIAL &&
2500 		    skb_checksum_help(skb))
2501 			use_scrq_send_direct = false;
2502 	}
2503 
2504 	if (skb_shinfo(skb)->nr_frags) {
2505 		int cur, i;
2506 
2507 		/* Copy the head */
2508 		skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
2509 		cur = skb_headlen(skb);
2510 
2511 		/* Copy the frags */
2512 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2513 			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2514 
2515 			memcpy(dst + cur, skb_frag_address(frag),
2516 			       skb_frag_size(frag));
2517 			cur += skb_frag_size(frag);
2518 		}
2519 	} else {
2520 		skb_copy_from_linear_data(skb, dst, skb->len);
2521 	}
2522 
2523 	/* post changes to long_term_buff *dst before VIOS accessing it */
2524 	dma_wmb();
2525 
2526 	tx_pool->consumer_index =
2527 	    (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
2528 
2529 	tx_buff = &tx_pool->tx_buff[bufidx];
2530 
2531 	/* Sanity checks on our free map to make sure it points to an index
2532 	 * that is not being occupied by another skb. If skb memory is
2533 	 * not freed then we see congestion control kick in and halt tx.
2534 	 */
2535 	if (unlikely(tx_buff->skb)) {
2536 		dev_warn_ratelimited(dev, "TX free map points to untracked skb (%s %d idx=%d)\n",
2537 				     skb_is_gso(skb) ? "tso_pool" : "tx_pool",
2538 				     queue_num, bufidx);
2539 		dev_kfree_skb_any(tx_buff->skb);
2540 	}
2541 
2542 	tx_buff->skb = skb;
2543 	tx_buff->index = bufidx;
2544 	tx_buff->pool_index = queue_num;
2545 	skblen = skb->len;
2546 
2547 	memset(&tx_crq, 0, sizeof(tx_crq));
2548 	tx_crq.v1.first = IBMVNIC_CRQ_CMD;
2549 	tx_crq.v1.type = IBMVNIC_TX_DESC;
2550 	tx_crq.v1.n_crq_elem = 1;
2551 	tx_crq.v1.n_sge = 1;
2552 	tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
2553 
2554 	if (skb_is_gso(skb))
2555 		tx_crq.v1.correlator =
2556 			cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK);
2557 	else
2558 		tx_crq.v1.correlator = cpu_to_be32(bufidx);
2559 	tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id);
2560 	tx_crq.v1.sge_len = cpu_to_be32(skb->len);
2561 	tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
2562 
2563 	if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
2564 		tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
2565 		tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2566 	}
2567 
2568 	if (skb->protocol == htons(ETH_P_IP)) {
2569 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2570 		proto = ip_hdr(skb)->protocol;
2571 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2572 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2573 		proto = ipv6_hdr(skb)->nexthdr;
2574 	}
2575 
2576 	if (proto == IPPROTO_TCP)
2577 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2578 	else if (proto == IPPROTO_UDP)
2579 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2580 
2581 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2582 		tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2583 		hdrs += 2;
2584 	}
2585 	if (skb_is_gso(skb)) {
2586 		tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2587 		tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2588 		hdrs += 2;
2589 	} else if (use_scrq_send_direct) {
2590 		/* See above comment, CSO disabled with direct xmit */
2591 		tx_crq.v1.flags1 &= ~(IBMVNIC_TX_CHKSUM_OFFLOAD);
2592 		ind_bufp->index = 1;
2593 		tx_buff->num_entries = 1;
2594 		netdev_tx_sent_queue(txq, skb->len);
2595 		ind_bufp->indir_arr[0] = tx_crq;
2596 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, false);
2597 		if (lpar_rc != H_SUCCESS)
2598 			goto tx_err;
2599 
2600 		tx_dpackets++;
2601 		goto early_exit;
2602 	}
2603 
2604 	if ((*hdrs >> 7) & 1)
2605 		build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2606 
2607 	tx_crq.v1.n_crq_elem = num_entries;
2608 	tx_buff->num_entries = num_entries;
2609 	/* flush buffer if current entry can not fit */
2610 	if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2611 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2612 		if (lpar_rc != H_SUCCESS)
2613 			goto tx_flush_err;
2614 	}
2615 
2616 	indir_arr[0] = tx_crq;
2617 	memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2618 	       num_entries * sizeof(struct ibmvnic_generic_scrq));
2619 
2620 	ind_bufp->index += num_entries;
2621 	if (__netdev_tx_sent_queue(txq, skb->len,
2622 				   netdev_xmit_more() &&
2623 				   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2624 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2625 		if (lpar_rc != H_SUCCESS)
2626 			goto tx_err;
2627 	}
2628 
2629 	tx_bpackets++;
2630 
2631 early_exit:
2632 	if (atomic_add_return(num_entries, &tx_scrq->used)
2633 					>= adapter->req_tx_entries_per_subcrq) {
2634 		netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2635 		netif_stop_subqueue(netdev, queue_num);
2636 	}
2637 
2638 	tx_bytes += skblen;
2639 	txq_trans_cond_update(txq);
2640 	ret = NETDEV_TX_OK;
2641 	goto out;
2642 
2643 tx_flush_err:
2644 	dev_kfree_skb_any(skb);
2645 	tx_buff->skb = NULL;
2646 	tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2647 				  tx_pool->num_buffers - 1 :
2648 				  tx_pool->consumer_index - 1;
2649 	tx_dropped++;
2650 tx_err:
2651 	if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2652 		dev_err_ratelimited(dev, "tx: send failed\n");
2653 
2654 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2655 		/* Disable TX and report carrier off if queue is closed
2656 		 * or pending failover.
2657 		 * Firmware guarantees that a signal will be sent to the
2658 		 * driver, triggering a reset or some other action.
2659 		 */
2660 		netif_tx_stop_all_queues(netdev);
2661 		netif_carrier_off(netdev);
2662 	}
2663 out:
2664 	rcu_read_unlock();
2665 	netdev->stats.tx_dropped += tx_dropped;
2666 	netdev->stats.tx_bytes += tx_bytes;
2667 	netdev->stats.tx_packets += tx_bpackets + tx_dpackets;
2668 	adapter->tx_send_failed += tx_send_failed;
2669 	adapter->tx_map_failed += tx_map_failed;
2670 	adapter->tx_stats_buffers[queue_num].batched_packets += tx_bpackets;
2671 	adapter->tx_stats_buffers[queue_num].direct_packets += tx_dpackets;
2672 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2673 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2674 
2675 	return ret;
2676 }
2677 
ibmvnic_set_multi(struct net_device * netdev)2678 static void ibmvnic_set_multi(struct net_device *netdev)
2679 {
2680 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2681 	struct netdev_hw_addr *ha;
2682 	union ibmvnic_crq crq;
2683 
2684 	memset(&crq, 0, sizeof(crq));
2685 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
2686 	crq.request_capability.cmd = REQUEST_CAPABILITY;
2687 
2688 	if (netdev->flags & IFF_PROMISC) {
2689 		if (!adapter->promisc_supported)
2690 			return;
2691 	} else {
2692 		if (netdev->flags & IFF_ALLMULTI) {
2693 			/* Accept all multicast */
2694 			memset(&crq, 0, sizeof(crq));
2695 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2696 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2697 			crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2698 			ibmvnic_send_crq(adapter, &crq);
2699 		} else if (netdev_mc_empty(netdev)) {
2700 			/* Reject all multicast */
2701 			memset(&crq, 0, sizeof(crq));
2702 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2703 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2704 			crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2705 			ibmvnic_send_crq(adapter, &crq);
2706 		} else {
2707 			/* Accept one or more multicast(s) */
2708 			netdev_for_each_mc_addr(ha, netdev) {
2709 				memset(&crq, 0, sizeof(crq));
2710 				crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2711 				crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2712 				crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2713 				ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2714 						ha->addr);
2715 				ibmvnic_send_crq(adapter, &crq);
2716 			}
2717 		}
2718 	}
2719 }
2720 
__ibmvnic_set_mac(struct net_device * netdev,u8 * dev_addr)2721 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2722 {
2723 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2724 	union ibmvnic_crq crq;
2725 	int rc;
2726 
2727 	if (!is_valid_ether_addr(dev_addr)) {
2728 		rc = -EADDRNOTAVAIL;
2729 		goto err;
2730 	}
2731 
2732 	memset(&crq, 0, sizeof(crq));
2733 	crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2734 	crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2735 	ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2736 
2737 	mutex_lock(&adapter->fw_lock);
2738 	adapter->fw_done_rc = 0;
2739 	reinit_completion(&adapter->fw_done);
2740 
2741 	rc = ibmvnic_send_crq(adapter, &crq);
2742 	if (rc) {
2743 		rc = -EIO;
2744 		mutex_unlock(&adapter->fw_lock);
2745 		goto err;
2746 	}
2747 
2748 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2749 	/* netdev->dev_addr is changed in handle_change_mac_rsp function */
2750 	if (rc || adapter->fw_done_rc) {
2751 		rc = -EIO;
2752 		mutex_unlock(&adapter->fw_lock);
2753 		goto err;
2754 	}
2755 	mutex_unlock(&adapter->fw_lock);
2756 	return 0;
2757 err:
2758 	ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2759 	return rc;
2760 }
2761 
ibmvnic_set_mac(struct net_device * netdev,void * p)2762 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2763 {
2764 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2765 	struct sockaddr *addr = p;
2766 	int rc;
2767 
2768 	rc = 0;
2769 	if (!is_valid_ether_addr(addr->sa_data))
2770 		return -EADDRNOTAVAIL;
2771 
2772 	ether_addr_copy(adapter->mac_addr, addr->sa_data);
2773 	if (adapter->state != VNIC_PROBED)
2774 		rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2775 
2776 	return rc;
2777 }
2778 
reset_reason_to_string(enum ibmvnic_reset_reason reason)2779 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2780 {
2781 	switch (reason) {
2782 	case VNIC_RESET_FAILOVER:
2783 		return "FAILOVER";
2784 	case VNIC_RESET_MOBILITY:
2785 		return "MOBILITY";
2786 	case VNIC_RESET_FATAL:
2787 		return "FATAL";
2788 	case VNIC_RESET_NON_FATAL:
2789 		return "NON_FATAL";
2790 	case VNIC_RESET_TIMEOUT:
2791 		return "TIMEOUT";
2792 	case VNIC_RESET_CHANGE_PARAM:
2793 		return "CHANGE_PARAM";
2794 	case VNIC_RESET_PASSIVE_INIT:
2795 		return "PASSIVE_INIT";
2796 	}
2797 	return "UNKNOWN";
2798 }
2799 
2800 /*
2801  * Initialize the init_done completion and return code values. We
2802  * can get a transport event just after registering the CRQ and the
2803  * tasklet will use this to communicate the transport event. To ensure
2804  * we don't miss the notification/error, initialize these _before_
2805  * regisering the CRQ.
2806  */
reinit_init_done(struct ibmvnic_adapter * adapter)2807 static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
2808 {
2809 	reinit_completion(&adapter->init_done);
2810 	adapter->init_done_rc = 0;
2811 }
2812 
2813 /*
2814  * do_reset returns zero if we are able to keep processing reset events, or
2815  * non-zero if we hit a fatal error and must halt.
2816  */
do_reset(struct ibmvnic_adapter * adapter,struct ibmvnic_rwi * rwi,u32 reset_state)2817 static int do_reset(struct ibmvnic_adapter *adapter,
2818 		    struct ibmvnic_rwi *rwi, u32 reset_state)
2819 {
2820 	struct net_device *netdev = adapter->netdev;
2821 	u64 old_num_rx_queues, old_num_tx_queues;
2822 	u64 old_num_rx_slots, old_num_tx_slots;
2823 	int rc;
2824 
2825 	netdev_dbg(adapter->netdev,
2826 		   "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2827 		   adapter_state_to_string(adapter->state),
2828 		   adapter->failover_pending,
2829 		   reset_reason_to_string(rwi->reset_reason),
2830 		   adapter_state_to_string(reset_state));
2831 
2832 	adapter->reset_reason = rwi->reset_reason;
2833 	/* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2834 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2835 		rtnl_lock();
2836 
2837 	/* Now that we have the rtnl lock, clear any pending failover.
2838 	 * This will ensure ibmvnic_open() has either completed or will
2839 	 * block until failover is complete.
2840 	 */
2841 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2842 		adapter->failover_pending = false;
2843 
2844 	/* read the state and check (again) after getting rtnl */
2845 	reset_state = adapter->state;
2846 
2847 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2848 		rc = -EBUSY;
2849 		goto out;
2850 	}
2851 
2852 	netif_carrier_off(netdev);
2853 
2854 	old_num_rx_queues = adapter->req_rx_queues;
2855 	old_num_tx_queues = adapter->req_tx_queues;
2856 	old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2857 	old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2858 
2859 	ibmvnic_cleanup(netdev);
2860 
2861 	if (reset_state == VNIC_OPEN &&
2862 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
2863 	    adapter->reset_reason != VNIC_RESET_FAILOVER) {
2864 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2865 			rc = __ibmvnic_close(netdev);
2866 			if (rc)
2867 				goto out;
2868 		} else {
2869 			adapter->state = VNIC_CLOSING;
2870 
2871 			/* Release the RTNL lock before link state change and
2872 			 * re-acquire after the link state change to allow
2873 			 * linkwatch_event to grab the RTNL lock and run during
2874 			 * a reset.
2875 			 */
2876 			rtnl_unlock();
2877 			rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2878 			rtnl_lock();
2879 			if (rc)
2880 				goto out;
2881 
2882 			if (adapter->state == VNIC_OPEN) {
2883 				/* When we dropped rtnl, ibmvnic_open() got
2884 				 * it and noticed that we are resetting and
2885 				 * set the adapter state to OPEN. Update our
2886 				 * new "target" state, and resume the reset
2887 				 * from VNIC_CLOSING state.
2888 				 */
2889 				netdev_dbg(netdev,
2890 					   "Open changed state from %s, updating.\n",
2891 					   adapter_state_to_string(reset_state));
2892 				reset_state = VNIC_OPEN;
2893 				adapter->state = VNIC_CLOSING;
2894 			}
2895 
2896 			if (adapter->state != VNIC_CLOSING) {
2897 				/* If someone else changed the adapter state
2898 				 * when we dropped the rtnl, fail the reset
2899 				 */
2900 				rc = -EAGAIN;
2901 				goto out;
2902 			}
2903 			adapter->state = VNIC_CLOSED;
2904 		}
2905 	}
2906 
2907 	if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2908 		release_resources(adapter);
2909 		release_sub_crqs(adapter, 1);
2910 		release_crq_queue(adapter);
2911 	}
2912 
2913 	if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2914 		/* remove the closed state so when we call open it appears
2915 		 * we are coming from the probed state.
2916 		 */
2917 		adapter->state = VNIC_PROBED;
2918 
2919 		reinit_init_done(adapter);
2920 
2921 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2922 			rc = init_crq_queue(adapter);
2923 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2924 			rc = ibmvnic_reenable_crq_queue(adapter);
2925 			release_sub_crqs(adapter, 1);
2926 		} else {
2927 			rc = ibmvnic_reset_crq(adapter);
2928 			if (rc == H_CLOSED || rc == H_SUCCESS) {
2929 				rc = vio_enable_interrupts(adapter->vdev);
2930 				if (rc)
2931 					netdev_err(adapter->netdev,
2932 						   "Reset failed to enable interrupts. rc=%d\n",
2933 						   rc);
2934 			}
2935 		}
2936 
2937 		if (rc) {
2938 			netdev_err(adapter->netdev,
2939 				   "Reset couldn't initialize crq. rc=%d\n", rc);
2940 			goto out;
2941 		}
2942 
2943 		rc = ibmvnic_reset_init(adapter, true);
2944 		if (rc)
2945 			goto out;
2946 
2947 		/* If the adapter was in PROBE or DOWN state prior to the reset,
2948 		 * exit here.
2949 		 */
2950 		if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2951 			rc = 0;
2952 			goto out;
2953 		}
2954 
2955 		rc = ibmvnic_login(netdev);
2956 		if (rc)
2957 			goto out;
2958 
2959 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2960 			rc = init_resources(adapter);
2961 			if (rc)
2962 				goto out;
2963 		} else if (adapter->req_rx_queues != old_num_rx_queues ||
2964 		    adapter->req_tx_queues != old_num_tx_queues ||
2965 		    adapter->req_rx_add_entries_per_subcrq !=
2966 		    old_num_rx_slots ||
2967 		    adapter->req_tx_entries_per_subcrq !=
2968 		    old_num_tx_slots ||
2969 		    !adapter->rx_pool ||
2970 		    !adapter->tso_pool ||
2971 		    !adapter->tx_pool) {
2972 			release_napi(adapter);
2973 			release_vpd_data(adapter);
2974 
2975 			rc = init_resources(adapter);
2976 			if (rc)
2977 				goto out;
2978 
2979 		} else {
2980 			rc = init_tx_pools(netdev);
2981 			if (rc) {
2982 				netdev_dbg(netdev,
2983 					   "init tx pools failed (%d)\n",
2984 					   rc);
2985 				goto out;
2986 			}
2987 
2988 			rc = init_rx_pools(netdev);
2989 			if (rc) {
2990 				netdev_dbg(netdev,
2991 					   "init rx pools failed (%d)\n",
2992 					   rc);
2993 				goto out;
2994 			}
2995 		}
2996 		ibmvnic_disable_irqs(adapter);
2997 	}
2998 	adapter->state = VNIC_CLOSED;
2999 
3000 	if (reset_state == VNIC_CLOSED) {
3001 		rc = 0;
3002 		goto out;
3003 	}
3004 
3005 	rc = __ibmvnic_open(netdev);
3006 	if (rc) {
3007 		rc = IBMVNIC_OPEN_FAILED;
3008 		goto out;
3009 	}
3010 
3011 	/* refresh device's multicast list */
3012 	ibmvnic_set_multi(netdev);
3013 
3014 	if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
3015 	    adapter->reset_reason == VNIC_RESET_MOBILITY)
3016 		__netdev_notify_peers(netdev);
3017 
3018 	rc = 0;
3019 
3020 out:
3021 	/* restore the adapter state if reset failed */
3022 	if (rc)
3023 		adapter->state = reset_state;
3024 	/* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
3025 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
3026 		rtnl_unlock();
3027 
3028 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
3029 		   adapter_state_to_string(adapter->state),
3030 		   adapter->failover_pending, rc);
3031 	return rc;
3032 }
3033 
do_hard_reset(struct ibmvnic_adapter * adapter,struct ibmvnic_rwi * rwi,u32 reset_state)3034 static int do_hard_reset(struct ibmvnic_adapter *adapter,
3035 			 struct ibmvnic_rwi *rwi, u32 reset_state)
3036 {
3037 	struct net_device *netdev = adapter->netdev;
3038 	int rc;
3039 
3040 	netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
3041 		   reset_reason_to_string(rwi->reset_reason));
3042 
3043 	/* read the state and check (again) after getting rtnl */
3044 	reset_state = adapter->state;
3045 
3046 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
3047 		rc = -EBUSY;
3048 		goto out;
3049 	}
3050 
3051 	netif_carrier_off(netdev);
3052 	adapter->reset_reason = rwi->reset_reason;
3053 
3054 	ibmvnic_cleanup(netdev);
3055 	release_resources(adapter);
3056 	release_sub_crqs(adapter, 0);
3057 	release_crq_queue(adapter);
3058 
3059 	/* remove the closed state so when we call open it appears
3060 	 * we are coming from the probed state.
3061 	 */
3062 	adapter->state = VNIC_PROBED;
3063 
3064 	reinit_init_done(adapter);
3065 
3066 	rc = init_crq_queue(adapter);
3067 	if (rc) {
3068 		netdev_err(adapter->netdev,
3069 			   "Couldn't initialize crq. rc=%d\n", rc);
3070 		goto out;
3071 	}
3072 
3073 	rc = ibmvnic_reset_init(adapter, false);
3074 	if (rc)
3075 		goto out;
3076 
3077 	/* If the adapter was in PROBE or DOWN state prior to the reset,
3078 	 * exit here.
3079 	 */
3080 	if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
3081 		goto out;
3082 
3083 	rc = ibmvnic_login(netdev);
3084 	if (rc)
3085 		goto out;
3086 
3087 	rc = init_resources(adapter);
3088 	if (rc)
3089 		goto out;
3090 
3091 	ibmvnic_disable_irqs(adapter);
3092 	adapter->state = VNIC_CLOSED;
3093 
3094 	if (reset_state == VNIC_CLOSED)
3095 		goto out;
3096 
3097 	rc = __ibmvnic_open(netdev);
3098 	if (rc) {
3099 		rc = IBMVNIC_OPEN_FAILED;
3100 		goto out;
3101 	}
3102 
3103 	__netdev_notify_peers(netdev);
3104 out:
3105 	/* restore adapter state if reset failed */
3106 	if (rc)
3107 		adapter->state = reset_state;
3108 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
3109 		   adapter_state_to_string(adapter->state),
3110 		   adapter->failover_pending, rc);
3111 	return rc;
3112 }
3113 
get_next_rwi(struct ibmvnic_adapter * adapter)3114 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
3115 {
3116 	struct ibmvnic_rwi *rwi;
3117 	unsigned long flags;
3118 
3119 	spin_lock_irqsave(&adapter->rwi_lock, flags);
3120 
3121 	if (!list_empty(&adapter->rwi_list)) {
3122 		rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
3123 				       list);
3124 		list_del(&rwi->list);
3125 	} else {
3126 		rwi = NULL;
3127 	}
3128 
3129 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3130 	return rwi;
3131 }
3132 
3133 /**
3134  * do_passive_init - complete probing when partner device is detected.
3135  * @adapter: ibmvnic_adapter struct
3136  *
3137  * If the ibmvnic device does not have a partner device to communicate with at boot
3138  * and that partner device comes online at a later time, this function is called
3139  * to complete the initialization process of ibmvnic device.
3140  * Caller is expected to hold rtnl_lock().
3141  *
3142  * Returns non-zero if sub-CRQs are not initialized properly leaving the device
3143  * in the down state.
3144  * Returns 0 upon success and the device is in PROBED state.
3145  */
3146 
do_passive_init(struct ibmvnic_adapter * adapter)3147 static int do_passive_init(struct ibmvnic_adapter *adapter)
3148 {
3149 	unsigned long timeout = msecs_to_jiffies(30000);
3150 	struct net_device *netdev = adapter->netdev;
3151 	struct device *dev = &adapter->vdev->dev;
3152 	int rc;
3153 
3154 	netdev_dbg(netdev, "Partner device found, probing.\n");
3155 
3156 	adapter->state = VNIC_PROBING;
3157 	reinit_completion(&adapter->init_done);
3158 	adapter->init_done_rc = 0;
3159 	adapter->crq.active = true;
3160 
3161 	rc = send_crq_init_complete(adapter);
3162 	if (rc)
3163 		goto out;
3164 
3165 	rc = send_version_xchg(adapter);
3166 	if (rc)
3167 		netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
3168 
3169 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
3170 		dev_err(dev, "Initialization sequence timed out\n");
3171 		rc = -ETIMEDOUT;
3172 		goto out;
3173 	}
3174 
3175 	rc = init_sub_crqs(adapter);
3176 	if (rc) {
3177 		dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
3178 		goto out;
3179 	}
3180 
3181 	rc = init_sub_crq_irqs(adapter);
3182 	if (rc) {
3183 		dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
3184 		goto init_failed;
3185 	}
3186 
3187 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
3188 	netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
3189 	netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
3190 
3191 	adapter->state = VNIC_PROBED;
3192 	netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
3193 
3194 	return 0;
3195 
3196 init_failed:
3197 	release_sub_crqs(adapter, 1);
3198 out:
3199 	adapter->state = VNIC_DOWN;
3200 	return rc;
3201 }
3202 
__ibmvnic_reset(struct work_struct * work)3203 static void __ibmvnic_reset(struct work_struct *work)
3204 {
3205 	struct ibmvnic_adapter *adapter;
3206 	unsigned int timeout = 5000;
3207 	struct ibmvnic_rwi *tmprwi;
3208 	bool saved_state = false;
3209 	struct ibmvnic_rwi *rwi;
3210 	unsigned long flags;
3211 	struct device *dev;
3212 	bool need_reset;
3213 	int num_fails = 0;
3214 	u32 reset_state;
3215 	int rc = 0;
3216 
3217 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
3218 		dev = &adapter->vdev->dev;
3219 
3220 	/* Wait for ibmvnic_probe() to complete. If probe is taking too long
3221 	 * or if another reset is in progress, defer work for now. If probe
3222 	 * eventually fails it will flush and terminate our work.
3223 	 *
3224 	 * Three possibilities here:
3225 	 * 1. Adpater being removed  - just return
3226 	 * 2. Timed out on probe or another reset in progress - delay the work
3227 	 * 3. Completed probe - perform any resets in queue
3228 	 */
3229 	if (adapter->state == VNIC_PROBING &&
3230 	    !wait_for_completion_timeout(&adapter->probe_done, timeout)) {
3231 		dev_err(dev, "Reset thread timed out on probe");
3232 		queue_delayed_work(system_long_wq,
3233 				   &adapter->ibmvnic_delayed_reset,
3234 				   IBMVNIC_RESET_DELAY);
3235 		return;
3236 	}
3237 
3238 	/* adapter is done with probe (i.e state is never VNIC_PROBING now) */
3239 	if (adapter->state == VNIC_REMOVING)
3240 		return;
3241 
3242 	/* ->rwi_list is stable now (no one else is removing entries) */
3243 
3244 	/* ibmvnic_probe() may have purged the reset queue after we were
3245 	 * scheduled to process a reset so there maybe no resets to process.
3246 	 * Before setting the ->resetting bit though, we have to make sure
3247 	 * that there is infact a reset to process. Otherwise we may race
3248 	 * with ibmvnic_open() and end up leaving the vnic down:
3249 	 *
3250 	 *	__ibmvnic_reset()	    ibmvnic_open()
3251 	 *	-----------------	    --------------
3252 	 *
3253 	 *  set ->resetting bit
3254 	 *  				find ->resetting bit is set
3255 	 *  				set ->state to IBMVNIC_OPEN (i.e
3256 	 *  				assume reset will open device)
3257 	 *  				return
3258 	 *  find reset queue empty
3259 	 *  return
3260 	 *
3261 	 *  	Neither performed vnic login/open and vnic stays down
3262 	 *
3263 	 * If we hold the lock and conditionally set the bit, either we
3264 	 * or ibmvnic_open() will complete the open.
3265 	 */
3266 	need_reset = false;
3267 	spin_lock(&adapter->rwi_lock);
3268 	if (!list_empty(&adapter->rwi_list)) {
3269 		if (test_and_set_bit_lock(0, &adapter->resetting)) {
3270 			queue_delayed_work(system_long_wq,
3271 					   &adapter->ibmvnic_delayed_reset,
3272 					   IBMVNIC_RESET_DELAY);
3273 		} else {
3274 			need_reset = true;
3275 		}
3276 	}
3277 	spin_unlock(&adapter->rwi_lock);
3278 
3279 	if (!need_reset)
3280 		return;
3281 
3282 	rwi = get_next_rwi(adapter);
3283 	while (rwi) {
3284 		spin_lock_irqsave(&adapter->state_lock, flags);
3285 
3286 		if (adapter->state == VNIC_REMOVING ||
3287 		    adapter->state == VNIC_REMOVED) {
3288 			spin_unlock_irqrestore(&adapter->state_lock, flags);
3289 			kfree(rwi);
3290 			rc = EBUSY;
3291 			break;
3292 		}
3293 
3294 		if (!saved_state) {
3295 			reset_state = adapter->state;
3296 			saved_state = true;
3297 		}
3298 		spin_unlock_irqrestore(&adapter->state_lock, flags);
3299 
3300 		if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
3301 			rtnl_lock();
3302 			rc = do_passive_init(adapter);
3303 			rtnl_unlock();
3304 			if (!rc)
3305 				netif_carrier_on(adapter->netdev);
3306 		} else if (adapter->force_reset_recovery) {
3307 			/* Since we are doing a hard reset now, clear the
3308 			 * failover_pending flag so we don't ignore any
3309 			 * future MOBILITY or other resets.
3310 			 */
3311 			adapter->failover_pending = false;
3312 
3313 			/* Transport event occurred during previous reset */
3314 			if (adapter->wait_for_reset) {
3315 				/* Previous was CHANGE_PARAM; caller locked */
3316 				adapter->force_reset_recovery = false;
3317 				rc = do_hard_reset(adapter, rwi, reset_state);
3318 			} else {
3319 				rtnl_lock();
3320 				adapter->force_reset_recovery = false;
3321 				rc = do_hard_reset(adapter, rwi, reset_state);
3322 				rtnl_unlock();
3323 			}
3324 			if (rc)
3325 				num_fails++;
3326 			else
3327 				num_fails = 0;
3328 
3329 			/* If auto-priority-failover is enabled we can get
3330 			 * back to back failovers during resets, resulting
3331 			 * in at least two failed resets (from high-priority
3332 			 * backing device to low-priority one and then back)
3333 			 * If resets continue to fail beyond that, give the
3334 			 * adapter some time to settle down before retrying.
3335 			 */
3336 			if (num_fails >= 3) {
3337 				netdev_dbg(adapter->netdev,
3338 					   "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
3339 					   adapter_state_to_string(adapter->state),
3340 					   num_fails);
3341 				set_current_state(TASK_UNINTERRUPTIBLE);
3342 				schedule_timeout(60 * HZ);
3343 			}
3344 		} else {
3345 			rc = do_reset(adapter, rwi, reset_state);
3346 		}
3347 		tmprwi = rwi;
3348 		adapter->last_reset_time = jiffies;
3349 
3350 		if (rc)
3351 			netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
3352 
3353 		rwi = get_next_rwi(adapter);
3354 
3355 		/*
3356 		 * If there are no resets queued and the previous reset failed,
3357 		 * the adapter would be in an undefined state. So retry the
3358 		 * previous reset as a hard reset.
3359 		 *
3360 		 * Else, free the previous rwi and, if there is another reset
3361 		 * queued, process the new reset even if previous reset failed
3362 		 * (the previous reset could have failed because of a fail
3363 		 * over for instance, so process the fail over).
3364 		 */
3365 		if (!rwi && rc)
3366 			rwi = tmprwi;
3367 		else
3368 			kfree(tmprwi);
3369 
3370 		if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
3371 			    rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
3372 			adapter->force_reset_recovery = true;
3373 	}
3374 
3375 	if (adapter->wait_for_reset) {
3376 		adapter->reset_done_rc = rc;
3377 		complete(&adapter->reset_done);
3378 	}
3379 
3380 	clear_bit_unlock(0, &adapter->resetting);
3381 
3382 	netdev_dbg(adapter->netdev,
3383 		   "[S:%s FRR:%d WFR:%d] Done processing resets\n",
3384 		   adapter_state_to_string(adapter->state),
3385 		   adapter->force_reset_recovery,
3386 		   adapter->wait_for_reset);
3387 }
3388 
__ibmvnic_delayed_reset(struct work_struct * work)3389 static void __ibmvnic_delayed_reset(struct work_struct *work)
3390 {
3391 	struct ibmvnic_adapter *adapter;
3392 
3393 	adapter = container_of(work, struct ibmvnic_adapter,
3394 			       ibmvnic_delayed_reset.work);
3395 	__ibmvnic_reset(&adapter->ibmvnic_reset);
3396 }
3397 
flush_reset_queue(struct ibmvnic_adapter * adapter)3398 static void flush_reset_queue(struct ibmvnic_adapter *adapter)
3399 {
3400 	struct list_head *entry, *tmp_entry;
3401 
3402 	if (!list_empty(&adapter->rwi_list)) {
3403 		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
3404 			list_del(entry);
3405 			kfree(list_entry(entry, struct ibmvnic_rwi, list));
3406 		}
3407 	}
3408 }
3409 
ibmvnic_reset(struct ibmvnic_adapter * adapter,enum ibmvnic_reset_reason reason)3410 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
3411 			 enum ibmvnic_reset_reason reason)
3412 {
3413 	struct net_device *netdev = adapter->netdev;
3414 	struct ibmvnic_rwi *rwi, *tmp;
3415 	unsigned long flags;
3416 	int ret;
3417 
3418 	spin_lock_irqsave(&adapter->rwi_lock, flags);
3419 
3420 	/* If failover is pending don't schedule any other reset.
3421 	 * Instead let the failover complete. If there is already a
3422 	 * a failover reset scheduled, we will detect and drop the
3423 	 * duplicate reset when walking the ->rwi_list below.
3424 	 */
3425 	if (adapter->state == VNIC_REMOVING ||
3426 	    adapter->state == VNIC_REMOVED ||
3427 	    (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
3428 		ret = EBUSY;
3429 		netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
3430 		goto err;
3431 	}
3432 
3433 	list_for_each_entry(tmp, &adapter->rwi_list, list) {
3434 		if (tmp->reset_reason == reason) {
3435 			netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
3436 				   reset_reason_to_string(reason));
3437 			ret = EBUSY;
3438 			goto err;
3439 		}
3440 	}
3441 
3442 	rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
3443 	if (!rwi) {
3444 		ret = ENOMEM;
3445 		goto err;
3446 	}
3447 	/* if we just received a transport event,
3448 	 * flush reset queue and process this reset
3449 	 */
3450 	if (adapter->force_reset_recovery)
3451 		flush_reset_queue(adapter);
3452 
3453 	rwi->reset_reason = reason;
3454 	list_add_tail(&rwi->list, &adapter->rwi_list);
3455 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
3456 		   reset_reason_to_string(reason));
3457 	queue_work(system_long_wq, &adapter->ibmvnic_reset);
3458 
3459 	ret = 0;
3460 err:
3461 	/* ibmvnic_close() below can block, so drop the lock first */
3462 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3463 
3464 	if (ret == ENOMEM)
3465 		ibmvnic_close(netdev);
3466 
3467 	return -ret;
3468 }
3469 
ibmvnic_tx_timeout(struct net_device * dev,unsigned int txqueue)3470 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
3471 {
3472 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3473 
3474 	if (test_bit(0, &adapter->resetting)) {
3475 		netdev_err(adapter->netdev,
3476 			   "Adapter is resetting, skip timeout reset\n");
3477 		return;
3478 	}
3479 	/* No queuing up reset until at least 5 seconds (default watchdog val)
3480 	 * after last reset
3481 	 */
3482 	if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
3483 		netdev_dbg(dev, "Not yet time to tx timeout.\n");
3484 		return;
3485 	}
3486 	ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
3487 }
3488 
remove_buff_from_pool(struct ibmvnic_adapter * adapter,struct ibmvnic_rx_buff * rx_buff)3489 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
3490 				  struct ibmvnic_rx_buff *rx_buff)
3491 {
3492 	struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
3493 
3494 	rx_buff->skb = NULL;
3495 
3496 	pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
3497 	pool->next_alloc = (pool->next_alloc + 1) % pool->size;
3498 
3499 	atomic_dec(&pool->available);
3500 }
3501 
ibmvnic_poll(struct napi_struct * napi,int budget)3502 static int ibmvnic_poll(struct napi_struct *napi, int budget)
3503 {
3504 	struct ibmvnic_sub_crq_queue *rx_scrq;
3505 	struct ibmvnic_adapter *adapter;
3506 	struct net_device *netdev;
3507 	int frames_processed;
3508 	int scrq_num;
3509 
3510 	netdev = napi->dev;
3511 	adapter = netdev_priv(netdev);
3512 	scrq_num = (int)(napi - adapter->napi);
3513 	frames_processed = 0;
3514 	rx_scrq = adapter->rx_scrq[scrq_num];
3515 
3516 restart_poll:
3517 	while (frames_processed < budget) {
3518 		struct sk_buff *skb;
3519 		struct ibmvnic_rx_buff *rx_buff;
3520 		union sub_crq *next;
3521 		u32 length;
3522 		u16 offset;
3523 		u8 flags = 0;
3524 
3525 		if (unlikely(test_bit(0, &adapter->resetting) &&
3526 			     adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
3527 			enable_scrq_irq(adapter, rx_scrq);
3528 			napi_complete_done(napi, frames_processed);
3529 			return frames_processed;
3530 		}
3531 
3532 		if (!pending_scrq(adapter, rx_scrq))
3533 			break;
3534 		next = ibmvnic_next_scrq(adapter, rx_scrq);
3535 		rx_buff = (struct ibmvnic_rx_buff *)
3536 			  be64_to_cpu(next->rx_comp.correlator);
3537 		/* do error checking */
3538 		if (next->rx_comp.rc) {
3539 			netdev_dbg(netdev, "rx buffer returned with rc %x\n",
3540 				   be16_to_cpu(next->rx_comp.rc));
3541 			/* free the entry */
3542 			next->rx_comp.first = 0;
3543 			dev_kfree_skb_any(rx_buff->skb);
3544 			remove_buff_from_pool(adapter, rx_buff);
3545 			continue;
3546 		} else if (!rx_buff->skb) {
3547 			/* free the entry */
3548 			next->rx_comp.first = 0;
3549 			remove_buff_from_pool(adapter, rx_buff);
3550 			continue;
3551 		}
3552 
3553 		length = be32_to_cpu(next->rx_comp.len);
3554 		offset = be16_to_cpu(next->rx_comp.off_frame_data);
3555 		flags = next->rx_comp.flags;
3556 		skb = rx_buff->skb;
3557 		/* load long_term_buff before copying to skb */
3558 		dma_rmb();
3559 		skb_copy_to_linear_data(skb, rx_buff->data + offset,
3560 					length);
3561 
3562 		/* VLAN Header has been stripped by the system firmware and
3563 		 * needs to be inserted by the driver
3564 		 */
3565 		if (adapter->rx_vlan_header_insertion &&
3566 		    (flags & IBMVNIC_VLAN_STRIPPED))
3567 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3568 					       ntohs(next->rx_comp.vlan_tci));
3569 
3570 		/* free the entry */
3571 		next->rx_comp.first = 0;
3572 		remove_buff_from_pool(adapter, rx_buff);
3573 
3574 		skb_put(skb, length);
3575 		skb->protocol = eth_type_trans(skb, netdev);
3576 		skb_record_rx_queue(skb, scrq_num);
3577 
3578 		if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
3579 		    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
3580 			skb->ip_summed = CHECKSUM_UNNECESSARY;
3581 		}
3582 
3583 		length = skb->len;
3584 		napi_gro_receive(napi, skb); /* send it up */
3585 		netdev->stats.rx_packets++;
3586 		netdev->stats.rx_bytes += length;
3587 		adapter->rx_stats_buffers[scrq_num].packets++;
3588 		adapter->rx_stats_buffers[scrq_num].bytes += length;
3589 		frames_processed++;
3590 	}
3591 
3592 	if (adapter->state != VNIC_CLOSING &&
3593 	    ((atomic_read(&adapter->rx_pool[scrq_num].available) <
3594 	      adapter->req_rx_add_entries_per_subcrq / 2) ||
3595 	      frames_processed < budget))
3596 		replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
3597 	if (frames_processed < budget) {
3598 		if (napi_complete_done(napi, frames_processed)) {
3599 			enable_scrq_irq(adapter, rx_scrq);
3600 			if (pending_scrq(adapter, rx_scrq)) {
3601 				if (napi_reschedule(napi)) {
3602 					disable_scrq_irq(adapter, rx_scrq);
3603 					goto restart_poll;
3604 				}
3605 			}
3606 		}
3607 	}
3608 	return frames_processed;
3609 }
3610 
wait_for_reset(struct ibmvnic_adapter * adapter)3611 static int wait_for_reset(struct ibmvnic_adapter *adapter)
3612 {
3613 	int rc, ret;
3614 
3615 	adapter->fallback.mtu = adapter->req_mtu;
3616 	adapter->fallback.rx_queues = adapter->req_rx_queues;
3617 	adapter->fallback.tx_queues = adapter->req_tx_queues;
3618 	adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
3619 	adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
3620 
3621 	reinit_completion(&adapter->reset_done);
3622 	adapter->wait_for_reset = true;
3623 	rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3624 
3625 	if (rc) {
3626 		ret = rc;
3627 		goto out;
3628 	}
3629 	rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
3630 	if (rc) {
3631 		ret = -ENODEV;
3632 		goto out;
3633 	}
3634 
3635 	ret = 0;
3636 	if (adapter->reset_done_rc) {
3637 		ret = -EIO;
3638 		adapter->desired.mtu = adapter->fallback.mtu;
3639 		adapter->desired.rx_queues = adapter->fallback.rx_queues;
3640 		adapter->desired.tx_queues = adapter->fallback.tx_queues;
3641 		adapter->desired.rx_entries = adapter->fallback.rx_entries;
3642 		adapter->desired.tx_entries = adapter->fallback.tx_entries;
3643 
3644 		reinit_completion(&adapter->reset_done);
3645 		adapter->wait_for_reset = true;
3646 		rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3647 		if (rc) {
3648 			ret = rc;
3649 			goto out;
3650 		}
3651 		rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
3652 						 60000);
3653 		if (rc) {
3654 			ret = -ENODEV;
3655 			goto out;
3656 		}
3657 	}
3658 out:
3659 	adapter->wait_for_reset = false;
3660 
3661 	return ret;
3662 }
3663 
ibmvnic_change_mtu(struct net_device * netdev,int new_mtu)3664 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
3665 {
3666 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3667 
3668 	adapter->desired.mtu = new_mtu + ETH_HLEN;
3669 
3670 	return wait_for_reset(adapter);
3671 }
3672 
ibmvnic_features_check(struct sk_buff * skb,struct net_device * dev,netdev_features_t features)3673 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3674 						struct net_device *dev,
3675 						netdev_features_t features)
3676 {
3677 	/* Some backing hardware adapters can not
3678 	 * handle packets with a MSS less than 224
3679 	 * or with only one segment.
3680 	 */
3681 	if (skb_is_gso(skb)) {
3682 		if (skb_shinfo(skb)->gso_size < 224 ||
3683 		    skb_shinfo(skb)->gso_segs == 1)
3684 			features &= ~NETIF_F_GSO_MASK;
3685 	}
3686 
3687 	return features;
3688 }
3689 
3690 static const struct net_device_ops ibmvnic_netdev_ops = {
3691 	.ndo_open		= ibmvnic_open,
3692 	.ndo_stop		= ibmvnic_close,
3693 	.ndo_start_xmit		= ibmvnic_xmit,
3694 	.ndo_set_rx_mode	= ibmvnic_set_multi,
3695 	.ndo_set_mac_address	= ibmvnic_set_mac,
3696 	.ndo_validate_addr	= eth_validate_addr,
3697 	.ndo_tx_timeout		= ibmvnic_tx_timeout,
3698 	.ndo_change_mtu		= ibmvnic_change_mtu,
3699 	.ndo_features_check     = ibmvnic_features_check,
3700 };
3701 
3702 /* ethtool functions */
3703 
ibmvnic_get_link_ksettings(struct net_device * netdev,struct ethtool_link_ksettings * cmd)3704 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3705 				      struct ethtool_link_ksettings *cmd)
3706 {
3707 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3708 	int rc;
3709 
3710 	rc = send_query_phys_parms(adapter);
3711 	if (rc) {
3712 		adapter->speed = SPEED_UNKNOWN;
3713 		adapter->duplex = DUPLEX_UNKNOWN;
3714 	}
3715 	cmd->base.speed = adapter->speed;
3716 	cmd->base.duplex = adapter->duplex;
3717 	cmd->base.port = PORT_FIBRE;
3718 	cmd->base.phy_address = 0;
3719 	cmd->base.autoneg = AUTONEG_ENABLE;
3720 
3721 	return 0;
3722 }
3723 
ibmvnic_get_drvinfo(struct net_device * netdev,struct ethtool_drvinfo * info)3724 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3725 				struct ethtool_drvinfo *info)
3726 {
3727 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3728 
3729 	strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3730 	strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3731 	strscpy(info->fw_version, adapter->fw_version,
3732 		sizeof(info->fw_version));
3733 }
3734 
ibmvnic_get_msglevel(struct net_device * netdev)3735 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3736 {
3737 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3738 
3739 	return adapter->msg_enable;
3740 }
3741 
ibmvnic_set_msglevel(struct net_device * netdev,u32 data)3742 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3743 {
3744 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3745 
3746 	adapter->msg_enable = data;
3747 }
3748 
ibmvnic_get_link(struct net_device * netdev)3749 static u32 ibmvnic_get_link(struct net_device *netdev)
3750 {
3751 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3752 
3753 	/* Don't need to send a query because we request a logical link up at
3754 	 * init and then we wait for link state indications
3755 	 */
3756 	return adapter->logical_link_state;
3757 }
3758 
ibmvnic_get_ringparam(struct net_device * netdev,struct ethtool_ringparam * ring,struct kernel_ethtool_ringparam * kernel_ring,struct netlink_ext_ack * extack)3759 static void ibmvnic_get_ringparam(struct net_device *netdev,
3760 				  struct ethtool_ringparam *ring,
3761 				  struct kernel_ethtool_ringparam *kernel_ring,
3762 				  struct netlink_ext_ack *extack)
3763 {
3764 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3765 
3766 	ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3767 	ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3768 	ring->rx_mini_max_pending = 0;
3769 	ring->rx_jumbo_max_pending = 0;
3770 	ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3771 	ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3772 	ring->rx_mini_pending = 0;
3773 	ring->rx_jumbo_pending = 0;
3774 }
3775 
ibmvnic_set_ringparam(struct net_device * netdev,struct ethtool_ringparam * ring,struct kernel_ethtool_ringparam * kernel_ring,struct netlink_ext_ack * extack)3776 static int ibmvnic_set_ringparam(struct net_device *netdev,
3777 				 struct ethtool_ringparam *ring,
3778 				 struct kernel_ethtool_ringparam *kernel_ring,
3779 				 struct netlink_ext_ack *extack)
3780 {
3781 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3782 
3783 	if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
3784 	    ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
3785 		netdev_err(netdev, "Invalid request.\n");
3786 		netdev_err(netdev, "Max tx buffers = %llu\n",
3787 			   adapter->max_rx_add_entries_per_subcrq);
3788 		netdev_err(netdev, "Max rx buffers = %llu\n",
3789 			   adapter->max_tx_entries_per_subcrq);
3790 		return -EINVAL;
3791 	}
3792 
3793 	adapter->desired.rx_entries = ring->rx_pending;
3794 	adapter->desired.tx_entries = ring->tx_pending;
3795 
3796 	return wait_for_reset(adapter);
3797 }
3798 
ibmvnic_get_channels(struct net_device * netdev,struct ethtool_channels * channels)3799 static void ibmvnic_get_channels(struct net_device *netdev,
3800 				 struct ethtool_channels *channels)
3801 {
3802 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3803 
3804 	channels->max_rx = adapter->max_rx_queues;
3805 	channels->max_tx = adapter->max_tx_queues;
3806 	channels->max_other = 0;
3807 	channels->max_combined = 0;
3808 	channels->rx_count = adapter->req_rx_queues;
3809 	channels->tx_count = adapter->req_tx_queues;
3810 	channels->other_count = 0;
3811 	channels->combined_count = 0;
3812 }
3813 
ibmvnic_set_channels(struct net_device * netdev,struct ethtool_channels * channels)3814 static int ibmvnic_set_channels(struct net_device *netdev,
3815 				struct ethtool_channels *channels)
3816 {
3817 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3818 
3819 	adapter->desired.rx_queues = channels->rx_count;
3820 	adapter->desired.tx_queues = channels->tx_count;
3821 
3822 	return wait_for_reset(adapter);
3823 }
3824 
ibmvnic_get_strings(struct net_device * dev,u32 stringset,u8 * data)3825 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3826 {
3827 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3828 	int i;
3829 
3830 	if (stringset != ETH_SS_STATS)
3831 		return;
3832 
3833 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
3834 		memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3835 
3836 	for (i = 0; i < adapter->req_tx_queues; i++) {
3837 		snprintf(data, ETH_GSTRING_LEN, "tx%d_batched_packets", i);
3838 		data += ETH_GSTRING_LEN;
3839 
3840 		snprintf(data, ETH_GSTRING_LEN, "tx%d_direct_packets", i);
3841 		data += ETH_GSTRING_LEN;
3842 
3843 		snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3844 		data += ETH_GSTRING_LEN;
3845 
3846 		snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
3847 		data += ETH_GSTRING_LEN;
3848 	}
3849 
3850 	for (i = 0; i < adapter->req_rx_queues; i++) {
3851 		snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3852 		data += ETH_GSTRING_LEN;
3853 
3854 		snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3855 		data += ETH_GSTRING_LEN;
3856 
3857 		snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3858 		data += ETH_GSTRING_LEN;
3859 	}
3860 }
3861 
ibmvnic_get_sset_count(struct net_device * dev,int sset)3862 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3863 {
3864 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3865 
3866 	switch (sset) {
3867 	case ETH_SS_STATS:
3868 		return ARRAY_SIZE(ibmvnic_stats) +
3869 		       adapter->req_tx_queues * NUM_TX_STATS +
3870 		       adapter->req_rx_queues * NUM_RX_STATS;
3871 	default:
3872 		return -EOPNOTSUPP;
3873 	}
3874 }
3875 
ibmvnic_get_ethtool_stats(struct net_device * dev,struct ethtool_stats * stats,u64 * data)3876 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3877 				      struct ethtool_stats *stats, u64 *data)
3878 {
3879 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3880 	union ibmvnic_crq crq;
3881 	int i, j;
3882 	int rc;
3883 
3884 	memset(&crq, 0, sizeof(crq));
3885 	crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3886 	crq.request_statistics.cmd = REQUEST_STATISTICS;
3887 	crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3888 	crq.request_statistics.len =
3889 	    cpu_to_be32(sizeof(struct ibmvnic_statistics));
3890 
3891 	/* Wait for data to be written */
3892 	reinit_completion(&adapter->stats_done);
3893 	rc = ibmvnic_send_crq(adapter, &crq);
3894 	if (rc)
3895 		return;
3896 	rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3897 	if (rc)
3898 		return;
3899 
3900 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3901 		data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3902 				      (adapter, ibmvnic_stats[i].offset));
3903 
3904 	for (j = 0; j < adapter->req_tx_queues; j++) {
3905 		data[i] = adapter->tx_stats_buffers[j].batched_packets;
3906 		i++;
3907 		data[i] = adapter->tx_stats_buffers[j].direct_packets;
3908 		i++;
3909 		data[i] = adapter->tx_stats_buffers[j].bytes;
3910 		i++;
3911 		data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3912 		i++;
3913 	}
3914 
3915 	for (j = 0; j < adapter->req_rx_queues; j++) {
3916 		data[i] = adapter->rx_stats_buffers[j].packets;
3917 		i++;
3918 		data[i] = adapter->rx_stats_buffers[j].bytes;
3919 		i++;
3920 		data[i] = adapter->rx_stats_buffers[j].interrupts;
3921 		i++;
3922 	}
3923 }
3924 
3925 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3926 	.get_drvinfo		= ibmvnic_get_drvinfo,
3927 	.get_msglevel		= ibmvnic_get_msglevel,
3928 	.set_msglevel		= ibmvnic_set_msglevel,
3929 	.get_link		= ibmvnic_get_link,
3930 	.get_ringparam		= ibmvnic_get_ringparam,
3931 	.set_ringparam		= ibmvnic_set_ringparam,
3932 	.get_channels		= ibmvnic_get_channels,
3933 	.set_channels		= ibmvnic_set_channels,
3934 	.get_strings            = ibmvnic_get_strings,
3935 	.get_sset_count         = ibmvnic_get_sset_count,
3936 	.get_ethtool_stats	= ibmvnic_get_ethtool_stats,
3937 	.get_link_ksettings	= ibmvnic_get_link_ksettings,
3938 };
3939 
3940 /* Routines for managing CRQs/sCRQs  */
3941 
reset_one_sub_crq_queue(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)3942 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3943 				   struct ibmvnic_sub_crq_queue *scrq)
3944 {
3945 	int rc;
3946 
3947 	if (!scrq) {
3948 		netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3949 		return -EINVAL;
3950 	}
3951 
3952 	if (scrq->irq) {
3953 		free_irq(scrq->irq, scrq);
3954 		irq_dispose_mapping(scrq->irq);
3955 		scrq->irq = 0;
3956 	}
3957 
3958 	if (scrq->msgs) {
3959 		memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3960 		atomic_set(&scrq->used, 0);
3961 		scrq->cur = 0;
3962 		scrq->ind_buf.index = 0;
3963 	} else {
3964 		netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3965 		return -EINVAL;
3966 	}
3967 
3968 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3969 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3970 	return rc;
3971 }
3972 
reset_sub_crq_queues(struct ibmvnic_adapter * adapter)3973 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3974 {
3975 	int i, rc;
3976 
3977 	if (!adapter->tx_scrq || !adapter->rx_scrq)
3978 		return -EINVAL;
3979 
3980 	ibmvnic_clean_affinity(adapter);
3981 
3982 	for (i = 0; i < adapter->req_tx_queues; i++) {
3983 		netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3984 		rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3985 		if (rc)
3986 			return rc;
3987 	}
3988 
3989 	for (i = 0; i < adapter->req_rx_queues; i++) {
3990 		netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3991 		rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3992 		if (rc)
3993 			return rc;
3994 	}
3995 
3996 	return rc;
3997 }
3998 
release_sub_crq_queue(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq,bool do_h_free)3999 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
4000 				  struct ibmvnic_sub_crq_queue *scrq,
4001 				  bool do_h_free)
4002 {
4003 	struct device *dev = &adapter->vdev->dev;
4004 	long rc;
4005 
4006 	netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
4007 
4008 	if (do_h_free) {
4009 		/* Close the sub-crqs */
4010 		do {
4011 			rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
4012 						adapter->vdev->unit_address,
4013 						scrq->crq_num);
4014 		} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4015 
4016 		if (rc) {
4017 			netdev_err(adapter->netdev,
4018 				   "Failed to release sub-CRQ %16lx, rc = %ld\n",
4019 				   scrq->crq_num, rc);
4020 		}
4021 	}
4022 
4023 	dma_free_coherent(dev,
4024 			  IBMVNIC_IND_ARR_SZ,
4025 			  scrq->ind_buf.indir_arr,
4026 			  scrq->ind_buf.indir_dma);
4027 
4028 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
4029 			 DMA_BIDIRECTIONAL);
4030 	free_pages((unsigned long)scrq->msgs, 2);
4031 	free_cpumask_var(scrq->affinity_mask);
4032 	kfree(scrq);
4033 }
4034 
init_sub_crq_queue(struct ibmvnic_adapter * adapter)4035 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
4036 							*adapter)
4037 {
4038 	struct device *dev = &adapter->vdev->dev;
4039 	struct ibmvnic_sub_crq_queue *scrq;
4040 	int rc;
4041 
4042 	scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
4043 	if (!scrq)
4044 		return NULL;
4045 
4046 	scrq->msgs =
4047 		(union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
4048 	if (!scrq->msgs) {
4049 		dev_warn(dev, "Couldn't allocate crq queue messages page\n");
4050 		goto zero_page_failed;
4051 	}
4052 	if (!zalloc_cpumask_var(&scrq->affinity_mask, GFP_KERNEL))
4053 		goto cpumask_alloc_failed;
4054 
4055 	scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
4056 					 DMA_BIDIRECTIONAL);
4057 	if (dma_mapping_error(dev, scrq->msg_token)) {
4058 		dev_warn(dev, "Couldn't map crq queue messages page\n");
4059 		goto map_failed;
4060 	}
4061 
4062 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
4063 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
4064 
4065 	if (rc == H_RESOURCE)
4066 		rc = ibmvnic_reset_crq(adapter);
4067 
4068 	if (rc == H_CLOSED) {
4069 		dev_warn(dev, "Partner adapter not ready, waiting.\n");
4070 	} else if (rc) {
4071 		dev_warn(dev, "Error %d registering sub-crq\n", rc);
4072 		goto reg_failed;
4073 	}
4074 
4075 	scrq->adapter = adapter;
4076 	scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
4077 	scrq->ind_buf.index = 0;
4078 
4079 	scrq->ind_buf.indir_arr =
4080 		dma_alloc_coherent(dev,
4081 				   IBMVNIC_IND_ARR_SZ,
4082 				   &scrq->ind_buf.indir_dma,
4083 				   GFP_KERNEL);
4084 
4085 	if (!scrq->ind_buf.indir_arr)
4086 		goto indir_failed;
4087 
4088 	spin_lock_init(&scrq->lock);
4089 
4090 	netdev_dbg(adapter->netdev,
4091 		   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
4092 		   scrq->crq_num, scrq->hw_irq, scrq->irq);
4093 
4094 	return scrq;
4095 
4096 indir_failed:
4097 	do {
4098 		rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
4099 					adapter->vdev->unit_address,
4100 					scrq->crq_num);
4101 	} while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
4102 reg_failed:
4103 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
4104 			 DMA_BIDIRECTIONAL);
4105 map_failed:
4106 	free_cpumask_var(scrq->affinity_mask);
4107 cpumask_alloc_failed:
4108 	free_pages((unsigned long)scrq->msgs, 2);
4109 zero_page_failed:
4110 	kfree(scrq);
4111 
4112 	return NULL;
4113 }
4114 
release_sub_crqs(struct ibmvnic_adapter * adapter,bool do_h_free)4115 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
4116 {
4117 	int i;
4118 
4119 	ibmvnic_clean_affinity(adapter);
4120 	if (adapter->tx_scrq) {
4121 		for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
4122 			if (!adapter->tx_scrq[i])
4123 				continue;
4124 
4125 			netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
4126 				   i);
4127 			ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
4128 			if (adapter->tx_scrq[i]->irq) {
4129 				free_irq(adapter->tx_scrq[i]->irq,
4130 					 adapter->tx_scrq[i]);
4131 				irq_dispose_mapping(adapter->tx_scrq[i]->irq);
4132 				adapter->tx_scrq[i]->irq = 0;
4133 			}
4134 
4135 			release_sub_crq_queue(adapter, adapter->tx_scrq[i],
4136 					      do_h_free);
4137 		}
4138 
4139 		kfree(adapter->tx_scrq);
4140 		adapter->tx_scrq = NULL;
4141 		adapter->num_active_tx_scrqs = 0;
4142 	}
4143 
4144 	/* Clean any remaining outstanding SKBs
4145 	 * we freed the irq so we won't be hearing
4146 	 * from them
4147 	 */
4148 	clean_tx_pools(adapter);
4149 
4150 	if (adapter->rx_scrq) {
4151 		for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
4152 			if (!adapter->rx_scrq[i])
4153 				continue;
4154 
4155 			netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
4156 				   i);
4157 			if (adapter->rx_scrq[i]->irq) {
4158 				free_irq(adapter->rx_scrq[i]->irq,
4159 					 adapter->rx_scrq[i]);
4160 				irq_dispose_mapping(adapter->rx_scrq[i]->irq);
4161 				adapter->rx_scrq[i]->irq = 0;
4162 			}
4163 
4164 			release_sub_crq_queue(adapter, adapter->rx_scrq[i],
4165 					      do_h_free);
4166 		}
4167 
4168 		kfree(adapter->rx_scrq);
4169 		adapter->rx_scrq = NULL;
4170 		adapter->num_active_rx_scrqs = 0;
4171 	}
4172 }
4173 
disable_scrq_irq(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4174 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
4175 			    struct ibmvnic_sub_crq_queue *scrq)
4176 {
4177 	struct device *dev = &adapter->vdev->dev;
4178 	unsigned long rc;
4179 
4180 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4181 				H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4182 	if (rc)
4183 		dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
4184 			scrq->hw_irq, rc);
4185 	return rc;
4186 }
4187 
4188 /* We can not use the IRQ chip EOI handler because that has the
4189  * unintended effect of changing the interrupt priority.
4190  */
ibmvnic_xics_eoi(struct device * dev,struct ibmvnic_sub_crq_queue * scrq)4191 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq)
4192 {
4193 	u64 val = 0xff000000 | scrq->hw_irq;
4194 	unsigned long rc;
4195 
4196 	rc = plpar_hcall_norets(H_EOI, val);
4197 	if (rc)
4198 		dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc);
4199 }
4200 
4201 /* Due to a firmware bug, the hypervisor can send an interrupt to a
4202  * transmit or receive queue just prior to a partition migration.
4203  * Force an EOI after migration.
4204  */
ibmvnic_clear_pending_interrupt(struct device * dev,struct ibmvnic_sub_crq_queue * scrq)4205 static void ibmvnic_clear_pending_interrupt(struct device *dev,
4206 					    struct ibmvnic_sub_crq_queue *scrq)
4207 {
4208 	if (!xive_enabled())
4209 		ibmvnic_xics_eoi(dev, scrq);
4210 }
4211 
enable_scrq_irq(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4212 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
4213 			   struct ibmvnic_sub_crq_queue *scrq)
4214 {
4215 	struct device *dev = &adapter->vdev->dev;
4216 	unsigned long rc;
4217 
4218 	if (scrq->hw_irq > 0x100000000ULL) {
4219 		dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
4220 		return 1;
4221 	}
4222 
4223 	if (test_bit(0, &adapter->resetting) &&
4224 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
4225 		ibmvnic_clear_pending_interrupt(dev, scrq);
4226 	}
4227 
4228 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4229 				H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4230 	if (rc)
4231 		dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
4232 			scrq->hw_irq, rc);
4233 	return rc;
4234 }
4235 
ibmvnic_complete_tx(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4236 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
4237 			       struct ibmvnic_sub_crq_queue *scrq)
4238 {
4239 	struct device *dev = &adapter->vdev->dev;
4240 	struct ibmvnic_tx_pool *tx_pool;
4241 	struct ibmvnic_tx_buff *txbuff;
4242 	struct netdev_queue *txq;
4243 	union sub_crq *next;
4244 	int index;
4245 	int i;
4246 
4247 restart_loop:
4248 	while (pending_scrq(adapter, scrq)) {
4249 		unsigned int pool = scrq->pool_index;
4250 		int num_entries = 0;
4251 		int total_bytes = 0;
4252 		int num_packets = 0;
4253 
4254 		next = ibmvnic_next_scrq(adapter, scrq);
4255 		for (i = 0; i < next->tx_comp.num_comps; i++) {
4256 			index = be32_to_cpu(next->tx_comp.correlators[i]);
4257 			if (index & IBMVNIC_TSO_POOL_MASK) {
4258 				tx_pool = &adapter->tso_pool[pool];
4259 				index &= ~IBMVNIC_TSO_POOL_MASK;
4260 			} else {
4261 				tx_pool = &adapter->tx_pool[pool];
4262 			}
4263 
4264 			txbuff = &tx_pool->tx_buff[index];
4265 			num_packets++;
4266 			num_entries += txbuff->num_entries;
4267 			if (txbuff->skb) {
4268 				total_bytes += txbuff->skb->len;
4269 				if (next->tx_comp.rcs[i]) {
4270 					dev_err(dev, "tx error %x\n",
4271 						next->tx_comp.rcs[i]);
4272 					dev_kfree_skb_irq(txbuff->skb);
4273 				} else {
4274 					dev_consume_skb_irq(txbuff->skb);
4275 				}
4276 				txbuff->skb = NULL;
4277 			} else {
4278 				netdev_warn(adapter->netdev,
4279 					    "TX completion received with NULL socket buffer\n");
4280 			}
4281 			tx_pool->free_map[tx_pool->producer_index] = index;
4282 			tx_pool->producer_index =
4283 				(tx_pool->producer_index + 1) %
4284 					tx_pool->num_buffers;
4285 		}
4286 		/* remove tx_comp scrq*/
4287 		next->tx_comp.first = 0;
4288 
4289 		txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
4290 		netdev_tx_completed_queue(txq, num_packets, total_bytes);
4291 
4292 		if (atomic_sub_return(num_entries, &scrq->used) <=
4293 		    (adapter->req_tx_entries_per_subcrq / 2) &&
4294 		    __netif_subqueue_stopped(adapter->netdev,
4295 					     scrq->pool_index)) {
4296 			rcu_read_lock();
4297 			if (adapter->tx_queues_active) {
4298 				netif_wake_subqueue(adapter->netdev,
4299 						    scrq->pool_index);
4300 				netdev_dbg(adapter->netdev,
4301 					   "Started queue %d\n",
4302 					   scrq->pool_index);
4303 			}
4304 			rcu_read_unlock();
4305 		}
4306 	}
4307 
4308 	enable_scrq_irq(adapter, scrq);
4309 
4310 	if (pending_scrq(adapter, scrq)) {
4311 		disable_scrq_irq(adapter, scrq);
4312 		goto restart_loop;
4313 	}
4314 
4315 	return 0;
4316 }
4317 
ibmvnic_interrupt_tx(int irq,void * instance)4318 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
4319 {
4320 	struct ibmvnic_sub_crq_queue *scrq = instance;
4321 	struct ibmvnic_adapter *adapter = scrq->adapter;
4322 
4323 	disable_scrq_irq(adapter, scrq);
4324 	ibmvnic_complete_tx(adapter, scrq);
4325 
4326 	return IRQ_HANDLED;
4327 }
4328 
ibmvnic_interrupt_rx(int irq,void * instance)4329 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
4330 {
4331 	struct ibmvnic_sub_crq_queue *scrq = instance;
4332 	struct ibmvnic_adapter *adapter = scrq->adapter;
4333 
4334 	/* When booting a kdump kernel we can hit pending interrupts
4335 	 * prior to completing driver initialization.
4336 	 */
4337 	if (unlikely(adapter->state != VNIC_OPEN))
4338 		return IRQ_NONE;
4339 
4340 	adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
4341 
4342 	if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
4343 		disable_scrq_irq(adapter, scrq);
4344 		__napi_schedule(&adapter->napi[scrq->scrq_num]);
4345 	}
4346 
4347 	return IRQ_HANDLED;
4348 }
4349 
init_sub_crq_irqs(struct ibmvnic_adapter * adapter)4350 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
4351 {
4352 	struct device *dev = &adapter->vdev->dev;
4353 	struct ibmvnic_sub_crq_queue *scrq;
4354 	int i = 0, j = 0;
4355 	int rc = 0;
4356 
4357 	for (i = 0; i < adapter->req_tx_queues; i++) {
4358 		netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
4359 			   i);
4360 		scrq = adapter->tx_scrq[i];
4361 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4362 
4363 		if (!scrq->irq) {
4364 			rc = -EINVAL;
4365 			dev_err(dev, "Error mapping irq\n");
4366 			goto req_tx_irq_failed;
4367 		}
4368 
4369 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
4370 			 adapter->vdev->unit_address, i);
4371 		rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
4372 				 0, scrq->name, scrq);
4373 
4374 		if (rc) {
4375 			dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
4376 				scrq->irq, rc);
4377 			irq_dispose_mapping(scrq->irq);
4378 			goto req_tx_irq_failed;
4379 		}
4380 	}
4381 
4382 	for (i = 0; i < adapter->req_rx_queues; i++) {
4383 		netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
4384 			   i);
4385 		scrq = adapter->rx_scrq[i];
4386 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4387 		if (!scrq->irq) {
4388 			rc = -EINVAL;
4389 			dev_err(dev, "Error mapping irq\n");
4390 			goto req_rx_irq_failed;
4391 		}
4392 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
4393 			 adapter->vdev->unit_address, i);
4394 		rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
4395 				 0, scrq->name, scrq);
4396 		if (rc) {
4397 			dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
4398 				scrq->irq, rc);
4399 			irq_dispose_mapping(scrq->irq);
4400 			goto req_rx_irq_failed;
4401 		}
4402 	}
4403 
4404 	cpus_read_lock();
4405 	ibmvnic_set_affinity(adapter);
4406 	cpus_read_unlock();
4407 
4408 	return rc;
4409 
4410 req_rx_irq_failed:
4411 	for (j = 0; j < i; j++) {
4412 		free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
4413 		irq_dispose_mapping(adapter->rx_scrq[j]->irq);
4414 	}
4415 	i = adapter->req_tx_queues;
4416 req_tx_irq_failed:
4417 	for (j = 0; j < i; j++) {
4418 		free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
4419 		irq_dispose_mapping(adapter->tx_scrq[j]->irq);
4420 	}
4421 	release_sub_crqs(adapter, 1);
4422 	return rc;
4423 }
4424 
init_sub_crqs(struct ibmvnic_adapter * adapter)4425 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
4426 {
4427 	struct device *dev = &adapter->vdev->dev;
4428 	struct ibmvnic_sub_crq_queue **allqueues;
4429 	int registered_queues = 0;
4430 	int total_queues;
4431 	int more = 0;
4432 	int i;
4433 
4434 	total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
4435 
4436 	allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
4437 	if (!allqueues)
4438 		return -ENOMEM;
4439 
4440 	for (i = 0; i < total_queues; i++) {
4441 		allqueues[i] = init_sub_crq_queue(adapter);
4442 		if (!allqueues[i]) {
4443 			dev_warn(dev, "Couldn't allocate all sub-crqs\n");
4444 			break;
4445 		}
4446 		registered_queues++;
4447 	}
4448 
4449 	/* Make sure we were able to register the minimum number of queues */
4450 	if (registered_queues <
4451 	    adapter->min_tx_queues + adapter->min_rx_queues) {
4452 		dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
4453 		goto tx_failed;
4454 	}
4455 
4456 	/* Distribute the failed allocated queues*/
4457 	for (i = 0; i < total_queues - registered_queues + more ; i++) {
4458 		netdev_dbg(adapter->netdev, "Reducing number of queues\n");
4459 		switch (i % 3) {
4460 		case 0:
4461 			if (adapter->req_rx_queues > adapter->min_rx_queues)
4462 				adapter->req_rx_queues--;
4463 			else
4464 				more++;
4465 			break;
4466 		case 1:
4467 			if (adapter->req_tx_queues > adapter->min_tx_queues)
4468 				adapter->req_tx_queues--;
4469 			else
4470 				more++;
4471 			break;
4472 		}
4473 	}
4474 
4475 	adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
4476 				   sizeof(*adapter->tx_scrq), GFP_KERNEL);
4477 	if (!adapter->tx_scrq)
4478 		goto tx_failed;
4479 
4480 	for (i = 0; i < adapter->req_tx_queues; i++) {
4481 		adapter->tx_scrq[i] = allqueues[i];
4482 		adapter->tx_scrq[i]->pool_index = i;
4483 		adapter->num_active_tx_scrqs++;
4484 	}
4485 
4486 	adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
4487 				   sizeof(*adapter->rx_scrq), GFP_KERNEL);
4488 	if (!adapter->rx_scrq)
4489 		goto rx_failed;
4490 
4491 	for (i = 0; i < adapter->req_rx_queues; i++) {
4492 		adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
4493 		adapter->rx_scrq[i]->scrq_num = i;
4494 		adapter->num_active_rx_scrqs++;
4495 	}
4496 
4497 	kfree(allqueues);
4498 	return 0;
4499 
4500 rx_failed:
4501 	kfree(adapter->tx_scrq);
4502 	adapter->tx_scrq = NULL;
4503 tx_failed:
4504 	for (i = 0; i < registered_queues; i++)
4505 		release_sub_crq_queue(adapter, allqueues[i], 1);
4506 	kfree(allqueues);
4507 	return -ENOMEM;
4508 }
4509 
send_request_cap(struct ibmvnic_adapter * adapter,int retry)4510 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
4511 {
4512 	struct device *dev = &adapter->vdev->dev;
4513 	union ibmvnic_crq crq;
4514 	int max_entries;
4515 	int cap_reqs;
4516 
4517 	/* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
4518 	 * the PROMISC flag). Initialize this count upfront. When the tasklet
4519 	 * receives a response to all of these, it will send the next protocol
4520 	 * message (QUERY_IP_OFFLOAD).
4521 	 */
4522 	if (!(adapter->netdev->flags & IFF_PROMISC) ||
4523 	    adapter->promisc_supported)
4524 		cap_reqs = 7;
4525 	else
4526 		cap_reqs = 6;
4527 
4528 	if (!retry) {
4529 		/* Sub-CRQ entries are 32 byte long */
4530 		int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
4531 
4532 		atomic_set(&adapter->running_cap_crqs, cap_reqs);
4533 
4534 		if (adapter->min_tx_entries_per_subcrq > entries_page ||
4535 		    adapter->min_rx_add_entries_per_subcrq > entries_page) {
4536 			dev_err(dev, "Fatal, invalid entries per sub-crq\n");
4537 			return;
4538 		}
4539 
4540 		if (adapter->desired.mtu)
4541 			adapter->req_mtu = adapter->desired.mtu;
4542 		else
4543 			adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
4544 
4545 		if (!adapter->desired.tx_entries)
4546 			adapter->desired.tx_entries =
4547 					adapter->max_tx_entries_per_subcrq;
4548 		if (!adapter->desired.rx_entries)
4549 			adapter->desired.rx_entries =
4550 					adapter->max_rx_add_entries_per_subcrq;
4551 
4552 		max_entries = IBMVNIC_LTB_SET_SIZE /
4553 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
4554 
4555 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4556 			adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) {
4557 			adapter->desired.tx_entries = max_entries;
4558 		}
4559 
4560 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4561 			adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) {
4562 			adapter->desired.rx_entries = max_entries;
4563 		}
4564 
4565 		if (adapter->desired.tx_entries)
4566 			adapter->req_tx_entries_per_subcrq =
4567 					adapter->desired.tx_entries;
4568 		else
4569 			adapter->req_tx_entries_per_subcrq =
4570 					adapter->max_tx_entries_per_subcrq;
4571 
4572 		if (adapter->desired.rx_entries)
4573 			adapter->req_rx_add_entries_per_subcrq =
4574 					adapter->desired.rx_entries;
4575 		else
4576 			adapter->req_rx_add_entries_per_subcrq =
4577 					adapter->max_rx_add_entries_per_subcrq;
4578 
4579 		if (adapter->desired.tx_queues)
4580 			adapter->req_tx_queues =
4581 					adapter->desired.tx_queues;
4582 		else
4583 			adapter->req_tx_queues =
4584 					adapter->opt_tx_comp_sub_queues;
4585 
4586 		if (adapter->desired.rx_queues)
4587 			adapter->req_rx_queues =
4588 					adapter->desired.rx_queues;
4589 		else
4590 			adapter->req_rx_queues =
4591 					adapter->opt_rx_comp_queues;
4592 
4593 		adapter->req_rx_add_queues = adapter->max_rx_add_queues;
4594 	} else {
4595 		atomic_add(cap_reqs, &adapter->running_cap_crqs);
4596 	}
4597 	memset(&crq, 0, sizeof(crq));
4598 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
4599 	crq.request_capability.cmd = REQUEST_CAPABILITY;
4600 
4601 	crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
4602 	crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
4603 	cap_reqs--;
4604 	ibmvnic_send_crq(adapter, &crq);
4605 
4606 	crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
4607 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
4608 	cap_reqs--;
4609 	ibmvnic_send_crq(adapter, &crq);
4610 
4611 	crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
4612 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
4613 	cap_reqs--;
4614 	ibmvnic_send_crq(adapter, &crq);
4615 
4616 	crq.request_capability.capability =
4617 	    cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
4618 	crq.request_capability.number =
4619 	    cpu_to_be64(adapter->req_tx_entries_per_subcrq);
4620 	cap_reqs--;
4621 	ibmvnic_send_crq(adapter, &crq);
4622 
4623 	crq.request_capability.capability =
4624 	    cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
4625 	crq.request_capability.number =
4626 	    cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
4627 	cap_reqs--;
4628 	ibmvnic_send_crq(adapter, &crq);
4629 
4630 	crq.request_capability.capability = cpu_to_be16(REQ_MTU);
4631 	crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
4632 	cap_reqs--;
4633 	ibmvnic_send_crq(adapter, &crq);
4634 
4635 	if (adapter->netdev->flags & IFF_PROMISC) {
4636 		if (adapter->promisc_supported) {
4637 			crq.request_capability.capability =
4638 			    cpu_to_be16(PROMISC_REQUESTED);
4639 			crq.request_capability.number = cpu_to_be64(1);
4640 			cap_reqs--;
4641 			ibmvnic_send_crq(adapter, &crq);
4642 		}
4643 	} else {
4644 		crq.request_capability.capability =
4645 		    cpu_to_be16(PROMISC_REQUESTED);
4646 		crq.request_capability.number = cpu_to_be64(0);
4647 		cap_reqs--;
4648 		ibmvnic_send_crq(adapter, &crq);
4649 	}
4650 
4651 	/* Keep at end to catch any discrepancy between expected and actual
4652 	 * CRQs sent.
4653 	 */
4654 	WARN_ON(cap_reqs != 0);
4655 }
4656 
pending_scrq(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4657 static int pending_scrq(struct ibmvnic_adapter *adapter,
4658 			struct ibmvnic_sub_crq_queue *scrq)
4659 {
4660 	union sub_crq *entry = &scrq->msgs[scrq->cur];
4661 	int rc;
4662 
4663 	rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4664 
4665 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4666 	 * contents of the SCRQ descriptor
4667 	 */
4668 	dma_rmb();
4669 
4670 	return rc;
4671 }
4672 
ibmvnic_next_scrq(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4673 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4674 					struct ibmvnic_sub_crq_queue *scrq)
4675 {
4676 	union sub_crq *entry;
4677 	unsigned long flags;
4678 
4679 	spin_lock_irqsave(&scrq->lock, flags);
4680 	entry = &scrq->msgs[scrq->cur];
4681 	if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4682 		if (++scrq->cur == scrq->size)
4683 			scrq->cur = 0;
4684 	} else {
4685 		entry = NULL;
4686 	}
4687 	spin_unlock_irqrestore(&scrq->lock, flags);
4688 
4689 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4690 	 * contents of the SCRQ descriptor
4691 	 */
4692 	dma_rmb();
4693 
4694 	return entry;
4695 }
4696 
ibmvnic_next_crq(struct ibmvnic_adapter * adapter)4697 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4698 {
4699 	struct ibmvnic_crq_queue *queue = &adapter->crq;
4700 	union ibmvnic_crq *crq;
4701 
4702 	crq = &queue->msgs[queue->cur];
4703 	if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4704 		if (++queue->cur == queue->size)
4705 			queue->cur = 0;
4706 	} else {
4707 		crq = NULL;
4708 	}
4709 
4710 	return crq;
4711 }
4712 
print_subcrq_error(struct device * dev,int rc,const char * func)4713 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4714 {
4715 	switch (rc) {
4716 	case H_PARAMETER:
4717 		dev_warn_ratelimited(dev,
4718 				     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4719 				     func, rc);
4720 		break;
4721 	case H_CLOSED:
4722 		dev_warn_ratelimited(dev,
4723 				     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4724 				     func, rc);
4725 		break;
4726 	default:
4727 		dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4728 		break;
4729 	}
4730 }
4731 
send_subcrq_indirect(struct ibmvnic_adapter * adapter,u64 remote_handle,u64 ioba,u64 num_entries)4732 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4733 				u64 remote_handle, u64 ioba, u64 num_entries)
4734 {
4735 	unsigned int ua = adapter->vdev->unit_address;
4736 	struct device *dev = &adapter->vdev->dev;
4737 	int rc;
4738 
4739 	/* Make sure the hypervisor sees the complete request */
4740 	dma_wmb();
4741 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4742 				cpu_to_be64(remote_handle),
4743 				ioba, num_entries);
4744 
4745 	if (rc)
4746 		print_subcrq_error(dev, rc, __func__);
4747 
4748 	return rc;
4749 }
4750 
ibmvnic_send_crq(struct ibmvnic_adapter * adapter,union ibmvnic_crq * crq)4751 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4752 			    union ibmvnic_crq *crq)
4753 {
4754 	unsigned int ua = adapter->vdev->unit_address;
4755 	struct device *dev = &adapter->vdev->dev;
4756 	u64 *u64_crq = (u64 *)crq;
4757 	int rc;
4758 
4759 	netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4760 		   (unsigned long)cpu_to_be64(u64_crq[0]),
4761 		   (unsigned long)cpu_to_be64(u64_crq[1]));
4762 
4763 	if (!adapter->crq.active &&
4764 	    crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4765 		dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4766 		return -EINVAL;
4767 	}
4768 
4769 	/* Make sure the hypervisor sees the complete request */
4770 	dma_wmb();
4771 
4772 	rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4773 				cpu_to_be64(u64_crq[0]),
4774 				cpu_to_be64(u64_crq[1]));
4775 
4776 	if (rc) {
4777 		if (rc == H_CLOSED) {
4778 			dev_warn(dev, "CRQ Queue closed\n");
4779 			/* do not reset, report the fail, wait for passive init from server */
4780 		}
4781 
4782 		dev_warn(dev, "Send error (rc=%d)\n", rc);
4783 	}
4784 
4785 	return rc;
4786 }
4787 
ibmvnic_send_crq_init(struct ibmvnic_adapter * adapter)4788 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4789 {
4790 	struct device *dev = &adapter->vdev->dev;
4791 	union ibmvnic_crq crq;
4792 	int retries = 100;
4793 	int rc;
4794 
4795 	memset(&crq, 0, sizeof(crq));
4796 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4797 	crq.generic.cmd = IBMVNIC_CRQ_INIT;
4798 	netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4799 
4800 	do {
4801 		rc = ibmvnic_send_crq(adapter, &crq);
4802 		if (rc != H_CLOSED)
4803 			break;
4804 		retries--;
4805 		msleep(50);
4806 
4807 	} while (retries > 0);
4808 
4809 	if (rc) {
4810 		dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4811 		return rc;
4812 	}
4813 
4814 	return 0;
4815 }
4816 
4817 struct vnic_login_client_data {
4818 	u8	type;
4819 	__be16	len;
4820 	char	name[];
4821 } __packed;
4822 
vnic_client_data_len(struct ibmvnic_adapter * adapter)4823 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4824 {
4825 	int len;
4826 
4827 	/* Calculate the amount of buffer space needed for the
4828 	 * vnic client data in the login buffer. There are four entries,
4829 	 * OS name, LPAR name, device name, and a null last entry.
4830 	 */
4831 	len = 4 * sizeof(struct vnic_login_client_data);
4832 	len += 6; /* "Linux" plus NULL */
4833 	len += strlen(utsname()->nodename) + 1;
4834 	len += strlen(adapter->netdev->name) + 1;
4835 
4836 	return len;
4837 }
4838 
vnic_add_client_data(struct ibmvnic_adapter * adapter,struct vnic_login_client_data * vlcd)4839 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4840 				 struct vnic_login_client_data *vlcd)
4841 {
4842 	const char *os_name = "Linux";
4843 	int len;
4844 
4845 	/* Type 1 - LPAR OS */
4846 	vlcd->type = 1;
4847 	len = strlen(os_name) + 1;
4848 	vlcd->len = cpu_to_be16(len);
4849 	strscpy(vlcd->name, os_name, len);
4850 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4851 
4852 	/* Type 2 - LPAR name */
4853 	vlcd->type = 2;
4854 	len = strlen(utsname()->nodename) + 1;
4855 	vlcd->len = cpu_to_be16(len);
4856 	strscpy(vlcd->name, utsname()->nodename, len);
4857 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4858 
4859 	/* Type 3 - device name */
4860 	vlcd->type = 3;
4861 	len = strlen(adapter->netdev->name) + 1;
4862 	vlcd->len = cpu_to_be16(len);
4863 	strscpy(vlcd->name, adapter->netdev->name, len);
4864 }
4865 
send_login(struct ibmvnic_adapter * adapter)4866 static int send_login(struct ibmvnic_adapter *adapter)
4867 {
4868 	struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4869 	struct ibmvnic_login_buffer *login_buffer;
4870 	struct device *dev = &adapter->vdev->dev;
4871 	struct vnic_login_client_data *vlcd;
4872 	dma_addr_t rsp_buffer_token;
4873 	dma_addr_t buffer_token;
4874 	size_t rsp_buffer_size;
4875 	union ibmvnic_crq crq;
4876 	int client_data_len;
4877 	size_t buffer_size;
4878 	__be64 *tx_list_p;
4879 	__be64 *rx_list_p;
4880 	int rc;
4881 	int i;
4882 
4883 	if (!adapter->tx_scrq || !adapter->rx_scrq) {
4884 		netdev_err(adapter->netdev,
4885 			   "RX or TX queues are not allocated, device login failed\n");
4886 		return -ENOMEM;
4887 	}
4888 
4889 	release_login_buffer(adapter);
4890 	release_login_rsp_buffer(adapter);
4891 
4892 	client_data_len = vnic_client_data_len(adapter);
4893 
4894 	buffer_size =
4895 	    sizeof(struct ibmvnic_login_buffer) +
4896 	    sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4897 	    client_data_len;
4898 
4899 	login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4900 	if (!login_buffer)
4901 		goto buf_alloc_failed;
4902 
4903 	buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4904 				      DMA_TO_DEVICE);
4905 	if (dma_mapping_error(dev, buffer_token)) {
4906 		dev_err(dev, "Couldn't map login buffer\n");
4907 		goto buf_map_failed;
4908 	}
4909 
4910 	rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4911 			  sizeof(u64) * adapter->req_tx_queues +
4912 			  sizeof(u64) * adapter->req_rx_queues +
4913 			  sizeof(u64) * adapter->req_rx_queues +
4914 			  sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4915 
4916 	login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4917 	if (!login_rsp_buffer)
4918 		goto buf_rsp_alloc_failed;
4919 
4920 	rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4921 					  rsp_buffer_size, DMA_FROM_DEVICE);
4922 	if (dma_mapping_error(dev, rsp_buffer_token)) {
4923 		dev_err(dev, "Couldn't map login rsp buffer\n");
4924 		goto buf_rsp_map_failed;
4925 	}
4926 
4927 	adapter->login_buf = login_buffer;
4928 	adapter->login_buf_token = buffer_token;
4929 	adapter->login_buf_sz = buffer_size;
4930 	adapter->login_rsp_buf = login_rsp_buffer;
4931 	adapter->login_rsp_buf_token = rsp_buffer_token;
4932 	adapter->login_rsp_buf_sz = rsp_buffer_size;
4933 
4934 	login_buffer->len = cpu_to_be32(buffer_size);
4935 	login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4936 	login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4937 	login_buffer->off_txcomp_subcrqs =
4938 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4939 	login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4940 	login_buffer->off_rxcomp_subcrqs =
4941 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4942 			sizeof(u64) * adapter->req_tx_queues);
4943 	login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4944 	login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4945 
4946 	tx_list_p = (__be64 *)((char *)login_buffer +
4947 				      sizeof(struct ibmvnic_login_buffer));
4948 	rx_list_p = (__be64 *)((char *)login_buffer +
4949 				      sizeof(struct ibmvnic_login_buffer) +
4950 				      sizeof(u64) * adapter->req_tx_queues);
4951 
4952 	for (i = 0; i < adapter->req_tx_queues; i++) {
4953 		if (adapter->tx_scrq[i]) {
4954 			tx_list_p[i] =
4955 				cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4956 		}
4957 	}
4958 
4959 	for (i = 0; i < adapter->req_rx_queues; i++) {
4960 		if (adapter->rx_scrq[i]) {
4961 			rx_list_p[i] =
4962 				cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4963 		}
4964 	}
4965 
4966 	/* Insert vNIC login client data */
4967 	vlcd = (struct vnic_login_client_data *)
4968 		((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4969 	login_buffer->client_data_offset =
4970 			cpu_to_be32((char *)vlcd - (char *)login_buffer);
4971 	login_buffer->client_data_len = cpu_to_be32(client_data_len);
4972 
4973 	vnic_add_client_data(adapter, vlcd);
4974 
4975 	netdev_dbg(adapter->netdev, "Login Buffer:\n");
4976 	for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4977 		netdev_dbg(adapter->netdev, "%016lx\n",
4978 			   ((unsigned long *)(adapter->login_buf))[i]);
4979 	}
4980 
4981 	memset(&crq, 0, sizeof(crq));
4982 	crq.login.first = IBMVNIC_CRQ_CMD;
4983 	crq.login.cmd = LOGIN;
4984 	crq.login.ioba = cpu_to_be32(buffer_token);
4985 	crq.login.len = cpu_to_be32(buffer_size);
4986 
4987 	adapter->login_pending = true;
4988 	rc = ibmvnic_send_crq(adapter, &crq);
4989 	if (rc) {
4990 		adapter->login_pending = false;
4991 		netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4992 		goto buf_send_failed;
4993 	}
4994 
4995 	return 0;
4996 
4997 buf_send_failed:
4998 	dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size,
4999 			 DMA_FROM_DEVICE);
5000 buf_rsp_map_failed:
5001 	kfree(login_rsp_buffer);
5002 	adapter->login_rsp_buf = NULL;
5003 buf_rsp_alloc_failed:
5004 	dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
5005 buf_map_failed:
5006 	kfree(login_buffer);
5007 	adapter->login_buf = NULL;
5008 buf_alloc_failed:
5009 	return -ENOMEM;
5010 }
5011 
send_request_map(struct ibmvnic_adapter * adapter,dma_addr_t addr,u32 len,u8 map_id)5012 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
5013 			    u32 len, u8 map_id)
5014 {
5015 	union ibmvnic_crq crq;
5016 
5017 	memset(&crq, 0, sizeof(crq));
5018 	crq.request_map.first = IBMVNIC_CRQ_CMD;
5019 	crq.request_map.cmd = REQUEST_MAP;
5020 	crq.request_map.map_id = map_id;
5021 	crq.request_map.ioba = cpu_to_be32(addr);
5022 	crq.request_map.len = cpu_to_be32(len);
5023 	return ibmvnic_send_crq(adapter, &crq);
5024 }
5025 
send_request_unmap(struct ibmvnic_adapter * adapter,u8 map_id)5026 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
5027 {
5028 	union ibmvnic_crq crq;
5029 
5030 	memset(&crq, 0, sizeof(crq));
5031 	crq.request_unmap.first = IBMVNIC_CRQ_CMD;
5032 	crq.request_unmap.cmd = REQUEST_UNMAP;
5033 	crq.request_unmap.map_id = map_id;
5034 	return ibmvnic_send_crq(adapter, &crq);
5035 }
5036 
send_query_map(struct ibmvnic_adapter * adapter)5037 static void send_query_map(struct ibmvnic_adapter *adapter)
5038 {
5039 	union ibmvnic_crq crq;
5040 
5041 	memset(&crq, 0, sizeof(crq));
5042 	crq.query_map.first = IBMVNIC_CRQ_CMD;
5043 	crq.query_map.cmd = QUERY_MAP;
5044 	ibmvnic_send_crq(adapter, &crq);
5045 }
5046 
5047 /* Send a series of CRQs requesting various capabilities of the VNIC server */
send_query_cap(struct ibmvnic_adapter * adapter)5048 static void send_query_cap(struct ibmvnic_adapter *adapter)
5049 {
5050 	union ibmvnic_crq crq;
5051 	int cap_reqs;
5052 
5053 	/* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
5054 	 * upfront. When the tasklet receives a response to all of these, it
5055 	 * can send out the next protocol messaage (REQUEST_CAPABILITY).
5056 	 */
5057 	cap_reqs = 25;
5058 
5059 	atomic_set(&adapter->running_cap_crqs, cap_reqs);
5060 
5061 	memset(&crq, 0, sizeof(crq));
5062 	crq.query_capability.first = IBMVNIC_CRQ_CMD;
5063 	crq.query_capability.cmd = QUERY_CAPABILITY;
5064 
5065 	crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
5066 	ibmvnic_send_crq(adapter, &crq);
5067 	cap_reqs--;
5068 
5069 	crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
5070 	ibmvnic_send_crq(adapter, &crq);
5071 	cap_reqs--;
5072 
5073 	crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
5074 	ibmvnic_send_crq(adapter, &crq);
5075 	cap_reqs--;
5076 
5077 	crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
5078 	ibmvnic_send_crq(adapter, &crq);
5079 	cap_reqs--;
5080 
5081 	crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
5082 	ibmvnic_send_crq(adapter, &crq);
5083 	cap_reqs--;
5084 
5085 	crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
5086 	ibmvnic_send_crq(adapter, &crq);
5087 	cap_reqs--;
5088 
5089 	crq.query_capability.capability =
5090 	    cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
5091 	ibmvnic_send_crq(adapter, &crq);
5092 	cap_reqs--;
5093 
5094 	crq.query_capability.capability =
5095 	    cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
5096 	ibmvnic_send_crq(adapter, &crq);
5097 	cap_reqs--;
5098 
5099 	crq.query_capability.capability =
5100 	    cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
5101 	ibmvnic_send_crq(adapter, &crq);
5102 	cap_reqs--;
5103 
5104 	crq.query_capability.capability =
5105 	    cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
5106 	ibmvnic_send_crq(adapter, &crq);
5107 	cap_reqs--;
5108 
5109 	crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
5110 	ibmvnic_send_crq(adapter, &crq);
5111 	cap_reqs--;
5112 
5113 	crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
5114 	ibmvnic_send_crq(adapter, &crq);
5115 	cap_reqs--;
5116 
5117 	crq.query_capability.capability = cpu_to_be16(MIN_MTU);
5118 	ibmvnic_send_crq(adapter, &crq);
5119 	cap_reqs--;
5120 
5121 	crq.query_capability.capability = cpu_to_be16(MAX_MTU);
5122 	ibmvnic_send_crq(adapter, &crq);
5123 	cap_reqs--;
5124 
5125 	crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
5126 	ibmvnic_send_crq(adapter, &crq);
5127 	cap_reqs--;
5128 
5129 	crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
5130 	ibmvnic_send_crq(adapter, &crq);
5131 	cap_reqs--;
5132 
5133 	crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
5134 	ibmvnic_send_crq(adapter, &crq);
5135 	cap_reqs--;
5136 
5137 	crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
5138 	ibmvnic_send_crq(adapter, &crq);
5139 	cap_reqs--;
5140 
5141 	crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
5142 	ibmvnic_send_crq(adapter, &crq);
5143 	cap_reqs--;
5144 
5145 	crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
5146 	ibmvnic_send_crq(adapter, &crq);
5147 	cap_reqs--;
5148 
5149 	crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
5150 	ibmvnic_send_crq(adapter, &crq);
5151 	cap_reqs--;
5152 
5153 	crq.query_capability.capability =
5154 			cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
5155 	ibmvnic_send_crq(adapter, &crq);
5156 	cap_reqs--;
5157 
5158 	crq.query_capability.capability =
5159 			cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
5160 	ibmvnic_send_crq(adapter, &crq);
5161 	cap_reqs--;
5162 
5163 	crq.query_capability.capability =
5164 			cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
5165 	ibmvnic_send_crq(adapter, &crq);
5166 	cap_reqs--;
5167 
5168 	crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
5169 
5170 	ibmvnic_send_crq(adapter, &crq);
5171 	cap_reqs--;
5172 
5173 	/* Keep at end to catch any discrepancy between expected and actual
5174 	 * CRQs sent.
5175 	 */
5176 	WARN_ON(cap_reqs != 0);
5177 }
5178 
send_query_ip_offload(struct ibmvnic_adapter * adapter)5179 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
5180 {
5181 	int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
5182 	struct device *dev = &adapter->vdev->dev;
5183 	union ibmvnic_crq crq;
5184 
5185 	adapter->ip_offload_tok =
5186 		dma_map_single(dev,
5187 			       &adapter->ip_offload_buf,
5188 			       buf_sz,
5189 			       DMA_FROM_DEVICE);
5190 
5191 	if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
5192 		if (!firmware_has_feature(FW_FEATURE_CMO))
5193 			dev_err(dev, "Couldn't map offload buffer\n");
5194 		return;
5195 	}
5196 
5197 	memset(&crq, 0, sizeof(crq));
5198 	crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
5199 	crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
5200 	crq.query_ip_offload.len = cpu_to_be32(buf_sz);
5201 	crq.query_ip_offload.ioba =
5202 	    cpu_to_be32(adapter->ip_offload_tok);
5203 
5204 	ibmvnic_send_crq(adapter, &crq);
5205 }
5206 
send_control_ip_offload(struct ibmvnic_adapter * adapter)5207 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
5208 {
5209 	struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
5210 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5211 	struct device *dev = &adapter->vdev->dev;
5212 	netdev_features_t old_hw_features = 0;
5213 	union ibmvnic_crq crq;
5214 
5215 	adapter->ip_offload_ctrl_tok =
5216 		dma_map_single(dev,
5217 			       ctrl_buf,
5218 			       sizeof(adapter->ip_offload_ctrl),
5219 			       DMA_TO_DEVICE);
5220 
5221 	if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
5222 		dev_err(dev, "Couldn't map ip offload control buffer\n");
5223 		return;
5224 	}
5225 
5226 	ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5227 	ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
5228 	ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
5229 	ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
5230 	ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
5231 	ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
5232 	ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
5233 	ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
5234 	ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
5235 	ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
5236 
5237 	/* large_rx disabled for now, additional features needed */
5238 	ctrl_buf->large_rx_ipv4 = 0;
5239 	ctrl_buf->large_rx_ipv6 = 0;
5240 
5241 	if (adapter->state != VNIC_PROBING) {
5242 		old_hw_features = adapter->netdev->hw_features;
5243 		adapter->netdev->hw_features = 0;
5244 	}
5245 
5246 	adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
5247 
5248 	if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
5249 		adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
5250 
5251 	if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
5252 		adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
5253 
5254 	if ((adapter->netdev->features &
5255 	    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
5256 		adapter->netdev->hw_features |= NETIF_F_RXCSUM;
5257 
5258 	if (buf->large_tx_ipv4)
5259 		adapter->netdev->hw_features |= NETIF_F_TSO;
5260 	if (buf->large_tx_ipv6)
5261 		adapter->netdev->hw_features |= NETIF_F_TSO6;
5262 
5263 	if (adapter->state == VNIC_PROBING) {
5264 		adapter->netdev->features |= adapter->netdev->hw_features;
5265 	} else if (old_hw_features != adapter->netdev->hw_features) {
5266 		netdev_features_t tmp = 0;
5267 
5268 		/* disable features no longer supported */
5269 		adapter->netdev->features &= adapter->netdev->hw_features;
5270 		/* turn on features now supported if previously enabled */
5271 		tmp = (old_hw_features ^ adapter->netdev->hw_features) &
5272 			adapter->netdev->hw_features;
5273 		adapter->netdev->features |=
5274 				tmp & adapter->netdev->wanted_features;
5275 	}
5276 
5277 	memset(&crq, 0, sizeof(crq));
5278 	crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
5279 	crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
5280 	crq.control_ip_offload.len =
5281 	    cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5282 	crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
5283 	ibmvnic_send_crq(adapter, &crq);
5284 }
5285 
handle_vpd_size_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5286 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
5287 				struct ibmvnic_adapter *adapter)
5288 {
5289 	struct device *dev = &adapter->vdev->dev;
5290 
5291 	if (crq->get_vpd_size_rsp.rc.code) {
5292 		dev_err(dev, "Error retrieving VPD size, rc=%x\n",
5293 			crq->get_vpd_size_rsp.rc.code);
5294 		complete(&adapter->fw_done);
5295 		return;
5296 	}
5297 
5298 	adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
5299 	complete(&adapter->fw_done);
5300 }
5301 
handle_vpd_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5302 static void handle_vpd_rsp(union ibmvnic_crq *crq,
5303 			   struct ibmvnic_adapter *adapter)
5304 {
5305 	struct device *dev = &adapter->vdev->dev;
5306 	unsigned char *substr = NULL;
5307 	u8 fw_level_len = 0;
5308 
5309 	memset(adapter->fw_version, 0, 32);
5310 
5311 	dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
5312 			 DMA_FROM_DEVICE);
5313 
5314 	if (crq->get_vpd_rsp.rc.code) {
5315 		dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
5316 			crq->get_vpd_rsp.rc.code);
5317 		goto complete;
5318 	}
5319 
5320 	/* get the position of the firmware version info
5321 	 * located after the ASCII 'RM' substring in the buffer
5322 	 */
5323 	substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
5324 	if (!substr) {
5325 		dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
5326 		goto complete;
5327 	}
5328 
5329 	/* get length of firmware level ASCII substring */
5330 	if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
5331 		fw_level_len = *(substr + 2);
5332 	} else {
5333 		dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
5334 		goto complete;
5335 	}
5336 
5337 	/* copy firmware version string from vpd into adapter */
5338 	if ((substr + 3 + fw_level_len) <
5339 	    (adapter->vpd->buff + adapter->vpd->len)) {
5340 		strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
5341 	} else {
5342 		dev_info(dev, "FW substr extrapolated VPD buff\n");
5343 	}
5344 
5345 complete:
5346 	if (adapter->fw_version[0] == '\0')
5347 		strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
5348 	complete(&adapter->fw_done);
5349 }
5350 
handle_query_ip_offload_rsp(struct ibmvnic_adapter * adapter)5351 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
5352 {
5353 	struct device *dev = &adapter->vdev->dev;
5354 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5355 	int i;
5356 
5357 	dma_unmap_single(dev, adapter->ip_offload_tok,
5358 			 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
5359 
5360 	netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
5361 	for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
5362 		netdev_dbg(adapter->netdev, "%016lx\n",
5363 			   ((unsigned long *)(buf))[i]);
5364 
5365 	netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
5366 	netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
5367 	netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
5368 		   buf->tcp_ipv4_chksum);
5369 	netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
5370 		   buf->tcp_ipv6_chksum);
5371 	netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
5372 		   buf->udp_ipv4_chksum);
5373 	netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
5374 		   buf->udp_ipv6_chksum);
5375 	netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
5376 		   buf->large_tx_ipv4);
5377 	netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
5378 		   buf->large_tx_ipv6);
5379 	netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
5380 		   buf->large_rx_ipv4);
5381 	netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
5382 		   buf->large_rx_ipv6);
5383 	netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
5384 		   buf->max_ipv4_header_size);
5385 	netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
5386 		   buf->max_ipv6_header_size);
5387 	netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
5388 		   buf->max_tcp_header_size);
5389 	netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
5390 		   buf->max_udp_header_size);
5391 	netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
5392 		   buf->max_large_tx_size);
5393 	netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
5394 		   buf->max_large_rx_size);
5395 	netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
5396 		   buf->ipv6_extension_header);
5397 	netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
5398 		   buf->tcp_pseudosum_req);
5399 	netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
5400 		   buf->num_ipv6_ext_headers);
5401 	netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
5402 		   buf->off_ipv6_ext_headers);
5403 
5404 	send_control_ip_offload(adapter);
5405 }
5406 
ibmvnic_fw_err_cause(u16 cause)5407 static const char *ibmvnic_fw_err_cause(u16 cause)
5408 {
5409 	switch (cause) {
5410 	case ADAPTER_PROBLEM:
5411 		return "adapter problem";
5412 	case BUS_PROBLEM:
5413 		return "bus problem";
5414 	case FW_PROBLEM:
5415 		return "firmware problem";
5416 	case DD_PROBLEM:
5417 		return "device driver problem";
5418 	case EEH_RECOVERY:
5419 		return "EEH recovery";
5420 	case FW_UPDATED:
5421 		return "firmware updated";
5422 	case LOW_MEMORY:
5423 		return "low Memory";
5424 	default:
5425 		return "unknown";
5426 	}
5427 }
5428 
handle_error_indication(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5429 static void handle_error_indication(union ibmvnic_crq *crq,
5430 				    struct ibmvnic_adapter *adapter)
5431 {
5432 	struct device *dev = &adapter->vdev->dev;
5433 	u16 cause;
5434 
5435 	cause = be16_to_cpu(crq->error_indication.error_cause);
5436 
5437 	dev_warn_ratelimited(dev,
5438 			     "Firmware reports %serror, cause: %s. Starting recovery...\n",
5439 			     crq->error_indication.flags
5440 				& IBMVNIC_FATAL_ERROR ? "FATAL " : "",
5441 			     ibmvnic_fw_err_cause(cause));
5442 
5443 	if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
5444 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5445 	else
5446 		ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
5447 }
5448 
handle_change_mac_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5449 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
5450 				 struct ibmvnic_adapter *adapter)
5451 {
5452 	struct net_device *netdev = adapter->netdev;
5453 	struct device *dev = &adapter->vdev->dev;
5454 	long rc;
5455 
5456 	rc = crq->change_mac_addr_rsp.rc.code;
5457 	if (rc) {
5458 		dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
5459 		goto out;
5460 	}
5461 	/* crq->change_mac_addr.mac_addr is the requested one
5462 	 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
5463 	 */
5464 	eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
5465 	ether_addr_copy(adapter->mac_addr,
5466 			&crq->change_mac_addr_rsp.mac_addr[0]);
5467 out:
5468 	complete(&adapter->fw_done);
5469 	return rc;
5470 }
5471 
handle_request_cap_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5472 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
5473 				   struct ibmvnic_adapter *adapter)
5474 {
5475 	struct device *dev = &adapter->vdev->dev;
5476 	u64 *req_value;
5477 	char *name;
5478 
5479 	atomic_dec(&adapter->running_cap_crqs);
5480 	netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
5481 		   atomic_read(&adapter->running_cap_crqs));
5482 	switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
5483 	case REQ_TX_QUEUES:
5484 		req_value = &adapter->req_tx_queues;
5485 		name = "tx";
5486 		break;
5487 	case REQ_RX_QUEUES:
5488 		req_value = &adapter->req_rx_queues;
5489 		name = "rx";
5490 		break;
5491 	case REQ_RX_ADD_QUEUES:
5492 		req_value = &adapter->req_rx_add_queues;
5493 		name = "rx_add";
5494 		break;
5495 	case REQ_TX_ENTRIES_PER_SUBCRQ:
5496 		req_value = &adapter->req_tx_entries_per_subcrq;
5497 		name = "tx_entries_per_subcrq";
5498 		break;
5499 	case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
5500 		req_value = &adapter->req_rx_add_entries_per_subcrq;
5501 		name = "rx_add_entries_per_subcrq";
5502 		break;
5503 	case REQ_MTU:
5504 		req_value = &adapter->req_mtu;
5505 		name = "mtu";
5506 		break;
5507 	case PROMISC_REQUESTED:
5508 		req_value = &adapter->promisc;
5509 		name = "promisc";
5510 		break;
5511 	default:
5512 		dev_err(dev, "Got invalid cap request rsp %d\n",
5513 			crq->request_capability.capability);
5514 		return;
5515 	}
5516 
5517 	switch (crq->request_capability_rsp.rc.code) {
5518 	case SUCCESS:
5519 		break;
5520 	case PARTIALSUCCESS:
5521 		dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
5522 			 *req_value,
5523 			 (long)be64_to_cpu(crq->request_capability_rsp.number),
5524 			 name);
5525 
5526 		if (be16_to_cpu(crq->request_capability_rsp.capability) ==
5527 		    REQ_MTU) {
5528 			pr_err("mtu of %llu is not supported. Reverting.\n",
5529 			       *req_value);
5530 			*req_value = adapter->fallback.mtu;
5531 		} else {
5532 			*req_value =
5533 				be64_to_cpu(crq->request_capability_rsp.number);
5534 		}
5535 
5536 		send_request_cap(adapter, 1);
5537 		return;
5538 	default:
5539 		dev_err(dev, "Error %d in request cap rsp\n",
5540 			crq->request_capability_rsp.rc.code);
5541 		return;
5542 	}
5543 
5544 	/* Done receiving requested capabilities, query IP offload support */
5545 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5546 		send_query_ip_offload(adapter);
5547 }
5548 
handle_login_rsp(union ibmvnic_crq * login_rsp_crq,struct ibmvnic_adapter * adapter)5549 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
5550 			    struct ibmvnic_adapter *adapter)
5551 {
5552 	struct device *dev = &adapter->vdev->dev;
5553 	struct net_device *netdev = adapter->netdev;
5554 	struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
5555 	struct ibmvnic_login_buffer *login = adapter->login_buf;
5556 	u64 *tx_handle_array;
5557 	u64 *rx_handle_array;
5558 	int num_tx_pools;
5559 	int num_rx_pools;
5560 	u64 *size_array;
5561 	u32 rsp_len;
5562 	int i;
5563 
5564 	/* CHECK: Test/set of login_pending does not need to be atomic
5565 	 * because only ibmvnic_tasklet tests/clears this.
5566 	 */
5567 	if (!adapter->login_pending) {
5568 		netdev_warn(netdev, "Ignoring unexpected login response\n");
5569 		return 0;
5570 	}
5571 	adapter->login_pending = false;
5572 
5573 	/* If the number of queues requested can't be allocated by the
5574 	 * server, the login response will return with code 1. We will need
5575 	 * to resend the login buffer with fewer queues requested.
5576 	 */
5577 	if (login_rsp_crq->generic.rc.code) {
5578 		adapter->init_done_rc = login_rsp_crq->generic.rc.code;
5579 		complete(&adapter->init_done);
5580 		return 0;
5581 	}
5582 
5583 	if (adapter->failover_pending) {
5584 		adapter->init_done_rc = -EAGAIN;
5585 		netdev_dbg(netdev, "Failover pending, ignoring login response\n");
5586 		complete(&adapter->init_done);
5587 		/* login response buffer will be released on reset */
5588 		return 0;
5589 	}
5590 
5591 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
5592 
5593 	netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
5594 	for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
5595 		netdev_dbg(adapter->netdev, "%016lx\n",
5596 			   ((unsigned long *)(adapter->login_rsp_buf))[i]);
5597 	}
5598 
5599 	/* Sanity checks */
5600 	if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
5601 	    (be32_to_cpu(login->num_rxcomp_subcrqs) *
5602 	     adapter->req_rx_add_queues !=
5603 	     be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
5604 		dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
5605 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5606 		return -EIO;
5607 	}
5608 
5609 	rsp_len = be32_to_cpu(login_rsp->len);
5610 	if (be32_to_cpu(login->login_rsp_len) < rsp_len ||
5611 	    rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) ||
5612 	    rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) ||
5613 	    rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) ||
5614 	    rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) {
5615 		/* This can happen if a login request times out and there are
5616 		 * 2 outstanding login requests sent, the LOGIN_RSP crq
5617 		 * could have been for the older login request. So we are
5618 		 * parsing the newer response buffer which may be incomplete
5619 		 */
5620 		dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n");
5621 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5622 		return -EIO;
5623 	}
5624 
5625 	size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5626 		be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
5627 	/* variable buffer sizes are not supported, so just read the
5628 	 * first entry.
5629 	 */
5630 	adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
5631 
5632 	num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
5633 	num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
5634 
5635 	tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5636 				  be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
5637 	rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5638 				  be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
5639 
5640 	for (i = 0; i < num_tx_pools; i++)
5641 		adapter->tx_scrq[i]->handle = tx_handle_array[i];
5642 
5643 	for (i = 0; i < num_rx_pools; i++)
5644 		adapter->rx_scrq[i]->handle = rx_handle_array[i];
5645 
5646 	adapter->num_active_tx_scrqs = num_tx_pools;
5647 	adapter->num_active_rx_scrqs = num_rx_pools;
5648 	release_login_rsp_buffer(adapter);
5649 	release_login_buffer(adapter);
5650 	complete(&adapter->init_done);
5651 
5652 	return 0;
5653 }
5654 
handle_request_unmap_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5655 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
5656 				     struct ibmvnic_adapter *adapter)
5657 {
5658 	struct device *dev = &adapter->vdev->dev;
5659 	long rc;
5660 
5661 	rc = crq->request_unmap_rsp.rc.code;
5662 	if (rc)
5663 		dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
5664 }
5665 
handle_query_map_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5666 static void handle_query_map_rsp(union ibmvnic_crq *crq,
5667 				 struct ibmvnic_adapter *adapter)
5668 {
5669 	struct net_device *netdev = adapter->netdev;
5670 	struct device *dev = &adapter->vdev->dev;
5671 	long rc;
5672 
5673 	rc = crq->query_map_rsp.rc.code;
5674 	if (rc) {
5675 		dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5676 		return;
5677 	}
5678 	netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5679 		   crq->query_map_rsp.page_size,
5680 		   __be32_to_cpu(crq->query_map_rsp.tot_pages),
5681 		   __be32_to_cpu(crq->query_map_rsp.free_pages));
5682 }
5683 
handle_query_cap_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5684 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5685 				 struct ibmvnic_adapter *adapter)
5686 {
5687 	struct net_device *netdev = adapter->netdev;
5688 	struct device *dev = &adapter->vdev->dev;
5689 	long rc;
5690 
5691 	atomic_dec(&adapter->running_cap_crqs);
5692 	netdev_dbg(netdev, "Outstanding queries: %d\n",
5693 		   atomic_read(&adapter->running_cap_crqs));
5694 	rc = crq->query_capability.rc.code;
5695 	if (rc) {
5696 		dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5697 		goto out;
5698 	}
5699 
5700 	switch (be16_to_cpu(crq->query_capability.capability)) {
5701 	case MIN_TX_QUEUES:
5702 		adapter->min_tx_queues =
5703 		    be64_to_cpu(crq->query_capability.number);
5704 		netdev_dbg(netdev, "min_tx_queues = %lld\n",
5705 			   adapter->min_tx_queues);
5706 		break;
5707 	case MIN_RX_QUEUES:
5708 		adapter->min_rx_queues =
5709 		    be64_to_cpu(crq->query_capability.number);
5710 		netdev_dbg(netdev, "min_rx_queues = %lld\n",
5711 			   adapter->min_rx_queues);
5712 		break;
5713 	case MIN_RX_ADD_QUEUES:
5714 		adapter->min_rx_add_queues =
5715 		    be64_to_cpu(crq->query_capability.number);
5716 		netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5717 			   adapter->min_rx_add_queues);
5718 		break;
5719 	case MAX_TX_QUEUES:
5720 		adapter->max_tx_queues =
5721 		    be64_to_cpu(crq->query_capability.number);
5722 		netdev_dbg(netdev, "max_tx_queues = %lld\n",
5723 			   adapter->max_tx_queues);
5724 		break;
5725 	case MAX_RX_QUEUES:
5726 		adapter->max_rx_queues =
5727 		    be64_to_cpu(crq->query_capability.number);
5728 		netdev_dbg(netdev, "max_rx_queues = %lld\n",
5729 			   adapter->max_rx_queues);
5730 		break;
5731 	case MAX_RX_ADD_QUEUES:
5732 		adapter->max_rx_add_queues =
5733 		    be64_to_cpu(crq->query_capability.number);
5734 		netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5735 			   adapter->max_rx_add_queues);
5736 		break;
5737 	case MIN_TX_ENTRIES_PER_SUBCRQ:
5738 		adapter->min_tx_entries_per_subcrq =
5739 		    be64_to_cpu(crq->query_capability.number);
5740 		netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5741 			   adapter->min_tx_entries_per_subcrq);
5742 		break;
5743 	case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5744 		adapter->min_rx_add_entries_per_subcrq =
5745 		    be64_to_cpu(crq->query_capability.number);
5746 		netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5747 			   adapter->min_rx_add_entries_per_subcrq);
5748 		break;
5749 	case MAX_TX_ENTRIES_PER_SUBCRQ:
5750 		adapter->max_tx_entries_per_subcrq =
5751 		    be64_to_cpu(crq->query_capability.number);
5752 		netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5753 			   adapter->max_tx_entries_per_subcrq);
5754 		break;
5755 	case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5756 		adapter->max_rx_add_entries_per_subcrq =
5757 		    be64_to_cpu(crq->query_capability.number);
5758 		netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5759 			   adapter->max_rx_add_entries_per_subcrq);
5760 		break;
5761 	case TCP_IP_OFFLOAD:
5762 		adapter->tcp_ip_offload =
5763 		    be64_to_cpu(crq->query_capability.number);
5764 		netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5765 			   adapter->tcp_ip_offload);
5766 		break;
5767 	case PROMISC_SUPPORTED:
5768 		adapter->promisc_supported =
5769 		    be64_to_cpu(crq->query_capability.number);
5770 		netdev_dbg(netdev, "promisc_supported = %lld\n",
5771 			   adapter->promisc_supported);
5772 		break;
5773 	case MIN_MTU:
5774 		adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5775 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5776 		netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5777 		break;
5778 	case MAX_MTU:
5779 		adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5780 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5781 		netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5782 		break;
5783 	case MAX_MULTICAST_FILTERS:
5784 		adapter->max_multicast_filters =
5785 		    be64_to_cpu(crq->query_capability.number);
5786 		netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5787 			   adapter->max_multicast_filters);
5788 		break;
5789 	case VLAN_HEADER_INSERTION:
5790 		adapter->vlan_header_insertion =
5791 		    be64_to_cpu(crq->query_capability.number);
5792 		if (adapter->vlan_header_insertion)
5793 			netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5794 		netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5795 			   adapter->vlan_header_insertion);
5796 		break;
5797 	case RX_VLAN_HEADER_INSERTION:
5798 		adapter->rx_vlan_header_insertion =
5799 		    be64_to_cpu(crq->query_capability.number);
5800 		netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5801 			   adapter->rx_vlan_header_insertion);
5802 		break;
5803 	case MAX_TX_SG_ENTRIES:
5804 		adapter->max_tx_sg_entries =
5805 		    be64_to_cpu(crq->query_capability.number);
5806 		netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5807 			   adapter->max_tx_sg_entries);
5808 		break;
5809 	case RX_SG_SUPPORTED:
5810 		adapter->rx_sg_supported =
5811 		    be64_to_cpu(crq->query_capability.number);
5812 		netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5813 			   adapter->rx_sg_supported);
5814 		break;
5815 	case OPT_TX_COMP_SUB_QUEUES:
5816 		adapter->opt_tx_comp_sub_queues =
5817 		    be64_to_cpu(crq->query_capability.number);
5818 		netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5819 			   adapter->opt_tx_comp_sub_queues);
5820 		break;
5821 	case OPT_RX_COMP_QUEUES:
5822 		adapter->opt_rx_comp_queues =
5823 		    be64_to_cpu(crq->query_capability.number);
5824 		netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5825 			   adapter->opt_rx_comp_queues);
5826 		break;
5827 	case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5828 		adapter->opt_rx_bufadd_q_per_rx_comp_q =
5829 		    be64_to_cpu(crq->query_capability.number);
5830 		netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5831 			   adapter->opt_rx_bufadd_q_per_rx_comp_q);
5832 		break;
5833 	case OPT_TX_ENTRIES_PER_SUBCRQ:
5834 		adapter->opt_tx_entries_per_subcrq =
5835 		    be64_to_cpu(crq->query_capability.number);
5836 		netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5837 			   adapter->opt_tx_entries_per_subcrq);
5838 		break;
5839 	case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5840 		adapter->opt_rxba_entries_per_subcrq =
5841 		    be64_to_cpu(crq->query_capability.number);
5842 		netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5843 			   adapter->opt_rxba_entries_per_subcrq);
5844 		break;
5845 	case TX_RX_DESC_REQ:
5846 		adapter->tx_rx_desc_req = crq->query_capability.number;
5847 		netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5848 			   adapter->tx_rx_desc_req);
5849 		break;
5850 
5851 	default:
5852 		netdev_err(netdev, "Got invalid cap rsp %d\n",
5853 			   crq->query_capability.capability);
5854 	}
5855 
5856 out:
5857 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5858 		send_request_cap(adapter, 0);
5859 }
5860 
send_query_phys_parms(struct ibmvnic_adapter * adapter)5861 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5862 {
5863 	union ibmvnic_crq crq;
5864 	int rc;
5865 
5866 	memset(&crq, 0, sizeof(crq));
5867 	crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5868 	crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5869 
5870 	mutex_lock(&adapter->fw_lock);
5871 	adapter->fw_done_rc = 0;
5872 	reinit_completion(&adapter->fw_done);
5873 
5874 	rc = ibmvnic_send_crq(adapter, &crq);
5875 	if (rc) {
5876 		mutex_unlock(&adapter->fw_lock);
5877 		return rc;
5878 	}
5879 
5880 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5881 	if (rc) {
5882 		mutex_unlock(&adapter->fw_lock);
5883 		return rc;
5884 	}
5885 
5886 	mutex_unlock(&adapter->fw_lock);
5887 	return adapter->fw_done_rc ? -EIO : 0;
5888 }
5889 
handle_query_phys_parms_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5890 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5891 				       struct ibmvnic_adapter *adapter)
5892 {
5893 	struct net_device *netdev = adapter->netdev;
5894 	int rc;
5895 	__be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5896 
5897 	rc = crq->query_phys_parms_rsp.rc.code;
5898 	if (rc) {
5899 		netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5900 		return rc;
5901 	}
5902 	switch (rspeed) {
5903 	case IBMVNIC_10MBPS:
5904 		adapter->speed = SPEED_10;
5905 		break;
5906 	case IBMVNIC_100MBPS:
5907 		adapter->speed = SPEED_100;
5908 		break;
5909 	case IBMVNIC_1GBPS:
5910 		adapter->speed = SPEED_1000;
5911 		break;
5912 	case IBMVNIC_10GBPS:
5913 		adapter->speed = SPEED_10000;
5914 		break;
5915 	case IBMVNIC_25GBPS:
5916 		adapter->speed = SPEED_25000;
5917 		break;
5918 	case IBMVNIC_40GBPS:
5919 		adapter->speed = SPEED_40000;
5920 		break;
5921 	case IBMVNIC_50GBPS:
5922 		adapter->speed = SPEED_50000;
5923 		break;
5924 	case IBMVNIC_100GBPS:
5925 		adapter->speed = SPEED_100000;
5926 		break;
5927 	case IBMVNIC_200GBPS:
5928 		adapter->speed = SPEED_200000;
5929 		break;
5930 	default:
5931 		if (netif_carrier_ok(netdev))
5932 			netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5933 		adapter->speed = SPEED_UNKNOWN;
5934 	}
5935 	if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5936 		adapter->duplex = DUPLEX_FULL;
5937 	else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5938 		adapter->duplex = DUPLEX_HALF;
5939 	else
5940 		adapter->duplex = DUPLEX_UNKNOWN;
5941 
5942 	return rc;
5943 }
5944 
ibmvnic_handle_crq(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5945 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5946 			       struct ibmvnic_adapter *adapter)
5947 {
5948 	struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5949 	struct net_device *netdev = adapter->netdev;
5950 	struct device *dev = &adapter->vdev->dev;
5951 	u64 *u64_crq = (u64 *)crq;
5952 	long rc;
5953 
5954 	netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5955 		   (unsigned long)cpu_to_be64(u64_crq[0]),
5956 		   (unsigned long)cpu_to_be64(u64_crq[1]));
5957 	switch (gen_crq->first) {
5958 	case IBMVNIC_CRQ_INIT_RSP:
5959 		switch (gen_crq->cmd) {
5960 		case IBMVNIC_CRQ_INIT:
5961 			dev_info(dev, "Partner initialized\n");
5962 			adapter->from_passive_init = true;
5963 			/* Discard any stale login responses from prev reset.
5964 			 * CHECK: should we clear even on INIT_COMPLETE?
5965 			 */
5966 			adapter->login_pending = false;
5967 
5968 			if (adapter->state == VNIC_DOWN)
5969 				rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5970 			else
5971 				rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5972 
5973 			if (rc && rc != -EBUSY) {
5974 				/* We were unable to schedule the failover
5975 				 * reset either because the adapter was still
5976 				 * probing (eg: during kexec) or we could not
5977 				 * allocate memory. Clear the failover_pending
5978 				 * flag since no one else will. We ignore
5979 				 * EBUSY because it means either FAILOVER reset
5980 				 * is already scheduled or the adapter is
5981 				 * being removed.
5982 				 */
5983 				netdev_err(netdev,
5984 					   "Error %ld scheduling failover reset\n",
5985 					   rc);
5986 				adapter->failover_pending = false;
5987 			}
5988 
5989 			if (!completion_done(&adapter->init_done)) {
5990 				if (!adapter->init_done_rc)
5991 					adapter->init_done_rc = -EAGAIN;
5992 				complete(&adapter->init_done);
5993 			}
5994 
5995 			break;
5996 		case IBMVNIC_CRQ_INIT_COMPLETE:
5997 			dev_info(dev, "Partner initialization complete\n");
5998 			adapter->crq.active = true;
5999 			send_version_xchg(adapter);
6000 			break;
6001 		default:
6002 			dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
6003 		}
6004 		return;
6005 	case IBMVNIC_CRQ_XPORT_EVENT:
6006 		netif_carrier_off(netdev);
6007 		adapter->crq.active = false;
6008 		/* terminate any thread waiting for a response
6009 		 * from the device
6010 		 */
6011 		if (!completion_done(&adapter->fw_done)) {
6012 			adapter->fw_done_rc = -EIO;
6013 			complete(&adapter->fw_done);
6014 		}
6015 
6016 		/* if we got here during crq-init, retry crq-init */
6017 		if (!completion_done(&adapter->init_done)) {
6018 			adapter->init_done_rc = -EAGAIN;
6019 			complete(&adapter->init_done);
6020 		}
6021 
6022 		if (!completion_done(&adapter->stats_done))
6023 			complete(&adapter->stats_done);
6024 		if (test_bit(0, &adapter->resetting))
6025 			adapter->force_reset_recovery = true;
6026 		if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
6027 			dev_info(dev, "Migrated, re-enabling adapter\n");
6028 			ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
6029 		} else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
6030 			dev_info(dev, "Backing device failover detected\n");
6031 			adapter->failover_pending = true;
6032 		} else {
6033 			/* The adapter lost the connection */
6034 			dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
6035 				gen_crq->cmd);
6036 			ibmvnic_reset(adapter, VNIC_RESET_FATAL);
6037 		}
6038 		return;
6039 	case IBMVNIC_CRQ_CMD_RSP:
6040 		break;
6041 	default:
6042 		dev_err(dev, "Got an invalid msg type 0x%02x\n",
6043 			gen_crq->first);
6044 		return;
6045 	}
6046 
6047 	switch (gen_crq->cmd) {
6048 	case VERSION_EXCHANGE_RSP:
6049 		rc = crq->version_exchange_rsp.rc.code;
6050 		if (rc) {
6051 			dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
6052 			break;
6053 		}
6054 		ibmvnic_version =
6055 			    be16_to_cpu(crq->version_exchange_rsp.version);
6056 		dev_info(dev, "Partner protocol version is %d\n",
6057 			 ibmvnic_version);
6058 		send_query_cap(adapter);
6059 		break;
6060 	case QUERY_CAPABILITY_RSP:
6061 		handle_query_cap_rsp(crq, adapter);
6062 		break;
6063 	case QUERY_MAP_RSP:
6064 		handle_query_map_rsp(crq, adapter);
6065 		break;
6066 	case REQUEST_MAP_RSP:
6067 		adapter->fw_done_rc = crq->request_map_rsp.rc.code;
6068 		complete(&adapter->fw_done);
6069 		break;
6070 	case REQUEST_UNMAP_RSP:
6071 		handle_request_unmap_rsp(crq, adapter);
6072 		break;
6073 	case REQUEST_CAPABILITY_RSP:
6074 		handle_request_cap_rsp(crq, adapter);
6075 		break;
6076 	case LOGIN_RSP:
6077 		netdev_dbg(netdev, "Got Login Response\n");
6078 		handle_login_rsp(crq, adapter);
6079 		break;
6080 	case LOGICAL_LINK_STATE_RSP:
6081 		netdev_dbg(netdev,
6082 			   "Got Logical Link State Response, state: %d rc: %d\n",
6083 			   crq->logical_link_state_rsp.link_state,
6084 			   crq->logical_link_state_rsp.rc.code);
6085 		adapter->logical_link_state =
6086 		    crq->logical_link_state_rsp.link_state;
6087 		adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
6088 		complete(&adapter->init_done);
6089 		break;
6090 	case LINK_STATE_INDICATION:
6091 		netdev_dbg(netdev, "Got Logical Link State Indication\n");
6092 		adapter->phys_link_state =
6093 		    crq->link_state_indication.phys_link_state;
6094 		adapter->logical_link_state =
6095 		    crq->link_state_indication.logical_link_state;
6096 		if (adapter->phys_link_state && adapter->logical_link_state)
6097 			netif_carrier_on(netdev);
6098 		else
6099 			netif_carrier_off(netdev);
6100 		break;
6101 	case CHANGE_MAC_ADDR_RSP:
6102 		netdev_dbg(netdev, "Got MAC address change Response\n");
6103 		adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
6104 		break;
6105 	case ERROR_INDICATION:
6106 		netdev_dbg(netdev, "Got Error Indication\n");
6107 		handle_error_indication(crq, adapter);
6108 		break;
6109 	case REQUEST_STATISTICS_RSP:
6110 		netdev_dbg(netdev, "Got Statistics Response\n");
6111 		complete(&adapter->stats_done);
6112 		break;
6113 	case QUERY_IP_OFFLOAD_RSP:
6114 		netdev_dbg(netdev, "Got Query IP offload Response\n");
6115 		handle_query_ip_offload_rsp(adapter);
6116 		break;
6117 	case MULTICAST_CTRL_RSP:
6118 		netdev_dbg(netdev, "Got multicast control Response\n");
6119 		break;
6120 	case CONTROL_IP_OFFLOAD_RSP:
6121 		netdev_dbg(netdev, "Got Control IP offload Response\n");
6122 		dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
6123 				 sizeof(adapter->ip_offload_ctrl),
6124 				 DMA_TO_DEVICE);
6125 		complete(&adapter->init_done);
6126 		break;
6127 	case COLLECT_FW_TRACE_RSP:
6128 		netdev_dbg(netdev, "Got Collect firmware trace Response\n");
6129 		complete(&adapter->fw_done);
6130 		break;
6131 	case GET_VPD_SIZE_RSP:
6132 		handle_vpd_size_rsp(crq, adapter);
6133 		break;
6134 	case GET_VPD_RSP:
6135 		handle_vpd_rsp(crq, adapter);
6136 		break;
6137 	case QUERY_PHYS_PARMS_RSP:
6138 		adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
6139 		complete(&adapter->fw_done);
6140 		break;
6141 	default:
6142 		netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
6143 			   gen_crq->cmd);
6144 	}
6145 }
6146 
ibmvnic_interrupt(int irq,void * instance)6147 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
6148 {
6149 	struct ibmvnic_adapter *adapter = instance;
6150 
6151 	tasklet_schedule(&adapter->tasklet);
6152 	return IRQ_HANDLED;
6153 }
6154 
ibmvnic_tasklet(struct tasklet_struct * t)6155 static void ibmvnic_tasklet(struct tasklet_struct *t)
6156 {
6157 	struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
6158 	struct ibmvnic_crq_queue *queue = &adapter->crq;
6159 	union ibmvnic_crq *crq;
6160 	unsigned long flags;
6161 
6162 	spin_lock_irqsave(&queue->lock, flags);
6163 
6164 	/* Pull all the valid messages off the CRQ */
6165 	while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
6166 		/* This barrier makes sure ibmvnic_next_crq()'s
6167 		 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
6168 		 * before ibmvnic_handle_crq()'s
6169 		 * switch(gen_crq->first) and switch(gen_crq->cmd).
6170 		 */
6171 		dma_rmb();
6172 		ibmvnic_handle_crq(crq, adapter);
6173 		crq->generic.first = 0;
6174 	}
6175 
6176 	spin_unlock_irqrestore(&queue->lock, flags);
6177 }
6178 
ibmvnic_reenable_crq_queue(struct ibmvnic_adapter * adapter)6179 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
6180 {
6181 	struct vio_dev *vdev = adapter->vdev;
6182 	int rc;
6183 
6184 	do {
6185 		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
6186 	} while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
6187 
6188 	if (rc)
6189 		dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
6190 
6191 	return rc;
6192 }
6193 
ibmvnic_reset_crq(struct ibmvnic_adapter * adapter)6194 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
6195 {
6196 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6197 	struct device *dev = &adapter->vdev->dev;
6198 	struct vio_dev *vdev = adapter->vdev;
6199 	int rc;
6200 
6201 	/* Close the CRQ */
6202 	do {
6203 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6204 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6205 
6206 	/* Clean out the queue */
6207 	if (!crq->msgs)
6208 		return -EINVAL;
6209 
6210 	memset(crq->msgs, 0, PAGE_SIZE);
6211 	crq->cur = 0;
6212 	crq->active = false;
6213 
6214 	/* And re-open it again */
6215 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6216 				crq->msg_token, PAGE_SIZE);
6217 
6218 	if (rc == H_CLOSED)
6219 		/* Adapter is good, but other end is not ready */
6220 		dev_warn(dev, "Partner adapter not ready\n");
6221 	else if (rc != 0)
6222 		dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
6223 
6224 	return rc;
6225 }
6226 
release_crq_queue(struct ibmvnic_adapter * adapter)6227 static void release_crq_queue(struct ibmvnic_adapter *adapter)
6228 {
6229 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6230 	struct vio_dev *vdev = adapter->vdev;
6231 	long rc;
6232 
6233 	if (!crq->msgs)
6234 		return;
6235 
6236 	netdev_dbg(adapter->netdev, "Releasing CRQ\n");
6237 	free_irq(vdev->irq, adapter);
6238 	tasklet_kill(&adapter->tasklet);
6239 	do {
6240 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6241 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6242 
6243 	dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
6244 			 DMA_BIDIRECTIONAL);
6245 	free_page((unsigned long)crq->msgs);
6246 	crq->msgs = NULL;
6247 	crq->active = false;
6248 }
6249 
init_crq_queue(struct ibmvnic_adapter * adapter)6250 static int init_crq_queue(struct ibmvnic_adapter *adapter)
6251 {
6252 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6253 	struct device *dev = &adapter->vdev->dev;
6254 	struct vio_dev *vdev = adapter->vdev;
6255 	int rc, retrc = -ENOMEM;
6256 
6257 	if (crq->msgs)
6258 		return 0;
6259 
6260 	crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
6261 	/* Should we allocate more than one page? */
6262 
6263 	if (!crq->msgs)
6264 		return -ENOMEM;
6265 
6266 	crq->size = PAGE_SIZE / sizeof(*crq->msgs);
6267 	crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
6268 					DMA_BIDIRECTIONAL);
6269 	if (dma_mapping_error(dev, crq->msg_token))
6270 		goto map_failed;
6271 
6272 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6273 				crq->msg_token, PAGE_SIZE);
6274 
6275 	if (rc == H_RESOURCE)
6276 		/* maybe kexecing and resource is busy. try a reset */
6277 		rc = ibmvnic_reset_crq(adapter);
6278 	retrc = rc;
6279 
6280 	if (rc == H_CLOSED) {
6281 		dev_warn(dev, "Partner adapter not ready\n");
6282 	} else if (rc) {
6283 		dev_warn(dev, "Error %d opening adapter\n", rc);
6284 		goto reg_crq_failed;
6285 	}
6286 
6287 	retrc = 0;
6288 
6289 	tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
6290 
6291 	netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
6292 	snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
6293 		 adapter->vdev->unit_address);
6294 	rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
6295 	if (rc) {
6296 		dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
6297 			vdev->irq, rc);
6298 		goto req_irq_failed;
6299 	}
6300 
6301 	rc = vio_enable_interrupts(vdev);
6302 	if (rc) {
6303 		dev_err(dev, "Error %d enabling interrupts\n", rc);
6304 		goto req_irq_failed;
6305 	}
6306 
6307 	crq->cur = 0;
6308 	spin_lock_init(&crq->lock);
6309 
6310 	/* process any CRQs that were queued before we enabled interrupts */
6311 	tasklet_schedule(&adapter->tasklet);
6312 
6313 	return retrc;
6314 
6315 req_irq_failed:
6316 	tasklet_kill(&adapter->tasklet);
6317 	do {
6318 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6319 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6320 reg_crq_failed:
6321 	dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
6322 map_failed:
6323 	free_page((unsigned long)crq->msgs);
6324 	crq->msgs = NULL;
6325 	return retrc;
6326 }
6327 
ibmvnic_reset_init(struct ibmvnic_adapter * adapter,bool reset)6328 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
6329 {
6330 	struct device *dev = &adapter->vdev->dev;
6331 	unsigned long timeout = msecs_to_jiffies(20000);
6332 	u64 old_num_rx_queues = adapter->req_rx_queues;
6333 	u64 old_num_tx_queues = adapter->req_tx_queues;
6334 	int rc;
6335 
6336 	adapter->from_passive_init = false;
6337 
6338 	rc = ibmvnic_send_crq_init(adapter);
6339 	if (rc) {
6340 		dev_err(dev, "Send crq init failed with error %d\n", rc);
6341 		return rc;
6342 	}
6343 
6344 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
6345 		dev_err(dev, "Initialization sequence timed out\n");
6346 		return -ETIMEDOUT;
6347 	}
6348 
6349 	if (adapter->init_done_rc) {
6350 		release_crq_queue(adapter);
6351 		dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
6352 		return adapter->init_done_rc;
6353 	}
6354 
6355 	if (adapter->from_passive_init) {
6356 		adapter->state = VNIC_OPEN;
6357 		adapter->from_passive_init = false;
6358 		dev_err(dev, "CRQ-init failed, passive-init\n");
6359 		return -EINVAL;
6360 	}
6361 
6362 	if (reset &&
6363 	    test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
6364 	    adapter->reset_reason != VNIC_RESET_MOBILITY) {
6365 		if (adapter->req_rx_queues != old_num_rx_queues ||
6366 		    adapter->req_tx_queues != old_num_tx_queues) {
6367 			release_sub_crqs(adapter, 0);
6368 			rc = init_sub_crqs(adapter);
6369 		} else {
6370 			/* no need to reinitialize completely, but we do
6371 			 * need to clean up transmits that were in flight
6372 			 * when we processed the reset.  Failure to do so
6373 			 * will confound the upper layer, usually TCP, by
6374 			 * creating the illusion of transmits that are
6375 			 * awaiting completion.
6376 			 */
6377 			clean_tx_pools(adapter);
6378 
6379 			rc = reset_sub_crq_queues(adapter);
6380 		}
6381 	} else {
6382 		rc = init_sub_crqs(adapter);
6383 	}
6384 
6385 	if (rc) {
6386 		dev_err(dev, "Initialization of sub crqs failed\n");
6387 		release_crq_queue(adapter);
6388 		return rc;
6389 	}
6390 
6391 	rc = init_sub_crq_irqs(adapter);
6392 	if (rc) {
6393 		dev_err(dev, "Failed to initialize sub crq irqs\n");
6394 		release_crq_queue(adapter);
6395 	}
6396 
6397 	return rc;
6398 }
6399 
6400 static struct device_attribute dev_attr_failover;
6401 
ibmvnic_probe(struct vio_dev * dev,const struct vio_device_id * id)6402 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
6403 {
6404 	struct ibmvnic_adapter *adapter;
6405 	struct net_device *netdev;
6406 	unsigned char *mac_addr_p;
6407 	unsigned long flags;
6408 	bool init_success;
6409 	int rc;
6410 
6411 	dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
6412 		dev->unit_address);
6413 
6414 	mac_addr_p = (unsigned char *)vio_get_attribute(dev,
6415 							VETH_MAC_ADDR, NULL);
6416 	if (!mac_addr_p) {
6417 		dev_err(&dev->dev,
6418 			"(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
6419 			__FILE__, __LINE__);
6420 		return 0;
6421 	}
6422 
6423 	netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
6424 				   IBMVNIC_MAX_QUEUES);
6425 	if (!netdev)
6426 		return -ENOMEM;
6427 
6428 	adapter = netdev_priv(netdev);
6429 	adapter->state = VNIC_PROBING;
6430 	dev_set_drvdata(&dev->dev, netdev);
6431 	adapter->vdev = dev;
6432 	adapter->netdev = netdev;
6433 	adapter->login_pending = false;
6434 	memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
6435 	/* map_ids start at 1, so ensure map_id 0 is always "in-use" */
6436 	bitmap_set(adapter->map_ids, 0, 1);
6437 
6438 	ether_addr_copy(adapter->mac_addr, mac_addr_p);
6439 	eth_hw_addr_set(netdev, adapter->mac_addr);
6440 	netdev->irq = dev->irq;
6441 	netdev->netdev_ops = &ibmvnic_netdev_ops;
6442 	netdev->ethtool_ops = &ibmvnic_ethtool_ops;
6443 	SET_NETDEV_DEV(netdev, &dev->dev);
6444 
6445 	INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
6446 	INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
6447 			  __ibmvnic_delayed_reset);
6448 	INIT_LIST_HEAD(&adapter->rwi_list);
6449 	spin_lock_init(&adapter->rwi_lock);
6450 	spin_lock_init(&adapter->state_lock);
6451 	mutex_init(&adapter->fw_lock);
6452 	init_completion(&adapter->probe_done);
6453 	init_completion(&adapter->init_done);
6454 	init_completion(&adapter->fw_done);
6455 	init_completion(&adapter->reset_done);
6456 	init_completion(&adapter->stats_done);
6457 	clear_bit(0, &adapter->resetting);
6458 	adapter->prev_rx_buf_sz = 0;
6459 	adapter->prev_mtu = 0;
6460 
6461 	init_success = false;
6462 	do {
6463 		reinit_init_done(adapter);
6464 
6465 		/* clear any failovers we got in the previous pass
6466 		 * since we are reinitializing the CRQ
6467 		 */
6468 		adapter->failover_pending = false;
6469 
6470 		/* If we had already initialized CRQ, we may have one or
6471 		 * more resets queued already. Discard those and release
6472 		 * the CRQ before initializing the CRQ again.
6473 		 */
6474 		release_crq_queue(adapter);
6475 
6476 		/* Since we are still in PROBING state, __ibmvnic_reset()
6477 		 * will not access the ->rwi_list and since we released CRQ,
6478 		 * we won't get _new_ transport events. But there maybe an
6479 		 * ongoing ibmvnic_reset() call. So serialize access to
6480 		 * rwi_list. If we win the race, ibvmnic_reset() could add
6481 		 * a reset after we purged but thats ok - we just may end
6482 		 * up with an extra reset (i.e similar to having two or more
6483 		 * resets in the queue at once).
6484 		 * CHECK.
6485 		 */
6486 		spin_lock_irqsave(&adapter->rwi_lock, flags);
6487 		flush_reset_queue(adapter);
6488 		spin_unlock_irqrestore(&adapter->rwi_lock, flags);
6489 
6490 		rc = init_crq_queue(adapter);
6491 		if (rc) {
6492 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
6493 				rc);
6494 			goto ibmvnic_init_fail;
6495 		}
6496 
6497 		rc = ibmvnic_reset_init(adapter, false);
6498 	} while (rc == -EAGAIN);
6499 
6500 	/* We are ignoring the error from ibmvnic_reset_init() assuming that the
6501 	 * partner is not ready. CRQ is not active. When the partner becomes
6502 	 * ready, we will do the passive init reset.
6503 	 */
6504 
6505 	if (!rc)
6506 		init_success = true;
6507 
6508 	rc = init_stats_buffers(adapter);
6509 	if (rc)
6510 		goto ibmvnic_init_fail;
6511 
6512 	rc = init_stats_token(adapter);
6513 	if (rc)
6514 		goto ibmvnic_stats_fail;
6515 
6516 	rc = device_create_file(&dev->dev, &dev_attr_failover);
6517 	if (rc)
6518 		goto ibmvnic_dev_file_err;
6519 
6520 	netif_carrier_off(netdev);
6521 
6522 	if (init_success) {
6523 		adapter->state = VNIC_PROBED;
6524 		netdev->mtu = adapter->req_mtu - ETH_HLEN;
6525 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
6526 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
6527 	} else {
6528 		adapter->state = VNIC_DOWN;
6529 	}
6530 
6531 	adapter->wait_for_reset = false;
6532 	adapter->last_reset_time = jiffies;
6533 
6534 	rc = register_netdev(netdev);
6535 	if (rc) {
6536 		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
6537 		goto ibmvnic_register_fail;
6538 	}
6539 	dev_info(&dev->dev, "ibmvnic registered\n");
6540 
6541 	rc = ibmvnic_cpu_notif_add(adapter);
6542 	if (rc) {
6543 		netdev_err(netdev, "Registering cpu notifier failed\n");
6544 		goto cpu_notif_add_failed;
6545 	}
6546 
6547 	complete(&adapter->probe_done);
6548 
6549 	return 0;
6550 
6551 cpu_notif_add_failed:
6552 	unregister_netdev(netdev);
6553 
6554 ibmvnic_register_fail:
6555 	device_remove_file(&dev->dev, &dev_attr_failover);
6556 
6557 ibmvnic_dev_file_err:
6558 	release_stats_token(adapter);
6559 
6560 ibmvnic_stats_fail:
6561 	release_stats_buffers(adapter);
6562 
6563 ibmvnic_init_fail:
6564 	release_sub_crqs(adapter, 1);
6565 	release_crq_queue(adapter);
6566 
6567 	/* cleanup worker thread after releasing CRQ so we don't get
6568 	 * transport events (i.e new work items for the worker thread).
6569 	 */
6570 	adapter->state = VNIC_REMOVING;
6571 	complete(&adapter->probe_done);
6572 	flush_work(&adapter->ibmvnic_reset);
6573 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6574 
6575 	flush_reset_queue(adapter);
6576 
6577 	mutex_destroy(&adapter->fw_lock);
6578 	free_netdev(netdev);
6579 
6580 	return rc;
6581 }
6582 
ibmvnic_remove(struct vio_dev * dev)6583 static void ibmvnic_remove(struct vio_dev *dev)
6584 {
6585 	struct net_device *netdev = dev_get_drvdata(&dev->dev);
6586 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6587 	unsigned long flags;
6588 
6589 	spin_lock_irqsave(&adapter->state_lock, flags);
6590 
6591 	/* If ibmvnic_reset() is scheduling a reset, wait for it to
6592 	 * finish. Then, set the state to REMOVING to prevent it from
6593 	 * scheduling any more work and to have reset functions ignore
6594 	 * any resets that have already been scheduled. Drop the lock
6595 	 * after setting state, so __ibmvnic_reset() which is called
6596 	 * from the flush_work() below, can make progress.
6597 	 */
6598 	spin_lock(&adapter->rwi_lock);
6599 	adapter->state = VNIC_REMOVING;
6600 	spin_unlock(&adapter->rwi_lock);
6601 
6602 	spin_unlock_irqrestore(&adapter->state_lock, flags);
6603 
6604 	ibmvnic_cpu_notif_remove(adapter);
6605 
6606 	flush_work(&adapter->ibmvnic_reset);
6607 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6608 
6609 	rtnl_lock();
6610 	unregister_netdevice(netdev);
6611 
6612 	release_resources(adapter);
6613 	release_rx_pools(adapter);
6614 	release_tx_pools(adapter);
6615 	release_sub_crqs(adapter, 1);
6616 	release_crq_queue(adapter);
6617 
6618 	release_stats_token(adapter);
6619 	release_stats_buffers(adapter);
6620 
6621 	adapter->state = VNIC_REMOVED;
6622 
6623 	rtnl_unlock();
6624 	mutex_destroy(&adapter->fw_lock);
6625 	device_remove_file(&dev->dev, &dev_attr_failover);
6626 	free_netdev(netdev);
6627 	dev_set_drvdata(&dev->dev, NULL);
6628 }
6629 
failover_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)6630 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
6631 			      const char *buf, size_t count)
6632 {
6633 	struct net_device *netdev = dev_get_drvdata(dev);
6634 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6635 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
6636 	__be64 session_token;
6637 	long rc;
6638 
6639 	if (!sysfs_streq(buf, "1"))
6640 		return -EINVAL;
6641 
6642 	rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
6643 			 H_GET_SESSION_TOKEN, 0, 0, 0);
6644 	if (rc) {
6645 		netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
6646 			   rc);
6647 		goto last_resort;
6648 	}
6649 
6650 	session_token = (__be64)retbuf[0];
6651 	netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
6652 		   be64_to_cpu(session_token));
6653 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
6654 				H_SESSION_ERR_DETECTED, session_token, 0, 0);
6655 	if (rc) {
6656 		netdev_err(netdev,
6657 			   "H_VIOCTL initiated failover failed, rc %ld\n",
6658 			   rc);
6659 		goto last_resort;
6660 	}
6661 
6662 	return count;
6663 
6664 last_resort:
6665 	netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
6666 	ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
6667 
6668 	return count;
6669 }
6670 static DEVICE_ATTR_WO(failover);
6671 
ibmvnic_get_desired_dma(struct vio_dev * vdev)6672 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
6673 {
6674 	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
6675 	struct ibmvnic_adapter *adapter;
6676 	struct iommu_table *tbl;
6677 	unsigned long ret = 0;
6678 	int i;
6679 
6680 	tbl = get_iommu_table_base(&vdev->dev);
6681 
6682 	/* netdev inits at probe time along with the structures we need below*/
6683 	if (!netdev)
6684 		return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
6685 
6686 	adapter = netdev_priv(netdev);
6687 
6688 	ret += PAGE_SIZE; /* the crq message queue */
6689 	ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
6690 
6691 	for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
6692 		ret += 4 * PAGE_SIZE; /* the scrq message queue */
6693 
6694 	for (i = 0; i < adapter->num_active_rx_pools; i++)
6695 		ret += adapter->rx_pool[i].size *
6696 		    IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
6697 
6698 	return ret;
6699 }
6700 
ibmvnic_resume(struct device * dev)6701 static int ibmvnic_resume(struct device *dev)
6702 {
6703 	struct net_device *netdev = dev_get_drvdata(dev);
6704 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6705 
6706 	if (adapter->state != VNIC_OPEN)
6707 		return 0;
6708 
6709 	tasklet_schedule(&adapter->tasklet);
6710 
6711 	return 0;
6712 }
6713 
6714 static const struct vio_device_id ibmvnic_device_table[] = {
6715 	{"network", "IBM,vnic"},
6716 	{"", "" }
6717 };
6718 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
6719 
6720 static const struct dev_pm_ops ibmvnic_pm_ops = {
6721 	.resume = ibmvnic_resume
6722 };
6723 
6724 static struct vio_driver ibmvnic_driver = {
6725 	.id_table       = ibmvnic_device_table,
6726 	.probe          = ibmvnic_probe,
6727 	.remove         = ibmvnic_remove,
6728 	.get_desired_dma = ibmvnic_get_desired_dma,
6729 	.name		= ibmvnic_driver_name,
6730 	.pm		= &ibmvnic_pm_ops,
6731 };
6732 
6733 /* module functions */
ibmvnic_module_init(void)6734 static int __init ibmvnic_module_init(void)
6735 {
6736 	int ret;
6737 
6738 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "net/ibmvnic:online",
6739 				      ibmvnic_cpu_online,
6740 				      ibmvnic_cpu_down_prep);
6741 	if (ret < 0)
6742 		goto out;
6743 	ibmvnic_online = ret;
6744 	ret = cpuhp_setup_state_multi(CPUHP_IBMVNIC_DEAD, "net/ibmvnic:dead",
6745 				      NULL, ibmvnic_cpu_dead);
6746 	if (ret)
6747 		goto err_dead;
6748 
6749 	ret = vio_register_driver(&ibmvnic_driver);
6750 	if (ret)
6751 		goto err_vio_register;
6752 
6753 	pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
6754 		IBMVNIC_DRIVER_VERSION);
6755 
6756 	return 0;
6757 err_vio_register:
6758 	cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6759 err_dead:
6760 	cpuhp_remove_multi_state(ibmvnic_online);
6761 out:
6762 	return ret;
6763 }
6764 
ibmvnic_module_exit(void)6765 static void __exit ibmvnic_module_exit(void)
6766 {
6767 	vio_unregister_driver(&ibmvnic_driver);
6768 	cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6769 	cpuhp_remove_multi_state(ibmvnic_online);
6770 }
6771 
6772 module_init(ibmvnic_module_init);
6773 module_exit(ibmvnic_module_exit);
6774