xref: /linux/drivers/gpu/drm/xe/xe_guc_relay.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2023 Intel Corporation
4  */
5 
6 #include <linux/bitfield.h>
7 #include <linux/delay.h>
8 
9 #include <drm/drm_managed.h>
10 
11 #include <kunit/static_stub.h>
12 #include <kunit/test-bug.h>
13 
14 #include "abi/guc_actions_sriov_abi.h"
15 #include "abi/guc_relay_actions_abi.h"
16 #include "abi/guc_relay_communication_abi.h"
17 
18 #include "xe_assert.h"
19 #include "xe_device.h"
20 #include "xe_gt.h"
21 #include "xe_gt_sriov_printk.h"
22 #include "xe_gt_sriov_pf_service.h"
23 #include "xe_guc.h"
24 #include "xe_guc_ct.h"
25 #include "xe_guc_hxg_helpers.h"
26 #include "xe_guc_relay.h"
27 #include "xe_guc_relay_types.h"
28 #include "xe_sriov.h"
29 
30 /*
31  * How long should we wait for the response?
32  * XXX this value is subject for the profiling.
33  */
34 #define RELAY_TIMEOUT_MSEC	(2500)
35 
36 static void relays_worker_fn(struct work_struct *w);
37 
38 static struct xe_guc *relay_to_guc(struct xe_guc_relay *relay)
39 {
40 	return container_of(relay, struct xe_guc, relay);
41 }
42 
43 static struct xe_guc_ct *relay_to_ct(struct xe_guc_relay *relay)
44 {
45 	return &relay_to_guc(relay)->ct;
46 }
47 
48 static struct xe_gt *relay_to_gt(struct xe_guc_relay *relay)
49 {
50 	return guc_to_gt(relay_to_guc(relay));
51 }
52 
53 static struct xe_device *relay_to_xe(struct xe_guc_relay *relay)
54 {
55 	return gt_to_xe(relay_to_gt(relay));
56 }
57 
58 #define relay_assert(relay, condition)	xe_gt_assert(relay_to_gt(relay), condition)
59 #define relay_notice(relay, msg...)	xe_gt_sriov_notice(relay_to_gt(relay), "relay: " msg)
60 #define relay_debug(relay, msg...)	xe_gt_sriov_dbg_verbose(relay_to_gt(relay), "relay: " msg)
61 
62 static int relay_get_totalvfs(struct xe_guc_relay *relay)
63 {
64 	struct xe_device *xe = relay_to_xe(relay);
65 	struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
66 
67 	KUNIT_STATIC_STUB_REDIRECT(relay_get_totalvfs, relay);
68 	return IS_SRIOV_VF(xe) ? 0 : pci_sriov_get_totalvfs(pdev);
69 }
70 
71 static bool relay_is_ready(struct xe_guc_relay *relay)
72 {
73 	return mempool_initialized(&relay->pool);
74 }
75 
76 static u32 relay_get_next_rid(struct xe_guc_relay *relay)
77 {
78 	u32 rid;
79 
80 	spin_lock(&relay->lock);
81 	rid = ++relay->last_rid;
82 	spin_unlock(&relay->lock);
83 
84 	return rid;
85 }
86 
87 /**
88  * struct relay_transaction - internal data used to handle transactions
89  *
90  * Relation between struct relay_transaction members::
91  *
92  *                 <-------------------- GUC_CTB_MAX_DWORDS -------------->
93  *                                  <-------- GUC_RELAY_MSG_MAX_LEN --->
94  *                 <--- offset ---> <--- request_len ------->
95  *                +----------------+-------------------------+----------+--+
96  *                |                |                         |          |  |
97  *                +----------------+-------------------------+----------+--+
98  *                ^                ^
99  *               /                /
100  *    request_buf          request
101  *
102  *                 <-------------------- GUC_CTB_MAX_DWORDS -------------->
103  *                                  <-------- GUC_RELAY_MSG_MAX_LEN --->
104  *                 <--- offset ---> <--- response_len --->
105  *                +----------------+----------------------+-------------+--+
106  *                |                |                      |             |  |
107  *                +----------------+----------------------+-------------+--+
108  *                ^                ^
109  *               /                /
110  *   response_buf         response
111  */
112 struct relay_transaction {
113 	/**
114 	 * @incoming: indicates whether this transaction represents an incoming
115 	 *            request from the remote VF/PF or this transaction
116 	 *            represents outgoing request to the remote VF/PF.
117 	 */
118 	bool incoming;
119 
120 	/**
121 	 * @remote: PF/VF identifier of the origin (or target) of the relay
122 	 *          request message.
123 	 */
124 	u32 remote;
125 
126 	/** @rid: identifier of the VF/PF relay message. */
127 	u32 rid;
128 
129 	/**
130 	 * @request: points to the inner VF/PF request message, copied to the
131 	 *           #response_buf starting at #offset.
132 	 */
133 	u32 *request;
134 
135 	/** @request_len: length of the inner VF/PF request message. */
136 	u32 request_len;
137 
138 	/**
139 	 * @response: points to the placeholder buffer where inner VF/PF
140 	 *            response will be located, for outgoing transaction
141 	 *            this could be caller's buffer (if provided) otherwise
142 	 *            it points to the #response_buf starting at #offset.
143 	 */
144 	u32 *response;
145 
146 	/**
147 	 * @response_len: length of the inner VF/PF response message (only
148 	 *                if #status is 0), initially set to the size of the
149 	 *                placeholder buffer where response message will be
150 	 *                copied.
151 	 */
152 	u32 response_len;
153 
154 	/**
155 	 * @offset: offset to the start of the inner VF/PF relay message inside
156 	 *          buffers; this offset is equal the length of the outer GuC
157 	 *          relay header message.
158 	 */
159 	u32 offset;
160 
161 	/**
162 	 * @request_buf: buffer with VF/PF request message including outer
163 	 *               transport message.
164 	 */
165 	u32 request_buf[GUC_CTB_MAX_DWORDS];
166 
167 	/**
168 	 * @response_buf: buffer with VF/PF response message including outer
169 	 *                transport message.
170 	 */
171 	u32 response_buf[GUC_CTB_MAX_DWORDS];
172 
173 	/**
174 	 * @reply: status of the reply, 0 means that data pointed by the
175 	 *         #response is valid.
176 	 */
177 	int reply;
178 
179 	/** @done: completion of the outgoing transaction. */
180 	struct completion done;
181 
182 	/** @link: transaction list link */
183 	struct list_head link;
184 };
185 
186 static u32 prepare_pf2guc(u32 *msg, u32 target, u32 rid)
187 {
188 	msg[0] = FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
189 		 FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
190 		 FIELD_PREP(GUC_HXG_REQUEST_MSG_0_ACTION, XE_GUC_ACTION_PF2GUC_RELAY_TO_VF);
191 	msg[1] = FIELD_PREP(PF2GUC_RELAY_TO_VF_REQUEST_MSG_1_VFID, target);
192 	msg[2] = FIELD_PREP(PF2GUC_RELAY_TO_VF_REQUEST_MSG_2_RELAY_ID, rid);
193 
194 	return PF2GUC_RELAY_TO_VF_REQUEST_MSG_MIN_LEN;
195 }
196 
197 static u32 prepare_vf2guc(u32 *msg, u32 rid)
198 {
199 	msg[0] = FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
200 		 FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
201 		 FIELD_PREP(GUC_HXG_REQUEST_MSG_0_ACTION, XE_GUC_ACTION_VF2GUC_RELAY_TO_PF);
202 	msg[1] = FIELD_PREP(VF2GUC_RELAY_TO_PF_REQUEST_MSG_1_RELAY_ID, rid);
203 
204 	return VF2GUC_RELAY_TO_PF_REQUEST_MSG_MIN_LEN;
205 }
206 
207 static struct relay_transaction *
208 __relay_get_transaction(struct xe_guc_relay *relay, bool incoming, u32 remote, u32 rid,
209 			const u32 *action, u32 action_len, u32 *resp, u32 resp_size)
210 {
211 	struct relay_transaction *txn;
212 
213 	relay_assert(relay, action_len >= GUC_RELAY_MSG_MIN_LEN);
214 	relay_assert(relay, action_len <= GUC_RELAY_MSG_MAX_LEN);
215 	relay_assert(relay, !(!!resp ^ !!resp_size));
216 	relay_assert(relay, resp_size <= GUC_RELAY_MSG_MAX_LEN);
217 	relay_assert(relay, resp_size == 0 || resp_size >= GUC_RELAY_MSG_MIN_LEN);
218 
219 	if (unlikely(!relay_is_ready(relay)))
220 		return ERR_PTR(-ENODEV);
221 
222 	/*
223 	 * For incoming requests we can't use GFP_KERNEL as those are delivered
224 	 * with CTB lock held which is marked as used in the reclaim path.
225 	 * Btw, that's one of the reason why we use mempool here!
226 	 */
227 	txn = mempool_alloc(&relay->pool, incoming ? GFP_ATOMIC : GFP_KERNEL);
228 	if (!txn)
229 		return ERR_PTR(-ENOMEM);
230 
231 	txn->incoming = incoming;
232 	txn->remote = remote;
233 	txn->rid = rid;
234 	txn->offset = remote ?
235 		prepare_pf2guc(incoming ? txn->response_buf : txn->request_buf, remote, rid) :
236 		prepare_vf2guc(incoming ? txn->response_buf : txn->request_buf, rid);
237 
238 	relay_assert(relay, txn->offset);
239 	relay_assert(relay, txn->offset + GUC_RELAY_MSG_MAX_LEN <= ARRAY_SIZE(txn->request_buf));
240 	relay_assert(relay, txn->offset + GUC_RELAY_MSG_MAX_LEN <= ARRAY_SIZE(txn->response_buf));
241 
242 	txn->request = txn->request_buf + txn->offset;
243 	memcpy(&txn->request_buf[txn->offset], action, sizeof(u32) * action_len);
244 	txn->request_len = action_len;
245 
246 	txn->response = resp ?: txn->response_buf + txn->offset;
247 	txn->response_len = resp_size ?: GUC_RELAY_MSG_MAX_LEN;
248 	txn->reply = -ENOMSG;
249 	INIT_LIST_HEAD(&txn->link);
250 	init_completion(&txn->done);
251 
252 	return txn;
253 }
254 
255 static struct relay_transaction *
256 relay_new_transaction(struct xe_guc_relay *relay, u32 target, const u32 *action, u32 len,
257 		      u32 *resp, u32 resp_size)
258 {
259 	u32 rid = relay_get_next_rid(relay);
260 
261 	return __relay_get_transaction(relay, false, target, rid, action, len, resp, resp_size);
262 }
263 
264 static struct relay_transaction *
265 relay_new_incoming_transaction(struct xe_guc_relay *relay, u32 origin, u32 rid,
266 			       const u32 *action, u32 len)
267 {
268 	return __relay_get_transaction(relay, true, origin, rid, action, len, NULL, 0);
269 }
270 
271 static void relay_release_transaction(struct xe_guc_relay *relay, struct relay_transaction *txn)
272 {
273 	relay_assert(relay, list_empty(&txn->link));
274 
275 	txn->offset = 0;
276 	txn->response = NULL;
277 	txn->reply = -ESTALE;
278 	mempool_free(txn, &relay->pool);
279 }
280 
281 static int relay_send_transaction(struct xe_guc_relay *relay, struct relay_transaction *txn)
282 {
283 	u32 len = txn->incoming ? txn->response_len : txn->request_len;
284 	u32 *buf = txn->incoming ? txn->response_buf : txn->request_buf;
285 	u32 *msg = buf + txn->offset;
286 	int ret;
287 
288 	relay_assert(relay, txn->offset);
289 	relay_assert(relay, txn->offset + len <= GUC_CTB_MAX_DWORDS);
290 	relay_assert(relay, len >= GUC_RELAY_MSG_MIN_LEN);
291 	relay_assert(relay, len <= GUC_RELAY_MSG_MAX_LEN);
292 
293 	relay_debug(relay, "sending %s.%u to %u = %*ph\n",
294 		    guc_hxg_type_to_string(FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0])),
295 		    txn->rid, txn->remote, (int)sizeof(u32) * len, msg);
296 
297 	ret = xe_guc_ct_send_block(relay_to_ct(relay), buf, len + txn->offset);
298 
299 	if (unlikely(ret > 0)) {
300 		relay_notice(relay, "Unexpected data=%d from GuC, wrong ABI?\n", ret);
301 		ret = -EPROTO;
302 	}
303 	if (unlikely(ret < 0)) {
304 		relay_notice(relay, "Failed to send %s.%x to GuC (%pe) %*ph ...\n",
305 			     guc_hxg_type_to_string(FIELD_GET(GUC_HXG_MSG_0_TYPE, buf[0])),
306 			     FIELD_GET(GUC_HXG_REQUEST_MSG_0_ACTION, buf[0]),
307 			     ERR_PTR(ret), (int)sizeof(u32) * txn->offset, buf);
308 		relay_notice(relay, "Failed to send %s.%u to %u (%pe) %*ph\n",
309 			     guc_hxg_type_to_string(FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0])),
310 			     txn->rid, txn->remote, ERR_PTR(ret), (int)sizeof(u32) * len, msg);
311 	}
312 
313 	return ret;
314 }
315 
316 static void __fini_relay(struct drm_device *drm, void *arg)
317 {
318 	struct xe_guc_relay *relay = arg;
319 
320 	mempool_exit(&relay->pool);
321 }
322 
323 /**
324  * xe_guc_relay_init - Initialize a &xe_guc_relay
325  * @relay: the &xe_guc_relay to initialize
326  *
327  * Initialize remaining members of &xe_guc_relay that may depend
328  * on the SR-IOV mode.
329  *
330  * Return: 0 on success or a negative error code on failure.
331  */
332 int xe_guc_relay_init(struct xe_guc_relay *relay)
333 {
334 	const int XE_RELAY_MEMPOOL_MIN_NUM = 1;
335 	struct xe_device *xe = relay_to_xe(relay);
336 	int err;
337 
338 	relay_assert(relay, !relay_is_ready(relay));
339 
340 	if (!IS_SRIOV(xe))
341 		return 0;
342 
343 	spin_lock_init(&relay->lock);
344 	INIT_WORK(&relay->worker, relays_worker_fn);
345 	INIT_LIST_HEAD(&relay->pending_relays);
346 	INIT_LIST_HEAD(&relay->incoming_actions);
347 
348 	err = mempool_init_kmalloc_pool(&relay->pool, XE_RELAY_MEMPOOL_MIN_NUM +
349 					relay_get_totalvfs(relay),
350 					sizeof(struct relay_transaction));
351 	if (err)
352 		return err;
353 
354 	relay_debug(relay, "using mempool with %d elements\n", relay->pool.min_nr);
355 
356 	return drmm_add_action_or_reset(&xe->drm, __fini_relay, relay);
357 }
358 
359 static u32 to_relay_error(int err)
360 {
361 	/* XXX: assume that relay errors match errno codes */
362 	return err < 0 ? -err : GUC_RELAY_ERROR_UNDISCLOSED;
363 }
364 
365 static int from_relay_error(u32 error)
366 {
367 	/* XXX: assume that relay errors match errno codes */
368 	return error ? -error : -ENODATA;
369 }
370 
371 static u32 sanitize_relay_error(u32 error)
372 {
373 	/* XXX TBD if generic error codes will be allowed */
374 	if (!IS_ENABLED(CONFIG_DRM_XE_DEBUG))
375 		error = GUC_RELAY_ERROR_UNDISCLOSED;
376 	return error;
377 }
378 
379 static u32 sanitize_relay_error_hint(u32 hint)
380 {
381 	/* XXX TBD if generic error codes will be allowed */
382 	if (!IS_ENABLED(CONFIG_DRM_XE_DEBUG))
383 		hint = 0;
384 	return hint;
385 }
386 
387 static u32 prepare_error_reply(u32 *msg, u32 error, u32 hint)
388 {
389 	msg[0] = FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
390 		 FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_RESPONSE_FAILURE) |
391 		 FIELD_PREP(GUC_HXG_FAILURE_MSG_0_HINT, hint) |
392 		 FIELD_PREP(GUC_HXG_FAILURE_MSG_0_ERROR, error);
393 
394 	XE_WARN_ON(!FIELD_FIT(GUC_HXG_FAILURE_MSG_0_ERROR, error));
395 	XE_WARN_ON(!FIELD_FIT(GUC_HXG_FAILURE_MSG_0_HINT, hint));
396 
397 	return GUC_HXG_FAILURE_MSG_LEN;
398 }
399 
400 static void relay_testonly_nop(struct xe_guc_relay *relay)
401 {
402 	KUNIT_STATIC_STUB_REDIRECT(relay_testonly_nop, relay);
403 }
404 
405 static int relay_send_message_and_wait(struct xe_guc_relay *relay,
406 				       struct relay_transaction *txn,
407 				       u32 *buf, u32 buf_size)
408 {
409 	unsigned long timeout = msecs_to_jiffies(RELAY_TIMEOUT_MSEC);
410 	u32 *msg = &txn->request_buf[txn->offset];
411 	u32 len = txn->request_len;
412 	u32 type, action, data0;
413 	int ret;
414 	long n;
415 
416 	type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]);
417 	action = FIELD_GET(GUC_HXG_REQUEST_MSG_0_ACTION, msg[0]);
418 	data0 = FIELD_GET(GUC_HXG_REQUEST_MSG_0_DATA0, msg[0]);
419 
420 	relay_debug(relay, "%s.%u to %u action %#x:%u\n",
421 		    guc_hxg_type_to_string(type),
422 		    txn->rid, txn->remote, action, data0);
423 
424 	/* list ordering does not need to match RID ordering */
425 	spin_lock(&relay->lock);
426 	list_add_tail(&txn->link, &relay->pending_relays);
427 	spin_unlock(&relay->lock);
428 
429 resend:
430 	ret = relay_send_transaction(relay, txn);
431 	if (unlikely(ret < 0))
432 		goto unlink;
433 
434 wait:
435 	n = wait_for_completion_timeout(&txn->done, timeout);
436 	if (unlikely(n == 0 && txn->reply)) {
437 		ret = -ETIME;
438 		goto unlink;
439 	}
440 
441 	relay_debug(relay, "%u.%u reply %d after %u msec\n",
442 		    txn->remote, txn->rid, txn->reply, jiffies_to_msecs(timeout - n));
443 	if (unlikely(txn->reply)) {
444 		reinit_completion(&txn->done);
445 		if (txn->reply == -EAGAIN)
446 			goto resend;
447 		if (txn->reply == -EBUSY) {
448 			relay_testonly_nop(relay);
449 			goto wait;
450 		}
451 		if (txn->reply > 0)
452 			ret = from_relay_error(txn->reply);
453 		else
454 			ret = txn->reply;
455 		goto unlink;
456 	}
457 
458 	relay_debug(relay, "%u.%u response %*ph\n", txn->remote, txn->rid,
459 		    (int)sizeof(u32) * txn->response_len, txn->response);
460 	relay_assert(relay, txn->response_len >= GUC_RELAY_MSG_MIN_LEN);
461 	ret = txn->response_len;
462 
463 unlink:
464 	spin_lock(&relay->lock);
465 	list_del_init(&txn->link);
466 	spin_unlock(&relay->lock);
467 
468 	if (unlikely(ret < 0)) {
469 		relay_notice(relay, "Unsuccessful %s.%u %#x:%u to %u (%pe) %*ph\n",
470 			     guc_hxg_type_to_string(type), txn->rid,
471 			     action, data0, txn->remote, ERR_PTR(ret),
472 			     (int)sizeof(u32) * len, msg);
473 	}
474 
475 	return ret;
476 }
477 
478 static int relay_send_to(struct xe_guc_relay *relay, u32 target,
479 			 const u32 *msg, u32 len, u32 *buf, u32 buf_size)
480 {
481 	struct relay_transaction *txn;
482 	int ret;
483 
484 	relay_assert(relay, len >= GUC_RELAY_MSG_MIN_LEN);
485 	relay_assert(relay, len <= GUC_RELAY_MSG_MAX_LEN);
486 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_ORIGIN, msg[0]) == GUC_HXG_ORIGIN_HOST);
487 	relay_assert(relay, guc_hxg_type_is_action(FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0])));
488 
489 	if (unlikely(!relay_is_ready(relay)))
490 		return -ENODEV;
491 
492 	txn = relay_new_transaction(relay, target, msg, len, buf, buf_size);
493 	if (IS_ERR(txn))
494 		return PTR_ERR(txn);
495 
496 	switch (FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0])) {
497 	case GUC_HXG_TYPE_REQUEST:
498 		ret = relay_send_message_and_wait(relay, txn, buf, buf_size);
499 		break;
500 	case GUC_HXG_TYPE_FAST_REQUEST:
501 		relay_assert(relay, !GUC_HXG_TYPE_FAST_REQUEST);
502 		fallthrough;
503 	case GUC_HXG_TYPE_EVENT:
504 		ret = relay_send_transaction(relay, txn);
505 		break;
506 	default:
507 		ret = -EINVAL;
508 		break;
509 	}
510 
511 	relay_release_transaction(relay, txn);
512 	return ret;
513 }
514 
515 #ifdef CONFIG_PCI_IOV
516 /**
517  * xe_guc_relay_send_to_vf - Send a message to the VF.
518  * @relay: the &xe_guc_relay which will send the message
519  * @target: target VF number
520  * @msg: request message to be sent
521  * @len: length of the request message (in dwords, can't be 0)
522  * @buf: placeholder for the response message
523  * @buf_size: size of the response message placeholder (in dwords)
524  *
525  * This function can only be used by the driver running in the SR-IOV PF mode.
526  *
527  * Return: Non-negative response length (in dwords) or
528  *         a negative error code on failure.
529  */
530 int xe_guc_relay_send_to_vf(struct xe_guc_relay *relay, u32 target,
531 			    const u32 *msg, u32 len, u32 *buf, u32 buf_size)
532 {
533 	relay_assert(relay, IS_SRIOV_PF(relay_to_xe(relay)));
534 
535 	return relay_send_to(relay, target, msg, len, buf, buf_size);
536 }
537 #endif
538 
539 /**
540  * xe_guc_relay_send_to_pf - Send a message to the PF.
541  * @relay: the &xe_guc_relay which will send the message
542  * @msg: request message to be sent
543  * @len: length of the message (in dwords, can't be 0)
544  * @buf: placeholder for the response message
545  * @buf_size: size of the response message placeholder (in dwords)
546  *
547  * This function can only be used by driver running in SR-IOV VF mode.
548  *
549  * Return: Non-negative response length (in dwords) or
550  *         a negative error code on failure.
551  */
552 int xe_guc_relay_send_to_pf(struct xe_guc_relay *relay,
553 			    const u32 *msg, u32 len, u32 *buf, u32 buf_size)
554 {
555 	relay_assert(relay, IS_SRIOV_VF(relay_to_xe(relay)));
556 
557 	return relay_send_to(relay, PFID, msg, len, buf, buf_size);
558 }
559 
560 static int relay_handle_reply(struct xe_guc_relay *relay, u32 origin,
561 			      u32 rid, int reply, const u32 *msg, u32 len)
562 {
563 	struct relay_transaction *pending;
564 	int err = -ESRCH;
565 
566 	spin_lock(&relay->lock);
567 	list_for_each_entry(pending, &relay->pending_relays, link) {
568 		if (pending->remote != origin || pending->rid != rid) {
569 			relay_debug(relay, "%u.%u still awaits response\n",
570 				    pending->remote, pending->rid);
571 			continue;
572 		}
573 		err = 0; /* found! */
574 		if (reply == 0) {
575 			if (len > pending->response_len) {
576 				reply = -ENOBUFS;
577 				err = -ENOBUFS;
578 			} else {
579 				memcpy(pending->response, msg, 4 * len);
580 				pending->response_len = len;
581 			}
582 		}
583 		pending->reply = reply;
584 		complete_all(&pending->done);
585 		break;
586 	}
587 	spin_unlock(&relay->lock);
588 
589 	return err;
590 }
591 
592 static int relay_handle_failure(struct xe_guc_relay *relay, u32 origin,
593 				u32 rid, const u32 *msg, u32 len)
594 {
595 	int error = FIELD_GET(GUC_HXG_FAILURE_MSG_0_ERROR, msg[0]);
596 	u32 hint __maybe_unused = FIELD_GET(GUC_HXG_FAILURE_MSG_0_HINT, msg[0]);
597 
598 	relay_assert(relay, len);
599 	relay_debug(relay, "%u.%u error %#x (%pe) hint %u debug %*ph\n",
600 		    origin, rid, error, ERR_PTR(-error), hint, 4 * (len - 1), msg + 1);
601 
602 	return relay_handle_reply(relay, origin, rid, error ?: -EREMOTEIO, NULL, 0);
603 }
604 
605 static int relay_testloop_action_handler(struct xe_guc_relay *relay, u32 origin,
606 					 const u32 *msg, u32 len, u32 *response, u32 size)
607 {
608 	static ktime_t last_reply = 0;
609 	u32 type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]);
610 	u32 action = FIELD_GET(GUC_HXG_REQUEST_MSG_0_ACTION, msg[0]);
611 	u32 opcode = FIELD_GET(GUC_HXG_REQUEST_MSG_0_DATA0, msg[0]);
612 	ktime_t now = ktime_get();
613 	bool busy;
614 	int ret;
615 
616 	relay_assert(relay, guc_hxg_type_is_action(type));
617 	relay_assert(relay, action == GUC_RELAY_ACTION_VFXPF_TESTLOOP);
618 
619 	if (!IS_ENABLED(CONFIG_DRM_XE_DEBUG_SRIOV))
620 		return -ECONNREFUSED;
621 
622 	if (!last_reply)
623 		last_reply = now;
624 	busy = ktime_before(now, ktime_add_ms(last_reply, 2 * RELAY_TIMEOUT_MSEC));
625 	if (!busy)
626 		last_reply = now;
627 
628 	switch (opcode) {
629 	case VFXPF_TESTLOOP_OPCODE_NOP:
630 		if (type == GUC_HXG_TYPE_EVENT)
631 			return 0;
632 		return guc_hxg_msg_encode_success(response, 0);
633 	case VFXPF_TESTLOOP_OPCODE_BUSY:
634 		if (type == GUC_HXG_TYPE_EVENT)
635 			return -EPROTO;
636 		msleep(RELAY_TIMEOUT_MSEC / 8);
637 		if (busy)
638 			return -EINPROGRESS;
639 		return guc_hxg_msg_encode_success(response, 0);
640 	case VFXPF_TESTLOOP_OPCODE_RETRY:
641 		if (type == GUC_HXG_TYPE_EVENT)
642 			return -EPROTO;
643 		msleep(RELAY_TIMEOUT_MSEC / 8);
644 		if (busy)
645 			return guc_hxg_msg_encode_retry(response, 0);
646 		return guc_hxg_msg_encode_success(response, 0);
647 	case VFXPF_TESTLOOP_OPCODE_ECHO:
648 		if (type == GUC_HXG_TYPE_EVENT)
649 			return -EPROTO;
650 		if (size < len)
651 			return -ENOBUFS;
652 		ret = guc_hxg_msg_encode_success(response, len);
653 		memcpy(response + ret, msg + ret, (len - ret) * sizeof(u32));
654 		return len;
655 	case VFXPF_TESTLOOP_OPCODE_FAIL:
656 		return -EHWPOISON;
657 	default:
658 		break;
659 	}
660 
661 	relay_notice(relay, "Unexpected action %#x opcode %#x\n", action, opcode);
662 	return -EBADRQC;
663 }
664 
665 static int relay_action_handler(struct xe_guc_relay *relay, u32 origin,
666 				const u32 *msg, u32 len, u32 *response, u32 size)
667 {
668 	struct xe_gt *gt = relay_to_gt(relay);
669 	u32 type;
670 	int ret;
671 
672 	relay_assert(relay, len >= GUC_HXG_MSG_MIN_LEN);
673 
674 	if (FIELD_GET(GUC_HXG_REQUEST_MSG_0_ACTION, msg[0]) == GUC_RELAY_ACTION_VFXPF_TESTLOOP)
675 		return relay_testloop_action_handler(relay, origin, msg, len, response, size);
676 
677 	type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]);
678 
679 	if (IS_SRIOV_PF(relay_to_xe(relay)))
680 		ret = xe_gt_sriov_pf_service_process_request(gt, origin, msg, len, response, size);
681 	else
682 		ret = -EOPNOTSUPP;
683 
684 	if (type == GUC_HXG_TYPE_EVENT)
685 		relay_assert(relay, ret <= 0);
686 
687 	return ret;
688 }
689 
690 static struct relay_transaction *relay_dequeue_transaction(struct xe_guc_relay *relay)
691 {
692 	struct relay_transaction *txn;
693 
694 	spin_lock(&relay->lock);
695 	txn = list_first_entry_or_null(&relay->incoming_actions, struct relay_transaction, link);
696 	if (txn)
697 		list_del_init(&txn->link);
698 	spin_unlock(&relay->lock);
699 
700 	return txn;
701 }
702 
703 static void relay_process_incoming_action(struct xe_guc_relay *relay)
704 {
705 	struct relay_transaction *txn;
706 	bool again = false;
707 	u32 type;
708 	int ret;
709 
710 	txn = relay_dequeue_transaction(relay);
711 	if (!txn)
712 		return;
713 
714 	type = FIELD_GET(GUC_HXG_MSG_0_TYPE, txn->request_buf[txn->offset]);
715 
716 	ret = relay_action_handler(relay, txn->remote,
717 				   txn->request_buf + txn->offset, txn->request_len,
718 				   txn->response_buf + txn->offset,
719 				   ARRAY_SIZE(txn->response_buf) - txn->offset);
720 
721 	if (ret == -EINPROGRESS) {
722 		again = true;
723 		ret = guc_hxg_msg_encode_busy(txn->response_buf + txn->offset, 0);
724 	}
725 
726 	if (ret > 0) {
727 		txn->response_len = ret;
728 		ret = relay_send_transaction(relay, txn);
729 	}
730 
731 	if (ret < 0) {
732 		u32 error = to_relay_error(ret);
733 
734 		relay_notice(relay, "Failed to handle %s.%u from %u (%pe) %*ph\n",
735 			     guc_hxg_type_to_string(type), txn->rid, txn->remote,
736 			     ERR_PTR(ret), 4 * txn->request_len, txn->request_buf + txn->offset);
737 
738 		txn->response_len = prepare_error_reply(txn->response_buf + txn->offset,
739 							txn->remote ?
740 							sanitize_relay_error(error) : error,
741 							txn->remote ?
742 							sanitize_relay_error_hint(-ret) : -ret);
743 		ret = relay_send_transaction(relay, txn);
744 		again = false;
745 	}
746 
747 	if (again) {
748 		spin_lock(&relay->lock);
749 		list_add(&txn->link, &relay->incoming_actions);
750 		spin_unlock(&relay->lock);
751 		return;
752 	}
753 
754 	if (unlikely(ret < 0))
755 		relay_notice(relay, "Failed to process action.%u (%pe) %*ph\n",
756 			     txn->rid, ERR_PTR(ret), 4 * txn->request_len,
757 			     txn->request_buf + txn->offset);
758 
759 	relay_release_transaction(relay, txn);
760 }
761 
762 static bool relay_needs_worker(struct xe_guc_relay *relay)
763 {
764 	bool is_empty;
765 
766 	spin_lock(&relay->lock);
767 	is_empty = list_empty(&relay->incoming_actions);
768 	spin_unlock(&relay->lock);
769 
770 	return !is_empty;
771 
772 }
773 
774 static void relay_kick_worker(struct xe_guc_relay *relay)
775 {
776 	KUNIT_STATIC_STUB_REDIRECT(relay_kick_worker, relay);
777 	queue_work(relay_to_xe(relay)->sriov.wq, &relay->worker);
778 }
779 
780 static void relays_worker_fn(struct work_struct *w)
781 {
782 	struct xe_guc_relay *relay = container_of(w, struct xe_guc_relay, worker);
783 
784 	relay_process_incoming_action(relay);
785 
786 	if (relay_needs_worker(relay))
787 		relay_kick_worker(relay);
788 }
789 
790 static int relay_queue_action_msg(struct xe_guc_relay *relay, u32 origin, u32 rid,
791 				  const u32 *msg, u32 len)
792 {
793 	struct relay_transaction *txn;
794 
795 	txn = relay_new_incoming_transaction(relay, origin, rid, msg, len);
796 	if (IS_ERR(txn))
797 		return PTR_ERR(txn);
798 
799 	spin_lock(&relay->lock);
800 	list_add_tail(&txn->link, &relay->incoming_actions);
801 	spin_unlock(&relay->lock);
802 
803 	relay_kick_worker(relay);
804 	return 0;
805 }
806 
807 static int relay_process_msg(struct xe_guc_relay *relay, u32 origin, u32 rid,
808 			     const u32 *msg, u32 len)
809 {
810 	u32 type;
811 	int err;
812 
813 	if (unlikely(len < GUC_HXG_MSG_MIN_LEN))
814 		return -EPROTO;
815 
816 	if (FIELD_GET(GUC_HXG_MSG_0_ORIGIN, msg[0]) != GUC_HXG_ORIGIN_HOST)
817 		return -EPROTO;
818 
819 	type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]);
820 	relay_debug(relay, "received %s.%u from %u = %*ph\n",
821 		    guc_hxg_type_to_string(type), rid, origin, 4 * len, msg);
822 
823 	switch (type) {
824 	case GUC_HXG_TYPE_REQUEST:
825 	case GUC_HXG_TYPE_FAST_REQUEST:
826 	case GUC_HXG_TYPE_EVENT:
827 		err = relay_queue_action_msg(relay, origin, rid, msg, len);
828 		break;
829 	case GUC_HXG_TYPE_RESPONSE_SUCCESS:
830 		err = relay_handle_reply(relay, origin, rid, 0, msg, len);
831 		break;
832 	case GUC_HXG_TYPE_NO_RESPONSE_BUSY:
833 		err = relay_handle_reply(relay, origin, rid, -EBUSY, NULL, 0);
834 		break;
835 	case GUC_HXG_TYPE_NO_RESPONSE_RETRY:
836 		err = relay_handle_reply(relay, origin, rid, -EAGAIN, NULL, 0);
837 		break;
838 	case GUC_HXG_TYPE_RESPONSE_FAILURE:
839 		err = relay_handle_failure(relay, origin, rid, msg, len);
840 		break;
841 	default:
842 		err = -EBADRQC;
843 	}
844 
845 	if (unlikely(err))
846 		relay_notice(relay, "Failed to process %s.%u from %u (%pe) %*ph\n",
847 			     guc_hxg_type_to_string(type), rid, origin,
848 			     ERR_PTR(err), 4 * len, msg);
849 
850 	return err;
851 }
852 
853 /**
854  * xe_guc_relay_process_guc2vf - Handle relay notification message from the GuC.
855  * @relay: the &xe_guc_relay which will handle the message
856  * @msg: message to be handled
857  * @len: length of the message (in dwords)
858  *
859  * This function will handle relay messages received from the GuC.
860  *
861  * This function is can only be used if driver is running in SR-IOV mode.
862  *
863  * Return: 0 on success or a negative error code on failure.
864  */
865 int xe_guc_relay_process_guc2vf(struct xe_guc_relay *relay, const u32 *msg, u32 len)
866 {
867 	u32 rid;
868 
869 	relay_assert(relay, len >= GUC_HXG_MSG_MIN_LEN);
870 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_ORIGIN, msg[0]) == GUC_HXG_ORIGIN_GUC);
871 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]) == GUC_HXG_TYPE_EVENT);
872 	relay_assert(relay, FIELD_GET(GUC_HXG_EVENT_MSG_0_ACTION, msg[0]) ==
873 		     XE_GUC_ACTION_GUC2VF_RELAY_FROM_PF);
874 
875 	if (unlikely(!IS_SRIOV_VF(relay_to_xe(relay)) && !kunit_get_current_test()))
876 		return -EPERM;
877 
878 	if (unlikely(!relay_is_ready(relay)))
879 		return -ENODEV;
880 
881 	if (unlikely(len < GUC2VF_RELAY_FROM_PF_EVENT_MSG_MIN_LEN))
882 		return -EPROTO;
883 
884 	if (unlikely(len > GUC2VF_RELAY_FROM_PF_EVENT_MSG_MAX_LEN))
885 		return -EMSGSIZE;
886 
887 	if (unlikely(FIELD_GET(GUC_HXG_EVENT_MSG_0_DATA0, msg[0])))
888 		return -EPFNOSUPPORT;
889 
890 	rid = FIELD_GET(GUC2VF_RELAY_FROM_PF_EVENT_MSG_1_RELAY_ID, msg[1]);
891 
892 	return relay_process_msg(relay, PFID, rid,
893 				 msg + GUC2VF_RELAY_FROM_PF_EVENT_MSG_MIN_LEN,
894 				 len - GUC2VF_RELAY_FROM_PF_EVENT_MSG_MIN_LEN);
895 }
896 
897 #ifdef CONFIG_PCI_IOV
898 /**
899  * xe_guc_relay_process_guc2pf - Handle relay notification message from the GuC.
900  * @relay: the &xe_guc_relay which will handle the message
901  * @msg: message to be handled
902  * @len: length of the message (in dwords)
903  *
904  * This function will handle relay messages received from the GuC.
905  *
906  * This function can only be used if driver is running in SR-IOV PF mode.
907  *
908  * Return: 0 on success or a negative error code on failure.
909  */
910 int xe_guc_relay_process_guc2pf(struct xe_guc_relay *relay, const u32 *msg, u32 len)
911 {
912 	u32 origin, rid;
913 	int err;
914 
915 	relay_assert(relay, len >= GUC_HXG_EVENT_MSG_MIN_LEN);
916 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_ORIGIN, msg[0]) == GUC_HXG_ORIGIN_GUC);
917 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]) == GUC_HXG_TYPE_EVENT);
918 	relay_assert(relay, FIELD_GET(GUC_HXG_EVENT_MSG_0_ACTION, msg[0]) ==
919 		     XE_GUC_ACTION_GUC2PF_RELAY_FROM_VF);
920 
921 	if (unlikely(!IS_SRIOV_PF(relay_to_xe(relay)) && !kunit_get_current_test()))
922 		return -EPERM;
923 
924 	if (unlikely(!relay_is_ready(relay)))
925 		return -ENODEV;
926 
927 	if (unlikely(len < GUC2PF_RELAY_FROM_VF_EVENT_MSG_MIN_LEN))
928 		return -EPROTO;
929 
930 	if (unlikely(len > GUC2PF_RELAY_FROM_VF_EVENT_MSG_MAX_LEN))
931 		return -EMSGSIZE;
932 
933 	if (unlikely(FIELD_GET(GUC_HXG_EVENT_MSG_0_DATA0, msg[0])))
934 		return -EPFNOSUPPORT;
935 
936 	origin = FIELD_GET(GUC2PF_RELAY_FROM_VF_EVENT_MSG_1_VFID, msg[1]);
937 	rid = FIELD_GET(GUC2PF_RELAY_FROM_VF_EVENT_MSG_2_RELAY_ID, msg[2]);
938 
939 	if (unlikely(origin > relay_get_totalvfs(relay)))
940 		return -ENOENT;
941 
942 	err = relay_process_msg(relay, origin, rid,
943 				msg + GUC2PF_RELAY_FROM_VF_EVENT_MSG_MIN_LEN,
944 				len - GUC2PF_RELAY_FROM_VF_EVENT_MSG_MIN_LEN);
945 
946 	return err;
947 }
948 #endif
949 
950 #if IS_BUILTIN(CONFIG_DRM_XE_KUNIT_TEST)
951 #include "tests/xe_guc_relay_test.c"
952 #endif
953