xref: /linux/drivers/gpu/drm/xe/xe_guc_relay.c (revision ae22a94997b8a03dcb3c922857c203246711f9d4)
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_guc.h"
23 #include "xe_guc_ct.h"
24 #include "xe_guc_hxg_helpers.h"
25 #include "xe_guc_relay.h"
26 #include "xe_guc_relay_types.h"
27 #include "xe_sriov.h"
28 
29 /*
30  * How long should we wait for the response?
31  * XXX this value is subject for the profiling.
32  */
33 #define RELAY_TIMEOUT_MSEC	(2500)
34 
35 static void relays_worker_fn(struct work_struct *w);
36 
37 static struct xe_guc *relay_to_guc(struct xe_guc_relay *relay)
38 {
39 	return container_of(relay, struct xe_guc, relay);
40 }
41 
42 static struct xe_guc_ct *relay_to_ct(struct xe_guc_relay *relay)
43 {
44 	return &relay_to_guc(relay)->ct;
45 }
46 
47 static struct xe_gt *relay_to_gt(struct xe_guc_relay *relay)
48 {
49 	return guc_to_gt(relay_to_guc(relay));
50 }
51 
52 static struct xe_device *relay_to_xe(struct xe_guc_relay *relay)
53 {
54 	return gt_to_xe(relay_to_gt(relay));
55 }
56 
57 #define relay_assert(relay, condition)	xe_gt_assert(relay_to_gt(relay), condition)
58 #define relay_notice(relay, msg...)	xe_gt_sriov_notice(relay_to_gt(relay), "relay: " msg)
59 #define relay_debug(relay, msg...)	xe_gt_sriov_dbg_verbose(relay_to_gt(relay), "relay: " msg)
60 
61 static int relay_get_totalvfs(struct xe_guc_relay *relay)
62 {
63 	struct xe_device *xe = relay_to_xe(relay);
64 	struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
65 
66 	KUNIT_STATIC_STUB_REDIRECT(relay_get_totalvfs, relay);
67 	return IS_SRIOV_VF(xe) ? 0 : pci_sriov_get_totalvfs(pdev);
68 }
69 
70 static bool relay_is_ready(struct xe_guc_relay *relay)
71 {
72 	return mempool_initialized(&relay->pool);
73 }
74 
75 static u32 relay_get_next_rid(struct xe_guc_relay *relay)
76 {
77 	u32 rid;
78 
79 	spin_lock(&relay->lock);
80 	rid = ++relay->last_rid;
81 	spin_unlock(&relay->lock);
82 
83 	return rid;
84 }
85 
86 /**
87  * struct relay_transaction - internal data used to handle transactions
88  *
89  * Relation between struct relay_transaction members::
90  *
91  *                 <-------------------- GUC_CTB_MAX_DWORDS -------------->
92  *                                  <-------- GUC_RELAY_MSG_MAX_LEN --->
93  *                 <--- offset ---> <--- request_len ------->
94  *                +----------------+-------------------------+----------+--+
95  *                |                |                         |          |  |
96  *                +----------------+-------------------------+----------+--+
97  *                ^                ^
98  *               /                /
99  *    request_buf          request
100  *
101  *                 <-------------------- GUC_CTB_MAX_DWORDS -------------->
102  *                                  <-------- GUC_RELAY_MSG_MAX_LEN --->
103  *                 <--- offset ---> <--- response_len --->
104  *                +----------------+----------------------+-------------+--+
105  *                |                |                      |             |  |
106  *                +----------------+----------------------+-------------+--+
107  *                ^                ^
108  *               /                /
109  *   response_buf         response
110  */
111 struct relay_transaction {
112 	/**
113 	 * @incoming: indicates whether this transaction represents an incoming
114 	 *            request from the remote VF/PF or this transaction
115 	 *            represents outgoing request to the remote VF/PF.
116 	 */
117 	bool incoming;
118 
119 	/**
120 	 * @remote: PF/VF identifier of the origin (or target) of the relay
121 	 *          request message.
122 	 */
123 	u32 remote;
124 
125 	/** @rid: identifier of the VF/PF relay message. */
126 	u32 rid;
127 
128 	/**
129 	 * @request: points to the inner VF/PF request message, copied to the
130 	 *           #response_buf starting at #offset.
131 	 */
132 	u32 *request;
133 
134 	/** @request_len: length of the inner VF/PF request message. */
135 	u32 request_len;
136 
137 	/**
138 	 * @response: points to the placeholder buffer where inner VF/PF
139 	 *            response will be located, for outgoing transaction
140 	 *            this could be caller's buffer (if provided) otherwise
141 	 *            it points to the #response_buf starting at #offset.
142 	 */
143 	u32 *response;
144 
145 	/**
146 	 * @response_len: length of the inner VF/PF response message (only
147 	 *                if #status is 0), initially set to the size of the
148 	 *                placeholder buffer where response message will be
149 	 *                copied.
150 	 */
151 	u32 response_len;
152 
153 	/**
154 	 * @offset: offset to the start of the inner VF/PF relay message inside
155 	 *          buffers; this offset is equal the length of the outer GuC
156 	 *          relay header message.
157 	 */
158 	u32 offset;
159 
160 	/**
161 	 * @request_buf: buffer with VF/PF request message including outer
162 	 *               transport message.
163 	 */
164 	u32 request_buf[GUC_CTB_MAX_DWORDS];
165 
166 	/**
167 	 * @response_buf: buffer with VF/PF response message including outer
168 	 *                transport message.
169 	 */
170 	u32 response_buf[GUC_CTB_MAX_DWORDS];
171 
172 	/**
173 	 * @reply: status of the reply, 0 means that data pointed by the
174 	 *         #response is valid.
175 	 */
176 	int reply;
177 
178 	/** @done: completion of the outgoing transaction. */
179 	struct completion done;
180 
181 	/** @link: transaction list link */
182 	struct list_head link;
183 };
184 
185 static u32 prepare_pf2guc(u32 *msg, u32 target, u32 rid)
186 {
187 	msg[0] = FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
188 		 FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
189 		 FIELD_PREP(GUC_HXG_REQUEST_MSG_0_ACTION, XE_GUC_ACTION_PF2GUC_RELAY_TO_VF);
190 	msg[1] = FIELD_PREP(PF2GUC_RELAY_TO_VF_REQUEST_MSG_1_VFID, target);
191 	msg[2] = FIELD_PREP(PF2GUC_RELAY_TO_VF_REQUEST_MSG_2_RELAY_ID, rid);
192 
193 	return PF2GUC_RELAY_TO_VF_REQUEST_MSG_MIN_LEN;
194 }
195 
196 static u32 prepare_vf2guc(u32 *msg, u32 rid)
197 {
198 	msg[0] = FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
199 		 FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
200 		 FIELD_PREP(GUC_HXG_REQUEST_MSG_0_ACTION, XE_GUC_ACTION_VF2GUC_RELAY_TO_PF);
201 	msg[1] = FIELD_PREP(VF2GUC_RELAY_TO_PF_REQUEST_MSG_1_RELAY_ID, rid);
202 
203 	return VF2GUC_RELAY_TO_PF_REQUEST_MSG_MIN_LEN;
204 }
205 
206 static struct relay_transaction *
207 __relay_get_transaction(struct xe_guc_relay *relay, bool incoming, u32 remote, u32 rid,
208 			const u32 *action, u32 action_len, u32 *resp, u32 resp_size)
209 {
210 	struct relay_transaction *txn;
211 
212 	relay_assert(relay, action_len >= GUC_RELAY_MSG_MIN_LEN);
213 	relay_assert(relay, action_len <= GUC_RELAY_MSG_MAX_LEN);
214 	relay_assert(relay, !(!!resp ^ !!resp_size));
215 	relay_assert(relay, resp_size <= GUC_RELAY_MSG_MAX_LEN);
216 	relay_assert(relay, resp_size == 0 || resp_size >= GUC_RELAY_MSG_MIN_LEN);
217 
218 	if (unlikely(!relay_is_ready(relay)))
219 		return ERR_PTR(-ENODEV);
220 
221 	/*
222 	 * For incoming requests we can't use GFP_KERNEL as those are delivered
223 	 * with CTB lock held which is marked as used in the reclaim path.
224 	 * Btw, that's one of the reason why we use mempool here!
225 	 */
226 	txn = mempool_alloc(&relay->pool, incoming ? GFP_ATOMIC : GFP_KERNEL);
227 	if (!txn)
228 		return ERR_PTR(-ENOMEM);
229 
230 	txn->incoming = incoming;
231 	txn->remote = remote;
232 	txn->rid = rid;
233 	txn->offset = remote ?
234 		prepare_pf2guc(incoming ? txn->response_buf : txn->request_buf, remote, rid) :
235 		prepare_vf2guc(incoming ? txn->response_buf : txn->request_buf, rid);
236 
237 	relay_assert(relay, txn->offset);
238 	relay_assert(relay, txn->offset + GUC_RELAY_MSG_MAX_LEN <= ARRAY_SIZE(txn->request_buf));
239 	relay_assert(relay, txn->offset + GUC_RELAY_MSG_MAX_LEN <= ARRAY_SIZE(txn->response_buf));
240 
241 	txn->request = txn->request_buf + txn->offset;
242 	memcpy(&txn->request_buf[txn->offset], action, sizeof(u32) * action_len);
243 	txn->request_len = action_len;
244 
245 	txn->response = resp ?: txn->response_buf + txn->offset;
246 	txn->response_len = resp_size ?: GUC_RELAY_MSG_MAX_LEN;
247 	txn->reply = -ENOMSG;
248 	INIT_LIST_HEAD(&txn->link);
249 	init_completion(&txn->done);
250 
251 	return txn;
252 }
253 
254 static struct relay_transaction *
255 relay_new_transaction(struct xe_guc_relay *relay, u32 target, const u32 *action, u32 len,
256 		      u32 *resp, u32 resp_size)
257 {
258 	u32 rid = relay_get_next_rid(relay);
259 
260 	return __relay_get_transaction(relay, false, target, rid, action, len, resp, resp_size);
261 }
262 
263 static struct relay_transaction *
264 relay_new_incoming_transaction(struct xe_guc_relay *relay, u32 origin, u32 rid,
265 			       const u32 *action, u32 len)
266 {
267 	return __relay_get_transaction(relay, true, origin, rid, action, len, NULL, 0);
268 }
269 
270 static void relay_release_transaction(struct xe_guc_relay *relay, struct relay_transaction *txn)
271 {
272 	relay_assert(relay, list_empty(&txn->link));
273 
274 	txn->offset = 0;
275 	txn->response = NULL;
276 	txn->reply = -ESTALE;
277 	mempool_free(txn, &relay->pool);
278 }
279 
280 static int relay_send_transaction(struct xe_guc_relay *relay, struct relay_transaction *txn)
281 {
282 	u32 len = txn->incoming ? txn->response_len : txn->request_len;
283 	u32 *buf = txn->incoming ? txn->response_buf : txn->request_buf;
284 	u32 *msg = buf + txn->offset;
285 	int ret;
286 
287 	relay_assert(relay, txn->offset);
288 	relay_assert(relay, txn->offset + len <= GUC_CTB_MAX_DWORDS);
289 	relay_assert(relay, len >= GUC_RELAY_MSG_MIN_LEN);
290 	relay_assert(relay, len <= GUC_RELAY_MSG_MAX_LEN);
291 
292 	relay_debug(relay, "sending %s.%u to %u = %*ph\n",
293 		    guc_hxg_type_to_string(FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0])),
294 		    txn->rid, txn->remote, (int)sizeof(u32) * len, msg);
295 
296 	ret = xe_guc_ct_send_block(relay_to_ct(relay), buf, len + txn->offset);
297 
298 	if (unlikely(ret > 0)) {
299 		relay_notice(relay, "Unexpected data=%d from GuC, wrong ABI?\n", ret);
300 		ret = -EPROTO;
301 	}
302 	if (unlikely(ret < 0)) {
303 		relay_notice(relay, "Failed to send %s.%x to GuC (%pe) %*ph ...\n",
304 			     guc_hxg_type_to_string(FIELD_GET(GUC_HXG_MSG_0_TYPE, buf[0])),
305 			     FIELD_GET(GUC_HXG_REQUEST_MSG_0_ACTION, buf[0]),
306 			     ERR_PTR(ret), (int)sizeof(u32) * txn->offset, buf);
307 		relay_notice(relay, "Failed to send %s.%u to %u (%pe) %*ph\n",
308 			     guc_hxg_type_to_string(FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0])),
309 			     txn->rid, txn->remote, ERR_PTR(ret), (int)sizeof(u32) * len, msg);
310 	}
311 
312 	return ret;
313 }
314 
315 static void __fini_relay(struct drm_device *drm, void *arg)
316 {
317 	struct xe_guc_relay *relay = arg;
318 
319 	mempool_exit(&relay->pool);
320 }
321 
322 /**
323  * xe_guc_relay_init - Initialize a &xe_guc_relay
324  * @relay: the &xe_guc_relay to initialize
325  *
326  * Initialize remaining members of &xe_guc_relay that may depend
327  * on the SR-IOV mode.
328  *
329  * Return: 0 on success or a negative error code on failure.
330  */
331 int xe_guc_relay_init(struct xe_guc_relay *relay)
332 {
333 	const int XE_RELAY_MEMPOOL_MIN_NUM = 1;
334 	struct xe_device *xe = relay_to_xe(relay);
335 	int err;
336 
337 	relay_assert(relay, !relay_is_ready(relay));
338 
339 	if (!IS_SRIOV(xe))
340 		return 0;
341 
342 	spin_lock_init(&relay->lock);
343 	INIT_WORK(&relay->worker, relays_worker_fn);
344 	INIT_LIST_HEAD(&relay->pending_relays);
345 	INIT_LIST_HEAD(&relay->incoming_actions);
346 
347 	err = mempool_init_kmalloc_pool(&relay->pool, XE_RELAY_MEMPOOL_MIN_NUM +
348 					relay_get_totalvfs(relay),
349 					sizeof(struct relay_transaction));
350 	if (err)
351 		return err;
352 
353 	relay_debug(relay, "using mempool with %d elements\n", relay->pool.min_nr);
354 
355 	return drmm_add_action_or_reset(&xe->drm, __fini_relay, relay);
356 }
357 
358 static u32 to_relay_error(int err)
359 {
360 	/* XXX: assume that relay errors match errno codes */
361 	return err < 0 ? -err : GUC_RELAY_ERROR_UNDISCLOSED;
362 }
363 
364 static int from_relay_error(u32 error)
365 {
366 	/* XXX: assume that relay errors match errno codes */
367 	return error ? -error : -ENODATA;
368 }
369 
370 static u32 sanitize_relay_error(u32 error)
371 {
372 	/* XXX TBD if generic error codes will be allowed */
373 	if (!IS_ENABLED(CONFIG_DRM_XE_DEBUG))
374 		error = GUC_RELAY_ERROR_UNDISCLOSED;
375 	return error;
376 }
377 
378 static u32 sanitize_relay_error_hint(u32 hint)
379 {
380 	/* XXX TBD if generic error codes will be allowed */
381 	if (!IS_ENABLED(CONFIG_DRM_XE_DEBUG))
382 		hint = 0;
383 	return hint;
384 }
385 
386 static u32 prepare_error_reply(u32 *msg, u32 error, u32 hint)
387 {
388 	msg[0] = FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
389 		 FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_RESPONSE_FAILURE) |
390 		 FIELD_PREP(GUC_HXG_FAILURE_MSG_0_HINT, hint) |
391 		 FIELD_PREP(GUC_HXG_FAILURE_MSG_0_ERROR, error);
392 
393 	XE_WARN_ON(!FIELD_FIT(GUC_HXG_FAILURE_MSG_0_ERROR, error));
394 	XE_WARN_ON(!FIELD_FIT(GUC_HXG_FAILURE_MSG_0_HINT, hint));
395 
396 	return GUC_HXG_FAILURE_MSG_LEN;
397 }
398 
399 static void relay_testonly_nop(struct xe_guc_relay *relay)
400 {
401 	KUNIT_STATIC_STUB_REDIRECT(relay_testonly_nop, relay);
402 }
403 
404 static int relay_send_message_and_wait(struct xe_guc_relay *relay,
405 				       struct relay_transaction *txn,
406 				       u32 *buf, u32 buf_size)
407 {
408 	unsigned long timeout = msecs_to_jiffies(RELAY_TIMEOUT_MSEC);
409 	u32 *msg = &txn->request_buf[txn->offset];
410 	u32 len = txn->request_len;
411 	u32 type, action, data0;
412 	int ret;
413 	long n;
414 
415 	type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]);
416 	action = FIELD_GET(GUC_HXG_REQUEST_MSG_0_ACTION, msg[0]);
417 	data0 = FIELD_GET(GUC_HXG_REQUEST_MSG_0_DATA0, msg[0]);
418 
419 	relay_debug(relay, "%s.%u to %u action %#x:%u\n",
420 		    guc_hxg_type_to_string(type),
421 		    txn->rid, txn->remote, action, data0);
422 
423 	/* list ordering does not need to match RID ordering */
424 	spin_lock(&relay->lock);
425 	list_add_tail(&txn->link, &relay->pending_relays);
426 	spin_unlock(&relay->lock);
427 
428 resend:
429 	ret = relay_send_transaction(relay, txn);
430 	if (unlikely(ret < 0))
431 		goto unlink;
432 
433 wait:
434 	n = wait_for_completion_timeout(&txn->done, timeout);
435 	if (unlikely(n == 0 && txn->reply)) {
436 		ret = -ETIME;
437 		goto unlink;
438 	}
439 
440 	relay_debug(relay, "%u.%u reply %d after %u msec\n",
441 		    txn->remote, txn->rid, txn->reply, jiffies_to_msecs(timeout - n));
442 	if (unlikely(txn->reply)) {
443 		reinit_completion(&txn->done);
444 		if (txn->reply == -EAGAIN)
445 			goto resend;
446 		if (txn->reply == -EBUSY) {
447 			relay_testonly_nop(relay);
448 			goto wait;
449 		}
450 		if (txn->reply > 0)
451 			ret = from_relay_error(txn->reply);
452 		else
453 			ret = txn->reply;
454 		goto unlink;
455 	}
456 
457 	relay_debug(relay, "%u.%u response %*ph\n", txn->remote, txn->rid,
458 		    (int)sizeof(u32) * txn->response_len, txn->response);
459 	relay_assert(relay, txn->response_len >= GUC_RELAY_MSG_MIN_LEN);
460 	ret = txn->response_len;
461 
462 unlink:
463 	spin_lock(&relay->lock);
464 	list_del_init(&txn->link);
465 	spin_unlock(&relay->lock);
466 
467 	if (unlikely(ret < 0)) {
468 		relay_notice(relay, "Unsuccessful %s.%u %#x:%u to %u (%pe) %*ph\n",
469 			     guc_hxg_type_to_string(type), txn->rid,
470 			     action, data0, txn->remote, ERR_PTR(ret),
471 			     (int)sizeof(u32) * len, msg);
472 	}
473 
474 	return ret;
475 }
476 
477 static int relay_send_to(struct xe_guc_relay *relay, u32 target,
478 			 const u32 *msg, u32 len, u32 *buf, u32 buf_size)
479 {
480 	struct relay_transaction *txn;
481 	int ret;
482 
483 	relay_assert(relay, len >= GUC_RELAY_MSG_MIN_LEN);
484 	relay_assert(relay, len <= GUC_RELAY_MSG_MAX_LEN);
485 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_ORIGIN, msg[0]) == GUC_HXG_ORIGIN_HOST);
486 	relay_assert(relay, guc_hxg_type_is_action(FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0])));
487 
488 	if (unlikely(!relay_is_ready(relay)))
489 		return -ENODEV;
490 
491 	txn = relay_new_transaction(relay, target, msg, len, buf, buf_size);
492 	if (IS_ERR(txn))
493 		return PTR_ERR(txn);
494 
495 	switch (FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0])) {
496 	case GUC_HXG_TYPE_REQUEST:
497 		ret = relay_send_message_and_wait(relay, txn, buf, buf_size);
498 		break;
499 	case GUC_HXG_TYPE_FAST_REQUEST:
500 		relay_assert(relay, !GUC_HXG_TYPE_FAST_REQUEST);
501 		fallthrough;
502 	case GUC_HXG_TYPE_EVENT:
503 		ret = relay_send_transaction(relay, txn);
504 		break;
505 	default:
506 		ret = -EINVAL;
507 		break;
508 	}
509 
510 	relay_release_transaction(relay, txn);
511 	return ret;
512 }
513 
514 #ifdef CONFIG_PCI_IOV
515 /**
516  * xe_guc_relay_send_to_vf - Send a message to the VF.
517  * @relay: the &xe_guc_relay which will send the message
518  * @target: target VF number
519  * @msg: request message to be sent
520  * @len: length of the request message (in dwords, can't be 0)
521  * @buf: placeholder for the response message
522  * @buf_size: size of the response message placeholder (in dwords)
523  *
524  * This function can only be used by the driver running in the SR-IOV PF mode.
525  *
526  * Return: Non-negative response length (in dwords) or
527  *         a negative error code on failure.
528  */
529 int xe_guc_relay_send_to_vf(struct xe_guc_relay *relay, u32 target,
530 			    const u32 *msg, u32 len, u32 *buf, u32 buf_size)
531 {
532 	relay_assert(relay, IS_SRIOV_PF(relay_to_xe(relay)));
533 
534 	return relay_send_to(relay, target, msg, len, buf, buf_size);
535 }
536 #endif
537 
538 /**
539  * xe_guc_relay_send_to_pf - Send a message to the PF.
540  * @relay: the &xe_guc_relay which will send the message
541  * @msg: request message to be sent
542  * @len: length of the message (in dwords, can't be 0)
543  * @buf: placeholder for the response message
544  * @buf_size: size of the response message placeholder (in dwords)
545  *
546  * This function can only be used by driver running in SR-IOV VF mode.
547  *
548  * Return: Non-negative response length (in dwords) or
549  *         a negative error code on failure.
550  */
551 int xe_guc_relay_send_to_pf(struct xe_guc_relay *relay,
552 			    const u32 *msg, u32 len, u32 *buf, u32 buf_size)
553 {
554 	relay_assert(relay, IS_SRIOV_VF(relay_to_xe(relay)));
555 
556 	return relay_send_to(relay, PFID, msg, len, buf, buf_size);
557 }
558 
559 static int relay_handle_reply(struct xe_guc_relay *relay, u32 origin,
560 			      u32 rid, int reply, const u32 *msg, u32 len)
561 {
562 	struct relay_transaction *pending;
563 	int err = -ESRCH;
564 
565 	spin_lock(&relay->lock);
566 	list_for_each_entry(pending, &relay->pending_relays, link) {
567 		if (pending->remote != origin || pending->rid != rid) {
568 			relay_debug(relay, "%u.%u still awaits response\n",
569 				    pending->remote, pending->rid);
570 			continue;
571 		}
572 		err = 0; /* found! */
573 		if (reply == 0) {
574 			if (len > pending->response_len) {
575 				reply = -ENOBUFS;
576 				err = -ENOBUFS;
577 			} else {
578 				memcpy(pending->response, msg, 4 * len);
579 				pending->response_len = len;
580 			}
581 		}
582 		pending->reply = reply;
583 		complete_all(&pending->done);
584 		break;
585 	}
586 	spin_unlock(&relay->lock);
587 
588 	return err;
589 }
590 
591 static int relay_handle_failure(struct xe_guc_relay *relay, u32 origin,
592 				u32 rid, const u32 *msg, u32 len)
593 {
594 	int error = FIELD_GET(GUC_HXG_FAILURE_MSG_0_ERROR, msg[0]);
595 	u32 hint __maybe_unused = FIELD_GET(GUC_HXG_FAILURE_MSG_0_HINT, msg[0]);
596 
597 	relay_assert(relay, len);
598 	relay_debug(relay, "%u.%u error %#x (%pe) hint %u debug %*ph\n",
599 		    origin, rid, error, ERR_PTR(-error), hint, 4 * (len - 1), msg + 1);
600 
601 	return relay_handle_reply(relay, origin, rid, error ?: -EREMOTEIO, NULL, 0);
602 }
603 
604 static int relay_testloop_action_handler(struct xe_guc_relay *relay, u32 origin,
605 					 const u32 *msg, u32 len, u32 *response, u32 size)
606 {
607 	static ktime_t last_reply = 0;
608 	u32 type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]);
609 	u32 action = FIELD_GET(GUC_HXG_REQUEST_MSG_0_ACTION, msg[0]);
610 	u32 opcode = FIELD_GET(GUC_HXG_REQUEST_MSG_0_DATA0, msg[0]);
611 	ktime_t now = ktime_get();
612 	bool busy;
613 	int ret;
614 
615 	relay_assert(relay, guc_hxg_type_is_action(type));
616 	relay_assert(relay, action == GUC_RELAY_ACTION_VFXPF_TESTLOOP);
617 
618 	if (!IS_ENABLED(CONFIG_DRM_XE_DEBUG_SRIOV))
619 		return -ECONNREFUSED;
620 
621 	if (!last_reply)
622 		last_reply = now;
623 	busy = ktime_before(now, ktime_add_ms(last_reply, 2 * RELAY_TIMEOUT_MSEC));
624 	if (!busy)
625 		last_reply = now;
626 
627 	switch (opcode) {
628 	case VFXPF_TESTLOOP_OPCODE_NOP:
629 		if (type == GUC_HXG_TYPE_EVENT)
630 			return 0;
631 		return guc_hxg_msg_encode_success(response, 0);
632 	case VFXPF_TESTLOOP_OPCODE_BUSY:
633 		if (type == GUC_HXG_TYPE_EVENT)
634 			return -EPROTO;
635 		msleep(RELAY_TIMEOUT_MSEC / 8);
636 		if (busy)
637 			return -EINPROGRESS;
638 		return guc_hxg_msg_encode_success(response, 0);
639 	case VFXPF_TESTLOOP_OPCODE_RETRY:
640 		if (type == GUC_HXG_TYPE_EVENT)
641 			return -EPROTO;
642 		msleep(RELAY_TIMEOUT_MSEC / 8);
643 		if (busy)
644 			return guc_hxg_msg_encode_retry(response, 0);
645 		return guc_hxg_msg_encode_success(response, 0);
646 	case VFXPF_TESTLOOP_OPCODE_ECHO:
647 		if (type == GUC_HXG_TYPE_EVENT)
648 			return -EPROTO;
649 		if (size < len)
650 			return -ENOBUFS;
651 		ret = guc_hxg_msg_encode_success(response, len);
652 		memcpy(response + ret, msg + ret, (len - ret) * sizeof(u32));
653 		return len;
654 	case VFXPF_TESTLOOP_OPCODE_FAIL:
655 		return -EHWPOISON;
656 	default:
657 		break;
658 	}
659 
660 	relay_notice(relay, "Unexpected action %#x opcode %#x\n", action, opcode);
661 	return -EBADRQC;
662 }
663 
664 static int relay_action_handler(struct xe_guc_relay *relay, u32 origin,
665 				const u32 *msg, u32 len, u32 *response, u32 size)
666 {
667 	u32 type;
668 	int ret;
669 
670 	relay_assert(relay, len >= GUC_HXG_MSG_MIN_LEN);
671 
672 	if (FIELD_GET(GUC_HXG_REQUEST_MSG_0_ACTION, msg[0]) == GUC_RELAY_ACTION_VFXPF_TESTLOOP)
673 		return relay_testloop_action_handler(relay, origin, msg, len, response, size);
674 
675 	type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]);
676 
677 	/* XXX: PF services will be added later */
678 	ret = -EOPNOTSUPP;
679 
680 	if (type == GUC_HXG_TYPE_EVENT)
681 		relay_assert(relay, ret <= 0);
682 
683 	return ret;
684 }
685 
686 static struct relay_transaction *relay_dequeue_transaction(struct xe_guc_relay *relay)
687 {
688 	struct relay_transaction *txn;
689 
690 	spin_lock(&relay->lock);
691 	txn = list_first_entry_or_null(&relay->incoming_actions, struct relay_transaction, link);
692 	if (txn)
693 		list_del_init(&txn->link);
694 	spin_unlock(&relay->lock);
695 
696 	return txn;
697 }
698 
699 static void relay_process_incoming_action(struct xe_guc_relay *relay)
700 {
701 	struct relay_transaction *txn;
702 	bool again = false;
703 	u32 type;
704 	int ret;
705 
706 	txn = relay_dequeue_transaction(relay);
707 	if (!txn)
708 		return;
709 
710 	type = FIELD_GET(GUC_HXG_MSG_0_TYPE, txn->request_buf[txn->offset]);
711 
712 	ret = relay_action_handler(relay, txn->remote,
713 				   txn->request_buf + txn->offset, txn->request_len,
714 				   txn->response_buf + txn->offset,
715 				   ARRAY_SIZE(txn->response_buf) - txn->offset);
716 
717 	if (ret == -EINPROGRESS) {
718 		again = true;
719 		ret = guc_hxg_msg_encode_busy(txn->response_buf + txn->offset, 0);
720 	}
721 
722 	if (ret > 0) {
723 		txn->response_len = ret;
724 		ret = relay_send_transaction(relay, txn);
725 	}
726 
727 	if (ret < 0) {
728 		u32 error = to_relay_error(ret);
729 
730 		relay_notice(relay, "Failed to handle %s.%u from %u (%pe) %*ph\n",
731 			     guc_hxg_type_to_string(type), txn->rid, txn->remote,
732 			     ERR_PTR(ret), 4 * txn->request_len, txn->request_buf + txn->offset);
733 
734 		txn->response_len = prepare_error_reply(txn->response_buf + txn->offset,
735 							txn->remote ?
736 							sanitize_relay_error(error) : error,
737 							txn->remote ?
738 							sanitize_relay_error_hint(-ret) : -ret);
739 		ret = relay_send_transaction(relay, txn);
740 		again = false;
741 	}
742 
743 	if (again) {
744 		spin_lock(&relay->lock);
745 		list_add(&txn->link, &relay->incoming_actions);
746 		spin_unlock(&relay->lock);
747 		return;
748 	}
749 
750 	if (unlikely(ret < 0))
751 		relay_notice(relay, "Failed to process action.%u (%pe) %*ph\n",
752 			     txn->rid, ERR_PTR(ret), 4 * txn->request_len,
753 			     txn->request_buf + txn->offset);
754 
755 	relay_release_transaction(relay, txn);
756 }
757 
758 static bool relay_needs_worker(struct xe_guc_relay *relay)
759 {
760 	return !list_empty(&relay->incoming_actions);
761 }
762 
763 static void relay_kick_worker(struct xe_guc_relay *relay)
764 {
765 	KUNIT_STATIC_STUB_REDIRECT(relay_kick_worker, relay);
766 	queue_work(relay_to_xe(relay)->sriov.wq, &relay->worker);
767 }
768 
769 static void relays_worker_fn(struct work_struct *w)
770 {
771 	struct xe_guc_relay *relay = container_of(w, struct xe_guc_relay, worker);
772 
773 	relay_process_incoming_action(relay);
774 
775 	if (relay_needs_worker(relay))
776 		relay_kick_worker(relay);
777 }
778 
779 static int relay_queue_action_msg(struct xe_guc_relay *relay, u32 origin, u32 rid,
780 				  const u32 *msg, u32 len)
781 {
782 	struct relay_transaction *txn;
783 
784 	txn = relay_new_incoming_transaction(relay, origin, rid, msg, len);
785 	if (IS_ERR(txn))
786 		return PTR_ERR(txn);
787 
788 	spin_lock(&relay->lock);
789 	list_add_tail(&txn->link, &relay->incoming_actions);
790 	spin_unlock(&relay->lock);
791 
792 	relay_kick_worker(relay);
793 	return 0;
794 }
795 
796 static int relay_process_msg(struct xe_guc_relay *relay, u32 origin, u32 rid,
797 			     const u32 *msg, u32 len)
798 {
799 	u32 type;
800 	int err;
801 
802 	if (unlikely(len < GUC_HXG_MSG_MIN_LEN))
803 		return -EPROTO;
804 
805 	if (FIELD_GET(GUC_HXG_MSG_0_ORIGIN, msg[0]) != GUC_HXG_ORIGIN_HOST)
806 		return -EPROTO;
807 
808 	type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]);
809 	relay_debug(relay, "received %s.%u from %u = %*ph\n",
810 		    guc_hxg_type_to_string(type), rid, origin, 4 * len, msg);
811 
812 	switch (type) {
813 	case GUC_HXG_TYPE_REQUEST:
814 	case GUC_HXG_TYPE_FAST_REQUEST:
815 	case GUC_HXG_TYPE_EVENT:
816 		err = relay_queue_action_msg(relay, origin, rid, msg, len);
817 		break;
818 	case GUC_HXG_TYPE_RESPONSE_SUCCESS:
819 		err = relay_handle_reply(relay, origin, rid, 0, msg, len);
820 		break;
821 	case GUC_HXG_TYPE_NO_RESPONSE_BUSY:
822 		err = relay_handle_reply(relay, origin, rid, -EBUSY, NULL, 0);
823 		break;
824 	case GUC_HXG_TYPE_NO_RESPONSE_RETRY:
825 		err = relay_handle_reply(relay, origin, rid, -EAGAIN, NULL, 0);
826 		break;
827 	case GUC_HXG_TYPE_RESPONSE_FAILURE:
828 		err = relay_handle_failure(relay, origin, rid, msg, len);
829 		break;
830 	default:
831 		err = -EBADRQC;
832 	}
833 
834 	if (unlikely(err))
835 		relay_notice(relay, "Failed to process %s.%u from %u (%pe) %*ph\n",
836 			     guc_hxg_type_to_string(type), rid, origin,
837 			     ERR_PTR(err), 4 * len, msg);
838 
839 	return err;
840 }
841 
842 /**
843  * xe_guc_relay_process_guc2vf - Handle relay notification message from the GuC.
844  * @relay: the &xe_guc_relay which will handle the message
845  * @msg: message to be handled
846  * @len: length of the message (in dwords)
847  *
848  * This function will handle relay messages received from the GuC.
849  *
850  * This function is can only be used if driver is running in SR-IOV mode.
851  *
852  * Return: 0 on success or a negative error code on failure.
853  */
854 int xe_guc_relay_process_guc2vf(struct xe_guc_relay *relay, const u32 *msg, u32 len)
855 {
856 	u32 rid;
857 
858 	relay_assert(relay, len >= GUC_HXG_MSG_MIN_LEN);
859 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_ORIGIN, msg[0]) == GUC_HXG_ORIGIN_GUC);
860 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]) == GUC_HXG_TYPE_EVENT);
861 	relay_assert(relay, FIELD_GET(GUC_HXG_EVENT_MSG_0_ACTION, msg[0]) ==
862 		     XE_GUC_ACTION_GUC2VF_RELAY_FROM_PF);
863 
864 	if (unlikely(!IS_SRIOV_VF(relay_to_xe(relay)) && !kunit_get_current_test()))
865 		return -EPERM;
866 
867 	if (unlikely(!relay_is_ready(relay)))
868 		return -ENODEV;
869 
870 	if (unlikely(len < GUC2VF_RELAY_FROM_PF_EVENT_MSG_MIN_LEN))
871 		return -EPROTO;
872 
873 	if (unlikely(len > GUC2VF_RELAY_FROM_PF_EVENT_MSG_MAX_LEN))
874 		return -EMSGSIZE;
875 
876 	if (unlikely(FIELD_GET(GUC_HXG_EVENT_MSG_0_DATA0, msg[0])))
877 		return -EPFNOSUPPORT;
878 
879 	rid = FIELD_GET(GUC2VF_RELAY_FROM_PF_EVENT_MSG_1_RELAY_ID, msg[1]);
880 
881 	return relay_process_msg(relay, PFID, rid,
882 				 msg + GUC2VF_RELAY_FROM_PF_EVENT_MSG_MIN_LEN,
883 				 len - GUC2VF_RELAY_FROM_PF_EVENT_MSG_MIN_LEN);
884 }
885 
886 #ifdef CONFIG_PCI_IOV
887 /**
888  * xe_guc_relay_process_guc2pf - Handle relay notification message from the GuC.
889  * @relay: the &xe_guc_relay which will handle the message
890  * @msg: message to be handled
891  * @len: length of the message (in dwords)
892  *
893  * This function will handle relay messages received from the GuC.
894  *
895  * This function can only be used if driver is running in SR-IOV PF mode.
896  *
897  * Return: 0 on success or a negative error code on failure.
898  */
899 int xe_guc_relay_process_guc2pf(struct xe_guc_relay *relay, const u32 *msg, u32 len)
900 {
901 	u32 origin, rid;
902 	int err;
903 
904 	relay_assert(relay, len >= GUC_HXG_EVENT_MSG_MIN_LEN);
905 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_ORIGIN, msg[0]) == GUC_HXG_ORIGIN_GUC);
906 	relay_assert(relay, FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[0]) == GUC_HXG_TYPE_EVENT);
907 	relay_assert(relay, FIELD_GET(GUC_HXG_EVENT_MSG_0_ACTION, msg[0]) ==
908 		     XE_GUC_ACTION_GUC2PF_RELAY_FROM_VF);
909 
910 	if (unlikely(!IS_SRIOV_PF(relay_to_xe(relay)) && !kunit_get_current_test()))
911 		return -EPERM;
912 
913 	if (unlikely(!relay_is_ready(relay)))
914 		return -ENODEV;
915 
916 	if (unlikely(len < GUC2PF_RELAY_FROM_VF_EVENT_MSG_MIN_LEN))
917 		return -EPROTO;
918 
919 	if (unlikely(len > GUC2PF_RELAY_FROM_VF_EVENT_MSG_MAX_LEN))
920 		return -EMSGSIZE;
921 
922 	if (unlikely(FIELD_GET(GUC_HXG_EVENT_MSG_0_DATA0, msg[0])))
923 		return -EPFNOSUPPORT;
924 
925 	origin = FIELD_GET(GUC2PF_RELAY_FROM_VF_EVENT_MSG_1_VFID, msg[1]);
926 	rid = FIELD_GET(GUC2PF_RELAY_FROM_VF_EVENT_MSG_2_RELAY_ID, msg[2]);
927 
928 	if (unlikely(origin > relay_get_totalvfs(relay)))
929 		return -ENOENT;
930 
931 	err = relay_process_msg(relay, origin, rid,
932 				msg + GUC2PF_RELAY_FROM_VF_EVENT_MSG_MIN_LEN,
933 				len - GUC2PF_RELAY_FROM_VF_EVENT_MSG_MIN_LEN);
934 
935 	return err;
936 }
937 #endif
938 
939 #if IS_BUILTIN(CONFIG_DRM_XE_KUNIT_TEST)
940 #include "tests/xe_guc_relay_test.c"
941 #endif
942