xref: /linux/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c (revision 570f7e331f5febb30f1384817463c7e42b65ca7d)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include "amdgpu.h"
25 #include "nbio/nbio_2_3_offset.h"
26 #include "nbio/nbio_2_3_sh_mask.h"
27 #include "gc/gc_10_1_0_offset.h"
28 #include "gc/gc_10_1_0_sh_mask.h"
29 #include "soc15.h"
30 #include "navi10_ih.h"
31 #include "soc15_common.h"
32 #include "mxgpu_nv.h"
33 
34 #include "amdgpu_reset.h"
35 
36 static void xgpu_nv_mailbox_send_ack(struct amdgpu_device *adev)
37 {
38 	WREG8(NV_MAIBOX_CONTROL_RCV_OFFSET_BYTE, 2);
39 }
40 
41 static void xgpu_nv_mailbox_set_valid(struct amdgpu_device *adev, bool val)
42 {
43 	WREG8(NV_MAIBOX_CONTROL_TRN_OFFSET_BYTE, val ? 1 : 0);
44 }
45 
46 /*
47  * this peek_msg could *only* be called in IRQ routine becuase in IRQ routine
48  * RCV_MSG_VALID filed of BIF_BX_PF_MAILBOX_CONTROL must already be set to 1
49  * by host.
50  *
51  * if called no in IRQ routine, this peek_msg cannot guaranteed to return the
52  * correct value since it doesn't return the RCV_DW0 under the case that
53  * RCV_MSG_VALID is set by host.
54  */
55 static enum idh_event xgpu_nv_mailbox_peek_msg(struct amdgpu_device *adev)
56 {
57 	return RREG32_NO_KIQ(mmMAILBOX_MSGBUF_RCV_DW0);
58 }
59 
60 
61 static int xgpu_nv_mailbox_rcv_msg(struct amdgpu_device *adev,
62 				   enum idh_event event)
63 {
64 	int r = 0;
65 	u32 reg;
66 
67 	reg = RREG32_NO_KIQ(mmMAILBOX_MSGBUF_RCV_DW0);
68 	if (reg == IDH_FAIL)
69 		r = -EINVAL;
70 	if (reg == IDH_UNRECOV_ERR_NOTIFICATION)
71 		r = -ENODEV;
72 	else if (reg != event)
73 		return -ENOENT;
74 
75 	xgpu_nv_mailbox_send_ack(adev);
76 
77 	return r;
78 }
79 
80 static uint8_t xgpu_nv_peek_ack(struct amdgpu_device *adev)
81 {
82 	return RREG8(NV_MAIBOX_CONTROL_TRN_OFFSET_BYTE) & 2;
83 }
84 
85 static int xgpu_nv_poll_ack(struct amdgpu_device *adev)
86 {
87 	int timeout  = NV_MAILBOX_POLL_ACK_TIMEDOUT;
88 	u8 reg;
89 
90 	do {
91 		reg = RREG8(NV_MAIBOX_CONTROL_TRN_OFFSET_BYTE);
92 		if (reg & 2)
93 			return 0;
94 
95 		mdelay(5);
96 		timeout -= 5;
97 	} while (timeout > 1);
98 
99 	dev_err(adev->dev,
100 		"Doesn't get TRN_MSG_ACK from pf in %d msec\n",
101 		NV_MAILBOX_POLL_ACK_TIMEDOUT);
102 
103 	return -ETIME;
104 }
105 
106 static int xgpu_nv_poll_msg(struct amdgpu_device *adev, enum idh_event event)
107 {
108 	int r;
109 	uint64_t timeout, now;
110 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
111 
112 	now = (uint64_t)ktime_to_ms(ktime_get());
113 	timeout = now + NV_MAILBOX_POLL_MSG_TIMEDOUT;
114 
115 	do {
116 		r = xgpu_nv_mailbox_rcv_msg(adev, event);
117 		if (!r) {
118 			dev_dbg(adev->dev, "rcv_msg 0x%x after %llu ms\n",
119 					event, NV_MAILBOX_POLL_MSG_TIMEDOUT - timeout + now);
120 			return 0;
121 		} else if (r == -ENODEV) {
122 			if (!amdgpu_ras_is_rma(adev)) {
123 				ras->is_rma = true;
124 				dev_err(adev->dev, "VF is in an unrecoverable state. "
125 						"Runtime Services are halted.\n");
126 			}
127 			return r;
128 		}
129 
130 		msleep(10);
131 		now = (uint64_t)ktime_to_ms(ktime_get());
132 	} while (timeout > now);
133 
134 	dev_dbg(adev->dev, "nv_poll_msg timed out\n");
135 
136 	return -ETIME;
137 }
138 
139 static void xgpu_nv_mailbox_trans_msg (struct amdgpu_device *adev,
140 	      enum idh_request req, u32 data1, u32 data2, u32 data3)
141 {
142 	int r;
143 	uint8_t trn;
144 
145 	/* IMPORTANT:
146 	 * clear TRN_MSG_VALID valid to clear host's RCV_MSG_ACK
147 	 * and with host's RCV_MSG_ACK cleared hw automatically clear host's RCV_MSG_ACK
148 	 * which lead to VF's TRN_MSG_ACK cleared, otherwise below xgpu_nv_poll_ack()
149 	 * will return immediatly
150 	 */
151 	do {
152 		xgpu_nv_mailbox_set_valid(adev, false);
153 		trn = xgpu_nv_peek_ack(adev);
154 		if (trn) {
155 			dev_err_ratelimited(adev->dev, "trn=%x ACK should not assert! wait again !\n", trn);
156 			msleep(1);
157 		}
158 	} while (trn);
159 
160 	dev_dbg(adev->dev, "trans_msg req = 0x%x, data1 = 0x%x\n", req, data1);
161 	WREG32_NO_KIQ(mmMAILBOX_MSGBUF_TRN_DW0, req);
162 	WREG32_NO_KIQ(mmMAILBOX_MSGBUF_TRN_DW1, data1);
163 	WREG32_NO_KIQ(mmMAILBOX_MSGBUF_TRN_DW2, data2);
164 	WREG32_NO_KIQ(mmMAILBOX_MSGBUF_TRN_DW3, data3);
165 	xgpu_nv_mailbox_set_valid(adev, true);
166 
167 	/* start to poll ack */
168 	r = xgpu_nv_poll_ack(adev);
169 	if (r)
170 		dev_err(adev->dev, "Doesn't get ack from pf, continue\n");
171 
172 	xgpu_nv_mailbox_set_valid(adev, false);
173 }
174 
175 static int xgpu_nv_send_access_requests_with_param(struct amdgpu_device *adev,
176 			enum idh_request req, u32 data1, u32 data2, u32 data3)
177 {
178 	struct amdgpu_virt *virt = &adev->virt;
179 	int r = 0, retry = 1;
180 	enum idh_event event = -1;
181 
182 	mutex_lock(&virt->access_req_mutex);
183 send_request:
184 
185 	if (amdgpu_ras_is_rma(adev)) {
186 		r = -ENODEV;
187 		goto out;
188 	}
189 
190 	xgpu_nv_mailbox_trans_msg(adev, req, data1, data2, data3);
191 
192 	switch (req) {
193 	case IDH_REQ_GPU_INIT_ACCESS:
194 	case IDH_REQ_GPU_FINI_ACCESS:
195 	case IDH_REQ_GPU_RESET_ACCESS:
196 		event = IDH_READY_TO_ACCESS_GPU;
197 		break;
198 	case IDH_REQ_GPU_INIT_DATA:
199 		event = IDH_REQ_GPU_INIT_DATA_READY;
200 		break;
201 	case IDH_RAS_POISON:
202 		if (data1 != 0)
203 			event = IDH_RAS_POISON_READY;
204 		break;
205 	case IDH_REQ_RAS_ERROR_COUNT:
206 		event = IDH_RAS_ERROR_COUNT_READY;
207 		break;
208 	case IDH_REQ_RAS_CPER_DUMP:
209 		event = IDH_RAS_CPER_DUMP_READY;
210 		break;
211 	case IDH_REQ_RAS_CHK_CRITI:
212 		event = IDH_REQ_RAS_CHK_CRITI_READY;
213 		break;
214 	case IDH_REQ_RAS_REMOTE_CMD:
215 		event = IDH_REQ_RAS_REMOTE_CMD_READY;
216 		break;
217 	case IDH_REQ_PTL_UPDATE:
218 		event = IDH_PTL_UPDATE_READY;
219 		break;
220 	default:
221 		break;
222 	}
223 
224 	if (event != -1) {
225 		r = xgpu_nv_poll_msg(adev, event);
226 		if (r) {
227 			if (retry++ < 5)
228 				goto send_request;
229 
230 			if (req != IDH_REQ_GPU_INIT_DATA) {
231 				dev_err(adev->dev, "Doesn't get msg:%d from pf, error=%d\n", event, r);
232 				goto out;
233 			} else /* host doesn't support REQ_GPU_INIT_DATA handshake */
234 				adev->virt.req_init_data_ver = 0;
235 		} else {
236 			if (req == IDH_REQ_GPU_INIT_DATA) {
237 				switch (RREG32_NO_KIQ(mmMAILBOX_MSGBUF_RCV_DW1)) {
238 				case GPU_CRIT_REGION_V2:
239 					adev->virt.req_init_data_ver = GPU_CRIT_REGION_V2;
240 					adev->virt.init_data_header.offset =
241 						RREG32_NO_KIQ(mmMAILBOX_MSGBUF_RCV_DW2);
242 					adev->virt.init_data_header.size_kb =
243 						RREG32_NO_KIQ(mmMAILBOX_MSGBUF_RCV_DW3);
244 					break;
245 				default:
246 					adev->virt.req_init_data_ver = GPU_CRIT_REGION_V1;
247 					adev->virt.init_data_header.offset = -1;
248 					adev->virt.init_data_header.size_kb = 0;
249 					break;
250 				}
251 			}
252 		}
253 
254 		/* Retrieve checksum from mailbox2 */
255 		if (req == IDH_REQ_GPU_INIT_ACCESS || req == IDH_REQ_GPU_RESET_ACCESS) {
256 			adev->virt.fw_reserve.checksum_key =
257 				RREG32_NO_KIQ(mmMAILBOX_MSGBUF_RCV_DW2);
258 		}
259 
260 		/* Retrieve PTL response from mailbox */
261 		if (req == IDH_REQ_PTL_UPDATE) {
262 			u32 dw1 = RREG32_NO_KIQ(mmMAILBOX_MSGBUF_RCV_DW1);
263 			u32 dw2 = RREG32_NO_KIQ(mmMAILBOX_MSGBUF_RCV_DW2);
264 			u32 status = AMD_SRIOV_PTL_UNPACK_STATUS(dw1);
265 
266 			if (status == AMD_SRIOV_RESP_UNSUPPORTED) {
267 				r = -EOPNOTSUPP;
268 			} else if (status == AMD_SRIOV_RESP_FAIL) {
269 				r = -EIO;
270 			} else {
271 				adev->virt.ptl_state = AMD_SRIOV_PTL_UNPACK_STATE(dw1);
272 				adev->virt.ptl_pref_format1 = AMD_SRIOV_PTL_UNPACK_FMT1(dw2);
273 				adev->virt.ptl_pref_format2 = AMD_SRIOV_PTL_UNPACK_FMT2(dw2);
274 			}
275 		}
276 	}
277 
278 out:
279 	mutex_unlock(&virt->access_req_mutex);
280 
281 	return r;
282 }
283 
284 static int xgpu_nv_send_access_requests(struct amdgpu_device *adev,
285 					enum idh_request req)
286 {
287 	return xgpu_nv_send_access_requests_with_param(adev,
288 						req, 0, 0, 0);
289 }
290 
291 static int xgpu_nv_request_reset(struct amdgpu_device *adev)
292 {
293 	int ret, i = 0;
294 
295 	while (i < NV_MAILBOX_POLL_MSG_REP_MAX) {
296 		ret = xgpu_nv_send_access_requests(adev, IDH_REQ_GPU_RESET_ACCESS);
297 		if (!ret)
298 			break;
299 		i++;
300 	}
301 
302 	return ret;
303 }
304 
305 static int xgpu_nv_request_full_gpu_access(struct amdgpu_device *adev,
306 					   bool init)
307 {
308 	enum idh_request req;
309 
310 	req = init ? IDH_REQ_GPU_INIT_ACCESS : IDH_REQ_GPU_FINI_ACCESS;
311 	return xgpu_nv_send_access_requests(adev, req);
312 }
313 
314 static int xgpu_nv_release_full_gpu_access(struct amdgpu_device *adev,
315 					   bool init)
316 {
317 	enum idh_request req;
318 	int r = 0;
319 
320 	req = init ? IDH_REL_GPU_INIT_ACCESS : IDH_REL_GPU_FINI_ACCESS;
321 	r = xgpu_nv_send_access_requests(adev, req);
322 
323 	return r;
324 }
325 
326 static int xgpu_nv_request_init_data(struct amdgpu_device *adev)
327 {
328 	return xgpu_nv_send_access_requests_with_param(adev, IDH_REQ_GPU_INIT_DATA,
329 			0, GPU_CRIT_REGION_V2, 0);
330 }
331 
332 static int xgpu_nv_mailbox_ack_irq(struct amdgpu_device *adev,
333 					struct amdgpu_irq_src *source,
334 					struct amdgpu_iv_entry *entry)
335 {
336 	dev_dbg(adev->dev, "get ack intr and do nothing.\n");
337 	return 0;
338 }
339 
340 static int xgpu_nv_set_mailbox_ack_irq(struct amdgpu_device *adev,
341 					struct amdgpu_irq_src *source,
342 					unsigned type,
343 					enum amdgpu_interrupt_state state)
344 {
345 	u32 tmp = RREG32_NO_KIQ(mmMAILBOX_INT_CNTL);
346 
347 	if (state == AMDGPU_IRQ_STATE_ENABLE)
348 		tmp |= 2;
349 	else
350 		tmp &= ~2;
351 
352 	WREG32_NO_KIQ(mmMAILBOX_INT_CNTL, tmp);
353 
354 	return 0;
355 }
356 
357 static void xgpu_nv_ready_to_reset(struct amdgpu_device *adev)
358 {
359 	xgpu_nv_mailbox_trans_msg(adev, IDH_READY_TO_RESET, 0, 0, 0);
360 }
361 
362 static int xgpu_nv_wait_reset(struct amdgpu_device *adev)
363 {
364 	int timeout = NV_MAILBOX_POLL_FLR_TIMEDOUT;
365 	do {
366 		if (xgpu_nv_mailbox_peek_msg(adev) == IDH_FLR_NOTIFICATION_CMPL) {
367 			dev_dbg(adev->dev, "Got NV IDH_FLR_NOTIFICATION_CMPL after %d ms\n", NV_MAILBOX_POLL_FLR_TIMEDOUT - timeout);
368 			return 0;
369 		}
370 		msleep(10);
371 		timeout -= 10;
372 	} while (timeout > 1);
373 
374 	dev_dbg(adev->dev, "waiting NV IDH_FLR_NOTIFICATION_CMPL timeout\n");
375 	return -ETIME;
376 }
377 
378 static void xgpu_nv_mailbox_flr_work(struct work_struct *work)
379 {
380 	struct amdgpu_virt *virt = container_of(work, struct amdgpu_virt, flr_work);
381 	struct amdgpu_device *adev = container_of(virt, struct amdgpu_device, virt);
382 	struct amdgpu_reset_context reset_context = { 0 };
383 
384 	amdgpu_virt_fini_data_exchange(adev);
385 
386 	/* Trigger recovery for world switch failure if no TDR */
387 	if (amdgpu_device_should_recover_gpu(adev)
388 		&& (!amdgpu_device_has_job_running(adev) ||
389 		adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT ||
390 		adev->gfx_timeout == MAX_SCHEDULE_TIMEOUT ||
391 		adev->compute_timeout == MAX_SCHEDULE_TIMEOUT ||
392 		adev->video_timeout == MAX_SCHEDULE_TIMEOUT)) {
393 
394 		reset_context.method = AMD_RESET_METHOD_NONE;
395 		reset_context.reset_req_dev = adev;
396 		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
397 		set_bit(AMDGPU_HOST_FLR, &reset_context.flags);
398 
399 		amdgpu_device_gpu_recover(adev, NULL, &reset_context);
400 	}
401 }
402 
403 static void xgpu_nv_mailbox_req_bad_pages_work(struct work_struct *work)
404 {
405 	struct amdgpu_virt *virt = container_of(work, struct amdgpu_virt, req_bad_pages_work);
406 	struct amdgpu_device *adev = container_of(virt, struct amdgpu_device, virt);
407 
408 	if (down_read_trylock(&adev->reset_domain->sem)) {
409 		amdgpu_virt_fini_data_exchange(adev);
410 		amdgpu_virt_request_bad_pages(adev);
411 		up_read(&adev->reset_domain->sem);
412 	}
413 }
414 
415 /**
416  * xgpu_nv_mailbox_handle_bad_pages_work - Reinitialize the data exchange region to get fresh bad page information
417  * @work: pointer to the work_struct
418  *
419  * This work handler is triggered when bad pages are ready, and it reinitializes
420  * the data exchange region to retrieve updated bad page information from the host.
421  */
422 static void xgpu_nv_mailbox_handle_bad_pages_work(struct work_struct *work)
423 {
424 	struct amdgpu_virt *virt = container_of(work, struct amdgpu_virt, handle_bad_pages_work);
425 	struct amdgpu_device *adev = container_of(virt, struct amdgpu_device, virt);
426 
427 	if (down_read_trylock(&adev->reset_domain->sem)) {
428 		amdgpu_virt_fini_data_exchange(adev);
429 		amdgpu_virt_init_data_exchange(adev);
430 		up_read(&adev->reset_domain->sem);
431 	}
432 }
433 
434 static int xgpu_nv_set_mailbox_rcv_irq(struct amdgpu_device *adev,
435 				       struct amdgpu_irq_src *src,
436 				       unsigned type,
437 				       enum amdgpu_interrupt_state state)
438 {
439 	u32 tmp = RREG32_NO_KIQ(mmMAILBOX_INT_CNTL);
440 
441 	if (state == AMDGPU_IRQ_STATE_ENABLE)
442 		tmp |= 1;
443 	else
444 		tmp &= ~1;
445 
446 	WREG32_NO_KIQ(mmMAILBOX_INT_CNTL, tmp);
447 
448 	return 0;
449 }
450 
451 static int xgpu_nv_mailbox_rcv_irq(struct amdgpu_device *adev,
452 				   struct amdgpu_irq_src *source,
453 				   struct amdgpu_iv_entry *entry)
454 {
455 	enum idh_event event = xgpu_nv_mailbox_peek_msg(adev);
456 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
457 
458 	switch (event) {
459 	case IDH_RAS_BAD_PAGES_READY:
460 		xgpu_nv_mailbox_send_ack(adev);
461 		if (amdgpu_sriov_runtime(adev))
462 			schedule_work(&adev->virt.handle_bad_pages_work);
463 		break;
464 	case IDH_RAS_BAD_PAGES_NOTIFICATION:
465 		xgpu_nv_mailbox_send_ack(adev);
466 		if (amdgpu_sriov_runtime(adev))
467 			schedule_work(&adev->virt.req_bad_pages_work);
468 		break;
469 	case IDH_UNRECOV_ERR_NOTIFICATION:
470 		xgpu_nv_mailbox_send_ack(adev);
471 		if (!amdgpu_ras_is_rma(adev)) {
472 			ras->is_rma = true;
473 			dev_err(adev->dev, "VF is in an unrecoverable state. Runtime Services are halted.\n");
474 		}
475 
476 		if (amdgpu_sriov_runtime(adev))
477 			WARN_ONCE(!amdgpu_reset_domain_schedule(adev->reset_domain,
478 						&adev->virt.flr_work),
479 					"Failed to queue work! at %s",
480 					__func__);
481 		break;
482 	case IDH_FLR_NOTIFICATION:
483 		if (amdgpu_sriov_runtime(adev))
484 			WARN_ONCE(!amdgpu_reset_domain_schedule(adev->reset_domain,
485 				   &adev->virt.flr_work),
486 				  "Failed to queue work! at %s",
487 				  __func__);
488 		break;
489 		/* READY_TO_ACCESS_GPU is fetched by kernel polling, IRQ can ignore
490 		 * it byfar since that polling thread will handle it,
491 		 * other msg like flr complete is not handled here.
492 		 */
493 	case IDH_CLR_MSG_BUF:
494 	case IDH_FLR_NOTIFICATION_CMPL:
495 	case IDH_READY_TO_ACCESS_GPU:
496 	default:
497 		break;
498 	}
499 
500 	return 0;
501 }
502 
503 static const struct amdgpu_irq_src_funcs xgpu_nv_mailbox_ack_irq_funcs = {
504 	.set = xgpu_nv_set_mailbox_ack_irq,
505 	.process = xgpu_nv_mailbox_ack_irq,
506 };
507 
508 static const struct amdgpu_irq_src_funcs xgpu_nv_mailbox_rcv_irq_funcs = {
509 	.set = xgpu_nv_set_mailbox_rcv_irq,
510 	.process = xgpu_nv_mailbox_rcv_irq,
511 };
512 
513 void xgpu_nv_mailbox_set_irq_funcs(struct amdgpu_device *adev)
514 {
515 	adev->virt.ack_irq.num_types = 1;
516 	adev->virt.ack_irq.funcs = &xgpu_nv_mailbox_ack_irq_funcs;
517 	adev->virt.rcv_irq.num_types = 1;
518 	adev->virt.rcv_irq.funcs = &xgpu_nv_mailbox_rcv_irq_funcs;
519 }
520 
521 int xgpu_nv_mailbox_add_irq_id(struct amdgpu_device *adev)
522 {
523 	int r;
524 
525 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_BIF, 135, &adev->virt.rcv_irq);
526 	if (r)
527 		return r;
528 
529 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_BIF, 138, &adev->virt.ack_irq);
530 	if (r) {
531 		amdgpu_irq_put(adev, &adev->virt.rcv_irq, 0);
532 		return r;
533 	}
534 
535 	return 0;
536 }
537 
538 int xgpu_nv_mailbox_get_irq(struct amdgpu_device *adev)
539 {
540 	int r;
541 
542 	r = amdgpu_irq_get(adev, &adev->virt.rcv_irq, 0);
543 	if (r)
544 		return r;
545 	r = amdgpu_irq_get(adev, &adev->virt.ack_irq, 0);
546 	if (r) {
547 		amdgpu_irq_put(adev, &adev->virt.rcv_irq, 0);
548 		return r;
549 	}
550 
551 	INIT_WORK(&adev->virt.flr_work, xgpu_nv_mailbox_flr_work);
552 	INIT_WORK(&adev->virt.req_bad_pages_work, xgpu_nv_mailbox_req_bad_pages_work);
553 	INIT_WORK(&adev->virt.handle_bad_pages_work, xgpu_nv_mailbox_handle_bad_pages_work);
554 
555 	return 0;
556 }
557 
558 void xgpu_nv_mailbox_put_irq(struct amdgpu_device *adev)
559 {
560 	amdgpu_irq_put(adev, &adev->virt.ack_irq, 0);
561 	amdgpu_irq_put(adev, &adev->virt.rcv_irq, 0);
562 }
563 
564 static void xgpu_nv_ras_poison_handler(struct amdgpu_device *adev,
565 		enum amdgpu_ras_block block)
566 {
567 	if (amdgpu_ip_version(adev, UMC_HWIP, 0) < IP_VERSION(12, 0, 0)) {
568 		xgpu_nv_send_access_requests(adev, IDH_RAS_POISON);
569 	} else {
570 		amdgpu_virt_fini_data_exchange(adev);
571 		xgpu_nv_send_access_requests_with_param(adev,
572 					IDH_RAS_POISON,	block, 0, 0);
573 	}
574 }
575 
576 static bool xgpu_nv_rcvd_ras_intr(struct amdgpu_device *adev)
577 {
578 	enum idh_event msg = xgpu_nv_mailbox_peek_msg(adev);
579 
580 	return (msg == IDH_RAS_ERROR_DETECTED || msg == 0xFFFFFFFF);
581 }
582 
583 static int xgpu_nv_req_ras_err_count(struct amdgpu_device *adev)
584 {
585 	return xgpu_nv_send_access_requests(adev, IDH_REQ_RAS_ERROR_COUNT);
586 }
587 
588 static int xgpu_nv_req_ras_cper_dump(struct amdgpu_device *adev, u64 vf_rptr)
589 {
590 	uint32_t vf_rptr_hi, vf_rptr_lo;
591 
592 	vf_rptr_hi = (uint32_t)(vf_rptr >> 32);
593 	vf_rptr_lo = (uint32_t)(vf_rptr & 0xFFFFFFFF);
594 	return xgpu_nv_send_access_requests_with_param(
595 		adev, IDH_REQ_RAS_CPER_DUMP, vf_rptr_hi, vf_rptr_lo, 0);
596 }
597 
598 static int xgpu_nv_req_ras_bad_pages(struct amdgpu_device *adev)
599 {
600 	return xgpu_nv_send_access_requests(adev, IDH_REQ_RAS_BAD_PAGES);
601 }
602 
603 static int xgpu_nv_check_vf_critical_region(struct amdgpu_device *adev, u64 addr)
604 {
605 	uint32_t addr_hi, addr_lo;
606 
607 	addr_hi = (uint32_t)(addr >> 32);
608 	addr_lo = (uint32_t)(addr & 0xFFFFFFFF);
609 	return xgpu_nv_send_access_requests_with_param(
610 		adev, IDH_REQ_RAS_CHK_CRITI, addr_hi, addr_lo, 0);
611 }
612 
613 static int xgpu_nv_req_remote_ras_cmd(struct amdgpu_device *adev,
614 		u32 param1, u32 param2, u32 param3)
615 {
616 	return xgpu_nv_send_access_requests_with_param(
617 		adev, IDH_REQ_RAS_REMOTE_CMD, param1, param2, param3);
618 }
619 
620 static int xgpu_nv_req_ptl_update(struct amdgpu_device *adev,
621 				  u32 req_code, u32 ptl_state, u32 fmt1, u32 fmt2)
622 {
623 	return xgpu_nv_send_access_requests_with_param(adev, IDH_REQ_PTL_UPDATE,
624 			req_code, ptl_state, AMD_SRIOV_PTL_PACK_FORMATS(fmt1, fmt2));
625 }
626 
627 const struct amdgpu_virt_ops xgpu_nv_virt_ops = {
628 	.req_full_gpu	= xgpu_nv_request_full_gpu_access,
629 	.rel_full_gpu	= xgpu_nv_release_full_gpu_access,
630 	.req_init_data  = xgpu_nv_request_init_data,
631 	.reset_gpu = xgpu_nv_request_reset,
632 	.ready_to_reset = xgpu_nv_ready_to_reset,
633 	.wait_reset = xgpu_nv_wait_reset,
634 	.trans_msg = xgpu_nv_mailbox_trans_msg,
635 	.ras_poison_handler = xgpu_nv_ras_poison_handler,
636 	.rcvd_ras_intr = xgpu_nv_rcvd_ras_intr,
637 	.req_ras_err_count = xgpu_nv_req_ras_err_count,
638 	.req_ras_cper_dump = xgpu_nv_req_ras_cper_dump,
639 	.req_bad_pages = xgpu_nv_req_ras_bad_pages,
640 	.req_ras_chk_criti = xgpu_nv_check_vf_critical_region,
641 	.req_remote_ras_cmd = xgpu_nv_req_remote_ras_cmd,
642 	.req_ptl_update = xgpu_nv_req_ptl_update,
643 };
644