xref: /linux/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/r535.c (revision 110d3047a3ec033de00322b1a8068b1215efa97a)
1 /*
2  * Copyright 2023 Red Hat 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 #include "priv.h"
23 
24 #include <core/pci.h>
25 #include <subdev/timer.h>
26 #include <subdev/vfn.h>
27 #include <engine/fifo/chan.h>
28 #include <engine/sec2.h>
29 
30 #include <nvfw/fw.h>
31 
32 #include <nvrm/nvtypes.h>
33 #include <nvrm/535.113.01/common/sdk/nvidia/inc/class/cl0000.h>
34 #include <nvrm/535.113.01/common/sdk/nvidia/inc/class/cl0005.h>
35 #include <nvrm/535.113.01/common/sdk/nvidia/inc/class/cl0080.h>
36 #include <nvrm/535.113.01/common/sdk/nvidia/inc/class/cl2080.h>
37 #include <nvrm/535.113.01/common/sdk/nvidia/inc/ctrl/ctrl2080/ctrl2080event.h>
38 #include <nvrm/535.113.01/common/sdk/nvidia/inc/ctrl/ctrl2080/ctrl2080gpu.h>
39 #include <nvrm/535.113.01/common/sdk/nvidia/inc/ctrl/ctrl2080/ctrl2080internal.h>
40 #include <nvrm/535.113.01/common/sdk/nvidia/inc/nvos.h>
41 #include <nvrm/535.113.01/common/shared/msgq/inc/msgq/msgq_priv.h>
42 #include <nvrm/535.113.01/common/uproc/os/common/include/libos_init_args.h>
43 #include <nvrm/535.113.01/nvidia/arch/nvalloc/common/inc/gsp/gsp_fw_sr_meta.h>
44 #include <nvrm/535.113.01/nvidia/arch/nvalloc/common/inc/gsp/gsp_fw_wpr_meta.h>
45 #include <nvrm/535.113.01/nvidia/arch/nvalloc/common/inc/rmRiscvUcode.h>
46 #include <nvrm/535.113.01/nvidia/arch/nvalloc/common/inc/rmgspseq.h>
47 #include <nvrm/535.113.01/nvidia/generated/g_allclasses.h>
48 #include <nvrm/535.113.01/nvidia/generated/g_os_nvoc.h>
49 #include <nvrm/535.113.01/nvidia/generated/g_rpc-structures.h>
50 #include <nvrm/535.113.01/nvidia/inc/kernel/gpu/gsp/gsp_fw_heap.h>
51 #include <nvrm/535.113.01/nvidia/inc/kernel/gpu/gsp/gsp_init_args.h>
52 #include <nvrm/535.113.01/nvidia/inc/kernel/gpu/gsp/gsp_static_config.h>
53 #include <nvrm/535.113.01/nvidia/inc/kernel/gpu/intr/engine_idx.h>
54 #include <nvrm/535.113.01/nvidia/kernel/inc/vgpu/rpc_global_enums.h>
55 
56 #include <linux/acpi.h>
57 
58 #define GSP_MSG_MIN_SIZE GSP_PAGE_SIZE
59 #define GSP_MSG_MAX_SIZE GSP_PAGE_MIN_SIZE * 16
60 
61 struct r535_gsp_msg {
62 	u8 auth_tag_buffer[16];
63 	u8 aad_buffer[16];
64 	u32 checksum;
65 	u32 sequence;
66 	u32 elem_count;
67 	u32 pad;
68 	u8  data[];
69 };
70 
71 #define GSP_MSG_HDR_SIZE offsetof(struct r535_gsp_msg, data)
72 
73 static int
74 r535_rpc_status_to_errno(uint32_t rpc_status)
75 {
76 	switch (rpc_status) {
77 	case 0x55: /* NV_ERR_NOT_READY */
78 	case 0x66: /* NV_ERR_TIMEOUT_RETRY */
79 		return -EAGAIN;
80 	case 0x51: /* NV_ERR_NO_MEMORY */
81 		return -ENOMEM;
82 	default:
83 		return -EINVAL;
84 	}
85 }
86 
87 static void *
88 r535_gsp_msgq_wait(struct nvkm_gsp *gsp, u32 repc, u32 *prepc, int *ptime)
89 {
90 	struct r535_gsp_msg *mqe;
91 	u32 size, rptr = *gsp->msgq.rptr;
92 	int used;
93 	u8 *msg;
94 	u32 len;
95 
96 	size = DIV_ROUND_UP(GSP_MSG_HDR_SIZE + repc, GSP_PAGE_SIZE);
97 	if (WARN_ON(!size || size >= gsp->msgq.cnt))
98 		return ERR_PTR(-EINVAL);
99 
100 	do {
101 		u32 wptr = *gsp->msgq.wptr;
102 
103 		used = wptr + gsp->msgq.cnt - rptr;
104 		if (used >= gsp->msgq.cnt)
105 			used -= gsp->msgq.cnt;
106 		if (used >= size)
107 			break;
108 
109 		usleep_range(1, 2);
110 	} while (--(*ptime));
111 
112 	if (WARN_ON(!*ptime))
113 		return ERR_PTR(-ETIMEDOUT);
114 
115 	mqe = (void *)((u8 *)gsp->shm.msgq.ptr + 0x1000 + rptr * 0x1000);
116 
117 	if (prepc) {
118 		*prepc = (used * GSP_PAGE_SIZE) - sizeof(*mqe);
119 		return mqe->data;
120 	}
121 
122 	msg = kvmalloc(repc, GFP_KERNEL);
123 	if (!msg)
124 		return ERR_PTR(-ENOMEM);
125 
126 	len = ((gsp->msgq.cnt - rptr) * GSP_PAGE_SIZE) - sizeof(*mqe);
127 	len = min_t(u32, repc, len);
128 	memcpy(msg, mqe->data, len);
129 
130 	rptr += DIV_ROUND_UP(len, GSP_PAGE_SIZE);
131 	if (rptr == gsp->msgq.cnt)
132 		rptr = 0;
133 
134 	repc -= len;
135 
136 	if (repc) {
137 		mqe = (void *)((u8 *)gsp->shm.msgq.ptr + 0x1000 + 0 * 0x1000);
138 		memcpy(msg + len, mqe, repc);
139 
140 		rptr += DIV_ROUND_UP(repc, GSP_PAGE_SIZE);
141 	}
142 
143 	mb();
144 	(*gsp->msgq.rptr) = rptr;
145 	return msg;
146 }
147 
148 static void *
149 r535_gsp_msgq_recv(struct nvkm_gsp *gsp, u32 repc, int *ptime)
150 {
151 	return r535_gsp_msgq_wait(gsp, repc, NULL, ptime);
152 }
153 
154 static int
155 r535_gsp_cmdq_push(struct nvkm_gsp *gsp, void *argv)
156 {
157 	struct r535_gsp_msg *cmd = container_of(argv, typeof(*cmd), data);
158 	struct r535_gsp_msg *cqe;
159 	u32 argc = cmd->checksum;
160 	u64 *ptr = (void *)cmd;
161 	u64 *end;
162 	u64 csum = 0;
163 	int free, time = 1000000;
164 	u32 wptr, size;
165 	u32 off = 0;
166 
167 	argc = ALIGN(GSP_MSG_HDR_SIZE + argc, GSP_PAGE_SIZE);
168 
169 	end = (u64 *)((char *)ptr + argc);
170 	cmd->pad = 0;
171 	cmd->checksum = 0;
172 	cmd->sequence = gsp->cmdq.seq++;
173 	cmd->elem_count = DIV_ROUND_UP(argc, 0x1000);
174 
175 	while (ptr < end)
176 		csum ^= *ptr++;
177 
178 	cmd->checksum = upper_32_bits(csum) ^ lower_32_bits(csum);
179 
180 	wptr = *gsp->cmdq.wptr;
181 	do {
182 		do {
183 			free = *gsp->cmdq.rptr + gsp->cmdq.cnt - wptr - 1;
184 			if (free >= gsp->cmdq.cnt)
185 				free -= gsp->cmdq.cnt;
186 			if (free >= 1)
187 				break;
188 
189 			usleep_range(1, 2);
190 		} while(--time);
191 
192 		if (WARN_ON(!time)) {
193 			kvfree(cmd);
194 			return -ETIMEDOUT;
195 		}
196 
197 		cqe = (void *)((u8 *)gsp->shm.cmdq.ptr + 0x1000 + wptr * 0x1000);
198 		size = min_t(u32, argc, (gsp->cmdq.cnt - wptr) * GSP_PAGE_SIZE);
199 		memcpy(cqe, (u8 *)cmd + off, size);
200 
201 		wptr += DIV_ROUND_UP(size, 0x1000);
202 		if (wptr == gsp->cmdq.cnt)
203 			wptr = 0;
204 
205 		off  += size;
206 		argc -= size;
207 	} while(argc);
208 
209 	nvkm_trace(&gsp->subdev, "cmdq: wptr %d\n", wptr);
210 	wmb();
211 	(*gsp->cmdq.wptr) = wptr;
212 	mb();
213 
214 	nvkm_falcon_wr32(&gsp->falcon, 0xc00, 0x00000000);
215 
216 	kvfree(cmd);
217 	return 0;
218 }
219 
220 static void *
221 r535_gsp_cmdq_get(struct nvkm_gsp *gsp, u32 argc)
222 {
223 	struct r535_gsp_msg *cmd;
224 	u32 size = GSP_MSG_HDR_SIZE + argc;
225 
226 	size = ALIGN(size, GSP_MSG_MIN_SIZE);
227 	cmd = kvzalloc(size, GFP_KERNEL);
228 	if (!cmd)
229 		return ERR_PTR(-ENOMEM);
230 
231 	cmd->checksum = argc;
232 	return cmd->data;
233 }
234 
235 struct nvfw_gsp_rpc {
236 	u32 header_version;
237 	u32 signature;
238 	u32 length;
239 	u32 function;
240 	u32 rpc_result;
241 	u32 rpc_result_private;
242 	u32 sequence;
243 	union {
244 		u32 spare;
245 		u32 cpuRmGfid;
246 	};
247 	u8  data[];
248 };
249 
250 static void
251 r535_gsp_msg_done(struct nvkm_gsp *gsp, struct nvfw_gsp_rpc *msg)
252 {
253 	kvfree(msg);
254 }
255 
256 static void
257 r535_gsp_msg_dump(struct nvkm_gsp *gsp, struct nvfw_gsp_rpc *msg, int lvl)
258 {
259 	if (gsp->subdev.debug >= lvl) {
260 		nvkm_printk__(&gsp->subdev, lvl, info,
261 			      "msg fn:%d len:0x%x/0x%zx res:0x%x resp:0x%x\n",
262 			      msg->function, msg->length, msg->length - sizeof(*msg),
263 			      msg->rpc_result, msg->rpc_result_private);
264 		print_hex_dump(KERN_INFO, "msg: ", DUMP_PREFIX_OFFSET, 16, 1,
265 			       msg->data, msg->length - sizeof(*msg), true);
266 	}
267 }
268 
269 static struct nvfw_gsp_rpc *
270 r535_gsp_msg_recv(struct nvkm_gsp *gsp, int fn, u32 repc)
271 {
272 	struct nvkm_subdev *subdev = &gsp->subdev;
273 	struct nvfw_gsp_rpc *msg;
274 	int time = 4000000, i;
275 	u32 size;
276 
277 retry:
278 	msg = r535_gsp_msgq_wait(gsp, sizeof(*msg), &size, &time);
279 	if (IS_ERR_OR_NULL(msg))
280 		return msg;
281 
282 	msg = r535_gsp_msgq_recv(gsp, msg->length, &time);
283 	if (IS_ERR_OR_NULL(msg))
284 		return msg;
285 
286 	if (msg->rpc_result) {
287 		r535_gsp_msg_dump(gsp, msg, NV_DBG_ERROR);
288 		r535_gsp_msg_done(gsp, msg);
289 		return ERR_PTR(-EINVAL);
290 	}
291 
292 	r535_gsp_msg_dump(gsp, msg, NV_DBG_TRACE);
293 
294 	if (fn && msg->function == fn) {
295 		if (repc) {
296 			if (msg->length < sizeof(*msg) + repc) {
297 				nvkm_error(subdev, "msg len %d < %zd\n",
298 					   msg->length, sizeof(*msg) + repc);
299 				r535_gsp_msg_dump(gsp, msg, NV_DBG_ERROR);
300 				r535_gsp_msg_done(gsp, msg);
301 				return ERR_PTR(-EIO);
302 			}
303 
304 			return msg;
305 		}
306 
307 		r535_gsp_msg_done(gsp, msg);
308 		return NULL;
309 	}
310 
311 	for (i = 0; i < gsp->msgq.ntfy_nr; i++) {
312 		struct nvkm_gsp_msgq_ntfy *ntfy = &gsp->msgq.ntfy[i];
313 
314 		if (ntfy->fn == msg->function) {
315 			if (ntfy->func)
316 				ntfy->func(ntfy->priv, ntfy->fn, msg->data, msg->length - sizeof(*msg));
317 			break;
318 		}
319 	}
320 
321 	if (i == gsp->msgq.ntfy_nr)
322 		r535_gsp_msg_dump(gsp, msg, NV_DBG_WARN);
323 
324 	r535_gsp_msg_done(gsp, msg);
325 	if (fn)
326 		goto retry;
327 
328 	if (*gsp->msgq.rptr != *gsp->msgq.wptr)
329 		goto retry;
330 
331 	return NULL;
332 }
333 
334 static int
335 r535_gsp_msg_ntfy_add(struct nvkm_gsp *gsp, u32 fn, nvkm_gsp_msg_ntfy_func func, void *priv)
336 {
337 	int ret = 0;
338 
339 	mutex_lock(&gsp->msgq.mutex);
340 	if (WARN_ON(gsp->msgq.ntfy_nr >= ARRAY_SIZE(gsp->msgq.ntfy))) {
341 		ret = -ENOSPC;
342 	} else {
343 		gsp->msgq.ntfy[gsp->msgq.ntfy_nr].fn = fn;
344 		gsp->msgq.ntfy[gsp->msgq.ntfy_nr].func = func;
345 		gsp->msgq.ntfy[gsp->msgq.ntfy_nr].priv = priv;
346 		gsp->msgq.ntfy_nr++;
347 	}
348 	mutex_unlock(&gsp->msgq.mutex);
349 	return ret;
350 }
351 
352 static int
353 r535_gsp_rpc_poll(struct nvkm_gsp *gsp, u32 fn)
354 {
355 	void *repv;
356 
357 	mutex_lock(&gsp->cmdq.mutex);
358 	repv = r535_gsp_msg_recv(gsp, fn, 0);
359 	mutex_unlock(&gsp->cmdq.mutex);
360 	if (IS_ERR(repv))
361 		return PTR_ERR(repv);
362 
363 	return 0;
364 }
365 
366 static void *
367 r535_gsp_rpc_send(struct nvkm_gsp *gsp, void *argv, bool wait, u32 repc)
368 {
369 	struct nvfw_gsp_rpc *rpc = container_of(argv, typeof(*rpc), data);
370 	struct nvfw_gsp_rpc *msg;
371 	u32 fn = rpc->function;
372 	void *repv = NULL;
373 	int ret;
374 
375 	if (gsp->subdev.debug >= NV_DBG_TRACE) {
376 		nvkm_trace(&gsp->subdev, "rpc fn:%d len:0x%x/0x%zx\n", rpc->function,
377 			   rpc->length, rpc->length - sizeof(*rpc));
378 		print_hex_dump(KERN_INFO, "rpc: ", DUMP_PREFIX_OFFSET, 16, 1,
379 			       rpc->data, rpc->length - sizeof(*rpc), true);
380 	}
381 
382 	ret = r535_gsp_cmdq_push(gsp, rpc);
383 	if (ret)
384 		return ERR_PTR(ret);
385 
386 	if (wait) {
387 		msg = r535_gsp_msg_recv(gsp, fn, repc);
388 		if (!IS_ERR_OR_NULL(msg))
389 			repv = msg->data;
390 		else
391 			repv = msg;
392 	}
393 
394 	return repv;
395 }
396 
397 static void
398 r535_gsp_event_dtor(struct nvkm_gsp_event *event)
399 {
400 	struct nvkm_gsp_device *device = event->device;
401 	struct nvkm_gsp_client *client = device->object.client;
402 	struct nvkm_gsp *gsp = client->gsp;
403 
404 	mutex_lock(&gsp->client_id.mutex);
405 	if (event->func) {
406 		list_del(&event->head);
407 		event->func = NULL;
408 	}
409 	mutex_unlock(&gsp->client_id.mutex);
410 
411 	nvkm_gsp_rm_free(&event->object);
412 	event->device = NULL;
413 }
414 
415 static int
416 r535_gsp_device_event_get(struct nvkm_gsp_event *event)
417 {
418 	struct nvkm_gsp_device *device = event->device;
419 	NV2080_CTRL_EVENT_SET_NOTIFICATION_PARAMS *ctrl;
420 
421 	ctrl = nvkm_gsp_rm_ctrl_get(&device->subdevice,
422 				    NV2080_CTRL_CMD_EVENT_SET_NOTIFICATION, sizeof(*ctrl));
423 	if (IS_ERR(ctrl))
424 		return PTR_ERR(ctrl);
425 
426 	ctrl->event = event->id;
427 	ctrl->action = NV2080_CTRL_EVENT_SET_NOTIFICATION_ACTION_REPEAT;
428 	return nvkm_gsp_rm_ctrl_wr(&device->subdevice, ctrl);
429 }
430 
431 static int
432 r535_gsp_device_event_ctor(struct nvkm_gsp_device *device, u32 handle, u32 id,
433 			   nvkm_gsp_event_func func, struct nvkm_gsp_event *event)
434 {
435 	struct nvkm_gsp_client *client = device->object.client;
436 	struct nvkm_gsp *gsp = client->gsp;
437 	NV0005_ALLOC_PARAMETERS *args;
438 	int ret;
439 
440 	args = nvkm_gsp_rm_alloc_get(&device->subdevice, handle,
441 				     NV01_EVENT_KERNEL_CALLBACK_EX, sizeof(*args),
442 				     &event->object);
443 	if (IS_ERR(args))
444 		return PTR_ERR(args);
445 
446 	args->hParentClient = client->object.handle;
447 	args->hSrcResource = 0;
448 	args->hClass = NV01_EVENT_KERNEL_CALLBACK_EX;
449 	args->notifyIndex = NV01_EVENT_CLIENT_RM | id;
450 	args->data = NULL;
451 
452 	ret = nvkm_gsp_rm_alloc_wr(&event->object, args);
453 	if (ret)
454 		return ret;
455 
456 	event->device = device;
457 	event->id = id;
458 
459 	ret = r535_gsp_device_event_get(event);
460 	if (ret) {
461 		nvkm_gsp_event_dtor(event);
462 		return ret;
463 	}
464 
465 	mutex_lock(&gsp->client_id.mutex);
466 	event->func = func;
467 	list_add(&event->head, &client->events);
468 	mutex_unlock(&gsp->client_id.mutex);
469 	return 0;
470 }
471 
472 static void
473 r535_gsp_device_dtor(struct nvkm_gsp_device *device)
474 {
475 	nvkm_gsp_rm_free(&device->subdevice);
476 	nvkm_gsp_rm_free(&device->object);
477 }
478 
479 static int
480 r535_gsp_subdevice_ctor(struct nvkm_gsp_device *device)
481 {
482 	NV2080_ALLOC_PARAMETERS *args;
483 
484 	return nvkm_gsp_rm_alloc(&device->object, 0x5d1d0000, NV20_SUBDEVICE_0, sizeof(*args),
485 				 &device->subdevice);
486 }
487 
488 static int
489 r535_gsp_device_ctor(struct nvkm_gsp_client *client, struct nvkm_gsp_device *device)
490 {
491 	NV0080_ALLOC_PARAMETERS *args;
492 	int ret;
493 
494 	args = nvkm_gsp_rm_alloc_get(&client->object, 0xde1d0000, NV01_DEVICE_0, sizeof(*args),
495 				     &device->object);
496 	if (IS_ERR(args))
497 		return PTR_ERR(args);
498 
499 	args->hClientShare = client->object.handle;
500 
501 	ret = nvkm_gsp_rm_alloc_wr(&device->object, args);
502 	if (ret)
503 		return ret;
504 
505 	ret = r535_gsp_subdevice_ctor(device);
506 	if (ret)
507 		nvkm_gsp_rm_free(&device->object);
508 
509 	return ret;
510 }
511 
512 static void
513 r535_gsp_client_dtor(struct nvkm_gsp_client *client)
514 {
515 	struct nvkm_gsp *gsp = client->gsp;
516 
517 	nvkm_gsp_rm_free(&client->object);
518 
519 	mutex_lock(&gsp->client_id.mutex);
520 	idr_remove(&gsp->client_id.idr, client->object.handle & 0xffff);
521 	mutex_unlock(&gsp->client_id.mutex);
522 
523 	client->gsp = NULL;
524 }
525 
526 static int
527 r535_gsp_client_ctor(struct nvkm_gsp *gsp, struct nvkm_gsp_client *client)
528 {
529 	NV0000_ALLOC_PARAMETERS *args;
530 	int ret;
531 
532 	mutex_lock(&gsp->client_id.mutex);
533 	ret = idr_alloc(&gsp->client_id.idr, client, 0, 0xffff + 1, GFP_KERNEL);
534 	mutex_unlock(&gsp->client_id.mutex);
535 	if (ret < 0)
536 		return ret;
537 
538 	client->gsp = gsp;
539 	client->object.client = client;
540 	INIT_LIST_HEAD(&client->events);
541 
542 	args = nvkm_gsp_rm_alloc_get(&client->object, 0xc1d00000 | ret, NV01_ROOT, sizeof(*args),
543 				     &client->object);
544 	if (IS_ERR(args)) {
545 		r535_gsp_client_dtor(client);
546 		return ret;
547 	}
548 
549 	args->hClient = client->object.handle;
550 	args->processID = ~0;
551 
552 	ret = nvkm_gsp_rm_alloc_wr(&client->object, args);
553 	if (ret) {
554 		r535_gsp_client_dtor(client);
555 		return ret;
556 	}
557 
558 	return 0;
559 }
560 
561 static int
562 r535_gsp_rpc_rm_free(struct nvkm_gsp_object *object)
563 {
564 	struct nvkm_gsp_client *client = object->client;
565 	struct nvkm_gsp *gsp = client->gsp;
566 	rpc_free_v03_00 *rpc;
567 
568 	nvkm_debug(&gsp->subdev, "cli:0x%08x obj:0x%08x free\n",
569 		   client->object.handle, object->handle);
570 
571 	rpc = nvkm_gsp_rpc_get(gsp, NV_VGPU_MSG_FUNCTION_FREE, sizeof(*rpc));
572 	if (WARN_ON(IS_ERR_OR_NULL(rpc)))
573 		return -EIO;
574 
575 	rpc->params.hRoot = client->object.handle;
576 	rpc->params.hObjectParent = 0;
577 	rpc->params.hObjectOld = object->handle;
578 	return nvkm_gsp_rpc_wr(gsp, rpc, true);
579 }
580 
581 static void
582 r535_gsp_rpc_rm_alloc_done(struct nvkm_gsp_object *object, void *repv)
583 {
584 	rpc_gsp_rm_alloc_v03_00 *rpc = container_of(repv, typeof(*rpc), params);
585 
586 	nvkm_gsp_rpc_done(object->client->gsp, rpc);
587 }
588 
589 static void *
590 r535_gsp_rpc_rm_alloc_push(struct nvkm_gsp_object *object, void *argv, u32 repc)
591 {
592 	rpc_gsp_rm_alloc_v03_00 *rpc = container_of(argv, typeof(*rpc), params);
593 	struct nvkm_gsp *gsp = object->client->gsp;
594 	void *ret;
595 
596 	rpc = nvkm_gsp_rpc_push(gsp, rpc, true, sizeof(*rpc) + repc);
597 	if (IS_ERR_OR_NULL(rpc))
598 		return rpc;
599 
600 	if (rpc->status) {
601 		ret = ERR_PTR(r535_rpc_status_to_errno(rpc->status));
602 		if (PTR_ERR(ret) != -EAGAIN)
603 			nvkm_error(&gsp->subdev, "RM_ALLOC: 0x%x\n", rpc->status);
604 	} else {
605 		ret = repc ? rpc->params : NULL;
606 	}
607 
608 	nvkm_gsp_rpc_done(gsp, rpc);
609 
610 	return ret;
611 }
612 
613 static void *
614 r535_gsp_rpc_rm_alloc_get(struct nvkm_gsp_object *object, u32 oclass, u32 argc)
615 {
616 	struct nvkm_gsp_client *client = object->client;
617 	struct nvkm_gsp *gsp = client->gsp;
618 	rpc_gsp_rm_alloc_v03_00 *rpc;
619 
620 	nvkm_debug(&gsp->subdev, "cli:0x%08x obj:0x%08x new obj:0x%08x cls:0x%08x argc:%d\n",
621 		   client->object.handle, object->parent->handle, object->handle, oclass, argc);
622 
623 	rpc = nvkm_gsp_rpc_get(gsp, NV_VGPU_MSG_FUNCTION_GSP_RM_ALLOC, sizeof(*rpc) + argc);
624 	if (IS_ERR(rpc))
625 		return rpc;
626 
627 	rpc->hClient = client->object.handle;
628 	rpc->hParent = object->parent->handle;
629 	rpc->hObject = object->handle;
630 	rpc->hClass = oclass;
631 	rpc->status = 0;
632 	rpc->paramsSize = argc;
633 	return rpc->params;
634 }
635 
636 static void
637 r535_gsp_rpc_rm_ctrl_done(struct nvkm_gsp_object *object, void *repv)
638 {
639 	rpc_gsp_rm_control_v03_00 *rpc = container_of(repv, typeof(*rpc), params);
640 
641 	if (!repv)
642 		return;
643 	nvkm_gsp_rpc_done(object->client->gsp, rpc);
644 }
645 
646 static int
647 r535_gsp_rpc_rm_ctrl_push(struct nvkm_gsp_object *object, void **argv, u32 repc)
648 {
649 	rpc_gsp_rm_control_v03_00 *rpc = container_of((*argv), typeof(*rpc), params);
650 	struct nvkm_gsp *gsp = object->client->gsp;
651 	int ret = 0;
652 
653 	rpc = nvkm_gsp_rpc_push(gsp, rpc, true, repc);
654 	if (IS_ERR_OR_NULL(rpc)) {
655 		*argv = NULL;
656 		return PTR_ERR(rpc);
657 	}
658 
659 	if (rpc->status) {
660 		ret = r535_rpc_status_to_errno(rpc->status);
661 		if (ret != -EAGAIN)
662 			nvkm_error(&gsp->subdev, "cli:0x%08x obj:0x%08x ctrl cmd:0x%08x failed: 0x%08x\n",
663 				   object->client->object.handle, object->handle, rpc->cmd, rpc->status);
664 	}
665 
666 	if (repc)
667 		*argv = rpc->params;
668 	else
669 		nvkm_gsp_rpc_done(gsp, rpc);
670 
671 	return ret;
672 }
673 
674 static void *
675 r535_gsp_rpc_rm_ctrl_get(struct nvkm_gsp_object *object, u32 cmd, u32 argc)
676 {
677 	struct nvkm_gsp_client *client = object->client;
678 	struct nvkm_gsp *gsp = client->gsp;
679 	rpc_gsp_rm_control_v03_00 *rpc;
680 
681 	nvkm_debug(&gsp->subdev, "cli:0x%08x obj:0x%08x ctrl cmd:0x%08x argc:%d\n",
682 		   client->object.handle, object->handle, cmd, argc);
683 
684 	rpc = nvkm_gsp_rpc_get(gsp, NV_VGPU_MSG_FUNCTION_GSP_RM_CONTROL, sizeof(*rpc) + argc);
685 	if (IS_ERR(rpc))
686 		return rpc;
687 
688 	rpc->hClient    = client->object.handle;
689 	rpc->hObject    = object->handle;
690 	rpc->cmd	= cmd;
691 	rpc->status     = 0;
692 	rpc->paramsSize = argc;
693 	return rpc->params;
694 }
695 
696 static void
697 r535_gsp_rpc_done(struct nvkm_gsp *gsp, void *repv)
698 {
699 	struct nvfw_gsp_rpc *rpc = container_of(repv, typeof(*rpc), data);
700 
701 	r535_gsp_msg_done(gsp, rpc);
702 }
703 
704 static void *
705 r535_gsp_rpc_get(struct nvkm_gsp *gsp, u32 fn, u32 argc)
706 {
707 	struct nvfw_gsp_rpc *rpc;
708 
709 	rpc = r535_gsp_cmdq_get(gsp, ALIGN(sizeof(*rpc) + argc, sizeof(u64)));
710 	if (IS_ERR(rpc))
711 		return ERR_CAST(rpc);
712 
713 	rpc->header_version = 0x03000000;
714 	rpc->signature = ('C' << 24) | ('P' << 16) | ('R' << 8) | 'V';
715 	rpc->function = fn;
716 	rpc->rpc_result = 0xffffffff;
717 	rpc->rpc_result_private = 0xffffffff;
718 	rpc->length = sizeof(*rpc) + argc;
719 	return rpc->data;
720 }
721 
722 static void *
723 r535_gsp_rpc_push(struct nvkm_gsp *gsp, void *argv, bool wait, u32 repc)
724 {
725 	struct nvfw_gsp_rpc *rpc = container_of(argv, typeof(*rpc), data);
726 	struct r535_gsp_msg *cmd = container_of((void *)rpc, typeof(*cmd), data);
727 	const u32 max_msg_size = (16 * 0x1000) - sizeof(struct r535_gsp_msg);
728 	const u32 max_rpc_size = max_msg_size - sizeof(*rpc);
729 	u32 rpc_size = rpc->length - sizeof(*rpc);
730 	void *repv;
731 
732 	mutex_lock(&gsp->cmdq.mutex);
733 	if (rpc_size > max_rpc_size) {
734 		const u32 fn = rpc->function;
735 
736 		/* Adjust length, and send initial RPC. */
737 		rpc->length = sizeof(*rpc) + max_rpc_size;
738 		cmd->checksum = rpc->length;
739 
740 		repv = r535_gsp_rpc_send(gsp, argv, false, 0);
741 		if (IS_ERR(repv))
742 			goto done;
743 
744 		argv += max_rpc_size;
745 		rpc_size -= max_rpc_size;
746 
747 		/* Remaining chunks sent as CONTINUATION_RECORD RPCs. */
748 		while (rpc_size) {
749 			u32 size = min(rpc_size, max_rpc_size);
750 			void *next;
751 
752 			next = r535_gsp_rpc_get(gsp, NV_VGPU_MSG_FUNCTION_CONTINUATION_RECORD, size);
753 			if (IS_ERR(next)) {
754 				repv = next;
755 				goto done;
756 			}
757 
758 			memcpy(next, argv, size);
759 
760 			repv = r535_gsp_rpc_send(gsp, next, false, 0);
761 			if (IS_ERR(repv))
762 				goto done;
763 
764 			argv += size;
765 			rpc_size -= size;
766 		}
767 
768 		/* Wait for reply. */
769 		if (wait) {
770 			rpc = r535_gsp_msg_recv(gsp, fn, repc);
771 			if (!IS_ERR_OR_NULL(rpc))
772 				repv = rpc->data;
773 			else
774 				repv = rpc;
775 		} else {
776 			repv = NULL;
777 		}
778 	} else {
779 		repv = r535_gsp_rpc_send(gsp, argv, wait, repc);
780 	}
781 
782 done:
783 	mutex_unlock(&gsp->cmdq.mutex);
784 	return repv;
785 }
786 
787 const struct nvkm_gsp_rm
788 r535_gsp_rm = {
789 	.rpc_get = r535_gsp_rpc_get,
790 	.rpc_push = r535_gsp_rpc_push,
791 	.rpc_done = r535_gsp_rpc_done,
792 
793 	.rm_ctrl_get = r535_gsp_rpc_rm_ctrl_get,
794 	.rm_ctrl_push = r535_gsp_rpc_rm_ctrl_push,
795 	.rm_ctrl_done = r535_gsp_rpc_rm_ctrl_done,
796 
797 	.rm_alloc_get = r535_gsp_rpc_rm_alloc_get,
798 	.rm_alloc_push = r535_gsp_rpc_rm_alloc_push,
799 	.rm_alloc_done = r535_gsp_rpc_rm_alloc_done,
800 
801 	.rm_free = r535_gsp_rpc_rm_free,
802 
803 	.client_ctor = r535_gsp_client_ctor,
804 	.client_dtor = r535_gsp_client_dtor,
805 
806 	.device_ctor = r535_gsp_device_ctor,
807 	.device_dtor = r535_gsp_device_dtor,
808 
809 	.event_ctor = r535_gsp_device_event_ctor,
810 	.event_dtor = r535_gsp_event_dtor,
811 };
812 
813 static void
814 r535_gsp_msgq_work(struct work_struct *work)
815 {
816 	struct nvkm_gsp *gsp = container_of(work, typeof(*gsp), msgq.work);
817 
818 	mutex_lock(&gsp->cmdq.mutex);
819 	if (*gsp->msgq.rptr != *gsp->msgq.wptr)
820 		r535_gsp_msg_recv(gsp, 0, 0);
821 	mutex_unlock(&gsp->cmdq.mutex);
822 }
823 
824 static irqreturn_t
825 r535_gsp_intr(struct nvkm_inth *inth)
826 {
827 	struct nvkm_gsp *gsp = container_of(inth, typeof(*gsp), subdev.inth);
828 	struct nvkm_subdev *subdev = &gsp->subdev;
829 	u32 intr = nvkm_falcon_rd32(&gsp->falcon, 0x0008);
830 	u32 inte = nvkm_falcon_rd32(&gsp->falcon, gsp->falcon.func->addr2 +
831 						  gsp->falcon.func->riscv_irqmask);
832 	u32 stat = intr & inte;
833 
834 	if (!stat) {
835 		nvkm_debug(subdev, "inte %08x %08x\n", intr, inte);
836 		return IRQ_NONE;
837 	}
838 
839 	if (stat & 0x00000040) {
840 		nvkm_falcon_wr32(&gsp->falcon, 0x004, 0x00000040);
841 		schedule_work(&gsp->msgq.work);
842 		stat &= ~0x00000040;
843 	}
844 
845 	if (stat) {
846 		nvkm_error(subdev, "intr %08x\n", stat);
847 		nvkm_falcon_wr32(&gsp->falcon, 0x014, stat);
848 		nvkm_falcon_wr32(&gsp->falcon, 0x004, stat);
849 	}
850 
851 	nvkm_falcon_intr_retrigger(&gsp->falcon);
852 	return IRQ_HANDLED;
853 }
854 
855 static int
856 r535_gsp_intr_get_table(struct nvkm_gsp *gsp)
857 {
858 	NV2080_CTRL_INTERNAL_INTR_GET_KERNEL_TABLE_PARAMS *ctrl;
859 	int ret = 0;
860 
861 	ctrl = nvkm_gsp_rm_ctrl_get(&gsp->internal.device.subdevice,
862 				    NV2080_CTRL_CMD_INTERNAL_INTR_GET_KERNEL_TABLE, sizeof(*ctrl));
863 	if (IS_ERR(ctrl))
864 		return PTR_ERR(ctrl);
865 
866 	ret = nvkm_gsp_rm_ctrl_push(&gsp->internal.device.subdevice, &ctrl, sizeof(*ctrl));
867 	if (WARN_ON(ret)) {
868 		nvkm_gsp_rm_ctrl_done(&gsp->internal.device.subdevice, ctrl);
869 		return ret;
870 	}
871 
872 	for (unsigned i = 0; i < ctrl->tableLen; i++) {
873 		enum nvkm_subdev_type type;
874 		int inst;
875 
876 		nvkm_debug(&gsp->subdev,
877 			   "%2d: engineIdx %3d pmcIntrMask %08x stall %08x nonStall %08x\n", i,
878 			   ctrl->table[i].engineIdx, ctrl->table[i].pmcIntrMask,
879 			   ctrl->table[i].vectorStall, ctrl->table[i].vectorNonStall);
880 
881 		switch (ctrl->table[i].engineIdx) {
882 		case MC_ENGINE_IDX_GSP:
883 			type = NVKM_SUBDEV_GSP;
884 			inst = 0;
885 			break;
886 		case MC_ENGINE_IDX_DISP:
887 			type = NVKM_ENGINE_DISP;
888 			inst = 0;
889 			break;
890 		case MC_ENGINE_IDX_CE0 ... MC_ENGINE_IDX_CE9:
891 			type = NVKM_ENGINE_CE;
892 			inst = ctrl->table[i].engineIdx - MC_ENGINE_IDX_CE0;
893 			break;
894 		case MC_ENGINE_IDX_GR0:
895 			type = NVKM_ENGINE_GR;
896 			inst = 0;
897 			break;
898 		case MC_ENGINE_IDX_NVDEC0 ... MC_ENGINE_IDX_NVDEC7:
899 			type = NVKM_ENGINE_NVDEC;
900 			inst = ctrl->table[i].engineIdx - MC_ENGINE_IDX_NVDEC0;
901 			break;
902 		case MC_ENGINE_IDX_MSENC ... MC_ENGINE_IDX_MSENC2:
903 			type = NVKM_ENGINE_NVENC;
904 			inst = ctrl->table[i].engineIdx - MC_ENGINE_IDX_MSENC;
905 			break;
906 		case MC_ENGINE_IDX_NVJPEG0 ... MC_ENGINE_IDX_NVJPEG7:
907 			type = NVKM_ENGINE_NVJPG;
908 			inst = ctrl->table[i].engineIdx - MC_ENGINE_IDX_NVJPEG0;
909 			break;
910 		case MC_ENGINE_IDX_OFA0:
911 			type = NVKM_ENGINE_OFA;
912 			inst = 0;
913 			break;
914 		default:
915 			continue;
916 		}
917 
918 		if (WARN_ON(gsp->intr_nr == ARRAY_SIZE(gsp->intr))) {
919 			ret = -ENOSPC;
920 			break;
921 		}
922 
923 		gsp->intr[gsp->intr_nr].type = type;
924 		gsp->intr[gsp->intr_nr].inst = inst;
925 		gsp->intr[gsp->intr_nr].stall = ctrl->table[i].vectorStall;
926 		gsp->intr[gsp->intr_nr].nonstall = ctrl->table[i].vectorNonStall;
927 		gsp->intr_nr++;
928 	}
929 
930 	nvkm_gsp_rm_ctrl_done(&gsp->internal.device.subdevice, ctrl);
931 	return ret;
932 }
933 
934 static int
935 r535_gsp_rpc_get_gsp_static_info(struct nvkm_gsp *gsp)
936 {
937 	GspStaticConfigInfo *rpc;
938 	int last_usable = -1;
939 
940 	rpc = nvkm_gsp_rpc_rd(gsp, NV_VGPU_MSG_FUNCTION_GET_GSP_STATIC_INFO, sizeof(*rpc));
941 	if (IS_ERR(rpc))
942 		return PTR_ERR(rpc);
943 
944 	gsp->internal.client.object.client = &gsp->internal.client;
945 	gsp->internal.client.object.parent = NULL;
946 	gsp->internal.client.object.handle = rpc->hInternalClient;
947 	gsp->internal.client.gsp = gsp;
948 
949 	gsp->internal.device.object.client = &gsp->internal.client;
950 	gsp->internal.device.object.parent = &gsp->internal.client.object;
951 	gsp->internal.device.object.handle = rpc->hInternalDevice;
952 
953 	gsp->internal.device.subdevice.client = &gsp->internal.client;
954 	gsp->internal.device.subdevice.parent = &gsp->internal.device.object;
955 	gsp->internal.device.subdevice.handle = rpc->hInternalSubdevice;
956 
957 	gsp->bar.rm_bar1_pdb = rpc->bar1PdeBase;
958 	gsp->bar.rm_bar2_pdb = rpc->bar2PdeBase;
959 
960 	for (int i = 0; i < rpc->fbRegionInfoParams.numFBRegions; i++) {
961 		NV2080_CTRL_CMD_FB_GET_FB_REGION_FB_REGION_INFO *reg =
962 			&rpc->fbRegionInfoParams.fbRegion[i];
963 
964 		nvkm_debug(&gsp->subdev, "fb region %d: "
965 			   "%016llx-%016llx rsvd:%016llx perf:%08x comp:%d iso:%d prot:%d\n", i,
966 			   reg->base, reg->limit, reg->reserved, reg->performance,
967 			   reg->supportCompressed, reg->supportISO, reg->bProtected);
968 
969 		if (!reg->reserved && !reg->bProtected) {
970 			if (reg->supportCompressed && reg->supportISO &&
971 			    !WARN_ON_ONCE(gsp->fb.region_nr >= ARRAY_SIZE(gsp->fb.region))) {
972 					const u64 size = (reg->limit + 1) - reg->base;
973 
974 					gsp->fb.region[gsp->fb.region_nr].addr = reg->base;
975 					gsp->fb.region[gsp->fb.region_nr].size = size;
976 					gsp->fb.region_nr++;
977 			}
978 
979 			last_usable = i;
980 		}
981 	}
982 
983 	if (last_usable >= 0) {
984 		u32 rsvd_base = rpc->fbRegionInfoParams.fbRegion[last_usable].limit + 1;
985 
986 		gsp->fb.rsvd_size = gsp->fb.heap.addr - rsvd_base;
987 	}
988 
989 	for (int gpc = 0; gpc < ARRAY_SIZE(rpc->tpcInfo); gpc++) {
990 		if (rpc->gpcInfo.gpcMask & BIT(gpc)) {
991 			gsp->gr.tpcs += hweight32(rpc->tpcInfo[gpc].tpcMask);
992 			gsp->gr.gpcs++;
993 		}
994 	}
995 
996 	nvkm_gsp_rpc_done(gsp, rpc);
997 	return 0;
998 }
999 
1000 static int
1001 r535_gsp_postinit(struct nvkm_gsp *gsp)
1002 {
1003 	struct nvkm_device *device = gsp->subdev.device;
1004 	int ret;
1005 
1006 	ret = r535_gsp_rpc_get_gsp_static_info(gsp);
1007 	if (WARN_ON(ret))
1008 		return ret;
1009 
1010 	INIT_WORK(&gsp->msgq.work, r535_gsp_msgq_work);
1011 
1012 	ret = r535_gsp_intr_get_table(gsp);
1013 	if (WARN_ON(ret))
1014 		return ret;
1015 
1016 	ret = nvkm_gsp_intr_stall(gsp, gsp->subdev.type, gsp->subdev.inst);
1017 	if (WARN_ON(ret < 0))
1018 		return ret;
1019 
1020 	ret = nvkm_inth_add(&device->vfn->intr, ret, NVKM_INTR_PRIO_NORMAL, &gsp->subdev,
1021 			    r535_gsp_intr, &gsp->subdev.inth);
1022 	if (WARN_ON(ret))
1023 		return ret;
1024 
1025 	nvkm_inth_allow(&gsp->subdev.inth);
1026 	nvkm_wr32(device, 0x110004, 0x00000040);
1027 	return ret;
1028 }
1029 
1030 static int
1031 r535_gsp_rpc_unloading_guest_driver(struct nvkm_gsp *gsp, bool suspend)
1032 {
1033 	rpc_unloading_guest_driver_v1F_07 *rpc;
1034 
1035 	rpc = nvkm_gsp_rpc_get(gsp, NV_VGPU_MSG_FUNCTION_UNLOADING_GUEST_DRIVER, sizeof(*rpc));
1036 	if (IS_ERR(rpc))
1037 		return PTR_ERR(rpc);
1038 
1039 	if (suspend) {
1040 		rpc->bInPMTransition = 1;
1041 		rpc->bGc6Entering = 0;
1042 		rpc->newLevel = NV2080_CTRL_GPU_SET_POWER_STATE_GPU_LEVEL_3;
1043 	} else {
1044 		rpc->bInPMTransition = 0;
1045 		rpc->bGc6Entering = 0;
1046 		rpc->newLevel = NV2080_CTRL_GPU_SET_POWER_STATE_GPU_LEVEL_0;
1047 	}
1048 
1049 	return nvkm_gsp_rpc_wr(gsp, rpc, true);
1050 }
1051 
1052 /* dword only */
1053 struct nv_gsp_registry_entries {
1054 	const char *name;
1055 	u32 value;
1056 };
1057 
1058 static const struct nv_gsp_registry_entries r535_registry_entries[] = {
1059 	{ "RMSecBusResetEnable", 1 },
1060 	{ "RMForcePcieConfigSave", 1 },
1061 };
1062 #define NV_GSP_REG_NUM_ENTRIES ARRAY_SIZE(r535_registry_entries)
1063 
1064 static int
1065 r535_gsp_rpc_set_registry(struct nvkm_gsp *gsp)
1066 {
1067 	PACKED_REGISTRY_TABLE *rpc;
1068 	char *strings;
1069 	int str_offset;
1070 	int i;
1071 	size_t rpc_size = struct_size(rpc, entries, NV_GSP_REG_NUM_ENTRIES);
1072 
1073 	/* add strings + null terminator */
1074 	for (i = 0; i < NV_GSP_REG_NUM_ENTRIES; i++)
1075 		rpc_size += strlen(r535_registry_entries[i].name) + 1;
1076 
1077 	rpc = nvkm_gsp_rpc_get(gsp, NV_VGPU_MSG_FUNCTION_SET_REGISTRY, rpc_size);
1078 	if (IS_ERR(rpc))
1079 		return PTR_ERR(rpc);
1080 
1081 	rpc->size = sizeof(*rpc);
1082 	rpc->numEntries = NV_GSP_REG_NUM_ENTRIES;
1083 
1084 	str_offset = offsetof(typeof(*rpc), entries[NV_GSP_REG_NUM_ENTRIES]);
1085 	strings = (char *)&rpc->entries[NV_GSP_REG_NUM_ENTRIES];
1086 	for (i = 0; i < NV_GSP_REG_NUM_ENTRIES; i++) {
1087 		int name_len = strlen(r535_registry_entries[i].name) + 1;
1088 
1089 		rpc->entries[i].nameOffset = str_offset;
1090 		rpc->entries[i].type = 1;
1091 		rpc->entries[i].data = r535_registry_entries[i].value;
1092 		rpc->entries[i].length = 4;
1093 		memcpy(strings, r535_registry_entries[i].name, name_len);
1094 		strings += name_len;
1095 		str_offset += name_len;
1096 	}
1097 
1098 	return nvkm_gsp_rpc_wr(gsp, rpc, false);
1099 }
1100 
1101 #if defined(CONFIG_ACPI) && defined(CONFIG_X86)
1102 static void
1103 r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps)
1104 {
1105 	const guid_t NVOP_DSM_GUID =
1106 		GUID_INIT(0xA486D8F8, 0x0BDA, 0x471B,
1107 			  0xA7, 0x2B, 0x60, 0x42, 0xA6, 0xB5, 0xBE, 0xE0);
1108 	u64 NVOP_DSM_REV = 0x00000100;
1109 	union acpi_object argv4 = {
1110 		.buffer.type    = ACPI_TYPE_BUFFER,
1111 		.buffer.length  = 4,
1112 		.buffer.pointer = kmalloc(argv4.buffer.length, GFP_KERNEL),
1113 	}, *obj;
1114 
1115 	caps->status = 0xffff;
1116 
1117 	if (!acpi_check_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV, BIT_ULL(0x1a)))
1118 		return;
1119 
1120 	obj = acpi_evaluate_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV, 0x1a, &argv4);
1121 	if (!obj)
1122 		return;
1123 
1124 	if (WARN_ON(obj->type != ACPI_TYPE_BUFFER) ||
1125 	    WARN_ON(obj->buffer.length != 4))
1126 		return;
1127 
1128 	caps->status = 0;
1129 	caps->optimusCaps = *(u32 *)obj->buffer.pointer;
1130 
1131 	ACPI_FREE(obj);
1132 
1133 	kfree(argv4.buffer.pointer);
1134 }
1135 
1136 static void
1137 r535_gsp_acpi_jt(acpi_handle handle, JT_METHOD_DATA *jt)
1138 {
1139 	const guid_t JT_DSM_GUID =
1140 		GUID_INIT(0xCBECA351L, 0x067B, 0x4924,
1141 			  0x9C, 0xBD, 0xB4, 0x6B, 0x00, 0xB8, 0x6F, 0x34);
1142 	u64 JT_DSM_REV = 0x00000103;
1143 	u32 caps;
1144 	union acpi_object argv4 = {
1145 		.buffer.type    = ACPI_TYPE_BUFFER,
1146 		.buffer.length  = sizeof(caps),
1147 		.buffer.pointer = kmalloc(argv4.buffer.length, GFP_KERNEL),
1148 	}, *obj;
1149 
1150 	jt->status = 0xffff;
1151 
1152 	obj = acpi_evaluate_dsm(handle, &JT_DSM_GUID, JT_DSM_REV, 0x1, &argv4);
1153 	if (!obj)
1154 		return;
1155 
1156 	if (WARN_ON(obj->type != ACPI_TYPE_BUFFER) ||
1157 	    WARN_ON(obj->buffer.length != 4))
1158 		return;
1159 
1160 	jt->status = 0;
1161 	jt->jtCaps = *(u32 *)obj->buffer.pointer;
1162 	jt->jtRevId = (jt->jtCaps & 0xfff00000) >> 20;
1163 	jt->bSBIOSCaps = 0;
1164 
1165 	ACPI_FREE(obj);
1166 
1167 	kfree(argv4.buffer.pointer);
1168 }
1169 
1170 static void
1171 r535_gsp_acpi_mux_id(acpi_handle handle, u32 id, MUX_METHOD_DATA_ELEMENT *mode,
1172 						 MUX_METHOD_DATA_ELEMENT *part)
1173 {
1174 	union acpi_object mux_arg = { ACPI_TYPE_INTEGER };
1175 	struct acpi_object_list input = { 1, &mux_arg };
1176 	acpi_handle iter = NULL, handle_mux = NULL;
1177 	acpi_status status;
1178 	unsigned long long value;
1179 
1180 	mode->status = 0xffff;
1181 	part->status = 0xffff;
1182 
1183 	do {
1184 		status = acpi_get_next_object(ACPI_TYPE_DEVICE, handle, iter, &iter);
1185 		if (ACPI_FAILURE(status) || !iter)
1186 			return;
1187 
1188 		status = acpi_evaluate_integer(iter, "_ADR", NULL, &value);
1189 		if (ACPI_FAILURE(status) || value != id)
1190 			continue;
1191 
1192 		handle_mux = iter;
1193 	} while (!handle_mux);
1194 
1195 	if (!handle_mux)
1196 		return;
1197 
1198 	/* I -think- 0 means "acquire" according to nvidia's driver source */
1199 	input.pointer->integer.type = ACPI_TYPE_INTEGER;
1200 	input.pointer->integer.value = 0;
1201 
1202 	status = acpi_evaluate_integer(handle_mux, "MXDM", &input, &value);
1203 	if (ACPI_SUCCESS(status)) {
1204 		mode->acpiId = id;
1205 		mode->mode   = value;
1206 		mode->status = 0;
1207 	}
1208 
1209 	status = acpi_evaluate_integer(handle_mux, "MXDS", &input, &value);
1210 	if (ACPI_SUCCESS(status)) {
1211 		part->acpiId = id;
1212 		part->mode   = value;
1213 		part->status = 0;
1214 	}
1215 }
1216 
1217 static void
1218 r535_gsp_acpi_mux(acpi_handle handle, DOD_METHOD_DATA *dod, MUX_METHOD_DATA *mux)
1219 {
1220 	mux->tableLen = dod->acpiIdListLen / sizeof(dod->acpiIdList[0]);
1221 
1222 	for (int i = 0; i < mux->tableLen; i++) {
1223 		r535_gsp_acpi_mux_id(handle, dod->acpiIdList[i], &mux->acpiIdMuxModeTable[i],
1224 								 &mux->acpiIdMuxPartTable[i]);
1225 	}
1226 }
1227 
1228 static void
1229 r535_gsp_acpi_dod(acpi_handle handle, DOD_METHOD_DATA *dod)
1230 {
1231 	acpi_status status;
1232 	struct acpi_buffer output = { ACPI_ALLOCATE_BUFFER, NULL };
1233 	union acpi_object *_DOD;
1234 
1235 	dod->status = 0xffff;
1236 
1237 	status = acpi_evaluate_object(handle, "_DOD", NULL, &output);
1238 	if (ACPI_FAILURE(status))
1239 		return;
1240 
1241 	_DOD = output.pointer;
1242 
1243 	if (WARN_ON(_DOD->type != ACPI_TYPE_PACKAGE) ||
1244 	    WARN_ON(_DOD->package.count > ARRAY_SIZE(dod->acpiIdList)))
1245 		return;
1246 
1247 	for (int i = 0; i < _DOD->package.count; i++) {
1248 		if (WARN_ON(_DOD->package.elements[i].type != ACPI_TYPE_INTEGER))
1249 			return;
1250 
1251 		dod->acpiIdList[i] = _DOD->package.elements[i].integer.value;
1252 		dod->acpiIdListLen += sizeof(dod->acpiIdList[0]);
1253 	}
1254 
1255 	dod->status = 0;
1256 	kfree(output.pointer);
1257 }
1258 #endif
1259 
1260 static void
1261 r535_gsp_acpi_info(struct nvkm_gsp *gsp, ACPI_METHOD_DATA *acpi)
1262 {
1263 #if defined(CONFIG_ACPI) && defined(CONFIG_X86)
1264 	acpi_handle handle = ACPI_HANDLE(gsp->subdev.device->dev);
1265 
1266 	if (!handle)
1267 		return;
1268 
1269 	acpi->bValid = 1;
1270 
1271 	r535_gsp_acpi_dod(handle, &acpi->dodMethodData);
1272 	if (acpi->dodMethodData.status == 0)
1273 		r535_gsp_acpi_mux(handle, &acpi->dodMethodData, &acpi->muxMethodData);
1274 
1275 	r535_gsp_acpi_jt(handle, &acpi->jtMethodData);
1276 	r535_gsp_acpi_caps(handle, &acpi->capsMethodData);
1277 #endif
1278 }
1279 
1280 static int
1281 r535_gsp_rpc_set_system_info(struct nvkm_gsp *gsp)
1282 {
1283 	struct nvkm_device *device = gsp->subdev.device;
1284 	struct nvkm_device_pci *pdev = container_of(device, typeof(*pdev), device);
1285 	GspSystemInfo *info;
1286 
1287 	if (WARN_ON(device->type == NVKM_DEVICE_TEGRA))
1288 		return -ENOSYS;
1289 
1290 	info = nvkm_gsp_rpc_get(gsp, NV_VGPU_MSG_FUNCTION_GSP_SET_SYSTEM_INFO, sizeof(*info));
1291 	if (IS_ERR(info))
1292 		return PTR_ERR(info);
1293 
1294 	info->gpuPhysAddr = device->func->resource_addr(device, 0);
1295 	info->gpuPhysFbAddr = device->func->resource_addr(device, 1);
1296 	info->gpuPhysInstAddr = device->func->resource_addr(device, 3);
1297 	info->nvDomainBusDeviceFunc = pci_dev_id(pdev->pdev);
1298 	info->maxUserVa = TASK_SIZE;
1299 	info->pciConfigMirrorBase = 0x088000;
1300 	info->pciConfigMirrorSize = 0x001000;
1301 	r535_gsp_acpi_info(gsp, &info->acpiMethodData);
1302 
1303 	return nvkm_gsp_rpc_wr(gsp, info, false);
1304 }
1305 
1306 static int
1307 r535_gsp_msg_os_error_log(void *priv, u32 fn, void *repv, u32 repc)
1308 {
1309 	struct nvkm_gsp *gsp = priv;
1310 	struct nvkm_subdev *subdev = &gsp->subdev;
1311 	rpc_os_error_log_v17_00 *msg = repv;
1312 
1313 	if (WARN_ON(repc < sizeof(*msg)))
1314 		return -EINVAL;
1315 
1316 	nvkm_error(subdev, "Xid:%d %s\n", msg->exceptType, msg->errString);
1317 	return 0;
1318 }
1319 
1320 static int
1321 r535_gsp_msg_rc_triggered(void *priv, u32 fn, void *repv, u32 repc)
1322 {
1323 	rpc_rc_triggered_v17_02 *msg = repv;
1324 	struct nvkm_gsp *gsp = priv;
1325 	struct nvkm_subdev *subdev = &gsp->subdev;
1326 	struct nvkm_chan *chan;
1327 	unsigned long flags;
1328 
1329 	if (WARN_ON(repc < sizeof(*msg)))
1330 		return -EINVAL;
1331 
1332 	nvkm_error(subdev, "rc engn:%08x chid:%d type:%d scope:%d part:%d\n",
1333 		   msg->nv2080EngineType, msg->chid, msg->exceptType, msg->scope,
1334 		   msg->partitionAttributionId);
1335 
1336 	chan = nvkm_chan_get_chid(&subdev->device->fifo->engine, msg->chid / 8, &flags);
1337 	if (!chan) {
1338 		nvkm_error(subdev, "rc chid:%d not found!\n", msg->chid);
1339 		return 0;
1340 	}
1341 
1342 	nvkm_chan_error(chan, false);
1343 	nvkm_chan_put(&chan, flags);
1344 	return 0;
1345 }
1346 
1347 static int
1348 r535_gsp_msg_mmu_fault_queued(void *priv, u32 fn, void *repv, u32 repc)
1349 {
1350 	struct nvkm_gsp *gsp = priv;
1351 	struct nvkm_subdev *subdev = &gsp->subdev;
1352 
1353 	WARN_ON(repc != 0);
1354 
1355 	nvkm_error(subdev, "mmu fault queued\n");
1356 	return 0;
1357 }
1358 
1359 static int
1360 r535_gsp_msg_post_event(void *priv, u32 fn, void *repv, u32 repc)
1361 {
1362 	struct nvkm_gsp *gsp = priv;
1363 	struct nvkm_gsp_client *client;
1364 	struct nvkm_subdev *subdev = &gsp->subdev;
1365 	rpc_post_event_v17_00 *msg = repv;
1366 
1367 	if (WARN_ON(repc < sizeof(*msg)))
1368 		return -EINVAL;
1369 	if (WARN_ON(repc != sizeof(*msg) + msg->eventDataSize))
1370 		return -EINVAL;
1371 
1372 	nvkm_debug(subdev, "event: %08x %08x %d %08x %08x %d %d\n",
1373 		   msg->hClient, msg->hEvent, msg->notifyIndex, msg->data,
1374 		   msg->status, msg->eventDataSize, msg->bNotifyList);
1375 
1376 	mutex_lock(&gsp->client_id.mutex);
1377 	client = idr_find(&gsp->client_id.idr, msg->hClient & 0xffff);
1378 	if (client) {
1379 		struct nvkm_gsp_event *event;
1380 		bool handled = false;
1381 
1382 		list_for_each_entry(event, &client->events, head) {
1383 			if (event->object.handle == msg->hEvent) {
1384 				event->func(event, msg->eventData, msg->eventDataSize);
1385 				handled = true;
1386 			}
1387 		}
1388 
1389 		if (!handled) {
1390 			nvkm_error(subdev, "event: cid 0x%08x event 0x%08x not found!\n",
1391 				   msg->hClient, msg->hEvent);
1392 		}
1393 	} else {
1394 		nvkm_error(subdev, "event: cid 0x%08x not found!\n", msg->hClient);
1395 	}
1396 	mutex_unlock(&gsp->client_id.mutex);
1397 	return 0;
1398 }
1399 
1400 /**
1401  * r535_gsp_msg_run_cpu_sequencer() -- process I/O commands from the GSP
1402  *
1403  * The GSP sequencer is a list of I/O commands that the GSP can send to
1404  * the driver to perform for various purposes.  The most common usage is to
1405  * perform a special mid-initialization reset.
1406  */
1407 static int
1408 r535_gsp_msg_run_cpu_sequencer(void *priv, u32 fn, void *repv, u32 repc)
1409 {
1410 	struct nvkm_gsp *gsp = priv;
1411 	struct nvkm_subdev *subdev = &gsp->subdev;
1412 	struct nvkm_device *device = subdev->device;
1413 	rpc_run_cpu_sequencer_v17_00 *seq = repv;
1414 	int ptr = 0, ret;
1415 
1416 	nvkm_debug(subdev, "seq: %08x %08x\n", seq->bufferSizeDWord, seq->cmdIndex);
1417 
1418 	while (ptr < seq->cmdIndex) {
1419 		GSP_SEQUENCER_BUFFER_CMD *cmd = (void *)&seq->commandBuffer[ptr];
1420 
1421 		ptr += 1;
1422 		ptr += GSP_SEQUENCER_PAYLOAD_SIZE_DWORDS(cmd->opCode);
1423 
1424 		switch (cmd->opCode) {
1425 		case GSP_SEQ_BUF_OPCODE_REG_WRITE: {
1426 			u32 addr = cmd->payload.regWrite.addr;
1427 			u32 data = cmd->payload.regWrite.val;
1428 
1429 			nvkm_trace(subdev, "seq wr32 %06x %08x\n", addr, data);
1430 			nvkm_wr32(device, addr, data);
1431 		}
1432 			break;
1433 		case GSP_SEQ_BUF_OPCODE_REG_MODIFY: {
1434 			u32 addr = cmd->payload.regModify.addr;
1435 			u32 mask = cmd->payload.regModify.mask;
1436 			u32 data = cmd->payload.regModify.val;
1437 
1438 			nvkm_trace(subdev, "seq mask %06x %08x %08x\n", addr, mask, data);
1439 			nvkm_mask(device, addr, mask, data);
1440 		}
1441 			break;
1442 		case GSP_SEQ_BUF_OPCODE_REG_POLL: {
1443 			u32 addr = cmd->payload.regPoll.addr;
1444 			u32 mask = cmd->payload.regPoll.mask;
1445 			u32 data = cmd->payload.regPoll.val;
1446 			u32 usec = cmd->payload.regPoll.timeout ?: 4000000;
1447 			//u32 error = cmd->payload.regPoll.error;
1448 
1449 			nvkm_trace(subdev, "seq poll %06x %08x %08x %d\n", addr, mask, data, usec);
1450 			nvkm_rd32(device, addr);
1451 			nvkm_usec(device, usec,
1452 				if ((nvkm_rd32(device, addr) & mask) == data)
1453 					break;
1454 			);
1455 		}
1456 			break;
1457 		case GSP_SEQ_BUF_OPCODE_DELAY_US: {
1458 			u32 usec = cmd->payload.delayUs.val;
1459 
1460 			nvkm_trace(subdev, "seq usec %d\n", usec);
1461 			udelay(usec);
1462 		}
1463 			break;
1464 		case GSP_SEQ_BUF_OPCODE_REG_STORE: {
1465 			u32 addr = cmd->payload.regStore.addr;
1466 			u32 slot = cmd->payload.regStore.index;
1467 
1468 			seq->regSaveArea[slot] = nvkm_rd32(device, addr);
1469 			nvkm_trace(subdev, "seq save %08x -> %d: %08x\n", addr, slot,
1470 				   seq->regSaveArea[slot]);
1471 		}
1472 			break;
1473 		case GSP_SEQ_BUF_OPCODE_CORE_RESET:
1474 			nvkm_trace(subdev, "seq core reset\n");
1475 			nvkm_falcon_reset(&gsp->falcon);
1476 			nvkm_falcon_mask(&gsp->falcon, 0x624, 0x00000080, 0x00000080);
1477 			nvkm_falcon_wr32(&gsp->falcon, 0x10c, 0x00000000);
1478 			break;
1479 		case GSP_SEQ_BUF_OPCODE_CORE_START:
1480 			nvkm_trace(subdev, "seq core start\n");
1481 			if (nvkm_falcon_rd32(&gsp->falcon, 0x100) & 0x00000040)
1482 				nvkm_falcon_wr32(&gsp->falcon, 0x130, 0x00000002);
1483 			else
1484 				nvkm_falcon_wr32(&gsp->falcon, 0x100, 0x00000002);
1485 			break;
1486 		case GSP_SEQ_BUF_OPCODE_CORE_WAIT_FOR_HALT:
1487 			nvkm_trace(subdev, "seq core wait halt\n");
1488 			nvkm_msec(device, 2000,
1489 				if (nvkm_falcon_rd32(&gsp->falcon, 0x100) & 0x00000010)
1490 					break;
1491 			);
1492 			break;
1493 		case GSP_SEQ_BUF_OPCODE_CORE_RESUME: {
1494 			struct nvkm_sec2 *sec2 = device->sec2;
1495 			u32 mbox0;
1496 
1497 			nvkm_trace(subdev, "seq core resume\n");
1498 
1499 			ret = gsp->func->reset(gsp);
1500 			if (WARN_ON(ret))
1501 				return ret;
1502 
1503 			nvkm_falcon_wr32(&gsp->falcon, 0x040, lower_32_bits(gsp->libos.addr));
1504 			nvkm_falcon_wr32(&gsp->falcon, 0x044, upper_32_bits(gsp->libos.addr));
1505 
1506 			nvkm_falcon_start(&sec2->falcon);
1507 
1508 			if (nvkm_msec(device, 2000,
1509 				if (nvkm_rd32(device, 0x1180f8) & 0x04000000)
1510 					break;
1511 			) < 0)
1512 				return -ETIMEDOUT;
1513 
1514 			mbox0 = nvkm_falcon_rd32(&sec2->falcon, 0x040);
1515 			if (WARN_ON(mbox0)) {
1516 				nvkm_error(&gsp->subdev, "seq core resume sec2: 0x%x\n", mbox0);
1517 				return -EIO;
1518 			}
1519 
1520 			nvkm_falcon_wr32(&gsp->falcon, 0x080, gsp->boot.app_version);
1521 
1522 			if (WARN_ON(!nvkm_falcon_riscv_active(&gsp->falcon)))
1523 				return -EIO;
1524 		}
1525 			break;
1526 		default:
1527 			nvkm_error(subdev, "unknown sequencer opcode %08x\n", cmd->opCode);
1528 			return -EINVAL;
1529 		}
1530 	}
1531 
1532 	return 0;
1533 }
1534 
1535 static void
1536 nvkm_gsp_mem_dtor(struct nvkm_gsp *gsp, struct nvkm_gsp_mem *mem)
1537 {
1538 	if (mem->data) {
1539 		dma_free_coherent(gsp->subdev.device->dev, mem->size, mem->data, mem->addr);
1540 		mem->data = NULL;
1541 	}
1542 }
1543 
1544 static int
1545 nvkm_gsp_mem_ctor(struct nvkm_gsp *gsp, u32 size, struct nvkm_gsp_mem *mem)
1546 {
1547 	mem->size = size;
1548 	mem->data = dma_alloc_coherent(gsp->subdev.device->dev, size, &mem->addr, GFP_KERNEL);
1549 	if (WARN_ON(!mem->data))
1550 		return -ENOMEM;
1551 
1552 	return 0;
1553 }
1554 
1555 
1556 static int
1557 r535_gsp_booter_unload(struct nvkm_gsp *gsp, u32 mbox0, u32 mbox1)
1558 {
1559 	struct nvkm_subdev *subdev = &gsp->subdev;
1560 	struct nvkm_device *device = subdev->device;
1561 	u32 wpr2_hi;
1562 	int ret;
1563 
1564 	wpr2_hi = nvkm_rd32(device, 0x1fa828);
1565 	if (!wpr2_hi) {
1566 		nvkm_debug(subdev, "WPR2 not set - skipping booter unload\n");
1567 		return 0;
1568 	}
1569 
1570 	ret = nvkm_falcon_fw_boot(&gsp->booter.unload, &gsp->subdev, true, &mbox0, &mbox1, 0, 0);
1571 	if (WARN_ON(ret))
1572 		return ret;
1573 
1574 	wpr2_hi = nvkm_rd32(device, 0x1fa828);
1575 	if (WARN_ON(wpr2_hi))
1576 		return -EIO;
1577 
1578 	return 0;
1579 }
1580 
1581 static int
1582 r535_gsp_booter_load(struct nvkm_gsp *gsp, u32 mbox0, u32 mbox1)
1583 {
1584 	int ret;
1585 
1586 	ret = nvkm_falcon_fw_boot(&gsp->booter.load, &gsp->subdev, true, &mbox0, &mbox1, 0, 0);
1587 	if (ret)
1588 		return ret;
1589 
1590 	nvkm_falcon_wr32(&gsp->falcon, 0x080, gsp->boot.app_version);
1591 
1592 	if (WARN_ON(!nvkm_falcon_riscv_active(&gsp->falcon)))
1593 		return -EIO;
1594 
1595 	return 0;
1596 }
1597 
1598 static int
1599 r535_gsp_wpr_meta_init(struct nvkm_gsp *gsp)
1600 {
1601 	GspFwWprMeta *meta;
1602 	int ret;
1603 
1604 	ret = nvkm_gsp_mem_ctor(gsp, 0x1000, &gsp->wpr_meta);
1605 	if (ret)
1606 		return ret;
1607 
1608 	meta = gsp->wpr_meta.data;
1609 
1610 	meta->magic = GSP_FW_WPR_META_MAGIC;
1611 	meta->revision = GSP_FW_WPR_META_REVISION;
1612 
1613 	meta->sysmemAddrOfRadix3Elf = gsp->radix3.mem[0].addr;
1614 	meta->sizeOfRadix3Elf = gsp->fb.wpr2.elf.size;
1615 
1616 	meta->sysmemAddrOfBootloader = gsp->boot.fw.addr;
1617 	meta->sizeOfBootloader = gsp->boot.fw.size;
1618 	meta->bootloaderCodeOffset = gsp->boot.code_offset;
1619 	meta->bootloaderDataOffset = gsp->boot.data_offset;
1620 	meta->bootloaderManifestOffset = gsp->boot.manifest_offset;
1621 
1622 	meta->sysmemAddrOfSignature = gsp->sig.addr;
1623 	meta->sizeOfSignature = gsp->sig.size;
1624 
1625 	meta->gspFwRsvdStart = gsp->fb.heap.addr;
1626 	meta->nonWprHeapOffset = gsp->fb.heap.addr;
1627 	meta->nonWprHeapSize = gsp->fb.heap.size;
1628 	meta->gspFwWprStart = gsp->fb.wpr2.addr;
1629 	meta->gspFwHeapOffset = gsp->fb.wpr2.heap.addr;
1630 	meta->gspFwHeapSize = gsp->fb.wpr2.heap.size;
1631 	meta->gspFwOffset = gsp->fb.wpr2.elf.addr;
1632 	meta->bootBinOffset = gsp->fb.wpr2.boot.addr;
1633 	meta->frtsOffset = gsp->fb.wpr2.frts.addr;
1634 	meta->frtsSize = gsp->fb.wpr2.frts.size;
1635 	meta->gspFwWprEnd = ALIGN_DOWN(gsp->fb.bios.vga_workspace.addr, 0x20000);
1636 	meta->fbSize = gsp->fb.size;
1637 	meta->vgaWorkspaceOffset = gsp->fb.bios.vga_workspace.addr;
1638 	meta->vgaWorkspaceSize = gsp->fb.bios.vga_workspace.size;
1639 	meta->bootCount = 0;
1640 	meta->partitionRpcAddr = 0;
1641 	meta->partitionRpcRequestOffset = 0;
1642 	meta->partitionRpcReplyOffset = 0;
1643 	meta->verified = 0;
1644 	return 0;
1645 }
1646 
1647 static int
1648 r535_gsp_shared_init(struct nvkm_gsp *gsp)
1649 {
1650 	struct {
1651 		msgqTxHeader tx;
1652 		msgqRxHeader rx;
1653 	} *cmdq, *msgq;
1654 	int ret, i;
1655 
1656 	gsp->shm.cmdq.size = 0x40000;
1657 	gsp->shm.msgq.size = 0x40000;
1658 
1659 	gsp->shm.ptes.nr  = (gsp->shm.cmdq.size + gsp->shm.msgq.size) >> GSP_PAGE_SHIFT;
1660 	gsp->shm.ptes.nr += DIV_ROUND_UP(gsp->shm.ptes.nr * sizeof(u64), GSP_PAGE_SIZE);
1661 	gsp->shm.ptes.size = ALIGN(gsp->shm.ptes.nr * sizeof(u64), GSP_PAGE_SIZE);
1662 
1663 	ret = nvkm_gsp_mem_ctor(gsp, gsp->shm.ptes.size +
1664 				     gsp->shm.cmdq.size +
1665 				     gsp->shm.msgq.size,
1666 				&gsp->shm.mem);
1667 	if (ret)
1668 		return ret;
1669 
1670 	gsp->shm.ptes.ptr = gsp->shm.mem.data;
1671 	gsp->shm.cmdq.ptr = (u8 *)gsp->shm.ptes.ptr + gsp->shm.ptes.size;
1672 	gsp->shm.msgq.ptr = (u8 *)gsp->shm.cmdq.ptr + gsp->shm.cmdq.size;
1673 
1674 	for (i = 0; i < gsp->shm.ptes.nr; i++)
1675 		gsp->shm.ptes.ptr[i] = gsp->shm.mem.addr + (i << GSP_PAGE_SHIFT);
1676 
1677 	cmdq = gsp->shm.cmdq.ptr;
1678 	cmdq->tx.version = 0;
1679 	cmdq->tx.size = gsp->shm.cmdq.size;
1680 	cmdq->tx.entryOff = GSP_PAGE_SIZE;
1681 	cmdq->tx.msgSize = GSP_PAGE_SIZE;
1682 	cmdq->tx.msgCount = (cmdq->tx.size - cmdq->tx.entryOff) / cmdq->tx.msgSize;
1683 	cmdq->tx.writePtr = 0;
1684 	cmdq->tx.flags = 1;
1685 	cmdq->tx.rxHdrOff = offsetof(typeof(*cmdq), rx.readPtr);
1686 
1687 	msgq = gsp->shm.msgq.ptr;
1688 
1689 	gsp->cmdq.cnt = cmdq->tx.msgCount;
1690 	gsp->cmdq.wptr = &cmdq->tx.writePtr;
1691 	gsp->cmdq.rptr = &msgq->rx.readPtr;
1692 	gsp->msgq.cnt = cmdq->tx.msgCount;
1693 	gsp->msgq.wptr = &msgq->tx.writePtr;
1694 	gsp->msgq.rptr = &cmdq->rx.readPtr;
1695 	return 0;
1696 }
1697 
1698 static int
1699 r535_gsp_rmargs_init(struct nvkm_gsp *gsp, bool resume)
1700 {
1701 	GSP_ARGUMENTS_CACHED *args;
1702 	int ret;
1703 
1704 	if (!resume) {
1705 		ret = r535_gsp_shared_init(gsp);
1706 		if (ret)
1707 			return ret;
1708 
1709 		ret = nvkm_gsp_mem_ctor(gsp, 0x1000, &gsp->rmargs);
1710 		if (ret)
1711 			return ret;
1712 	}
1713 
1714 	args = gsp->rmargs.data;
1715 	args->messageQueueInitArguments.sharedMemPhysAddr = gsp->shm.mem.addr;
1716 	args->messageQueueInitArguments.pageTableEntryCount = gsp->shm.ptes.nr;
1717 	args->messageQueueInitArguments.cmdQueueOffset =
1718 		(u8 *)gsp->shm.cmdq.ptr - (u8 *)gsp->shm.mem.data;
1719 	args->messageQueueInitArguments.statQueueOffset =
1720 		(u8 *)gsp->shm.msgq.ptr - (u8 *)gsp->shm.mem.data;
1721 
1722 	if (!resume) {
1723 		args->srInitArguments.oldLevel = 0;
1724 		args->srInitArguments.flags = 0;
1725 		args->srInitArguments.bInPMTransition = 0;
1726 	} else {
1727 		args->srInitArguments.oldLevel = NV2080_CTRL_GPU_SET_POWER_STATE_GPU_LEVEL_3;
1728 		args->srInitArguments.flags = 0;
1729 		args->srInitArguments.bInPMTransition = 1;
1730 	}
1731 
1732 	return 0;
1733 }
1734 
1735 static inline u64
1736 r535_gsp_libos_id8(const char *name)
1737 {
1738 	u64 id = 0;
1739 
1740 	for (int i = 0; i < sizeof(id) && *name; i++, name++)
1741 		id = (id << 8) | *name;
1742 
1743 	return id;
1744 }
1745 
1746 /**
1747  * create_pte_array() - creates a PTE array of a physically contiguous buffer
1748  * @ptes: pointer to the array
1749  * @addr: base address of physically contiguous buffer (GSP_PAGE_SIZE aligned)
1750  * @size: size of the buffer
1751  *
1752  * GSP-RM sometimes expects physically-contiguous buffers to have an array of
1753  * "PTEs" for each page in that buffer.  Although in theory that allows for
1754  * the buffer to be physically discontiguous, GSP-RM does not currently
1755  * support that.
1756  *
1757  * In this case, the PTEs are DMA addresses of each page of the buffer.  Since
1758  * the buffer is physically contiguous, calculating all the PTEs is simple
1759  * math.
1760  *
1761  * See memdescGetPhysAddrsForGpu()
1762  */
1763 static void create_pte_array(u64 *ptes, dma_addr_t addr, size_t size)
1764 {
1765 	unsigned int num_pages = DIV_ROUND_UP_ULL(size, GSP_PAGE_SIZE);
1766 	unsigned int i;
1767 
1768 	for (i = 0; i < num_pages; i++)
1769 		ptes[i] = (u64)addr + (i << GSP_PAGE_SHIFT);
1770 }
1771 
1772 /**
1773  * r535_gsp_libos_init() -- create the libos arguments structure
1774  *
1775  * The logging buffers are byte queues that contain encoded printf-like
1776  * messages from GSP-RM.  They need to be decoded by a special application
1777  * that can parse the buffers.
1778  *
1779  * The 'loginit' buffer contains logs from early GSP-RM init and
1780  * exception dumps.  The 'logrm' buffer contains the subsequent logs. Both are
1781  * written to directly by GSP-RM and can be any multiple of GSP_PAGE_SIZE.
1782  *
1783  * The physical address map for the log buffer is stored in the buffer
1784  * itself, starting with offset 1. Offset 0 contains the "put" pointer.
1785  *
1786  * The GSP only understands 4K pages (GSP_PAGE_SIZE), so even if the kernel is
1787  * configured for a larger page size (e.g. 64K pages), we need to give
1788  * the GSP an array of 4K pages. Fortunately, since the buffer is
1789  * physically contiguous, it's simple math to calculate the addresses.
1790  *
1791  * The buffers must be a multiple of GSP_PAGE_SIZE.  GSP-RM also currently
1792  * ignores the @kind field for LOGINIT, LOGINTR, and LOGRM, but expects the
1793  * buffers to be physically contiguous anyway.
1794  *
1795  * The memory allocated for the arguments must remain until the GSP sends the
1796  * init_done RPC.
1797  *
1798  * See _kgspInitLibosLoggingStructures (allocates memory for buffers)
1799  * See kgspSetupLibosInitArgs_IMPL (creates pLibosInitArgs[] array)
1800  */
1801 static int
1802 r535_gsp_libos_init(struct nvkm_gsp *gsp)
1803 {
1804 	LibosMemoryRegionInitArgument *args;
1805 	int ret;
1806 
1807 	ret = nvkm_gsp_mem_ctor(gsp, 0x1000, &gsp->libos);
1808 	if (ret)
1809 		return ret;
1810 
1811 	args = gsp->libos.data;
1812 
1813 	ret = nvkm_gsp_mem_ctor(gsp, 0x10000, &gsp->loginit);
1814 	if (ret)
1815 		return ret;
1816 
1817 	args[0].id8  = r535_gsp_libos_id8("LOGINIT");
1818 	args[0].pa   = gsp->loginit.addr;
1819 	args[0].size = gsp->loginit.size;
1820 	args[0].kind = LIBOS_MEMORY_REGION_CONTIGUOUS;
1821 	args[0].loc  = LIBOS_MEMORY_REGION_LOC_SYSMEM;
1822 	create_pte_array(gsp->loginit.data + sizeof(u64), gsp->loginit.addr, gsp->loginit.size);
1823 
1824 	ret = nvkm_gsp_mem_ctor(gsp, 0x10000, &gsp->logintr);
1825 	if (ret)
1826 		return ret;
1827 
1828 	args[1].id8  = r535_gsp_libos_id8("LOGINTR");
1829 	args[1].pa   = gsp->logintr.addr;
1830 	args[1].size = gsp->logintr.size;
1831 	args[1].kind = LIBOS_MEMORY_REGION_CONTIGUOUS;
1832 	args[1].loc  = LIBOS_MEMORY_REGION_LOC_SYSMEM;
1833 	create_pte_array(gsp->logintr.data + sizeof(u64), gsp->logintr.addr, gsp->logintr.size);
1834 
1835 	ret = nvkm_gsp_mem_ctor(gsp, 0x10000, &gsp->logrm);
1836 	if (ret)
1837 		return ret;
1838 
1839 	args[2].id8  = r535_gsp_libos_id8("LOGRM");
1840 	args[2].pa   = gsp->logrm.addr;
1841 	args[2].size = gsp->logrm.size;
1842 	args[2].kind = LIBOS_MEMORY_REGION_CONTIGUOUS;
1843 	args[2].loc  = LIBOS_MEMORY_REGION_LOC_SYSMEM;
1844 	create_pte_array(gsp->logrm.data + sizeof(u64), gsp->logrm.addr, gsp->logrm.size);
1845 
1846 	ret = r535_gsp_rmargs_init(gsp, false);
1847 	if (ret)
1848 		return ret;
1849 
1850 	args[3].id8  = r535_gsp_libos_id8("RMARGS");
1851 	args[3].pa   = gsp->rmargs.addr;
1852 	args[3].size = gsp->rmargs.size;
1853 	args[3].kind = LIBOS_MEMORY_REGION_CONTIGUOUS;
1854 	args[3].loc  = LIBOS_MEMORY_REGION_LOC_SYSMEM;
1855 	return 0;
1856 }
1857 
1858 void
1859 nvkm_gsp_sg_free(struct nvkm_device *device, struct sg_table *sgt)
1860 {
1861 	struct scatterlist *sgl;
1862 	int i;
1863 
1864 	dma_unmap_sgtable(device->dev, sgt, DMA_BIDIRECTIONAL, 0);
1865 
1866 	for_each_sgtable_sg(sgt, sgl, i) {
1867 		struct page *page = sg_page(sgl);
1868 
1869 		__free_page(page);
1870 	}
1871 
1872 	sg_free_table(sgt);
1873 }
1874 
1875 int
1876 nvkm_gsp_sg(struct nvkm_device *device, u64 size, struct sg_table *sgt)
1877 {
1878 	const u64 pages = DIV_ROUND_UP(size, PAGE_SIZE);
1879 	struct scatterlist *sgl;
1880 	int ret, i;
1881 
1882 	ret = sg_alloc_table(sgt, pages, GFP_KERNEL);
1883 	if (ret)
1884 		return ret;
1885 
1886 	for_each_sgtable_sg(sgt, sgl, i) {
1887 		struct page *page = alloc_page(GFP_KERNEL);
1888 
1889 		if (!page) {
1890 			nvkm_gsp_sg_free(device, sgt);
1891 			return -ENOMEM;
1892 		}
1893 
1894 		sg_set_page(sgl, page, PAGE_SIZE, 0);
1895 	}
1896 
1897 	ret = dma_map_sgtable(device->dev, sgt, DMA_BIDIRECTIONAL, 0);
1898 	if (ret)
1899 		nvkm_gsp_sg_free(device, sgt);
1900 
1901 	return ret;
1902 }
1903 
1904 static void
1905 nvkm_gsp_radix3_dtor(struct nvkm_gsp *gsp, struct nvkm_gsp_radix3 *rx3)
1906 {
1907 	for (int i = ARRAY_SIZE(rx3->mem) - 1; i >= 0; i--)
1908 		nvkm_gsp_mem_dtor(gsp, &rx3->mem[i]);
1909 }
1910 
1911 /**
1912  * nvkm_gsp_radix3_sg - build a radix3 table from a S/G list
1913  *
1914  * The GSP uses a three-level page table, called radix3, to map the firmware.
1915  * Each 64-bit "pointer" in the table is either the bus address of an entry in
1916  * the next table (for levels 0 and 1) or the bus address of the next page in
1917  * the GSP firmware image itself.
1918  *
1919  * Level 0 contains a single entry in one page that points to the first page
1920  * of level 1.
1921  *
1922  * Level 1, since it's also only one page in size, contains up to 512 entries,
1923  * one for each page in Level 2.
1924  *
1925  * Level 2 can be up to 512 pages in size, and each of those entries points to
1926  * the next page of the firmware image.  Since there can be up to 512*512
1927  * pages, that limits the size of the firmware to 512*512*GSP_PAGE_SIZE = 1GB.
1928  *
1929  * Internally, the GSP has its window into system memory, but the base
1930  * physical address of the aperture is not 0.  In fact, it varies depending on
1931  * the GPU architecture.  Since the GPU is a PCI device, this window is
1932  * accessed via DMA and is therefore bound by IOMMU translation.  The end
1933  * result is that GSP-RM must translate the bus addresses in the table to GSP
1934  * physical addresses.  All this should happen transparently.
1935  *
1936  * Returns 0 on success, or negative error code
1937  *
1938  * See kgspCreateRadix3_IMPL
1939  */
1940 static int
1941 nvkm_gsp_radix3_sg(struct nvkm_device *device, struct sg_table *sgt, u64 size,
1942 		   struct nvkm_gsp_radix3 *rx3)
1943 {
1944 	u64 addr;
1945 
1946 	for (int i = ARRAY_SIZE(rx3->mem) - 1; i >= 0; i--) {
1947 		u64 *ptes;
1948 		int idx;
1949 
1950 		rx3->mem[i].size = ALIGN((size / GSP_PAGE_SIZE) * sizeof(u64), GSP_PAGE_SIZE);
1951 		rx3->mem[i].data = dma_alloc_coherent(device->dev, rx3->mem[i].size,
1952 						      &rx3->mem[i].addr, GFP_KERNEL);
1953 		if (WARN_ON(!rx3->mem[i].data))
1954 			return -ENOMEM;
1955 
1956 		ptes = rx3->mem[i].data;
1957 		if (i == 2) {
1958 			struct scatterlist *sgl;
1959 
1960 			for_each_sgtable_dma_sg(sgt, sgl, idx) {
1961 				for (int j = 0; j < sg_dma_len(sgl) / GSP_PAGE_SIZE; j++)
1962 					*ptes++ = sg_dma_address(sgl) + (GSP_PAGE_SIZE * j);
1963 			}
1964 		} else {
1965 			for (int j = 0; j < size / GSP_PAGE_SIZE; j++)
1966 				*ptes++ = addr + GSP_PAGE_SIZE * j;
1967 		}
1968 
1969 		size = rx3->mem[i].size;
1970 		addr = rx3->mem[i].addr;
1971 	}
1972 
1973 	return 0;
1974 }
1975 
1976 int
1977 r535_gsp_fini(struct nvkm_gsp *gsp, bool suspend)
1978 {
1979 	u32 mbox0 = 0xff, mbox1 = 0xff;
1980 	int ret;
1981 
1982 	if (!gsp->running)
1983 		return 0;
1984 
1985 	if (suspend) {
1986 		GspFwWprMeta *meta = gsp->wpr_meta.data;
1987 		u64 len = meta->gspFwWprEnd - meta->gspFwWprStart;
1988 		GspFwSRMeta *sr;
1989 
1990 		ret = nvkm_gsp_sg(gsp->subdev.device, len, &gsp->sr.sgt);
1991 		if (ret)
1992 			return ret;
1993 
1994 		ret = nvkm_gsp_radix3_sg(gsp->subdev.device, &gsp->sr.sgt, len, &gsp->sr.radix3);
1995 		if (ret)
1996 			return ret;
1997 
1998 		ret = nvkm_gsp_mem_ctor(gsp, sizeof(*sr), &gsp->sr.meta);
1999 		if (ret)
2000 			return ret;
2001 
2002 		sr = gsp->sr.meta.data;
2003 		sr->magic = GSP_FW_SR_META_MAGIC;
2004 		sr->revision = GSP_FW_SR_META_REVISION;
2005 		sr->sysmemAddrOfSuspendResumeData = gsp->sr.radix3.mem[0].addr;
2006 		sr->sizeOfSuspendResumeData = len;
2007 
2008 		mbox0 = lower_32_bits(gsp->sr.meta.addr);
2009 		mbox1 = upper_32_bits(gsp->sr.meta.addr);
2010 	}
2011 
2012 	ret = r535_gsp_rpc_unloading_guest_driver(gsp, suspend);
2013 	if (WARN_ON(ret))
2014 		return ret;
2015 
2016 	nvkm_msec(gsp->subdev.device, 2000,
2017 		if (nvkm_falcon_rd32(&gsp->falcon, 0x040) & 0x80000000)
2018 			break;
2019 	);
2020 
2021 	nvkm_falcon_reset(&gsp->falcon);
2022 
2023 	ret = nvkm_gsp_fwsec_sb(gsp);
2024 	WARN_ON(ret);
2025 
2026 	ret = r535_gsp_booter_unload(gsp, mbox0, mbox1);
2027 	WARN_ON(ret);
2028 
2029 	gsp->running = false;
2030 	return 0;
2031 }
2032 
2033 int
2034 r535_gsp_init(struct nvkm_gsp *gsp)
2035 {
2036 	u32 mbox0, mbox1;
2037 	int ret;
2038 
2039 	if (!gsp->sr.meta.data) {
2040 		mbox0 = lower_32_bits(gsp->wpr_meta.addr);
2041 		mbox1 = upper_32_bits(gsp->wpr_meta.addr);
2042 	} else {
2043 		r535_gsp_rmargs_init(gsp, true);
2044 
2045 		mbox0 = lower_32_bits(gsp->sr.meta.addr);
2046 		mbox1 = upper_32_bits(gsp->sr.meta.addr);
2047 	}
2048 
2049 	/* Execute booter to handle (eventually...) booting GSP-RM. */
2050 	ret = r535_gsp_booter_load(gsp, mbox0, mbox1);
2051 	if (WARN_ON(ret))
2052 		goto done;
2053 
2054 	ret = r535_gsp_rpc_poll(gsp, NV_VGPU_MSG_EVENT_GSP_INIT_DONE);
2055 	if (ret)
2056 		goto done;
2057 
2058 	gsp->running = true;
2059 
2060 done:
2061 	if (gsp->sr.meta.data) {
2062 		nvkm_gsp_mem_dtor(gsp, &gsp->sr.meta);
2063 		nvkm_gsp_radix3_dtor(gsp, &gsp->sr.radix3);
2064 		nvkm_gsp_sg_free(gsp->subdev.device, &gsp->sr.sgt);
2065 		return ret;
2066 	}
2067 
2068 	if (ret == 0)
2069 		ret = r535_gsp_postinit(gsp);
2070 
2071 	return ret;
2072 }
2073 
2074 static int
2075 r535_gsp_rm_boot_ctor(struct nvkm_gsp *gsp)
2076 {
2077 	const struct firmware *fw = gsp->fws.bl;
2078 	const struct nvfw_bin_hdr *hdr;
2079 	RM_RISCV_UCODE_DESC *desc;
2080 	int ret;
2081 
2082 	hdr = nvfw_bin_hdr(&gsp->subdev, fw->data);
2083 	desc = (void *)fw->data + hdr->header_offset;
2084 
2085 	ret = nvkm_gsp_mem_ctor(gsp, hdr->data_size, &gsp->boot.fw);
2086 	if (ret)
2087 		return ret;
2088 
2089 	memcpy(gsp->boot.fw.data, fw->data + hdr->data_offset, hdr->data_size);
2090 
2091 	gsp->boot.code_offset = desc->monitorCodeOffset;
2092 	gsp->boot.data_offset = desc->monitorDataOffset;
2093 	gsp->boot.manifest_offset = desc->manifestOffset;
2094 	gsp->boot.app_version = desc->appVersion;
2095 	return 0;
2096 }
2097 
2098 static const struct nvkm_firmware_func
2099 r535_gsp_fw = {
2100 	.type = NVKM_FIRMWARE_IMG_SGT,
2101 };
2102 
2103 static int
2104 r535_gsp_elf_section(struct nvkm_gsp *gsp, const char *name, const u8 **pdata, u64 *psize)
2105 {
2106 	const u8 *img = gsp->fws.rm->data;
2107 	const struct elf64_hdr *ehdr = (const struct elf64_hdr *)img;
2108 	const struct elf64_shdr *shdr = (const struct elf64_shdr *)&img[ehdr->e_shoff];
2109 	const char *names = &img[shdr[ehdr->e_shstrndx].sh_offset];
2110 
2111 	for (int i = 0; i < ehdr->e_shnum; i++, shdr++) {
2112 		if (!strcmp(&names[shdr->sh_name], name)) {
2113 			*pdata = &img[shdr->sh_offset];
2114 			*psize = shdr->sh_size;
2115 			return 0;
2116 		}
2117 	}
2118 
2119 	nvkm_error(&gsp->subdev, "section '%s' not found\n", name);
2120 	return -ENOENT;
2121 }
2122 
2123 static void
2124 r535_gsp_dtor_fws(struct nvkm_gsp *gsp)
2125 {
2126 	nvkm_firmware_put(gsp->fws.bl);
2127 	gsp->fws.bl = NULL;
2128 	nvkm_firmware_put(gsp->fws.booter.unload);
2129 	gsp->fws.booter.unload = NULL;
2130 	nvkm_firmware_put(gsp->fws.booter.load);
2131 	gsp->fws.booter.load = NULL;
2132 	nvkm_firmware_put(gsp->fws.rm);
2133 	gsp->fws.rm = NULL;
2134 }
2135 
2136 void
2137 r535_gsp_dtor(struct nvkm_gsp *gsp)
2138 {
2139 	idr_destroy(&gsp->client_id.idr);
2140 	mutex_destroy(&gsp->client_id.mutex);
2141 
2142 	nvkm_gsp_radix3_dtor(gsp, &gsp->radix3);
2143 	nvkm_gsp_mem_dtor(gsp, &gsp->sig);
2144 	nvkm_firmware_dtor(&gsp->fw);
2145 
2146 	nvkm_falcon_fw_dtor(&gsp->booter.unload);
2147 	nvkm_falcon_fw_dtor(&gsp->booter.load);
2148 
2149 	mutex_destroy(&gsp->msgq.mutex);
2150 	mutex_destroy(&gsp->cmdq.mutex);
2151 
2152 	r535_gsp_dtor_fws(gsp);
2153 }
2154 
2155 int
2156 r535_gsp_oneinit(struct nvkm_gsp *gsp)
2157 {
2158 	struct nvkm_device *device = gsp->subdev.device;
2159 	const u8 *data;
2160 	u64 size;
2161 	int ret;
2162 
2163 	mutex_init(&gsp->cmdq.mutex);
2164 	mutex_init(&gsp->msgq.mutex);
2165 
2166 	ret = gsp->func->booter.ctor(gsp, "booter-load", gsp->fws.booter.load,
2167 				     &device->sec2->falcon, &gsp->booter.load);
2168 	if (ret)
2169 		return ret;
2170 
2171 	ret = gsp->func->booter.ctor(gsp, "booter-unload", gsp->fws.booter.unload,
2172 				     &device->sec2->falcon, &gsp->booter.unload);
2173 	if (ret)
2174 		return ret;
2175 
2176 	/* Load GSP firmware from ELF image into DMA-accessible memory. */
2177 	ret = r535_gsp_elf_section(gsp, ".fwimage", &data, &size);
2178 	if (ret)
2179 		return ret;
2180 
2181 	ret = nvkm_firmware_ctor(&r535_gsp_fw, "gsp-rm", device, data, size, &gsp->fw);
2182 	if (ret)
2183 		return ret;
2184 
2185 	/* Load relevant signature from ELF image. */
2186 	ret = r535_gsp_elf_section(gsp, gsp->func->sig_section, &data, &size);
2187 	if (ret)
2188 		return ret;
2189 
2190 	ret = nvkm_gsp_mem_ctor(gsp, ALIGN(size, 256), &gsp->sig);
2191 	if (ret)
2192 		return ret;
2193 
2194 	memcpy(gsp->sig.data, data, size);
2195 
2196 	/* Build radix3 page table for ELF image. */
2197 	ret = nvkm_gsp_radix3_sg(device, &gsp->fw.mem.sgt, gsp->fw.len, &gsp->radix3);
2198 	if (ret)
2199 		return ret;
2200 
2201 	r535_gsp_msg_ntfy_add(gsp, NV_VGPU_MSG_EVENT_GSP_RUN_CPU_SEQUENCER,
2202 			      r535_gsp_msg_run_cpu_sequencer, gsp);
2203 	r535_gsp_msg_ntfy_add(gsp, NV_VGPU_MSG_EVENT_POST_EVENT, r535_gsp_msg_post_event, gsp);
2204 	r535_gsp_msg_ntfy_add(gsp, NV_VGPU_MSG_EVENT_RC_TRIGGERED,
2205 			      r535_gsp_msg_rc_triggered, gsp);
2206 	r535_gsp_msg_ntfy_add(gsp, NV_VGPU_MSG_EVENT_MMU_FAULT_QUEUED,
2207 			      r535_gsp_msg_mmu_fault_queued, gsp);
2208 	r535_gsp_msg_ntfy_add(gsp, NV_VGPU_MSG_EVENT_OS_ERROR_LOG, r535_gsp_msg_os_error_log, gsp);
2209 	r535_gsp_msg_ntfy_add(gsp, NV_VGPU_MSG_EVENT_PERF_BRIDGELESS_INFO_UPDATE, NULL, NULL);
2210 	r535_gsp_msg_ntfy_add(gsp, NV_VGPU_MSG_EVENT_UCODE_LIBOS_PRINT, NULL, NULL);
2211 	r535_gsp_msg_ntfy_add(gsp, NV_VGPU_MSG_EVENT_GSP_SEND_USER_SHARED_DATA, NULL, NULL);
2212 	ret = r535_gsp_rm_boot_ctor(gsp);
2213 	if (ret)
2214 		return ret;
2215 
2216 	/* Release FW images - we've copied them to DMA buffers now. */
2217 	r535_gsp_dtor_fws(gsp);
2218 
2219 	/* Calculate FB layout. */
2220 	gsp->fb.wpr2.frts.size = 0x100000;
2221 	gsp->fb.wpr2.frts.addr = ALIGN_DOWN(gsp->fb.bios.addr, 0x20000) - gsp->fb.wpr2.frts.size;
2222 
2223 	gsp->fb.wpr2.boot.size = gsp->boot.fw.size;
2224 	gsp->fb.wpr2.boot.addr = ALIGN_DOWN(gsp->fb.wpr2.frts.addr - gsp->fb.wpr2.boot.size, 0x1000);
2225 
2226 	gsp->fb.wpr2.elf.size = gsp->fw.len;
2227 	gsp->fb.wpr2.elf.addr = ALIGN_DOWN(gsp->fb.wpr2.boot.addr - gsp->fb.wpr2.elf.size, 0x10000);
2228 
2229 	{
2230 		u32 fb_size_gb = DIV_ROUND_UP_ULL(gsp->fb.size, 1 << 30);
2231 
2232 		gsp->fb.wpr2.heap.size =
2233 			gsp->func->wpr_heap.os_carveout_size +
2234 			gsp->func->wpr_heap.base_size +
2235 			ALIGN(GSP_FW_HEAP_PARAM_SIZE_PER_GB_FB * fb_size_gb, 1 << 20) +
2236 			ALIGN(GSP_FW_HEAP_PARAM_CLIENT_ALLOC_SIZE, 1 << 20);
2237 
2238 		gsp->fb.wpr2.heap.size = max(gsp->fb.wpr2.heap.size, gsp->func->wpr_heap.min_size);
2239 	}
2240 
2241 	gsp->fb.wpr2.heap.addr = ALIGN_DOWN(gsp->fb.wpr2.elf.addr - gsp->fb.wpr2.heap.size, 0x100000);
2242 	gsp->fb.wpr2.heap.size = ALIGN_DOWN(gsp->fb.wpr2.elf.addr - gsp->fb.wpr2.heap.addr, 0x100000);
2243 
2244 	gsp->fb.wpr2.addr = ALIGN_DOWN(gsp->fb.wpr2.heap.addr - sizeof(GspFwWprMeta), 0x100000);
2245 	gsp->fb.wpr2.size = gsp->fb.wpr2.frts.addr + gsp->fb.wpr2.frts.size - gsp->fb.wpr2.addr;
2246 
2247 	gsp->fb.heap.size = 0x100000;
2248 	gsp->fb.heap.addr = gsp->fb.wpr2.addr - gsp->fb.heap.size;
2249 
2250 	ret = nvkm_gsp_fwsec_frts(gsp);
2251 	if (WARN_ON(ret))
2252 		return ret;
2253 
2254 	ret = r535_gsp_libos_init(gsp);
2255 	if (WARN_ON(ret))
2256 		return ret;
2257 
2258 	ret = r535_gsp_wpr_meta_init(gsp);
2259 	if (WARN_ON(ret))
2260 		return ret;
2261 
2262 	ret = r535_gsp_rpc_set_system_info(gsp);
2263 	if (WARN_ON(ret))
2264 		return ret;
2265 
2266 	ret = r535_gsp_rpc_set_registry(gsp);
2267 	if (WARN_ON(ret))
2268 		return ret;
2269 
2270 	/* Reset GSP into RISC-V mode. */
2271 	ret = gsp->func->reset(gsp);
2272 	if (WARN_ON(ret))
2273 		return ret;
2274 
2275 	nvkm_falcon_wr32(&gsp->falcon, 0x040, lower_32_bits(gsp->libos.addr));
2276 	nvkm_falcon_wr32(&gsp->falcon, 0x044, upper_32_bits(gsp->libos.addr));
2277 
2278 	mutex_init(&gsp->client_id.mutex);
2279 	idr_init(&gsp->client_id.idr);
2280 	return 0;
2281 }
2282 
2283 static int
2284 r535_gsp_load_fw(struct nvkm_gsp *gsp, const char *name, const char *ver,
2285 		 const struct firmware **pfw)
2286 {
2287 	char fwname[64];
2288 
2289 	snprintf(fwname, sizeof(fwname), "gsp/%s-%s", name, ver);
2290 	return nvkm_firmware_get(&gsp->subdev, fwname, 0, pfw);
2291 }
2292 
2293 int
2294 r535_gsp_load(struct nvkm_gsp *gsp, int ver, const struct nvkm_gsp_fwif *fwif)
2295 {
2296 	struct nvkm_subdev *subdev = &gsp->subdev;
2297 	int ret;
2298 
2299 	if (!nvkm_boolopt(subdev->device->cfgopt, "NvGspRm", fwif->enable))
2300 		return -EINVAL;
2301 
2302 	if ((ret = r535_gsp_load_fw(gsp, "gsp", fwif->ver, &gsp->fws.rm)) ||
2303 	    (ret = r535_gsp_load_fw(gsp, "booter_load", fwif->ver, &gsp->fws.booter.load)) ||
2304 	    (ret = r535_gsp_load_fw(gsp, "booter_unload", fwif->ver, &gsp->fws.booter.unload)) ||
2305 	    (ret = r535_gsp_load_fw(gsp, "bootloader", fwif->ver, &gsp->fws.bl))) {
2306 		r535_gsp_dtor_fws(gsp);
2307 		return ret;
2308 	}
2309 
2310 	return 0;
2311 }
2312 
2313 #define NVKM_GSP_FIRMWARE(chip)                                  \
2314 MODULE_FIRMWARE("nvidia/"#chip"/gsp/booter_load-535.113.01.bin");   \
2315 MODULE_FIRMWARE("nvidia/"#chip"/gsp/booter_unload-535.113.01.bin"); \
2316 MODULE_FIRMWARE("nvidia/"#chip"/gsp/bootloader-535.113.01.bin");    \
2317 MODULE_FIRMWARE("nvidia/"#chip"/gsp/gsp-535.113.01.bin")
2318 
2319 NVKM_GSP_FIRMWARE(tu102);
2320 NVKM_GSP_FIRMWARE(tu104);
2321 NVKM_GSP_FIRMWARE(tu106);
2322 
2323 NVKM_GSP_FIRMWARE(tu116);
2324 NVKM_GSP_FIRMWARE(tu117);
2325 
2326 NVKM_GSP_FIRMWARE(ga100);
2327 
2328 NVKM_GSP_FIRMWARE(ga102);
2329 NVKM_GSP_FIRMWARE(ga103);
2330 NVKM_GSP_FIRMWARE(ga104);
2331 NVKM_GSP_FIRMWARE(ga106);
2332 NVKM_GSP_FIRMWARE(ga107);
2333 
2334 NVKM_GSP_FIRMWARE(ad102);
2335 NVKM_GSP_FIRMWARE(ad103);
2336 NVKM_GSP_FIRMWARE(ad104);
2337 NVKM_GSP_FIRMWARE(ad106);
2338 NVKM_GSP_FIRMWARE(ad107);
2339