1 /*
2 * Copyright 2018 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 "gk104.h"
23 #include "cgrp.h"
24 #include "changk104.h"
25 #include "user.h"
26
27 #include <core/client.h>
28 #include <core/gpuobj.h>
29 #include <subdev/bar.h>
30 #include <subdev/fault.h>
31 #include <subdev/top.h>
32 #include <subdev/timer.h>
33 #include <engine/sw.h>
34
35 #include <nvif/class.h>
36
37 static void
tu102_fifo_runlist_commit(struct gk104_fifo * fifo,int runl,struct nvkm_memory * mem,int nr)38 tu102_fifo_runlist_commit(struct gk104_fifo *fifo, int runl,
39 struct nvkm_memory *mem, int nr)
40 {
41 struct nvkm_device *device = fifo->base.engine.subdev.device;
42 u64 addr = nvkm_memory_addr(mem);
43 /*XXX: target? */
44
45 nvkm_wr32(device, 0x002b00 + (runl * 0x10), lower_32_bits(addr));
46 nvkm_wr32(device, 0x002b04 + (runl * 0x10), upper_32_bits(addr));
47 nvkm_wr32(device, 0x002b08 + (runl * 0x10), nr);
48
49 /*XXX: how to wait? can you even wait? */
50 }
51
52 static const struct gk104_fifo_runlist_func
53 tu102_fifo_runlist = {
54 .size = 16,
55 .cgrp = gv100_fifo_runlist_cgrp,
56 .chan = gv100_fifo_runlist_chan,
57 .commit = tu102_fifo_runlist_commit,
58 };
59
60 static const struct nvkm_enum
61 tu102_fifo_fault_engine[] = {
62 { 0x01, "DISPLAY" },
63 { 0x03, "PTP" },
64 { 0x06, "PWR_PMU" },
65 { 0x08, "IFB", NULL, NVKM_ENGINE_IFB },
66 { 0x09, "PERF" },
67 { 0x1f, "PHYSICAL" },
68 { 0x20, "HOST0" },
69 { 0x21, "HOST1" },
70 { 0x22, "HOST2" },
71 { 0x23, "HOST3" },
72 { 0x24, "HOST4" },
73 { 0x25, "HOST5" },
74 { 0x26, "HOST6" },
75 { 0x27, "HOST7" },
76 { 0x28, "HOST8" },
77 { 0x29, "HOST9" },
78 { 0x2a, "HOST10" },
79 { 0x2b, "HOST11" },
80 { 0x2c, "HOST12" },
81 { 0x2d, "HOST13" },
82 { 0x2e, "HOST14" },
83 { 0x80, "BAR1", NULL, NVKM_SUBDEV_BAR },
84 { 0xc0, "BAR2", NULL, NVKM_SUBDEV_INSTMEM },
85 {}
86 };
87
88 static void
tu102_fifo_pbdma_init(struct gk104_fifo * fifo)89 tu102_fifo_pbdma_init(struct gk104_fifo *fifo)
90 {
91 struct nvkm_device *device = fifo->base.engine.subdev.device;
92 const u32 mask = (1 << fifo->pbdma_nr) - 1;
93 /*XXX: this is a bit of a guess at this point in time. */
94 nvkm_mask(device, 0xb65000, 0x80000fff, 0x80000000 | mask);
95 }
96
97 static const struct gk104_fifo_pbdma_func
98 tu102_fifo_pbdma = {
99 .nr = gm200_fifo_pbdma_nr,
100 .init = tu102_fifo_pbdma_init,
101 .init_timeout = gk208_fifo_pbdma_init_timeout,
102 };
103
104 static const struct gk104_fifo_func
105 tu102_fifo = {
106 .pbdma = &tu102_fifo_pbdma,
107 .fault.access = gv100_fifo_fault_access,
108 .fault.engine = tu102_fifo_fault_engine,
109 .fault.reason = gv100_fifo_fault_reason,
110 .fault.hubclient = gv100_fifo_fault_hubclient,
111 .fault.gpcclient = gv100_fifo_fault_gpcclient,
112 .runlist = &tu102_fifo_runlist,
113 .user = {{-1,-1,VOLTA_USERMODE_A }, tu102_fifo_user_new },
114 .chan = {{ 0, 0,TURING_CHANNEL_GPFIFO_A}, tu102_fifo_gpfifo_new },
115 .cgrp_force = true,
116 };
117
118 static void
tu102_fifo_recover_work(struct work_struct * w)119 tu102_fifo_recover_work(struct work_struct *w)
120 {
121 struct gk104_fifo *fifo = container_of(w, typeof(*fifo), recover.work);
122 struct nvkm_device *device = fifo->base.engine.subdev.device;
123 struct nvkm_engine *engine;
124 unsigned long flags;
125 u32 engm, runm, todo;
126 int engn, runl;
127
128 spin_lock_irqsave(&fifo->base.lock, flags);
129 runm = fifo->recover.runm;
130 engm = fifo->recover.engm;
131 fifo->recover.engm = 0;
132 fifo->recover.runm = 0;
133 spin_unlock_irqrestore(&fifo->base.lock, flags);
134
135 nvkm_mask(device, 0x002630, runm, runm);
136
137 for (todo = engm; engn = __ffs(todo), todo; todo &= ~BIT(engn)) {
138 if ((engine = fifo->engine[engn].engine)) {
139 nvkm_subdev_fini(&engine->subdev, false);
140 WARN_ON(nvkm_subdev_init(&engine->subdev));
141 }
142 }
143
144 for (todo = runm; runl = __ffs(todo), todo; todo &= ~BIT(runl))
145 gk104_fifo_runlist_update(fifo, runl);
146
147 nvkm_mask(device, 0x002630, runm, 0x00000000);
148 }
149
150 static void tu102_fifo_recover_engn(struct gk104_fifo *fifo, int engn);
151
152 static void
tu102_fifo_recover_runl(struct gk104_fifo * fifo,int runl)153 tu102_fifo_recover_runl(struct gk104_fifo *fifo, int runl)
154 {
155 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
156 struct nvkm_device *device = subdev->device;
157 const u32 runm = BIT(runl);
158
159 assert_spin_locked(&fifo->base.lock);
160 if (fifo->recover.runm & runm)
161 return;
162 fifo->recover.runm |= runm;
163
164 /* Block runlist to prevent channel assignment(s) from changing. */
165 nvkm_mask(device, 0x002630, runm, runm);
166
167 /* Schedule recovery. */
168 nvkm_warn(subdev, "runlist %d: scheduled for recovery\n", runl);
169 schedule_work(&fifo->recover.work);
170 }
171
172 static struct gk104_fifo_chan *
tu102_fifo_recover_chid(struct gk104_fifo * fifo,int runl,int chid)173 tu102_fifo_recover_chid(struct gk104_fifo *fifo, int runl, int chid)
174 {
175 struct gk104_fifo_chan *chan;
176 struct nvkm_fifo_cgrp *cgrp;
177
178 list_for_each_entry(chan, &fifo->runlist[runl].chan, head) {
179 if (chan->base.chid == chid) {
180 list_del_init(&chan->head);
181 return chan;
182 }
183 }
184
185 list_for_each_entry(cgrp, &fifo->runlist[runl].cgrp, head) {
186 if (cgrp->id == chid) {
187 chan = list_first_entry(&cgrp->chan, typeof(*chan), head);
188 list_del_init(&chan->head);
189 if (!--cgrp->chan_nr)
190 list_del_init(&cgrp->head);
191 return chan;
192 }
193 }
194
195 return NULL;
196 }
197
198 static void
tu102_fifo_recover_chan(struct nvkm_fifo * base,int chid)199 tu102_fifo_recover_chan(struct nvkm_fifo *base, int chid)
200 {
201 struct gk104_fifo *fifo = gk104_fifo(base);
202 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
203 struct nvkm_device *device = subdev->device;
204 const u32 stat = nvkm_rd32(device, 0x800004 + (chid * 0x08));
205 const u32 runl = (stat & 0x000f0000) >> 16;
206 const bool used = (stat & 0x00000001);
207 unsigned long engn, engm = fifo->runlist[runl].engm;
208 struct gk104_fifo_chan *chan;
209
210 assert_spin_locked(&fifo->base.lock);
211 if (!used)
212 return;
213
214 /* Lookup SW state for channel, and mark it as dead. */
215 chan = tu102_fifo_recover_chid(fifo, runl, chid);
216 if (chan) {
217 chan->killed = true;
218 nvkm_fifo_kevent(&fifo->base, chid);
219 }
220
221 /* Disable channel. */
222 nvkm_wr32(device, 0x800004 + (chid * 0x08), stat | 0x00000800);
223 nvkm_warn(subdev, "channel %d: killed\n", chid);
224
225 /* Block channel assignments from changing during recovery. */
226 tu102_fifo_recover_runl(fifo, runl);
227
228 /* Schedule recovery for any engines the channel is on. */
229 for_each_set_bit(engn, &engm, fifo->engine_nr) {
230 struct gk104_fifo_engine_status status;
231
232 gk104_fifo_engine_status(fifo, engn, &status);
233 if (!status.chan || status.chan->id != chid)
234 continue;
235 tu102_fifo_recover_engn(fifo, engn);
236 }
237 }
238
239 static void
tu102_fifo_recover_engn(struct gk104_fifo * fifo,int engn)240 tu102_fifo_recover_engn(struct gk104_fifo *fifo, int engn)
241 {
242 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
243 struct nvkm_device *device = subdev->device;
244 const u32 runl = fifo->engine[engn].runl;
245 const u32 engm = BIT(engn);
246 struct gk104_fifo_engine_status status;
247
248 assert_spin_locked(&fifo->base.lock);
249 if (fifo->recover.engm & engm)
250 return;
251 fifo->recover.engm |= engm;
252
253 /* Block channel assignments from changing during recovery. */
254 tu102_fifo_recover_runl(fifo, runl);
255
256 /* Determine which channel (if any) is currently on the engine. */
257 gk104_fifo_engine_status(fifo, engn, &status);
258 if (status.chan) {
259 /* The channel is not longer viable, kill it. */
260 tu102_fifo_recover_chan(&fifo->base, status.chan->id);
261 }
262
263 /* Preempt the runlist */
264 nvkm_wr32(device, 0x2638, BIT(runl));
265
266 /* Schedule recovery. */
267 nvkm_warn(subdev, "engine %d: scheduled for recovery\n", engn);
268 schedule_work(&fifo->recover.work);
269 }
270
271 static void
tu102_fifo_fault(struct nvkm_fifo * base,struct nvkm_fault_data * info)272 tu102_fifo_fault(struct nvkm_fifo *base, struct nvkm_fault_data *info)
273 {
274 struct gk104_fifo *fifo = gk104_fifo(base);
275 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
276 struct nvkm_device *device = subdev->device;
277 const struct nvkm_enum *er, *ee, *ec, *ea;
278 struct nvkm_engine *engine = NULL;
279 struct nvkm_fifo_chan *chan;
280 unsigned long flags;
281 const char *en = "";
282 char ct[8] = "HUB/";
283 int engn;
284
285 er = nvkm_enum_find(fifo->func->fault.reason, info->reason);
286 ee = nvkm_enum_find(fifo->func->fault.engine, info->engine);
287 if (info->hub) {
288 ec = nvkm_enum_find(fifo->func->fault.hubclient, info->client);
289 } else {
290 ec = nvkm_enum_find(fifo->func->fault.gpcclient, info->client);
291 snprintf(ct, sizeof(ct), "GPC%d/", info->gpc);
292 }
293 ea = nvkm_enum_find(fifo->func->fault.access, info->access);
294
295 if (ee && ee->data2) {
296 switch (ee->data2) {
297 case NVKM_SUBDEV_BAR:
298 nvkm_bar_bar1_reset(device);
299 break;
300 case NVKM_SUBDEV_INSTMEM:
301 nvkm_bar_bar2_reset(device);
302 break;
303 case NVKM_ENGINE_IFB:
304 nvkm_mask(device, 0x001718, 0x00000000, 0x00000000);
305 break;
306 default:
307 engine = nvkm_device_engine(device, ee->data2, 0);
308 break;
309 }
310 }
311
312 if (ee == NULL) {
313 struct nvkm_subdev *subdev = nvkm_top_fault(device, info->engine);
314 if (subdev) {
315 if (subdev->func == &nvkm_engine)
316 engine = container_of(subdev, typeof(*engine), subdev);
317 en = engine->subdev.name;
318 }
319 } else {
320 en = ee->name;
321 }
322
323 spin_lock_irqsave(&fifo->base.lock, flags);
324 chan = nvkm_fifo_chan_inst_locked(&fifo->base, info->inst);
325
326 nvkm_error(subdev,
327 "fault %02x [%s] at %016llx engine %02x [%s] client %02x "
328 "[%s%s] reason %02x [%s] on channel %d [%010llx %s]\n",
329 info->access, ea ? ea->name : "", info->addr,
330 info->engine, ee ? ee->name : en,
331 info->client, ct, ec ? ec->name : "",
332 info->reason, er ? er->name : "", chan ? chan->chid : -1,
333 info->inst, chan ? chan->object.client->name : "unknown");
334
335 /* Kill the channel that caused the fault. */
336 if (chan)
337 tu102_fifo_recover_chan(&fifo->base, chan->chid);
338
339 /* Channel recovery will probably have already done this for the
340 * correct engine(s), but just in case we can't find the channel
341 * information...
342 */
343 for (engn = 0; engn < fifo->engine_nr && engine; engn++) {
344 if (fifo->engine[engn].engine == engine) {
345 tu102_fifo_recover_engn(fifo, engn);
346 break;
347 }
348 }
349
350 spin_unlock_irqrestore(&fifo->base.lock, flags);
351 }
352
353 static void
tu102_fifo_intr_ctxsw_timeout(struct gk104_fifo * fifo)354 tu102_fifo_intr_ctxsw_timeout(struct gk104_fifo *fifo)
355 {
356 struct nvkm_device *device = fifo->base.engine.subdev.device;
357 unsigned long flags, engm;
358 u32 engn;
359
360 spin_lock_irqsave(&fifo->base.lock, flags);
361
362 engm = nvkm_rd32(device, 0x2a30);
363 nvkm_wr32(device, 0x2a30, engm);
364
365 for_each_set_bit(engn, &engm, 32)
366 tu102_fifo_recover_engn(fifo, engn);
367
368 spin_unlock_irqrestore(&fifo->base.lock, flags);
369 }
370
371 static void
tu102_fifo_intr_sched(struct gk104_fifo * fifo)372 tu102_fifo_intr_sched(struct gk104_fifo *fifo)
373 {
374 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
375 struct nvkm_device *device = subdev->device;
376 u32 intr = nvkm_rd32(device, 0x00254c);
377 u32 code = intr & 0x000000ff;
378
379 nvkm_error(subdev, "SCHED_ERROR %02x\n", code);
380 }
381
382 static void
tu102_fifo_intr(struct nvkm_fifo * base)383 tu102_fifo_intr(struct nvkm_fifo *base)
384 {
385 struct gk104_fifo *fifo = gk104_fifo(base);
386 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
387 struct nvkm_device *device = subdev->device;
388 u32 mask = nvkm_rd32(device, 0x002140);
389 u32 stat = nvkm_rd32(device, 0x002100) & mask;
390
391 if (stat & 0x00000001) {
392 gk104_fifo_intr_bind(fifo);
393 nvkm_wr32(device, 0x002100, 0x00000001);
394 stat &= ~0x00000001;
395 }
396
397 if (stat & 0x00000002) {
398 tu102_fifo_intr_ctxsw_timeout(fifo);
399 stat &= ~0x00000002;
400 }
401
402 if (stat & 0x00000100) {
403 tu102_fifo_intr_sched(fifo);
404 nvkm_wr32(device, 0x002100, 0x00000100);
405 stat &= ~0x00000100;
406 }
407
408 if (stat & 0x00010000) {
409 gk104_fifo_intr_chsw(fifo);
410 nvkm_wr32(device, 0x002100, 0x00010000);
411 stat &= ~0x00010000;
412 }
413
414 if (stat & 0x20000000) {
415 u32 mask = nvkm_rd32(device, 0x0025a0);
416
417 while (mask) {
418 u32 unit = __ffs(mask);
419
420 gk104_fifo_intr_pbdma_0(fifo, unit);
421 gk104_fifo_intr_pbdma_1(fifo, unit);
422 nvkm_wr32(device, 0x0025a0, (1 << unit));
423 mask &= ~(1 << unit);
424 }
425 stat &= ~0x20000000;
426 }
427
428 if (stat & 0x40000000) {
429 gk104_fifo_intr_runlist(fifo);
430 stat &= ~0x40000000;
431 }
432
433 if (stat & 0x80000000) {
434 nvkm_wr32(device, 0x002100, 0x80000000);
435 gk104_fifo_intr_engine(fifo);
436 stat &= ~0x80000000;
437 }
438
439 if (stat) {
440 nvkm_error(subdev, "INTR %08x\n", stat);
441 nvkm_mask(device, 0x002140, stat, 0x00000000);
442 nvkm_wr32(device, 0x002100, stat);
443 }
444 }
445
446 static const struct nvkm_fifo_func
447 tu102_fifo_ = {
448 .dtor = gk104_fifo_dtor,
449 .oneinit = gk104_fifo_oneinit,
450 .info = gk104_fifo_info,
451 .init = gk104_fifo_init,
452 .fini = gk104_fifo_fini,
453 .intr = tu102_fifo_intr,
454 .fault = tu102_fifo_fault,
455 .engine_id = gk104_fifo_engine_id,
456 .id_engine = gk104_fifo_id_engine,
457 .uevent_init = gk104_fifo_uevent_init,
458 .uevent_fini = gk104_fifo_uevent_fini,
459 .recover_chan = tu102_fifo_recover_chan,
460 .class_get = gk104_fifo_class_get,
461 .class_new = gk104_fifo_class_new,
462 };
463
464 int
tu102_fifo_new(struct nvkm_device * device,enum nvkm_subdev_type type,int inst,struct nvkm_fifo ** pfifo)465 tu102_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
466 struct nvkm_fifo **pfifo)
467 {
468 struct gk104_fifo *fifo;
469
470 if (!(fifo = kzalloc(sizeof(*fifo), GFP_KERNEL)))
471 return -ENOMEM;
472 fifo->func = &tu102_fifo;
473 INIT_WORK(&fifo->recover.work, tu102_fifo_recover_work);
474 *pfifo = &fifo->base;
475
476 return nvkm_fifo_ctor(&tu102_fifo_, device, type, inst, 4096, &fifo->base);
477 }
478