1 /*
2  * Copyright 2012 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  * Authors: Ben Skeggs
23  */
24 #include "gf100.h"
25 #include "ctxgf100.h"
26 #include "fuc/os.h"
27 
28 #include <core/client.h>
29 #include <core/firmware.h>
30 #include <core/option.h>
31 #include <subdev/acr.h>
32 #include <subdev/fb.h>
33 #include <subdev/mc.h>
34 #include <subdev/pmu.h>
35 #include <subdev/therm.h>
36 #include <subdev/timer.h>
37 #include <engine/fifo.h>
38 
39 #include <nvif/class.h>
40 #include <nvif/cl9097.h>
41 #include <nvif/if900d.h>
42 #include <nvif/unpack.h>
43 
44 /*******************************************************************************
45  * Zero Bandwidth Clear
46  ******************************************************************************/
47 
48 static void
49 gf100_gr_zbc_clear_color(struct gf100_gr *gr, int zbc)
50 {
51 	struct nvkm_device *device = gr->base.engine.subdev.device;
52 	if (gr->zbc_color[zbc].format) {
53 		nvkm_wr32(device, 0x405804, gr->zbc_color[zbc].ds[0]);
54 		nvkm_wr32(device, 0x405808, gr->zbc_color[zbc].ds[1]);
55 		nvkm_wr32(device, 0x40580c, gr->zbc_color[zbc].ds[2]);
56 		nvkm_wr32(device, 0x405810, gr->zbc_color[zbc].ds[3]);
57 	}
58 	nvkm_wr32(device, 0x405814, gr->zbc_color[zbc].format);
59 	nvkm_wr32(device, 0x405820, zbc);
60 	nvkm_wr32(device, 0x405824, 0x00000004); /* TRIGGER | WRITE | COLOR */
61 }
62 
63 static int
64 gf100_gr_zbc_color_get(struct gf100_gr *gr, int format,
65 		       const u32 ds[4], const u32 l2[4])
66 {
67 	struct nvkm_ltc *ltc = gr->base.engine.subdev.device->ltc;
68 	int zbc = -ENOSPC, i;
69 
70 	for (i = ltc->zbc_color_min; i <= ltc->zbc_color_max; i++) {
71 		if (gr->zbc_color[i].format) {
72 			if (gr->zbc_color[i].format != format)
73 				continue;
74 			if (memcmp(gr->zbc_color[i].ds, ds, sizeof(
75 				   gr->zbc_color[i].ds)))
76 				continue;
77 			if (memcmp(gr->zbc_color[i].l2, l2, sizeof(
78 				   gr->zbc_color[i].l2))) {
79 				WARN_ON(1);
80 				return -EINVAL;
81 			}
82 			return i;
83 		} else {
84 			zbc = (zbc < 0) ? i : zbc;
85 		}
86 	}
87 
88 	if (zbc < 0)
89 		return zbc;
90 
91 	memcpy(gr->zbc_color[zbc].ds, ds, sizeof(gr->zbc_color[zbc].ds));
92 	memcpy(gr->zbc_color[zbc].l2, l2, sizeof(gr->zbc_color[zbc].l2));
93 	gr->zbc_color[zbc].format = format;
94 	nvkm_ltc_zbc_color_get(ltc, zbc, l2);
95 	gr->func->zbc->clear_color(gr, zbc);
96 	return zbc;
97 }
98 
99 static void
100 gf100_gr_zbc_clear_depth(struct gf100_gr *gr, int zbc)
101 {
102 	struct nvkm_device *device = gr->base.engine.subdev.device;
103 	if (gr->zbc_depth[zbc].format)
104 		nvkm_wr32(device, 0x405818, gr->zbc_depth[zbc].ds);
105 	nvkm_wr32(device, 0x40581c, gr->zbc_depth[zbc].format);
106 	nvkm_wr32(device, 0x405820, zbc);
107 	nvkm_wr32(device, 0x405824, 0x00000005); /* TRIGGER | WRITE | DEPTH */
108 }
109 
110 static int
111 gf100_gr_zbc_depth_get(struct gf100_gr *gr, int format,
112 		       const u32 ds, const u32 l2)
113 {
114 	struct nvkm_ltc *ltc = gr->base.engine.subdev.device->ltc;
115 	int zbc = -ENOSPC, i;
116 
117 	for (i = ltc->zbc_depth_min; i <= ltc->zbc_depth_max; i++) {
118 		if (gr->zbc_depth[i].format) {
119 			if (gr->zbc_depth[i].format != format)
120 				continue;
121 			if (gr->zbc_depth[i].ds != ds)
122 				continue;
123 			if (gr->zbc_depth[i].l2 != l2) {
124 				WARN_ON(1);
125 				return -EINVAL;
126 			}
127 			return i;
128 		} else {
129 			zbc = (zbc < 0) ? i : zbc;
130 		}
131 	}
132 
133 	if (zbc < 0)
134 		return zbc;
135 
136 	gr->zbc_depth[zbc].format = format;
137 	gr->zbc_depth[zbc].ds = ds;
138 	gr->zbc_depth[zbc].l2 = l2;
139 	nvkm_ltc_zbc_depth_get(ltc, zbc, l2);
140 	gr->func->zbc->clear_depth(gr, zbc);
141 	return zbc;
142 }
143 
144 const struct gf100_gr_func_zbc
145 gf100_gr_zbc = {
146 	.clear_color = gf100_gr_zbc_clear_color,
147 	.clear_depth = gf100_gr_zbc_clear_depth,
148 };
149 
150 /*******************************************************************************
151  * Graphics object classes
152  ******************************************************************************/
153 #define gf100_gr_object(p) container_of((p), struct gf100_gr_object, object)
154 
155 struct gf100_gr_object {
156 	struct nvkm_object object;
157 	struct gf100_gr_chan *chan;
158 };
159 
160 static int
161 gf100_fermi_mthd_zbc_color(struct nvkm_object *object, void *data, u32 size)
162 {
163 	struct gf100_gr *gr = gf100_gr(nvkm_gr(object->engine));
164 	union {
165 		struct fermi_a_zbc_color_v0 v0;
166 	} *args = data;
167 	int ret = -ENOSYS;
168 
169 	if (!(ret = nvif_unpack(ret, &data, &size, args->v0, 0, 0, false))) {
170 		switch (args->v0.format) {
171 		case FERMI_A_ZBC_COLOR_V0_FMT_ZERO:
172 		case FERMI_A_ZBC_COLOR_V0_FMT_UNORM_ONE:
173 		case FERMI_A_ZBC_COLOR_V0_FMT_RF32_GF32_BF32_AF32:
174 		case FERMI_A_ZBC_COLOR_V0_FMT_R16_G16_B16_A16:
175 		case FERMI_A_ZBC_COLOR_V0_FMT_RN16_GN16_BN16_AN16:
176 		case FERMI_A_ZBC_COLOR_V0_FMT_RS16_GS16_BS16_AS16:
177 		case FERMI_A_ZBC_COLOR_V0_FMT_RU16_GU16_BU16_AU16:
178 		case FERMI_A_ZBC_COLOR_V0_FMT_RF16_GF16_BF16_AF16:
179 		case FERMI_A_ZBC_COLOR_V0_FMT_A8R8G8B8:
180 		case FERMI_A_ZBC_COLOR_V0_FMT_A8RL8GL8BL8:
181 		case FERMI_A_ZBC_COLOR_V0_FMT_A2B10G10R10:
182 		case FERMI_A_ZBC_COLOR_V0_FMT_AU2BU10GU10RU10:
183 		case FERMI_A_ZBC_COLOR_V0_FMT_A8B8G8R8:
184 		case FERMI_A_ZBC_COLOR_V0_FMT_A8BL8GL8RL8:
185 		case FERMI_A_ZBC_COLOR_V0_FMT_AN8BN8GN8RN8:
186 		case FERMI_A_ZBC_COLOR_V0_FMT_AS8BS8GS8RS8:
187 		case FERMI_A_ZBC_COLOR_V0_FMT_AU8BU8GU8RU8:
188 		case FERMI_A_ZBC_COLOR_V0_FMT_A2R10G10B10:
189 		case FERMI_A_ZBC_COLOR_V0_FMT_BF10GF11RF11:
190 			ret = gf100_gr_zbc_color_get(gr, args->v0.format,
191 							   args->v0.ds,
192 							   args->v0.l2);
193 			if (ret >= 0) {
194 				args->v0.index = ret;
195 				return 0;
196 			}
197 			break;
198 		default:
199 			return -EINVAL;
200 		}
201 	}
202 
203 	return ret;
204 }
205 
206 static int
207 gf100_fermi_mthd_zbc_depth(struct nvkm_object *object, void *data, u32 size)
208 {
209 	struct gf100_gr *gr = gf100_gr(nvkm_gr(object->engine));
210 	union {
211 		struct fermi_a_zbc_depth_v0 v0;
212 	} *args = data;
213 	int ret = -ENOSYS;
214 
215 	if (!(ret = nvif_unpack(ret, &data, &size, args->v0, 0, 0, false))) {
216 		switch (args->v0.format) {
217 		case FERMI_A_ZBC_DEPTH_V0_FMT_FP32:
218 			ret = gf100_gr_zbc_depth_get(gr, args->v0.format,
219 							   args->v0.ds,
220 							   args->v0.l2);
221 			return (ret >= 0) ? 0 : -ENOSPC;
222 		default:
223 			return -EINVAL;
224 		}
225 	}
226 
227 	return ret;
228 }
229 
230 static int
231 gf100_fermi_mthd(struct nvkm_object *object, u32 mthd, void *data, u32 size)
232 {
233 	nvif_ioctl(object, "fermi mthd %08x\n", mthd);
234 	switch (mthd) {
235 	case FERMI_A_ZBC_COLOR:
236 		return gf100_fermi_mthd_zbc_color(object, data, size);
237 	case FERMI_A_ZBC_DEPTH:
238 		return gf100_fermi_mthd_zbc_depth(object, data, size);
239 	default:
240 		break;
241 	}
242 	return -EINVAL;
243 }
244 
245 const struct nvkm_object_func
246 gf100_fermi = {
247 	.mthd = gf100_fermi_mthd,
248 };
249 
250 static void
251 gf100_gr_mthd_set_shader_exceptions(struct nvkm_device *device, u32 data)
252 {
253 	nvkm_wr32(device, 0x419e44, data ? 0xffffffff : 0x00000000);
254 	nvkm_wr32(device, 0x419e4c, data ? 0xffffffff : 0x00000000);
255 }
256 
257 static bool
258 gf100_gr_mthd_sw(struct nvkm_device *device, u16 class, u32 mthd, u32 data)
259 {
260 	switch (class & 0x00ff) {
261 	case 0x97:
262 	case 0xc0:
263 		switch (mthd) {
264 		case 0x1528:
265 			gf100_gr_mthd_set_shader_exceptions(device, data);
266 			return true;
267 		default:
268 			break;
269 		}
270 		break;
271 	default:
272 		break;
273 	}
274 	return false;
275 }
276 
277 static const struct nvkm_object_func
278 gf100_gr_object_func = {
279 };
280 
281 static int
282 gf100_gr_object_new(const struct nvkm_oclass *oclass, void *data, u32 size,
283 		    struct nvkm_object **pobject)
284 {
285 	struct gf100_gr_chan *chan = gf100_gr_chan(oclass->parent);
286 	struct gf100_gr_object *object;
287 
288 	if (!(object = kzalloc(sizeof(*object), GFP_KERNEL)))
289 		return -ENOMEM;
290 	*pobject = &object->object;
291 
292 	nvkm_object_ctor(oclass->base.func ? oclass->base.func :
293 			 &gf100_gr_object_func, oclass, &object->object);
294 	object->chan = chan;
295 	return 0;
296 }
297 
298 static int
299 gf100_gr_object_get(struct nvkm_gr *base, int index, struct nvkm_sclass *sclass)
300 {
301 	struct gf100_gr *gr = gf100_gr(base);
302 	int c = 0;
303 
304 	while (gr->func->sclass[c].oclass) {
305 		if (c++ == index) {
306 			*sclass = gr->func->sclass[index];
307 			sclass->ctor = gf100_gr_object_new;
308 			return index;
309 		}
310 	}
311 
312 	return c;
313 }
314 
315 /*******************************************************************************
316  * PGRAPH context
317  ******************************************************************************/
318 
319 static int
320 gf100_gr_chan_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent,
321 		   int align, struct nvkm_gpuobj **pgpuobj)
322 {
323 	struct gf100_gr_chan *chan = gf100_gr_chan(object);
324 	struct gf100_gr *gr = chan->gr;
325 	int ret, i;
326 
327 	ret = nvkm_gpuobj_new(gr->base.engine.subdev.device, gr->size,
328 			      align, false, parent, pgpuobj);
329 	if (ret)
330 		return ret;
331 
332 	nvkm_kmap(*pgpuobj);
333 	for (i = 0; i < gr->size; i += 4)
334 		nvkm_wo32(*pgpuobj, i, gr->data[i / 4]);
335 
336 	if (!gr->firmware) {
337 		nvkm_wo32(*pgpuobj, 0x00, chan->mmio_nr / 2);
338 		nvkm_wo32(*pgpuobj, 0x04, chan->mmio_vma->addr >> 8);
339 	} else {
340 		nvkm_wo32(*pgpuobj, 0xf4, 0);
341 		nvkm_wo32(*pgpuobj, 0xf8, 0);
342 		nvkm_wo32(*pgpuobj, 0x10, chan->mmio_nr / 2);
343 		nvkm_wo32(*pgpuobj, 0x14, lower_32_bits(chan->mmio_vma->addr));
344 		nvkm_wo32(*pgpuobj, 0x18, upper_32_bits(chan->mmio_vma->addr));
345 		nvkm_wo32(*pgpuobj, 0x1c, 1);
346 		nvkm_wo32(*pgpuobj, 0x20, 0);
347 		nvkm_wo32(*pgpuobj, 0x28, 0);
348 		nvkm_wo32(*pgpuobj, 0x2c, 0);
349 	}
350 	nvkm_done(*pgpuobj);
351 	return 0;
352 }
353 
354 static void *
355 gf100_gr_chan_dtor(struct nvkm_object *object)
356 {
357 	struct gf100_gr_chan *chan = gf100_gr_chan(object);
358 
359 	nvkm_vmm_put(chan->vmm, &chan->mmio_vma);
360 	nvkm_memory_unref(&chan->mmio);
361 
362 	nvkm_vmm_put(chan->vmm, &chan->attrib_cb);
363 	nvkm_vmm_put(chan->vmm, &chan->unknown);
364 	nvkm_vmm_put(chan->vmm, &chan->bundle_cb);
365 	nvkm_vmm_put(chan->vmm, &chan->pagepool);
366 	nvkm_vmm_unref(&chan->vmm);
367 	return chan;
368 }
369 
370 static const struct nvkm_object_func
371 gf100_gr_chan = {
372 	.dtor = gf100_gr_chan_dtor,
373 	.bind = gf100_gr_chan_bind,
374 };
375 
376 static int
377 gf100_gr_chan_new(struct nvkm_gr *base, struct nvkm_chan *fifoch,
378 		  const struct nvkm_oclass *oclass,
379 		  struct nvkm_object **pobject)
380 {
381 	struct gf100_gr *gr = gf100_gr(base);
382 	struct gf100_gr_chan *chan;
383 	struct gf100_vmm_map_v0 args = { .priv = 1 };
384 	struct nvkm_device *device = gr->base.engine.subdev.device;
385 	int ret;
386 
387 	if (!(chan = kzalloc(sizeof(*chan), GFP_KERNEL)))
388 		return -ENOMEM;
389 	nvkm_object_ctor(&gf100_gr_chan, oclass, &chan->object);
390 	chan->gr = gr;
391 	chan->vmm = nvkm_vmm_ref(fifoch->vmm);
392 	*pobject = &chan->object;
393 
394 	/* Map pagepool. */
395 	ret = nvkm_vmm_get(chan->vmm, 12, nvkm_memory_size(gr->pagepool), &chan->pagepool);
396 	if (ret)
397 		return ret;
398 
399 	ret = nvkm_memory_map(gr->pagepool, 0, chan->vmm, chan->pagepool, &args, sizeof(args));
400 	if (ret)
401 		return ret;
402 
403 	/* Map bundle circular buffer. */
404 	ret = nvkm_vmm_get(chan->vmm, 12, nvkm_memory_size(gr->bundle_cb), &chan->bundle_cb);
405 	if (ret)
406 		return ret;
407 
408 	ret = nvkm_memory_map(gr->bundle_cb, 0, chan->vmm, chan->bundle_cb, &args, sizeof(args));
409 	if (ret)
410 		return ret;
411 
412 	/* Map attribute circular buffer. */
413 	ret = nvkm_vmm_get(chan->vmm, 12, nvkm_memory_size(gr->attrib_cb), &chan->attrib_cb);
414 	if (ret)
415 		return ret;
416 
417 	if (device->card_type < GP100) {
418 		ret = nvkm_memory_map(gr->attrib_cb, 0, chan->vmm, chan->attrib_cb, NULL, 0);
419 		if (ret)
420 			return ret;
421 	} else {
422 		ret = nvkm_memory_map(gr->attrib_cb, 0, chan->vmm, chan->attrib_cb,
423 				      &args, sizeof(args));;
424 		if (ret)
425 			return ret;
426 	}
427 
428 	/* Map some context buffer of unknown purpose. */
429 	if (gr->func->grctx->unknown_size) {
430 		ret = nvkm_vmm_get(chan->vmm, 12, nvkm_memory_size(gr->unknown), &chan->unknown);
431 		if (ret)
432 			return ret;
433 
434 		ret = nvkm_memory_map(gr->unknown, 0, chan->vmm, chan->unknown,
435 				      &args, sizeof(args));
436 		if (ret)
437 			return ret;
438 	}
439 
440 	/* Generate golden context image. */
441 	mutex_lock(&gr->fecs.mutex);
442 	if (gr->data == NULL) {
443 		ret = gf100_grctx_generate(gr, chan, fifoch->inst);
444 		if (ret) {
445 			nvkm_error(&base->engine.subdev, "failed to construct context\n");
446 			return ret;
447 		}
448 	}
449 	mutex_unlock(&gr->fecs.mutex);
450 
451 	/* allocate memory for a "mmio list" buffer that's used by the HUB
452 	 * fuc to modify some per-context register settings on first load
453 	 * of the context.
454 	 */
455 	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x1000, 0x100,
456 			      false, &chan->mmio);
457 	if (ret)
458 		return ret;
459 
460 	ret = nvkm_vmm_get(fifoch->vmm, 12, 0x1000, &chan->mmio_vma);
461 	if (ret)
462 		return ret;
463 
464 	ret = nvkm_memory_map(chan->mmio, 0, fifoch->vmm,
465 			      chan->mmio_vma, &args, sizeof(args));
466 	if (ret)
467 		return ret;
468 
469 	/* finally, fill in the mmio list and point the context at it */
470 	nvkm_kmap(chan->mmio);
471 	gr->func->grctx->pagepool(chan, chan->pagepool->addr);
472 	gr->func->grctx->bundle(chan, chan->bundle_cb->addr, gr->func->grctx->bundle_size);
473 	gr->func->grctx->attrib_cb(chan, chan->attrib_cb->addr, gr->func->grctx->attrib_cb_size(gr));
474 	gr->func->grctx->attrib(chan);
475 	if (gr->func->grctx->patch_ltc)
476 		gr->func->grctx->patch_ltc(chan);
477 	if (gr->func->grctx->unknown_size)
478 		gr->func->grctx->unknown(chan, chan->unknown->addr, gr->func->grctx->unknown_size);
479 	nvkm_done(chan->mmio);
480 	return 0;
481 }
482 
483 /*******************************************************************************
484  * PGRAPH register lists
485  ******************************************************************************/
486 
487 const struct gf100_gr_init
488 gf100_gr_init_main_0[] = {
489 	{ 0x400080,   1, 0x04, 0x003083c2 },
490 	{ 0x400088,   1, 0x04, 0x00006fe7 },
491 	{ 0x40008c,   1, 0x04, 0x00000000 },
492 	{ 0x400090,   1, 0x04, 0x00000030 },
493 	{ 0x40013c,   1, 0x04, 0x013901f7 },
494 	{ 0x400140,   1, 0x04, 0x00000100 },
495 	{ 0x400144,   1, 0x04, 0x00000000 },
496 	{ 0x400148,   1, 0x04, 0x00000110 },
497 	{ 0x400138,   1, 0x04, 0x00000000 },
498 	{ 0x400130,   2, 0x04, 0x00000000 },
499 	{ 0x400124,   1, 0x04, 0x00000002 },
500 	{}
501 };
502 
503 const struct gf100_gr_init
504 gf100_gr_init_fe_0[] = {
505 	{ 0x40415c,   1, 0x04, 0x00000000 },
506 	{ 0x404170,   1, 0x04, 0x00000000 },
507 	{}
508 };
509 
510 const struct gf100_gr_init
511 gf100_gr_init_pri_0[] = {
512 	{ 0x404488,   2, 0x04, 0x00000000 },
513 	{}
514 };
515 
516 const struct gf100_gr_init
517 gf100_gr_init_rstr2d_0[] = {
518 	{ 0x407808,   1, 0x04, 0x00000000 },
519 	{}
520 };
521 
522 const struct gf100_gr_init
523 gf100_gr_init_pd_0[] = {
524 	{ 0x406024,   1, 0x04, 0x00000000 },
525 	{}
526 };
527 
528 const struct gf100_gr_init
529 gf100_gr_init_ds_0[] = {
530 	{ 0x405844,   1, 0x04, 0x00ffffff },
531 	{ 0x405850,   1, 0x04, 0x00000000 },
532 	{ 0x405908,   1, 0x04, 0x00000000 },
533 	{}
534 };
535 
536 const struct gf100_gr_init
537 gf100_gr_init_scc_0[] = {
538 	{ 0x40803c,   1, 0x04, 0x00000000 },
539 	{}
540 };
541 
542 const struct gf100_gr_init
543 gf100_gr_init_prop_0[] = {
544 	{ 0x4184a0,   1, 0x04, 0x00000000 },
545 	{}
546 };
547 
548 const struct gf100_gr_init
549 gf100_gr_init_gpc_unk_0[] = {
550 	{ 0x418604,   1, 0x04, 0x00000000 },
551 	{ 0x418680,   1, 0x04, 0x00000000 },
552 	{ 0x418714,   1, 0x04, 0x80000000 },
553 	{ 0x418384,   1, 0x04, 0x00000000 },
554 	{}
555 };
556 
557 const struct gf100_gr_init
558 gf100_gr_init_setup_0[] = {
559 	{ 0x418814,   3, 0x04, 0x00000000 },
560 	{}
561 };
562 
563 const struct gf100_gr_init
564 gf100_gr_init_crstr_0[] = {
565 	{ 0x418b04,   1, 0x04, 0x00000000 },
566 	{}
567 };
568 
569 const struct gf100_gr_init
570 gf100_gr_init_setup_1[] = {
571 	{ 0x4188c8,   1, 0x04, 0x80000000 },
572 	{ 0x4188cc,   1, 0x04, 0x00000000 },
573 	{ 0x4188d0,   1, 0x04, 0x00010000 },
574 	{ 0x4188d4,   1, 0x04, 0x00000001 },
575 	{}
576 };
577 
578 const struct gf100_gr_init
579 gf100_gr_init_zcull_0[] = {
580 	{ 0x418910,   1, 0x04, 0x00010001 },
581 	{ 0x418914,   1, 0x04, 0x00000301 },
582 	{ 0x418918,   1, 0x04, 0x00800000 },
583 	{ 0x418980,   1, 0x04, 0x77777770 },
584 	{ 0x418984,   3, 0x04, 0x77777777 },
585 	{}
586 };
587 
588 const struct gf100_gr_init
589 gf100_gr_init_gpm_0[] = {
590 	{ 0x418c04,   1, 0x04, 0x00000000 },
591 	{ 0x418c88,   1, 0x04, 0x00000000 },
592 	{}
593 };
594 
595 const struct gf100_gr_init
596 gf100_gr_init_gpc_unk_1[] = {
597 	{ 0x418d00,   1, 0x04, 0x00000000 },
598 	{ 0x418f08,   1, 0x04, 0x00000000 },
599 	{ 0x418e00,   1, 0x04, 0x00000050 },
600 	{ 0x418e08,   1, 0x04, 0x00000000 },
601 	{}
602 };
603 
604 const struct gf100_gr_init
605 gf100_gr_init_gcc_0[] = {
606 	{ 0x41900c,   1, 0x04, 0x00000000 },
607 	{ 0x419018,   1, 0x04, 0x00000000 },
608 	{}
609 };
610 
611 const struct gf100_gr_init
612 gf100_gr_init_tpccs_0[] = {
613 	{ 0x419d08,   2, 0x04, 0x00000000 },
614 	{ 0x419d10,   1, 0x04, 0x00000014 },
615 	{}
616 };
617 
618 const struct gf100_gr_init
619 gf100_gr_init_tex_0[] = {
620 	{ 0x419ab0,   1, 0x04, 0x00000000 },
621 	{ 0x419ab8,   1, 0x04, 0x000000e7 },
622 	{ 0x419abc,   2, 0x04, 0x00000000 },
623 	{}
624 };
625 
626 const struct gf100_gr_init
627 gf100_gr_init_pe_0[] = {
628 	{ 0x41980c,   3, 0x04, 0x00000000 },
629 	{ 0x419844,   1, 0x04, 0x00000000 },
630 	{ 0x41984c,   1, 0x04, 0x00005bc5 },
631 	{ 0x419850,   4, 0x04, 0x00000000 },
632 	{}
633 };
634 
635 const struct gf100_gr_init
636 gf100_gr_init_l1c_0[] = {
637 	{ 0x419c98,   1, 0x04, 0x00000000 },
638 	{ 0x419ca8,   1, 0x04, 0x80000000 },
639 	{ 0x419cb4,   1, 0x04, 0x00000000 },
640 	{ 0x419cb8,   1, 0x04, 0x00008bf4 },
641 	{ 0x419cbc,   1, 0x04, 0x28137606 },
642 	{ 0x419cc0,   2, 0x04, 0x00000000 },
643 	{}
644 };
645 
646 const struct gf100_gr_init
647 gf100_gr_init_wwdx_0[] = {
648 	{ 0x419bd4,   1, 0x04, 0x00800000 },
649 	{ 0x419bdc,   1, 0x04, 0x00000000 },
650 	{}
651 };
652 
653 const struct gf100_gr_init
654 gf100_gr_init_tpccs_1[] = {
655 	{ 0x419d2c,   1, 0x04, 0x00000000 },
656 	{}
657 };
658 
659 const struct gf100_gr_init
660 gf100_gr_init_mpc_0[] = {
661 	{ 0x419c0c,   1, 0x04, 0x00000000 },
662 	{}
663 };
664 
665 static const struct gf100_gr_init
666 gf100_gr_init_sm_0[] = {
667 	{ 0x419e00,   1, 0x04, 0x00000000 },
668 	{ 0x419ea0,   1, 0x04, 0x00000000 },
669 	{ 0x419ea4,   1, 0x04, 0x00000100 },
670 	{ 0x419ea8,   1, 0x04, 0x00001100 },
671 	{ 0x419eac,   1, 0x04, 0x11100702 },
672 	{ 0x419eb0,   1, 0x04, 0x00000003 },
673 	{ 0x419eb4,   4, 0x04, 0x00000000 },
674 	{ 0x419ec8,   1, 0x04, 0x06060618 },
675 	{ 0x419ed0,   1, 0x04, 0x0eff0e38 },
676 	{ 0x419ed4,   1, 0x04, 0x011104f1 },
677 	{ 0x419edc,   1, 0x04, 0x00000000 },
678 	{ 0x419f00,   1, 0x04, 0x00000000 },
679 	{ 0x419f2c,   1, 0x04, 0x00000000 },
680 	{}
681 };
682 
683 const struct gf100_gr_init
684 gf100_gr_init_be_0[] = {
685 	{ 0x40880c,   1, 0x04, 0x00000000 },
686 	{ 0x408910,   9, 0x04, 0x00000000 },
687 	{ 0x408950,   1, 0x04, 0x00000000 },
688 	{ 0x408954,   1, 0x04, 0x0000ffff },
689 	{ 0x408984,   1, 0x04, 0x00000000 },
690 	{ 0x408988,   1, 0x04, 0x08040201 },
691 	{ 0x40898c,   1, 0x04, 0x80402010 },
692 	{}
693 };
694 
695 const struct gf100_gr_init
696 gf100_gr_init_fe_1[] = {
697 	{ 0x4040f0,   1, 0x04, 0x00000000 },
698 	{}
699 };
700 
701 const struct gf100_gr_init
702 gf100_gr_init_pe_1[] = {
703 	{ 0x419880,   1, 0x04, 0x00000002 },
704 	{}
705 };
706 
707 static const struct gf100_gr_pack
708 gf100_gr_pack_mmio[] = {
709 	{ gf100_gr_init_main_0 },
710 	{ gf100_gr_init_fe_0 },
711 	{ gf100_gr_init_pri_0 },
712 	{ gf100_gr_init_rstr2d_0 },
713 	{ gf100_gr_init_pd_0 },
714 	{ gf100_gr_init_ds_0 },
715 	{ gf100_gr_init_scc_0 },
716 	{ gf100_gr_init_prop_0 },
717 	{ gf100_gr_init_gpc_unk_0 },
718 	{ gf100_gr_init_setup_0 },
719 	{ gf100_gr_init_crstr_0 },
720 	{ gf100_gr_init_setup_1 },
721 	{ gf100_gr_init_zcull_0 },
722 	{ gf100_gr_init_gpm_0 },
723 	{ gf100_gr_init_gpc_unk_1 },
724 	{ gf100_gr_init_gcc_0 },
725 	{ gf100_gr_init_tpccs_0 },
726 	{ gf100_gr_init_tex_0 },
727 	{ gf100_gr_init_pe_0 },
728 	{ gf100_gr_init_l1c_0 },
729 	{ gf100_gr_init_wwdx_0 },
730 	{ gf100_gr_init_tpccs_1 },
731 	{ gf100_gr_init_mpc_0 },
732 	{ gf100_gr_init_sm_0 },
733 	{ gf100_gr_init_be_0 },
734 	{ gf100_gr_init_fe_1 },
735 	{ gf100_gr_init_pe_1 },
736 	{}
737 };
738 
739 /*******************************************************************************
740  * PGRAPH engine/subdev functions
741  ******************************************************************************/
742 
743 static u32
744 gf100_gr_ctxsw_inst(struct nvkm_gr *gr)
745 {
746 	return nvkm_rd32(gr->engine.subdev.device, 0x409b00);
747 }
748 
749 static int
750 gf100_gr_fecs_ctrl_ctxsw(struct gf100_gr *gr, u32 mthd)
751 {
752 	struct nvkm_device *device = gr->base.engine.subdev.device;
753 
754 	nvkm_wr32(device, 0x409804, 0xffffffff);
755 	nvkm_wr32(device, 0x409800, 0x00000000);
756 	nvkm_wr32(device, 0x409500, 0xffffffff);
757 	nvkm_wr32(device, 0x409504, mthd);
758 	nvkm_msec(device, 2000,
759 		u32 stat = nvkm_rd32(device, 0x409804);
760 		if (stat == 0x00000002)
761 			return -EIO;
762 		if (stat == 0x00000001)
763 			return 0;
764 	);
765 
766 	return -ETIMEDOUT;
767 }
768 
769 static int
770 gf100_gr_fecs_start_ctxsw(struct nvkm_gr *base)
771 {
772 	struct gf100_gr *gr = gf100_gr(base);
773 	int ret = 0;
774 
775 	mutex_lock(&gr->fecs.mutex);
776 	if (!--gr->fecs.disable) {
777 		if (WARN_ON(ret = gf100_gr_fecs_ctrl_ctxsw(gr, 0x39)))
778 			gr->fecs.disable++;
779 	}
780 	mutex_unlock(&gr->fecs.mutex);
781 	return ret;
782 }
783 
784 static int
785 gf100_gr_fecs_stop_ctxsw(struct nvkm_gr *base)
786 {
787 	struct gf100_gr *gr = gf100_gr(base);
788 	int ret = 0;
789 
790 	mutex_lock(&gr->fecs.mutex);
791 	if (!gr->fecs.disable++) {
792 		if (WARN_ON(ret = gf100_gr_fecs_ctrl_ctxsw(gr, 0x38)))
793 			gr->fecs.disable--;
794 	}
795 	mutex_unlock(&gr->fecs.mutex);
796 	return ret;
797 }
798 
799 static int
800 gf100_gr_fecs_halt_pipeline(struct gf100_gr *gr)
801 {
802 	int ret = 0;
803 
804 	if (gr->firmware) {
805 		mutex_lock(&gr->fecs.mutex);
806 		ret = gf100_gr_fecs_ctrl_ctxsw(gr, 0x04);
807 		mutex_unlock(&gr->fecs.mutex);
808 	}
809 
810 	return ret;
811 }
812 
813 int
814 gf100_gr_fecs_wfi_golden_save(struct gf100_gr *gr, u32 inst)
815 {
816 	struct nvkm_device *device = gr->base.engine.subdev.device;
817 
818 	nvkm_mask(device, 0x409800, 0x00000003, 0x00000000);
819 	nvkm_wr32(device, 0x409500, inst);
820 	nvkm_wr32(device, 0x409504, 0x00000009);
821 	nvkm_msec(device, 2000,
822 		u32 stat = nvkm_rd32(device, 0x409800);
823 		if (stat & 0x00000002)
824 			return -EIO;
825 		if (stat & 0x00000001)
826 			return 0;
827 	);
828 
829 	return -ETIMEDOUT;
830 }
831 
832 int
833 gf100_gr_fecs_bind_pointer(struct gf100_gr *gr, u32 inst)
834 {
835 	struct nvkm_device *device = gr->base.engine.subdev.device;
836 
837 	nvkm_mask(device, 0x409800, 0x00000030, 0x00000000);
838 	nvkm_wr32(device, 0x409500, inst);
839 	nvkm_wr32(device, 0x409504, 0x00000003);
840 	nvkm_msec(device, 2000,
841 		u32 stat = nvkm_rd32(device, 0x409800);
842 		if (stat & 0x00000020)
843 			return -EIO;
844 		if (stat & 0x00000010)
845 			return 0;
846 	);
847 
848 	return -ETIMEDOUT;
849 }
850 
851 static int
852 gf100_gr_fecs_set_reglist_virtual_address(struct gf100_gr *gr, u64 addr)
853 {
854 	struct nvkm_device *device = gr->base.engine.subdev.device;
855 
856 	nvkm_wr32(device, 0x409810, addr >> 8);
857 	nvkm_wr32(device, 0x409800, 0x00000000);
858 	nvkm_wr32(device, 0x409500, 0x00000001);
859 	nvkm_wr32(device, 0x409504, 0x00000032);
860 	nvkm_msec(device, 2000,
861 		if (nvkm_rd32(device, 0x409800) == 0x00000001)
862 			return 0;
863 	);
864 
865 	return -ETIMEDOUT;
866 }
867 
868 static int
869 gf100_gr_fecs_set_reglist_bind_instance(struct gf100_gr *gr, u32 inst)
870 {
871 	struct nvkm_device *device = gr->base.engine.subdev.device;
872 
873 	nvkm_wr32(device, 0x409810, inst);
874 	nvkm_wr32(device, 0x409800, 0x00000000);
875 	nvkm_wr32(device, 0x409500, 0x00000001);
876 	nvkm_wr32(device, 0x409504, 0x00000031);
877 	nvkm_msec(device, 2000,
878 		if (nvkm_rd32(device, 0x409800) == 0x00000001)
879 			return 0;
880 	);
881 
882 	return -ETIMEDOUT;
883 }
884 
885 static int
886 gf100_gr_fecs_discover_reglist_image_size(struct gf100_gr *gr, u32 *psize)
887 {
888 	struct nvkm_device *device = gr->base.engine.subdev.device;
889 
890 	nvkm_wr32(device, 0x409800, 0x00000000);
891 	nvkm_wr32(device, 0x409500, 0x00000001);
892 	nvkm_wr32(device, 0x409504, 0x00000030);
893 	nvkm_msec(device, 2000,
894 		if ((*psize = nvkm_rd32(device, 0x409800)))
895 			return 0;
896 	);
897 
898 	return -ETIMEDOUT;
899 }
900 
901 static int
902 gf100_gr_fecs_elpg_bind(struct gf100_gr *gr)
903 {
904 	u32 size;
905 	int ret;
906 
907 	ret = gf100_gr_fecs_discover_reglist_image_size(gr, &size);
908 	if (ret)
909 		return ret;
910 
911 	/*XXX: We need to allocate + map the above into PMU's inst block,
912 	 *     which which means we probably need a proper PMU before we
913 	 *     even bother.
914 	 */
915 
916 	ret = gf100_gr_fecs_set_reglist_bind_instance(gr, 0);
917 	if (ret)
918 		return ret;
919 
920 	return gf100_gr_fecs_set_reglist_virtual_address(gr, 0);
921 }
922 
923 static int
924 gf100_gr_fecs_discover_pm_image_size(struct gf100_gr *gr, u32 *psize)
925 {
926 	struct nvkm_device *device = gr->base.engine.subdev.device;
927 
928 	nvkm_wr32(device, 0x409800, 0x00000000);
929 	nvkm_wr32(device, 0x409500, 0x00000000);
930 	nvkm_wr32(device, 0x409504, 0x00000025);
931 	nvkm_msec(device, 2000,
932 		if ((*psize = nvkm_rd32(device, 0x409800)))
933 			return 0;
934 	);
935 
936 	return -ETIMEDOUT;
937 }
938 
939 static int
940 gf100_gr_fecs_discover_zcull_image_size(struct gf100_gr *gr, u32 *psize)
941 {
942 	struct nvkm_device *device = gr->base.engine.subdev.device;
943 
944 	nvkm_wr32(device, 0x409800, 0x00000000);
945 	nvkm_wr32(device, 0x409500, 0x00000000);
946 	nvkm_wr32(device, 0x409504, 0x00000016);
947 	nvkm_msec(device, 2000,
948 		if ((*psize = nvkm_rd32(device, 0x409800)))
949 			return 0;
950 	);
951 
952 	return -ETIMEDOUT;
953 }
954 
955 static int
956 gf100_gr_fecs_discover_image_size(struct gf100_gr *gr, u32 *psize)
957 {
958 	struct nvkm_device *device = gr->base.engine.subdev.device;
959 
960 	nvkm_wr32(device, 0x409800, 0x00000000);
961 	nvkm_wr32(device, 0x409500, 0x00000000);
962 	nvkm_wr32(device, 0x409504, 0x00000010);
963 	nvkm_msec(device, 2000,
964 		if ((*psize = nvkm_rd32(device, 0x409800)))
965 			return 0;
966 	);
967 
968 	return -ETIMEDOUT;
969 }
970 
971 static void
972 gf100_gr_fecs_set_watchdog_timeout(struct gf100_gr *gr, u32 timeout)
973 {
974 	struct nvkm_device *device = gr->base.engine.subdev.device;
975 
976 	nvkm_wr32(device, 0x409800, 0x00000000);
977 	nvkm_wr32(device, 0x409500, timeout);
978 	nvkm_wr32(device, 0x409504, 0x00000021);
979 }
980 
981 static bool
982 gf100_gr_chsw_load(struct nvkm_gr *base)
983 {
984 	struct gf100_gr *gr = gf100_gr(base);
985 	if (!gr->firmware) {
986 		u32 trace = nvkm_rd32(gr->base.engine.subdev.device, 0x40981c);
987 		if (trace & 0x00000040)
988 			return true;
989 	} else {
990 		u32 mthd = nvkm_rd32(gr->base.engine.subdev.device, 0x409808);
991 		if (mthd & 0x00080000)
992 			return true;
993 	}
994 	return false;
995 }
996 
997 int
998 gf100_gr_rops(struct gf100_gr *gr)
999 {
1000 	struct nvkm_device *device = gr->base.engine.subdev.device;
1001 	return (nvkm_rd32(device, 0x409604) & 0x001f0000) >> 16;
1002 }
1003 
1004 void
1005 gf100_gr_zbc_init(struct gf100_gr *gr)
1006 {
1007 	const u32  zero[] = { 0x00000000, 0x00000000, 0x00000000, 0x00000000,
1008 			      0x00000000, 0x00000000, 0x00000000, 0x00000000 };
1009 	const u32   one[] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000,
1010 			      0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
1011 	const u32 f32_0[] = { 0x00000000, 0x00000000, 0x00000000, 0x00000000,
1012 			      0x00000000, 0x00000000, 0x00000000, 0x00000000 };
1013 	const u32 f32_1[] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000,
1014 			      0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 };
1015 	struct nvkm_ltc *ltc = gr->base.engine.subdev.device->ltc;
1016 	int index, c = ltc->zbc_color_min, d = ltc->zbc_depth_min, s = ltc->zbc_depth_min;
1017 
1018 	if (!gr->zbc_color[0].format) {
1019 		gf100_gr_zbc_color_get(gr, 1,  & zero[0],   &zero[4]); c++;
1020 		gf100_gr_zbc_color_get(gr, 2,  &  one[0],    &one[4]); c++;
1021 		gf100_gr_zbc_color_get(gr, 4,  &f32_0[0],  &f32_0[4]); c++;
1022 		gf100_gr_zbc_color_get(gr, 4,  &f32_1[0],  &f32_1[4]); c++;
1023 		gf100_gr_zbc_depth_get(gr, 1, 0x00000000, 0x00000000); d++;
1024 		gf100_gr_zbc_depth_get(gr, 1, 0x3f800000, 0x3f800000); d++;
1025 		if (gr->func->zbc->stencil_get) {
1026 			gr->func->zbc->stencil_get(gr, 1, 0x00, 0x00); s++;
1027 			gr->func->zbc->stencil_get(gr, 1, 0x01, 0x01); s++;
1028 			gr->func->zbc->stencil_get(gr, 1, 0xff, 0xff); s++;
1029 		}
1030 	}
1031 
1032 	for (index = c; index <= ltc->zbc_color_max; index++)
1033 		gr->func->zbc->clear_color(gr, index);
1034 	for (index = d; index <= ltc->zbc_depth_max; index++)
1035 		gr->func->zbc->clear_depth(gr, index);
1036 
1037 	if (gr->func->zbc->clear_stencil) {
1038 		for (index = s; index <= ltc->zbc_depth_max; index++)
1039 			gr->func->zbc->clear_stencil(gr, index);
1040 	}
1041 }
1042 
1043 /**
1044  * Wait until GR goes idle. GR is considered idle if it is disabled by the
1045  * MC (0x200) register, or GR is not busy and a context switch is not in
1046  * progress.
1047  */
1048 int
1049 gf100_gr_wait_idle(struct gf100_gr *gr)
1050 {
1051 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1052 	struct nvkm_device *device = subdev->device;
1053 	unsigned long end_jiffies = jiffies + msecs_to_jiffies(2000);
1054 	bool gr_enabled, ctxsw_active, gr_busy;
1055 
1056 	do {
1057 		/*
1058 		 * required to make sure FIFO_ENGINE_STATUS (0x2640) is
1059 		 * up-to-date
1060 		 */
1061 		nvkm_rd32(device, 0x400700);
1062 
1063 		gr_enabled = nvkm_rd32(device, 0x200) & 0x1000;
1064 		ctxsw_active = nvkm_fifo_ctxsw_in_progress(&gr->base.engine);
1065 		gr_busy = nvkm_rd32(device, 0x40060c) & 0x1;
1066 
1067 		if (!gr_enabled || (!gr_busy && !ctxsw_active))
1068 			return 0;
1069 	} while (time_before(jiffies, end_jiffies));
1070 
1071 	nvkm_error(subdev,
1072 		   "wait for idle timeout (en: %d, ctxsw: %d, busy: %d)\n",
1073 		   gr_enabled, ctxsw_active, gr_busy);
1074 	return -EAGAIN;
1075 }
1076 
1077 void
1078 gf100_gr_mmio(struct gf100_gr *gr, const struct gf100_gr_pack *p)
1079 {
1080 	struct nvkm_device *device = gr->base.engine.subdev.device;
1081 	const struct gf100_gr_pack *pack;
1082 	const struct gf100_gr_init *init;
1083 
1084 	pack_for_each_init(init, pack, p) {
1085 		u32 next = init->addr + init->count * init->pitch;
1086 		u32 addr = init->addr;
1087 		while (addr < next) {
1088 			nvkm_wr32(device, addr, init->data);
1089 			addr += init->pitch;
1090 		}
1091 	}
1092 }
1093 
1094 void
1095 gf100_gr_icmd(struct gf100_gr *gr, const struct gf100_gr_pack *p)
1096 {
1097 	struct nvkm_device *device = gr->base.engine.subdev.device;
1098 	const struct gf100_gr_pack *pack;
1099 	const struct gf100_gr_init *init;
1100 	u64 data = 0;
1101 
1102 	nvkm_wr32(device, 0x400208, 0x80000000);
1103 
1104 	pack_for_each_init(init, pack, p) {
1105 		u32 next = init->addr + init->count * init->pitch;
1106 		u32 addr = init->addr;
1107 
1108 		if ((pack == p && init == p->init) || data != init->data) {
1109 			nvkm_wr32(device, 0x400204, init->data);
1110 			if (pack->type == 64)
1111 				nvkm_wr32(device, 0x40020c, upper_32_bits(init->data));
1112 			data = init->data;
1113 		}
1114 
1115 		while (addr < next) {
1116 			nvkm_wr32(device, 0x400200, addr);
1117 			/**
1118 			 * Wait for GR to go idle after submitting a
1119 			 * GO_IDLE bundle
1120 			 */
1121 			if ((addr & 0xffff) == 0xe100)
1122 				gf100_gr_wait_idle(gr);
1123 			nvkm_msec(device, 2000,
1124 				if (!(nvkm_rd32(device, 0x400700) & 0x00000004))
1125 					break;
1126 			);
1127 			addr += init->pitch;
1128 		}
1129 	}
1130 
1131 	nvkm_wr32(device, 0x400208, 0x00000000);
1132 }
1133 
1134 void
1135 gf100_gr_mthd(struct gf100_gr *gr, const struct gf100_gr_pack *p)
1136 {
1137 	struct nvkm_device *device = gr->base.engine.subdev.device;
1138 	const struct gf100_gr_pack *pack;
1139 	const struct gf100_gr_init *init;
1140 	u32 data = 0;
1141 
1142 	pack_for_each_init(init, pack, p) {
1143 		u32 ctrl = 0x80000000 | pack->type;
1144 		u32 next = init->addr + init->count * init->pitch;
1145 		u32 addr = init->addr;
1146 
1147 		if ((pack == p && init == p->init) || data != init->data) {
1148 			nvkm_wr32(device, 0x40448c, init->data);
1149 			data = init->data;
1150 		}
1151 
1152 		while (addr < next) {
1153 			nvkm_wr32(device, 0x404488, ctrl | (addr << 14));
1154 			addr += init->pitch;
1155 		}
1156 	}
1157 }
1158 
1159 u64
1160 gf100_gr_units(struct nvkm_gr *base)
1161 {
1162 	struct gf100_gr *gr = gf100_gr(base);
1163 	u64 cfg;
1164 
1165 	cfg  = (u32)gr->gpc_nr;
1166 	cfg |= (u32)gr->tpc_total << 8;
1167 	cfg |= (u64)gr->rop_nr << 32;
1168 
1169 	return cfg;
1170 }
1171 
1172 static const struct nvkm_bitfield gf100_dispatch_error[] = {
1173 	{ 0x00000001, "INJECTED_BUNDLE_ERROR" },
1174 	{ 0x00000002, "CLASS_SUBCH_MISMATCH" },
1175 	{ 0x00000004, "SUBCHSW_DURING_NOTIFY" },
1176 	{}
1177 };
1178 
1179 static const struct nvkm_bitfield gf100_m2mf_error[] = {
1180 	{ 0x00000001, "PUSH_TOO_MUCH_DATA" },
1181 	{ 0x00000002, "PUSH_NOT_ENOUGH_DATA" },
1182 	{}
1183 };
1184 
1185 static const struct nvkm_bitfield gf100_unk6_error[] = {
1186 	{ 0x00000001, "TEMP_TOO_SMALL" },
1187 	{}
1188 };
1189 
1190 static const struct nvkm_bitfield gf100_ccache_error[] = {
1191 	{ 0x00000001, "INTR" },
1192 	{ 0x00000002, "LDCONST_OOB" },
1193 	{}
1194 };
1195 
1196 static const struct nvkm_bitfield gf100_macro_error[] = {
1197 	{ 0x00000001, "TOO_FEW_PARAMS" },
1198 	{ 0x00000002, "TOO_MANY_PARAMS" },
1199 	{ 0x00000004, "ILLEGAL_OPCODE" },
1200 	{ 0x00000008, "DOUBLE_BRANCH" },
1201 	{ 0x00000010, "WATCHDOG" },
1202 	{}
1203 };
1204 
1205 static const struct nvkm_bitfield gk104_sked_error[] = {
1206 	{ 0x00000040, "CTA_RESUME" },
1207 	{ 0x00000080, "CONSTANT_BUFFER_SIZE" },
1208 	{ 0x00000200, "LOCAL_MEMORY_SIZE_POS" },
1209 	{ 0x00000400, "LOCAL_MEMORY_SIZE_NEG" },
1210 	{ 0x00000800, "WARP_CSTACK_SIZE" },
1211 	{ 0x00001000, "TOTAL_TEMP_SIZE" },
1212 	{ 0x00002000, "REGISTER_COUNT" },
1213 	{ 0x00040000, "TOTAL_THREADS" },
1214 	{ 0x00100000, "PROGRAM_OFFSET" },
1215 	{ 0x00200000, "SHARED_MEMORY_SIZE" },
1216 	{ 0x00800000, "CTA_THREAD_DIMENSION_ZERO" },
1217 	{ 0x01000000, "MEMORY_WINDOW_OVERLAP" },
1218 	{ 0x02000000, "SHARED_CONFIG_TOO_SMALL" },
1219 	{ 0x04000000, "TOTAL_REGISTER_COUNT" },
1220 	{}
1221 };
1222 
1223 static const struct nvkm_bitfield gf100_gpc_rop_error[] = {
1224 	{ 0x00000002, "RT_PITCH_OVERRUN" },
1225 	{ 0x00000010, "RT_WIDTH_OVERRUN" },
1226 	{ 0x00000020, "RT_HEIGHT_OVERRUN" },
1227 	{ 0x00000080, "ZETA_STORAGE_TYPE_MISMATCH" },
1228 	{ 0x00000100, "RT_STORAGE_TYPE_MISMATCH" },
1229 	{ 0x00000400, "RT_LINEAR_MISMATCH" },
1230 	{}
1231 };
1232 
1233 static void
1234 gf100_gr_trap_gpc_rop(struct gf100_gr *gr, int gpc)
1235 {
1236 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1237 	struct nvkm_device *device = subdev->device;
1238 	char error[128];
1239 	u32 trap[4];
1240 
1241 	trap[0] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0420)) & 0x3fffffff;
1242 	trap[1] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0434));
1243 	trap[2] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0438));
1244 	trap[3] = nvkm_rd32(device, GPC_UNIT(gpc, 0x043c));
1245 
1246 	nvkm_snprintbf(error, sizeof(error), gf100_gpc_rop_error, trap[0]);
1247 
1248 	nvkm_error(subdev, "GPC%d/PROP trap: %08x [%s] x = %u, y = %u, "
1249 			   "format = %x, storage type = %x\n",
1250 		   gpc, trap[0], error, trap[1] & 0xffff, trap[1] >> 16,
1251 		   (trap[2] >> 8) & 0x3f, trap[3] & 0xff);
1252 	nvkm_wr32(device, GPC_UNIT(gpc, 0x0420), 0xc0000000);
1253 }
1254 
1255 const struct nvkm_enum gf100_mp_warp_error[] = {
1256 	{ 0x01, "STACK_ERROR" },
1257 	{ 0x02, "API_STACK_ERROR" },
1258 	{ 0x03, "RET_EMPTY_STACK_ERROR" },
1259 	{ 0x04, "PC_WRAP" },
1260 	{ 0x05, "MISALIGNED_PC" },
1261 	{ 0x06, "PC_OVERFLOW" },
1262 	{ 0x07, "MISALIGNED_IMMC_ADDR" },
1263 	{ 0x08, "MISALIGNED_REG" },
1264 	{ 0x09, "ILLEGAL_INSTR_ENCODING" },
1265 	{ 0x0a, "ILLEGAL_SPH_INSTR_COMBO" },
1266 	{ 0x0b, "ILLEGAL_INSTR_PARAM" },
1267 	{ 0x0c, "INVALID_CONST_ADDR" },
1268 	{ 0x0d, "OOR_REG" },
1269 	{ 0x0e, "OOR_ADDR" },
1270 	{ 0x0f, "MISALIGNED_ADDR" },
1271 	{ 0x10, "INVALID_ADDR_SPACE" },
1272 	{ 0x11, "ILLEGAL_INSTR_PARAM2" },
1273 	{ 0x12, "INVALID_CONST_ADDR_LDC" },
1274 	{ 0x13, "GEOMETRY_SM_ERROR" },
1275 	{ 0x14, "DIVERGENT" },
1276 	{ 0x15, "WARP_EXIT" },
1277 	{}
1278 };
1279 
1280 const struct nvkm_bitfield gf100_mp_global_error[] = {
1281 	{ 0x00000001, "SM_TO_SM_FAULT" },
1282 	{ 0x00000002, "L1_ERROR" },
1283 	{ 0x00000004, "MULTIPLE_WARP_ERRORS" },
1284 	{ 0x00000008, "PHYSICAL_STACK_OVERFLOW" },
1285 	{ 0x00000010, "BPT_INT" },
1286 	{ 0x00000020, "BPT_PAUSE" },
1287 	{ 0x00000040, "SINGLE_STEP_COMPLETE" },
1288 	{ 0x20000000, "ECC_SEC_ERROR" },
1289 	{ 0x40000000, "ECC_DED_ERROR" },
1290 	{ 0x80000000, "TIMEOUT" },
1291 	{}
1292 };
1293 
1294 void
1295 gf100_gr_trap_mp(struct gf100_gr *gr, int gpc, int tpc)
1296 {
1297 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1298 	struct nvkm_device *device = subdev->device;
1299 	u32 werr = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x648));
1300 	u32 gerr = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x650));
1301 	const struct nvkm_enum *warp;
1302 	char glob[128];
1303 
1304 	nvkm_snprintbf(glob, sizeof(glob), gf100_mp_global_error, gerr);
1305 	warp = nvkm_enum_find(gf100_mp_warp_error, werr & 0xffff);
1306 
1307 	nvkm_error(subdev, "GPC%i/TPC%i/MP trap: "
1308 			   "global %08x [%s] warp %04x [%s]\n",
1309 		   gpc, tpc, gerr, glob, werr, warp ? warp->name : "");
1310 
1311 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x648), 0x00000000);
1312 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x650), gerr);
1313 }
1314 
1315 static void
1316 gf100_gr_trap_tpc(struct gf100_gr *gr, int gpc, int tpc)
1317 {
1318 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1319 	struct nvkm_device *device = subdev->device;
1320 	u32 stat = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0508));
1321 
1322 	if (stat & 0x00000001) {
1323 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0224));
1324 		nvkm_error(subdev, "GPC%d/TPC%d/TEX: %08x\n", gpc, tpc, trap);
1325 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x0224), 0xc0000000);
1326 		stat &= ~0x00000001;
1327 	}
1328 
1329 	if (stat & 0x00000002) {
1330 		gr->func->trap_mp(gr, gpc, tpc);
1331 		stat &= ~0x00000002;
1332 	}
1333 
1334 	if (stat & 0x00000004) {
1335 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0084));
1336 		nvkm_error(subdev, "GPC%d/TPC%d/POLY: %08x\n", gpc, tpc, trap);
1337 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x0084), 0xc0000000);
1338 		stat &= ~0x00000004;
1339 	}
1340 
1341 	if (stat & 0x00000008) {
1342 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x048c));
1343 		nvkm_error(subdev, "GPC%d/TPC%d/L1C: %08x\n", gpc, tpc, trap);
1344 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x048c), 0xc0000000);
1345 		stat &= ~0x00000008;
1346 	}
1347 
1348 	if (stat & 0x00000010) {
1349 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0430));
1350 		nvkm_error(subdev, "GPC%d/TPC%d/MPC: %08x\n", gpc, tpc, trap);
1351 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x0430), 0xc0000000);
1352 		stat &= ~0x00000010;
1353 	}
1354 
1355 	if (stat) {
1356 		nvkm_error(subdev, "GPC%d/TPC%d/%08x: unknown\n", gpc, tpc, stat);
1357 	}
1358 }
1359 
1360 static void
1361 gf100_gr_trap_gpc(struct gf100_gr *gr, int gpc)
1362 {
1363 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1364 	struct nvkm_device *device = subdev->device;
1365 	u32 stat = nvkm_rd32(device, GPC_UNIT(gpc, 0x2c90));
1366 	int tpc;
1367 
1368 	if (stat & 0x00000001) {
1369 		gf100_gr_trap_gpc_rop(gr, gpc);
1370 		stat &= ~0x00000001;
1371 	}
1372 
1373 	if (stat & 0x00000002) {
1374 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x0900));
1375 		nvkm_error(subdev, "GPC%d/ZCULL: %08x\n", gpc, trap);
1376 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0900), 0xc0000000);
1377 		stat &= ~0x00000002;
1378 	}
1379 
1380 	if (stat & 0x00000004) {
1381 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x1028));
1382 		nvkm_error(subdev, "GPC%d/CCACHE: %08x\n", gpc, trap);
1383 		nvkm_wr32(device, GPC_UNIT(gpc, 0x1028), 0xc0000000);
1384 		stat &= ~0x00000004;
1385 	}
1386 
1387 	if (stat & 0x00000008) {
1388 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x0824));
1389 		nvkm_error(subdev, "GPC%d/ESETUP: %08x\n", gpc, trap);
1390 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0824), 0xc0000000);
1391 		stat &= ~0x00000009;
1392 	}
1393 
1394 	for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++) {
1395 		u32 mask = 0x00010000 << tpc;
1396 		if (stat & mask) {
1397 			gf100_gr_trap_tpc(gr, gpc, tpc);
1398 			nvkm_wr32(device, GPC_UNIT(gpc, 0x2c90), mask);
1399 			stat &= ~mask;
1400 		}
1401 	}
1402 
1403 	if (stat) {
1404 		nvkm_error(subdev, "GPC%d/%08x: unknown\n", gpc, stat);
1405 	}
1406 }
1407 
1408 static void
1409 gf100_gr_trap_intr(struct gf100_gr *gr)
1410 {
1411 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1412 	struct nvkm_device *device = subdev->device;
1413 	char error[128];
1414 	u32 trap = nvkm_rd32(device, 0x400108);
1415 	int rop, gpc;
1416 
1417 	if (trap & 0x00000001) {
1418 		u32 stat = nvkm_rd32(device, 0x404000);
1419 
1420 		nvkm_snprintbf(error, sizeof(error), gf100_dispatch_error,
1421 			       stat & 0x3fffffff);
1422 		nvkm_error(subdev, "DISPATCH %08x [%s]\n", stat, error);
1423 		nvkm_wr32(device, 0x404000, 0xc0000000);
1424 		nvkm_wr32(device, 0x400108, 0x00000001);
1425 		trap &= ~0x00000001;
1426 	}
1427 
1428 	if (trap & 0x00000002) {
1429 		u32 stat = nvkm_rd32(device, 0x404600);
1430 
1431 		nvkm_snprintbf(error, sizeof(error), gf100_m2mf_error,
1432 			       stat & 0x3fffffff);
1433 		nvkm_error(subdev, "M2MF %08x [%s]\n", stat, error);
1434 
1435 		nvkm_wr32(device, 0x404600, 0xc0000000);
1436 		nvkm_wr32(device, 0x400108, 0x00000002);
1437 		trap &= ~0x00000002;
1438 	}
1439 
1440 	if (trap & 0x00000008) {
1441 		u32 stat = nvkm_rd32(device, 0x408030);
1442 
1443 		nvkm_snprintbf(error, sizeof(error), gf100_ccache_error,
1444 			       stat & 0x3fffffff);
1445 		nvkm_error(subdev, "CCACHE %08x [%s]\n", stat, error);
1446 		nvkm_wr32(device, 0x408030, 0xc0000000);
1447 		nvkm_wr32(device, 0x400108, 0x00000008);
1448 		trap &= ~0x00000008;
1449 	}
1450 
1451 	if (trap & 0x00000010) {
1452 		u32 stat = nvkm_rd32(device, 0x405840);
1453 		nvkm_error(subdev, "SHADER %08x, sph: 0x%06x, stage: 0x%02x\n",
1454 			   stat, stat & 0xffffff, (stat >> 24) & 0x3f);
1455 		nvkm_wr32(device, 0x405840, 0xc0000000);
1456 		nvkm_wr32(device, 0x400108, 0x00000010);
1457 		trap &= ~0x00000010;
1458 	}
1459 
1460 	if (trap & 0x00000040) {
1461 		u32 stat = nvkm_rd32(device, 0x40601c);
1462 
1463 		nvkm_snprintbf(error, sizeof(error), gf100_unk6_error,
1464 			       stat & 0x3fffffff);
1465 		nvkm_error(subdev, "UNK6 %08x [%s]\n", stat, error);
1466 
1467 		nvkm_wr32(device, 0x40601c, 0xc0000000);
1468 		nvkm_wr32(device, 0x400108, 0x00000040);
1469 		trap &= ~0x00000040;
1470 	}
1471 
1472 	if (trap & 0x00000080) {
1473 		u32 stat = nvkm_rd32(device, 0x404490);
1474 		u32 pc = nvkm_rd32(device, 0x404494);
1475 		u32 op = nvkm_rd32(device, 0x40449c);
1476 
1477 		nvkm_snprintbf(error, sizeof(error), gf100_macro_error,
1478 			       stat & 0x1fffffff);
1479 		nvkm_error(subdev, "MACRO %08x [%s], pc: 0x%03x%s, op: 0x%08x\n",
1480 			   stat, error, pc & 0x7ff,
1481 			   (pc & 0x10000000) ? "" : " (invalid)",
1482 			   op);
1483 
1484 		nvkm_wr32(device, 0x404490, 0xc0000000);
1485 		nvkm_wr32(device, 0x400108, 0x00000080);
1486 		trap &= ~0x00000080;
1487 	}
1488 
1489 	if (trap & 0x00000100) {
1490 		u32 stat = nvkm_rd32(device, 0x407020) & 0x3fffffff;
1491 
1492 		nvkm_snprintbf(error, sizeof(error), gk104_sked_error, stat);
1493 		nvkm_error(subdev, "SKED: %08x [%s]\n", stat, error);
1494 
1495 		if (stat)
1496 			nvkm_wr32(device, 0x407020, 0x40000000);
1497 		nvkm_wr32(device, 0x400108, 0x00000100);
1498 		trap &= ~0x00000100;
1499 	}
1500 
1501 	if (trap & 0x01000000) {
1502 		u32 stat = nvkm_rd32(device, 0x400118);
1503 		for (gpc = 0; stat && gpc < gr->gpc_nr; gpc++) {
1504 			u32 mask = 0x00000001 << gpc;
1505 			if (stat & mask) {
1506 				gf100_gr_trap_gpc(gr, gpc);
1507 				nvkm_wr32(device, 0x400118, mask);
1508 				stat &= ~mask;
1509 			}
1510 		}
1511 		nvkm_wr32(device, 0x400108, 0x01000000);
1512 		trap &= ~0x01000000;
1513 	}
1514 
1515 	if (trap & 0x02000000) {
1516 		for (rop = 0; rop < gr->rop_nr; rop++) {
1517 			u32 statz = nvkm_rd32(device, ROP_UNIT(rop, 0x070));
1518 			u32 statc = nvkm_rd32(device, ROP_UNIT(rop, 0x144));
1519 			nvkm_error(subdev, "ROP%d %08x %08x\n",
1520 				 rop, statz, statc);
1521 			nvkm_wr32(device, ROP_UNIT(rop, 0x070), 0xc0000000);
1522 			nvkm_wr32(device, ROP_UNIT(rop, 0x144), 0xc0000000);
1523 		}
1524 		nvkm_wr32(device, 0x400108, 0x02000000);
1525 		trap &= ~0x02000000;
1526 	}
1527 
1528 	if (trap) {
1529 		nvkm_error(subdev, "TRAP UNHANDLED %08x\n", trap);
1530 		nvkm_wr32(device, 0x400108, trap);
1531 	}
1532 }
1533 
1534 static void
1535 gf100_gr_ctxctl_debug_unit(struct gf100_gr *gr, u32 base)
1536 {
1537 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1538 	struct nvkm_device *device = subdev->device;
1539 	nvkm_error(subdev, "%06x - done %08x\n", base,
1540 		   nvkm_rd32(device, base + 0x400));
1541 	nvkm_error(subdev, "%06x - stat %08x %08x %08x %08x\n", base,
1542 		   nvkm_rd32(device, base + 0x800),
1543 		   nvkm_rd32(device, base + 0x804),
1544 		   nvkm_rd32(device, base + 0x808),
1545 		   nvkm_rd32(device, base + 0x80c));
1546 	nvkm_error(subdev, "%06x - stat %08x %08x %08x %08x\n", base,
1547 		   nvkm_rd32(device, base + 0x810),
1548 		   nvkm_rd32(device, base + 0x814),
1549 		   nvkm_rd32(device, base + 0x818),
1550 		   nvkm_rd32(device, base + 0x81c));
1551 }
1552 
1553 void
1554 gf100_gr_ctxctl_debug(struct gf100_gr *gr)
1555 {
1556 	struct nvkm_device *device = gr->base.engine.subdev.device;
1557 	u32 gpcnr = nvkm_rd32(device, 0x409604) & 0xffff;
1558 	u32 gpc;
1559 
1560 	gf100_gr_ctxctl_debug_unit(gr, 0x409000);
1561 	for (gpc = 0; gpc < gpcnr; gpc++)
1562 		gf100_gr_ctxctl_debug_unit(gr, 0x502000 + (gpc * 0x8000));
1563 }
1564 
1565 static void
1566 gf100_gr_ctxctl_isr(struct gf100_gr *gr)
1567 {
1568 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1569 	struct nvkm_device *device = subdev->device;
1570 	u32 stat = nvkm_rd32(device, 0x409c18);
1571 
1572 	if (!gr->firmware && (stat & 0x00000001)) {
1573 		u32 code = nvkm_rd32(device, 0x409814);
1574 		if (code == E_BAD_FWMTHD) {
1575 			u32 class = nvkm_rd32(device, 0x409808);
1576 			u32  addr = nvkm_rd32(device, 0x40980c);
1577 			u32  subc = (addr & 0x00070000) >> 16;
1578 			u32  mthd = (addr & 0x00003ffc);
1579 			u32  data = nvkm_rd32(device, 0x409810);
1580 
1581 			nvkm_error(subdev, "FECS MTHD subc %d class %04x "
1582 					   "mthd %04x data %08x\n",
1583 				   subc, class, mthd, data);
1584 		} else {
1585 			nvkm_error(subdev, "FECS ucode error %d\n", code);
1586 		}
1587 		nvkm_wr32(device, 0x409c20, 0x00000001);
1588 		stat &= ~0x00000001;
1589 	}
1590 
1591 	if (!gr->firmware && (stat & 0x00080000)) {
1592 		nvkm_error(subdev, "FECS watchdog timeout\n");
1593 		gf100_gr_ctxctl_debug(gr);
1594 		nvkm_wr32(device, 0x409c20, 0x00080000);
1595 		stat &= ~0x00080000;
1596 	}
1597 
1598 	if (stat) {
1599 		nvkm_error(subdev, "FECS %08x\n", stat);
1600 		gf100_gr_ctxctl_debug(gr);
1601 		nvkm_wr32(device, 0x409c20, stat);
1602 	}
1603 }
1604 
1605 static irqreturn_t
1606 gf100_gr_intr(struct nvkm_inth *inth)
1607 {
1608 	struct gf100_gr *gr = container_of(inth, typeof(*gr), base.engine.subdev.inth);
1609 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1610 	struct nvkm_device *device = subdev->device;
1611 	struct nvkm_chan *chan;
1612 	unsigned long flags;
1613 	u64 inst = nvkm_rd32(device, 0x409b00) & 0x0fffffff;
1614 	u32 stat = nvkm_rd32(device, 0x400100);
1615 	u32 addr = nvkm_rd32(device, 0x400704);
1616 	u32 mthd = (addr & 0x00003ffc);
1617 	u32 subc = (addr & 0x00070000) >> 16;
1618 	u32 data = nvkm_rd32(device, 0x400708);
1619 	u32 code = nvkm_rd32(device, 0x400110);
1620 	u32 class;
1621 	const char *name = "unknown";
1622 	int chid = -1;
1623 
1624 	chan = nvkm_chan_get_inst(&gr->base.engine, (u64)inst << 12, &flags);
1625 	if (chan) {
1626 		name = chan->name;
1627 		chid = chan->id;
1628 	}
1629 
1630 	if (device->card_type < NV_E0 || subc < 4)
1631 		class = nvkm_rd32(device, 0x404200 + (subc * 4));
1632 	else
1633 		class = 0x0000;
1634 
1635 	if (stat & 0x00000001) {
1636 		/*
1637 		 * notifier interrupt, only needed for cyclestats
1638 		 * can be safely ignored
1639 		 */
1640 		nvkm_wr32(device, 0x400100, 0x00000001);
1641 		stat &= ~0x00000001;
1642 	}
1643 
1644 	if (stat & 0x00000010) {
1645 		if (!gf100_gr_mthd_sw(device, class, mthd, data)) {
1646 			nvkm_error(subdev, "ILLEGAL_MTHD ch %d [%010llx %s] "
1647 				   "subc %d class %04x mthd %04x data %08x\n",
1648 				   chid, inst << 12, name, subc,
1649 				   class, mthd, data);
1650 		}
1651 		nvkm_wr32(device, 0x400100, 0x00000010);
1652 		stat &= ~0x00000010;
1653 	}
1654 
1655 	if (stat & 0x00000020) {
1656 		nvkm_error(subdev, "ILLEGAL_CLASS ch %d [%010llx %s] "
1657 			   "subc %d class %04x mthd %04x data %08x\n",
1658 			   chid, inst << 12, name, subc, class, mthd, data);
1659 		nvkm_wr32(device, 0x400100, 0x00000020);
1660 		stat &= ~0x00000020;
1661 	}
1662 
1663 	if (stat & 0x00100000) {
1664 		const struct nvkm_enum *en =
1665 			nvkm_enum_find(nv50_data_error_names, code);
1666 		nvkm_error(subdev, "DATA_ERROR %08x [%s] ch %d [%010llx %s] "
1667 				   "subc %d class %04x mthd %04x data %08x\n",
1668 			   code, en ? en->name : "", chid, inst << 12,
1669 			   name, subc, class, mthd, data);
1670 		nvkm_wr32(device, 0x400100, 0x00100000);
1671 		stat &= ~0x00100000;
1672 	}
1673 
1674 	if (stat & 0x00200000) {
1675 		nvkm_error(subdev, "TRAP ch %d [%010llx %s]\n",
1676 			   chid, inst << 12, name);
1677 		gf100_gr_trap_intr(gr);
1678 		nvkm_wr32(device, 0x400100, 0x00200000);
1679 		stat &= ~0x00200000;
1680 	}
1681 
1682 	if (stat & 0x00080000) {
1683 		gf100_gr_ctxctl_isr(gr);
1684 		nvkm_wr32(device, 0x400100, 0x00080000);
1685 		stat &= ~0x00080000;
1686 	}
1687 
1688 	if (stat) {
1689 		nvkm_error(subdev, "intr %08x\n", stat);
1690 		nvkm_wr32(device, 0x400100, stat);
1691 	}
1692 
1693 	nvkm_wr32(device, 0x400500, 0x00010001);
1694 	nvkm_chan_put(&chan, flags);
1695 	return IRQ_HANDLED;
1696 }
1697 
1698 static void
1699 gf100_gr_init_fw(struct nvkm_falcon *falcon,
1700 		 struct nvkm_blob *code, struct nvkm_blob *data)
1701 {
1702 	nvkm_falcon_load_dmem(falcon, data->data, 0x0, data->size, 0);
1703 	nvkm_falcon_load_imem(falcon, code->data, 0x0, code->size, 0, 0, false);
1704 }
1705 
1706 static void
1707 gf100_gr_init_csdata(struct gf100_gr *gr,
1708 		     const struct gf100_gr_pack *pack,
1709 		     u32 falcon, u32 starstar, u32 base)
1710 {
1711 	struct nvkm_device *device = gr->base.engine.subdev.device;
1712 	const struct gf100_gr_pack *iter;
1713 	const struct gf100_gr_init *init;
1714 	u32 addr = ~0, prev = ~0, xfer = 0;
1715 	u32 star, temp;
1716 
1717 	nvkm_wr32(device, falcon + 0x01c0, 0x02000000 + starstar);
1718 	star = nvkm_rd32(device, falcon + 0x01c4);
1719 	temp = nvkm_rd32(device, falcon + 0x01c4);
1720 	if (temp > star)
1721 		star = temp;
1722 	nvkm_wr32(device, falcon + 0x01c0, 0x01000000 + star);
1723 
1724 	pack_for_each_init(init, iter, pack) {
1725 		u32 head = init->addr - base;
1726 		u32 tail = head + init->count * init->pitch;
1727 		while (head < tail) {
1728 			if (head != prev + 4 || xfer >= 32) {
1729 				if (xfer) {
1730 					u32 data = ((--xfer << 26) | addr);
1731 					nvkm_wr32(device, falcon + 0x01c4, data);
1732 					star += 4;
1733 				}
1734 				addr = head;
1735 				xfer = 0;
1736 			}
1737 			prev = head;
1738 			xfer = xfer + 1;
1739 			head = head + init->pitch;
1740 		}
1741 	}
1742 
1743 	nvkm_wr32(device, falcon + 0x01c4, (--xfer << 26) | addr);
1744 	nvkm_wr32(device, falcon + 0x01c0, 0x01000004 + starstar);
1745 	nvkm_wr32(device, falcon + 0x01c4, star + 4);
1746 }
1747 
1748 /* Initialize context from an external (secure or not) firmware */
1749 static int
1750 gf100_gr_init_ctxctl_ext(struct gf100_gr *gr)
1751 {
1752 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1753 	struct nvkm_device *device = subdev->device;
1754 	u32 lsf_mask = 0;
1755 	int ret;
1756 
1757 	/* load fuc microcode */
1758 	nvkm_mc_unk260(device, 0);
1759 
1760 	/* securely-managed falcons must be reset using secure boot */
1761 
1762 	if (!nvkm_acr_managed_falcon(device, NVKM_ACR_LSF_FECS)) {
1763 		gf100_gr_init_fw(&gr->fecs.falcon, &gr->fecs.inst,
1764 						   &gr->fecs.data);
1765 	} else {
1766 		lsf_mask |= BIT(NVKM_ACR_LSF_FECS);
1767 	}
1768 
1769 	if (!nvkm_acr_managed_falcon(device, NVKM_ACR_LSF_GPCCS)) {
1770 		gf100_gr_init_fw(&gr->gpccs.falcon, &gr->gpccs.inst,
1771 						    &gr->gpccs.data);
1772 	} else {
1773 		lsf_mask |= BIT(NVKM_ACR_LSF_GPCCS);
1774 	}
1775 
1776 	if (lsf_mask) {
1777 		ret = nvkm_acr_bootstrap_falcons(device, lsf_mask);
1778 		if (ret)
1779 			return ret;
1780 	}
1781 
1782 	nvkm_mc_unk260(device, 1);
1783 
1784 	/* start both of them running */
1785 	nvkm_wr32(device, 0x409800, 0x00000000);
1786 	nvkm_wr32(device, 0x41a10c, 0x00000000);
1787 	nvkm_wr32(device, 0x40910c, 0x00000000);
1788 
1789 	nvkm_falcon_start(&gr->gpccs.falcon);
1790 	nvkm_falcon_start(&gr->fecs.falcon);
1791 
1792 	if (nvkm_msec(device, 2000,
1793 		if (nvkm_rd32(device, 0x409800) & 0x00000001)
1794 			break;
1795 	) < 0)
1796 		return -EBUSY;
1797 
1798 	gf100_gr_fecs_set_watchdog_timeout(gr, 0x7fffffff);
1799 
1800 	/* Determine how much memory is required to store main context image. */
1801 	ret = gf100_gr_fecs_discover_image_size(gr, &gr->size);
1802 	if (ret)
1803 		return ret;
1804 
1805 	/* Determine how much memory is required to store ZCULL image. */
1806 	ret = gf100_gr_fecs_discover_zcull_image_size(gr, &gr->size_zcull);
1807 	if (ret)
1808 		return ret;
1809 
1810 	/* Determine how much memory is required to store PerfMon image. */
1811 	ret = gf100_gr_fecs_discover_pm_image_size(gr, &gr->size_pm);
1812 	if (ret)
1813 		return ret;
1814 
1815 	/*XXX: We (likely) require PMU support to even bother with this.
1816 	 *
1817 	 *     Also, it seems like not all GPUs support ELPG.  Traces I
1818 	 *     have here show RM enabling it on Kepler/Turing, but none
1819 	 *     of the GPUs between those.  NVGPU decides this by PCIID.
1820 	 */
1821 	if (0) {
1822 		ret = gf100_gr_fecs_elpg_bind(gr);
1823 		if (ret)
1824 			return ret;
1825 	}
1826 
1827 	return 0;
1828 }
1829 
1830 static int
1831 gf100_gr_init_ctxctl_int(struct gf100_gr *gr)
1832 {
1833 	const struct gf100_grctx_func *grctx = gr->func->grctx;
1834 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1835 	struct nvkm_device *device = subdev->device;
1836 
1837 	if (!gr->func->fecs.ucode) {
1838 		return -ENOSYS;
1839 	}
1840 
1841 	/* load HUB microcode */
1842 	nvkm_mc_unk260(device, 0);
1843 	nvkm_falcon_load_dmem(&gr->fecs.falcon,
1844 			      gr->func->fecs.ucode->data.data, 0x0,
1845 			      gr->func->fecs.ucode->data.size, 0);
1846 	nvkm_falcon_load_imem(&gr->fecs.falcon,
1847 			      gr->func->fecs.ucode->code.data, 0x0,
1848 			      gr->func->fecs.ucode->code.size, 0, 0, false);
1849 
1850 	/* load GPC microcode */
1851 	nvkm_falcon_load_dmem(&gr->gpccs.falcon,
1852 			      gr->func->gpccs.ucode->data.data, 0x0,
1853 			      gr->func->gpccs.ucode->data.size, 0);
1854 	nvkm_falcon_load_imem(&gr->gpccs.falcon,
1855 			      gr->func->gpccs.ucode->code.data, 0x0,
1856 			      gr->func->gpccs.ucode->code.size, 0, 0, false);
1857 	nvkm_mc_unk260(device, 1);
1858 
1859 	/* load register lists */
1860 	gf100_gr_init_csdata(gr, grctx->hub, 0x409000, 0x000, 0x000000);
1861 	gf100_gr_init_csdata(gr, grctx->gpc_0, 0x41a000, 0x000, 0x418000);
1862 	gf100_gr_init_csdata(gr, grctx->gpc_1, 0x41a000, 0x000, 0x418000);
1863 	gf100_gr_init_csdata(gr, grctx->tpc, 0x41a000, 0x004, 0x419800);
1864 	gf100_gr_init_csdata(gr, grctx->ppc, 0x41a000, 0x008, 0x41be00);
1865 
1866 	/* start HUB ucode running, it'll init the GPCs */
1867 	nvkm_wr32(device, 0x40910c, 0x00000000);
1868 	nvkm_wr32(device, 0x409100, 0x00000002);
1869 	if (nvkm_msec(device, 2000,
1870 		if (nvkm_rd32(device, 0x409800) & 0x80000000)
1871 			break;
1872 	) < 0) {
1873 		gf100_gr_ctxctl_debug(gr);
1874 		return -EBUSY;
1875 	}
1876 
1877 	gr->size = nvkm_rd32(device, 0x409804);
1878 	return 0;
1879 }
1880 
1881 int
1882 gf100_gr_init_ctxctl(struct gf100_gr *gr)
1883 {
1884 	int ret;
1885 
1886 	if (gr->firmware)
1887 		ret = gf100_gr_init_ctxctl_ext(gr);
1888 	else
1889 		ret = gf100_gr_init_ctxctl_int(gr);
1890 
1891 	return ret;
1892 }
1893 
1894 int
1895 gf100_gr_oneinit_sm_id(struct gf100_gr *gr)
1896 {
1897 	int tpc, gpc;
1898 
1899 	for (tpc = 0; tpc < gr->tpc_max; tpc++) {
1900 		for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
1901 			if (tpc < gr->tpc_nr[gpc]) {
1902 				gr->sm[gr->sm_nr].gpc = gpc;
1903 				gr->sm[gr->sm_nr].tpc = tpc;
1904 				gr->sm_nr++;
1905 			}
1906 		}
1907 	}
1908 
1909 	return 0;
1910 }
1911 
1912 void
1913 gf100_gr_oneinit_tiles(struct gf100_gr *gr)
1914 {
1915 	static const u8 primes[] = {
1916 		3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61
1917 	};
1918 	int init_frac[GPC_MAX], init_err[GPC_MAX], run_err[GPC_MAX], i, j;
1919 	u32 mul_factor, comm_denom;
1920 	u8  gpc_map[GPC_MAX];
1921 	bool sorted;
1922 
1923 	switch (gr->tpc_total) {
1924 	case 15: gr->screen_tile_row_offset = 0x06; break;
1925 	case 14: gr->screen_tile_row_offset = 0x05; break;
1926 	case 13: gr->screen_tile_row_offset = 0x02; break;
1927 	case 11: gr->screen_tile_row_offset = 0x07; break;
1928 	case 10: gr->screen_tile_row_offset = 0x06; break;
1929 	case  7:
1930 	case  5: gr->screen_tile_row_offset = 0x01; break;
1931 	case  3: gr->screen_tile_row_offset = 0x02; break;
1932 	case  2:
1933 	case  1: gr->screen_tile_row_offset = 0x01; break;
1934 	default: gr->screen_tile_row_offset = 0x03;
1935 		for (i = 0; i < ARRAY_SIZE(primes); i++) {
1936 			if (gr->tpc_total % primes[i]) {
1937 				gr->screen_tile_row_offset = primes[i];
1938 				break;
1939 			}
1940 		}
1941 		break;
1942 	}
1943 
1944 	/* Sort GPCs by TPC count, highest-to-lowest. */
1945 	for (i = 0; i < gr->gpc_nr; i++)
1946 		gpc_map[i] = i;
1947 	sorted = false;
1948 
1949 	while (!sorted) {
1950 		for (sorted = true, i = 0; i < gr->gpc_nr - 1; i++) {
1951 			if (gr->tpc_nr[gpc_map[i + 1]] >
1952 			    gr->tpc_nr[gpc_map[i + 0]]) {
1953 				u8 swap = gpc_map[i];
1954 				gpc_map[i + 0] = gpc_map[i + 1];
1955 				gpc_map[i + 1] = swap;
1956 				sorted = false;
1957 			}
1958 		}
1959 	}
1960 
1961 	/* Determine tile->GPC mapping */
1962 	mul_factor = gr->gpc_nr * gr->tpc_max;
1963 	if (mul_factor & 1)
1964 		mul_factor = 2;
1965 	else
1966 		mul_factor = 1;
1967 
1968 	comm_denom = gr->gpc_nr * gr->tpc_max * mul_factor;
1969 
1970 	for (i = 0; i < gr->gpc_nr; i++) {
1971 		init_frac[i] = gr->tpc_nr[gpc_map[i]] * gr->gpc_nr * mul_factor;
1972 		init_err[i] = i * gr->tpc_max * mul_factor - comm_denom/2;
1973 		run_err[i] = init_frac[i] + init_err[i];
1974 	}
1975 
1976 	for (i = 0; i < gr->tpc_total;) {
1977 		for (j = 0; j < gr->gpc_nr; j++) {
1978 			if ((run_err[j] * 2) >= comm_denom) {
1979 				gr->tile[i++] = gpc_map[j];
1980 				run_err[j] += init_frac[j] - comm_denom;
1981 			} else {
1982 				run_err[j] += init_frac[j];
1983 			}
1984 		}
1985 	}
1986 }
1987 
1988 static int
1989 gf100_gr_oneinit(struct nvkm_gr *base)
1990 {
1991 	struct gf100_gr *gr = gf100_gr(base);
1992 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1993 	struct nvkm_device *device = subdev->device;
1994 	struct nvkm_intr *intr = &device->mc->intr;
1995 	enum nvkm_intr_type intr_type = NVKM_INTR_SUBDEV;
1996 	int ret, i, j;
1997 
1998 	if (gr->func->oneinit_intr)
1999 		intr = gr->func->oneinit_intr(gr, &intr_type);
2000 
2001 	ret = nvkm_inth_add(intr, intr_type, NVKM_INTR_PRIO_NORMAL, &gr->base.engine.subdev,
2002 			    gf100_gr_intr, &gr->base.engine.subdev.inth);
2003 	if (ret)
2004 		return ret;
2005 
2006 	nvkm_pmu_pgob(device->pmu, false);
2007 
2008 	gr->rop_nr = gr->func->rops(gr);
2009 	gr->gpc_nr = nvkm_rd32(device, 0x409604) & 0x0000001f;
2010 	for (i = 0; i < gr->gpc_nr; i++) {
2011 		gr->tpc_nr[i]  = nvkm_rd32(device, GPC_UNIT(i, 0x2608));
2012 		gr->tpc_max = max(gr->tpc_max, gr->tpc_nr[i]);
2013 		gr->tpc_total += gr->tpc_nr[i];
2014 		for (j = 0; j < gr->func->ppc_nr; j++) {
2015 			gr->ppc_tpc_mask[i][j] =
2016 				nvkm_rd32(device, GPC_UNIT(i, 0x0c30 + (j * 4)));
2017 			if (gr->ppc_tpc_mask[i][j] == 0)
2018 				continue;
2019 
2020 			gr->ppc_nr[i]++;
2021 
2022 			gr->ppc_mask[i] |= (1 << j);
2023 			gr->ppc_tpc_nr[i][j] = hweight8(gr->ppc_tpc_mask[i][j]);
2024 			if (gr->ppc_tpc_min == 0 ||
2025 			    gr->ppc_tpc_min > gr->ppc_tpc_nr[i][j])
2026 				gr->ppc_tpc_min = gr->ppc_tpc_nr[i][j];
2027 			if (gr->ppc_tpc_max < gr->ppc_tpc_nr[i][j])
2028 				gr->ppc_tpc_max = gr->ppc_tpc_nr[i][j];
2029 		}
2030 
2031 		gr->ppc_total += gr->ppc_nr[i];
2032 	}
2033 
2034 	/* Allocate global context buffers. */
2035 	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, gr->func->grctx->pagepool_size,
2036 			      0x100, false, &gr->pagepool);
2037 	if (ret)
2038 		return ret;
2039 
2040 	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, gr->func->grctx->bundle_size,
2041 			      0x100, false, &gr->bundle_cb);
2042 	if (ret)
2043 		return ret;
2044 
2045 	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, gr->func->grctx->attrib_cb_size(gr),
2046 			      0x1000, false, &gr->attrib_cb);
2047 	if (ret)
2048 		return ret;
2049 
2050 	if (gr->func->grctx->unknown_size) {
2051 		ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, gr->func->grctx->unknown_size,
2052 				      0x100, false, &gr->unknown);
2053 		if (ret)
2054 			return ret;
2055 	}
2056 
2057 	memset(gr->tile, 0xff, sizeof(gr->tile));
2058 	gr->func->oneinit_tiles(gr);
2059 
2060 	return gr->func->oneinit_sm_id(gr);
2061 }
2062 
2063 static int
2064 gf100_gr_init_(struct nvkm_gr *base)
2065 {
2066 	struct gf100_gr *gr = gf100_gr(base);
2067 	struct nvkm_subdev *subdev = &base->engine.subdev;
2068 	struct nvkm_device *device = subdev->device;
2069 	bool reset = device->chipset == 0x137 || device->chipset == 0x138;
2070 	int ret;
2071 
2072 	/* On certain GP107/GP108 boards, we trigger a weird issue where
2073 	 * GR will stop responding to PRI accesses after we've asked the
2074 	 * SEC2 RTOS to boot the GR falcons.  This happens with far more
2075 	 * frequency when cold-booting a board (ie. returning from D3).
2076 	 *
2077 	 * The root cause for this is not known and has proven difficult
2078 	 * to isolate, with many avenues being dead-ends.
2079 	 *
2080 	 * A workaround was discovered by Karol, whereby putting GR into
2081 	 * reset for an extended period right before initialisation
2082 	 * prevents the problem from occuring.
2083 	 *
2084 	 * XXX: As RM does not require any such workaround, this is more
2085 	 *      of a hack than a true fix.
2086 	 */
2087 	reset = nvkm_boolopt(device->cfgopt, "NvGrResetWar", reset);
2088 	if (reset) {
2089 		nvkm_mask(device, 0x000200, 0x00001000, 0x00000000);
2090 		nvkm_rd32(device, 0x000200);
2091 		msleep(50);
2092 		nvkm_mask(device, 0x000200, 0x00001000, 0x00001000);
2093 		nvkm_rd32(device, 0x000200);
2094 	}
2095 
2096 	nvkm_pmu_pgob(gr->base.engine.subdev.device->pmu, false);
2097 
2098 	ret = nvkm_falcon_get(&gr->fecs.falcon, subdev);
2099 	if (ret)
2100 		return ret;
2101 
2102 	ret = nvkm_falcon_get(&gr->gpccs.falcon, subdev);
2103 	if (ret)
2104 		return ret;
2105 
2106 	ret = gr->func->init(gr);
2107 	if (ret)
2108 		return ret;
2109 
2110 	nvkm_inth_allow(&subdev->inth);
2111 	return 0;
2112 }
2113 
2114 static int
2115 gf100_gr_fini(struct nvkm_gr *base, bool suspend)
2116 {
2117 	struct gf100_gr *gr = gf100_gr(base);
2118 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
2119 
2120 	nvkm_inth_block(&subdev->inth);
2121 
2122 	nvkm_falcon_put(&gr->gpccs.falcon, subdev);
2123 	nvkm_falcon_put(&gr->fecs.falcon, subdev);
2124 	return 0;
2125 }
2126 
2127 static void *
2128 gf100_gr_dtor(struct nvkm_gr *base)
2129 {
2130 	struct gf100_gr *gr = gf100_gr(base);
2131 
2132 	kfree(gr->data);
2133 
2134 	nvkm_memory_unref(&gr->unknown);
2135 	nvkm_memory_unref(&gr->attrib_cb);
2136 	nvkm_memory_unref(&gr->bundle_cb);
2137 	nvkm_memory_unref(&gr->pagepool);
2138 
2139 	nvkm_falcon_dtor(&gr->gpccs.falcon);
2140 	nvkm_falcon_dtor(&gr->fecs.falcon);
2141 
2142 	nvkm_blob_dtor(&gr->fecs.inst);
2143 	nvkm_blob_dtor(&gr->fecs.data);
2144 	nvkm_blob_dtor(&gr->gpccs.inst);
2145 	nvkm_blob_dtor(&gr->gpccs.data);
2146 
2147 	vfree(gr->bundle64);
2148 	vfree(gr->bundle_veid);
2149 	vfree(gr->bundle);
2150 	vfree(gr->method);
2151 	vfree(gr->sw_ctx);
2152 	vfree(gr->sw_nonctx);
2153 	vfree(gr->sw_nonctx1);
2154 	vfree(gr->sw_nonctx2);
2155 	vfree(gr->sw_nonctx3);
2156 	vfree(gr->sw_nonctx4);
2157 
2158 	return gr;
2159 }
2160 
2161 static const struct nvkm_falcon_func
2162 gf100_gr_flcn = {
2163 	.load_imem = nvkm_falcon_v1_load_imem,
2164 	.load_dmem = nvkm_falcon_v1_load_dmem,
2165 	.start = nvkm_falcon_v1_start,
2166 };
2167 
2168 void
2169 gf100_gr_init_num_tpc_per_gpc(struct gf100_gr *gr, bool pd, bool ds)
2170 {
2171 	struct nvkm_device *device = gr->base.engine.subdev.device;
2172 	int gpc, i, j;
2173 	u32 data;
2174 
2175 	for (gpc = 0, i = 0; i < 4; i++) {
2176 		for (data = 0, j = 0; j < 8 && gpc < gr->gpc_nr; j++, gpc++)
2177 			data |= gr->tpc_nr[gpc] << (j * 4);
2178 		if (pd)
2179 			nvkm_wr32(device, 0x406028 + (i * 4), data);
2180 		if (ds)
2181 			nvkm_wr32(device, 0x405870 + (i * 4), data);
2182 	}
2183 }
2184 
2185 void
2186 gf100_gr_init_400054(struct gf100_gr *gr)
2187 {
2188 	nvkm_wr32(gr->base.engine.subdev.device, 0x400054, 0x34ce3464);
2189 }
2190 
2191 void
2192 gf100_gr_init_exception2(struct gf100_gr *gr)
2193 {
2194 	struct nvkm_device *device = gr->base.engine.subdev.device;
2195 
2196 	nvkm_wr32(device, 0x40011c, 0xffffffff);
2197 	nvkm_wr32(device, 0x400134, 0xffffffff);
2198 }
2199 
2200 void
2201 gf100_gr_init_rop_exceptions(struct gf100_gr *gr)
2202 {
2203 	struct nvkm_device *device = gr->base.engine.subdev.device;
2204 	int rop;
2205 
2206 	for (rop = 0; rop < gr->rop_nr; rop++) {
2207 		nvkm_wr32(device, ROP_UNIT(rop, 0x144), 0x40000000);
2208 		nvkm_wr32(device, ROP_UNIT(rop, 0x070), 0x40000000);
2209 		nvkm_wr32(device, ROP_UNIT(rop, 0x204), 0xffffffff);
2210 		nvkm_wr32(device, ROP_UNIT(rop, 0x208), 0xffffffff);
2211 	}
2212 }
2213 
2214 void
2215 gf100_gr_init_shader_exceptions(struct gf100_gr *gr, int gpc, int tpc)
2216 {
2217 	struct nvkm_device *device = gr->base.engine.subdev.device;
2218 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x644), 0x001ffffe);
2219 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x64c), 0x0000000f);
2220 }
2221 
2222 void
2223 gf100_gr_init_tex_hww_esr(struct gf100_gr *gr, int gpc, int tpc)
2224 {
2225 	struct nvkm_device *device = gr->base.engine.subdev.device;
2226 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x224), 0xc0000000);
2227 }
2228 
2229 void
2230 gf100_gr_init_419eb4(struct gf100_gr *gr)
2231 {
2232 	struct nvkm_device *device = gr->base.engine.subdev.device;
2233 	nvkm_mask(device, 0x419eb4, 0x00001000, 0x00001000);
2234 }
2235 
2236 void
2237 gf100_gr_init_419cc0(struct gf100_gr *gr)
2238 {
2239 	struct nvkm_device *device = gr->base.engine.subdev.device;
2240 	int gpc, tpc;
2241 
2242 	nvkm_mask(device, 0x419cc0, 0x00000008, 0x00000008);
2243 
2244 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
2245 		for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++)
2246 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x48c), 0xc0000000);
2247 	}
2248 }
2249 
2250 void
2251 gf100_gr_init_40601c(struct gf100_gr *gr)
2252 {
2253 	nvkm_wr32(gr->base.engine.subdev.device, 0x40601c, 0xc0000000);
2254 }
2255 
2256 void
2257 gf100_gr_init_fecs_exceptions(struct gf100_gr *gr)
2258 {
2259 	const u32 data = gr->firmware ? 0x000e0000 : 0x000e0001;
2260 	nvkm_wr32(gr->base.engine.subdev.device, 0x409c24, data);
2261 }
2262 
2263 void
2264 gf100_gr_init_gpc_mmu(struct gf100_gr *gr)
2265 {
2266 	struct nvkm_device *device = gr->base.engine.subdev.device;
2267 	struct nvkm_fb *fb = device->fb;
2268 
2269 	nvkm_wr32(device, 0x418880, nvkm_rd32(device, 0x100c80) & 0x00000001);
2270 	nvkm_wr32(device, 0x4188a4, 0x03000000);
2271 	nvkm_wr32(device, 0x418888, 0x00000000);
2272 	nvkm_wr32(device, 0x41888c, 0x00000000);
2273 	nvkm_wr32(device, 0x418890, 0x00000000);
2274 	nvkm_wr32(device, 0x418894, 0x00000000);
2275 	nvkm_wr32(device, 0x4188b4, nvkm_memory_addr(fb->mmu_wr) >> 8);
2276 	nvkm_wr32(device, 0x4188b8, nvkm_memory_addr(fb->mmu_rd) >> 8);
2277 }
2278 
2279 void
2280 gf100_gr_init_num_active_ltcs(struct gf100_gr *gr)
2281 {
2282 	struct nvkm_device *device = gr->base.engine.subdev.device;
2283 	nvkm_wr32(device, GPC_BCAST(0x08ac), nvkm_rd32(device, 0x100800));
2284 }
2285 
2286 void
2287 gf100_gr_init_zcull(struct gf100_gr *gr)
2288 {
2289 	struct nvkm_device *device = gr->base.engine.subdev.device;
2290 	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, gr->tpc_total);
2291 	const u8 tile_nr = ALIGN(gr->tpc_total, 32);
2292 	u8 bank[GPC_MAX] = {}, gpc, i, j;
2293 	u32 data;
2294 
2295 	for (i = 0; i < tile_nr; i += 8) {
2296 		for (data = 0, j = 0; j < 8 && i + j < gr->tpc_total; j++) {
2297 			data |= bank[gr->tile[i + j]] << (j * 4);
2298 			bank[gr->tile[i + j]]++;
2299 		}
2300 		nvkm_wr32(device, GPC_BCAST(0x0980 + ((i / 8) * 4)), data);
2301 	}
2302 
2303 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
2304 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0914),
2305 			  gr->screen_tile_row_offset << 8 | gr->tpc_nr[gpc]);
2306 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0910), 0x00040000 |
2307 							 gr->tpc_total);
2308 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0918), magicgpc918);
2309 	}
2310 
2311 	nvkm_wr32(device, GPC_BCAST(0x1bd4), magicgpc918);
2312 }
2313 
2314 void
2315 gf100_gr_init_vsc_stream_master(struct gf100_gr *gr)
2316 {
2317 	struct nvkm_device *device = gr->base.engine.subdev.device;
2318 	nvkm_mask(device, TPC_UNIT(0, 0, 0x05c), 0x00000001, 0x00000001);
2319 }
2320 
2321 static int
2322 gf100_gr_reset(struct nvkm_gr *base)
2323 {
2324 	struct nvkm_subdev *subdev = &base->engine.subdev;
2325 	struct nvkm_device *device = subdev->device;
2326 	struct gf100_gr *gr = gf100_gr(base);
2327 
2328 	nvkm_mask(device, 0x400500, 0x00000001, 0x00000000);
2329 
2330 	WARN_ON(gf100_gr_fecs_halt_pipeline(gr));
2331 
2332 	subdev->func->fini(subdev, false);
2333 	nvkm_mc_disable(device, subdev->type, subdev->inst);
2334 	if (gr->func->gpccs.reset)
2335 		gr->func->gpccs.reset(gr);
2336 
2337 	nvkm_mc_enable(device, subdev->type, subdev->inst);
2338 	return subdev->func->init(subdev);
2339 }
2340 
2341 int
2342 gf100_gr_init(struct gf100_gr *gr)
2343 {
2344 	struct nvkm_device *device = gr->base.engine.subdev.device;
2345 	int gpc, tpc;
2346 
2347 	nvkm_mask(device, 0x400500, 0x00010001, 0x00000000);
2348 
2349 	gr->func->init_gpc_mmu(gr);
2350 
2351 	if (gr->sw_nonctx1) {
2352 		gf100_gr_mmio(gr, gr->sw_nonctx1);
2353 		gf100_gr_mmio(gr, gr->sw_nonctx2);
2354 		gf100_gr_mmio(gr, gr->sw_nonctx3);
2355 		gf100_gr_mmio(gr, gr->sw_nonctx4);
2356 	} else
2357 	if (gr->sw_nonctx) {
2358 		gf100_gr_mmio(gr, gr->sw_nonctx);
2359 	} else {
2360 		gf100_gr_mmio(gr, gr->func->mmio);
2361 	}
2362 
2363 	gf100_gr_wait_idle(gr);
2364 
2365 	if (gr->func->init_r405a14)
2366 		gr->func->init_r405a14(gr);
2367 
2368 	if (gr->func->clkgate_pack)
2369 		nvkm_therm_clkgate_init(device->therm, gr->func->clkgate_pack);
2370 
2371 	if (gr->func->init_bios)
2372 		gr->func->init_bios(gr);
2373 
2374 	gr->func->init_vsc_stream_master(gr);
2375 	gr->func->init_zcull(gr);
2376 	gr->func->init_num_active_ltcs(gr);
2377 	if (gr->func->init_rop_active_fbps)
2378 		gr->func->init_rop_active_fbps(gr);
2379 	if (gr->func->init_bios_2)
2380 		gr->func->init_bios_2(gr);
2381 	if (gr->func->init_swdx_pes_mask)
2382 		gr->func->init_swdx_pes_mask(gr);
2383 	if (gr->func->init_fs)
2384 		gr->func->init_fs(gr);
2385 
2386 	nvkm_wr32(device, 0x400500, 0x00010001);
2387 
2388 	nvkm_wr32(device, 0x400100, 0xffffffff);
2389 	nvkm_wr32(device, 0x40013c, 0xffffffff);
2390 	nvkm_wr32(device, 0x400124, 0x00000002);
2391 
2392 	gr->func->init_fecs_exceptions(gr);
2393 
2394 	if (gr->func->init_40a790)
2395 		gr->func->init_40a790(gr);
2396 
2397 	if (gr->func->init_ds_hww_esr_2)
2398 		gr->func->init_ds_hww_esr_2(gr);
2399 
2400 	nvkm_wr32(device, 0x404000, 0xc0000000);
2401 	nvkm_wr32(device, 0x404600, 0xc0000000);
2402 	nvkm_wr32(device, 0x408030, 0xc0000000);
2403 
2404 	if (gr->func->init_40601c)
2405 		gr->func->init_40601c(gr);
2406 
2407 	nvkm_wr32(device, 0x406018, 0xc0000000);
2408 	nvkm_wr32(device, 0x404490, 0xc0000000);
2409 
2410 	if (gr->func->init_sked_hww_esr)
2411 		gr->func->init_sked_hww_esr(gr);
2412 
2413 	nvkm_wr32(device, 0x405840, 0xc0000000);
2414 	nvkm_wr32(device, 0x405844, 0x00ffffff);
2415 
2416 	if (gr->func->init_419cc0)
2417 		gr->func->init_419cc0(gr);
2418 	if (gr->func->init_419eb4)
2419 		gr->func->init_419eb4(gr);
2420 	if (gr->func->init_419c9c)
2421 		gr->func->init_419c9c(gr);
2422 
2423 	if (gr->func->init_ppc_exceptions)
2424 		gr->func->init_ppc_exceptions(gr);
2425 
2426 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
2427 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0420), 0xc0000000);
2428 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0900), 0xc0000000);
2429 		nvkm_wr32(device, GPC_UNIT(gpc, 0x1028), 0xc0000000);
2430 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0824), 0xc0000000);
2431 		for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++) {
2432 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x508), 0xffffffff);
2433 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x50c), 0xffffffff);
2434 			if (gr->func->init_tex_hww_esr)
2435 				gr->func->init_tex_hww_esr(gr, gpc, tpc);
2436 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x084), 0xc0000000);
2437 			if (gr->func->init_504430)
2438 				gr->func->init_504430(gr, gpc, tpc);
2439 			gr->func->init_shader_exceptions(gr, gpc, tpc);
2440 		}
2441 		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c90), 0xffffffff);
2442 		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c94), 0xffffffff);
2443 	}
2444 
2445 	gr->func->init_rop_exceptions(gr);
2446 
2447 	nvkm_wr32(device, 0x400108, 0xffffffff);
2448 	nvkm_wr32(device, 0x400138, 0xffffffff);
2449 	nvkm_wr32(device, 0x400118, 0xffffffff);
2450 	nvkm_wr32(device, 0x400130, 0xffffffff);
2451 	if (gr->func->init_exception2)
2452 		gr->func->init_exception2(gr);
2453 
2454 	if (gr->func->init_400054)
2455 		gr->func->init_400054(gr);
2456 
2457 	gf100_gr_zbc_init(gr);
2458 
2459 	if (gr->func->init_4188a4)
2460 		gr->func->init_4188a4(gr);
2461 
2462 	return gf100_gr_init_ctxctl(gr);
2463 }
2464 
2465 void
2466 gf100_gr_fecs_reset(struct gf100_gr *gr)
2467 {
2468 	struct nvkm_device *device = gr->base.engine.subdev.device;
2469 
2470 	nvkm_wr32(device, 0x409614, 0x00000070);
2471 	nvkm_usec(device, 10, NVKM_DELAY);
2472 	nvkm_mask(device, 0x409614, 0x00000700, 0x00000700);
2473 	nvkm_usec(device, 10, NVKM_DELAY);
2474 	nvkm_rd32(device, 0x409614);
2475 }
2476 
2477 #include "fuc/hubgf100.fuc3.h"
2478 
2479 struct gf100_gr_ucode
2480 gf100_gr_fecs_ucode = {
2481 	.code.data = gf100_grhub_code,
2482 	.code.size = sizeof(gf100_grhub_code),
2483 	.data.data = gf100_grhub_data,
2484 	.data.size = sizeof(gf100_grhub_data),
2485 };
2486 
2487 #include "fuc/gpcgf100.fuc3.h"
2488 
2489 struct gf100_gr_ucode
2490 gf100_gr_gpccs_ucode = {
2491 	.code.data = gf100_grgpc_code,
2492 	.code.size = sizeof(gf100_grgpc_code),
2493 	.data.data = gf100_grgpc_data,
2494 	.data.size = sizeof(gf100_grgpc_data),
2495 };
2496 
2497 static int
2498 gf100_gr_nonstall(struct nvkm_gr *base)
2499 {
2500 	struct gf100_gr *gr = gf100_gr(base);
2501 
2502 	if (gr->func->nonstall)
2503 		return gr->func->nonstall(gr);
2504 
2505 	return -EINVAL;
2506 }
2507 
2508 static const struct nvkm_gr_func
2509 gf100_gr_ = {
2510 	.dtor = gf100_gr_dtor,
2511 	.oneinit = gf100_gr_oneinit,
2512 	.init = gf100_gr_init_,
2513 	.fini = gf100_gr_fini,
2514 	.nonstall = gf100_gr_nonstall,
2515 	.reset = gf100_gr_reset,
2516 	.units = gf100_gr_units,
2517 	.chan_new = gf100_gr_chan_new,
2518 	.object_get = gf100_gr_object_get,
2519 	.chsw_load = gf100_gr_chsw_load,
2520 	.ctxsw.pause = gf100_gr_fecs_stop_ctxsw,
2521 	.ctxsw.resume = gf100_gr_fecs_start_ctxsw,
2522 	.ctxsw.inst = gf100_gr_ctxsw_inst,
2523 };
2524 
2525 static const struct gf100_gr_func
2526 gf100_gr = {
2527 	.oneinit_tiles = gf100_gr_oneinit_tiles,
2528 	.oneinit_sm_id = gf100_gr_oneinit_sm_id,
2529 	.init = gf100_gr_init,
2530 	.init_gpc_mmu = gf100_gr_init_gpc_mmu,
2531 	.init_vsc_stream_master = gf100_gr_init_vsc_stream_master,
2532 	.init_zcull = gf100_gr_init_zcull,
2533 	.init_num_active_ltcs = gf100_gr_init_num_active_ltcs,
2534 	.init_fecs_exceptions = gf100_gr_init_fecs_exceptions,
2535 	.init_40601c = gf100_gr_init_40601c,
2536 	.init_419cc0 = gf100_gr_init_419cc0,
2537 	.init_419eb4 = gf100_gr_init_419eb4,
2538 	.init_tex_hww_esr = gf100_gr_init_tex_hww_esr,
2539 	.init_shader_exceptions = gf100_gr_init_shader_exceptions,
2540 	.init_rop_exceptions = gf100_gr_init_rop_exceptions,
2541 	.init_exception2 = gf100_gr_init_exception2,
2542 	.init_400054 = gf100_gr_init_400054,
2543 	.trap_mp = gf100_gr_trap_mp,
2544 	.mmio = gf100_gr_pack_mmio,
2545 	.fecs.ucode = &gf100_gr_fecs_ucode,
2546 	.fecs.reset = gf100_gr_fecs_reset,
2547 	.gpccs.ucode = &gf100_gr_gpccs_ucode,
2548 	.rops = gf100_gr_rops,
2549 	.grctx = &gf100_grctx,
2550 	.zbc = &gf100_gr_zbc,
2551 	.sclass = {
2552 		{ -1, -1, FERMI_TWOD_A },
2553 		{ -1, -1, FERMI_MEMORY_TO_MEMORY_FORMAT_A },
2554 		{ -1, -1, FERMI_A, &gf100_fermi },
2555 		{ -1, -1, FERMI_COMPUTE_A },
2556 		{}
2557 	}
2558 };
2559 
2560 int
2561 gf100_gr_nofw(struct gf100_gr *gr, int ver, const struct gf100_gr_fwif *fwif)
2562 {
2563 	gr->firmware = false;
2564 	return 0;
2565 }
2566 
2567 static int
2568 gf100_gr_load_fw(struct gf100_gr *gr, const char *name,
2569 		 struct nvkm_blob *blob)
2570 {
2571 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
2572 	struct nvkm_device *device = subdev->device;
2573 	const struct firmware *fw;
2574 	char f[32];
2575 	int ret;
2576 
2577 	snprintf(f, sizeof(f), "nouveau/nv%02x_%s", device->chipset, name);
2578 	ret = request_firmware(&fw, f, device->dev);
2579 	if (ret) {
2580 		snprintf(f, sizeof(f), "nouveau/%s", name);
2581 		ret = request_firmware(&fw, f, device->dev);
2582 		if (ret) {
2583 			nvkm_error(subdev, "failed to load %s\n", name);
2584 			return ret;
2585 		}
2586 	}
2587 
2588 	blob->size = fw->size;
2589 	blob->data = kmemdup(fw->data, blob->size, GFP_KERNEL);
2590 	release_firmware(fw);
2591 	return (blob->data != NULL) ? 0 : -ENOMEM;
2592 }
2593 
2594 int
2595 gf100_gr_load(struct gf100_gr *gr, int ver, const struct gf100_gr_fwif *fwif)
2596 {
2597 	struct nvkm_device *device = gr->base.engine.subdev.device;
2598 
2599 	if (!nvkm_boolopt(device->cfgopt, "NvGrUseFW", false))
2600 		return -EINVAL;
2601 
2602 	if (gf100_gr_load_fw(gr, "fuc409c", &gr->fecs.inst) ||
2603 	    gf100_gr_load_fw(gr, "fuc409d", &gr->fecs.data) ||
2604 	    gf100_gr_load_fw(gr, "fuc41ac", &gr->gpccs.inst) ||
2605 	    gf100_gr_load_fw(gr, "fuc41ad", &gr->gpccs.data))
2606 		return -ENOENT;
2607 
2608 	gr->firmware = true;
2609 	return 0;
2610 }
2611 
2612 static const struct gf100_gr_fwif
2613 gf100_gr_fwif[] = {
2614 	{ -1, gf100_gr_load, &gf100_gr },
2615 	{ -1, gf100_gr_nofw, &gf100_gr },
2616 	{}
2617 };
2618 
2619 int
2620 gf100_gr_new_(const struct gf100_gr_fwif *fwif, struct nvkm_device *device,
2621 	      enum nvkm_subdev_type type, int inst, struct nvkm_gr **pgr)
2622 {
2623 	struct gf100_gr *gr;
2624 	int ret;
2625 
2626 	if (!(gr = kzalloc(sizeof(*gr), GFP_KERNEL)))
2627 		return -ENOMEM;
2628 	*pgr = &gr->base;
2629 
2630 	ret = nvkm_gr_ctor(&gf100_gr_, device, type, inst, true, &gr->base);
2631 	if (ret)
2632 		return ret;
2633 
2634 	fwif = nvkm_firmware_load(&gr->base.engine.subdev, fwif, "Gr", gr);
2635 	if (IS_ERR(fwif))
2636 		return PTR_ERR(fwif);
2637 
2638 	gr->func = fwif->func;
2639 
2640 	ret = nvkm_falcon_ctor(&gf100_gr_flcn, &gr->base.engine.subdev,
2641 			       "fecs", 0x409000, &gr->fecs.falcon);
2642 	if (ret)
2643 		return ret;
2644 
2645 	mutex_init(&gr->fecs.mutex);
2646 
2647 	ret = nvkm_falcon_ctor(&gf100_gr_flcn, &gr->base.engine.subdev,
2648 			       "gpccs", 0x41a000, &gr->gpccs.falcon);
2649 	if (ret)
2650 		return ret;
2651 
2652 	return 0;
2653 }
2654 
2655 int
2656 gf100_gr_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst, struct nvkm_gr **pgr)
2657 {
2658 	return gf100_gr_new_(gf100_gr_fwif, device, type, inst, pgr);
2659 }
2660