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/handle.h>
30 #include <core/option.h>
31 #include <engine/fifo.h>
32 #include <subdev/fb.h>
33 #include <subdev/mc.h>
34 #include <subdev/timer.h>
35 
36 #include <nvif/class.h>
37 #include <nvif/unpack.h>
38 
39 /*******************************************************************************
40  * Zero Bandwidth Clear
41  ******************************************************************************/
42 
43 static void
44 gf100_gr_zbc_clear_color(struct gf100_gr *gr, int zbc)
45 {
46 	struct nvkm_device *device = gr->base.engine.subdev.device;
47 	if (gr->zbc_color[zbc].format) {
48 		nvkm_wr32(device, 0x405804, gr->zbc_color[zbc].ds[0]);
49 		nvkm_wr32(device, 0x405808, gr->zbc_color[zbc].ds[1]);
50 		nvkm_wr32(device, 0x40580c, gr->zbc_color[zbc].ds[2]);
51 		nvkm_wr32(device, 0x405810, gr->zbc_color[zbc].ds[3]);
52 	}
53 	nvkm_wr32(device, 0x405814, gr->zbc_color[zbc].format);
54 	nvkm_wr32(device, 0x405820, zbc);
55 	nvkm_wr32(device, 0x405824, 0x00000004); /* TRIGGER | WRITE | COLOR */
56 }
57 
58 static int
59 gf100_gr_zbc_color_get(struct gf100_gr *gr, int format,
60 		       const u32 ds[4], const u32 l2[4])
61 {
62 	struct nvkm_ltc *ltc = nvkm_ltc(gr);
63 	int zbc = -ENOSPC, i;
64 
65 	for (i = ltc->zbc_min; i <= ltc->zbc_max; i++) {
66 		if (gr->zbc_color[i].format) {
67 			if (gr->zbc_color[i].format != format)
68 				continue;
69 			if (memcmp(gr->zbc_color[i].ds, ds, sizeof(
70 				   gr->zbc_color[i].ds)))
71 				continue;
72 			if (memcmp(gr->zbc_color[i].l2, l2, sizeof(
73 				   gr->zbc_color[i].l2))) {
74 				WARN_ON(1);
75 				return -EINVAL;
76 			}
77 			return i;
78 		} else {
79 			zbc = (zbc < 0) ? i : zbc;
80 		}
81 	}
82 
83 	if (zbc < 0)
84 		return zbc;
85 
86 	memcpy(gr->zbc_color[zbc].ds, ds, sizeof(gr->zbc_color[zbc].ds));
87 	memcpy(gr->zbc_color[zbc].l2, l2, sizeof(gr->zbc_color[zbc].l2));
88 	gr->zbc_color[zbc].format = format;
89 	ltc->zbc_color_get(ltc, zbc, l2);
90 	gf100_gr_zbc_clear_color(gr, zbc);
91 	return zbc;
92 }
93 
94 static void
95 gf100_gr_zbc_clear_depth(struct gf100_gr *gr, int zbc)
96 {
97 	struct nvkm_device *device = gr->base.engine.subdev.device;
98 	if (gr->zbc_depth[zbc].format)
99 		nvkm_wr32(device, 0x405818, gr->zbc_depth[zbc].ds);
100 	nvkm_wr32(device, 0x40581c, gr->zbc_depth[zbc].format);
101 	nvkm_wr32(device, 0x405820, zbc);
102 	nvkm_wr32(device, 0x405824, 0x00000005); /* TRIGGER | WRITE | DEPTH */
103 }
104 
105 static int
106 gf100_gr_zbc_depth_get(struct gf100_gr *gr, int format,
107 		       const u32 ds, const u32 l2)
108 {
109 	struct nvkm_ltc *ltc = nvkm_ltc(gr);
110 	int zbc = -ENOSPC, i;
111 
112 	for (i = ltc->zbc_min; i <= ltc->zbc_max; i++) {
113 		if (gr->zbc_depth[i].format) {
114 			if (gr->zbc_depth[i].format != format)
115 				continue;
116 			if (gr->zbc_depth[i].ds != ds)
117 				continue;
118 			if (gr->zbc_depth[i].l2 != l2) {
119 				WARN_ON(1);
120 				return -EINVAL;
121 			}
122 			return i;
123 		} else {
124 			zbc = (zbc < 0) ? i : zbc;
125 		}
126 	}
127 
128 	if (zbc < 0)
129 		return zbc;
130 
131 	gr->zbc_depth[zbc].format = format;
132 	gr->zbc_depth[zbc].ds = ds;
133 	gr->zbc_depth[zbc].l2 = l2;
134 	ltc->zbc_depth_get(ltc, zbc, l2);
135 	gf100_gr_zbc_clear_depth(gr, zbc);
136 	return zbc;
137 }
138 
139 /*******************************************************************************
140  * Graphics object classes
141  ******************************************************************************/
142 
143 static int
144 gf100_fermi_mthd_zbc_color(struct nvkm_object *object, void *data, u32 size)
145 {
146 	struct gf100_gr *gr = (void *)object->engine;
147 	union {
148 		struct fermi_a_zbc_color_v0 v0;
149 	} *args = data;
150 	int ret;
151 
152 	if (nvif_unpack(args->v0, 0, 0, false)) {
153 		switch (args->v0.format) {
154 		case FERMI_A_ZBC_COLOR_V0_FMT_ZERO:
155 		case FERMI_A_ZBC_COLOR_V0_FMT_UNORM_ONE:
156 		case FERMI_A_ZBC_COLOR_V0_FMT_RF32_GF32_BF32_AF32:
157 		case FERMI_A_ZBC_COLOR_V0_FMT_R16_G16_B16_A16:
158 		case FERMI_A_ZBC_COLOR_V0_FMT_RN16_GN16_BN16_AN16:
159 		case FERMI_A_ZBC_COLOR_V0_FMT_RS16_GS16_BS16_AS16:
160 		case FERMI_A_ZBC_COLOR_V0_FMT_RU16_GU16_BU16_AU16:
161 		case FERMI_A_ZBC_COLOR_V0_FMT_RF16_GF16_BF16_AF16:
162 		case FERMI_A_ZBC_COLOR_V0_FMT_A8R8G8B8:
163 		case FERMI_A_ZBC_COLOR_V0_FMT_A8RL8GL8BL8:
164 		case FERMI_A_ZBC_COLOR_V0_FMT_A2B10G10R10:
165 		case FERMI_A_ZBC_COLOR_V0_FMT_AU2BU10GU10RU10:
166 		case FERMI_A_ZBC_COLOR_V0_FMT_A8B8G8R8:
167 		case FERMI_A_ZBC_COLOR_V0_FMT_A8BL8GL8RL8:
168 		case FERMI_A_ZBC_COLOR_V0_FMT_AN8BN8GN8RN8:
169 		case FERMI_A_ZBC_COLOR_V0_FMT_AS8BS8GS8RS8:
170 		case FERMI_A_ZBC_COLOR_V0_FMT_AU8BU8GU8RU8:
171 		case FERMI_A_ZBC_COLOR_V0_FMT_A2R10G10B10:
172 		case FERMI_A_ZBC_COLOR_V0_FMT_BF10GF11RF11:
173 			ret = gf100_gr_zbc_color_get(gr, args->v0.format,
174 							   args->v0.ds,
175 							   args->v0.l2);
176 			if (ret >= 0) {
177 				args->v0.index = ret;
178 				return 0;
179 			}
180 			break;
181 		default:
182 			return -EINVAL;
183 		}
184 	}
185 
186 	return ret;
187 }
188 
189 static int
190 gf100_fermi_mthd_zbc_depth(struct nvkm_object *object, void *data, u32 size)
191 {
192 	struct gf100_gr *gr = (void *)object->engine;
193 	union {
194 		struct fermi_a_zbc_depth_v0 v0;
195 	} *args = data;
196 	int ret;
197 
198 	if (nvif_unpack(args->v0, 0, 0, false)) {
199 		switch (args->v0.format) {
200 		case FERMI_A_ZBC_DEPTH_V0_FMT_FP32:
201 			ret = gf100_gr_zbc_depth_get(gr, args->v0.format,
202 							   args->v0.ds,
203 							   args->v0.l2);
204 			return (ret >= 0) ? 0 : -ENOSPC;
205 		default:
206 			return -EINVAL;
207 		}
208 	}
209 
210 	return ret;
211 }
212 
213 static int
214 gf100_fermi_mthd(struct nvkm_object *object, u32 mthd, void *data, u32 size)
215 {
216 	switch (mthd) {
217 	case FERMI_A_ZBC_COLOR:
218 		return gf100_fermi_mthd_zbc_color(object, data, size);
219 	case FERMI_A_ZBC_DEPTH:
220 		return gf100_fermi_mthd_zbc_depth(object, data, size);
221 	default:
222 		break;
223 	}
224 	return -EINVAL;
225 }
226 
227 struct nvkm_ofuncs
228 gf100_fermi_ofuncs = {
229 	.ctor = _nvkm_object_ctor,
230 	.dtor = nvkm_object_destroy,
231 	.init = nvkm_object_init,
232 	.fini = nvkm_object_fini,
233 	.mthd = gf100_fermi_mthd,
234 };
235 
236 static int
237 gf100_gr_set_shader_exceptions(struct nvkm_object *object, u32 mthd,
238 			       void *pdata, u32 size)
239 {
240 	struct gf100_gr *gr = (void *)object->engine;
241 	struct nvkm_device *device = gr->base.engine.subdev.device;
242 	if (size >= sizeof(u32)) {
243 		u32 data = *(u32 *)pdata ? 0xffffffff : 0x00000000;
244 		nvkm_wr32(device, 0x419e44, data);
245 		nvkm_wr32(device, 0x419e4c, data);
246 		return 0;
247 	}
248 	return -EINVAL;
249 }
250 
251 struct nvkm_omthds
252 gf100_gr_9097_omthds[] = {
253 	{ 0x1528, 0x1528, gf100_gr_set_shader_exceptions },
254 	{}
255 };
256 
257 struct nvkm_omthds
258 gf100_gr_90c0_omthds[] = {
259 	{ 0x1528, 0x1528, gf100_gr_set_shader_exceptions },
260 	{}
261 };
262 
263 struct nvkm_oclass
264 gf100_gr_sclass[] = {
265 	{ FERMI_TWOD_A, &nvkm_object_ofuncs },
266 	{ FERMI_MEMORY_TO_MEMORY_FORMAT_A, &nvkm_object_ofuncs },
267 	{ FERMI_A, &gf100_fermi_ofuncs, gf100_gr_9097_omthds },
268 	{ FERMI_COMPUTE_A, &nvkm_object_ofuncs, gf100_gr_90c0_omthds },
269 	{}
270 };
271 
272 /*******************************************************************************
273  * PGRAPH context
274  ******************************************************************************/
275 
276 int
277 gf100_gr_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
278 		      struct nvkm_oclass *oclass, void *args, u32 size,
279 		      struct nvkm_object **pobject)
280 {
281 	struct nvkm_vm *vm = nvkm_client(parent)->vm;
282 	struct gf100_gr *gr = (void *)engine;
283 	struct gf100_gr_data *data = gr->mmio_data;
284 	struct gf100_gr_mmio *mmio = gr->mmio_list;
285 	struct gf100_gr_chan *chan;
286 	struct nvkm_gpuobj *image;
287 	int ret, i;
288 
289 	/* allocate memory for context, and fill with default values */
290 	ret = nvkm_gr_context_create(parent, engine, oclass, NULL,
291 				     gr->size, 0x100,
292 				     NVOBJ_FLAG_ZERO_ALLOC, &chan);
293 	*pobject = nv_object(chan);
294 	if (ret)
295 		return ret;
296 
297 	/* allocate memory for a "mmio list" buffer that's used by the HUB
298 	 * fuc to modify some per-context register settings on first load
299 	 * of the context.
300 	 */
301 	ret = nvkm_gpuobj_new(nv_object(chan), NULL, 0x1000, 0x100, 0,
302 			      &chan->mmio);
303 	if (ret)
304 		return ret;
305 
306 	ret = nvkm_gpuobj_map_vm(nv_gpuobj(chan->mmio), vm,
307 				 NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS,
308 				 &chan->mmio_vma);
309 	if (ret)
310 		return ret;
311 
312 	/* allocate buffers referenced by mmio list */
313 	for (i = 0; data->size && i < ARRAY_SIZE(gr->mmio_data); i++) {
314 		ret = nvkm_gpuobj_new(nv_object(chan), NULL, data->size,
315 				      data->align, 0, &chan->data[i].mem);
316 		if (ret)
317 			return ret;
318 
319 		ret = nvkm_gpuobj_map_vm(chan->data[i].mem, vm, data->access,
320 					 &chan->data[i].vma);
321 		if (ret)
322 			return ret;
323 
324 		data++;
325 	}
326 
327 	/* finally, fill in the mmio list and point the context at it */
328 	nvkm_kmap(chan->mmio);
329 	for (i = 0; mmio->addr && i < ARRAY_SIZE(gr->mmio_list); i++) {
330 		u32 addr = mmio->addr;
331 		u32 data = mmio->data;
332 
333 		if (mmio->buffer >= 0) {
334 			u64 info = chan->data[mmio->buffer].vma.offset;
335 			data |= info >> mmio->shift;
336 		}
337 
338 		nvkm_wo32(chan->mmio, chan->mmio_nr++ * 4, addr);
339 		nvkm_wo32(chan->mmio, chan->mmio_nr++ * 4, data);
340 		mmio++;
341 	}
342 	nvkm_done(chan->mmio);
343 
344 	image = &chan->base.base.gpuobj;
345 	nvkm_kmap(image);
346 	for (i = 0; i < gr->size; i += 4)
347 		nvkm_wo32(image, i, gr->data[i / 4]);
348 
349 	if (!gr->firmware) {
350 		nvkm_wo32(image, 0x00, chan->mmio_nr / 2);
351 		nvkm_wo32(image, 0x04, chan->mmio_vma.offset >> 8);
352 	} else {
353 		nvkm_wo32(image, 0xf4, 0);
354 		nvkm_wo32(image, 0xf8, 0);
355 		nvkm_wo32(image, 0x10, chan->mmio_nr / 2);
356 		nvkm_wo32(image, 0x14, lower_32_bits(chan->mmio_vma.offset));
357 		nvkm_wo32(image, 0x18, upper_32_bits(chan->mmio_vma.offset));
358 		nvkm_wo32(image, 0x1c, 1);
359 		nvkm_wo32(image, 0x20, 0);
360 		nvkm_wo32(image, 0x28, 0);
361 		nvkm_wo32(image, 0x2c, 0);
362 	}
363 	nvkm_done(image);
364 
365 	return 0;
366 }
367 
368 void
369 gf100_gr_context_dtor(struct nvkm_object *object)
370 {
371 	struct gf100_gr_chan *chan = (void *)object;
372 	int i;
373 
374 	for (i = 0; i < ARRAY_SIZE(chan->data); i++) {
375 		nvkm_gpuobj_unmap(&chan->data[i].vma);
376 		nvkm_gpuobj_ref(NULL, &chan->data[i].mem);
377 	}
378 
379 	nvkm_gpuobj_unmap(&chan->mmio_vma);
380 	nvkm_gpuobj_ref(NULL, &chan->mmio);
381 
382 	nvkm_gr_context_destroy(&chan->base);
383 }
384 
385 /*******************************************************************************
386  * PGRAPH register lists
387  ******************************************************************************/
388 
389 const struct gf100_gr_init
390 gf100_gr_init_main_0[] = {
391 	{ 0x400080,   1, 0x04, 0x003083c2 },
392 	{ 0x400088,   1, 0x04, 0x00006fe7 },
393 	{ 0x40008c,   1, 0x04, 0x00000000 },
394 	{ 0x400090,   1, 0x04, 0x00000030 },
395 	{ 0x40013c,   1, 0x04, 0x013901f7 },
396 	{ 0x400140,   1, 0x04, 0x00000100 },
397 	{ 0x400144,   1, 0x04, 0x00000000 },
398 	{ 0x400148,   1, 0x04, 0x00000110 },
399 	{ 0x400138,   1, 0x04, 0x00000000 },
400 	{ 0x400130,   2, 0x04, 0x00000000 },
401 	{ 0x400124,   1, 0x04, 0x00000002 },
402 	{}
403 };
404 
405 const struct gf100_gr_init
406 gf100_gr_init_fe_0[] = {
407 	{ 0x40415c,   1, 0x04, 0x00000000 },
408 	{ 0x404170,   1, 0x04, 0x00000000 },
409 	{}
410 };
411 
412 const struct gf100_gr_init
413 gf100_gr_init_pri_0[] = {
414 	{ 0x404488,   2, 0x04, 0x00000000 },
415 	{}
416 };
417 
418 const struct gf100_gr_init
419 gf100_gr_init_rstr2d_0[] = {
420 	{ 0x407808,   1, 0x04, 0x00000000 },
421 	{}
422 };
423 
424 const struct gf100_gr_init
425 gf100_gr_init_pd_0[] = {
426 	{ 0x406024,   1, 0x04, 0x00000000 },
427 	{}
428 };
429 
430 const struct gf100_gr_init
431 gf100_gr_init_ds_0[] = {
432 	{ 0x405844,   1, 0x04, 0x00ffffff },
433 	{ 0x405850,   1, 0x04, 0x00000000 },
434 	{ 0x405908,   1, 0x04, 0x00000000 },
435 	{}
436 };
437 
438 const struct gf100_gr_init
439 gf100_gr_init_scc_0[] = {
440 	{ 0x40803c,   1, 0x04, 0x00000000 },
441 	{}
442 };
443 
444 const struct gf100_gr_init
445 gf100_gr_init_prop_0[] = {
446 	{ 0x4184a0,   1, 0x04, 0x00000000 },
447 	{}
448 };
449 
450 const struct gf100_gr_init
451 gf100_gr_init_gpc_unk_0[] = {
452 	{ 0x418604,   1, 0x04, 0x00000000 },
453 	{ 0x418680,   1, 0x04, 0x00000000 },
454 	{ 0x418714,   1, 0x04, 0x80000000 },
455 	{ 0x418384,   1, 0x04, 0x00000000 },
456 	{}
457 };
458 
459 const struct gf100_gr_init
460 gf100_gr_init_setup_0[] = {
461 	{ 0x418814,   3, 0x04, 0x00000000 },
462 	{}
463 };
464 
465 const struct gf100_gr_init
466 gf100_gr_init_crstr_0[] = {
467 	{ 0x418b04,   1, 0x04, 0x00000000 },
468 	{}
469 };
470 
471 const struct gf100_gr_init
472 gf100_gr_init_setup_1[] = {
473 	{ 0x4188c8,   1, 0x04, 0x80000000 },
474 	{ 0x4188cc,   1, 0x04, 0x00000000 },
475 	{ 0x4188d0,   1, 0x04, 0x00010000 },
476 	{ 0x4188d4,   1, 0x04, 0x00000001 },
477 	{}
478 };
479 
480 const struct gf100_gr_init
481 gf100_gr_init_zcull_0[] = {
482 	{ 0x418910,   1, 0x04, 0x00010001 },
483 	{ 0x418914,   1, 0x04, 0x00000301 },
484 	{ 0x418918,   1, 0x04, 0x00800000 },
485 	{ 0x418980,   1, 0x04, 0x77777770 },
486 	{ 0x418984,   3, 0x04, 0x77777777 },
487 	{}
488 };
489 
490 const struct gf100_gr_init
491 gf100_gr_init_gpm_0[] = {
492 	{ 0x418c04,   1, 0x04, 0x00000000 },
493 	{ 0x418c88,   1, 0x04, 0x00000000 },
494 	{}
495 };
496 
497 const struct gf100_gr_init
498 gf100_gr_init_gpc_unk_1[] = {
499 	{ 0x418d00,   1, 0x04, 0x00000000 },
500 	{ 0x418f08,   1, 0x04, 0x00000000 },
501 	{ 0x418e00,   1, 0x04, 0x00000050 },
502 	{ 0x418e08,   1, 0x04, 0x00000000 },
503 	{}
504 };
505 
506 const struct gf100_gr_init
507 gf100_gr_init_gcc_0[] = {
508 	{ 0x41900c,   1, 0x04, 0x00000000 },
509 	{ 0x419018,   1, 0x04, 0x00000000 },
510 	{}
511 };
512 
513 const struct gf100_gr_init
514 gf100_gr_init_tpccs_0[] = {
515 	{ 0x419d08,   2, 0x04, 0x00000000 },
516 	{ 0x419d10,   1, 0x04, 0x00000014 },
517 	{}
518 };
519 
520 const struct gf100_gr_init
521 gf100_gr_init_tex_0[] = {
522 	{ 0x419ab0,   1, 0x04, 0x00000000 },
523 	{ 0x419ab8,   1, 0x04, 0x000000e7 },
524 	{ 0x419abc,   2, 0x04, 0x00000000 },
525 	{}
526 };
527 
528 const struct gf100_gr_init
529 gf100_gr_init_pe_0[] = {
530 	{ 0x41980c,   3, 0x04, 0x00000000 },
531 	{ 0x419844,   1, 0x04, 0x00000000 },
532 	{ 0x41984c,   1, 0x04, 0x00005bc5 },
533 	{ 0x419850,   4, 0x04, 0x00000000 },
534 	{}
535 };
536 
537 const struct gf100_gr_init
538 gf100_gr_init_l1c_0[] = {
539 	{ 0x419c98,   1, 0x04, 0x00000000 },
540 	{ 0x419ca8,   1, 0x04, 0x80000000 },
541 	{ 0x419cb4,   1, 0x04, 0x00000000 },
542 	{ 0x419cb8,   1, 0x04, 0x00008bf4 },
543 	{ 0x419cbc,   1, 0x04, 0x28137606 },
544 	{ 0x419cc0,   2, 0x04, 0x00000000 },
545 	{}
546 };
547 
548 const struct gf100_gr_init
549 gf100_gr_init_wwdx_0[] = {
550 	{ 0x419bd4,   1, 0x04, 0x00800000 },
551 	{ 0x419bdc,   1, 0x04, 0x00000000 },
552 	{}
553 };
554 
555 const struct gf100_gr_init
556 gf100_gr_init_tpccs_1[] = {
557 	{ 0x419d2c,   1, 0x04, 0x00000000 },
558 	{}
559 };
560 
561 const struct gf100_gr_init
562 gf100_gr_init_mpc_0[] = {
563 	{ 0x419c0c,   1, 0x04, 0x00000000 },
564 	{}
565 };
566 
567 static const struct gf100_gr_init
568 gf100_gr_init_sm_0[] = {
569 	{ 0x419e00,   1, 0x04, 0x00000000 },
570 	{ 0x419ea0,   1, 0x04, 0x00000000 },
571 	{ 0x419ea4,   1, 0x04, 0x00000100 },
572 	{ 0x419ea8,   1, 0x04, 0x00001100 },
573 	{ 0x419eac,   1, 0x04, 0x11100702 },
574 	{ 0x419eb0,   1, 0x04, 0x00000003 },
575 	{ 0x419eb4,   4, 0x04, 0x00000000 },
576 	{ 0x419ec8,   1, 0x04, 0x06060618 },
577 	{ 0x419ed0,   1, 0x04, 0x0eff0e38 },
578 	{ 0x419ed4,   1, 0x04, 0x011104f1 },
579 	{ 0x419edc,   1, 0x04, 0x00000000 },
580 	{ 0x419f00,   1, 0x04, 0x00000000 },
581 	{ 0x419f2c,   1, 0x04, 0x00000000 },
582 	{}
583 };
584 
585 const struct gf100_gr_init
586 gf100_gr_init_be_0[] = {
587 	{ 0x40880c,   1, 0x04, 0x00000000 },
588 	{ 0x408910,   9, 0x04, 0x00000000 },
589 	{ 0x408950,   1, 0x04, 0x00000000 },
590 	{ 0x408954,   1, 0x04, 0x0000ffff },
591 	{ 0x408984,   1, 0x04, 0x00000000 },
592 	{ 0x408988,   1, 0x04, 0x08040201 },
593 	{ 0x40898c,   1, 0x04, 0x80402010 },
594 	{}
595 };
596 
597 const struct gf100_gr_init
598 gf100_gr_init_fe_1[] = {
599 	{ 0x4040f0,   1, 0x04, 0x00000000 },
600 	{}
601 };
602 
603 const struct gf100_gr_init
604 gf100_gr_init_pe_1[] = {
605 	{ 0x419880,   1, 0x04, 0x00000002 },
606 	{}
607 };
608 
609 static const struct gf100_gr_pack
610 gf100_gr_pack_mmio[] = {
611 	{ gf100_gr_init_main_0 },
612 	{ gf100_gr_init_fe_0 },
613 	{ gf100_gr_init_pri_0 },
614 	{ gf100_gr_init_rstr2d_0 },
615 	{ gf100_gr_init_pd_0 },
616 	{ gf100_gr_init_ds_0 },
617 	{ gf100_gr_init_scc_0 },
618 	{ gf100_gr_init_prop_0 },
619 	{ gf100_gr_init_gpc_unk_0 },
620 	{ gf100_gr_init_setup_0 },
621 	{ gf100_gr_init_crstr_0 },
622 	{ gf100_gr_init_setup_1 },
623 	{ gf100_gr_init_zcull_0 },
624 	{ gf100_gr_init_gpm_0 },
625 	{ gf100_gr_init_gpc_unk_1 },
626 	{ gf100_gr_init_gcc_0 },
627 	{ gf100_gr_init_tpccs_0 },
628 	{ gf100_gr_init_tex_0 },
629 	{ gf100_gr_init_pe_0 },
630 	{ gf100_gr_init_l1c_0 },
631 	{ gf100_gr_init_wwdx_0 },
632 	{ gf100_gr_init_tpccs_1 },
633 	{ gf100_gr_init_mpc_0 },
634 	{ gf100_gr_init_sm_0 },
635 	{ gf100_gr_init_be_0 },
636 	{ gf100_gr_init_fe_1 },
637 	{ gf100_gr_init_pe_1 },
638 	{}
639 };
640 
641 /*******************************************************************************
642  * PGRAPH engine/subdev functions
643  ******************************************************************************/
644 
645 void
646 gf100_gr_zbc_init(struct gf100_gr *gr)
647 {
648 	const u32  zero[] = { 0x00000000, 0x00000000, 0x00000000, 0x00000000,
649 			      0x00000000, 0x00000000, 0x00000000, 0x00000000 };
650 	const u32   one[] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000,
651 			      0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
652 	const u32 f32_0[] = { 0x00000000, 0x00000000, 0x00000000, 0x00000000,
653 			      0x00000000, 0x00000000, 0x00000000, 0x00000000 };
654 	const u32 f32_1[] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000,
655 			      0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 };
656 	struct nvkm_ltc *ltc = nvkm_ltc(gr);
657 	int index;
658 
659 	if (!gr->zbc_color[0].format) {
660 		gf100_gr_zbc_color_get(gr, 1,  & zero[0],   &zero[4]);
661 		gf100_gr_zbc_color_get(gr, 2,  &  one[0],    &one[4]);
662 		gf100_gr_zbc_color_get(gr, 4,  &f32_0[0],  &f32_0[4]);
663 		gf100_gr_zbc_color_get(gr, 4,  &f32_1[0],  &f32_1[4]);
664 		gf100_gr_zbc_depth_get(gr, 1, 0x00000000, 0x00000000);
665 		gf100_gr_zbc_depth_get(gr, 1, 0x3f800000, 0x3f800000);
666 	}
667 
668 	for (index = ltc->zbc_min; index <= ltc->zbc_max; index++)
669 		gf100_gr_zbc_clear_color(gr, index);
670 	for (index = ltc->zbc_min; index <= ltc->zbc_max; index++)
671 		gf100_gr_zbc_clear_depth(gr, index);
672 }
673 
674 /**
675  * Wait until GR goes idle. GR is considered idle if it is disabled by the
676  * MC (0x200) register, or GR is not busy and a context switch is not in
677  * progress.
678  */
679 int
680 gf100_gr_wait_idle(struct gf100_gr *gr)
681 {
682 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
683 	struct nvkm_device *device = subdev->device;
684 	unsigned long end_jiffies = jiffies + msecs_to_jiffies(2000);
685 	bool gr_enabled, ctxsw_active, gr_busy;
686 
687 	do {
688 		/*
689 		 * required to make sure FIFO_ENGINE_STATUS (0x2640) is
690 		 * up-to-date
691 		 */
692 		nvkm_rd32(device, 0x400700);
693 
694 		gr_enabled = nvkm_rd32(device, 0x200) & 0x1000;
695 		ctxsw_active = nvkm_rd32(device, 0x2640) & 0x8000;
696 		gr_busy = nvkm_rd32(device, 0x40060c) & 0x1;
697 
698 		if (!gr_enabled || (!gr_busy && !ctxsw_active))
699 			return 0;
700 	} while (time_before(jiffies, end_jiffies));
701 
702 	nvkm_error(subdev,
703 		   "wait for idle timeout (en: %d, ctxsw: %d, busy: %d)\n",
704 		   gr_enabled, ctxsw_active, gr_busy);
705 	return -EAGAIN;
706 }
707 
708 void
709 gf100_gr_mmio(struct gf100_gr *gr, const struct gf100_gr_pack *p)
710 {
711 	struct nvkm_device *device = gr->base.engine.subdev.device;
712 	const struct gf100_gr_pack *pack;
713 	const struct gf100_gr_init *init;
714 
715 	pack_for_each_init(init, pack, p) {
716 		u32 next = init->addr + init->count * init->pitch;
717 		u32 addr = init->addr;
718 		while (addr < next) {
719 			nvkm_wr32(device, addr, init->data);
720 			addr += init->pitch;
721 		}
722 	}
723 }
724 
725 void
726 gf100_gr_icmd(struct gf100_gr *gr, const struct gf100_gr_pack *p)
727 {
728 	struct nvkm_device *device = gr->base.engine.subdev.device;
729 	const struct gf100_gr_pack *pack;
730 	const struct gf100_gr_init *init;
731 	u32 data = 0;
732 
733 	nvkm_wr32(device, 0x400208, 0x80000000);
734 
735 	pack_for_each_init(init, pack, p) {
736 		u32 next = init->addr + init->count * init->pitch;
737 		u32 addr = init->addr;
738 
739 		if ((pack == p && init == p->init) || data != init->data) {
740 			nvkm_wr32(device, 0x400204, init->data);
741 			data = init->data;
742 		}
743 
744 		while (addr < next) {
745 			nvkm_wr32(device, 0x400200, addr);
746 			/**
747 			 * Wait for GR to go idle after submitting a
748 			 * GO_IDLE bundle
749 			 */
750 			if ((addr & 0xffff) == 0xe100)
751 				gf100_gr_wait_idle(gr);
752 			nvkm_msec(device, 2000,
753 				if (!(nvkm_rd32(device, 0x400700) & 0x00000004))
754 					break;
755 			);
756 			addr += init->pitch;
757 		}
758 	}
759 
760 	nvkm_wr32(device, 0x400208, 0x00000000);
761 }
762 
763 void
764 gf100_gr_mthd(struct gf100_gr *gr, const struct gf100_gr_pack *p)
765 {
766 	struct nvkm_device *device = gr->base.engine.subdev.device;
767 	const struct gf100_gr_pack *pack;
768 	const struct gf100_gr_init *init;
769 	u32 data = 0;
770 
771 	pack_for_each_init(init, pack, p) {
772 		u32 ctrl = 0x80000000 | pack->type;
773 		u32 next = init->addr + init->count * init->pitch;
774 		u32 addr = init->addr;
775 
776 		if ((pack == p && init == p->init) || data != init->data) {
777 			nvkm_wr32(device, 0x40448c, init->data);
778 			data = init->data;
779 		}
780 
781 		while (addr < next) {
782 			nvkm_wr32(device, 0x404488, ctrl | (addr << 14));
783 			addr += init->pitch;
784 		}
785 	}
786 }
787 
788 u64
789 gf100_gr_units(struct nvkm_gr *obj)
790 {
791 	struct gf100_gr *gr = container_of(obj, typeof(*gr), base);
792 	u64 cfg;
793 
794 	cfg  = (u32)gr->gpc_nr;
795 	cfg |= (u32)gr->tpc_total << 8;
796 	cfg |= (u64)gr->rop_nr << 32;
797 
798 	return cfg;
799 }
800 
801 static const struct nvkm_bitfield gk104_sked_error[] = {
802 	{ 0x00000080, "CONSTANT_BUFFER_SIZE" },
803 	{ 0x00000200, "LOCAL_MEMORY_SIZE_POS" },
804 	{ 0x00000400, "LOCAL_MEMORY_SIZE_NEG" },
805 	{ 0x00000800, "WARP_CSTACK_SIZE" },
806 	{ 0x00001000, "TOTAL_TEMP_SIZE" },
807 	{ 0x00002000, "REGISTER_COUNT" },
808 	{ 0x00040000, "TOTAL_THREADS" },
809 	{ 0x00100000, "PROGRAM_OFFSET" },
810 	{ 0x00200000, "SHARED_MEMORY_SIZE" },
811 	{ 0x02000000, "SHARED_CONFIG_TOO_SMALL" },
812 	{ 0x04000000, "TOTAL_REGISTER_COUNT" },
813 	{}
814 };
815 
816 static const struct nvkm_bitfield gf100_gpc_rop_error[] = {
817 	{ 0x00000002, "RT_PITCH_OVERRUN" },
818 	{ 0x00000010, "RT_WIDTH_OVERRUN" },
819 	{ 0x00000020, "RT_HEIGHT_OVERRUN" },
820 	{ 0x00000080, "ZETA_STORAGE_TYPE_MISMATCH" },
821 	{ 0x00000100, "RT_STORAGE_TYPE_MISMATCH" },
822 	{ 0x00000400, "RT_LINEAR_MISMATCH" },
823 	{}
824 };
825 
826 static void
827 gf100_gr_trap_gpc_rop(struct gf100_gr *gr, int gpc)
828 {
829 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
830 	struct nvkm_device *device = subdev->device;
831 	char error[128];
832 	u32 trap[4];
833 
834 	trap[0] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0420)) & 0x3fffffff;
835 	trap[1] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0434));
836 	trap[2] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0438));
837 	trap[3] = nvkm_rd32(device, GPC_UNIT(gpc, 0x043c));
838 
839 	nvkm_snprintbf(error, sizeof(error), gf100_gpc_rop_error, trap[0]);
840 
841 	nvkm_error(subdev, "GPC%d/PROP trap: %08x [%s] x = %u, y = %u, "
842 			   "format = %x, storage type = %x\n",
843 		   gpc, trap[0], error, trap[1] & 0xffff, trap[1] >> 16,
844 		   (trap[2] >> 8) & 0x3f, trap[3] & 0xff);
845 	nvkm_wr32(device, GPC_UNIT(gpc, 0x0420), 0xc0000000);
846 }
847 
848 static const struct nvkm_enum gf100_mp_warp_error[] = {
849 	{ 0x00, "NO_ERROR" },
850 	{ 0x01, "STACK_MISMATCH" },
851 	{ 0x05, "MISALIGNED_PC" },
852 	{ 0x08, "MISALIGNED_GPR" },
853 	{ 0x09, "INVALID_OPCODE" },
854 	{ 0x0d, "GPR_OUT_OF_BOUNDS" },
855 	{ 0x0e, "MEM_OUT_OF_BOUNDS" },
856 	{ 0x0f, "UNALIGNED_MEM_ACCESS" },
857 	{ 0x11, "INVALID_PARAM" },
858 	{}
859 };
860 
861 static const struct nvkm_bitfield gf100_mp_global_error[] = {
862 	{ 0x00000004, "MULTIPLE_WARP_ERRORS" },
863 	{ 0x00000008, "OUT_OF_STACK_SPACE" },
864 	{}
865 };
866 
867 static void
868 gf100_gr_trap_mp(struct gf100_gr *gr, int gpc, int tpc)
869 {
870 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
871 	struct nvkm_device *device = subdev->device;
872 	u32 werr = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x648));
873 	u32 gerr = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x650));
874 	const struct nvkm_enum *warp;
875 	char glob[128];
876 
877 	nvkm_snprintbf(glob, sizeof(glob), gf100_mp_global_error, gerr);
878 	warp = nvkm_enum_find(gf100_mp_warp_error, werr & 0xffff);
879 
880 	nvkm_error(subdev, "GPC%i/TPC%i/MP trap: "
881 			   "global %08x [%s] warp %04x [%s]\n",
882 		   gpc, tpc, gerr, glob, werr, warp ? warp->name : "");
883 
884 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x648), 0x00000000);
885 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x650), gerr);
886 }
887 
888 static void
889 gf100_gr_trap_tpc(struct gf100_gr *gr, int gpc, int tpc)
890 {
891 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
892 	struct nvkm_device *device = subdev->device;
893 	u32 stat = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0508));
894 
895 	if (stat & 0x00000001) {
896 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0224));
897 		nvkm_error(subdev, "GPC%d/TPC%d/TEX: %08x\n", gpc, tpc, trap);
898 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x0224), 0xc0000000);
899 		stat &= ~0x00000001;
900 	}
901 
902 	if (stat & 0x00000002) {
903 		gf100_gr_trap_mp(gr, gpc, tpc);
904 		stat &= ~0x00000002;
905 	}
906 
907 	if (stat & 0x00000004) {
908 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0084));
909 		nvkm_error(subdev, "GPC%d/TPC%d/POLY: %08x\n", gpc, tpc, trap);
910 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x0084), 0xc0000000);
911 		stat &= ~0x00000004;
912 	}
913 
914 	if (stat & 0x00000008) {
915 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x048c));
916 		nvkm_error(subdev, "GPC%d/TPC%d/L1C: %08x\n", gpc, tpc, trap);
917 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x048c), 0xc0000000);
918 		stat &= ~0x00000008;
919 	}
920 
921 	if (stat) {
922 		nvkm_error(subdev, "GPC%d/TPC%d/%08x: unknown\n", gpc, tpc, stat);
923 	}
924 }
925 
926 static void
927 gf100_gr_trap_gpc(struct gf100_gr *gr, int gpc)
928 {
929 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
930 	struct nvkm_device *device = subdev->device;
931 	u32 stat = nvkm_rd32(device, GPC_UNIT(gpc, 0x2c90));
932 	int tpc;
933 
934 	if (stat & 0x00000001) {
935 		gf100_gr_trap_gpc_rop(gr, gpc);
936 		stat &= ~0x00000001;
937 	}
938 
939 	if (stat & 0x00000002) {
940 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x0900));
941 		nvkm_error(subdev, "GPC%d/ZCULL: %08x\n", gpc, trap);
942 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0900), 0xc0000000);
943 		stat &= ~0x00000002;
944 	}
945 
946 	if (stat & 0x00000004) {
947 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x1028));
948 		nvkm_error(subdev, "GPC%d/CCACHE: %08x\n", gpc, trap);
949 		nvkm_wr32(device, GPC_UNIT(gpc, 0x1028), 0xc0000000);
950 		stat &= ~0x00000004;
951 	}
952 
953 	if (stat & 0x00000008) {
954 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x0824));
955 		nvkm_error(subdev, "GPC%d/ESETUP: %08x\n", gpc, trap);
956 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0824), 0xc0000000);
957 		stat &= ~0x00000009;
958 	}
959 
960 	for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++) {
961 		u32 mask = 0x00010000 << tpc;
962 		if (stat & mask) {
963 			gf100_gr_trap_tpc(gr, gpc, tpc);
964 			nvkm_wr32(device, GPC_UNIT(gpc, 0x2c90), mask);
965 			stat &= ~mask;
966 		}
967 	}
968 
969 	if (stat) {
970 		nvkm_error(subdev, "GPC%d/%08x: unknown\n", gpc, stat);
971 	}
972 }
973 
974 static void
975 gf100_gr_trap_intr(struct gf100_gr *gr)
976 {
977 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
978 	struct nvkm_device *device = subdev->device;
979 	u32 trap = nvkm_rd32(device, 0x400108);
980 	int rop, gpc;
981 
982 	if (trap & 0x00000001) {
983 		u32 stat = nvkm_rd32(device, 0x404000);
984 		nvkm_error(subdev, "DISPATCH %08x\n", stat);
985 		nvkm_wr32(device, 0x404000, 0xc0000000);
986 		nvkm_wr32(device, 0x400108, 0x00000001);
987 		trap &= ~0x00000001;
988 	}
989 
990 	if (trap & 0x00000002) {
991 		u32 stat = nvkm_rd32(device, 0x404600);
992 		nvkm_error(subdev, "M2MF %08x\n", stat);
993 		nvkm_wr32(device, 0x404600, 0xc0000000);
994 		nvkm_wr32(device, 0x400108, 0x00000002);
995 		trap &= ~0x00000002;
996 	}
997 
998 	if (trap & 0x00000008) {
999 		u32 stat = nvkm_rd32(device, 0x408030);
1000 		nvkm_error(subdev, "CCACHE %08x\n", stat);
1001 		nvkm_wr32(device, 0x408030, 0xc0000000);
1002 		nvkm_wr32(device, 0x400108, 0x00000008);
1003 		trap &= ~0x00000008;
1004 	}
1005 
1006 	if (trap & 0x00000010) {
1007 		u32 stat = nvkm_rd32(device, 0x405840);
1008 		nvkm_error(subdev, "SHADER %08x\n", stat);
1009 		nvkm_wr32(device, 0x405840, 0xc0000000);
1010 		nvkm_wr32(device, 0x400108, 0x00000010);
1011 		trap &= ~0x00000010;
1012 	}
1013 
1014 	if (trap & 0x00000040) {
1015 		u32 stat = nvkm_rd32(device, 0x40601c);
1016 		nvkm_error(subdev, "UNK6 %08x\n", stat);
1017 		nvkm_wr32(device, 0x40601c, 0xc0000000);
1018 		nvkm_wr32(device, 0x400108, 0x00000040);
1019 		trap &= ~0x00000040;
1020 	}
1021 
1022 	if (trap & 0x00000080) {
1023 		u32 stat = nvkm_rd32(device, 0x404490);
1024 		nvkm_error(subdev, "MACRO %08x\n", stat);
1025 		nvkm_wr32(device, 0x404490, 0xc0000000);
1026 		nvkm_wr32(device, 0x400108, 0x00000080);
1027 		trap &= ~0x00000080;
1028 	}
1029 
1030 	if (trap & 0x00000100) {
1031 		u32 stat = nvkm_rd32(device, 0x407020) & 0x3fffffff;
1032 		char sked[128];
1033 
1034 		nvkm_snprintbf(sked, sizeof(sked), gk104_sked_error, stat);
1035 		nvkm_error(subdev, "SKED: %08x [%s]\n", stat, sked);
1036 
1037 		if (stat)
1038 			nvkm_wr32(device, 0x407020, 0x40000000);
1039 		nvkm_wr32(device, 0x400108, 0x00000100);
1040 		trap &= ~0x00000100;
1041 	}
1042 
1043 	if (trap & 0x01000000) {
1044 		u32 stat = nvkm_rd32(device, 0x400118);
1045 		for (gpc = 0; stat && gpc < gr->gpc_nr; gpc++) {
1046 			u32 mask = 0x00000001 << gpc;
1047 			if (stat & mask) {
1048 				gf100_gr_trap_gpc(gr, gpc);
1049 				nvkm_wr32(device, 0x400118, mask);
1050 				stat &= ~mask;
1051 			}
1052 		}
1053 		nvkm_wr32(device, 0x400108, 0x01000000);
1054 		trap &= ~0x01000000;
1055 	}
1056 
1057 	if (trap & 0x02000000) {
1058 		for (rop = 0; rop < gr->rop_nr; rop++) {
1059 			u32 statz = nvkm_rd32(device, ROP_UNIT(rop, 0x070));
1060 			u32 statc = nvkm_rd32(device, ROP_UNIT(rop, 0x144));
1061 			nvkm_error(subdev, "ROP%d %08x %08x\n",
1062 				 rop, statz, statc);
1063 			nvkm_wr32(device, ROP_UNIT(rop, 0x070), 0xc0000000);
1064 			nvkm_wr32(device, ROP_UNIT(rop, 0x144), 0xc0000000);
1065 		}
1066 		nvkm_wr32(device, 0x400108, 0x02000000);
1067 		trap &= ~0x02000000;
1068 	}
1069 
1070 	if (trap) {
1071 		nvkm_error(subdev, "TRAP UNHANDLED %08x\n", trap);
1072 		nvkm_wr32(device, 0x400108, trap);
1073 	}
1074 }
1075 
1076 static void
1077 gf100_gr_ctxctl_debug_unit(struct gf100_gr *gr, u32 base)
1078 {
1079 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1080 	struct nvkm_device *device = subdev->device;
1081 	nvkm_error(subdev, "%06x - done %08x\n", base,
1082 		   nvkm_rd32(device, base + 0x400));
1083 	nvkm_error(subdev, "%06x - stat %08x %08x %08x %08x\n", base,
1084 		   nvkm_rd32(device, base + 0x800),
1085 		   nvkm_rd32(device, base + 0x804),
1086 		   nvkm_rd32(device, base + 0x808),
1087 		   nvkm_rd32(device, base + 0x80c));
1088 	nvkm_error(subdev, "%06x - stat %08x %08x %08x %08x\n", base,
1089 		   nvkm_rd32(device, base + 0x810),
1090 		   nvkm_rd32(device, base + 0x814),
1091 		   nvkm_rd32(device, base + 0x818),
1092 		   nvkm_rd32(device, base + 0x81c));
1093 }
1094 
1095 void
1096 gf100_gr_ctxctl_debug(struct gf100_gr *gr)
1097 {
1098 	struct nvkm_device *device = gr->base.engine.subdev.device;
1099 	u32 gpcnr = nvkm_rd32(device, 0x409604) & 0xffff;
1100 	u32 gpc;
1101 
1102 	gf100_gr_ctxctl_debug_unit(gr, 0x409000);
1103 	for (gpc = 0; gpc < gpcnr; gpc++)
1104 		gf100_gr_ctxctl_debug_unit(gr, 0x502000 + (gpc * 0x8000));
1105 }
1106 
1107 static void
1108 gf100_gr_ctxctl_isr(struct gf100_gr *gr)
1109 {
1110 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1111 	struct nvkm_device *device = subdev->device;
1112 	u32 stat = nvkm_rd32(device, 0x409c18);
1113 
1114 	if (stat & 0x00000001) {
1115 		u32 code = nvkm_rd32(device, 0x409814);
1116 		if (code == E_BAD_FWMTHD) {
1117 			u32 class = nvkm_rd32(device, 0x409808);
1118 			u32  addr = nvkm_rd32(device, 0x40980c);
1119 			u32  subc = (addr & 0x00070000) >> 16;
1120 			u32  mthd = (addr & 0x00003ffc);
1121 			u32  data = nvkm_rd32(device, 0x409810);
1122 
1123 			nvkm_error(subdev, "FECS MTHD subc %d class %04x "
1124 					   "mthd %04x data %08x\n",
1125 				   subc, class, mthd, data);
1126 
1127 			nvkm_wr32(device, 0x409c20, 0x00000001);
1128 			stat &= ~0x00000001;
1129 		} else {
1130 			nvkm_error(subdev, "FECS ucode error %d\n", code);
1131 		}
1132 	}
1133 
1134 	if (stat & 0x00080000) {
1135 		nvkm_error(subdev, "FECS watchdog timeout\n");
1136 		gf100_gr_ctxctl_debug(gr);
1137 		nvkm_wr32(device, 0x409c20, 0x00080000);
1138 		stat &= ~0x00080000;
1139 	}
1140 
1141 	if (stat) {
1142 		nvkm_error(subdev, "FECS %08x\n", stat);
1143 		gf100_gr_ctxctl_debug(gr);
1144 		nvkm_wr32(device, 0x409c20, stat);
1145 	}
1146 }
1147 
1148 static void
1149 gf100_gr_intr(struct nvkm_subdev *subdev)
1150 {
1151 	struct gf100_gr *gr = (void *)subdev;
1152 	struct nvkm_device *device = gr->base.engine.subdev.device;
1153 	struct nvkm_fifo *fifo = device->fifo;
1154 	struct nvkm_engine *engine = nv_engine(subdev);
1155 	struct nvkm_object *engctx;
1156 	struct nvkm_handle *handle;
1157 	u64 inst = nvkm_rd32(device, 0x409b00) & 0x0fffffff;
1158 	u32 stat = nvkm_rd32(device, 0x400100);
1159 	u32 addr = nvkm_rd32(device, 0x400704);
1160 	u32 mthd = (addr & 0x00003ffc);
1161 	u32 subc = (addr & 0x00070000) >> 16;
1162 	u32 data = nvkm_rd32(device, 0x400708);
1163 	u32 code = nvkm_rd32(device, 0x400110);
1164 	u32 class;
1165 	int chid;
1166 
1167 	if (nv_device(gr)->card_type < NV_E0 || subc < 4)
1168 		class = nvkm_rd32(device, 0x404200 + (subc * 4));
1169 	else
1170 		class = 0x0000;
1171 
1172 	engctx = nvkm_engctx_get(engine, inst);
1173 	chid   = fifo->chid(fifo, engctx);
1174 
1175 	if (stat & 0x00000001) {
1176 		/*
1177 		 * notifier interrupt, only needed for cyclestats
1178 		 * can be safely ignored
1179 		 */
1180 		nvkm_wr32(device, 0x400100, 0x00000001);
1181 		stat &= ~0x00000001;
1182 	}
1183 
1184 	if (stat & 0x00000010) {
1185 		handle = nvkm_handle_get_class(engctx, class);
1186 		if (!handle || nv_call(handle->object, mthd, data)) {
1187 			nvkm_error(subdev, "ILLEGAL_MTHD ch %d [%010llx %s] "
1188 				   "subc %d class %04x mthd %04x data %08x\n",
1189 				   chid, inst << 12, nvkm_client_name(engctx),
1190 				   subc, class, mthd, data);
1191 		}
1192 		nvkm_handle_put(handle);
1193 		nvkm_wr32(device, 0x400100, 0x00000010);
1194 		stat &= ~0x00000010;
1195 	}
1196 
1197 	if (stat & 0x00000020) {
1198 		nvkm_error(subdev, "ILLEGAL_CLASS ch %d [%010llx %s] "
1199 			   "subc %d class %04x mthd %04x data %08x\n",
1200 			   chid, inst << 12, nvkm_client_name(engctx), subc,
1201 			   class, mthd, data);
1202 		nvkm_wr32(device, 0x400100, 0x00000020);
1203 		stat &= ~0x00000020;
1204 	}
1205 
1206 	if (stat & 0x00100000) {
1207 		const struct nvkm_enum *en =
1208 			nvkm_enum_find(nv50_data_error_names, code);
1209 		nvkm_error(subdev, "DATA_ERROR %08x [%s] ch %d [%010llx %s] "
1210 				   "subc %d class %04x mthd %04x data %08x\n",
1211 			   code, en ? en->name : "", chid, inst << 12,
1212 			   nvkm_client_name(engctx), subc, class, mthd, data);
1213 		nvkm_wr32(device, 0x400100, 0x00100000);
1214 		stat &= ~0x00100000;
1215 	}
1216 
1217 	if (stat & 0x00200000) {
1218 		nvkm_error(subdev, "TRAP ch %d [%010llx %s]\n",
1219 			   chid, inst << 12,
1220 			   nvkm_client_name(engctx));
1221 		gf100_gr_trap_intr(gr);
1222 		nvkm_wr32(device, 0x400100, 0x00200000);
1223 		stat &= ~0x00200000;
1224 	}
1225 
1226 	if (stat & 0x00080000) {
1227 		gf100_gr_ctxctl_isr(gr);
1228 		nvkm_wr32(device, 0x400100, 0x00080000);
1229 		stat &= ~0x00080000;
1230 	}
1231 
1232 	if (stat) {
1233 		nvkm_error(subdev, "intr %08x\n", stat);
1234 		nvkm_wr32(device, 0x400100, stat);
1235 	}
1236 
1237 	nvkm_wr32(device, 0x400500, 0x00010001);
1238 	nvkm_engctx_put(engctx);
1239 }
1240 
1241 void
1242 gf100_gr_init_fw(struct gf100_gr *gr, u32 fuc_base,
1243 		 struct gf100_gr_fuc *code, struct gf100_gr_fuc *data)
1244 {
1245 	struct nvkm_device *device = gr->base.engine.subdev.device;
1246 	int i;
1247 
1248 	nvkm_wr32(device, fuc_base + 0x01c0, 0x01000000);
1249 	for (i = 0; i < data->size / 4; i++)
1250 		nvkm_wr32(device, fuc_base + 0x01c4, data->data[i]);
1251 
1252 	nvkm_wr32(device, fuc_base + 0x0180, 0x01000000);
1253 	for (i = 0; i < code->size / 4; i++) {
1254 		if ((i & 0x3f) == 0)
1255 			nvkm_wr32(device, fuc_base + 0x0188, i >> 6);
1256 		nvkm_wr32(device, fuc_base + 0x0184, code->data[i]);
1257 	}
1258 
1259 	/* code must be padded to 0x40 words */
1260 	for (; i & 0x3f; i++)
1261 		nvkm_wr32(device, fuc_base + 0x0184, 0);
1262 }
1263 
1264 static void
1265 gf100_gr_init_csdata(struct gf100_gr *gr,
1266 		     const struct gf100_gr_pack *pack,
1267 		     u32 falcon, u32 starstar, u32 base)
1268 {
1269 	struct nvkm_device *device = gr->base.engine.subdev.device;
1270 	const struct gf100_gr_pack *iter;
1271 	const struct gf100_gr_init *init;
1272 	u32 addr = ~0, prev = ~0, xfer = 0;
1273 	u32 star, temp;
1274 
1275 	nvkm_wr32(device, falcon + 0x01c0, 0x02000000 + starstar);
1276 	star = nvkm_rd32(device, falcon + 0x01c4);
1277 	temp = nvkm_rd32(device, falcon + 0x01c4);
1278 	if (temp > star)
1279 		star = temp;
1280 	nvkm_wr32(device, falcon + 0x01c0, 0x01000000 + star);
1281 
1282 	pack_for_each_init(init, iter, pack) {
1283 		u32 head = init->addr - base;
1284 		u32 tail = head + init->count * init->pitch;
1285 		while (head < tail) {
1286 			if (head != prev + 4 || xfer >= 32) {
1287 				if (xfer) {
1288 					u32 data = ((--xfer << 26) | addr);
1289 					nvkm_wr32(device, falcon + 0x01c4, data);
1290 					star += 4;
1291 				}
1292 				addr = head;
1293 				xfer = 0;
1294 			}
1295 			prev = head;
1296 			xfer = xfer + 1;
1297 			head = head + init->pitch;
1298 		}
1299 	}
1300 
1301 	nvkm_wr32(device, falcon + 0x01c4, (--xfer << 26) | addr);
1302 	nvkm_wr32(device, falcon + 0x01c0, 0x01000004 + starstar);
1303 	nvkm_wr32(device, falcon + 0x01c4, star + 4);
1304 }
1305 
1306 int
1307 gf100_gr_init_ctxctl(struct gf100_gr *gr)
1308 {
1309 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1310 	struct nvkm_device *device = subdev->device;
1311 	struct gf100_gr_oclass *oclass = (void *)nv_object(gr)->oclass;
1312 	struct gf100_grctx_oclass *cclass = (void *)nv_engine(gr)->cclass;
1313 	int i;
1314 
1315 	if (gr->firmware) {
1316 		/* load fuc microcode */
1317 		nvkm_mc(gr)->unk260(nvkm_mc(gr), 0);
1318 		gf100_gr_init_fw(gr, 0x409000, &gr->fuc409c,
1319 						 &gr->fuc409d);
1320 		gf100_gr_init_fw(gr, 0x41a000, &gr->fuc41ac,
1321 						 &gr->fuc41ad);
1322 		nvkm_mc(gr)->unk260(nvkm_mc(gr), 1);
1323 
1324 		/* start both of them running */
1325 		nvkm_wr32(device, 0x409840, 0xffffffff);
1326 		nvkm_wr32(device, 0x41a10c, 0x00000000);
1327 		nvkm_wr32(device, 0x40910c, 0x00000000);
1328 		nvkm_wr32(device, 0x41a100, 0x00000002);
1329 		nvkm_wr32(device, 0x409100, 0x00000002);
1330 		if (nvkm_msec(device, 2000,
1331 			if (nvkm_rd32(device, 0x409800) & 0x00000001)
1332 				break;
1333 		) < 0)
1334 			return -EBUSY;
1335 
1336 		nvkm_wr32(device, 0x409840, 0xffffffff);
1337 		nvkm_wr32(device, 0x409500, 0x7fffffff);
1338 		nvkm_wr32(device, 0x409504, 0x00000021);
1339 
1340 		nvkm_wr32(device, 0x409840, 0xffffffff);
1341 		nvkm_wr32(device, 0x409500, 0x00000000);
1342 		nvkm_wr32(device, 0x409504, 0x00000010);
1343 		if (nvkm_msec(device, 2000,
1344 			if ((gr->size = nvkm_rd32(device, 0x409800)))
1345 				break;
1346 		) < 0)
1347 			return -EBUSY;
1348 
1349 		nvkm_wr32(device, 0x409840, 0xffffffff);
1350 		nvkm_wr32(device, 0x409500, 0x00000000);
1351 		nvkm_wr32(device, 0x409504, 0x00000016);
1352 		if (nvkm_msec(device, 2000,
1353 			if (nvkm_rd32(device, 0x409800))
1354 				break;
1355 		) < 0)
1356 			return -EBUSY;
1357 
1358 		nvkm_wr32(device, 0x409840, 0xffffffff);
1359 		nvkm_wr32(device, 0x409500, 0x00000000);
1360 		nvkm_wr32(device, 0x409504, 0x00000025);
1361 		if (nvkm_msec(device, 2000,
1362 			if (nvkm_rd32(device, 0x409800))
1363 				break;
1364 		) < 0)
1365 			return -EBUSY;
1366 
1367 		if (nv_device(gr)->chipset >= 0xe0) {
1368 			nvkm_wr32(device, 0x409800, 0x00000000);
1369 			nvkm_wr32(device, 0x409500, 0x00000001);
1370 			nvkm_wr32(device, 0x409504, 0x00000030);
1371 			if (nvkm_msec(device, 2000,
1372 				if (nvkm_rd32(device, 0x409800))
1373 					break;
1374 			) < 0)
1375 				return -EBUSY;
1376 
1377 			nvkm_wr32(device, 0x409810, 0xb00095c8);
1378 			nvkm_wr32(device, 0x409800, 0x00000000);
1379 			nvkm_wr32(device, 0x409500, 0x00000001);
1380 			nvkm_wr32(device, 0x409504, 0x00000031);
1381 			if (nvkm_msec(device, 2000,
1382 				if (nvkm_rd32(device, 0x409800))
1383 					break;
1384 			) < 0)
1385 				return -EBUSY;
1386 
1387 			nvkm_wr32(device, 0x409810, 0x00080420);
1388 			nvkm_wr32(device, 0x409800, 0x00000000);
1389 			nvkm_wr32(device, 0x409500, 0x00000001);
1390 			nvkm_wr32(device, 0x409504, 0x00000032);
1391 			if (nvkm_msec(device, 2000,
1392 				if (nvkm_rd32(device, 0x409800))
1393 					break;
1394 			) < 0)
1395 				return -EBUSY;
1396 
1397 			nvkm_wr32(device, 0x409614, 0x00000070);
1398 			nvkm_wr32(device, 0x409614, 0x00000770);
1399 			nvkm_wr32(device, 0x40802c, 0x00000001);
1400 		}
1401 
1402 		if (gr->data == NULL) {
1403 			int ret = gf100_grctx_generate(gr);
1404 			if (ret) {
1405 				nvkm_error(subdev, "failed to construct context\n");
1406 				return ret;
1407 			}
1408 		}
1409 
1410 		return 0;
1411 	} else
1412 	if (!oclass->fecs.ucode) {
1413 		return -ENOSYS;
1414 	}
1415 
1416 	/* load HUB microcode */
1417 	nvkm_mc(gr)->unk260(nvkm_mc(gr), 0);
1418 	nvkm_wr32(device, 0x4091c0, 0x01000000);
1419 	for (i = 0; i < oclass->fecs.ucode->data.size / 4; i++)
1420 		nvkm_wr32(device, 0x4091c4, oclass->fecs.ucode->data.data[i]);
1421 
1422 	nvkm_wr32(device, 0x409180, 0x01000000);
1423 	for (i = 0; i < oclass->fecs.ucode->code.size / 4; i++) {
1424 		if ((i & 0x3f) == 0)
1425 			nvkm_wr32(device, 0x409188, i >> 6);
1426 		nvkm_wr32(device, 0x409184, oclass->fecs.ucode->code.data[i]);
1427 	}
1428 
1429 	/* load GPC microcode */
1430 	nvkm_wr32(device, 0x41a1c0, 0x01000000);
1431 	for (i = 0; i < oclass->gpccs.ucode->data.size / 4; i++)
1432 		nvkm_wr32(device, 0x41a1c4, oclass->gpccs.ucode->data.data[i]);
1433 
1434 	nvkm_wr32(device, 0x41a180, 0x01000000);
1435 	for (i = 0; i < oclass->gpccs.ucode->code.size / 4; i++) {
1436 		if ((i & 0x3f) == 0)
1437 			nvkm_wr32(device, 0x41a188, i >> 6);
1438 		nvkm_wr32(device, 0x41a184, oclass->gpccs.ucode->code.data[i]);
1439 	}
1440 	nvkm_mc(gr)->unk260(nvkm_mc(gr), 1);
1441 
1442 	/* load register lists */
1443 	gf100_gr_init_csdata(gr, cclass->hub, 0x409000, 0x000, 0x000000);
1444 	gf100_gr_init_csdata(gr, cclass->gpc, 0x41a000, 0x000, 0x418000);
1445 	gf100_gr_init_csdata(gr, cclass->tpc, 0x41a000, 0x004, 0x419800);
1446 	gf100_gr_init_csdata(gr, cclass->ppc, 0x41a000, 0x008, 0x41be00);
1447 
1448 	/* start HUB ucode running, it'll init the GPCs */
1449 	nvkm_wr32(device, 0x40910c, 0x00000000);
1450 	nvkm_wr32(device, 0x409100, 0x00000002);
1451 	if (nvkm_msec(device, 2000,
1452 		if (nvkm_rd32(device, 0x409800) & 0x80000000)
1453 			break;
1454 	) < 0) {
1455 		gf100_gr_ctxctl_debug(gr);
1456 		return -EBUSY;
1457 	}
1458 
1459 	gr->size = nvkm_rd32(device, 0x409804);
1460 	if (gr->data == NULL) {
1461 		int ret = gf100_grctx_generate(gr);
1462 		if (ret) {
1463 			nvkm_error(subdev, "failed to construct context\n");
1464 			return ret;
1465 		}
1466 	}
1467 
1468 	return 0;
1469 }
1470 
1471 int
1472 gf100_gr_init(struct nvkm_object *object)
1473 {
1474 	struct gf100_gr *gr = (void *)object;
1475 	struct nvkm_device *device = gr->base.engine.subdev.device;
1476 	struct gf100_gr_oclass *oclass = (void *)object->oclass;
1477 	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, gr->tpc_total);
1478 	u32 data[TPC_MAX / 8] = {};
1479 	u8  tpcnr[GPC_MAX];
1480 	int gpc, tpc, rop;
1481 	int ret, i;
1482 
1483 	ret = nvkm_gr_init(&gr->base);
1484 	if (ret)
1485 		return ret;
1486 
1487 	nvkm_wr32(device, GPC_BCAST(0x0880), 0x00000000);
1488 	nvkm_wr32(device, GPC_BCAST(0x08a4), 0x00000000);
1489 	nvkm_wr32(device, GPC_BCAST(0x0888), 0x00000000);
1490 	nvkm_wr32(device, GPC_BCAST(0x088c), 0x00000000);
1491 	nvkm_wr32(device, GPC_BCAST(0x0890), 0x00000000);
1492 	nvkm_wr32(device, GPC_BCAST(0x0894), 0x00000000);
1493 	nvkm_wr32(device, GPC_BCAST(0x08b4), gr->unk4188b4->addr >> 8);
1494 	nvkm_wr32(device, GPC_BCAST(0x08b8), gr->unk4188b8->addr >> 8);
1495 
1496 	gf100_gr_mmio(gr, oclass->mmio);
1497 
1498 	memcpy(tpcnr, gr->tpc_nr, sizeof(gr->tpc_nr));
1499 	for (i = 0, gpc = -1; i < gr->tpc_total; i++) {
1500 		do {
1501 			gpc = (gpc + 1) % gr->gpc_nr;
1502 		} while (!tpcnr[gpc]);
1503 		tpc = gr->tpc_nr[gpc] - tpcnr[gpc]--;
1504 
1505 		data[i / 8] |= tpc << ((i % 8) * 4);
1506 	}
1507 
1508 	nvkm_wr32(device, GPC_BCAST(0x0980), data[0]);
1509 	nvkm_wr32(device, GPC_BCAST(0x0984), data[1]);
1510 	nvkm_wr32(device, GPC_BCAST(0x0988), data[2]);
1511 	nvkm_wr32(device, GPC_BCAST(0x098c), data[3]);
1512 
1513 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
1514 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0914),
1515 			gr->magic_not_rop_nr << 8 | gr->tpc_nr[gpc]);
1516 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0910), 0x00040000 |
1517 			gr->tpc_total);
1518 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0918), magicgpc918);
1519 	}
1520 
1521 	if (nv_device(gr)->chipset != 0xd7)
1522 		nvkm_wr32(device, GPC_BCAST(0x1bd4), magicgpc918);
1523 	else
1524 		nvkm_wr32(device, GPC_BCAST(0x3fd4), magicgpc918);
1525 
1526 	nvkm_wr32(device, GPC_BCAST(0x08ac), nvkm_rd32(device, 0x100800));
1527 
1528 	nvkm_wr32(device, 0x400500, 0x00010001);
1529 
1530 	nvkm_wr32(device, 0x400100, 0xffffffff);
1531 	nvkm_wr32(device, 0x40013c, 0xffffffff);
1532 
1533 	nvkm_wr32(device, 0x409c24, 0x000f0000);
1534 	nvkm_wr32(device, 0x404000, 0xc0000000);
1535 	nvkm_wr32(device, 0x404600, 0xc0000000);
1536 	nvkm_wr32(device, 0x408030, 0xc0000000);
1537 	nvkm_wr32(device, 0x40601c, 0xc0000000);
1538 	nvkm_wr32(device, 0x404490, 0xc0000000);
1539 	nvkm_wr32(device, 0x406018, 0xc0000000);
1540 	nvkm_wr32(device, 0x405840, 0xc0000000);
1541 	nvkm_wr32(device, 0x405844, 0x00ffffff);
1542 	nvkm_mask(device, 0x419cc0, 0x00000008, 0x00000008);
1543 	nvkm_mask(device, 0x419eb4, 0x00001000, 0x00001000);
1544 
1545 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
1546 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0420), 0xc0000000);
1547 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0900), 0xc0000000);
1548 		nvkm_wr32(device, GPC_UNIT(gpc, 0x1028), 0xc0000000);
1549 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0824), 0xc0000000);
1550 		for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++) {
1551 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x508), 0xffffffff);
1552 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x50c), 0xffffffff);
1553 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x224), 0xc0000000);
1554 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x48c), 0xc0000000);
1555 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x084), 0xc0000000);
1556 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x644), 0x001ffffe);
1557 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x64c), 0x0000000f);
1558 		}
1559 		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c90), 0xffffffff);
1560 		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c94), 0xffffffff);
1561 	}
1562 
1563 	for (rop = 0; rop < gr->rop_nr; rop++) {
1564 		nvkm_wr32(device, ROP_UNIT(rop, 0x144), 0xc0000000);
1565 		nvkm_wr32(device, ROP_UNIT(rop, 0x070), 0xc0000000);
1566 		nvkm_wr32(device, ROP_UNIT(rop, 0x204), 0xffffffff);
1567 		nvkm_wr32(device, ROP_UNIT(rop, 0x208), 0xffffffff);
1568 	}
1569 
1570 	nvkm_wr32(device, 0x400108, 0xffffffff);
1571 	nvkm_wr32(device, 0x400138, 0xffffffff);
1572 	nvkm_wr32(device, 0x400118, 0xffffffff);
1573 	nvkm_wr32(device, 0x400130, 0xffffffff);
1574 	nvkm_wr32(device, 0x40011c, 0xffffffff);
1575 	nvkm_wr32(device, 0x400134, 0xffffffff);
1576 
1577 	nvkm_wr32(device, 0x400054, 0x34ce3464);
1578 
1579 	gf100_gr_zbc_init(gr);
1580 
1581 	return gf100_gr_init_ctxctl(gr);
1582 }
1583 
1584 void
1585 gf100_gr_dtor_fw(struct gf100_gr_fuc *fuc)
1586 {
1587 	kfree(fuc->data);
1588 	fuc->data = NULL;
1589 }
1590 
1591 int
1592 gf100_gr_ctor_fw(struct gf100_gr *gr, const char *fwname,
1593 		 struct gf100_gr_fuc *fuc)
1594 {
1595 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1596 	struct nvkm_device *device = subdev->device;
1597 	const struct firmware *fw;
1598 	char f[64];
1599 	char cname[16];
1600 	int ret;
1601 	int i;
1602 
1603 	/* Convert device name to lowercase */
1604 	strncpy(cname, device->cname, sizeof(cname));
1605 	cname[sizeof(cname) - 1] = '\0';
1606 	i = strlen(cname);
1607 	while (i) {
1608 		--i;
1609 		cname[i] = tolower(cname[i]);
1610 	}
1611 
1612 	snprintf(f, sizeof(f), "nvidia/%s/%s.bin", cname, fwname);
1613 	ret = request_firmware(&fw, f, nv_device_base(device));
1614 	if (ret) {
1615 		nvkm_error(subdev, "failed to load %s\n", fwname);
1616 		return ret;
1617 	}
1618 
1619 	fuc->size = fw->size;
1620 	fuc->data = kmemdup(fw->data, fuc->size, GFP_KERNEL);
1621 	release_firmware(fw);
1622 	return (fuc->data != NULL) ? 0 : -ENOMEM;
1623 }
1624 
1625 void
1626 gf100_gr_dtor(struct nvkm_object *object)
1627 {
1628 	struct gf100_gr *gr = (void *)object;
1629 
1630 	kfree(gr->data);
1631 
1632 	gf100_gr_dtor_fw(&gr->fuc409c);
1633 	gf100_gr_dtor_fw(&gr->fuc409d);
1634 	gf100_gr_dtor_fw(&gr->fuc41ac);
1635 	gf100_gr_dtor_fw(&gr->fuc41ad);
1636 
1637 	nvkm_gpuobj_ref(NULL, &gr->unk4188b8);
1638 	nvkm_gpuobj_ref(NULL, &gr->unk4188b4);
1639 
1640 	nvkm_gr_destroy(&gr->base);
1641 }
1642 
1643 int
1644 gf100_gr_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
1645 	      struct nvkm_oclass *bclass, void *data, u32 size,
1646 	      struct nvkm_object **pobject)
1647 {
1648 	struct gf100_gr_oclass *oclass = (void *)bclass;
1649 	struct nvkm_device *device = (void *)parent;
1650 	struct gf100_gr *gr;
1651 	bool use_ext_fw, enable;
1652 	int ret, i, j;
1653 
1654 	use_ext_fw = nvkm_boolopt(device->cfgopt, "NvGrUseFW",
1655 				  oclass->fecs.ucode == NULL);
1656 	enable = use_ext_fw || oclass->fecs.ucode != NULL;
1657 
1658 	ret = nvkm_gr_create(parent, engine, bclass, enable, &gr);
1659 	*pobject = nv_object(gr);
1660 	if (ret)
1661 		return ret;
1662 
1663 	nv_subdev(gr)->unit = 0x08001000;
1664 	nv_subdev(gr)->intr = gf100_gr_intr;
1665 
1666 	gr->base.units = gf100_gr_units;
1667 
1668 	if (use_ext_fw) {
1669 		nvkm_info(&gr->base.engine.subdev, "using external firmware\n");
1670 		if (gf100_gr_ctor_fw(gr, "fecs_inst", &gr->fuc409c) ||
1671 		    gf100_gr_ctor_fw(gr, "fecs_data", &gr->fuc409d) ||
1672 		    gf100_gr_ctor_fw(gr, "gpccs_inst", &gr->fuc41ac) ||
1673 		    gf100_gr_ctor_fw(gr, "gpccs_data", &gr->fuc41ad))
1674 			return -ENODEV;
1675 		gr->firmware = true;
1676 	}
1677 
1678 	ret = nvkm_gpuobj_new(nv_object(gr), NULL, 0x1000, 256, 0,
1679 			      &gr->unk4188b4);
1680 	if (ret)
1681 		return ret;
1682 
1683 	ret = nvkm_gpuobj_new(nv_object(gr), NULL, 0x1000, 256, 0,
1684 			      &gr->unk4188b8);
1685 	if (ret)
1686 		return ret;
1687 
1688 	nvkm_kmap(gr->unk4188b4);
1689 	for (i = 0; i < 0x1000; i += 4)
1690 		nvkm_wo32(gr->unk4188b4, i, 0x00000010);
1691 	nvkm_done(gr->unk4188b4);
1692 
1693 	nvkm_kmap(gr->unk4188b8);
1694 	for (i = 0; i < 0x1000; i += 4)
1695 		nvkm_wo32(gr->unk4188b8, i, 0x00000010);
1696 	nvkm_done(gr->unk4188b8);
1697 
1698 	gr->rop_nr = (nvkm_rd32(device, 0x409604) & 0x001f0000) >> 16;
1699 	gr->gpc_nr =  nvkm_rd32(device, 0x409604) & 0x0000001f;
1700 	for (i = 0; i < gr->gpc_nr; i++) {
1701 		gr->tpc_nr[i]  = nvkm_rd32(device, GPC_UNIT(i, 0x2608));
1702 		gr->tpc_total += gr->tpc_nr[i];
1703 		gr->ppc_nr[i]  = oclass->ppc_nr;
1704 		for (j = 0; j < gr->ppc_nr[i]; j++) {
1705 			u8 mask = nvkm_rd32(device, GPC_UNIT(i, 0x0c30 + (j * 4)));
1706 			gr->ppc_tpc_nr[i][j] = hweight8(mask);
1707 		}
1708 	}
1709 
1710 	/*XXX: these need figuring out... though it might not even matter */
1711 	switch (nv_device(gr)->chipset) {
1712 	case 0xc0:
1713 		if (gr->tpc_total == 11) { /* 465, 3/4/4/0, 4 */
1714 			gr->magic_not_rop_nr = 0x07;
1715 		} else
1716 		if (gr->tpc_total == 14) { /* 470, 3/3/4/4, 5 */
1717 			gr->magic_not_rop_nr = 0x05;
1718 		} else
1719 		if (gr->tpc_total == 15) { /* 480, 3/4/4/4, 6 */
1720 			gr->magic_not_rop_nr = 0x06;
1721 		}
1722 		break;
1723 	case 0xc3: /* 450, 4/0/0/0, 2 */
1724 		gr->magic_not_rop_nr = 0x03;
1725 		break;
1726 	case 0xc4: /* 460, 3/4/0/0, 4 */
1727 		gr->magic_not_rop_nr = 0x01;
1728 		break;
1729 	case 0xc1: /* 2/0/0/0, 1 */
1730 		gr->magic_not_rop_nr = 0x01;
1731 		break;
1732 	case 0xc8: /* 4/4/3/4, 5 */
1733 		gr->magic_not_rop_nr = 0x06;
1734 		break;
1735 	case 0xce: /* 4/4/0/0, 4 */
1736 		gr->magic_not_rop_nr = 0x03;
1737 		break;
1738 	case 0xcf: /* 4/0/0/0, 3 */
1739 		gr->magic_not_rop_nr = 0x03;
1740 		break;
1741 	case 0xd7:
1742 	case 0xd9: /* 1/0/0/0, 1 */
1743 	case 0xea: /* gk20a */
1744 	case 0x12b: /* gm20b */
1745 		gr->magic_not_rop_nr = 0x01;
1746 		break;
1747 	}
1748 
1749 	nv_engine(gr)->cclass = *oclass->cclass;
1750 	nv_engine(gr)->sclass =  oclass->sclass;
1751 	return 0;
1752 }
1753 
1754 #include "fuc/hubgf100.fuc3.h"
1755 
1756 struct gf100_gr_ucode
1757 gf100_gr_fecs_ucode = {
1758 	.code.data = gf100_grhub_code,
1759 	.code.size = sizeof(gf100_grhub_code),
1760 	.data.data = gf100_grhub_data,
1761 	.data.size = sizeof(gf100_grhub_data),
1762 };
1763 
1764 #include "fuc/gpcgf100.fuc3.h"
1765 
1766 struct gf100_gr_ucode
1767 gf100_gr_gpccs_ucode = {
1768 	.code.data = gf100_grgpc_code,
1769 	.code.size = sizeof(gf100_grgpc_code),
1770 	.data.data = gf100_grgpc_data,
1771 	.data.size = sizeof(gf100_grgpc_data),
1772 };
1773 
1774 struct nvkm_oclass *
1775 gf100_gr_oclass = &(struct gf100_gr_oclass) {
1776 	.base.handle = NV_ENGINE(GR, 0xc0),
1777 	.base.ofuncs = &(struct nvkm_ofuncs) {
1778 		.ctor = gf100_gr_ctor,
1779 		.dtor = gf100_gr_dtor,
1780 		.init = gf100_gr_init,
1781 		.fini = _nvkm_gr_fini,
1782 	},
1783 	.cclass = &gf100_grctx_oclass,
1784 	.sclass =  gf100_gr_sclass,
1785 	.mmio = gf100_gr_pack_mmio,
1786 	.fecs.ucode = &gf100_gr_fecs_ucode,
1787 	.gpccs.ucode = &gf100_gr_gpccs_ucode,
1788 }.base;
1789