xref: /netbsd-src/sys/external/bsd/drm2/dist/drm/nouveau/nvkm/engine/gr/nouveau_nvkm_engine_gr_gf100.c (revision 798b8d11ecd8257a8e35c3396210f98abf3d9ade)
1 /*	$NetBSD: nouveau_nvkm_engine_gr_gf100.c,v 1.7 2021/12/19 11:34:45 riastradh Exp $	*/
2 
3 /*
4  * Copyright 2012 Red Hat Inc.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Ben Skeggs
25  */
26 #include <sys/cdefs.h>
27 __KERNEL_RCSID(0, "$NetBSD: nouveau_nvkm_engine_gr_gf100.c,v 1.7 2021/12/19 11:34:45 riastradh Exp $");
28 
29 #include "gf100.h"
30 #include "ctxgf100.h"
31 #include "fuc/os.h"
32 
33 #include <core/client.h>
34 #include <core/firmware.h>
35 #include <core/option.h>
36 #include <subdev/acr.h>
37 #include <subdev/fb.h>
38 #include <subdev/mc.h>
39 #include <subdev/pmu.h>
40 #include <subdev/therm.h>
41 #include <subdev/timer.h>
42 #include <engine/fifo.h>
43 
44 #include <nvif/class.h>
45 #include <nvif/cl9097.h>
46 #include <nvif/if900d.h>
47 #include <nvif/unpack.h>
48 
49 #include <linux/nbsd-namespace.h>
50 
51 /*******************************************************************************
52  * Zero Bandwidth Clear
53  ******************************************************************************/
54 
55 static void
gf100_gr_zbc_clear_color(struct gf100_gr * gr,int zbc)56 gf100_gr_zbc_clear_color(struct gf100_gr *gr, int zbc)
57 {
58 	struct nvkm_device *device = gr->base.engine.subdev.device;
59 	if (gr->zbc_color[zbc].format) {
60 		nvkm_wr32(device, 0x405804, gr->zbc_color[zbc].ds[0]);
61 		nvkm_wr32(device, 0x405808, gr->zbc_color[zbc].ds[1]);
62 		nvkm_wr32(device, 0x40580c, gr->zbc_color[zbc].ds[2]);
63 		nvkm_wr32(device, 0x405810, gr->zbc_color[zbc].ds[3]);
64 	}
65 	nvkm_wr32(device, 0x405814, gr->zbc_color[zbc].format);
66 	nvkm_wr32(device, 0x405820, zbc);
67 	nvkm_wr32(device, 0x405824, 0x00000004); /* TRIGGER | WRITE | COLOR */
68 }
69 
70 static int
gf100_gr_zbc_color_get(struct gf100_gr * gr,int format,const u32 ds[4],const u32 l2[4])71 gf100_gr_zbc_color_get(struct gf100_gr *gr, int format,
72 		       const u32 ds[4], const u32 l2[4])
73 {
74 	struct nvkm_ltc *ltc = gr->base.engine.subdev.device->ltc;
75 	int zbc = -ENOSPC, i;
76 
77 	for (i = ltc->zbc_min; i <= ltc->zbc_max; i++) {
78 		if (gr->zbc_color[i].format) {
79 			if (gr->zbc_color[i].format != format)
80 				continue;
81 			if (memcmp(gr->zbc_color[i].ds, ds, sizeof(
82 				   gr->zbc_color[i].ds)))
83 				continue;
84 			if (memcmp(gr->zbc_color[i].l2, l2, sizeof(
85 				   gr->zbc_color[i].l2))) {
86 				WARN_ON(1);
87 				return -EINVAL;
88 			}
89 			return i;
90 		} else {
91 			zbc = (zbc < 0) ? i : zbc;
92 		}
93 	}
94 
95 	if (zbc < 0)
96 		return zbc;
97 
98 	memcpy(gr->zbc_color[zbc].ds, ds, sizeof(gr->zbc_color[zbc].ds));
99 	memcpy(gr->zbc_color[zbc].l2, l2, sizeof(gr->zbc_color[zbc].l2));
100 	gr->zbc_color[zbc].format = format;
101 	nvkm_ltc_zbc_color_get(ltc, zbc, l2);
102 	gr->func->zbc->clear_color(gr, zbc);
103 	return zbc;
104 }
105 
106 static void
gf100_gr_zbc_clear_depth(struct gf100_gr * gr,int zbc)107 gf100_gr_zbc_clear_depth(struct gf100_gr *gr, int zbc)
108 {
109 	struct nvkm_device *device = gr->base.engine.subdev.device;
110 	if (gr->zbc_depth[zbc].format)
111 		nvkm_wr32(device, 0x405818, gr->zbc_depth[zbc].ds);
112 	nvkm_wr32(device, 0x40581c, gr->zbc_depth[zbc].format);
113 	nvkm_wr32(device, 0x405820, zbc);
114 	nvkm_wr32(device, 0x405824, 0x00000005); /* TRIGGER | WRITE | DEPTH */
115 }
116 
117 static int
gf100_gr_zbc_depth_get(struct gf100_gr * gr,int format,const u32 ds,const u32 l2)118 gf100_gr_zbc_depth_get(struct gf100_gr *gr, int format,
119 		       const u32 ds, const u32 l2)
120 {
121 	struct nvkm_ltc *ltc = gr->base.engine.subdev.device->ltc;
122 	int zbc = -ENOSPC, i;
123 
124 	for (i = ltc->zbc_min; i <= ltc->zbc_max; i++) {
125 		if (gr->zbc_depth[i].format) {
126 			if (gr->zbc_depth[i].format != format)
127 				continue;
128 			if (gr->zbc_depth[i].ds != ds)
129 				continue;
130 			if (gr->zbc_depth[i].l2 != l2) {
131 				WARN_ON(1);
132 				return -EINVAL;
133 			}
134 			return i;
135 		} else {
136 			zbc = (zbc < 0) ? i : zbc;
137 		}
138 	}
139 
140 	if (zbc < 0)
141 		return zbc;
142 
143 	gr->zbc_depth[zbc].format = format;
144 	gr->zbc_depth[zbc].ds = ds;
145 	gr->zbc_depth[zbc].l2 = l2;
146 	nvkm_ltc_zbc_depth_get(ltc, zbc, l2);
147 	gr->func->zbc->clear_depth(gr, zbc);
148 	return zbc;
149 }
150 
151 const struct gf100_gr_func_zbc
152 gf100_gr_zbc = {
153 	.clear_color = gf100_gr_zbc_clear_color,
154 	.clear_depth = gf100_gr_zbc_clear_depth,
155 };
156 
157 /*******************************************************************************
158  * Graphics object classes
159  ******************************************************************************/
160 #define gf100_gr_object(p) container_of((p), struct gf100_gr_object, object)
161 
162 struct gf100_gr_object {
163 	struct nvkm_object object;
164 	struct gf100_gr_chan *chan;
165 };
166 
167 static int
gf100_fermi_mthd_zbc_color(struct nvkm_object * object,void * data,u32 size)168 gf100_fermi_mthd_zbc_color(struct nvkm_object *object, void *data, u32 size)
169 {
170 	struct gf100_gr *gr = gf100_gr(nvkm_gr(object->engine));
171 	union {
172 		struct fermi_a_zbc_color_v0 v0;
173 	} *args = data;
174 	int ret = -ENOSYS;
175 
176 	if (!(ret = nvif_unpack(ret, &data, &size, args->v0, 0, 0, false))) {
177 		switch (args->v0.format) {
178 		case FERMI_A_ZBC_COLOR_V0_FMT_ZERO:
179 		case FERMI_A_ZBC_COLOR_V0_FMT_UNORM_ONE:
180 		case FERMI_A_ZBC_COLOR_V0_FMT_RF32_GF32_BF32_AF32:
181 		case FERMI_A_ZBC_COLOR_V0_FMT_R16_G16_B16_A16:
182 		case FERMI_A_ZBC_COLOR_V0_FMT_RN16_GN16_BN16_AN16:
183 		case FERMI_A_ZBC_COLOR_V0_FMT_RS16_GS16_BS16_AS16:
184 		case FERMI_A_ZBC_COLOR_V0_FMT_RU16_GU16_BU16_AU16:
185 		case FERMI_A_ZBC_COLOR_V0_FMT_RF16_GF16_BF16_AF16:
186 		case FERMI_A_ZBC_COLOR_V0_FMT_A8R8G8B8:
187 		case FERMI_A_ZBC_COLOR_V0_FMT_A8RL8GL8BL8:
188 		case FERMI_A_ZBC_COLOR_V0_FMT_A2B10G10R10:
189 		case FERMI_A_ZBC_COLOR_V0_FMT_AU2BU10GU10RU10:
190 		case FERMI_A_ZBC_COLOR_V0_FMT_A8B8G8R8:
191 		case FERMI_A_ZBC_COLOR_V0_FMT_A8BL8GL8RL8:
192 		case FERMI_A_ZBC_COLOR_V0_FMT_AN8BN8GN8RN8:
193 		case FERMI_A_ZBC_COLOR_V0_FMT_AS8BS8GS8RS8:
194 		case FERMI_A_ZBC_COLOR_V0_FMT_AU8BU8GU8RU8:
195 		case FERMI_A_ZBC_COLOR_V0_FMT_A2R10G10B10:
196 		case FERMI_A_ZBC_COLOR_V0_FMT_BF10GF11RF11:
197 			ret = gf100_gr_zbc_color_get(gr, args->v0.format,
198 							   args->v0.ds,
199 							   args->v0.l2);
200 			if (ret >= 0) {
201 				args->v0.index = ret;
202 				return 0;
203 			}
204 			break;
205 		default:
206 			return -EINVAL;
207 		}
208 	}
209 
210 	return ret;
211 }
212 
213 static int
gf100_fermi_mthd_zbc_depth(struct nvkm_object * object,void * data,u32 size)214 gf100_fermi_mthd_zbc_depth(struct nvkm_object *object, void *data, u32 size)
215 {
216 	struct gf100_gr *gr = gf100_gr(nvkm_gr(object->engine));
217 	union {
218 		struct fermi_a_zbc_depth_v0 v0;
219 	} *args = data;
220 	int ret = -ENOSYS;
221 
222 	if (!(ret = nvif_unpack(ret, &data, &size, args->v0, 0, 0, false))) {
223 		switch (args->v0.format) {
224 		case FERMI_A_ZBC_DEPTH_V0_FMT_FP32:
225 			ret = gf100_gr_zbc_depth_get(gr, args->v0.format,
226 							   args->v0.ds,
227 							   args->v0.l2);
228 			return (ret >= 0) ? 0 : -ENOSPC;
229 		default:
230 			return -EINVAL;
231 		}
232 	}
233 
234 	return ret;
235 }
236 
237 static int
gf100_fermi_mthd(struct nvkm_object * object,u32 mthd,void * data,u32 size)238 gf100_fermi_mthd(struct nvkm_object *object, u32 mthd, void *data, u32 size)
239 {
240 	nvif_ioctl(object, "fermi mthd %08x\n", mthd);
241 	switch (mthd) {
242 	case FERMI_A_ZBC_COLOR:
243 		return gf100_fermi_mthd_zbc_color(object, data, size);
244 	case FERMI_A_ZBC_DEPTH:
245 		return gf100_fermi_mthd_zbc_depth(object, data, size);
246 	default:
247 		break;
248 	}
249 	return -EINVAL;
250 }
251 
252 const struct nvkm_object_func
253 gf100_fermi = {
254 	.mthd = gf100_fermi_mthd,
255 };
256 
257 static void
gf100_gr_mthd_set_shader_exceptions(struct nvkm_device * device,u32 data)258 gf100_gr_mthd_set_shader_exceptions(struct nvkm_device *device, u32 data)
259 {
260 	nvkm_wr32(device, 0x419e44, data ? 0xffffffff : 0x00000000);
261 	nvkm_wr32(device, 0x419e4c, data ? 0xffffffff : 0x00000000);
262 }
263 
264 static bool
gf100_gr_mthd_sw(struct nvkm_device * device,u16 class,u32 mthd,u32 data)265 gf100_gr_mthd_sw(struct nvkm_device *device, u16 class, u32 mthd, u32 data)
266 {
267 	switch (class & 0x00ff) {
268 	case 0x97:
269 	case 0xc0:
270 		switch (mthd) {
271 		case 0x1528:
272 			gf100_gr_mthd_set_shader_exceptions(device, data);
273 			return true;
274 		default:
275 			break;
276 		}
277 		break;
278 	default:
279 		break;
280 	}
281 	return false;
282 }
283 
284 static const struct nvkm_object_func
285 gf100_gr_object_func = {
286 };
287 
288 static int
gf100_gr_object_new(const struct nvkm_oclass * oclass,void * data,u32 size,struct nvkm_object ** pobject)289 gf100_gr_object_new(const struct nvkm_oclass *oclass, void *data, u32 size,
290 		    struct nvkm_object **pobject)
291 {
292 	struct gf100_gr_chan *chan = gf100_gr_chan(oclass->parent);
293 	struct gf100_gr_object *object;
294 
295 	if (!(object = kzalloc(sizeof(*object), GFP_KERNEL)))
296 		return -ENOMEM;
297 	*pobject = &object->object;
298 
299 	nvkm_object_ctor(oclass->base.func ? oclass->base.func :
300 			 &gf100_gr_object_func, oclass, &object->object);
301 	object->chan = chan;
302 	return 0;
303 }
304 
305 static int
gf100_gr_object_get(struct nvkm_gr * base,int index,struct nvkm_sclass * sclass)306 gf100_gr_object_get(struct nvkm_gr *base, int index, struct nvkm_sclass *sclass)
307 {
308 	struct gf100_gr *gr = gf100_gr(base);
309 	int c = 0;
310 
311 	while (gr->func->sclass[c].oclass) {
312 		if (c++ == index) {
313 			*sclass = gr->func->sclass[index];
314 			sclass->ctor = gf100_gr_object_new;
315 			return index;
316 		}
317 	}
318 
319 	return c;
320 }
321 
322 /*******************************************************************************
323  * PGRAPH context
324  ******************************************************************************/
325 
326 static int
gf100_gr_chan_bind(struct nvkm_object * object,struct nvkm_gpuobj * parent,int align,struct nvkm_gpuobj ** pgpuobj)327 gf100_gr_chan_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent,
328 		   int align, struct nvkm_gpuobj **pgpuobj)
329 {
330 	struct gf100_gr_chan *chan = gf100_gr_chan(object);
331 	struct gf100_gr *gr = chan->gr;
332 	int ret, i;
333 
334 	ret = nvkm_gpuobj_new(gr->base.engine.subdev.device, gr->size,
335 			      align, false, parent, pgpuobj);
336 	if (ret)
337 		return ret;
338 
339 	nvkm_kmap(*pgpuobj);
340 	for (i = 0; i < gr->size; i += 4)
341 		nvkm_wo32(*pgpuobj, i, gr->data[i / 4]);
342 
343 	if (!gr->firmware) {
344 		nvkm_wo32(*pgpuobj, 0x00, chan->mmio_nr / 2);
345 		nvkm_wo32(*pgpuobj, 0x04, chan->mmio_vma->addr >> 8);
346 	} else {
347 		nvkm_wo32(*pgpuobj, 0xf4, 0);
348 		nvkm_wo32(*pgpuobj, 0xf8, 0);
349 		nvkm_wo32(*pgpuobj, 0x10, chan->mmio_nr / 2);
350 		nvkm_wo32(*pgpuobj, 0x14, lower_32_bits(chan->mmio_vma->addr));
351 		nvkm_wo32(*pgpuobj, 0x18, upper_32_bits(chan->mmio_vma->addr));
352 		nvkm_wo32(*pgpuobj, 0x1c, 1);
353 		nvkm_wo32(*pgpuobj, 0x20, 0);
354 		nvkm_wo32(*pgpuobj, 0x28, 0);
355 		nvkm_wo32(*pgpuobj, 0x2c, 0);
356 	}
357 	nvkm_done(*pgpuobj);
358 	return 0;
359 }
360 
361 static void *
gf100_gr_chan_dtor(struct nvkm_object * object)362 gf100_gr_chan_dtor(struct nvkm_object *object)
363 {
364 	struct gf100_gr_chan *chan = gf100_gr_chan(object);
365 	int i;
366 
367 	for (i = 0; i < ARRAY_SIZE(chan->data); i++) {
368 		nvkm_vmm_put(chan->vmm, &chan->data[i].vma);
369 		nvkm_memory_unref(&chan->data[i].mem);
370 	}
371 
372 	nvkm_vmm_put(chan->vmm, &chan->mmio_vma);
373 	nvkm_memory_unref(&chan->mmio);
374 	nvkm_vmm_unref(&chan->vmm);
375 	return chan;
376 }
377 
378 static const struct nvkm_object_func
379 gf100_gr_chan = {
380 	.dtor = gf100_gr_chan_dtor,
381 	.bind = gf100_gr_chan_bind,
382 };
383 
384 static int
gf100_gr_chan_new(struct nvkm_gr * base,struct nvkm_fifo_chan * fifoch,const struct nvkm_oclass * oclass,struct nvkm_object ** pobject)385 gf100_gr_chan_new(struct nvkm_gr *base, struct nvkm_fifo_chan *fifoch,
386 		  const struct nvkm_oclass *oclass,
387 		  struct nvkm_object **pobject)
388 {
389 	struct gf100_gr *gr = gf100_gr(base);
390 	struct gf100_gr_data *data = gr->mmio_data;
391 	struct gf100_gr_mmio *mmio = gr->mmio_list;
392 	struct gf100_gr_chan *chan;
393 	struct gf100_vmm_map_v0 args = { .priv = 1 };
394 	struct nvkm_device *device = gr->base.engine.subdev.device;
395 	int ret, i;
396 
397 	if (!(chan = kzalloc(sizeof(*chan), GFP_KERNEL)))
398 		return -ENOMEM;
399 	nvkm_object_ctor(&gf100_gr_chan, oclass, &chan->object);
400 	chan->gr = gr;
401 	chan->vmm = nvkm_vmm_ref(fifoch->vmm);
402 	*pobject = &chan->object;
403 
404 	/* allocate memory for a "mmio list" buffer that's used by the HUB
405 	 * fuc to modify some per-context register settings on first load
406 	 * of the context.
407 	 */
408 	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x1000, 0x100,
409 			      false, &chan->mmio);
410 	if (ret)
411 		return ret;
412 
413 	ret = nvkm_vmm_get(fifoch->vmm, 12, 0x1000, &chan->mmio_vma);
414 	if (ret)
415 		return ret;
416 
417 	ret = nvkm_memory_map(chan->mmio, 0, fifoch->vmm,
418 			      chan->mmio_vma, &args, sizeof(args));
419 	if (ret)
420 		return ret;
421 
422 	/* allocate buffers referenced by mmio list */
423 	for (i = 0; data->size && i < ARRAY_SIZE(gr->mmio_data); i++) {
424 		ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST,
425 				      data->size, data->align, false,
426 				      &chan->data[i].mem);
427 		if (ret)
428 			return ret;
429 
430 		ret = nvkm_vmm_get(fifoch->vmm, 12,
431 				   nvkm_memory_size(chan->data[i].mem),
432 				   &chan->data[i].vma);
433 		if (ret)
434 			return ret;
435 
436 		args.priv = data->priv;
437 
438 		ret = nvkm_memory_map(chan->data[i].mem, 0, chan->vmm,
439 				      chan->data[i].vma, &args, sizeof(args));
440 		if (ret)
441 			return ret;
442 
443 		data++;
444 	}
445 
446 	/* finally, fill in the mmio list and point the context at it */
447 	nvkm_kmap(chan->mmio);
448 	for (i = 0; mmio->addr && i < ARRAY_SIZE(gr->mmio_list); i++) {
449 		u32 addr = mmio->addr;
450 		u32 data = mmio->data;
451 
452 		if (mmio->buffer >= 0) {
453 			u64 info = chan->data[mmio->buffer].vma->addr;
454 			data |= info >> mmio->shift;
455 		}
456 
457 		nvkm_wo32(chan->mmio, chan->mmio_nr++ * 4, addr);
458 		nvkm_wo32(chan->mmio, chan->mmio_nr++ * 4, data);
459 		mmio++;
460 	}
461 	nvkm_done(chan->mmio);
462 	return 0;
463 }
464 
465 /*******************************************************************************
466  * PGRAPH register lists
467  ******************************************************************************/
468 
469 const struct gf100_gr_init
470 gf100_gr_init_main_0[] = {
471 	{ 0x400080,   1, 0x04, 0x003083c2 },
472 	{ 0x400088,   1, 0x04, 0x00006fe7 },
473 	{ 0x40008c,   1, 0x04, 0x00000000 },
474 	{ 0x400090,   1, 0x04, 0x00000030 },
475 	{ 0x40013c,   1, 0x04, 0x013901f7 },
476 	{ 0x400140,   1, 0x04, 0x00000100 },
477 	{ 0x400144,   1, 0x04, 0x00000000 },
478 	{ 0x400148,   1, 0x04, 0x00000110 },
479 	{ 0x400138,   1, 0x04, 0x00000000 },
480 	{ 0x400130,   2, 0x04, 0x00000000 },
481 	{ 0x400124,   1, 0x04, 0x00000002 },
482 	{}
483 };
484 
485 const struct gf100_gr_init
486 gf100_gr_init_fe_0[] = {
487 	{ 0x40415c,   1, 0x04, 0x00000000 },
488 	{ 0x404170,   1, 0x04, 0x00000000 },
489 	{}
490 };
491 
492 const struct gf100_gr_init
493 gf100_gr_init_pri_0[] = {
494 	{ 0x404488,   2, 0x04, 0x00000000 },
495 	{}
496 };
497 
498 const struct gf100_gr_init
499 gf100_gr_init_rstr2d_0[] = {
500 	{ 0x407808,   1, 0x04, 0x00000000 },
501 	{}
502 };
503 
504 const struct gf100_gr_init
505 gf100_gr_init_pd_0[] = {
506 	{ 0x406024,   1, 0x04, 0x00000000 },
507 	{}
508 };
509 
510 const struct gf100_gr_init
511 gf100_gr_init_ds_0[] = {
512 	{ 0x405844,   1, 0x04, 0x00ffffff },
513 	{ 0x405850,   1, 0x04, 0x00000000 },
514 	{ 0x405908,   1, 0x04, 0x00000000 },
515 	{}
516 };
517 
518 const struct gf100_gr_init
519 gf100_gr_init_scc_0[] = {
520 	{ 0x40803c,   1, 0x04, 0x00000000 },
521 	{}
522 };
523 
524 const struct gf100_gr_init
525 gf100_gr_init_prop_0[] = {
526 	{ 0x4184a0,   1, 0x04, 0x00000000 },
527 	{}
528 };
529 
530 const struct gf100_gr_init
531 gf100_gr_init_gpc_unk_0[] = {
532 	{ 0x418604,   1, 0x04, 0x00000000 },
533 	{ 0x418680,   1, 0x04, 0x00000000 },
534 	{ 0x418714,   1, 0x04, 0x80000000 },
535 	{ 0x418384,   1, 0x04, 0x00000000 },
536 	{}
537 };
538 
539 const struct gf100_gr_init
540 gf100_gr_init_setup_0[] = {
541 	{ 0x418814,   3, 0x04, 0x00000000 },
542 	{}
543 };
544 
545 const struct gf100_gr_init
546 gf100_gr_init_crstr_0[] = {
547 	{ 0x418b04,   1, 0x04, 0x00000000 },
548 	{}
549 };
550 
551 const struct gf100_gr_init
552 gf100_gr_init_setup_1[] = {
553 	{ 0x4188c8,   1, 0x04, 0x80000000 },
554 	{ 0x4188cc,   1, 0x04, 0x00000000 },
555 	{ 0x4188d0,   1, 0x04, 0x00010000 },
556 	{ 0x4188d4,   1, 0x04, 0x00000001 },
557 	{}
558 };
559 
560 const struct gf100_gr_init
561 gf100_gr_init_zcull_0[] = {
562 	{ 0x418910,   1, 0x04, 0x00010001 },
563 	{ 0x418914,   1, 0x04, 0x00000301 },
564 	{ 0x418918,   1, 0x04, 0x00800000 },
565 	{ 0x418980,   1, 0x04, 0x77777770 },
566 	{ 0x418984,   3, 0x04, 0x77777777 },
567 	{}
568 };
569 
570 const struct gf100_gr_init
571 gf100_gr_init_gpm_0[] = {
572 	{ 0x418c04,   1, 0x04, 0x00000000 },
573 	{ 0x418c88,   1, 0x04, 0x00000000 },
574 	{}
575 };
576 
577 const struct gf100_gr_init
578 gf100_gr_init_gpc_unk_1[] = {
579 	{ 0x418d00,   1, 0x04, 0x00000000 },
580 	{ 0x418f08,   1, 0x04, 0x00000000 },
581 	{ 0x418e00,   1, 0x04, 0x00000050 },
582 	{ 0x418e08,   1, 0x04, 0x00000000 },
583 	{}
584 };
585 
586 const struct gf100_gr_init
587 gf100_gr_init_gcc_0[] = {
588 	{ 0x41900c,   1, 0x04, 0x00000000 },
589 	{ 0x419018,   1, 0x04, 0x00000000 },
590 	{}
591 };
592 
593 const struct gf100_gr_init
594 gf100_gr_init_tpccs_0[] = {
595 	{ 0x419d08,   2, 0x04, 0x00000000 },
596 	{ 0x419d10,   1, 0x04, 0x00000014 },
597 	{}
598 };
599 
600 const struct gf100_gr_init
601 gf100_gr_init_tex_0[] = {
602 	{ 0x419ab0,   1, 0x04, 0x00000000 },
603 	{ 0x419ab8,   1, 0x04, 0x000000e7 },
604 	{ 0x419abc,   2, 0x04, 0x00000000 },
605 	{}
606 };
607 
608 const struct gf100_gr_init
609 gf100_gr_init_pe_0[] = {
610 	{ 0x41980c,   3, 0x04, 0x00000000 },
611 	{ 0x419844,   1, 0x04, 0x00000000 },
612 	{ 0x41984c,   1, 0x04, 0x00005bc5 },
613 	{ 0x419850,   4, 0x04, 0x00000000 },
614 	{}
615 };
616 
617 const struct gf100_gr_init
618 gf100_gr_init_l1c_0[] = {
619 	{ 0x419c98,   1, 0x04, 0x00000000 },
620 	{ 0x419ca8,   1, 0x04, 0x80000000 },
621 	{ 0x419cb4,   1, 0x04, 0x00000000 },
622 	{ 0x419cb8,   1, 0x04, 0x00008bf4 },
623 	{ 0x419cbc,   1, 0x04, 0x28137606 },
624 	{ 0x419cc0,   2, 0x04, 0x00000000 },
625 	{}
626 };
627 
628 const struct gf100_gr_init
629 gf100_gr_init_wwdx_0[] = {
630 	{ 0x419bd4,   1, 0x04, 0x00800000 },
631 	{ 0x419bdc,   1, 0x04, 0x00000000 },
632 	{}
633 };
634 
635 const struct gf100_gr_init
636 gf100_gr_init_tpccs_1[] = {
637 	{ 0x419d2c,   1, 0x04, 0x00000000 },
638 	{}
639 };
640 
641 const struct gf100_gr_init
642 gf100_gr_init_mpc_0[] = {
643 	{ 0x419c0c,   1, 0x04, 0x00000000 },
644 	{}
645 };
646 
647 static const struct gf100_gr_init
648 gf100_gr_init_sm_0[] = {
649 	{ 0x419e00,   1, 0x04, 0x00000000 },
650 	{ 0x419ea0,   1, 0x04, 0x00000000 },
651 	{ 0x419ea4,   1, 0x04, 0x00000100 },
652 	{ 0x419ea8,   1, 0x04, 0x00001100 },
653 	{ 0x419eac,   1, 0x04, 0x11100702 },
654 	{ 0x419eb0,   1, 0x04, 0x00000003 },
655 	{ 0x419eb4,   4, 0x04, 0x00000000 },
656 	{ 0x419ec8,   1, 0x04, 0x06060618 },
657 	{ 0x419ed0,   1, 0x04, 0x0eff0e38 },
658 	{ 0x419ed4,   1, 0x04, 0x011104f1 },
659 	{ 0x419edc,   1, 0x04, 0x00000000 },
660 	{ 0x419f00,   1, 0x04, 0x00000000 },
661 	{ 0x419f2c,   1, 0x04, 0x00000000 },
662 	{}
663 };
664 
665 const struct gf100_gr_init
666 gf100_gr_init_be_0[] = {
667 	{ 0x40880c,   1, 0x04, 0x00000000 },
668 	{ 0x408910,   9, 0x04, 0x00000000 },
669 	{ 0x408950,   1, 0x04, 0x00000000 },
670 	{ 0x408954,   1, 0x04, 0x0000ffff },
671 	{ 0x408984,   1, 0x04, 0x00000000 },
672 	{ 0x408988,   1, 0x04, 0x08040201 },
673 	{ 0x40898c,   1, 0x04, 0x80402010 },
674 	{}
675 };
676 
677 const struct gf100_gr_init
678 gf100_gr_init_fe_1[] = {
679 	{ 0x4040f0,   1, 0x04, 0x00000000 },
680 	{}
681 };
682 
683 const struct gf100_gr_init
684 gf100_gr_init_pe_1[] = {
685 	{ 0x419880,   1, 0x04, 0x00000002 },
686 	{}
687 };
688 
689 static const struct gf100_gr_pack
690 gf100_gr_pack_mmio[] = {
691 	{ gf100_gr_init_main_0 },
692 	{ gf100_gr_init_fe_0 },
693 	{ gf100_gr_init_pri_0 },
694 	{ gf100_gr_init_rstr2d_0 },
695 	{ gf100_gr_init_pd_0 },
696 	{ gf100_gr_init_ds_0 },
697 	{ gf100_gr_init_scc_0 },
698 	{ gf100_gr_init_prop_0 },
699 	{ gf100_gr_init_gpc_unk_0 },
700 	{ gf100_gr_init_setup_0 },
701 	{ gf100_gr_init_crstr_0 },
702 	{ gf100_gr_init_setup_1 },
703 	{ gf100_gr_init_zcull_0 },
704 	{ gf100_gr_init_gpm_0 },
705 	{ gf100_gr_init_gpc_unk_1 },
706 	{ gf100_gr_init_gcc_0 },
707 	{ gf100_gr_init_tpccs_0 },
708 	{ gf100_gr_init_tex_0 },
709 	{ gf100_gr_init_pe_0 },
710 	{ gf100_gr_init_l1c_0 },
711 	{ gf100_gr_init_wwdx_0 },
712 	{ gf100_gr_init_tpccs_1 },
713 	{ gf100_gr_init_mpc_0 },
714 	{ gf100_gr_init_sm_0 },
715 	{ gf100_gr_init_be_0 },
716 	{ gf100_gr_init_fe_1 },
717 	{ gf100_gr_init_pe_1 },
718 	{}
719 };
720 
721 /*******************************************************************************
722  * PGRAPH engine/subdev functions
723  ******************************************************************************/
724 
725 static u32
gf100_gr_ctxsw_inst(struct nvkm_gr * gr)726 gf100_gr_ctxsw_inst(struct nvkm_gr *gr)
727 {
728 	return nvkm_rd32(gr->engine.subdev.device, 0x409b00);
729 }
730 
731 static int
gf100_gr_fecs_ctrl_ctxsw(struct gf100_gr * gr,u32 mthd)732 gf100_gr_fecs_ctrl_ctxsw(struct gf100_gr *gr, u32 mthd)
733 {
734 	struct nvkm_device *device = gr->base.engine.subdev.device;
735 
736 	nvkm_wr32(device, 0x409804, 0xffffffff);
737 	nvkm_wr32(device, 0x409840, 0xffffffff);
738 	nvkm_wr32(device, 0x409500, 0xffffffff);
739 	nvkm_wr32(device, 0x409504, mthd);
740 	nvkm_msec(device, 2000,
741 		u32 stat = nvkm_rd32(device, 0x409804);
742 		if (stat == 0x00000002)
743 			return -EIO;
744 		if (stat == 0x00000001)
745 			return 0;
746 	);
747 
748 	return -ETIMEDOUT;
749 }
750 
751 static int
gf100_gr_fecs_start_ctxsw(struct nvkm_gr * base)752 gf100_gr_fecs_start_ctxsw(struct nvkm_gr *base)
753 {
754 	struct gf100_gr *gr = gf100_gr(base);
755 	int ret = 0;
756 
757 	mutex_lock(&gr->fecs.mutex);
758 	if (!--gr->fecs.disable) {
759 		if (WARN_ON(ret = gf100_gr_fecs_ctrl_ctxsw(gr, 0x39)))
760 			gr->fecs.disable++;
761 	}
762 	mutex_unlock(&gr->fecs.mutex);
763 	return ret;
764 }
765 
766 static int
gf100_gr_fecs_stop_ctxsw(struct nvkm_gr * base)767 gf100_gr_fecs_stop_ctxsw(struct nvkm_gr *base)
768 {
769 	struct gf100_gr *gr = gf100_gr(base);
770 	int ret = 0;
771 
772 	mutex_lock(&gr->fecs.mutex);
773 	if (!gr->fecs.disable++) {
774 		if (WARN_ON(ret = gf100_gr_fecs_ctrl_ctxsw(gr, 0x38)))
775 			gr->fecs.disable--;
776 	}
777 	mutex_unlock(&gr->fecs.mutex);
778 	return ret;
779 }
780 
781 int
gf100_gr_fecs_bind_pointer(struct gf100_gr * gr,u32 inst)782 gf100_gr_fecs_bind_pointer(struct gf100_gr *gr, u32 inst)
783 {
784 	struct nvkm_device *device = gr->base.engine.subdev.device;
785 
786 	nvkm_wr32(device, 0x409840, 0x00000030);
787 	nvkm_wr32(device, 0x409500, inst);
788 	nvkm_wr32(device, 0x409504, 0x00000003);
789 	nvkm_msec(device, 2000,
790 		u32 stat = nvkm_rd32(device, 0x409800);
791 		if (stat & 0x00000020)
792 			return -EIO;
793 		if (stat & 0x00000010)
794 			return 0;
795 	);
796 
797 	return -ETIMEDOUT;
798 }
799 
800 static int
gf100_gr_fecs_set_reglist_virtual_address(struct gf100_gr * gr,u64 addr)801 gf100_gr_fecs_set_reglist_virtual_address(struct gf100_gr *gr, u64 addr)
802 {
803 	struct nvkm_device *device = gr->base.engine.subdev.device;
804 
805 	nvkm_wr32(device, 0x409810, addr >> 8);
806 	nvkm_wr32(device, 0x409800, 0x00000000);
807 	nvkm_wr32(device, 0x409500, 0x00000001);
808 	nvkm_wr32(device, 0x409504, 0x00000032);
809 	nvkm_msec(device, 2000,
810 		if (nvkm_rd32(device, 0x409800) == 0x00000001)
811 			return 0;
812 	);
813 
814 	return -ETIMEDOUT;
815 }
816 
817 static int
gf100_gr_fecs_set_reglist_bind_instance(struct gf100_gr * gr,u32 inst)818 gf100_gr_fecs_set_reglist_bind_instance(struct gf100_gr *gr, u32 inst)
819 {
820 	struct nvkm_device *device = gr->base.engine.subdev.device;
821 
822 	nvkm_wr32(device, 0x409810, inst);
823 	nvkm_wr32(device, 0x409800, 0x00000000);
824 	nvkm_wr32(device, 0x409500, 0x00000001);
825 	nvkm_wr32(device, 0x409504, 0x00000031);
826 	nvkm_msec(device, 2000,
827 		if (nvkm_rd32(device, 0x409800) == 0x00000001)
828 			return 0;
829 	);
830 
831 	return -ETIMEDOUT;
832 }
833 
834 static int
gf100_gr_fecs_discover_reglist_image_size(struct gf100_gr * gr,u32 * psize)835 gf100_gr_fecs_discover_reglist_image_size(struct gf100_gr *gr, u32 *psize)
836 {
837 	struct nvkm_device *device = gr->base.engine.subdev.device;
838 
839 	nvkm_wr32(device, 0x409800, 0x00000000);
840 	nvkm_wr32(device, 0x409500, 0x00000001);
841 	nvkm_wr32(device, 0x409504, 0x00000030);
842 	nvkm_msec(device, 2000,
843 		if ((*psize = nvkm_rd32(device, 0x409800)))
844 			return 0;
845 	);
846 
847 	return -ETIMEDOUT;
848 }
849 
850 static int
gf100_gr_fecs_elpg_bind(struct gf100_gr * gr)851 gf100_gr_fecs_elpg_bind(struct gf100_gr *gr)
852 {
853 	u32 size;
854 	int ret;
855 
856 	ret = gf100_gr_fecs_discover_reglist_image_size(gr, &size);
857 	if (ret)
858 		return ret;
859 
860 	/*XXX: We need to allocate + map the above into PMU's inst block,
861 	 *     which which means we probably need a proper PMU before we
862 	 *     even bother.
863 	 */
864 
865 	ret = gf100_gr_fecs_set_reglist_bind_instance(gr, 0);
866 	if (ret)
867 		return ret;
868 
869 	return gf100_gr_fecs_set_reglist_virtual_address(gr, 0);
870 }
871 
872 static int
gf100_gr_fecs_discover_pm_image_size(struct gf100_gr * gr,u32 * psize)873 gf100_gr_fecs_discover_pm_image_size(struct gf100_gr *gr, u32 *psize)
874 {
875 	struct nvkm_device *device = gr->base.engine.subdev.device;
876 
877 	nvkm_wr32(device, 0x409840, 0xffffffff);
878 	nvkm_wr32(device, 0x409500, 0x00000000);
879 	nvkm_wr32(device, 0x409504, 0x00000025);
880 	nvkm_msec(device, 2000,
881 		if ((*psize = nvkm_rd32(device, 0x409800)))
882 			return 0;
883 	);
884 
885 	return -ETIMEDOUT;
886 }
887 
888 static int
gf100_gr_fecs_discover_zcull_image_size(struct gf100_gr * gr,u32 * psize)889 gf100_gr_fecs_discover_zcull_image_size(struct gf100_gr *gr, u32 *psize)
890 {
891 	struct nvkm_device *device = gr->base.engine.subdev.device;
892 
893 	nvkm_wr32(device, 0x409840, 0xffffffff);
894 	nvkm_wr32(device, 0x409500, 0x00000000);
895 	nvkm_wr32(device, 0x409504, 0x00000016);
896 	nvkm_msec(device, 2000,
897 		if ((*psize = nvkm_rd32(device, 0x409800)))
898 			return 0;
899 	);
900 
901 	return -ETIMEDOUT;
902 }
903 
904 static int
gf100_gr_fecs_discover_image_size(struct gf100_gr * gr,u32 * psize)905 gf100_gr_fecs_discover_image_size(struct gf100_gr *gr, u32 *psize)
906 {
907 	struct nvkm_device *device = gr->base.engine.subdev.device;
908 
909 	nvkm_wr32(device, 0x409840, 0xffffffff);
910 	nvkm_wr32(device, 0x409500, 0x00000000);
911 	nvkm_wr32(device, 0x409504, 0x00000010);
912 	nvkm_msec(device, 2000,
913 		if ((*psize = nvkm_rd32(device, 0x409800)))
914 			return 0;
915 	);
916 
917 	return -ETIMEDOUT;
918 }
919 
920 static void
gf100_gr_fecs_set_watchdog_timeout(struct gf100_gr * gr,u32 timeout)921 gf100_gr_fecs_set_watchdog_timeout(struct gf100_gr *gr, u32 timeout)
922 {
923 	struct nvkm_device *device = gr->base.engine.subdev.device;
924 
925 	nvkm_wr32(device, 0x409840, 0xffffffff);
926 	nvkm_wr32(device, 0x409500, timeout);
927 	nvkm_wr32(device, 0x409504, 0x00000021);
928 }
929 
930 static bool
gf100_gr_chsw_load(struct nvkm_gr * base)931 gf100_gr_chsw_load(struct nvkm_gr *base)
932 {
933 	struct gf100_gr *gr = gf100_gr(base);
934 	if (!gr->firmware) {
935 		u32 trace = nvkm_rd32(gr->base.engine.subdev.device, 0x40981c);
936 		if (trace & 0x00000040)
937 			return true;
938 	} else {
939 		u32 mthd = nvkm_rd32(gr->base.engine.subdev.device, 0x409808);
940 		if (mthd & 0x00080000)
941 			return true;
942 	}
943 	return false;
944 }
945 
946 int
gf100_gr_rops(struct gf100_gr * gr)947 gf100_gr_rops(struct gf100_gr *gr)
948 {
949 	struct nvkm_device *device = gr->base.engine.subdev.device;
950 	return (nvkm_rd32(device, 0x409604) & 0x001f0000) >> 16;
951 }
952 
953 void
gf100_gr_zbc_init(struct gf100_gr * gr)954 gf100_gr_zbc_init(struct gf100_gr *gr)
955 {
956 	const u32  zero[] = { 0x00000000, 0x00000000, 0x00000000, 0x00000000,
957 			      0x00000000, 0x00000000, 0x00000000, 0x00000000 };
958 	const u32   one[] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000,
959 			      0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
960 	const u32 f32_0[] = { 0x00000000, 0x00000000, 0x00000000, 0x00000000,
961 			      0x00000000, 0x00000000, 0x00000000, 0x00000000 };
962 	const u32 f32_1[] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000,
963 			      0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 };
964 	struct nvkm_ltc *ltc = gr->base.engine.subdev.device->ltc;
965 	int index, c = ltc->zbc_min, d = ltc->zbc_min, s = ltc->zbc_min;
966 
967 	if (!gr->zbc_color[0].format) {
968 		gf100_gr_zbc_color_get(gr, 1,  & zero[0],   &zero[4]); c++;
969 		gf100_gr_zbc_color_get(gr, 2,  &  one[0],    &one[4]); c++;
970 		gf100_gr_zbc_color_get(gr, 4,  &f32_0[0],  &f32_0[4]); c++;
971 		gf100_gr_zbc_color_get(gr, 4,  &f32_1[0],  &f32_1[4]); c++;
972 		gf100_gr_zbc_depth_get(gr, 1, 0x00000000, 0x00000000); d++;
973 		gf100_gr_zbc_depth_get(gr, 1, 0x3f800000, 0x3f800000); d++;
974 		if (gr->func->zbc->stencil_get) {
975 			gr->func->zbc->stencil_get(gr, 1, 0x00, 0x00); s++;
976 			gr->func->zbc->stencil_get(gr, 1, 0x01, 0x01); s++;
977 			gr->func->zbc->stencil_get(gr, 1, 0xff, 0xff); s++;
978 		}
979 	}
980 
981 	for (index = c; index <= ltc->zbc_max; index++)
982 		gr->func->zbc->clear_color(gr, index);
983 	for (index = d; index <= ltc->zbc_max; index++)
984 		gr->func->zbc->clear_depth(gr, index);
985 
986 	if (gr->func->zbc->clear_stencil) {
987 		for (index = s; index <= ltc->zbc_max; index++)
988 			gr->func->zbc->clear_stencil(gr, index);
989 	}
990 }
991 
992 /**
993  * Wait until GR goes idle. GR is considered idle if it is disabled by the
994  * MC (0x200) register, or GR is not busy and a context switch is not in
995  * progress.
996  */
997 int
gf100_gr_wait_idle(struct gf100_gr * gr)998 gf100_gr_wait_idle(struct gf100_gr *gr)
999 {
1000 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1001 	struct nvkm_device *device = subdev->device;
1002 	unsigned long end_jiffies = jiffies + msecs_to_jiffies(2000);
1003 	bool gr_enabled, ctxsw_active, gr_busy;
1004 
1005 	do {
1006 		/*
1007 		 * required to make sure FIFO_ENGINE_STATUS (0x2640) is
1008 		 * up-to-date
1009 		 */
1010 		nvkm_rd32(device, 0x400700);
1011 
1012 		gr_enabled = nvkm_rd32(device, 0x200) & 0x1000;
1013 		ctxsw_active = nvkm_rd32(device, 0x2640) & 0x8000;
1014 		gr_busy = nvkm_rd32(device, 0x40060c) & 0x1;
1015 
1016 		if (!gr_enabled || (!gr_busy && !ctxsw_active))
1017 			return 0;
1018 	} while (time_before(jiffies, end_jiffies));
1019 
1020 	nvkm_error(subdev,
1021 		   "wait for idle timeout (en: %d, ctxsw: %d, busy: %d)\n",
1022 		   gr_enabled, ctxsw_active, gr_busy);
1023 	return -EAGAIN;
1024 }
1025 
1026 void
gf100_gr_mmio(struct gf100_gr * gr,const struct gf100_gr_pack * p)1027 gf100_gr_mmio(struct gf100_gr *gr, const struct gf100_gr_pack *p)
1028 {
1029 	struct nvkm_device *device = gr->base.engine.subdev.device;
1030 	const struct gf100_gr_pack *pack;
1031 	const struct gf100_gr_init *init;
1032 
1033 	pack_for_each_init(init, pack, p) {
1034 		u32 next = init->addr + init->count * init->pitch;
1035 		u32 addr = init->addr;
1036 		while (addr < next) {
1037 			nvkm_wr32(device, addr, init->data);
1038 			addr += init->pitch;
1039 		}
1040 	}
1041 }
1042 
1043 void
gf100_gr_icmd(struct gf100_gr * gr,const struct gf100_gr_pack * p)1044 gf100_gr_icmd(struct gf100_gr *gr, const struct gf100_gr_pack *p)
1045 {
1046 	struct nvkm_device *device = gr->base.engine.subdev.device;
1047 	const struct gf100_gr_pack *pack;
1048 	const struct gf100_gr_init *init;
1049 	u32 data = 0;
1050 
1051 	nvkm_wr32(device, 0x400208, 0x80000000);
1052 
1053 	pack_for_each_init(init, pack, p) {
1054 		u32 next = init->addr + init->count * init->pitch;
1055 		u32 addr = init->addr;
1056 
1057 		if ((pack == p && init == p->init) || data != init->data) {
1058 			nvkm_wr32(device, 0x400204, init->data);
1059 			data = init->data;
1060 		}
1061 
1062 		while (addr < next) {
1063 			nvkm_wr32(device, 0x400200, addr);
1064 			/**
1065 			 * Wait for GR to go idle after submitting a
1066 			 * GO_IDLE bundle
1067 			 */
1068 			if ((addr & 0xffff) == 0xe100)
1069 				gf100_gr_wait_idle(gr);
1070 			nvkm_msec(device, 2000,
1071 				if (!(nvkm_rd32(device, 0x400700) & 0x00000004))
1072 					break;
1073 			);
1074 			addr += init->pitch;
1075 		}
1076 	}
1077 
1078 	nvkm_wr32(device, 0x400208, 0x00000000);
1079 }
1080 
1081 void
gf100_gr_mthd(struct gf100_gr * gr,const struct gf100_gr_pack * p)1082 gf100_gr_mthd(struct gf100_gr *gr, const struct gf100_gr_pack *p)
1083 {
1084 	struct nvkm_device *device = gr->base.engine.subdev.device;
1085 	const struct gf100_gr_pack *pack;
1086 	const struct gf100_gr_init *init;
1087 	u32 data = 0;
1088 
1089 	pack_for_each_init(init, pack, p) {
1090 		u32 ctrl = 0x80000000 | pack->type;
1091 		u32 next = init->addr + init->count * init->pitch;
1092 		u32 addr = init->addr;
1093 
1094 		if ((pack == p && init == p->init) || data != init->data) {
1095 			nvkm_wr32(device, 0x40448c, init->data);
1096 			data = init->data;
1097 		}
1098 
1099 		while (addr < next) {
1100 			nvkm_wr32(device, 0x404488, ctrl | (addr << 14));
1101 			addr += init->pitch;
1102 		}
1103 	}
1104 }
1105 
1106 u64
gf100_gr_units(struct nvkm_gr * base)1107 gf100_gr_units(struct nvkm_gr *base)
1108 {
1109 	struct gf100_gr *gr = gf100_gr(base);
1110 	u64 cfg;
1111 
1112 	cfg  = (u32)gr->gpc_nr;
1113 	cfg |= (u32)gr->tpc_total << 8;
1114 	cfg |= (u64)gr->rop_nr << 32;
1115 
1116 	return cfg;
1117 }
1118 
1119 static const struct nvkm_bitfield gf100_dispatch_error[] = {
1120 	{ 0x00000001, "INJECTED_BUNDLE_ERROR" },
1121 	{ 0x00000002, "CLASS_SUBCH_MISMATCH" },
1122 	{ 0x00000004, "SUBCHSW_DURING_NOTIFY" },
1123 	{}
1124 };
1125 
1126 static const struct nvkm_bitfield gf100_m2mf_error[] = {
1127 	{ 0x00000001, "PUSH_TOO_MUCH_DATA" },
1128 	{ 0x00000002, "PUSH_NOT_ENOUGH_DATA" },
1129 	{}
1130 };
1131 
1132 static const struct nvkm_bitfield gf100_unk6_error[] = {
1133 	{ 0x00000001, "TEMP_TOO_SMALL" },
1134 	{}
1135 };
1136 
1137 static const struct nvkm_bitfield gf100_ccache_error[] = {
1138 	{ 0x00000001, "INTR" },
1139 	{ 0x00000002, "LDCONST_OOB" },
1140 	{}
1141 };
1142 
1143 static const struct nvkm_bitfield gf100_macro_error[] = {
1144 	{ 0x00000001, "TOO_FEW_PARAMS" },
1145 	{ 0x00000002, "TOO_MANY_PARAMS" },
1146 	{ 0x00000004, "ILLEGAL_OPCODE" },
1147 	{ 0x00000008, "DOUBLE_BRANCH" },
1148 	{ 0x00000010, "WATCHDOG" },
1149 	{}
1150 };
1151 
1152 static const struct nvkm_bitfield gk104_sked_error[] = {
1153 	{ 0x00000040, "CTA_RESUME" },
1154 	{ 0x00000080, "CONSTANT_BUFFER_SIZE" },
1155 	{ 0x00000200, "LOCAL_MEMORY_SIZE_POS" },
1156 	{ 0x00000400, "LOCAL_MEMORY_SIZE_NEG" },
1157 	{ 0x00000800, "WARP_CSTACK_SIZE" },
1158 	{ 0x00001000, "TOTAL_TEMP_SIZE" },
1159 	{ 0x00002000, "REGISTER_COUNT" },
1160 	{ 0x00040000, "TOTAL_THREADS" },
1161 	{ 0x00100000, "PROGRAM_OFFSET" },
1162 	{ 0x00200000, "SHARED_MEMORY_SIZE" },
1163 	{ 0x00800000, "CTA_THREAD_DIMENSION_ZERO" },
1164 	{ 0x01000000, "MEMORY_WINDOW_OVERLAP" },
1165 	{ 0x02000000, "SHARED_CONFIG_TOO_SMALL" },
1166 	{ 0x04000000, "TOTAL_REGISTER_COUNT" },
1167 	{}
1168 };
1169 
1170 static const struct nvkm_bitfield gf100_gpc_rop_error[] = {
1171 	{ 0x00000002, "RT_PITCH_OVERRUN" },
1172 	{ 0x00000010, "RT_WIDTH_OVERRUN" },
1173 	{ 0x00000020, "RT_HEIGHT_OVERRUN" },
1174 	{ 0x00000080, "ZETA_STORAGE_TYPE_MISMATCH" },
1175 	{ 0x00000100, "RT_STORAGE_TYPE_MISMATCH" },
1176 	{ 0x00000400, "RT_LINEAR_MISMATCH" },
1177 	{}
1178 };
1179 
1180 static void
gf100_gr_trap_gpc_rop(struct gf100_gr * gr,int gpc)1181 gf100_gr_trap_gpc_rop(struct gf100_gr *gr, int gpc)
1182 {
1183 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1184 	struct nvkm_device *device = subdev->device;
1185 	char error[128];
1186 	u32 trap[4];
1187 
1188 	trap[0] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0420)) & 0x3fffffff;
1189 	trap[1] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0434));
1190 	trap[2] = nvkm_rd32(device, GPC_UNIT(gpc, 0x0438));
1191 	trap[3] = nvkm_rd32(device, GPC_UNIT(gpc, 0x043c));
1192 
1193 	nvkm_snprintbf(error, sizeof(error), gf100_gpc_rop_error, trap[0]);
1194 
1195 	nvkm_error(subdev, "GPC%d/PROP trap: %08x [%s] x = %u, y = %u, "
1196 			   "format = %x, storage type = %x\n",
1197 		   gpc, trap[0], error, trap[1] & 0xffff, trap[1] >> 16,
1198 		   (trap[2] >> 8) & 0x3f, trap[3] & 0xff);
1199 	nvkm_wr32(device, GPC_UNIT(gpc, 0x0420), 0xc0000000);
1200 }
1201 
1202 const struct nvkm_enum gf100_mp_warp_error[] = {
1203 	{ 0x01, "STACK_ERROR" },
1204 	{ 0x02, "API_STACK_ERROR" },
1205 	{ 0x03, "RET_EMPTY_STACK_ERROR" },
1206 	{ 0x04, "PC_WRAP" },
1207 	{ 0x05, "MISALIGNED_PC" },
1208 	{ 0x06, "PC_OVERFLOW" },
1209 	{ 0x07, "MISALIGNED_IMMC_ADDR" },
1210 	{ 0x08, "MISALIGNED_REG" },
1211 	{ 0x09, "ILLEGAL_INSTR_ENCODING" },
1212 	{ 0x0a, "ILLEGAL_SPH_INSTR_COMBO" },
1213 	{ 0x0b, "ILLEGAL_INSTR_PARAM" },
1214 	{ 0x0c, "INVALID_CONST_ADDR" },
1215 	{ 0x0d, "OOR_REG" },
1216 	{ 0x0e, "OOR_ADDR" },
1217 	{ 0x0f, "MISALIGNED_ADDR" },
1218 	{ 0x10, "INVALID_ADDR_SPACE" },
1219 	{ 0x11, "ILLEGAL_INSTR_PARAM2" },
1220 	{ 0x12, "INVALID_CONST_ADDR_LDC" },
1221 	{ 0x13, "GEOMETRY_SM_ERROR" },
1222 	{ 0x14, "DIVERGENT" },
1223 	{ 0x15, "WARP_EXIT" },
1224 	{}
1225 };
1226 
1227 const struct nvkm_bitfield gf100_mp_global_error[] = {
1228 	{ 0x00000001, "SM_TO_SM_FAULT" },
1229 	{ 0x00000002, "L1_ERROR" },
1230 	{ 0x00000004, "MULTIPLE_WARP_ERRORS" },
1231 	{ 0x00000008, "PHYSICAL_STACK_OVERFLOW" },
1232 	{ 0x00000010, "BPT_INT" },
1233 	{ 0x00000020, "BPT_PAUSE" },
1234 	{ 0x00000040, "SINGLE_STEP_COMPLETE" },
1235 	{ 0x20000000, "ECC_SEC_ERROR" },
1236 	{ 0x40000000, "ECC_DED_ERROR" },
1237 	{ 0x80000000, "TIMEOUT" },
1238 	{}
1239 };
1240 
1241 void
gf100_gr_trap_mp(struct gf100_gr * gr,int gpc,int tpc)1242 gf100_gr_trap_mp(struct gf100_gr *gr, int gpc, int tpc)
1243 {
1244 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1245 	struct nvkm_device *device = subdev->device;
1246 	u32 werr = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x648));
1247 	u32 gerr = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x650));
1248 	const struct nvkm_enum *warp;
1249 	char glob[128];
1250 
1251 	nvkm_snprintbf(glob, sizeof(glob), gf100_mp_global_error, gerr);
1252 	warp = nvkm_enum_find(gf100_mp_warp_error, werr & 0xffff);
1253 
1254 	nvkm_error(subdev, "GPC%i/TPC%i/MP trap: "
1255 			   "global %08x [%s] warp %04x [%s]\n",
1256 		   gpc, tpc, gerr, glob, werr, warp ? warp->name : "");
1257 
1258 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x648), 0x00000000);
1259 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x650), gerr);
1260 }
1261 
1262 static void
gf100_gr_trap_tpc(struct gf100_gr * gr,int gpc,int tpc)1263 gf100_gr_trap_tpc(struct gf100_gr *gr, int gpc, int tpc)
1264 {
1265 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1266 	struct nvkm_device *device = subdev->device;
1267 	u32 stat = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0508));
1268 
1269 	if (stat & 0x00000001) {
1270 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0224));
1271 		nvkm_error(subdev, "GPC%d/TPC%d/TEX: %08x\n", gpc, tpc, trap);
1272 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x0224), 0xc0000000);
1273 		stat &= ~0x00000001;
1274 	}
1275 
1276 	if (stat & 0x00000002) {
1277 		gr->func->trap_mp(gr, gpc, tpc);
1278 		stat &= ~0x00000002;
1279 	}
1280 
1281 	if (stat & 0x00000004) {
1282 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0084));
1283 		nvkm_error(subdev, "GPC%d/TPC%d/POLY: %08x\n", gpc, tpc, trap);
1284 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x0084), 0xc0000000);
1285 		stat &= ~0x00000004;
1286 	}
1287 
1288 	if (stat & 0x00000008) {
1289 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x048c));
1290 		nvkm_error(subdev, "GPC%d/TPC%d/L1C: %08x\n", gpc, tpc, trap);
1291 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x048c), 0xc0000000);
1292 		stat &= ~0x00000008;
1293 	}
1294 
1295 	if (stat & 0x00000010) {
1296 		u32 trap = nvkm_rd32(device, TPC_UNIT(gpc, tpc, 0x0430));
1297 		nvkm_error(subdev, "GPC%d/TPC%d/MPC: %08x\n", gpc, tpc, trap);
1298 		nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x0430), 0xc0000000);
1299 		stat &= ~0x00000010;
1300 	}
1301 
1302 	if (stat) {
1303 		nvkm_error(subdev, "GPC%d/TPC%d/%08x: unknown\n", gpc, tpc, stat);
1304 	}
1305 }
1306 
1307 static void
gf100_gr_trap_gpc(struct gf100_gr * gr,int gpc)1308 gf100_gr_trap_gpc(struct gf100_gr *gr, int gpc)
1309 {
1310 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1311 	struct nvkm_device *device = subdev->device;
1312 	u32 stat = nvkm_rd32(device, GPC_UNIT(gpc, 0x2c90));
1313 	int tpc;
1314 
1315 	if (stat & 0x00000001) {
1316 		gf100_gr_trap_gpc_rop(gr, gpc);
1317 		stat &= ~0x00000001;
1318 	}
1319 
1320 	if (stat & 0x00000002) {
1321 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x0900));
1322 		nvkm_error(subdev, "GPC%d/ZCULL: %08x\n", gpc, trap);
1323 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0900), 0xc0000000);
1324 		stat &= ~0x00000002;
1325 	}
1326 
1327 	if (stat & 0x00000004) {
1328 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x1028));
1329 		nvkm_error(subdev, "GPC%d/CCACHE: %08x\n", gpc, trap);
1330 		nvkm_wr32(device, GPC_UNIT(gpc, 0x1028), 0xc0000000);
1331 		stat &= ~0x00000004;
1332 	}
1333 
1334 	if (stat & 0x00000008) {
1335 		u32 trap = nvkm_rd32(device, GPC_UNIT(gpc, 0x0824));
1336 		nvkm_error(subdev, "GPC%d/ESETUP: %08x\n", gpc, trap);
1337 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0824), 0xc0000000);
1338 		stat &= ~0x00000009;
1339 	}
1340 
1341 	for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++) {
1342 		u32 mask = 0x00010000 << tpc;
1343 		if (stat & mask) {
1344 			gf100_gr_trap_tpc(gr, gpc, tpc);
1345 			nvkm_wr32(device, GPC_UNIT(gpc, 0x2c90), mask);
1346 			stat &= ~mask;
1347 		}
1348 	}
1349 
1350 	if (stat) {
1351 		nvkm_error(subdev, "GPC%d/%08x: unknown\n", gpc, stat);
1352 	}
1353 }
1354 
1355 static void
gf100_gr_trap_intr(struct gf100_gr * gr)1356 gf100_gr_trap_intr(struct gf100_gr *gr)
1357 {
1358 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1359 	struct nvkm_device *device = subdev->device;
1360 	char error[128];
1361 	u32 trap = nvkm_rd32(device, 0x400108);
1362 	int rop, gpc;
1363 
1364 	if (trap & 0x00000001) {
1365 		u32 stat = nvkm_rd32(device, 0x404000);
1366 
1367 		nvkm_snprintbf(error, sizeof(error), gf100_dispatch_error,
1368 			       stat & 0x3fffffff);
1369 		nvkm_error(subdev, "DISPATCH %08x [%s]\n", stat, error);
1370 		nvkm_wr32(device, 0x404000, 0xc0000000);
1371 		nvkm_wr32(device, 0x400108, 0x00000001);
1372 		trap &= ~0x00000001;
1373 	}
1374 
1375 	if (trap & 0x00000002) {
1376 		u32 stat = nvkm_rd32(device, 0x404600);
1377 
1378 		nvkm_snprintbf(error, sizeof(error), gf100_m2mf_error,
1379 			       stat & 0x3fffffff);
1380 		nvkm_error(subdev, "M2MF %08x [%s]\n", stat, error);
1381 
1382 		nvkm_wr32(device, 0x404600, 0xc0000000);
1383 		nvkm_wr32(device, 0x400108, 0x00000002);
1384 		trap &= ~0x00000002;
1385 	}
1386 
1387 	if (trap & 0x00000008) {
1388 		u32 stat = nvkm_rd32(device, 0x408030);
1389 
1390 		nvkm_snprintbf(error, sizeof(error), gf100_ccache_error,
1391 			       stat & 0x3fffffff);
1392 		nvkm_error(subdev, "CCACHE %08x [%s]\n", stat, error);
1393 		nvkm_wr32(device, 0x408030, 0xc0000000);
1394 		nvkm_wr32(device, 0x400108, 0x00000008);
1395 		trap &= ~0x00000008;
1396 	}
1397 
1398 	if (trap & 0x00000010) {
1399 		u32 stat = nvkm_rd32(device, 0x405840);
1400 		nvkm_error(subdev, "SHADER %08x, sph: 0x%06x, stage: 0x%02x\n",
1401 			   stat, stat & 0xffffff, (stat >> 24) & 0x3f);
1402 		nvkm_wr32(device, 0x405840, 0xc0000000);
1403 		nvkm_wr32(device, 0x400108, 0x00000010);
1404 		trap &= ~0x00000010;
1405 	}
1406 
1407 	if (trap & 0x00000040) {
1408 		u32 stat = nvkm_rd32(device, 0x40601c);
1409 
1410 		nvkm_snprintbf(error, sizeof(error), gf100_unk6_error,
1411 			       stat & 0x3fffffff);
1412 		nvkm_error(subdev, "UNK6 %08x [%s]\n", stat, error);
1413 
1414 		nvkm_wr32(device, 0x40601c, 0xc0000000);
1415 		nvkm_wr32(device, 0x400108, 0x00000040);
1416 		trap &= ~0x00000040;
1417 	}
1418 
1419 	if (trap & 0x00000080) {
1420 		u32 stat = nvkm_rd32(device, 0x404490);
1421 		u32 pc = nvkm_rd32(device, 0x404494);
1422 		u32 op = nvkm_rd32(device, 0x40449c);
1423 
1424 		nvkm_snprintbf(error, sizeof(error), gf100_macro_error,
1425 			       stat & 0x1fffffff);
1426 		nvkm_error(subdev, "MACRO %08x [%s], pc: 0x%03x%s, op: 0x%08x\n",
1427 			   stat, error, pc & 0x7ff,
1428 			   (pc & 0x10000000) ? "" : " (invalid)",
1429 			   op);
1430 
1431 		nvkm_wr32(device, 0x404490, 0xc0000000);
1432 		nvkm_wr32(device, 0x400108, 0x00000080);
1433 		trap &= ~0x00000080;
1434 	}
1435 
1436 	if (trap & 0x00000100) {
1437 		u32 stat = nvkm_rd32(device, 0x407020) & 0x3fffffff;
1438 
1439 		nvkm_snprintbf(error, sizeof(error), gk104_sked_error, stat);
1440 		nvkm_error(subdev, "SKED: %08x [%s]\n", stat, error);
1441 
1442 		if (stat)
1443 			nvkm_wr32(device, 0x407020, 0x40000000);
1444 		nvkm_wr32(device, 0x400108, 0x00000100);
1445 		trap &= ~0x00000100;
1446 	}
1447 
1448 	if (trap & 0x01000000) {
1449 		u32 stat = nvkm_rd32(device, 0x400118);
1450 		for (gpc = 0; stat && gpc < gr->gpc_nr; gpc++) {
1451 			u32 mask = 0x00000001 << gpc;
1452 			if (stat & mask) {
1453 				gf100_gr_trap_gpc(gr, gpc);
1454 				nvkm_wr32(device, 0x400118, mask);
1455 				stat &= ~mask;
1456 			}
1457 		}
1458 		nvkm_wr32(device, 0x400108, 0x01000000);
1459 		trap &= ~0x01000000;
1460 	}
1461 
1462 	if (trap & 0x02000000) {
1463 		for (rop = 0; rop < gr->rop_nr; rop++) {
1464 			u32 statz = nvkm_rd32(device, ROP_UNIT(rop, 0x070));
1465 			u32 statc = nvkm_rd32(device, ROP_UNIT(rop, 0x144));
1466 			nvkm_error(subdev, "ROP%d %08x %08x\n",
1467 				 rop, statz, statc);
1468 			nvkm_wr32(device, ROP_UNIT(rop, 0x070), 0xc0000000);
1469 			nvkm_wr32(device, ROP_UNIT(rop, 0x144), 0xc0000000);
1470 		}
1471 		nvkm_wr32(device, 0x400108, 0x02000000);
1472 		trap &= ~0x02000000;
1473 	}
1474 
1475 	if (trap) {
1476 		nvkm_error(subdev, "TRAP UNHANDLED %08x\n", trap);
1477 		nvkm_wr32(device, 0x400108, trap);
1478 	}
1479 }
1480 
1481 static void
gf100_gr_ctxctl_debug_unit(struct gf100_gr * gr,u32 base)1482 gf100_gr_ctxctl_debug_unit(struct gf100_gr *gr, u32 base)
1483 {
1484 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1485 	struct nvkm_device *device = subdev->device;
1486 	nvkm_error(subdev, "%06x - done %08x\n", base,
1487 		   nvkm_rd32(device, base + 0x400));
1488 	nvkm_error(subdev, "%06x - stat %08x %08x %08x %08x\n", base,
1489 		   nvkm_rd32(device, base + 0x800),
1490 		   nvkm_rd32(device, base + 0x804),
1491 		   nvkm_rd32(device, base + 0x808),
1492 		   nvkm_rd32(device, base + 0x80c));
1493 	nvkm_error(subdev, "%06x - stat %08x %08x %08x %08x\n", base,
1494 		   nvkm_rd32(device, base + 0x810),
1495 		   nvkm_rd32(device, base + 0x814),
1496 		   nvkm_rd32(device, base + 0x818),
1497 		   nvkm_rd32(device, base + 0x81c));
1498 }
1499 
1500 void
gf100_gr_ctxctl_debug(struct gf100_gr * gr)1501 gf100_gr_ctxctl_debug(struct gf100_gr *gr)
1502 {
1503 	struct nvkm_device *device = gr->base.engine.subdev.device;
1504 	u32 gpcnr = nvkm_rd32(device, 0x409604) & 0xffff;
1505 	u32 gpc;
1506 
1507 	gf100_gr_ctxctl_debug_unit(gr, 0x409000);
1508 	for (gpc = 0; gpc < gpcnr; gpc++)
1509 		gf100_gr_ctxctl_debug_unit(gr, 0x502000 + (gpc * 0x8000));
1510 }
1511 
1512 static void
gf100_gr_ctxctl_isr(struct gf100_gr * gr)1513 gf100_gr_ctxctl_isr(struct gf100_gr *gr)
1514 {
1515 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1516 	struct nvkm_device *device = subdev->device;
1517 	u32 stat = nvkm_rd32(device, 0x409c18);
1518 
1519 	if (!gr->firmware && (stat & 0x00000001)) {
1520 		u32 code = nvkm_rd32(device, 0x409814);
1521 		if (code == E_BAD_FWMTHD) {
1522 			u32 class = nvkm_rd32(device, 0x409808);
1523 			u32  addr = nvkm_rd32(device, 0x40980c);
1524 			u32  subc = (addr & 0x00070000) >> 16;
1525 			u32  mthd = (addr & 0x00003ffc);
1526 			u32  data = nvkm_rd32(device, 0x409810);
1527 
1528 			nvkm_error(subdev, "FECS MTHD subc %d class %04x "
1529 					   "mthd %04x data %08x\n",
1530 				   subc, class, mthd, data);
1531 		} else {
1532 			nvkm_error(subdev, "FECS ucode error %d\n", code);
1533 		}
1534 		nvkm_wr32(device, 0x409c20, 0x00000001);
1535 		stat &= ~0x00000001;
1536 	}
1537 
1538 	if (!gr->firmware && (stat & 0x00080000)) {
1539 		nvkm_error(subdev, "FECS watchdog timeout\n");
1540 		gf100_gr_ctxctl_debug(gr);
1541 		nvkm_wr32(device, 0x409c20, 0x00080000);
1542 		stat &= ~0x00080000;
1543 	}
1544 
1545 	if (stat) {
1546 		nvkm_error(subdev, "FECS %08x\n", stat);
1547 		gf100_gr_ctxctl_debug(gr);
1548 		nvkm_wr32(device, 0x409c20, stat);
1549 	}
1550 }
1551 
1552 static void
gf100_gr_intr(struct nvkm_gr * base)1553 gf100_gr_intr(struct nvkm_gr *base)
1554 {
1555 	struct gf100_gr *gr = gf100_gr(base);
1556 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1557 	struct nvkm_device *device = subdev->device;
1558 	struct nvkm_fifo_chan *chan;
1559 	unsigned long flags;
1560 	u64 inst = nvkm_rd32(device, 0x409b00) & 0x0fffffff;
1561 	u32 stat = nvkm_rd32(device, 0x400100);
1562 	u32 addr = nvkm_rd32(device, 0x400704);
1563 	u32 mthd = (addr & 0x00003ffc);
1564 	u32 subc = (addr & 0x00070000) >> 16;
1565 	u32 data = nvkm_rd32(device, 0x400708);
1566 	u32 code = nvkm_rd32(device, 0x400110);
1567 	u32 class;
1568 	const char *name = "unknown";
1569 	int chid = -1;
1570 
1571 	chan = nvkm_fifo_chan_inst(device->fifo, (u64)inst << 12, &flags);
1572 	if (chan) {
1573 		name = chan->object.client->name;
1574 		chid = chan->chid;
1575 	}
1576 
1577 	if (device->card_type < NV_E0 || subc < 4)
1578 		class = nvkm_rd32(device, 0x404200 + (subc * 4));
1579 	else
1580 		class = 0x0000;
1581 
1582 	if (stat & 0x00000001) {
1583 		/*
1584 		 * notifier interrupt, only needed for cyclestats
1585 		 * can be safely ignored
1586 		 */
1587 		nvkm_wr32(device, 0x400100, 0x00000001);
1588 		stat &= ~0x00000001;
1589 	}
1590 
1591 	if (stat & 0x00000010) {
1592 		if (!gf100_gr_mthd_sw(device, class, mthd, data)) {
1593 			nvkm_error(subdev, "ILLEGAL_MTHD ch %d [%010"PRIx64" %s] "
1594 				   "subc %d class %04x mthd %04x data %08x\n",
1595 				   chid, inst << 12, name, subc,
1596 				   class, mthd, data);
1597 		}
1598 		nvkm_wr32(device, 0x400100, 0x00000010);
1599 		stat &= ~0x00000010;
1600 	}
1601 
1602 	if (stat & 0x00000020) {
1603 		nvkm_error(subdev, "ILLEGAL_CLASS ch %d [%010"PRIx64" %s] "
1604 			   "subc %d class %04x mthd %04x data %08x\n",
1605 			   chid, inst << 12, name, subc, class, mthd, data);
1606 		nvkm_wr32(device, 0x400100, 0x00000020);
1607 		stat &= ~0x00000020;
1608 	}
1609 
1610 	if (stat & 0x00100000) {
1611 		const struct nvkm_enum *en =
1612 			nvkm_enum_find(nv50_data_error_names, code);
1613 		nvkm_error(subdev, "DATA_ERROR %08x [%s] ch %d [%010"PRIx64" %s] "
1614 				   "subc %d class %04x mthd %04x data %08x\n",
1615 			   code, en ? en->name : "", chid, inst << 12,
1616 			   name, subc, class, mthd, data);
1617 		nvkm_wr32(device, 0x400100, 0x00100000);
1618 		stat &= ~0x00100000;
1619 	}
1620 
1621 	if (stat & 0x00200000) {
1622 		nvkm_error(subdev, "TRAP ch %d [%010"PRIx64" %s]\n",
1623 			   chid, inst << 12, name);
1624 		gf100_gr_trap_intr(gr);
1625 		nvkm_wr32(device, 0x400100, 0x00200000);
1626 		stat &= ~0x00200000;
1627 	}
1628 
1629 	if (stat & 0x00080000) {
1630 		gf100_gr_ctxctl_isr(gr);
1631 		nvkm_wr32(device, 0x400100, 0x00080000);
1632 		stat &= ~0x00080000;
1633 	}
1634 
1635 	if (stat) {
1636 		nvkm_error(subdev, "intr %08x\n", stat);
1637 		nvkm_wr32(device, 0x400100, stat);
1638 	}
1639 
1640 	nvkm_wr32(device, 0x400500, 0x00010001);
1641 	nvkm_fifo_chan_put(device->fifo, flags, &chan);
1642 }
1643 
1644 static void
gf100_gr_init_fw(struct nvkm_falcon * falcon,struct nvkm_blob * code,struct nvkm_blob * data)1645 gf100_gr_init_fw(struct nvkm_falcon *falcon,
1646 		 struct nvkm_blob *code, struct nvkm_blob *data)
1647 {
1648 	nvkm_falcon_load_dmem(falcon, data->data, 0x0, data->size, 0);
1649 	nvkm_falcon_load_imem(falcon, code->data, 0x0, code->size, 0, 0, false);
1650 }
1651 
1652 static void
gf100_gr_init_csdata(struct gf100_gr * gr,const struct gf100_gr_pack * pack,u32 falcon,u32 starstar,u32 base)1653 gf100_gr_init_csdata(struct gf100_gr *gr,
1654 		     const struct gf100_gr_pack *pack,
1655 		     u32 falcon, u32 starstar, u32 base)
1656 {
1657 	struct nvkm_device *device = gr->base.engine.subdev.device;
1658 	const struct gf100_gr_pack *iter;
1659 	const struct gf100_gr_init *init;
1660 	u32 addr = ~0, prev = ~0, xfer = 0;
1661 	u32 star, temp;
1662 
1663 	nvkm_wr32(device, falcon + 0x01c0, 0x02000000 + starstar);
1664 	star = nvkm_rd32(device, falcon + 0x01c4);
1665 	temp = nvkm_rd32(device, falcon + 0x01c4);
1666 	if (temp > star)
1667 		star = temp;
1668 	nvkm_wr32(device, falcon + 0x01c0, 0x01000000 + star);
1669 
1670 	pack_for_each_init(init, iter, pack) {
1671 		u32 head = init->addr - base;
1672 		u32 tail = head + init->count * init->pitch;
1673 		while (head < tail) {
1674 			if (head != prev + 4 || xfer >= 32) {
1675 				if (xfer) {
1676 					u32 data = ((--xfer << 26) | addr);
1677 					nvkm_wr32(device, falcon + 0x01c4, data);
1678 					star += 4;
1679 				}
1680 				addr = head;
1681 				xfer = 0;
1682 			}
1683 			prev = head;
1684 			xfer = xfer + 1;
1685 			head = head + init->pitch;
1686 		}
1687 	}
1688 
1689 	nvkm_wr32(device, falcon + 0x01c4, (--xfer << 26) | addr);
1690 	nvkm_wr32(device, falcon + 0x01c0, 0x01000004 + starstar);
1691 	nvkm_wr32(device, falcon + 0x01c4, star + 4);
1692 }
1693 
1694 /* Initialize context from an external (secure or not) firmware */
1695 static int
gf100_gr_init_ctxctl_ext(struct gf100_gr * gr)1696 gf100_gr_init_ctxctl_ext(struct gf100_gr *gr)
1697 {
1698 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1699 	struct nvkm_device *device = subdev->device;
1700 	u32 lsf_mask = 0;
1701 	int ret;
1702 
1703 	/* load fuc microcode */
1704 	nvkm_mc_unk260(device, 0);
1705 
1706 	/* securely-managed falcons must be reset using secure boot */
1707 
1708 	if (!nvkm_acr_managed_falcon(device, NVKM_ACR_LSF_FECS)) {
1709 		gf100_gr_init_fw(&gr->fecs.falcon, &gr->fecs.inst,
1710 						   &gr->fecs.data);
1711 	} else {
1712 		lsf_mask |= BIT(NVKM_ACR_LSF_FECS);
1713 	}
1714 
1715 	if (!nvkm_acr_managed_falcon(device, NVKM_ACR_LSF_GPCCS)) {
1716 		gf100_gr_init_fw(&gr->gpccs.falcon, &gr->gpccs.inst,
1717 						    &gr->gpccs.data);
1718 	} else {
1719 		lsf_mask |= BIT(NVKM_ACR_LSF_GPCCS);
1720 	}
1721 
1722 	if (lsf_mask) {
1723 		ret = nvkm_acr_bootstrap_falcons(device, lsf_mask);
1724 		if (ret)
1725 			return ret;
1726 	}
1727 
1728 	nvkm_mc_unk260(device, 1);
1729 
1730 	/* start both of them running */
1731 	nvkm_wr32(device, 0x409840, 0xffffffff);
1732 	nvkm_wr32(device, 0x41a10c, 0x00000000);
1733 	nvkm_wr32(device, 0x40910c, 0x00000000);
1734 
1735 	nvkm_falcon_start(&gr->gpccs.falcon);
1736 	nvkm_falcon_start(&gr->fecs.falcon);
1737 
1738 	if (nvkm_msec(device, 2000,
1739 		if (nvkm_rd32(device, 0x409800) & 0x00000001)
1740 			break;
1741 	) < 0)
1742 		return -EBUSY;
1743 
1744 	gf100_gr_fecs_set_watchdog_timeout(gr, 0x7fffffff);
1745 
1746 	/* Determine how much memory is required to store main context image. */
1747 	ret = gf100_gr_fecs_discover_image_size(gr, &gr->size);
1748 	if (ret)
1749 		return ret;
1750 
1751 	/* Determine how much memory is required to store ZCULL image. */
1752 	ret = gf100_gr_fecs_discover_zcull_image_size(gr, &gr->size_zcull);
1753 	if (ret)
1754 		return ret;
1755 
1756 	/* Determine how much memory is required to store PerfMon image. */
1757 	ret = gf100_gr_fecs_discover_pm_image_size(gr, &gr->size_pm);
1758 	if (ret)
1759 		return ret;
1760 
1761 	/*XXX: We (likely) require PMU support to even bother with this.
1762 	 *
1763 	 *     Also, it seems like not all GPUs support ELPG.  Traces I
1764 	 *     have here show RM enabling it on Kepler/Turing, but none
1765 	 *     of the GPUs between those.  NVGPU decides this by PCIID.
1766 	 */
1767 	if (0) {
1768 		ret = gf100_gr_fecs_elpg_bind(gr);
1769 		if (ret)
1770 			return ret;
1771 	}
1772 
1773 	/* Generate golden context image. */
1774 	if (gr->data == NULL) {
1775 		int ret = gf100_grctx_generate(gr);
1776 		if (ret) {
1777 			nvkm_error(subdev, "failed to construct context\n");
1778 			return ret;
1779 		}
1780 	}
1781 
1782 	return 0;
1783 }
1784 
1785 static int
gf100_gr_init_ctxctl_int(struct gf100_gr * gr)1786 gf100_gr_init_ctxctl_int(struct gf100_gr *gr)
1787 {
1788 	const struct gf100_grctx_func *grctx = gr->func->grctx;
1789 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1790 	struct nvkm_device *device = subdev->device;
1791 
1792 	if (!gr->func->fecs.ucode) {
1793 		return -ENOSYS;
1794 	}
1795 
1796 	/* load HUB microcode */
1797 	nvkm_mc_unk260(device, 0);
1798 	nvkm_falcon_load_dmem(&gr->fecs.falcon,
1799 			      gr->func->fecs.ucode->data.data, 0x0,
1800 			      gr->func->fecs.ucode->data.size, 0);
1801 	nvkm_falcon_load_imem(&gr->fecs.falcon,
1802 			      gr->func->fecs.ucode->code.data, 0x0,
1803 			      gr->func->fecs.ucode->code.size, 0, 0, false);
1804 
1805 	/* load GPC microcode */
1806 	nvkm_falcon_load_dmem(&gr->gpccs.falcon,
1807 			      gr->func->gpccs.ucode->data.data, 0x0,
1808 			      gr->func->gpccs.ucode->data.size, 0);
1809 	nvkm_falcon_load_imem(&gr->gpccs.falcon,
1810 			      gr->func->gpccs.ucode->code.data, 0x0,
1811 			      gr->func->gpccs.ucode->code.size, 0, 0, false);
1812 	nvkm_mc_unk260(device, 1);
1813 
1814 	/* load register lists */
1815 	gf100_gr_init_csdata(gr, grctx->hub, 0x409000, 0x000, 0x000000);
1816 	gf100_gr_init_csdata(gr, grctx->gpc_0, 0x41a000, 0x000, 0x418000);
1817 	gf100_gr_init_csdata(gr, grctx->gpc_1, 0x41a000, 0x000, 0x418000);
1818 	gf100_gr_init_csdata(gr, grctx->tpc, 0x41a000, 0x004, 0x419800);
1819 	gf100_gr_init_csdata(gr, grctx->ppc, 0x41a000, 0x008, 0x41be00);
1820 
1821 	/* start HUB ucode running, it'll init the GPCs */
1822 	nvkm_wr32(device, 0x40910c, 0x00000000);
1823 	nvkm_wr32(device, 0x409100, 0x00000002);
1824 	if (nvkm_msec(device, 2000,
1825 		if (nvkm_rd32(device, 0x409800) & 0x80000000)
1826 			break;
1827 	) < 0) {
1828 		gf100_gr_ctxctl_debug(gr);
1829 		return -EBUSY;
1830 	}
1831 
1832 	gr->size = nvkm_rd32(device, 0x409804);
1833 	if (gr->data == NULL) {
1834 		int ret = gf100_grctx_generate(gr);
1835 		if (ret) {
1836 			nvkm_error(subdev, "failed to construct context\n");
1837 			return ret;
1838 		}
1839 	}
1840 
1841 	return 0;
1842 }
1843 
1844 int
gf100_gr_init_ctxctl(struct gf100_gr * gr)1845 gf100_gr_init_ctxctl(struct gf100_gr *gr)
1846 {
1847 	int ret;
1848 
1849 	if (gr->firmware)
1850 		ret = gf100_gr_init_ctxctl_ext(gr);
1851 	else
1852 		ret = gf100_gr_init_ctxctl_int(gr);
1853 
1854 	return ret;
1855 }
1856 
1857 void
gf100_gr_oneinit_sm_id(struct gf100_gr * gr)1858 gf100_gr_oneinit_sm_id(struct gf100_gr *gr)
1859 {
1860 	int tpc, gpc;
1861 	for (tpc = 0; tpc < gr->tpc_max; tpc++) {
1862 		for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
1863 			if (tpc < gr->tpc_nr[gpc]) {
1864 				gr->sm[gr->sm_nr].gpc = gpc;
1865 				gr->sm[gr->sm_nr].tpc = tpc;
1866 				gr->sm_nr++;
1867 			}
1868 		}
1869 	}
1870 }
1871 
1872 void
gf100_gr_oneinit_tiles(struct gf100_gr * gr)1873 gf100_gr_oneinit_tiles(struct gf100_gr *gr)
1874 {
1875 	static const u8 primes[] = {
1876 		3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61
1877 	};
1878 	int init_frac[GPC_MAX], init_err[GPC_MAX], run_err[GPC_MAX], i, j;
1879 	u32 mul_factor, comm_denom;
1880 	u8  gpc_map[GPC_MAX];
1881 	bool sorted;
1882 
1883 	switch (gr->tpc_total) {
1884 	case 15: gr->screen_tile_row_offset = 0x06; break;
1885 	case 14: gr->screen_tile_row_offset = 0x05; break;
1886 	case 13: gr->screen_tile_row_offset = 0x02; break;
1887 	case 11: gr->screen_tile_row_offset = 0x07; break;
1888 	case 10: gr->screen_tile_row_offset = 0x06; break;
1889 	case  7:
1890 	case  5: gr->screen_tile_row_offset = 0x01; break;
1891 	case  3: gr->screen_tile_row_offset = 0x02; break;
1892 	case  2:
1893 	case  1: gr->screen_tile_row_offset = 0x01; break;
1894 	default: gr->screen_tile_row_offset = 0x03;
1895 		for (i = 0; i < ARRAY_SIZE(primes); i++) {
1896 			if (gr->tpc_total % primes[i]) {
1897 				gr->screen_tile_row_offset = primes[i];
1898 				break;
1899 			}
1900 		}
1901 		break;
1902 	}
1903 
1904 	/* Sort GPCs by TPC count, highest-to-lowest. */
1905 	for (i = 0; i < gr->gpc_nr; i++)
1906 		gpc_map[i] = i;
1907 	sorted = false;
1908 
1909 	while (!sorted) {
1910 		for (sorted = true, i = 0; i < gr->gpc_nr - 1; i++) {
1911 			if (gr->tpc_nr[gpc_map[i + 1]] >
1912 			    gr->tpc_nr[gpc_map[i + 0]]) {
1913 				u8 swap = gpc_map[i];
1914 				gpc_map[i + 0] = gpc_map[i + 1];
1915 				gpc_map[i + 1] = swap;
1916 				sorted = false;
1917 			}
1918 		}
1919 	}
1920 
1921 	/* Determine tile->GPC mapping */
1922 	mul_factor = gr->gpc_nr * gr->tpc_max;
1923 	if (mul_factor & 1)
1924 		mul_factor = 2;
1925 	else
1926 		mul_factor = 1;
1927 
1928 	comm_denom = gr->gpc_nr * gr->tpc_max * mul_factor;
1929 
1930 	for (i = 0; i < gr->gpc_nr; i++) {
1931 		init_frac[i] = gr->tpc_nr[gpc_map[i]] * gr->gpc_nr * mul_factor;
1932 		 init_err[i] = i * gr->tpc_max * mul_factor - comm_denom/2;
1933 		  run_err[i] = init_frac[i] + init_err[i];
1934 	}
1935 
1936 	for (i = 0; i < gr->tpc_total;) {
1937 		for (j = 0; j < gr->gpc_nr; j++) {
1938 			if ((run_err[j] * 2) >= comm_denom) {
1939 				gr->tile[i++] = gpc_map[j];
1940 				run_err[j] += init_frac[j] - comm_denom;
1941 			} else {
1942 				run_err[j] += init_frac[j];
1943 			}
1944 		}
1945 	}
1946 }
1947 
1948 static int
gf100_gr_oneinit(struct nvkm_gr * base)1949 gf100_gr_oneinit(struct nvkm_gr *base)
1950 {
1951 	struct gf100_gr *gr = gf100_gr(base);
1952 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
1953 	struct nvkm_device *device = subdev->device;
1954 	int i, j;
1955 
1956 	nvkm_pmu_pgob(device->pmu, false);
1957 
1958 	gr->rop_nr = gr->func->rops(gr);
1959 	gr->gpc_nr = nvkm_rd32(device, 0x409604) & 0x0000001f;
1960 	for (i = 0; i < gr->gpc_nr; i++) {
1961 		gr->tpc_nr[i]  = nvkm_rd32(device, GPC_UNIT(i, 0x2608));
1962 		gr->tpc_max = max(gr->tpc_max, gr->tpc_nr[i]);
1963 		gr->tpc_total += gr->tpc_nr[i];
1964 		gr->ppc_nr[i]  = gr->func->ppc_nr;
1965 		for (j = 0; j < gr->ppc_nr[i]; j++) {
1966 			gr->ppc_tpc_mask[i][j] =
1967 				nvkm_rd32(device, GPC_UNIT(i, 0x0c30 + (j * 4)));
1968 			if (gr->ppc_tpc_mask[i][j] == 0)
1969 				continue;
1970 			gr->ppc_mask[i] |= (1 << j);
1971 			gr->ppc_tpc_nr[i][j] = hweight8(gr->ppc_tpc_mask[i][j]);
1972 			if (gr->ppc_tpc_min == 0 ||
1973 			    gr->ppc_tpc_min > gr->ppc_tpc_nr[i][j])
1974 				gr->ppc_tpc_min = gr->ppc_tpc_nr[i][j];
1975 			if (gr->ppc_tpc_max < gr->ppc_tpc_nr[i][j])
1976 				gr->ppc_tpc_max = gr->ppc_tpc_nr[i][j];
1977 		}
1978 	}
1979 
1980 	memset(gr->tile, 0xff, sizeof(gr->tile));
1981 	gr->func->oneinit_tiles(gr);
1982 	gr->func->oneinit_sm_id(gr);
1983 	return 0;
1984 }
1985 
1986 static int
gf100_gr_init_(struct nvkm_gr * base)1987 gf100_gr_init_(struct nvkm_gr *base)
1988 {
1989 	struct gf100_gr *gr = gf100_gr(base);
1990 	struct nvkm_subdev *subdev = &base->engine.subdev;
1991 	u32 ret;
1992 
1993 	nvkm_pmu_pgob(gr->base.engine.subdev.device->pmu, false);
1994 
1995 	ret = nvkm_falcon_get(&gr->fecs.falcon, subdev);
1996 	if (ret)
1997 		return ret;
1998 
1999 	ret = nvkm_falcon_get(&gr->gpccs.falcon, subdev);
2000 	if (ret)
2001 		return ret;
2002 
2003 	return gr->func->init(gr);
2004 }
2005 
2006 static int
gf100_gr_fini(struct nvkm_gr * base,bool suspend)2007 gf100_gr_fini(struct nvkm_gr *base, bool suspend)
2008 {
2009 	struct gf100_gr *gr = gf100_gr(base);
2010 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
2011 	nvkm_falcon_put(&gr->gpccs.falcon, subdev);
2012 	nvkm_falcon_put(&gr->fecs.falcon, subdev);
2013 	return 0;
2014 }
2015 
2016 static void *
gf100_gr_dtor(struct nvkm_gr * base)2017 gf100_gr_dtor(struct nvkm_gr *base)
2018 {
2019 	struct gf100_gr *gr = gf100_gr(base);
2020 
2021 	kfree(gr->data);
2022 
2023 	nvkm_falcon_dtor(&gr->gpccs.falcon);
2024 	nvkm_falcon_dtor(&gr->fecs.falcon);
2025 
2026 	nvkm_blob_dtor(&gr->fecs.inst);
2027 	nvkm_blob_dtor(&gr->fecs.data);
2028 	nvkm_blob_dtor(&gr->gpccs.inst);
2029 	nvkm_blob_dtor(&gr->gpccs.data);
2030 
2031 	vfree(gr->bundle);
2032 	vfree(gr->method);
2033 	vfree(gr->sw_ctx);
2034 	vfree(gr->sw_nonctx);
2035 
2036 	mutex_destroy(&gr->fecs.mutex);
2037 
2038 	return gr;
2039 }
2040 
2041 static const struct nvkm_gr_func
2042 gf100_gr_ = {
2043 	.dtor = gf100_gr_dtor,
2044 	.oneinit = gf100_gr_oneinit,
2045 	.init = gf100_gr_init_,
2046 	.fini = gf100_gr_fini,
2047 	.intr = gf100_gr_intr,
2048 	.units = gf100_gr_units,
2049 	.chan_new = gf100_gr_chan_new,
2050 	.object_get = gf100_gr_object_get,
2051 	.chsw_load = gf100_gr_chsw_load,
2052 	.ctxsw.pause = gf100_gr_fecs_stop_ctxsw,
2053 	.ctxsw.resume = gf100_gr_fecs_start_ctxsw,
2054 	.ctxsw.inst = gf100_gr_ctxsw_inst,
2055 };
2056 
2057 static const struct nvkm_falcon_func
2058 gf100_gr_flcn = {
2059 	.fbif = 0x600,
2060 	.load_imem = nvkm_falcon_v1_load_imem,
2061 	.load_dmem = nvkm_falcon_v1_load_dmem,
2062 	.read_dmem = nvkm_falcon_v1_read_dmem,
2063 	.bind_context = nvkm_falcon_v1_bind_context,
2064 	.wait_for_halt = nvkm_falcon_v1_wait_for_halt,
2065 	.clear_interrupt = nvkm_falcon_v1_clear_interrupt,
2066 	.set_start_addr = nvkm_falcon_v1_set_start_addr,
2067 	.start = nvkm_falcon_v1_start,
2068 	.enable = nvkm_falcon_v1_enable,
2069 	.disable = nvkm_falcon_v1_disable,
2070 };
2071 
2072 int
gf100_gr_new_(const struct gf100_gr_fwif * fwif,struct nvkm_device * device,int index,struct nvkm_gr ** pgr)2073 gf100_gr_new_(const struct gf100_gr_fwif *fwif,
2074 	      struct nvkm_device *device, int index, struct nvkm_gr **pgr)
2075 {
2076 	struct gf100_gr *gr;
2077 	int ret;
2078 
2079 	if (!(gr = kzalloc(sizeof(*gr), GFP_KERNEL)))
2080 		return -ENOMEM;
2081 	*pgr = &gr->base;
2082 
2083 	ret = nvkm_gr_ctor(&gf100_gr_, device, index, true, &gr->base);
2084 	if (ret)
2085 		return ret;
2086 
2087 	fwif = nvkm_firmware_load(&gr->base.engine.subdev, fwif, "Gr", gr);
2088 	if (IS_ERR(fwif))
2089 		return -ENODEV;
2090 
2091 	gr->func = fwif->func;
2092 
2093 	ret = nvkm_falcon_ctor(&gf100_gr_flcn, &gr->base.engine.subdev,
2094 			       "fecs", 0x409000, &gr->fecs.falcon);
2095 	if (ret)
2096 		return ret;
2097 
2098 	mutex_init(&gr->fecs.mutex);
2099 
2100 	ret = nvkm_falcon_ctor(&gf100_gr_flcn, &gr->base.engine.subdev,
2101 			       "gpccs", 0x41a000, &gr->gpccs.falcon);
2102 	if (ret)
2103 		return ret;
2104 
2105 	return 0;
2106 }
2107 
2108 void
gf100_gr_init_num_tpc_per_gpc(struct gf100_gr * gr,bool pd,bool ds)2109 gf100_gr_init_num_tpc_per_gpc(struct gf100_gr *gr, bool pd, bool ds)
2110 {
2111 	struct nvkm_device *device = gr->base.engine.subdev.device;
2112 	int gpc, i, j;
2113 	u32 data;
2114 
2115 	for (gpc = 0, i = 0; i < 4; i++) {
2116 		for (data = 0, j = 0; j < 8 && gpc < gr->gpc_nr; j++, gpc++)
2117 			data |= gr->tpc_nr[gpc] << (j * 4);
2118 		if (pd)
2119 			nvkm_wr32(device, 0x406028 + (i * 4), data);
2120 		if (ds)
2121 			nvkm_wr32(device, 0x405870 + (i * 4), data);
2122 	}
2123 }
2124 
2125 void
gf100_gr_init_400054(struct gf100_gr * gr)2126 gf100_gr_init_400054(struct gf100_gr *gr)
2127 {
2128 	nvkm_wr32(gr->base.engine.subdev.device, 0x400054, 0x34ce3464);
2129 }
2130 
2131 void
gf100_gr_init_shader_exceptions(struct gf100_gr * gr,int gpc,int tpc)2132 gf100_gr_init_shader_exceptions(struct gf100_gr *gr, int gpc, int tpc)
2133 {
2134 	struct nvkm_device *device = gr->base.engine.subdev.device;
2135 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x644), 0x001ffffe);
2136 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x64c), 0x0000000f);
2137 }
2138 
2139 void
gf100_gr_init_tex_hww_esr(struct gf100_gr * gr,int gpc,int tpc)2140 gf100_gr_init_tex_hww_esr(struct gf100_gr *gr, int gpc, int tpc)
2141 {
2142 	struct nvkm_device *device = gr->base.engine.subdev.device;
2143 	nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x224), 0xc0000000);
2144 }
2145 
2146 void
gf100_gr_init_419eb4(struct gf100_gr * gr)2147 gf100_gr_init_419eb4(struct gf100_gr *gr)
2148 {
2149 	struct nvkm_device *device = gr->base.engine.subdev.device;
2150 	nvkm_mask(device, 0x419eb4, 0x00001000, 0x00001000);
2151 }
2152 
2153 void
gf100_gr_init_419cc0(struct gf100_gr * gr)2154 gf100_gr_init_419cc0(struct gf100_gr *gr)
2155 {
2156 	struct nvkm_device *device = gr->base.engine.subdev.device;
2157 	int gpc, tpc;
2158 
2159 	nvkm_mask(device, 0x419cc0, 0x00000008, 0x00000008);
2160 
2161 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
2162 		for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++)
2163 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x48c), 0xc0000000);
2164 	}
2165 }
2166 
2167 void
gf100_gr_init_40601c(struct gf100_gr * gr)2168 gf100_gr_init_40601c(struct gf100_gr *gr)
2169 {
2170 	nvkm_wr32(gr->base.engine.subdev.device, 0x40601c, 0xc0000000);
2171 }
2172 
2173 void
gf100_gr_init_fecs_exceptions(struct gf100_gr * gr)2174 gf100_gr_init_fecs_exceptions(struct gf100_gr *gr)
2175 {
2176 	const u32 data = gr->firmware ? 0x000e0000 : 0x000e0001;
2177 	nvkm_wr32(gr->base.engine.subdev.device, 0x409c24, data);
2178 }
2179 
2180 void
gf100_gr_init_gpc_mmu(struct gf100_gr * gr)2181 gf100_gr_init_gpc_mmu(struct gf100_gr *gr)
2182 {
2183 	struct nvkm_device *device = gr->base.engine.subdev.device;
2184 	struct nvkm_fb *fb = device->fb;
2185 
2186 	nvkm_wr32(device, 0x418880, nvkm_rd32(device, 0x100c80) & 0x00000001);
2187 	nvkm_wr32(device, 0x4188a4, 0x03000000);
2188 	nvkm_wr32(device, 0x418888, 0x00000000);
2189 	nvkm_wr32(device, 0x41888c, 0x00000000);
2190 	nvkm_wr32(device, 0x418890, 0x00000000);
2191 	nvkm_wr32(device, 0x418894, 0x00000000);
2192 	nvkm_wr32(device, 0x4188b4, nvkm_memory_addr(fb->mmu_wr) >> 8);
2193 	nvkm_wr32(device, 0x4188b8, nvkm_memory_addr(fb->mmu_rd) >> 8);
2194 }
2195 
2196 void
gf100_gr_init_num_active_ltcs(struct gf100_gr * gr)2197 gf100_gr_init_num_active_ltcs(struct gf100_gr *gr)
2198 {
2199 	struct nvkm_device *device = gr->base.engine.subdev.device;
2200 	nvkm_wr32(device, GPC_BCAST(0x08ac), nvkm_rd32(device, 0x100800));
2201 }
2202 
2203 void
gf100_gr_init_zcull(struct gf100_gr * gr)2204 gf100_gr_init_zcull(struct gf100_gr *gr)
2205 {
2206 	struct nvkm_device *device = gr->base.engine.subdev.device;
2207 	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, gr->tpc_total);
2208 	const u8 tile_nr = ALIGN(gr->tpc_total, 32);
2209 	u8 bank[GPC_MAX] = {}, gpc, i, j;
2210 	u32 data;
2211 
2212 	for (i = 0; i < tile_nr; i += 8) {
2213 		for (data = 0, j = 0; j < 8 && i + j < gr->tpc_total; j++) {
2214 			data |= bank[gr->tile[i + j]] << (j * 4);
2215 			bank[gr->tile[i + j]]++;
2216 		}
2217 		nvkm_wr32(device, GPC_BCAST(0x0980 + ((i / 8) * 4)), data);
2218 	}
2219 
2220 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
2221 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0914),
2222 			  gr->screen_tile_row_offset << 8 | gr->tpc_nr[gpc]);
2223 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0910), 0x00040000 |
2224 							 gr->tpc_total);
2225 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0918), magicgpc918);
2226 	}
2227 
2228 	nvkm_wr32(device, GPC_BCAST(0x1bd4), magicgpc918);
2229 }
2230 
2231 void
gf100_gr_init_vsc_stream_master(struct gf100_gr * gr)2232 gf100_gr_init_vsc_stream_master(struct gf100_gr *gr)
2233 {
2234 	struct nvkm_device *device = gr->base.engine.subdev.device;
2235 	nvkm_mask(device, TPC_UNIT(0, 0, 0x05c), 0x00000001, 0x00000001);
2236 }
2237 
2238 int
gf100_gr_init(struct gf100_gr * gr)2239 gf100_gr_init(struct gf100_gr *gr)
2240 {
2241 	struct nvkm_device *device = gr->base.engine.subdev.device;
2242 	int gpc, tpc, rop;
2243 
2244 	if (gr->func->init_419bd8)
2245 		gr->func->init_419bd8(gr);
2246 
2247 	gr->func->init_gpc_mmu(gr);
2248 
2249 	if (gr->sw_nonctx)
2250 		gf100_gr_mmio(gr, gr->sw_nonctx);
2251 	else
2252 		gf100_gr_mmio(gr, gr->func->mmio);
2253 
2254 	gf100_gr_wait_idle(gr);
2255 
2256 	if (gr->func->init_r405a14)
2257 		gr->func->init_r405a14(gr);
2258 
2259 	if (gr->func->clkgate_pack)
2260 		nvkm_therm_clkgate_init(device->therm, gr->func->clkgate_pack);
2261 
2262 	if (gr->func->init_bios)
2263 		gr->func->init_bios(gr);
2264 
2265 	gr->func->init_vsc_stream_master(gr);
2266 	gr->func->init_zcull(gr);
2267 	gr->func->init_num_active_ltcs(gr);
2268 	if (gr->func->init_rop_active_fbps)
2269 		gr->func->init_rop_active_fbps(gr);
2270 	if (gr->func->init_bios_2)
2271 		gr->func->init_bios_2(gr);
2272 	if (gr->func->init_swdx_pes_mask)
2273 		gr->func->init_swdx_pes_mask(gr);
2274 	if (gr->func->init_fs)
2275 		gr->func->init_fs(gr);
2276 
2277 	nvkm_wr32(device, 0x400500, 0x00010001);
2278 
2279 	nvkm_wr32(device, 0x400100, 0xffffffff);
2280 	nvkm_wr32(device, 0x40013c, 0xffffffff);
2281 	nvkm_wr32(device, 0x400124, 0x00000002);
2282 
2283 	gr->func->init_fecs_exceptions(gr);
2284 	if (gr->func->init_ds_hww_esr_2)
2285 		gr->func->init_ds_hww_esr_2(gr);
2286 
2287 	nvkm_wr32(device, 0x404000, 0xc0000000);
2288 	nvkm_wr32(device, 0x404600, 0xc0000000);
2289 	nvkm_wr32(device, 0x408030, 0xc0000000);
2290 
2291 	if (gr->func->init_40601c)
2292 		gr->func->init_40601c(gr);
2293 
2294 	nvkm_wr32(device, 0x406018, 0xc0000000);
2295 	nvkm_wr32(device, 0x404490, 0xc0000000);
2296 
2297 	if (gr->func->init_sked_hww_esr)
2298 		gr->func->init_sked_hww_esr(gr);
2299 
2300 	nvkm_wr32(device, 0x405840, 0xc0000000);
2301 	nvkm_wr32(device, 0x405844, 0x00ffffff);
2302 
2303 	if (gr->func->init_419cc0)
2304 		gr->func->init_419cc0(gr);
2305 	if (gr->func->init_419eb4)
2306 		gr->func->init_419eb4(gr);
2307 	if (gr->func->init_419c9c)
2308 		gr->func->init_419c9c(gr);
2309 
2310 	if (gr->func->init_ppc_exceptions)
2311 		gr->func->init_ppc_exceptions(gr);
2312 
2313 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
2314 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0420), 0xc0000000);
2315 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0900), 0xc0000000);
2316 		nvkm_wr32(device, GPC_UNIT(gpc, 0x1028), 0xc0000000);
2317 		nvkm_wr32(device, GPC_UNIT(gpc, 0x0824), 0xc0000000);
2318 		for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++) {
2319 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x508), 0xffffffff);
2320 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x50c), 0xffffffff);
2321 			if (gr->func->init_tex_hww_esr)
2322 				gr->func->init_tex_hww_esr(gr, gpc, tpc);
2323 			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x084), 0xc0000000);
2324 			if (gr->func->init_504430)
2325 				gr->func->init_504430(gr, gpc, tpc);
2326 			gr->func->init_shader_exceptions(gr, gpc, tpc);
2327 		}
2328 		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c90), 0xffffffff);
2329 		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c94), 0xffffffff);
2330 	}
2331 
2332 	for (rop = 0; rop < gr->rop_nr; rop++) {
2333 		nvkm_wr32(device, ROP_UNIT(rop, 0x144), 0x40000000);
2334 		nvkm_wr32(device, ROP_UNIT(rop, 0x070), 0x40000000);
2335 		nvkm_wr32(device, ROP_UNIT(rop, 0x204), 0xffffffff);
2336 		nvkm_wr32(device, ROP_UNIT(rop, 0x208), 0xffffffff);
2337 	}
2338 
2339 	nvkm_wr32(device, 0x400108, 0xffffffff);
2340 	nvkm_wr32(device, 0x400138, 0xffffffff);
2341 	nvkm_wr32(device, 0x400118, 0xffffffff);
2342 	nvkm_wr32(device, 0x400130, 0xffffffff);
2343 	nvkm_wr32(device, 0x40011c, 0xffffffff);
2344 	nvkm_wr32(device, 0x400134, 0xffffffff);
2345 
2346 	if (gr->func->init_400054)
2347 		gr->func->init_400054(gr);
2348 
2349 	gf100_gr_zbc_init(gr);
2350 
2351 	if (gr->func->init_4188a4)
2352 		gr->func->init_4188a4(gr);
2353 
2354 	return gf100_gr_init_ctxctl(gr);
2355 }
2356 
2357 #include "fuc/hubgf100.fuc3.h"
2358 
2359 struct gf100_gr_ucode
2360 gf100_gr_fecs_ucode = {
2361 	.code.data = gf100_grhub_code,
2362 	.code.size = sizeof(gf100_grhub_code),
2363 	.data.data = gf100_grhub_data,
2364 	.data.size = sizeof(gf100_grhub_data),
2365 };
2366 
2367 #include "fuc/gpcgf100.fuc3.h"
2368 
2369 struct gf100_gr_ucode
2370 gf100_gr_gpccs_ucode = {
2371 	.code.data = gf100_grgpc_code,
2372 	.code.size = sizeof(gf100_grgpc_code),
2373 	.data.data = gf100_grgpc_data,
2374 	.data.size = sizeof(gf100_grgpc_data),
2375 };
2376 
2377 static const struct gf100_gr_func
2378 gf100_gr = {
2379 	.oneinit_tiles = gf100_gr_oneinit_tiles,
2380 	.oneinit_sm_id = gf100_gr_oneinit_sm_id,
2381 	.init = gf100_gr_init,
2382 	.init_gpc_mmu = gf100_gr_init_gpc_mmu,
2383 	.init_vsc_stream_master = gf100_gr_init_vsc_stream_master,
2384 	.init_zcull = gf100_gr_init_zcull,
2385 	.init_num_active_ltcs = gf100_gr_init_num_active_ltcs,
2386 	.init_fecs_exceptions = gf100_gr_init_fecs_exceptions,
2387 	.init_40601c = gf100_gr_init_40601c,
2388 	.init_419cc0 = gf100_gr_init_419cc0,
2389 	.init_419eb4 = gf100_gr_init_419eb4,
2390 	.init_tex_hww_esr = gf100_gr_init_tex_hww_esr,
2391 	.init_shader_exceptions = gf100_gr_init_shader_exceptions,
2392 	.init_400054 = gf100_gr_init_400054,
2393 	.trap_mp = gf100_gr_trap_mp,
2394 	.mmio = gf100_gr_pack_mmio,
2395 	.fecs.ucode = &gf100_gr_fecs_ucode,
2396 	.gpccs.ucode = &gf100_gr_gpccs_ucode,
2397 	.rops = gf100_gr_rops,
2398 	.grctx = &gf100_grctx,
2399 	.zbc = &gf100_gr_zbc,
2400 	.sclass = {
2401 		{ -1, -1, FERMI_TWOD_A },
2402 		{ -1, -1, FERMI_MEMORY_TO_MEMORY_FORMAT_A },
2403 		{ -1, -1, FERMI_A, &gf100_fermi },
2404 		{ -1, -1, FERMI_COMPUTE_A },
2405 		{}
2406 	}
2407 };
2408 
2409 int
gf100_gr_nofw(struct gf100_gr * gr,int ver,const struct gf100_gr_fwif * fwif)2410 gf100_gr_nofw(struct gf100_gr *gr, int ver, const struct gf100_gr_fwif *fwif)
2411 {
2412 	gr->firmware = false;
2413 	return 0;
2414 }
2415 
2416 static int
gf100_gr_load_fw(struct gf100_gr * gr,const char * name,struct nvkm_blob * blob)2417 gf100_gr_load_fw(struct gf100_gr *gr, const char *name,
2418 		 struct nvkm_blob *blob)
2419 {
2420 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
2421 	struct nvkm_device *device = subdev->device;
2422 	const struct firmware *fw;
2423 	char f[32];
2424 	int ret;
2425 
2426 	snprintf(f, sizeof(f), "nouveau/nv%02x_%s", device->chipset, name);
2427 	ret = request_firmware(&fw, f, device->dev);
2428 	if (ret) {
2429 		snprintf(f, sizeof(f), "nouveau/%s", name);
2430 		ret = request_firmware(&fw, f, device->dev);
2431 		if (ret) {
2432 			nvkm_error(subdev, "failed to load %s\n", name);
2433 			return ret;
2434 		}
2435 	}
2436 
2437 	blob->size = fw->size;
2438 	blob->data = kmemdup(fw->data, blob->size, GFP_KERNEL);
2439 	release_firmware(fw);
2440 	return (blob->data != NULL) ? 0 : -ENOMEM;
2441 }
2442 
2443 int
gf100_gr_load(struct gf100_gr * gr,int ver,const struct gf100_gr_fwif * fwif)2444 gf100_gr_load(struct gf100_gr *gr, int ver, const struct gf100_gr_fwif *fwif)
2445 {
2446 	struct nvkm_device *device = gr->base.engine.subdev.device;
2447 
2448 	if (!nvkm_boolopt(device->cfgopt, "NvGrUseFW", false))
2449 		return -EINVAL;
2450 
2451 	if (gf100_gr_load_fw(gr, "fuc409c", &gr->fecs.inst) ||
2452 	    gf100_gr_load_fw(gr, "fuc409d", &gr->fecs.data) ||
2453 	    gf100_gr_load_fw(gr, "fuc41ac", &gr->gpccs.inst) ||
2454 	    gf100_gr_load_fw(gr, "fuc41ad", &gr->gpccs.data))
2455 		return -ENOENT;
2456 
2457 	gr->firmware = true;
2458 	return 0;
2459 }
2460 
2461 static const struct gf100_gr_fwif
2462 gf100_gr_fwif[] = {
2463 	{ -1, gf100_gr_load, &gf100_gr },
2464 	{ -1, gf100_gr_nofw, &gf100_gr },
2465 	{}
2466 };
2467 
2468 int
gf100_gr_new(struct nvkm_device * device,int index,struct nvkm_gr ** pgr)2469 gf100_gr_new(struct nvkm_device *device, int index, struct nvkm_gr **pgr)
2470 {
2471 	return gf100_gr_new_(gf100_gr_fwif, device, index, pgr);
2472 }
2473