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 "nv50.h"
25
26 #include <core/client.h>
27 #include <core/gpuobj.h>
28 #include <engine/fifo.h>
29
30 #include <nvif/class.h>
31
32 u64
nv50_gr_units(struct nvkm_gr * gr)33 nv50_gr_units(struct nvkm_gr *gr)
34 {
35 return nvkm_rd32(gr->engine.subdev.device, 0x1540);
36 }
37
38 /*******************************************************************************
39 * Graphics object classes
40 ******************************************************************************/
41
42 static int
nv50_gr_object_bind(struct nvkm_object * object,struct nvkm_gpuobj * parent,int align,struct nvkm_gpuobj ** pgpuobj)43 nv50_gr_object_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent,
44 int align, struct nvkm_gpuobj **pgpuobj)
45 {
46 int ret = nvkm_gpuobj_new(object->engine->subdev.device, 16,
47 align, false, parent, pgpuobj);
48 if (ret == 0) {
49 nvkm_kmap(*pgpuobj);
50 nvkm_wo32(*pgpuobj, 0x00, object->oclass);
51 nvkm_wo32(*pgpuobj, 0x04, 0x00000000);
52 nvkm_wo32(*pgpuobj, 0x08, 0x00000000);
53 nvkm_wo32(*pgpuobj, 0x0c, 0x00000000);
54 nvkm_done(*pgpuobj);
55 }
56 return ret;
57 }
58
59 const struct nvkm_object_func
60 nv50_gr_object = {
61 .bind = nv50_gr_object_bind,
62 };
63
64 /*******************************************************************************
65 * PGRAPH context
66 ******************************************************************************/
67
68 static int
nv50_gr_chan_bind(struct nvkm_object * object,struct nvkm_gpuobj * parent,int align,struct nvkm_gpuobj ** pgpuobj)69 nv50_gr_chan_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent,
70 int align, struct nvkm_gpuobj **pgpuobj)
71 {
72 struct nv50_gr *gr = nv50_gr_chan(object)->gr;
73 int ret = nvkm_gpuobj_new(gr->base.engine.subdev.device, gr->size,
74 align, true, parent, pgpuobj);
75 if (ret == 0) {
76 nvkm_kmap(*pgpuobj);
77 nv50_grctx_fill(gr->base.engine.subdev.device, *pgpuobj);
78 nvkm_done(*pgpuobj);
79 }
80 return ret;
81 }
82
83 static const struct nvkm_object_func
84 nv50_gr_chan = {
85 .bind = nv50_gr_chan_bind,
86 };
87
88 int
nv50_gr_chan_new(struct nvkm_gr * base,struct nvkm_fifo_chan * fifoch,const struct nvkm_oclass * oclass,struct nvkm_object ** pobject)89 nv50_gr_chan_new(struct nvkm_gr *base, struct nvkm_fifo_chan *fifoch,
90 const struct nvkm_oclass *oclass, struct nvkm_object **pobject)
91 {
92 struct nv50_gr *gr = nv50_gr(base);
93 struct nv50_gr_chan *chan;
94
95 if (!(chan = kzalloc(sizeof(*chan), GFP_KERNEL)))
96 return -ENOMEM;
97 nvkm_object_ctor(&nv50_gr_chan, oclass, &chan->object);
98 chan->gr = gr;
99 *pobject = &chan->object;
100 return 0;
101 }
102
103 /*******************************************************************************
104 * PGRAPH engine/subdev functions
105 ******************************************************************************/
106
107 static const struct nvkm_bitfield nv50_mp_exec_errors[] = {
108 { 0x01, "STACK_UNDERFLOW" },
109 { 0x02, "STACK_MISMATCH" },
110 { 0x04, "QUADON_ACTIVE" },
111 { 0x08, "TIMEOUT" },
112 { 0x10, "INVALID_OPCODE" },
113 { 0x20, "PM_OVERFLOW" },
114 { 0x40, "BREAKPOINT" },
115 {}
116 };
117
118 static const struct nvkm_bitfield nv50_mpc_traps[] = {
119 { 0x0000001, "LOCAL_LIMIT_READ" },
120 { 0x0000010, "LOCAL_LIMIT_WRITE" },
121 { 0x0000040, "STACK_LIMIT" },
122 { 0x0000100, "GLOBAL_LIMIT_READ" },
123 { 0x0001000, "GLOBAL_LIMIT_WRITE" },
124 { 0x0010000, "MP0" },
125 { 0x0020000, "MP1" },
126 { 0x0040000, "GLOBAL_LIMIT_RED" },
127 { 0x0400000, "GLOBAL_LIMIT_ATOM" },
128 { 0x4000000, "MP2" },
129 {}
130 };
131
132 static const struct nvkm_bitfield nv50_tex_traps[] = {
133 { 0x00000001, "" }, /* any bit set? */
134 { 0x00000002, "FAULT" },
135 { 0x00000004, "STORAGE_TYPE_MISMATCH" },
136 { 0x00000008, "LINEAR_MISMATCH" },
137 { 0x00000020, "WRONG_MEMTYPE" },
138 {}
139 };
140
141 static const struct nvkm_bitfield nv50_gr_trap_m2mf[] = {
142 { 0x00000001, "NOTIFY" },
143 { 0x00000002, "IN" },
144 { 0x00000004, "OUT" },
145 {}
146 };
147
148 static const struct nvkm_bitfield nv50_gr_trap_vfetch[] = {
149 { 0x00000001, "FAULT" },
150 {}
151 };
152
153 static const struct nvkm_bitfield nv50_gr_trap_strmout[] = {
154 { 0x00000001, "FAULT" },
155 {}
156 };
157
158 static const struct nvkm_bitfield nv50_gr_trap_ccache[] = {
159 { 0x00000001, "FAULT" },
160 {}
161 };
162
163 /* There must be a *lot* of these. Will take some time to gather them up. */
164 const struct nvkm_enum nv50_data_error_names[] = {
165 { 0x00000003, "INVALID_OPERATION", NULL },
166 { 0x00000004, "INVALID_VALUE", NULL },
167 { 0x00000005, "INVALID_ENUM", NULL },
168 { 0x00000008, "INVALID_OBJECT", NULL },
169 { 0x00000009, "READ_ONLY_OBJECT", NULL },
170 { 0x0000000a, "SUPERVISOR_OBJECT", NULL },
171 { 0x0000000b, "INVALID_ADDRESS_ALIGNMENT", NULL },
172 { 0x0000000c, "INVALID_BITFIELD", NULL },
173 { 0x0000000d, "BEGIN_END_ACTIVE", NULL },
174 { 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT", NULL },
175 { 0x0000000f, "VIEWPORT_ID_NEEDS_GP", NULL },
176 { 0x00000010, "RT_DOUBLE_BIND", NULL },
177 { 0x00000011, "RT_TYPES_MISMATCH", NULL },
178 { 0x00000012, "RT_LINEAR_WITH_ZETA", NULL },
179 { 0x00000015, "FP_TOO_FEW_REGS", NULL },
180 { 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH", NULL },
181 { 0x00000017, "RT_LINEAR_WITH_MSAA", NULL },
182 { 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT", NULL },
183 { 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT", NULL },
184 { 0x0000001a, "RT_INVALID_ALIGNMENT", NULL },
185 { 0x0000001b, "SAMPLER_OVER_LIMIT", NULL },
186 { 0x0000001c, "TEXTURE_OVER_LIMIT", NULL },
187 { 0x0000001e, "GP_TOO_MANY_OUTPUTS", NULL },
188 { 0x0000001f, "RT_BPP128_WITH_MS8", NULL },
189 { 0x00000021, "Z_OUT_OF_BOUNDS", NULL },
190 { 0x00000023, "XY_OUT_OF_BOUNDS", NULL },
191 { 0x00000024, "VP_ZERO_INPUTS", NULL },
192 { 0x00000027, "CP_MORE_PARAMS_THAN_SHARED", NULL },
193 { 0x00000028, "CP_NO_REG_SPACE_STRIPED", NULL },
194 { 0x00000029, "CP_NO_REG_SPACE_PACKED", NULL },
195 { 0x0000002a, "CP_NOT_ENOUGH_WARPS", NULL },
196 { 0x0000002b, "CP_BLOCK_SIZE_MISMATCH", NULL },
197 { 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS", NULL },
198 { 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS", NULL },
199 { 0x0000002e, "CP_NO_BLOCKDIM_LATCH", NULL },
200 { 0x00000031, "ENG2D_FORMAT_MISMATCH", NULL },
201 { 0x0000003f, "PRIMITIVE_ID_NEEDS_GP", NULL },
202 { 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT", NULL },
203 { 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT", NULL },
204 { 0x00000046, "LAYER_ID_NEEDS_GP", NULL },
205 { 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT", NULL },
206 { 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT", NULL },
207 {}
208 };
209
210 static const struct nvkm_bitfield nv50_gr_intr_name[] = {
211 { 0x00000001, "NOTIFY" },
212 { 0x00000002, "COMPUTE_QUERY" },
213 { 0x00000010, "ILLEGAL_MTHD" },
214 { 0x00000020, "ILLEGAL_CLASS" },
215 { 0x00000040, "DOUBLE_NOTIFY" },
216 { 0x00001000, "CONTEXT_SWITCH" },
217 { 0x00010000, "BUFFER_NOTIFY" },
218 { 0x00100000, "DATA_ERROR" },
219 { 0x00200000, "TRAP" },
220 { 0x01000000, "SINGLE_STEP" },
221 {}
222 };
223
224 static const struct nvkm_bitfield nv50_gr_trap_prop[] = {
225 { 0x00000004, "SURF_WIDTH_OVERRUN" },
226 { 0x00000008, "SURF_HEIGHT_OVERRUN" },
227 { 0x00000010, "DST2D_FAULT" },
228 { 0x00000020, "ZETA_FAULT" },
229 { 0x00000040, "RT_FAULT" },
230 { 0x00000080, "CUDA_FAULT" },
231 { 0x00000100, "DST2D_STORAGE_TYPE_MISMATCH" },
232 { 0x00000200, "ZETA_STORAGE_TYPE_MISMATCH" },
233 { 0x00000400, "RT_STORAGE_TYPE_MISMATCH" },
234 { 0x00000800, "DST2D_LINEAR_MISMATCH" },
235 { 0x00001000, "RT_LINEAR_MISMATCH" },
236 {}
237 };
238
239 static void
nv50_gr_prop_trap(struct nv50_gr * gr,u32 ustatus_addr,u32 ustatus,u32 tp)240 nv50_gr_prop_trap(struct nv50_gr *gr, u32 ustatus_addr, u32 ustatus, u32 tp)
241 {
242 struct nvkm_subdev *subdev = &gr->base.engine.subdev;
243 struct nvkm_device *device = subdev->device;
244 u32 e0c = nvkm_rd32(device, ustatus_addr + 0x04);
245 u32 e10 = nvkm_rd32(device, ustatus_addr + 0x08);
246 u32 e14 = nvkm_rd32(device, ustatus_addr + 0x0c);
247 u32 e18 = nvkm_rd32(device, ustatus_addr + 0x10);
248 u32 e1c = nvkm_rd32(device, ustatus_addr + 0x14);
249 u32 e20 = nvkm_rd32(device, ustatus_addr + 0x18);
250 u32 e24 = nvkm_rd32(device, ustatus_addr + 0x1c);
251 char msg[128];
252
253 /* CUDA memory: l[], g[] or stack. */
254 if (ustatus & 0x00000080) {
255 if (e18 & 0x80000000) {
256 /* g[] read fault? */
257 nvkm_error(subdev, "TRAP_PROP - TP %d - CUDA_FAULT - Global read fault at address %02x%08x\n",
258 tp, e14, e10 | ((e18 >> 24) & 0x1f));
259 e18 &= ~0x1f000000;
260 } else if (e18 & 0xc) {
261 /* g[] write fault? */
262 nvkm_error(subdev, "TRAP_PROP - TP %d - CUDA_FAULT - Global write fault at address %02x%08x\n",
263 tp, e14, e10 | ((e18 >> 7) & 0x1f));
264 e18 &= ~0x00000f80;
265 } else {
266 nvkm_error(subdev, "TRAP_PROP - TP %d - Unknown CUDA fault at address %02x%08x\n",
267 tp, e14, e10);
268 }
269 ustatus &= ~0x00000080;
270 }
271 if (ustatus) {
272 nvkm_snprintbf(msg, sizeof(msg), nv50_gr_trap_prop, ustatus);
273 nvkm_error(subdev, "TRAP_PROP - TP %d - %08x [%s] - "
274 "Address %02x%08x\n",
275 tp, ustatus, msg, e14, e10);
276 }
277 nvkm_error(subdev, "TRAP_PROP - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
278 tp, e0c, e18, e1c, e20, e24);
279 }
280
281 static void
nv50_gr_mp_trap(struct nv50_gr * gr,int tpid,int display)282 nv50_gr_mp_trap(struct nv50_gr *gr, int tpid, int display)
283 {
284 struct nvkm_subdev *subdev = &gr->base.engine.subdev;
285 struct nvkm_device *device = subdev->device;
286 u32 units = nvkm_rd32(device, 0x1540);
287 u32 addr, mp10, status, pc, oplow, ophigh;
288 char msg[128];
289 int i;
290 int mps = 0;
291 for (i = 0; i < 4; i++) {
292 if (!(units & 1 << (i+24)))
293 continue;
294 if (device->chipset < 0xa0)
295 addr = 0x408200 + (tpid << 12) + (i << 7);
296 else
297 addr = 0x408100 + (tpid << 11) + (i << 7);
298 mp10 = nvkm_rd32(device, addr + 0x10);
299 status = nvkm_rd32(device, addr + 0x14);
300 if (!status)
301 continue;
302 if (display) {
303 nvkm_rd32(device, addr + 0x20);
304 pc = nvkm_rd32(device, addr + 0x24);
305 oplow = nvkm_rd32(device, addr + 0x70);
306 ophigh = nvkm_rd32(device, addr + 0x74);
307 nvkm_snprintbf(msg, sizeof(msg),
308 nv50_mp_exec_errors, status);
309 nvkm_error(subdev, "TRAP_MP_EXEC - TP %d MP %d: "
310 "%08x [%s] at %06x warp %d, "
311 "opcode %08x %08x\n",
312 tpid, i, status, msg, pc & 0xffffff,
313 pc >> 24, oplow, ophigh);
314 }
315 nvkm_wr32(device, addr + 0x10, mp10);
316 nvkm_wr32(device, addr + 0x14, 0);
317 mps++;
318 }
319 if (!mps && display)
320 nvkm_error(subdev, "TRAP_MP_EXEC - TP %d: "
321 "No MPs claiming errors?\n", tpid);
322 }
323
324 static void
nv50_gr_tp_trap(struct nv50_gr * gr,int type,u32 ustatus_old,u32 ustatus_new,int display,const char * name)325 nv50_gr_tp_trap(struct nv50_gr *gr, int type, u32 ustatus_old,
326 u32 ustatus_new, int display, const char *name)
327 {
328 struct nvkm_subdev *subdev = &gr->base.engine.subdev;
329 struct nvkm_device *device = subdev->device;
330 u32 units = nvkm_rd32(device, 0x1540);
331 int tps = 0;
332 int i, r;
333 char msg[128];
334 u32 ustatus_addr, ustatus;
335 for (i = 0; i < 16; i++) {
336 if (!(units & (1 << i)))
337 continue;
338 if (device->chipset < 0xa0)
339 ustatus_addr = ustatus_old + (i << 12);
340 else
341 ustatus_addr = ustatus_new + (i << 11);
342 ustatus = nvkm_rd32(device, ustatus_addr) & 0x7fffffff;
343 if (!ustatus)
344 continue;
345 tps++;
346 switch (type) {
347 case 6: /* texture error... unknown for now */
348 if (display) {
349 nvkm_error(subdev, "magic set %d:\n", i);
350 for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
351 nvkm_error(subdev, "\t%08x: %08x\n", r,
352 nvkm_rd32(device, r));
353 if (ustatus) {
354 nvkm_snprintbf(msg, sizeof(msg),
355 nv50_tex_traps, ustatus);
356 nvkm_error(subdev,
357 "%s - TP%d: %08x [%s]\n",
358 name, i, ustatus, msg);
359 ustatus = 0;
360 }
361 }
362 break;
363 case 7: /* MP error */
364 if (ustatus & 0x04030000) {
365 nv50_gr_mp_trap(gr, i, display);
366 ustatus &= ~0x04030000;
367 }
368 if (ustatus && display) {
369 nvkm_snprintbf(msg, sizeof(msg),
370 nv50_mpc_traps, ustatus);
371 nvkm_error(subdev, "%s - TP%d: %08x [%s]\n",
372 name, i, ustatus, msg);
373 ustatus = 0;
374 }
375 break;
376 case 8: /* PROP error */
377 if (display)
378 nv50_gr_prop_trap(
379 gr, ustatus_addr, ustatus, i);
380 ustatus = 0;
381 break;
382 }
383 if (ustatus) {
384 if (display)
385 nvkm_error(subdev, "%s - TP%d: Unhandled ustatus %08x\n", name, i, ustatus);
386 }
387 nvkm_wr32(device, ustatus_addr, 0xc0000000);
388 }
389
390 if (!tps && display)
391 nvkm_warn(subdev, "%s - No TPs claiming errors?\n", name);
392 }
393
394 static int
nv50_gr_trap_handler(struct nv50_gr * gr,u32 display,int chid,u64 inst,const char * name)395 nv50_gr_trap_handler(struct nv50_gr *gr, u32 display,
396 int chid, u64 inst, const char *name)
397 {
398 struct nvkm_subdev *subdev = &gr->base.engine.subdev;
399 struct nvkm_device *device = subdev->device;
400 u32 status = nvkm_rd32(device, 0x400108);
401 u32 ustatus;
402 char msg[128];
403
404 if (!status && display) {
405 nvkm_error(subdev, "TRAP: no units reporting traps?\n");
406 return 1;
407 }
408
409 /* DISPATCH: Relays commands to other units and handles NOTIFY,
410 * COND, QUERY. If you get a trap from it, the command is still stuck
411 * in DISPATCH and you need to do something about it. */
412 if (status & 0x001) {
413 ustatus = nvkm_rd32(device, 0x400804) & 0x7fffffff;
414 if (!ustatus && display) {
415 nvkm_error(subdev, "TRAP_DISPATCH - no ustatus?\n");
416 }
417
418 nvkm_wr32(device, 0x400500, 0x00000000);
419
420 /* Known to be triggered by screwed up NOTIFY and COND... */
421 if (ustatus & 0x00000001) {
422 u32 addr = nvkm_rd32(device, 0x400808);
423 u32 subc = (addr & 0x00070000) >> 16;
424 u32 mthd = (addr & 0x00001ffc);
425 u32 datal = nvkm_rd32(device, 0x40080c);
426 u32 datah = nvkm_rd32(device, 0x400810);
427 u32 class = nvkm_rd32(device, 0x400814);
428 u32 r848 = nvkm_rd32(device, 0x400848);
429
430 nvkm_error(subdev, "TRAP DISPATCH_FAULT\n");
431 if (display && (addr & 0x80000000)) {
432 nvkm_error(subdev,
433 "ch %d [%010llx %s] subc %d "
434 "class %04x mthd %04x data %08x%08x "
435 "400808 %08x 400848 %08x\n",
436 chid, inst, name, subc, class, mthd,
437 datah, datal, addr, r848);
438 } else
439 if (display) {
440 nvkm_error(subdev, "no stuck command?\n");
441 }
442
443 nvkm_wr32(device, 0x400808, 0);
444 nvkm_wr32(device, 0x4008e8, nvkm_rd32(device, 0x4008e8) & 3);
445 nvkm_wr32(device, 0x400848, 0);
446 ustatus &= ~0x00000001;
447 }
448
449 if (ustatus & 0x00000002) {
450 u32 addr = nvkm_rd32(device, 0x40084c);
451 u32 subc = (addr & 0x00070000) >> 16;
452 u32 mthd = (addr & 0x00001ffc);
453 u32 data = nvkm_rd32(device, 0x40085c);
454 u32 class = nvkm_rd32(device, 0x400814);
455
456 nvkm_error(subdev, "TRAP DISPATCH_QUERY\n");
457 if (display && (addr & 0x80000000)) {
458 nvkm_error(subdev,
459 "ch %d [%010llx %s] subc %d "
460 "class %04x mthd %04x data %08x "
461 "40084c %08x\n", chid, inst, name,
462 subc, class, mthd, data, addr);
463 } else
464 if (display) {
465 nvkm_error(subdev, "no stuck command?\n");
466 }
467
468 nvkm_wr32(device, 0x40084c, 0);
469 ustatus &= ~0x00000002;
470 }
471
472 if (ustatus && display) {
473 nvkm_error(subdev, "TRAP_DISPATCH "
474 "(unknown %08x)\n", ustatus);
475 }
476
477 nvkm_wr32(device, 0x400804, 0xc0000000);
478 nvkm_wr32(device, 0x400108, 0x001);
479 status &= ~0x001;
480 if (!status)
481 return 0;
482 }
483
484 /* M2MF: Memory to memory copy engine. */
485 if (status & 0x002) {
486 u32 ustatus = nvkm_rd32(device, 0x406800) & 0x7fffffff;
487 if (display) {
488 nvkm_snprintbf(msg, sizeof(msg),
489 nv50_gr_trap_m2mf, ustatus);
490 nvkm_error(subdev, "TRAP_M2MF %08x [%s]\n",
491 ustatus, msg);
492 nvkm_error(subdev, "TRAP_M2MF %08x %08x %08x %08x\n",
493 nvkm_rd32(device, 0x406804),
494 nvkm_rd32(device, 0x406808),
495 nvkm_rd32(device, 0x40680c),
496 nvkm_rd32(device, 0x406810));
497 }
498
499 /* No sane way found yet -- just reset the bugger. */
500 nvkm_wr32(device, 0x400040, 2);
501 nvkm_wr32(device, 0x400040, 0);
502 nvkm_wr32(device, 0x406800, 0xc0000000);
503 nvkm_wr32(device, 0x400108, 0x002);
504 status &= ~0x002;
505 }
506
507 /* VFETCH: Fetches data from vertex buffers. */
508 if (status & 0x004) {
509 u32 ustatus = nvkm_rd32(device, 0x400c04) & 0x7fffffff;
510 if (display) {
511 nvkm_snprintbf(msg, sizeof(msg),
512 nv50_gr_trap_vfetch, ustatus);
513 nvkm_error(subdev, "TRAP_VFETCH %08x [%s]\n",
514 ustatus, msg);
515 nvkm_error(subdev, "TRAP_VFETCH %08x %08x %08x %08x\n",
516 nvkm_rd32(device, 0x400c00),
517 nvkm_rd32(device, 0x400c08),
518 nvkm_rd32(device, 0x400c0c),
519 nvkm_rd32(device, 0x400c10));
520 }
521
522 nvkm_wr32(device, 0x400c04, 0xc0000000);
523 nvkm_wr32(device, 0x400108, 0x004);
524 status &= ~0x004;
525 }
526
527 /* STRMOUT: DirectX streamout / OpenGL transform feedback. */
528 if (status & 0x008) {
529 ustatus = nvkm_rd32(device, 0x401800) & 0x7fffffff;
530 if (display) {
531 nvkm_snprintbf(msg, sizeof(msg),
532 nv50_gr_trap_strmout, ustatus);
533 nvkm_error(subdev, "TRAP_STRMOUT %08x [%s]\n",
534 ustatus, msg);
535 nvkm_error(subdev, "TRAP_STRMOUT %08x %08x %08x %08x\n",
536 nvkm_rd32(device, 0x401804),
537 nvkm_rd32(device, 0x401808),
538 nvkm_rd32(device, 0x40180c),
539 nvkm_rd32(device, 0x401810));
540 }
541
542 /* No sane way found yet -- just reset the bugger. */
543 nvkm_wr32(device, 0x400040, 0x80);
544 nvkm_wr32(device, 0x400040, 0);
545 nvkm_wr32(device, 0x401800, 0xc0000000);
546 nvkm_wr32(device, 0x400108, 0x008);
547 status &= ~0x008;
548 }
549
550 /* CCACHE: Handles code and c[] caches and fills them. */
551 if (status & 0x010) {
552 ustatus = nvkm_rd32(device, 0x405018) & 0x7fffffff;
553 if (display) {
554 nvkm_snprintbf(msg, sizeof(msg),
555 nv50_gr_trap_ccache, ustatus);
556 nvkm_error(subdev, "TRAP_CCACHE %08x [%s]\n",
557 ustatus, msg);
558 nvkm_error(subdev, "TRAP_CCACHE %08x %08x %08x %08x "
559 "%08x %08x %08x\n",
560 nvkm_rd32(device, 0x405000),
561 nvkm_rd32(device, 0x405004),
562 nvkm_rd32(device, 0x405008),
563 nvkm_rd32(device, 0x40500c),
564 nvkm_rd32(device, 0x405010),
565 nvkm_rd32(device, 0x405014),
566 nvkm_rd32(device, 0x40501c));
567 }
568
569 nvkm_wr32(device, 0x405018, 0xc0000000);
570 nvkm_wr32(device, 0x400108, 0x010);
571 status &= ~0x010;
572 }
573
574 /* Unknown, not seen yet... 0x402000 is the only trap status reg
575 * remaining, so try to handle it anyway. Perhaps related to that
576 * unknown DMA slot on tesla? */
577 if (status & 0x20) {
578 ustatus = nvkm_rd32(device, 0x402000) & 0x7fffffff;
579 if (display)
580 nvkm_error(subdev, "TRAP_UNKC04 %08x\n", ustatus);
581 nvkm_wr32(device, 0x402000, 0xc0000000);
582 /* no status modifiction on purpose */
583 }
584
585 /* TEXTURE: CUDA texturing units */
586 if (status & 0x040) {
587 nv50_gr_tp_trap(gr, 6, 0x408900, 0x408600, display,
588 "TRAP_TEXTURE");
589 nvkm_wr32(device, 0x400108, 0x040);
590 status &= ~0x040;
591 }
592
593 /* MP: CUDA execution engines. */
594 if (status & 0x080) {
595 nv50_gr_tp_trap(gr, 7, 0x408314, 0x40831c, display,
596 "TRAP_MP");
597 nvkm_wr32(device, 0x400108, 0x080);
598 status &= ~0x080;
599 }
600
601 /* PROP: Handles TP-initiated uncached memory accesses:
602 * l[], g[], stack, 2d surfaces, render targets. */
603 if (status & 0x100) {
604 nv50_gr_tp_trap(gr, 8, 0x408e08, 0x408708, display,
605 "TRAP_PROP");
606 nvkm_wr32(device, 0x400108, 0x100);
607 status &= ~0x100;
608 }
609
610 if (status) {
611 if (display)
612 nvkm_error(subdev, "TRAP: unknown %08x\n", status);
613 nvkm_wr32(device, 0x400108, status);
614 }
615
616 return 1;
617 }
618
619 void
nv50_gr_intr(struct nvkm_gr * base)620 nv50_gr_intr(struct nvkm_gr *base)
621 {
622 struct nv50_gr *gr = nv50_gr(base);
623 struct nvkm_subdev *subdev = &gr->base.engine.subdev;
624 struct nvkm_device *device = subdev->device;
625 struct nvkm_fifo_chan *chan;
626 u32 stat = nvkm_rd32(device, 0x400100);
627 u32 inst = nvkm_rd32(device, 0x40032c) & 0x0fffffff;
628 u32 addr = nvkm_rd32(device, 0x400704);
629 u32 subc = (addr & 0x00070000) >> 16;
630 u32 mthd = (addr & 0x00001ffc);
631 u32 data = nvkm_rd32(device, 0x400708);
632 u32 class = nvkm_rd32(device, 0x400814);
633 u32 show = stat, show_bitfield = stat;
634 const struct nvkm_enum *en;
635 unsigned long flags;
636 const char *name = "unknown";
637 char msg[128];
638 int chid = -1;
639
640 chan = nvkm_fifo_chan_inst(device->fifo, (u64)inst << 12, &flags);
641 if (chan) {
642 name = chan->object.client->name;
643 chid = chan->chid;
644 }
645
646 if (show & 0x00100000) {
647 u32 ecode = nvkm_rd32(device, 0x400110);
648 en = nvkm_enum_find(nv50_data_error_names, ecode);
649 nvkm_error(subdev, "DATA_ERROR %08x [%s]\n",
650 ecode, en ? en->name : "");
651 show_bitfield &= ~0x00100000;
652 }
653
654 if (stat & 0x00200000) {
655 if (!nv50_gr_trap_handler(gr, show, chid, (u64)inst << 12, name))
656 show &= ~0x00200000;
657 show_bitfield &= ~0x00200000;
658 }
659
660 nvkm_wr32(device, 0x400100, stat);
661 nvkm_wr32(device, 0x400500, 0x00010001);
662
663 if (show) {
664 show &= show_bitfield;
665 nvkm_snprintbf(msg, sizeof(msg), nv50_gr_intr_name, show);
666 nvkm_error(subdev, "%08x [%s] ch %d [%010llx %s] subc %d "
667 "class %04x mthd %04x data %08x\n",
668 stat, msg, chid, (u64)inst << 12, name,
669 subc, class, mthd, data);
670 }
671
672 if (nvkm_rd32(device, 0x400824) & (1 << 31))
673 nvkm_wr32(device, 0x400824, nvkm_rd32(device, 0x400824) & ~(1 << 31));
674
675 nvkm_fifo_chan_put(device->fifo, flags, &chan);
676 }
677
678 int
nv50_gr_init(struct nvkm_gr * base)679 nv50_gr_init(struct nvkm_gr *base)
680 {
681 struct nv50_gr *gr = nv50_gr(base);
682 struct nvkm_device *device = gr->base.engine.subdev.device;
683 int ret, units, i;
684
685 /* NV_PGRAPH_DEBUG_3_HW_CTX_SWITCH_ENABLED */
686 nvkm_wr32(device, 0x40008c, 0x00000004);
687
688 /* reset/enable traps and interrupts */
689 nvkm_wr32(device, 0x400804, 0xc0000000);
690 nvkm_wr32(device, 0x406800, 0xc0000000);
691 nvkm_wr32(device, 0x400c04, 0xc0000000);
692 nvkm_wr32(device, 0x401800, 0xc0000000);
693 nvkm_wr32(device, 0x405018, 0xc0000000);
694 nvkm_wr32(device, 0x402000, 0xc0000000);
695
696 units = nvkm_rd32(device, 0x001540);
697 for (i = 0; i < 16; i++) {
698 if (!(units & (1 << i)))
699 continue;
700
701 if (device->chipset < 0xa0) {
702 nvkm_wr32(device, 0x408900 + (i << 12), 0xc0000000);
703 nvkm_wr32(device, 0x408e08 + (i << 12), 0xc0000000);
704 nvkm_wr32(device, 0x408314 + (i << 12), 0xc0000000);
705 } else {
706 nvkm_wr32(device, 0x408600 + (i << 11), 0xc0000000);
707 nvkm_wr32(device, 0x408708 + (i << 11), 0xc0000000);
708 nvkm_wr32(device, 0x40831c + (i << 11), 0xc0000000);
709 }
710 }
711
712 nvkm_wr32(device, 0x400108, 0xffffffff);
713 nvkm_wr32(device, 0x400138, 0xffffffff);
714 nvkm_wr32(device, 0x400100, 0xffffffff);
715 nvkm_wr32(device, 0x40013c, 0xffffffff);
716 nvkm_wr32(device, 0x400500, 0x00010001);
717
718 /* upload context program, initialise ctxctl defaults */
719 ret = nv50_grctx_init(device, &gr->size);
720 if (ret)
721 return ret;
722
723 nvkm_wr32(device, 0x400824, 0x00000000);
724 nvkm_wr32(device, 0x400828, 0x00000000);
725 nvkm_wr32(device, 0x40082c, 0x00000000);
726 nvkm_wr32(device, 0x400830, 0x00000000);
727 nvkm_wr32(device, 0x40032c, 0x00000000);
728 nvkm_wr32(device, 0x400330, 0x00000000);
729
730 /* some unknown zcull magic */
731 switch (device->chipset & 0xf0) {
732 case 0x50:
733 case 0x80:
734 case 0x90:
735 nvkm_wr32(device, 0x402ca8, 0x00000800);
736 break;
737 case 0xa0:
738 default:
739 if (device->chipset == 0xa0 ||
740 device->chipset == 0xaa ||
741 device->chipset == 0xac) {
742 nvkm_wr32(device, 0x402ca8, 0x00000802);
743 } else {
744 nvkm_wr32(device, 0x402cc0, 0x00000000);
745 nvkm_wr32(device, 0x402ca8, 0x00000002);
746 }
747
748 break;
749 }
750
751 /* zero out zcull regions */
752 for (i = 0; i < 8; i++) {
753 nvkm_wr32(device, 0x402c20 + (i * 0x10), 0x00000000);
754 nvkm_wr32(device, 0x402c24 + (i * 0x10), 0x00000000);
755 nvkm_wr32(device, 0x402c28 + (i * 0x10), 0x00000000);
756 nvkm_wr32(device, 0x402c2c + (i * 0x10), 0x00000000);
757 }
758
759 return 0;
760 }
761
762 int
nv50_gr_new_(const struct nvkm_gr_func * func,struct nvkm_device * device,int index,struct nvkm_gr ** pgr)763 nv50_gr_new_(const struct nvkm_gr_func *func, struct nvkm_device *device,
764 int index, struct nvkm_gr **pgr)
765 {
766 struct nv50_gr *gr;
767
768 if (!(gr = kzalloc(sizeof(*gr), GFP_KERNEL)))
769 return -ENOMEM;
770 spin_lock_init(&gr->lock);
771 *pgr = &gr->base;
772
773 return nvkm_gr_ctor(func, device, index, true, &gr->base);
774 }
775
776 static const struct nvkm_gr_func
777 nv50_gr = {
778 .init = nv50_gr_init,
779 .intr = nv50_gr_intr,
780 .chan_new = nv50_gr_chan_new,
781 .units = nv50_gr_units,
782 .sclass = {
783 { -1, -1, NV_NULL_CLASS, &nv50_gr_object },
784 { -1, -1, NV50_TWOD, &nv50_gr_object },
785 { -1, -1, NV50_MEMORY_TO_MEMORY_FORMAT, &nv50_gr_object },
786 { -1, -1, NV50_TESLA, &nv50_gr_object },
787 { -1, -1, NV50_COMPUTE, &nv50_gr_object },
788 {}
789 }
790 };
791
792 int
nv50_gr_new(struct nvkm_device * device,int index,struct nvkm_gr ** pgr)793 nv50_gr_new(struct nvkm_device *device, int index, struct nvkm_gr **pgr)
794 {
795 return nv50_gr_new_(&nv50_gr, device, index, pgr);
796 }
797