]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/gpu/drm/nouveau/core/engine/graph/nv50.c
spi: s3c64xx: Let spi core handle validating transfer length
[mirror_ubuntu-jammy-kernel.git] / drivers / gpu / drm / nouveau / core / engine / graph / nv50.c
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
25 #include <core/os.h>
26 #include <core/class.h>
27 #include <core/client.h>
28 #include <core/handle.h>
29 #include <core/engctx.h>
30 #include <core/enum.h>
31
32 #include <subdev/fb.h>
33 #include <subdev/vm.h>
34 #include <subdev/timer.h>
35
36 #include <engine/fifo.h>
37 #include <engine/graph.h>
38
39 #include "nv50.h"
40
41 struct nv50_graph_priv {
42 struct nouveau_graph base;
43 spinlock_t lock;
44 u32 size;
45 };
46
47 struct nv50_graph_chan {
48 struct nouveau_graph_chan base;
49 };
50
51 static u64
52 nv50_graph_units(struct nouveau_graph *graph)
53 {
54 struct nv50_graph_priv *priv = (void *)graph;
55
56 return nv_rd32(priv, 0x1540);
57 }
58
59 /*******************************************************************************
60 * Graphics object classes
61 ******************************************************************************/
62
63 static int
64 nv50_graph_object_ctor(struct nouveau_object *parent,
65 struct nouveau_object *engine,
66 struct nouveau_oclass *oclass, void *data, u32 size,
67 struct nouveau_object **pobject)
68 {
69 struct nouveau_gpuobj *obj;
70 int ret;
71
72 ret = nouveau_gpuobj_create(parent, engine, oclass, 0, parent,
73 16, 16, 0, &obj);
74 *pobject = nv_object(obj);
75 if (ret)
76 return ret;
77
78 nv_wo32(obj, 0x00, nv_mclass(obj));
79 nv_wo32(obj, 0x04, 0x00000000);
80 nv_wo32(obj, 0x08, 0x00000000);
81 nv_wo32(obj, 0x0c, 0x00000000);
82 return 0;
83 }
84
85 static struct nouveau_ofuncs
86 nv50_graph_ofuncs = {
87 .ctor = nv50_graph_object_ctor,
88 .dtor = _nouveau_gpuobj_dtor,
89 .init = _nouveau_gpuobj_init,
90 .fini = _nouveau_gpuobj_fini,
91 .rd32 = _nouveau_gpuobj_rd32,
92 .wr32 = _nouveau_gpuobj_wr32,
93 };
94
95 static struct nouveau_oclass
96 nv50_graph_sclass[] = {
97 { 0x0030, &nv50_graph_ofuncs },
98 { 0x502d, &nv50_graph_ofuncs },
99 { 0x5039, &nv50_graph_ofuncs },
100 { 0x5097, &nv50_graph_ofuncs },
101 { 0x50c0, &nv50_graph_ofuncs },
102 {}
103 };
104
105 static struct nouveau_oclass
106 nv84_graph_sclass[] = {
107 { 0x0030, &nv50_graph_ofuncs },
108 { 0x502d, &nv50_graph_ofuncs },
109 { 0x5039, &nv50_graph_ofuncs },
110 { 0x50c0, &nv50_graph_ofuncs },
111 { 0x8297, &nv50_graph_ofuncs },
112 {}
113 };
114
115 static struct nouveau_oclass
116 nva0_graph_sclass[] = {
117 { 0x0030, &nv50_graph_ofuncs },
118 { 0x502d, &nv50_graph_ofuncs },
119 { 0x5039, &nv50_graph_ofuncs },
120 { 0x50c0, &nv50_graph_ofuncs },
121 { 0x8397, &nv50_graph_ofuncs },
122 {}
123 };
124
125 static struct nouveau_oclass
126 nva3_graph_sclass[] = {
127 { 0x0030, &nv50_graph_ofuncs },
128 { 0x502d, &nv50_graph_ofuncs },
129 { 0x5039, &nv50_graph_ofuncs },
130 { 0x50c0, &nv50_graph_ofuncs },
131 { 0x8597, &nv50_graph_ofuncs },
132 { 0x85c0, &nv50_graph_ofuncs },
133 {}
134 };
135
136 static struct nouveau_oclass
137 nvaf_graph_sclass[] = {
138 { 0x0030, &nv50_graph_ofuncs },
139 { 0x502d, &nv50_graph_ofuncs },
140 { 0x5039, &nv50_graph_ofuncs },
141 { 0x50c0, &nv50_graph_ofuncs },
142 { 0x85c0, &nv50_graph_ofuncs },
143 { 0x8697, &nv50_graph_ofuncs },
144 {}
145 };
146
147 /*******************************************************************************
148 * PGRAPH context
149 ******************************************************************************/
150
151 static int
152 nv50_graph_context_ctor(struct nouveau_object *parent,
153 struct nouveau_object *engine,
154 struct nouveau_oclass *oclass, void *data, u32 size,
155 struct nouveau_object **pobject)
156 {
157 struct nv50_graph_priv *priv = (void *)engine;
158 struct nv50_graph_chan *chan;
159 int ret;
160
161 ret = nouveau_graph_context_create(parent, engine, oclass, NULL,
162 priv->size, 0,
163 NVOBJ_FLAG_ZERO_ALLOC, &chan);
164 *pobject = nv_object(chan);
165 if (ret)
166 return ret;
167
168 nv50_grctx_fill(nv_device(priv), nv_gpuobj(chan));
169 return 0;
170 }
171
172 static struct nouveau_oclass
173 nv50_graph_cclass = {
174 .handle = NV_ENGCTX(GR, 0x50),
175 .ofuncs = &(struct nouveau_ofuncs) {
176 .ctor = nv50_graph_context_ctor,
177 .dtor = _nouveau_graph_context_dtor,
178 .init = _nouveau_graph_context_init,
179 .fini = _nouveau_graph_context_fini,
180 .rd32 = _nouveau_graph_context_rd32,
181 .wr32 = _nouveau_graph_context_wr32,
182 },
183 };
184
185 /*******************************************************************************
186 * PGRAPH engine/subdev functions
187 ******************************************************************************/
188
189 static const struct nouveau_bitfield nv50_pgraph_status[] = {
190 { 0x00000001, "BUSY" }, /* set when any bit is set */
191 { 0x00000002, "DISPATCH" },
192 { 0x00000004, "UNK2" },
193 { 0x00000008, "UNK3" },
194 { 0x00000010, "UNK4" },
195 { 0x00000020, "UNK5" },
196 { 0x00000040, "M2MF" },
197 { 0x00000080, "UNK7" },
198 { 0x00000100, "CTXPROG" },
199 { 0x00000200, "VFETCH" },
200 { 0x00000400, "CCACHE_UNK4" },
201 { 0x00000800, "STRMOUT_GSCHED_UNK5" },
202 { 0x00001000, "UNK14XX" },
203 { 0x00002000, "UNK24XX_CSCHED" },
204 { 0x00004000, "UNK1CXX" },
205 { 0x00008000, "CLIPID" },
206 { 0x00010000, "ZCULL" },
207 { 0x00020000, "ENG2D" },
208 { 0x00040000, "UNK34XX" },
209 { 0x00080000, "TPRAST" },
210 { 0x00100000, "TPROP" },
211 { 0x00200000, "TEX" },
212 { 0x00400000, "TPVP" },
213 { 0x00800000, "MP" },
214 { 0x01000000, "ROP" },
215 {}
216 };
217
218 static const char *const nv50_pgraph_vstatus_0[] = {
219 "VFETCH", "CCACHE", "UNK4", "UNK5", "GSCHED", "STRMOUT", "UNK14XX", NULL
220 };
221
222 static const char *const nv50_pgraph_vstatus_1[] = {
223 "TPRAST", "TPROP", "TEXTURE", "TPVP", "MP", NULL
224 };
225
226 static const char *const nv50_pgraph_vstatus_2[] = {
227 "UNK24XX", "CSCHED", "UNK1CXX", "CLIPID", "ZCULL", "ENG2D", "UNK34XX",
228 "ROP", NULL
229 };
230
231 static void nouveau_pgraph_vstatus_print(struct nv50_graph_priv *priv, int r,
232 const char *const units[], u32 status)
233 {
234 int i;
235
236 nv_error(priv, "PGRAPH_VSTATUS%d: 0x%08x", r, status);
237
238 for (i = 0; units[i] && status; i++) {
239 if ((status & 7) == 1)
240 pr_cont(" %s", units[i]);
241 status >>= 3;
242 }
243 if (status)
244 pr_cont(" (invalid: 0x%x)", status);
245 pr_cont("\n");
246 }
247
248 static int
249 nv84_graph_tlb_flush(struct nouveau_engine *engine)
250 {
251 struct nouveau_timer *ptimer = nouveau_timer(engine);
252 struct nv50_graph_priv *priv = (void *)engine;
253 bool idle, timeout = false;
254 unsigned long flags;
255 u64 start;
256 u32 tmp;
257
258 spin_lock_irqsave(&priv->lock, flags);
259 nv_mask(priv, 0x400500, 0x00000001, 0x00000000);
260
261 start = ptimer->read(ptimer);
262 do {
263 idle = true;
264
265 for (tmp = nv_rd32(priv, 0x400380); tmp && idle; tmp >>= 3) {
266 if ((tmp & 7) == 1)
267 idle = false;
268 }
269
270 for (tmp = nv_rd32(priv, 0x400384); tmp && idle; tmp >>= 3) {
271 if ((tmp & 7) == 1)
272 idle = false;
273 }
274
275 for (tmp = nv_rd32(priv, 0x400388); tmp && idle; tmp >>= 3) {
276 if ((tmp & 7) == 1)
277 idle = false;
278 }
279 } while (!idle &&
280 !(timeout = ptimer->read(ptimer) - start > 2000000000));
281
282 if (timeout) {
283 nv_error(priv, "PGRAPH TLB flush idle timeout fail\n");
284
285 tmp = nv_rd32(priv, 0x400700);
286 nv_error(priv, "PGRAPH_STATUS : 0x%08x", tmp);
287 nouveau_bitfield_print(nv50_pgraph_status, tmp);
288 pr_cont("\n");
289
290 nouveau_pgraph_vstatus_print(priv, 0, nv50_pgraph_vstatus_0,
291 nv_rd32(priv, 0x400380));
292 nouveau_pgraph_vstatus_print(priv, 1, nv50_pgraph_vstatus_1,
293 nv_rd32(priv, 0x400384));
294 nouveau_pgraph_vstatus_print(priv, 2, nv50_pgraph_vstatus_2,
295 nv_rd32(priv, 0x400388));
296 }
297
298
299 nv_wr32(priv, 0x100c80, 0x00000001);
300 if (!nv_wait(priv, 0x100c80, 0x00000001, 0x00000000))
301 nv_error(priv, "vm flush timeout\n");
302 nv_mask(priv, 0x400500, 0x00000001, 0x00000001);
303 spin_unlock_irqrestore(&priv->lock, flags);
304 return timeout ? -EBUSY : 0;
305 }
306
307 static const struct nouveau_bitfield nv50_mp_exec_errors[] = {
308 { 0x01, "STACK_UNDERFLOW" },
309 { 0x02, "STACK_MISMATCH" },
310 { 0x04, "QUADON_ACTIVE" },
311 { 0x08, "TIMEOUT" },
312 { 0x10, "INVALID_OPCODE" },
313 { 0x20, "PM_OVERFLOW" },
314 { 0x40, "BREAKPOINT" },
315 {}
316 };
317
318 static const struct nouveau_bitfield nv50_mpc_traps[] = {
319 { 0x0000001, "LOCAL_LIMIT_READ" },
320 { 0x0000010, "LOCAL_LIMIT_WRITE" },
321 { 0x0000040, "STACK_LIMIT" },
322 { 0x0000100, "GLOBAL_LIMIT_READ" },
323 { 0x0001000, "GLOBAL_LIMIT_WRITE" },
324 { 0x0010000, "MP0" },
325 { 0x0020000, "MP1" },
326 { 0x0040000, "GLOBAL_LIMIT_RED" },
327 { 0x0400000, "GLOBAL_LIMIT_ATOM" },
328 { 0x4000000, "MP2" },
329 {}
330 };
331
332 static const struct nouveau_bitfield nv50_graph_trap_m2mf[] = {
333 { 0x00000001, "NOTIFY" },
334 { 0x00000002, "IN" },
335 { 0x00000004, "OUT" },
336 {}
337 };
338
339 static const struct nouveau_bitfield nv50_graph_trap_vfetch[] = {
340 { 0x00000001, "FAULT" },
341 {}
342 };
343
344 static const struct nouveau_bitfield nv50_graph_trap_strmout[] = {
345 { 0x00000001, "FAULT" },
346 {}
347 };
348
349 static const struct nouveau_bitfield nv50_graph_trap_ccache[] = {
350 { 0x00000001, "FAULT" },
351 {}
352 };
353
354 /* There must be a *lot* of these. Will take some time to gather them up. */
355 const struct nouveau_enum nv50_data_error_names[] = {
356 { 0x00000003, "INVALID_OPERATION", NULL },
357 { 0x00000004, "INVALID_VALUE", NULL },
358 { 0x00000005, "INVALID_ENUM", NULL },
359 { 0x00000008, "INVALID_OBJECT", NULL },
360 { 0x00000009, "READ_ONLY_OBJECT", NULL },
361 { 0x0000000a, "SUPERVISOR_OBJECT", NULL },
362 { 0x0000000b, "INVALID_ADDRESS_ALIGNMENT", NULL },
363 { 0x0000000c, "INVALID_BITFIELD", NULL },
364 { 0x0000000d, "BEGIN_END_ACTIVE", NULL },
365 { 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT", NULL },
366 { 0x0000000f, "VIEWPORT_ID_NEEDS_GP", NULL },
367 { 0x00000010, "RT_DOUBLE_BIND", NULL },
368 { 0x00000011, "RT_TYPES_MISMATCH", NULL },
369 { 0x00000012, "RT_LINEAR_WITH_ZETA", NULL },
370 { 0x00000015, "FP_TOO_FEW_REGS", NULL },
371 { 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH", NULL },
372 { 0x00000017, "RT_LINEAR_WITH_MSAA", NULL },
373 { 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT", NULL },
374 { 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT", NULL },
375 { 0x0000001a, "RT_INVALID_ALIGNMENT", NULL },
376 { 0x0000001b, "SAMPLER_OVER_LIMIT", NULL },
377 { 0x0000001c, "TEXTURE_OVER_LIMIT", NULL },
378 { 0x0000001e, "GP_TOO_MANY_OUTPUTS", NULL },
379 { 0x0000001f, "RT_BPP128_WITH_MS8", NULL },
380 { 0x00000021, "Z_OUT_OF_BOUNDS", NULL },
381 { 0x00000023, "XY_OUT_OF_BOUNDS", NULL },
382 { 0x00000024, "VP_ZERO_INPUTS", NULL },
383 { 0x00000027, "CP_MORE_PARAMS_THAN_SHARED", NULL },
384 { 0x00000028, "CP_NO_REG_SPACE_STRIPED", NULL },
385 { 0x00000029, "CP_NO_REG_SPACE_PACKED", NULL },
386 { 0x0000002a, "CP_NOT_ENOUGH_WARPS", NULL },
387 { 0x0000002b, "CP_BLOCK_SIZE_MISMATCH", NULL },
388 { 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS", NULL },
389 { 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS", NULL },
390 { 0x0000002e, "CP_NO_BLOCKDIM_LATCH", NULL },
391 { 0x00000031, "ENG2D_FORMAT_MISMATCH", NULL },
392 { 0x0000003f, "PRIMITIVE_ID_NEEDS_GP", NULL },
393 { 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT", NULL },
394 { 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT", NULL },
395 { 0x00000046, "LAYER_ID_NEEDS_GP", NULL },
396 { 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT", NULL },
397 { 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT", NULL },
398 {}
399 };
400
401 static const struct nouveau_bitfield nv50_graph_intr_name[] = {
402 { 0x00000001, "NOTIFY" },
403 { 0x00000002, "COMPUTE_QUERY" },
404 { 0x00000010, "ILLEGAL_MTHD" },
405 { 0x00000020, "ILLEGAL_CLASS" },
406 { 0x00000040, "DOUBLE_NOTIFY" },
407 { 0x00001000, "CONTEXT_SWITCH" },
408 { 0x00010000, "BUFFER_NOTIFY" },
409 { 0x00100000, "DATA_ERROR" },
410 { 0x00200000, "TRAP" },
411 { 0x01000000, "SINGLE_STEP" },
412 {}
413 };
414
415 static const struct nouveau_bitfield nv50_graph_trap_prop[] = {
416 { 0x00000004, "SURF_WIDTH_OVERRUN" },
417 { 0x00000008, "SURF_HEIGHT_OVERRUN" },
418 { 0x00000010, "DST2D_FAULT" },
419 { 0x00000020, "ZETA_FAULT" },
420 { 0x00000040, "RT_FAULT" },
421 { 0x00000080, "CUDA_FAULT" },
422 { 0x00000100, "DST2D_STORAGE_TYPE_MISMATCH" },
423 { 0x00000200, "ZETA_STORAGE_TYPE_MISMATCH" },
424 { 0x00000400, "RT_STORAGE_TYPE_MISMATCH" },
425 { 0x00000800, "DST2D_LINEAR_MISMATCH" },
426 { 0x00001000, "RT_LINEAR_MISMATCH" },
427 {}
428 };
429
430 static void
431 nv50_priv_prop_trap(struct nv50_graph_priv *priv,
432 u32 ustatus_addr, u32 ustatus, u32 tp)
433 {
434 u32 e0c = nv_rd32(priv, ustatus_addr + 0x04);
435 u32 e10 = nv_rd32(priv, ustatus_addr + 0x08);
436 u32 e14 = nv_rd32(priv, ustatus_addr + 0x0c);
437 u32 e18 = nv_rd32(priv, ustatus_addr + 0x10);
438 u32 e1c = nv_rd32(priv, ustatus_addr + 0x14);
439 u32 e20 = nv_rd32(priv, ustatus_addr + 0x18);
440 u32 e24 = nv_rd32(priv, ustatus_addr + 0x1c);
441
442 /* CUDA memory: l[], g[] or stack. */
443 if (ustatus & 0x00000080) {
444 if (e18 & 0x80000000) {
445 /* g[] read fault? */
446 nv_error(priv, "TRAP_PROP - TP %d - CUDA_FAULT - Global read fault at address %02x%08x\n",
447 tp, e14, e10 | ((e18 >> 24) & 0x1f));
448 e18 &= ~0x1f000000;
449 } else if (e18 & 0xc) {
450 /* g[] write fault? */
451 nv_error(priv, "TRAP_PROP - TP %d - CUDA_FAULT - Global write fault at address %02x%08x\n",
452 tp, e14, e10 | ((e18 >> 7) & 0x1f));
453 e18 &= ~0x00000f80;
454 } else {
455 nv_error(priv, "TRAP_PROP - TP %d - Unknown CUDA fault at address %02x%08x\n",
456 tp, e14, e10);
457 }
458 ustatus &= ~0x00000080;
459 }
460 if (ustatus) {
461 nv_error(priv, "TRAP_PROP - TP %d -", tp);
462 nouveau_bitfield_print(nv50_graph_trap_prop, ustatus);
463 pr_cont(" - Address %02x%08x\n", e14, e10);
464 }
465 nv_error(priv, "TRAP_PROP - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
466 tp, e0c, e18, e1c, e20, e24);
467 }
468
469 static void
470 nv50_priv_mp_trap(struct nv50_graph_priv *priv, int tpid, int display)
471 {
472 u32 units = nv_rd32(priv, 0x1540);
473 u32 addr, mp10, status, pc, oplow, ophigh;
474 int i;
475 int mps = 0;
476 for (i = 0; i < 4; i++) {
477 if (!(units & 1 << (i+24)))
478 continue;
479 if (nv_device(priv)->chipset < 0xa0)
480 addr = 0x408200 + (tpid << 12) + (i << 7);
481 else
482 addr = 0x408100 + (tpid << 11) + (i << 7);
483 mp10 = nv_rd32(priv, addr + 0x10);
484 status = nv_rd32(priv, addr + 0x14);
485 if (!status)
486 continue;
487 if (display) {
488 nv_rd32(priv, addr + 0x20);
489 pc = nv_rd32(priv, addr + 0x24);
490 oplow = nv_rd32(priv, addr + 0x70);
491 ophigh = nv_rd32(priv, addr + 0x74);
492 nv_error(priv, "TRAP_MP_EXEC - "
493 "TP %d MP %d:", tpid, i);
494 nouveau_bitfield_print(nv50_mp_exec_errors, status);
495 pr_cont(" at %06x warp %d, opcode %08x %08x\n",
496 pc&0xffffff, pc >> 24,
497 oplow, ophigh);
498 }
499 nv_wr32(priv, addr + 0x10, mp10);
500 nv_wr32(priv, addr + 0x14, 0);
501 mps++;
502 }
503 if (!mps && display)
504 nv_error(priv, "TRAP_MP_EXEC - TP %d: "
505 "No MPs claiming errors?\n", tpid);
506 }
507
508 static void
509 nv50_priv_tp_trap(struct nv50_graph_priv *priv, int type, u32 ustatus_old,
510 u32 ustatus_new, int display, const char *name)
511 {
512 int tps = 0;
513 u32 units = nv_rd32(priv, 0x1540);
514 int i, r;
515 u32 ustatus_addr, ustatus;
516 for (i = 0; i < 16; i++) {
517 if (!(units & (1 << i)))
518 continue;
519 if (nv_device(priv)->chipset < 0xa0)
520 ustatus_addr = ustatus_old + (i << 12);
521 else
522 ustatus_addr = ustatus_new + (i << 11);
523 ustatus = nv_rd32(priv, ustatus_addr) & 0x7fffffff;
524 if (!ustatus)
525 continue;
526 tps++;
527 switch (type) {
528 case 6: /* texture error... unknown for now */
529 if (display) {
530 nv_error(priv, "magic set %d:\n", i);
531 for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
532 nv_error(priv, "\t0x%08x: 0x%08x\n", r,
533 nv_rd32(priv, r));
534 }
535 break;
536 case 7: /* MP error */
537 if (ustatus & 0x04030000) {
538 nv50_priv_mp_trap(priv, i, display);
539 ustatus &= ~0x04030000;
540 }
541 if (ustatus && display) {
542 nv_error("%s - TP%d:", name, i);
543 nouveau_bitfield_print(nv50_mpc_traps, ustatus);
544 pr_cont("\n");
545 ustatus = 0;
546 }
547 break;
548 case 8: /* PROP error */
549 if (display)
550 nv50_priv_prop_trap(
551 priv, ustatus_addr, ustatus, i);
552 ustatus = 0;
553 break;
554 }
555 if (ustatus) {
556 if (display)
557 nv_error(priv, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus);
558 }
559 nv_wr32(priv, ustatus_addr, 0xc0000000);
560 }
561
562 if (!tps && display)
563 nv_warn(priv, "%s - No TPs claiming errors?\n", name);
564 }
565
566 static int
567 nv50_graph_trap_handler(struct nv50_graph_priv *priv, u32 display,
568 int chid, u64 inst, struct nouveau_object *engctx)
569 {
570 u32 status = nv_rd32(priv, 0x400108);
571 u32 ustatus;
572
573 if (!status && display) {
574 nv_error(priv, "TRAP: no units reporting traps?\n");
575 return 1;
576 }
577
578 /* DISPATCH: Relays commands to other units and handles NOTIFY,
579 * COND, QUERY. If you get a trap from it, the command is still stuck
580 * in DISPATCH and you need to do something about it. */
581 if (status & 0x001) {
582 ustatus = nv_rd32(priv, 0x400804) & 0x7fffffff;
583 if (!ustatus && display) {
584 nv_error(priv, "TRAP_DISPATCH - no ustatus?\n");
585 }
586
587 nv_wr32(priv, 0x400500, 0x00000000);
588
589 /* Known to be triggered by screwed up NOTIFY and COND... */
590 if (ustatus & 0x00000001) {
591 u32 addr = nv_rd32(priv, 0x400808);
592 u32 subc = (addr & 0x00070000) >> 16;
593 u32 mthd = (addr & 0x00001ffc);
594 u32 datal = nv_rd32(priv, 0x40080c);
595 u32 datah = nv_rd32(priv, 0x400810);
596 u32 class = nv_rd32(priv, 0x400814);
597 u32 r848 = nv_rd32(priv, 0x400848);
598
599 nv_error(priv, "TRAP DISPATCH_FAULT\n");
600 if (display && (addr & 0x80000000)) {
601 nv_error(priv,
602 "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x%08x 400808 0x%08x 400848 0x%08x\n",
603 chid, inst,
604 nouveau_client_name(engctx), subc,
605 class, mthd, datah, datal, addr, r848);
606 } else
607 if (display) {
608 nv_error(priv, "no stuck command?\n");
609 }
610
611 nv_wr32(priv, 0x400808, 0);
612 nv_wr32(priv, 0x4008e8, nv_rd32(priv, 0x4008e8) & 3);
613 nv_wr32(priv, 0x400848, 0);
614 ustatus &= ~0x00000001;
615 }
616
617 if (ustatus & 0x00000002) {
618 u32 addr = nv_rd32(priv, 0x40084c);
619 u32 subc = (addr & 0x00070000) >> 16;
620 u32 mthd = (addr & 0x00001ffc);
621 u32 data = nv_rd32(priv, 0x40085c);
622 u32 class = nv_rd32(priv, 0x400814);
623
624 nv_error(priv, "TRAP DISPATCH_QUERY\n");
625 if (display && (addr & 0x80000000)) {
626 nv_error(priv,
627 "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x 40084c 0x%08x\n",
628 chid, inst,
629 nouveau_client_name(engctx), subc,
630 class, mthd, data, addr);
631 } else
632 if (display) {
633 nv_error(priv, "no stuck command?\n");
634 }
635
636 nv_wr32(priv, 0x40084c, 0);
637 ustatus &= ~0x00000002;
638 }
639
640 if (ustatus && display) {
641 nv_error(priv, "TRAP_DISPATCH (unknown "
642 "0x%08x)\n", ustatus);
643 }
644
645 nv_wr32(priv, 0x400804, 0xc0000000);
646 nv_wr32(priv, 0x400108, 0x001);
647 status &= ~0x001;
648 if (!status)
649 return 0;
650 }
651
652 /* M2MF: Memory to memory copy engine. */
653 if (status & 0x002) {
654 u32 ustatus = nv_rd32(priv, 0x406800) & 0x7fffffff;
655 if (display) {
656 nv_error(priv, "TRAP_M2MF");
657 nouveau_bitfield_print(nv50_graph_trap_m2mf, ustatus);
658 pr_cont("\n");
659 nv_error(priv, "TRAP_M2MF %08x %08x %08x %08x\n",
660 nv_rd32(priv, 0x406804), nv_rd32(priv, 0x406808),
661 nv_rd32(priv, 0x40680c), nv_rd32(priv, 0x406810));
662
663 }
664
665 /* No sane way found yet -- just reset the bugger. */
666 nv_wr32(priv, 0x400040, 2);
667 nv_wr32(priv, 0x400040, 0);
668 nv_wr32(priv, 0x406800, 0xc0000000);
669 nv_wr32(priv, 0x400108, 0x002);
670 status &= ~0x002;
671 }
672
673 /* VFETCH: Fetches data from vertex buffers. */
674 if (status & 0x004) {
675 u32 ustatus = nv_rd32(priv, 0x400c04) & 0x7fffffff;
676 if (display) {
677 nv_error(priv, "TRAP_VFETCH");
678 nouveau_bitfield_print(nv50_graph_trap_vfetch, ustatus);
679 pr_cont("\n");
680 nv_error(priv, "TRAP_VFETCH %08x %08x %08x %08x\n",
681 nv_rd32(priv, 0x400c00), nv_rd32(priv, 0x400c08),
682 nv_rd32(priv, 0x400c0c), nv_rd32(priv, 0x400c10));
683 }
684
685 nv_wr32(priv, 0x400c04, 0xc0000000);
686 nv_wr32(priv, 0x400108, 0x004);
687 status &= ~0x004;
688 }
689
690 /* STRMOUT: DirectX streamout / OpenGL transform feedback. */
691 if (status & 0x008) {
692 ustatus = nv_rd32(priv, 0x401800) & 0x7fffffff;
693 if (display) {
694 nv_error(priv, "TRAP_STRMOUT");
695 nouveau_bitfield_print(nv50_graph_trap_strmout, ustatus);
696 pr_cont("\n");
697 nv_error(priv, "TRAP_STRMOUT %08x %08x %08x %08x\n",
698 nv_rd32(priv, 0x401804), nv_rd32(priv, 0x401808),
699 nv_rd32(priv, 0x40180c), nv_rd32(priv, 0x401810));
700
701 }
702
703 /* No sane way found yet -- just reset the bugger. */
704 nv_wr32(priv, 0x400040, 0x80);
705 nv_wr32(priv, 0x400040, 0);
706 nv_wr32(priv, 0x401800, 0xc0000000);
707 nv_wr32(priv, 0x400108, 0x008);
708 status &= ~0x008;
709 }
710
711 /* CCACHE: Handles code and c[] caches and fills them. */
712 if (status & 0x010) {
713 ustatus = nv_rd32(priv, 0x405018) & 0x7fffffff;
714 if (display) {
715 nv_error(priv, "TRAP_CCACHE");
716 nouveau_bitfield_print(nv50_graph_trap_ccache, ustatus);
717 pr_cont("\n");
718 nv_error(priv, "TRAP_CCACHE %08x %08x %08x %08x"
719 " %08x %08x %08x\n",
720 nv_rd32(priv, 0x405000), nv_rd32(priv, 0x405004),
721 nv_rd32(priv, 0x405008), nv_rd32(priv, 0x40500c),
722 nv_rd32(priv, 0x405010), nv_rd32(priv, 0x405014),
723 nv_rd32(priv, 0x40501c));
724
725 }
726
727 nv_wr32(priv, 0x405018, 0xc0000000);
728 nv_wr32(priv, 0x400108, 0x010);
729 status &= ~0x010;
730 }
731
732 /* Unknown, not seen yet... 0x402000 is the only trap status reg
733 * remaining, so try to handle it anyway. Perhaps related to that
734 * unknown DMA slot on tesla? */
735 if (status & 0x20) {
736 ustatus = nv_rd32(priv, 0x402000) & 0x7fffffff;
737 if (display)
738 nv_error(priv, "TRAP_UNKC04 0x%08x\n", ustatus);
739 nv_wr32(priv, 0x402000, 0xc0000000);
740 /* no status modifiction on purpose */
741 }
742
743 /* TEXTURE: CUDA texturing units */
744 if (status & 0x040) {
745 nv50_priv_tp_trap(priv, 6, 0x408900, 0x408600, display,
746 "TRAP_TEXTURE");
747 nv_wr32(priv, 0x400108, 0x040);
748 status &= ~0x040;
749 }
750
751 /* MP: CUDA execution engines. */
752 if (status & 0x080) {
753 nv50_priv_tp_trap(priv, 7, 0x408314, 0x40831c, display,
754 "TRAP_MP");
755 nv_wr32(priv, 0x400108, 0x080);
756 status &= ~0x080;
757 }
758
759 /* PROP: Handles TP-initiated uncached memory accesses:
760 * l[], g[], stack, 2d surfaces, render targets. */
761 if (status & 0x100) {
762 nv50_priv_tp_trap(priv, 8, 0x408e08, 0x408708, display,
763 "TRAP_PROP");
764 nv_wr32(priv, 0x400108, 0x100);
765 status &= ~0x100;
766 }
767
768 if (status) {
769 if (display)
770 nv_error(priv, "TRAP: unknown 0x%08x\n", status);
771 nv_wr32(priv, 0x400108, status);
772 }
773
774 return 1;
775 }
776
777 static void
778 nv50_graph_intr(struct nouveau_subdev *subdev)
779 {
780 struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
781 struct nouveau_engine *engine = nv_engine(subdev);
782 struct nouveau_object *engctx;
783 struct nouveau_handle *handle = NULL;
784 struct nv50_graph_priv *priv = (void *)subdev;
785 u32 stat = nv_rd32(priv, 0x400100);
786 u32 inst = nv_rd32(priv, 0x40032c) & 0x0fffffff;
787 u32 addr = nv_rd32(priv, 0x400704);
788 u32 subc = (addr & 0x00070000) >> 16;
789 u32 mthd = (addr & 0x00001ffc);
790 u32 data = nv_rd32(priv, 0x400708);
791 u32 class = nv_rd32(priv, 0x400814);
792 u32 show = stat, show_bitfield = stat;
793 int chid;
794
795 engctx = nouveau_engctx_get(engine, inst);
796 chid = pfifo->chid(pfifo, engctx);
797
798 if (stat & 0x00000010) {
799 handle = nouveau_handle_get_class(engctx, class);
800 if (handle && !nv_call(handle->object, mthd, data))
801 show &= ~0x00000010;
802 nouveau_handle_put(handle);
803 }
804
805 if (show & 0x00100000) {
806 u32 ecode = nv_rd32(priv, 0x400110);
807 nv_error(priv, "DATA_ERROR ");
808 nouveau_enum_print(nv50_data_error_names, ecode);
809 pr_cont("\n");
810 show_bitfield &= ~0x00100000;
811 }
812
813 if (stat & 0x00200000) {
814 if (!nv50_graph_trap_handler(priv, show, chid, (u64)inst << 12,
815 engctx))
816 show &= ~0x00200000;
817 show_bitfield &= ~0x00200000;
818 }
819
820 nv_wr32(priv, 0x400100, stat);
821 nv_wr32(priv, 0x400500, 0x00010001);
822
823 if (show) {
824 show &= show_bitfield;
825 if (show) {
826 nv_error(priv, "%s", "");
827 nouveau_bitfield_print(nv50_graph_intr_name, show);
828 pr_cont("\n");
829 }
830 nv_error(priv,
831 "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x\n",
832 chid, (u64)inst << 12, nouveau_client_name(engctx),
833 subc, class, mthd, data);
834 }
835
836 if (nv_rd32(priv, 0x400824) & (1 << 31))
837 nv_wr32(priv, 0x400824, nv_rd32(priv, 0x400824) & ~(1 << 31));
838
839 nouveau_engctx_put(engctx);
840 }
841
842 static int
843 nv50_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
844 struct nouveau_oclass *oclass, void *data, u32 size,
845 struct nouveau_object **pobject)
846 {
847 struct nv50_graph_priv *priv;
848 int ret;
849
850 ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
851 *pobject = nv_object(priv);
852 if (ret)
853 return ret;
854
855 nv_subdev(priv)->unit = 0x00201000;
856 nv_subdev(priv)->intr = nv50_graph_intr;
857 nv_engine(priv)->cclass = &nv50_graph_cclass;
858
859 priv->base.units = nv50_graph_units;
860
861 switch (nv_device(priv)->chipset) {
862 case 0x50:
863 nv_engine(priv)->sclass = nv50_graph_sclass;
864 break;
865 case 0x84:
866 case 0x86:
867 case 0x92:
868 case 0x94:
869 case 0x96:
870 case 0x98:
871 nv_engine(priv)->sclass = nv84_graph_sclass;
872 break;
873 case 0xa0:
874 case 0xaa:
875 case 0xac:
876 nv_engine(priv)->sclass = nva0_graph_sclass;
877 break;
878 case 0xa3:
879 case 0xa5:
880 case 0xa8:
881 nv_engine(priv)->sclass = nva3_graph_sclass;
882 break;
883 case 0xaf:
884 nv_engine(priv)->sclass = nvaf_graph_sclass;
885 break;
886
887 };
888
889 /* unfortunate hw bug workaround... */
890 if (nv_device(priv)->chipset != 0x50 &&
891 nv_device(priv)->chipset != 0xac)
892 nv_engine(priv)->tlb_flush = nv84_graph_tlb_flush;
893
894 spin_lock_init(&priv->lock);
895 return 0;
896 }
897
898 static int
899 nv50_graph_init(struct nouveau_object *object)
900 {
901 struct nv50_graph_priv *priv = (void *)object;
902 int ret, units, i;
903
904 ret = nouveau_graph_init(&priv->base);
905 if (ret)
906 return ret;
907
908 /* NV_PGRAPH_DEBUG_3_HW_CTX_SWITCH_ENABLED */
909 nv_wr32(priv, 0x40008c, 0x00000004);
910
911 /* reset/enable traps and interrupts */
912 nv_wr32(priv, 0x400804, 0xc0000000);
913 nv_wr32(priv, 0x406800, 0xc0000000);
914 nv_wr32(priv, 0x400c04, 0xc0000000);
915 nv_wr32(priv, 0x401800, 0xc0000000);
916 nv_wr32(priv, 0x405018, 0xc0000000);
917 nv_wr32(priv, 0x402000, 0xc0000000);
918
919 units = nv_rd32(priv, 0x001540);
920 for (i = 0; i < 16; i++) {
921 if (!(units & (1 << i)))
922 continue;
923
924 if (nv_device(priv)->chipset < 0xa0) {
925 nv_wr32(priv, 0x408900 + (i << 12), 0xc0000000);
926 nv_wr32(priv, 0x408e08 + (i << 12), 0xc0000000);
927 nv_wr32(priv, 0x408314 + (i << 12), 0xc0000000);
928 } else {
929 nv_wr32(priv, 0x408600 + (i << 11), 0xc0000000);
930 nv_wr32(priv, 0x408708 + (i << 11), 0xc0000000);
931 nv_wr32(priv, 0x40831c + (i << 11), 0xc0000000);
932 }
933 }
934
935 nv_wr32(priv, 0x400108, 0xffffffff);
936 nv_wr32(priv, 0x400138, 0xffffffff);
937 nv_wr32(priv, 0x400100, 0xffffffff);
938 nv_wr32(priv, 0x40013c, 0xffffffff);
939 nv_wr32(priv, 0x400500, 0x00010001);
940
941 /* upload context program, initialise ctxctl defaults */
942 ret = nv50_grctx_init(nv_device(priv), &priv->size);
943 if (ret)
944 return ret;
945
946 nv_wr32(priv, 0x400824, 0x00000000);
947 nv_wr32(priv, 0x400828, 0x00000000);
948 nv_wr32(priv, 0x40082c, 0x00000000);
949 nv_wr32(priv, 0x400830, 0x00000000);
950 nv_wr32(priv, 0x40032c, 0x00000000);
951 nv_wr32(priv, 0x400330, 0x00000000);
952
953 /* some unknown zcull magic */
954 switch (nv_device(priv)->chipset & 0xf0) {
955 case 0x50:
956 case 0x80:
957 case 0x90:
958 nv_wr32(priv, 0x402ca8, 0x00000800);
959 break;
960 case 0xa0:
961 default:
962 nv_wr32(priv, 0x402cc0, 0x00000000);
963 if (nv_device(priv)->chipset == 0xa0 ||
964 nv_device(priv)->chipset == 0xaa ||
965 nv_device(priv)->chipset == 0xac) {
966 nv_wr32(priv, 0x402ca8, 0x00000802);
967 } else {
968 nv_wr32(priv, 0x402cc0, 0x00000000);
969 nv_wr32(priv, 0x402ca8, 0x00000002);
970 }
971
972 break;
973 }
974
975 /* zero out zcull regions */
976 for (i = 0; i < 8; i++) {
977 nv_wr32(priv, 0x402c20 + (i * 8), 0x00000000);
978 nv_wr32(priv, 0x402c24 + (i * 8), 0x00000000);
979 nv_wr32(priv, 0x402c28 + (i * 8), 0x00000000);
980 nv_wr32(priv, 0x402c2c + (i * 8), 0x00000000);
981 }
982 return 0;
983 }
984
985 struct nouveau_oclass
986 nv50_graph_oclass = {
987 .handle = NV_ENGINE(GR, 0x50),
988 .ofuncs = &(struct nouveau_ofuncs) {
989 .ctor = nv50_graph_ctor,
990 .dtor = _nouveau_graph_dtor,
991 .init = nv50_graph_init,
992 .fini = _nouveau_graph_fini,
993 },
994 };