2 * Copyright 2012 Red Hat Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
26 #include <core/client.h>
27 #include <core/handle.h>
28 #include <engine/fifo.h>
29 #include <subdev/timer.h>
38 struct nvkm_gr_chan base
;
42 nv50_gr_units(struct nvkm_gr
*gr
)
44 struct nv50_gr_priv
*priv
= (void *)gr
;
46 return nv_rd32(priv
, 0x1540);
49 /*******************************************************************************
50 * Graphics object classes
51 ******************************************************************************/
54 nv50_gr_object_ctor(struct nvkm_object
*parent
, struct nvkm_object
*engine
,
55 struct nvkm_oclass
*oclass
, void *data
, u32 size
,
56 struct nvkm_object
**pobject
)
58 struct nvkm_gpuobj
*obj
;
61 ret
= nvkm_gpuobj_create(parent
, engine
, oclass
, 0, parent
,
63 *pobject
= nv_object(obj
);
67 nv_wo32(obj
, 0x00, nv_mclass(obj
));
68 nv_wo32(obj
, 0x04, 0x00000000);
69 nv_wo32(obj
, 0x08, 0x00000000);
70 nv_wo32(obj
, 0x0c, 0x00000000);
74 static struct nvkm_ofuncs
76 .ctor
= nv50_gr_object_ctor
,
77 .dtor
= _nvkm_gpuobj_dtor
,
78 .init
= _nvkm_gpuobj_init
,
79 .fini
= _nvkm_gpuobj_fini
,
80 .rd32
= _nvkm_gpuobj_rd32
,
81 .wr32
= _nvkm_gpuobj_wr32
,
84 static struct nvkm_oclass
86 { 0x0030, &nv50_gr_ofuncs
},
87 { 0x502d, &nv50_gr_ofuncs
},
88 { 0x5039, &nv50_gr_ofuncs
},
89 { 0x5097, &nv50_gr_ofuncs
},
90 { 0x50c0, &nv50_gr_ofuncs
},
94 static struct nvkm_oclass
96 { 0x0030, &nv50_gr_ofuncs
},
97 { 0x502d, &nv50_gr_ofuncs
},
98 { 0x5039, &nv50_gr_ofuncs
},
99 { 0x50c0, &nv50_gr_ofuncs
},
100 { 0x8297, &nv50_gr_ofuncs
},
104 static struct nvkm_oclass
105 gt200_gr_sclass
[] = {
106 { 0x0030, &nv50_gr_ofuncs
},
107 { 0x502d, &nv50_gr_ofuncs
},
108 { 0x5039, &nv50_gr_ofuncs
},
109 { 0x50c0, &nv50_gr_ofuncs
},
110 { 0x8397, &nv50_gr_ofuncs
},
114 static struct nvkm_oclass
115 gt215_gr_sclass
[] = {
116 { 0x0030, &nv50_gr_ofuncs
},
117 { 0x502d, &nv50_gr_ofuncs
},
118 { 0x5039, &nv50_gr_ofuncs
},
119 { 0x50c0, &nv50_gr_ofuncs
},
120 { 0x8597, &nv50_gr_ofuncs
},
121 { 0x85c0, &nv50_gr_ofuncs
},
125 static struct nvkm_oclass
126 mcp89_gr_sclass
[] = {
127 { 0x0030, &nv50_gr_ofuncs
},
128 { 0x502d, &nv50_gr_ofuncs
},
129 { 0x5039, &nv50_gr_ofuncs
},
130 { 0x50c0, &nv50_gr_ofuncs
},
131 { 0x85c0, &nv50_gr_ofuncs
},
132 { 0x8697, &nv50_gr_ofuncs
},
136 /*******************************************************************************
138 ******************************************************************************/
141 nv50_gr_context_ctor(struct nvkm_object
*parent
, struct nvkm_object
*engine
,
142 struct nvkm_oclass
*oclass
, void *data
, u32 size
,
143 struct nvkm_object
**pobject
)
145 struct nv50_gr_priv
*priv
= (void *)engine
;
146 struct nv50_gr_chan
*chan
;
149 ret
= nvkm_gr_context_create(parent
, engine
, oclass
, NULL
, priv
->size
,
150 0, NVOBJ_FLAG_ZERO_ALLOC
, &chan
);
151 *pobject
= nv_object(chan
);
155 nv50_grctx_fill(nv_device(priv
), nv_gpuobj(chan
));
159 static struct nvkm_oclass
161 .handle
= NV_ENGCTX(GR
, 0x50),
162 .ofuncs
= &(struct nvkm_ofuncs
) {
163 .ctor
= nv50_gr_context_ctor
,
164 .dtor
= _nvkm_gr_context_dtor
,
165 .init
= _nvkm_gr_context_init
,
166 .fini
= _nvkm_gr_context_fini
,
167 .rd32
= _nvkm_gr_context_rd32
,
168 .wr32
= _nvkm_gr_context_wr32
,
172 /*******************************************************************************
173 * PGRAPH engine/subdev functions
174 ******************************************************************************/
176 static const struct nvkm_bitfield nv50_pgr_status
[] = {
177 { 0x00000001, "BUSY" }, /* set when any bit is set */
178 { 0x00000002, "DISPATCH" },
179 { 0x00000004, "UNK2" },
180 { 0x00000008, "UNK3" },
181 { 0x00000010, "UNK4" },
182 { 0x00000020, "UNK5" },
183 { 0x00000040, "M2MF" },
184 { 0x00000080, "UNK7" },
185 { 0x00000100, "CTXPROG" },
186 { 0x00000200, "VFETCH" },
187 { 0x00000400, "CCACHE_PREGEOM" },
188 { 0x00000800, "STRMOUT_VATTR_POSTGEOM" },
189 { 0x00001000, "VCLIP" },
190 { 0x00002000, "RATTR_APLANE" },
191 { 0x00004000, "TRAST" },
192 { 0x00008000, "CLIPID" },
193 { 0x00010000, "ZCULL" },
194 { 0x00020000, "ENG2D" },
195 { 0x00040000, "RMASK" },
196 { 0x00080000, "TPC_RAST" },
197 { 0x00100000, "TPC_PROP" },
198 { 0x00200000, "TPC_TEX" },
199 { 0x00400000, "TPC_GEOM" },
200 { 0x00800000, "TPC_MP" },
201 { 0x01000000, "ROP" },
205 static const char *const nv50_pgr_vstatus_0
[] = {
206 "VFETCH", "CCACHE", "PREGEOM", "POSTGEOM", "VATTR", "STRMOUT", "VCLIP",
210 static const char *const nv50_pgr_vstatus_1
[] = {
211 "TPC_RAST", "TPC_PROP", "TPC_TEX", "TPC_GEOM", "TPC_MP", NULL
214 static const char *const nv50_pgr_vstatus_2
[] = {
215 "RATTR", "APLANE", "TRAST", "CLIPID", "ZCULL", "ENG2D", "RMASK",
220 nvkm_pgr_vstatus_print(struct nv50_gr_priv
*priv
, int r
,
221 const char *const units
[], u32 status
)
225 nv_error(priv
, "PGRAPH_VSTATUS%d: 0x%08x", r
, status
);
227 for (i
= 0; units
[i
] && status
; i
++) {
228 if ((status
& 7) == 1)
229 pr_cont(" %s", units
[i
]);
233 pr_cont(" (invalid: 0x%x)", status
);
238 g84_gr_tlb_flush(struct nvkm_engine
*engine
)
240 struct nvkm_timer
*ptimer
= nvkm_timer(engine
);
241 struct nv50_gr_priv
*priv
= (void *)engine
;
242 bool idle
, timeout
= false;
247 spin_lock_irqsave(&priv
->lock
, flags
);
248 nv_mask(priv
, 0x400500, 0x00000001, 0x00000000);
250 start
= ptimer
->read(ptimer
);
254 for (tmp
= nv_rd32(priv
, 0x400380); tmp
&& idle
; tmp
>>= 3) {
259 for (tmp
= nv_rd32(priv
, 0x400384); tmp
&& idle
; tmp
>>= 3) {
264 for (tmp
= nv_rd32(priv
, 0x400388); tmp
&& idle
; tmp
>>= 3) {
269 !(timeout
= ptimer
->read(ptimer
) - start
> 2000000000));
272 nv_error(priv
, "PGRAPH TLB flush idle timeout fail\n");
274 tmp
= nv_rd32(priv
, 0x400700);
275 nv_error(priv
, "PGRAPH_STATUS : 0x%08x", tmp
);
276 nvkm_bitfield_print(nv50_pgr_status
, tmp
);
279 nvkm_pgr_vstatus_print(priv
, 0, nv50_pgr_vstatus_0
,
280 nv_rd32(priv
, 0x400380));
281 nvkm_pgr_vstatus_print(priv
, 1, nv50_pgr_vstatus_1
,
282 nv_rd32(priv
, 0x400384));
283 nvkm_pgr_vstatus_print(priv
, 2, nv50_pgr_vstatus_2
,
284 nv_rd32(priv
, 0x400388));
288 nv_wr32(priv
, 0x100c80, 0x00000001);
289 if (!nv_wait(priv
, 0x100c80, 0x00000001, 0x00000000))
290 nv_error(priv
, "vm flush timeout\n");
291 nv_mask(priv
, 0x400500, 0x00000001, 0x00000001);
292 spin_unlock_irqrestore(&priv
->lock
, flags
);
293 return timeout
? -EBUSY
: 0;
296 static const struct nvkm_bitfield nv50_mp_exec_errors
[] = {
297 { 0x01, "STACK_UNDERFLOW" },
298 { 0x02, "STACK_MISMATCH" },
299 { 0x04, "QUADON_ACTIVE" },
301 { 0x10, "INVALID_OPCODE" },
302 { 0x20, "PM_OVERFLOW" },
303 { 0x40, "BREAKPOINT" },
307 static const struct nvkm_bitfield nv50_mpc_traps
[] = {
308 { 0x0000001, "LOCAL_LIMIT_READ" },
309 { 0x0000010, "LOCAL_LIMIT_WRITE" },
310 { 0x0000040, "STACK_LIMIT" },
311 { 0x0000100, "GLOBAL_LIMIT_READ" },
312 { 0x0001000, "GLOBAL_LIMIT_WRITE" },
313 { 0x0010000, "MP0" },
314 { 0x0020000, "MP1" },
315 { 0x0040000, "GLOBAL_LIMIT_RED" },
316 { 0x0400000, "GLOBAL_LIMIT_ATOM" },
317 { 0x4000000, "MP2" },
321 static const struct nvkm_bitfield nv50_tex_traps
[] = {
322 { 0x00000001, "" }, /* any bit set? */
323 { 0x00000002, "FAULT" },
324 { 0x00000004, "STORAGE_TYPE_MISMATCH" },
325 { 0x00000008, "LINEAR_MISMATCH" },
326 { 0x00000020, "WRONG_MEMTYPE" },
330 static const struct nvkm_bitfield nv50_gr_trap_m2mf
[] = {
331 { 0x00000001, "NOTIFY" },
332 { 0x00000002, "IN" },
333 { 0x00000004, "OUT" },
337 static const struct nvkm_bitfield nv50_gr_trap_vfetch
[] = {
338 { 0x00000001, "FAULT" },
342 static const struct nvkm_bitfield nv50_gr_trap_strmout
[] = {
343 { 0x00000001, "FAULT" },
347 static const struct nvkm_bitfield nv50_gr_trap_ccache
[] = {
348 { 0x00000001, "FAULT" },
352 /* There must be a *lot* of these. Will take some time to gather them up. */
353 const struct nvkm_enum nv50_data_error_names
[] = {
354 { 0x00000003, "INVALID_OPERATION", NULL
},
355 { 0x00000004, "INVALID_VALUE", NULL
},
356 { 0x00000005, "INVALID_ENUM", NULL
},
357 { 0x00000008, "INVALID_OBJECT", NULL
},
358 { 0x00000009, "READ_ONLY_OBJECT", NULL
},
359 { 0x0000000a, "SUPERVISOR_OBJECT", NULL
},
360 { 0x0000000b, "INVALID_ADDRESS_ALIGNMENT", NULL
},
361 { 0x0000000c, "INVALID_BITFIELD", NULL
},
362 { 0x0000000d, "BEGIN_END_ACTIVE", NULL
},
363 { 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT", NULL
},
364 { 0x0000000f, "VIEWPORT_ID_NEEDS_GP", NULL
},
365 { 0x00000010, "RT_DOUBLE_BIND", NULL
},
366 { 0x00000011, "RT_TYPES_MISMATCH", NULL
},
367 { 0x00000012, "RT_LINEAR_WITH_ZETA", NULL
},
368 { 0x00000015, "FP_TOO_FEW_REGS", NULL
},
369 { 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH", NULL
},
370 { 0x00000017, "RT_LINEAR_WITH_MSAA", NULL
},
371 { 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT", NULL
},
372 { 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT", NULL
},
373 { 0x0000001a, "RT_INVALID_ALIGNMENT", NULL
},
374 { 0x0000001b, "SAMPLER_OVER_LIMIT", NULL
},
375 { 0x0000001c, "TEXTURE_OVER_LIMIT", NULL
},
376 { 0x0000001e, "GP_TOO_MANY_OUTPUTS", NULL
},
377 { 0x0000001f, "RT_BPP128_WITH_MS8", NULL
},
378 { 0x00000021, "Z_OUT_OF_BOUNDS", NULL
},
379 { 0x00000023, "XY_OUT_OF_BOUNDS", NULL
},
380 { 0x00000024, "VP_ZERO_INPUTS", NULL
},
381 { 0x00000027, "CP_MORE_PARAMS_THAN_SHARED", NULL
},
382 { 0x00000028, "CP_NO_REG_SPACE_STRIPED", NULL
},
383 { 0x00000029, "CP_NO_REG_SPACE_PACKED", NULL
},
384 { 0x0000002a, "CP_NOT_ENOUGH_WARPS", NULL
},
385 { 0x0000002b, "CP_BLOCK_SIZE_MISMATCH", NULL
},
386 { 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS", NULL
},
387 { 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS", NULL
},
388 { 0x0000002e, "CP_NO_BLOCKDIM_LATCH", NULL
},
389 { 0x00000031, "ENG2D_FORMAT_MISMATCH", NULL
},
390 { 0x0000003f, "PRIMITIVE_ID_NEEDS_GP", NULL
},
391 { 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT", NULL
},
392 { 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT", NULL
},
393 { 0x00000046, "LAYER_ID_NEEDS_GP", NULL
},
394 { 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT", NULL
},
395 { 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT", NULL
},
399 static const struct nvkm_bitfield nv50_gr_intr_name
[] = {
400 { 0x00000001, "NOTIFY" },
401 { 0x00000002, "COMPUTE_QUERY" },
402 { 0x00000010, "ILLEGAL_MTHD" },
403 { 0x00000020, "ILLEGAL_CLASS" },
404 { 0x00000040, "DOUBLE_NOTIFY" },
405 { 0x00001000, "CONTEXT_SWITCH" },
406 { 0x00010000, "BUFFER_NOTIFY" },
407 { 0x00100000, "DATA_ERROR" },
408 { 0x00200000, "TRAP" },
409 { 0x01000000, "SINGLE_STEP" },
413 static const struct nvkm_bitfield nv50_gr_trap_prop
[] = {
414 { 0x00000004, "SURF_WIDTH_OVERRUN" },
415 { 0x00000008, "SURF_HEIGHT_OVERRUN" },
416 { 0x00000010, "DST2D_FAULT" },
417 { 0x00000020, "ZETA_FAULT" },
418 { 0x00000040, "RT_FAULT" },
419 { 0x00000080, "CUDA_FAULT" },
420 { 0x00000100, "DST2D_STORAGE_TYPE_MISMATCH" },
421 { 0x00000200, "ZETA_STORAGE_TYPE_MISMATCH" },
422 { 0x00000400, "RT_STORAGE_TYPE_MISMATCH" },
423 { 0x00000800, "DST2D_LINEAR_MISMATCH" },
424 { 0x00001000, "RT_LINEAR_MISMATCH" },
429 nv50_priv_prop_trap(struct nv50_gr_priv
*priv
,
430 u32 ustatus_addr
, u32 ustatus
, u32 tp
)
432 u32 e0c
= nv_rd32(priv
, ustatus_addr
+ 0x04);
433 u32 e10
= nv_rd32(priv
, ustatus_addr
+ 0x08);
434 u32 e14
= nv_rd32(priv
, ustatus_addr
+ 0x0c);
435 u32 e18
= nv_rd32(priv
, ustatus_addr
+ 0x10);
436 u32 e1c
= nv_rd32(priv
, ustatus_addr
+ 0x14);
437 u32 e20
= nv_rd32(priv
, ustatus_addr
+ 0x18);
438 u32 e24
= nv_rd32(priv
, ustatus_addr
+ 0x1c);
440 /* CUDA memory: l[], g[] or stack. */
441 if (ustatus
& 0x00000080) {
442 if (e18
& 0x80000000) {
443 /* g[] read fault? */
444 nv_error(priv
, "TRAP_PROP - TP %d - CUDA_FAULT - Global read fault at address %02x%08x\n",
445 tp
, e14
, e10
| ((e18
>> 24) & 0x1f));
447 } else if (e18
& 0xc) {
448 /* g[] write fault? */
449 nv_error(priv
, "TRAP_PROP - TP %d - CUDA_FAULT - Global write fault at address %02x%08x\n",
450 tp
, e14
, e10
| ((e18
>> 7) & 0x1f));
453 nv_error(priv
, "TRAP_PROP - TP %d - Unknown CUDA fault at address %02x%08x\n",
456 ustatus
&= ~0x00000080;
459 nv_error(priv
, "TRAP_PROP - TP %d -", tp
);
460 nvkm_bitfield_print(nv50_gr_trap_prop
, ustatus
);
461 pr_cont(" - Address %02x%08x\n", e14
, e10
);
463 nv_error(priv
, "TRAP_PROP - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
464 tp
, e0c
, e18
, e1c
, e20
, e24
);
468 nv50_priv_mp_trap(struct nv50_gr_priv
*priv
, int tpid
, int display
)
470 u32 units
= nv_rd32(priv
, 0x1540);
471 u32 addr
, mp10
, status
, pc
, oplow
, ophigh
;
474 for (i
= 0; i
< 4; i
++) {
475 if (!(units
& 1 << (i
+24)))
477 if (nv_device(priv
)->chipset
< 0xa0)
478 addr
= 0x408200 + (tpid
<< 12) + (i
<< 7);
480 addr
= 0x408100 + (tpid
<< 11) + (i
<< 7);
481 mp10
= nv_rd32(priv
, addr
+ 0x10);
482 status
= nv_rd32(priv
, addr
+ 0x14);
486 nv_rd32(priv
, addr
+ 0x20);
487 pc
= nv_rd32(priv
, addr
+ 0x24);
488 oplow
= nv_rd32(priv
, addr
+ 0x70);
489 ophigh
= nv_rd32(priv
, addr
+ 0x74);
490 nv_error(priv
, "TRAP_MP_EXEC - "
491 "TP %d MP %d:", tpid
, i
);
492 nvkm_bitfield_print(nv50_mp_exec_errors
, status
);
493 pr_cont(" at %06x warp %d, opcode %08x %08x\n",
494 pc
&0xffffff, pc
>> 24,
497 nv_wr32(priv
, addr
+ 0x10, mp10
);
498 nv_wr32(priv
, addr
+ 0x14, 0);
502 nv_error(priv
, "TRAP_MP_EXEC - TP %d: "
503 "No MPs claiming errors?\n", tpid
);
507 nv50_priv_tp_trap(struct nv50_gr_priv
*priv
, int type
, u32 ustatus_old
,
508 u32 ustatus_new
, int display
, const char *name
)
511 u32 units
= nv_rd32(priv
, 0x1540);
513 u32 ustatus_addr
, ustatus
;
514 for (i
= 0; i
< 16; i
++) {
515 if (!(units
& (1 << i
)))
517 if (nv_device(priv
)->chipset
< 0xa0)
518 ustatus_addr
= ustatus_old
+ (i
<< 12);
520 ustatus_addr
= ustatus_new
+ (i
<< 11);
521 ustatus
= nv_rd32(priv
, ustatus_addr
) & 0x7fffffff;
526 case 6: /* texture error... unknown for now */
528 nv_error(priv
, "magic set %d:\n", i
);
529 for (r
= ustatus_addr
+ 4; r
<= ustatus_addr
+ 0x10; r
+= 4)
530 nv_error(priv
, "\t0x%08x: 0x%08x\n", r
,
533 nv_error(priv
, "%s - TP%d:", name
, i
);
534 nvkm_bitfield_print(nv50_tex_traps
,
541 case 7: /* MP error */
542 if (ustatus
& 0x04030000) {
543 nv50_priv_mp_trap(priv
, i
, display
);
544 ustatus
&= ~0x04030000;
546 if (ustatus
&& display
) {
547 nv_error(priv
, "%s - TP%d:", name
, i
);
548 nvkm_bitfield_print(nv50_mpc_traps
, ustatus
);
553 case 8: /* PROP error */
556 priv
, ustatus_addr
, ustatus
, i
);
562 nv_error(priv
, "%s - TP%d: Unhandled ustatus 0x%08x\n", name
, i
, ustatus
);
564 nv_wr32(priv
, ustatus_addr
, 0xc0000000);
568 nv_warn(priv
, "%s - No TPs claiming errors?\n", name
);
572 nv50_gr_trap_handler(struct nv50_gr_priv
*priv
, u32 display
,
573 int chid
, u64 inst
, struct nvkm_object
*engctx
)
575 u32 status
= nv_rd32(priv
, 0x400108);
578 if (!status
&& display
) {
579 nv_error(priv
, "TRAP: no units reporting traps?\n");
583 /* DISPATCH: Relays commands to other units and handles NOTIFY,
584 * COND, QUERY. If you get a trap from it, the command is still stuck
585 * in DISPATCH and you need to do something about it. */
586 if (status
& 0x001) {
587 ustatus
= nv_rd32(priv
, 0x400804) & 0x7fffffff;
588 if (!ustatus
&& display
) {
589 nv_error(priv
, "TRAP_DISPATCH - no ustatus?\n");
592 nv_wr32(priv
, 0x400500, 0x00000000);
594 /* Known to be triggered by screwed up NOTIFY and COND... */
595 if (ustatus
& 0x00000001) {
596 u32 addr
= nv_rd32(priv
, 0x400808);
597 u32 subc
= (addr
& 0x00070000) >> 16;
598 u32 mthd
= (addr
& 0x00001ffc);
599 u32 datal
= nv_rd32(priv
, 0x40080c);
600 u32 datah
= nv_rd32(priv
, 0x400810);
601 u32
class = nv_rd32(priv
, 0x400814);
602 u32 r848
= nv_rd32(priv
, 0x400848);
604 nv_error(priv
, "TRAP DISPATCH_FAULT\n");
605 if (display
&& (addr
& 0x80000000)) {
607 "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x%08x 400808 0x%08x 400848 0x%08x\n",
609 nvkm_client_name(engctx
), subc
,
610 class, mthd
, datah
, datal
, addr
, r848
);
613 nv_error(priv
, "no stuck command?\n");
616 nv_wr32(priv
, 0x400808, 0);
617 nv_wr32(priv
, 0x4008e8, nv_rd32(priv
, 0x4008e8) & 3);
618 nv_wr32(priv
, 0x400848, 0);
619 ustatus
&= ~0x00000001;
622 if (ustatus
& 0x00000002) {
623 u32 addr
= nv_rd32(priv
, 0x40084c);
624 u32 subc
= (addr
& 0x00070000) >> 16;
625 u32 mthd
= (addr
& 0x00001ffc);
626 u32 data
= nv_rd32(priv
, 0x40085c);
627 u32
class = nv_rd32(priv
, 0x400814);
629 nv_error(priv
, "TRAP DISPATCH_QUERY\n");
630 if (display
&& (addr
& 0x80000000)) {
632 "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x 40084c 0x%08x\n",
634 nvkm_client_name(engctx
), subc
,
635 class, mthd
, data
, addr
);
638 nv_error(priv
, "no stuck command?\n");
641 nv_wr32(priv
, 0x40084c, 0);
642 ustatus
&= ~0x00000002;
645 if (ustatus
&& display
) {
646 nv_error(priv
, "TRAP_DISPATCH (unknown "
647 "0x%08x)\n", ustatus
);
650 nv_wr32(priv
, 0x400804, 0xc0000000);
651 nv_wr32(priv
, 0x400108, 0x001);
657 /* M2MF: Memory to memory copy engine. */
658 if (status
& 0x002) {
659 u32 ustatus
= nv_rd32(priv
, 0x406800) & 0x7fffffff;
661 nv_error(priv
, "TRAP_M2MF");
662 nvkm_bitfield_print(nv50_gr_trap_m2mf
, ustatus
);
664 nv_error(priv
, "TRAP_M2MF %08x %08x %08x %08x\n",
665 nv_rd32(priv
, 0x406804), nv_rd32(priv
, 0x406808),
666 nv_rd32(priv
, 0x40680c), nv_rd32(priv
, 0x406810));
670 /* No sane way found yet -- just reset the bugger. */
671 nv_wr32(priv
, 0x400040, 2);
672 nv_wr32(priv
, 0x400040, 0);
673 nv_wr32(priv
, 0x406800, 0xc0000000);
674 nv_wr32(priv
, 0x400108, 0x002);
678 /* VFETCH: Fetches data from vertex buffers. */
679 if (status
& 0x004) {
680 u32 ustatus
= nv_rd32(priv
, 0x400c04) & 0x7fffffff;
682 nv_error(priv
, "TRAP_VFETCH");
683 nvkm_bitfield_print(nv50_gr_trap_vfetch
, ustatus
);
685 nv_error(priv
, "TRAP_VFETCH %08x %08x %08x %08x\n",
686 nv_rd32(priv
, 0x400c00), nv_rd32(priv
, 0x400c08),
687 nv_rd32(priv
, 0x400c0c), nv_rd32(priv
, 0x400c10));
690 nv_wr32(priv
, 0x400c04, 0xc0000000);
691 nv_wr32(priv
, 0x400108, 0x004);
695 /* STRMOUT: DirectX streamout / OpenGL transform feedback. */
696 if (status
& 0x008) {
697 ustatus
= nv_rd32(priv
, 0x401800) & 0x7fffffff;
699 nv_error(priv
, "TRAP_STRMOUT");
700 nvkm_bitfield_print(nv50_gr_trap_strmout
, ustatus
);
702 nv_error(priv
, "TRAP_STRMOUT %08x %08x %08x %08x\n",
703 nv_rd32(priv
, 0x401804), nv_rd32(priv
, 0x401808),
704 nv_rd32(priv
, 0x40180c), nv_rd32(priv
, 0x401810));
708 /* No sane way found yet -- just reset the bugger. */
709 nv_wr32(priv
, 0x400040, 0x80);
710 nv_wr32(priv
, 0x400040, 0);
711 nv_wr32(priv
, 0x401800, 0xc0000000);
712 nv_wr32(priv
, 0x400108, 0x008);
716 /* CCACHE: Handles code and c[] caches and fills them. */
717 if (status
& 0x010) {
718 ustatus
= nv_rd32(priv
, 0x405018) & 0x7fffffff;
720 nv_error(priv
, "TRAP_CCACHE");
721 nvkm_bitfield_print(nv50_gr_trap_ccache
, ustatus
);
723 nv_error(priv
, "TRAP_CCACHE %08x %08x %08x %08x"
725 nv_rd32(priv
, 0x405000), nv_rd32(priv
, 0x405004),
726 nv_rd32(priv
, 0x405008), nv_rd32(priv
, 0x40500c),
727 nv_rd32(priv
, 0x405010), nv_rd32(priv
, 0x405014),
728 nv_rd32(priv
, 0x40501c));
732 nv_wr32(priv
, 0x405018, 0xc0000000);
733 nv_wr32(priv
, 0x400108, 0x010);
737 /* Unknown, not seen yet... 0x402000 is the only trap status reg
738 * remaining, so try to handle it anyway. Perhaps related to that
739 * unknown DMA slot on tesla? */
741 ustatus
= nv_rd32(priv
, 0x402000) & 0x7fffffff;
743 nv_error(priv
, "TRAP_UNKC04 0x%08x\n", ustatus
);
744 nv_wr32(priv
, 0x402000, 0xc0000000);
745 /* no status modifiction on purpose */
748 /* TEXTURE: CUDA texturing units */
749 if (status
& 0x040) {
750 nv50_priv_tp_trap(priv
, 6, 0x408900, 0x408600, display
,
752 nv_wr32(priv
, 0x400108, 0x040);
756 /* MP: CUDA execution engines. */
757 if (status
& 0x080) {
758 nv50_priv_tp_trap(priv
, 7, 0x408314, 0x40831c, display
,
760 nv_wr32(priv
, 0x400108, 0x080);
764 /* PROP: Handles TP-initiated uncached memory accesses:
765 * l[], g[], stack, 2d surfaces, render targets. */
766 if (status
& 0x100) {
767 nv50_priv_tp_trap(priv
, 8, 0x408e08, 0x408708, display
,
769 nv_wr32(priv
, 0x400108, 0x100);
775 nv_error(priv
, "TRAP: unknown 0x%08x\n", status
);
776 nv_wr32(priv
, 0x400108, status
);
783 nv50_gr_intr(struct nvkm_subdev
*subdev
)
785 struct nvkm_fifo
*pfifo
= nvkm_fifo(subdev
);
786 struct nvkm_engine
*engine
= nv_engine(subdev
);
787 struct nvkm_object
*engctx
;
788 struct nvkm_handle
*handle
= NULL
;
789 struct nv50_gr_priv
*priv
= (void *)subdev
;
790 u32 stat
= nv_rd32(priv
, 0x400100);
791 u32 inst
= nv_rd32(priv
, 0x40032c) & 0x0fffffff;
792 u32 addr
= nv_rd32(priv
, 0x400704);
793 u32 subc
= (addr
& 0x00070000) >> 16;
794 u32 mthd
= (addr
& 0x00001ffc);
795 u32 data
= nv_rd32(priv
, 0x400708);
796 u32
class = nv_rd32(priv
, 0x400814);
797 u32 show
= stat
, show_bitfield
= stat
;
800 engctx
= nvkm_engctx_get(engine
, inst
);
801 chid
= pfifo
->chid(pfifo
, engctx
);
803 if (stat
& 0x00000010) {
804 handle
= nvkm_handle_get_class(engctx
, class);
805 if (handle
&& !nv_call(handle
->object
, mthd
, data
))
807 nvkm_handle_put(handle
);
810 if (show
& 0x00100000) {
811 u32 ecode
= nv_rd32(priv
, 0x400110);
812 nv_error(priv
, "DATA_ERROR ");
813 nvkm_enum_print(nv50_data_error_names
, ecode
);
815 show_bitfield
&= ~0x00100000;
818 if (stat
& 0x00200000) {
819 if (!nv50_gr_trap_handler(priv
, show
, chid
, (u64
)inst
<< 12,
822 show_bitfield
&= ~0x00200000;
825 nv_wr32(priv
, 0x400100, stat
);
826 nv_wr32(priv
, 0x400500, 0x00010001);
829 show
&= show_bitfield
;
831 nv_error(priv
, "%s", "");
832 nvkm_bitfield_print(nv50_gr_intr_name
, show
);
836 "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x\n",
837 chid
, (u64
)inst
<< 12, nvkm_client_name(engctx
),
838 subc
, class, mthd
, data
);
841 if (nv_rd32(priv
, 0x400824) & (1 << 31))
842 nv_wr32(priv
, 0x400824, nv_rd32(priv
, 0x400824) & ~(1 << 31));
844 nvkm_engctx_put(engctx
);
848 nv50_gr_ctor(struct nvkm_object
*parent
, struct nvkm_object
*engine
,
849 struct nvkm_oclass
*oclass
, void *data
, u32 size
,
850 struct nvkm_object
**pobject
)
852 struct nv50_gr_priv
*priv
;
855 ret
= nvkm_gr_create(parent
, engine
, oclass
, true, &priv
);
856 *pobject
= nv_object(priv
);
860 nv_subdev(priv
)->unit
= 0x00201000;
861 nv_subdev(priv
)->intr
= nv50_gr_intr
;
862 nv_engine(priv
)->cclass
= &nv50_gr_cclass
;
864 priv
->base
.units
= nv50_gr_units
;
866 switch (nv_device(priv
)->chipset
) {
868 nv_engine(priv
)->sclass
= nv50_gr_sclass
;
876 nv_engine(priv
)->sclass
= g84_gr_sclass
;
881 nv_engine(priv
)->sclass
= gt200_gr_sclass
;
886 nv_engine(priv
)->sclass
= gt215_gr_sclass
;
889 nv_engine(priv
)->sclass
= mcp89_gr_sclass
;
894 /* unfortunate hw bug workaround... */
895 if (nv_device(priv
)->chipset
!= 0x50 &&
896 nv_device(priv
)->chipset
!= 0xac)
897 nv_engine(priv
)->tlb_flush
= g84_gr_tlb_flush
;
899 spin_lock_init(&priv
->lock
);
904 nv50_gr_init(struct nvkm_object
*object
)
906 struct nv50_gr_priv
*priv
= (void *)object
;
909 ret
= nvkm_gr_init(&priv
->base
);
913 /* NV_PGRAPH_DEBUG_3_HW_CTX_SWITCH_ENABLED */
914 nv_wr32(priv
, 0x40008c, 0x00000004);
916 /* reset/enable traps and interrupts */
917 nv_wr32(priv
, 0x400804, 0xc0000000);
918 nv_wr32(priv
, 0x406800, 0xc0000000);
919 nv_wr32(priv
, 0x400c04, 0xc0000000);
920 nv_wr32(priv
, 0x401800, 0xc0000000);
921 nv_wr32(priv
, 0x405018, 0xc0000000);
922 nv_wr32(priv
, 0x402000, 0xc0000000);
924 units
= nv_rd32(priv
, 0x001540);
925 for (i
= 0; i
< 16; i
++) {
926 if (!(units
& (1 << i
)))
929 if (nv_device(priv
)->chipset
< 0xa0) {
930 nv_wr32(priv
, 0x408900 + (i
<< 12), 0xc0000000);
931 nv_wr32(priv
, 0x408e08 + (i
<< 12), 0xc0000000);
932 nv_wr32(priv
, 0x408314 + (i
<< 12), 0xc0000000);
934 nv_wr32(priv
, 0x408600 + (i
<< 11), 0xc0000000);
935 nv_wr32(priv
, 0x408708 + (i
<< 11), 0xc0000000);
936 nv_wr32(priv
, 0x40831c + (i
<< 11), 0xc0000000);
940 nv_wr32(priv
, 0x400108, 0xffffffff);
941 nv_wr32(priv
, 0x400138, 0xffffffff);
942 nv_wr32(priv
, 0x400100, 0xffffffff);
943 nv_wr32(priv
, 0x40013c, 0xffffffff);
944 nv_wr32(priv
, 0x400500, 0x00010001);
946 /* upload context program, initialise ctxctl defaults */
947 ret
= nv50_grctx_init(nv_device(priv
), &priv
->size
);
951 nv_wr32(priv
, 0x400824, 0x00000000);
952 nv_wr32(priv
, 0x400828, 0x00000000);
953 nv_wr32(priv
, 0x40082c, 0x00000000);
954 nv_wr32(priv
, 0x400830, 0x00000000);
955 nv_wr32(priv
, 0x40032c, 0x00000000);
956 nv_wr32(priv
, 0x400330, 0x00000000);
958 /* some unknown zcull magic */
959 switch (nv_device(priv
)->chipset
& 0xf0) {
963 nv_wr32(priv
, 0x402ca8, 0x00000800);
967 if (nv_device(priv
)->chipset
== 0xa0 ||
968 nv_device(priv
)->chipset
== 0xaa ||
969 nv_device(priv
)->chipset
== 0xac) {
970 nv_wr32(priv
, 0x402ca8, 0x00000802);
972 nv_wr32(priv
, 0x402cc0, 0x00000000);
973 nv_wr32(priv
, 0x402ca8, 0x00000002);
979 /* zero out zcull regions */
980 for (i
= 0; i
< 8; i
++) {
981 nv_wr32(priv
, 0x402c20 + (i
* 0x10), 0x00000000);
982 nv_wr32(priv
, 0x402c24 + (i
* 0x10), 0x00000000);
983 nv_wr32(priv
, 0x402c28 + (i
* 0x10), 0x00000000);
984 nv_wr32(priv
, 0x402c2c + (i
* 0x10), 0x00000000);
991 .handle
= NV_ENGINE(GR
, 0x50),
992 .ofuncs
= &(struct nvkm_ofuncs
) {
993 .ctor
= nv50_gr_ctor
,
994 .dtor
= _nvkm_gr_dtor
,
995 .init
= nv50_gr_init
,
996 .fini
= _nvkm_gr_fini
,