xref: /linux/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
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/device.h>
28 #include <core/handle.h>
29 #include <engine/fifo.h>
30 #include <subdev/timer.h>
31 
32 struct nv50_gr_priv {
33 	struct nvkm_gr base;
34 	spinlock_t lock;
35 	u32 size;
36 };
37 
38 struct nv50_gr_chan {
39 	struct nvkm_gr_chan base;
40 };
41 
42 static u64
43 nv50_gr_units(struct nvkm_gr *gr)
44 {
45 	struct nv50_gr_priv *priv = (void *)gr;
46 
47 	return nv_rd32(priv, 0x1540);
48 }
49 
50 /*******************************************************************************
51  * Graphics object classes
52  ******************************************************************************/
53 
54 static int
55 nv50_gr_object_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
56 		    struct nvkm_oclass *oclass, void *data, u32 size,
57 		    struct nvkm_object **pobject)
58 {
59 	struct nvkm_gpuobj *obj;
60 	int ret;
61 
62 	ret = nvkm_gpuobj_create(parent, engine, oclass, 0, parent,
63 				 16, 16, 0, &obj);
64 	*pobject = nv_object(obj);
65 	if (ret)
66 		return ret;
67 
68 	nv_wo32(obj, 0x00, nv_mclass(obj));
69 	nv_wo32(obj, 0x04, 0x00000000);
70 	nv_wo32(obj, 0x08, 0x00000000);
71 	nv_wo32(obj, 0x0c, 0x00000000);
72 	return 0;
73 }
74 
75 static struct nvkm_ofuncs
76 nv50_gr_ofuncs = {
77 	.ctor = nv50_gr_object_ctor,
78 	.dtor = _nvkm_gpuobj_dtor,
79 	.init = _nvkm_gpuobj_init,
80 	.fini = _nvkm_gpuobj_fini,
81 	.rd32 = _nvkm_gpuobj_rd32,
82 	.wr32 = _nvkm_gpuobj_wr32,
83 };
84 
85 static struct nvkm_oclass
86 nv50_gr_sclass[] = {
87 	{ 0x0030, &nv50_gr_ofuncs },
88 	{ 0x502d, &nv50_gr_ofuncs },
89 	{ 0x5039, &nv50_gr_ofuncs },
90 	{ 0x5097, &nv50_gr_ofuncs },
91 	{ 0x50c0, &nv50_gr_ofuncs },
92 	{}
93 };
94 
95 static struct nvkm_oclass
96 g84_gr_sclass[] = {
97 	{ 0x0030, &nv50_gr_ofuncs },
98 	{ 0x502d, &nv50_gr_ofuncs },
99 	{ 0x5039, &nv50_gr_ofuncs },
100 	{ 0x50c0, &nv50_gr_ofuncs },
101 	{ 0x8297, &nv50_gr_ofuncs },
102 	{}
103 };
104 
105 static struct nvkm_oclass
106 gt200_gr_sclass[] = {
107 	{ 0x0030, &nv50_gr_ofuncs },
108 	{ 0x502d, &nv50_gr_ofuncs },
109 	{ 0x5039, &nv50_gr_ofuncs },
110 	{ 0x50c0, &nv50_gr_ofuncs },
111 	{ 0x8397, &nv50_gr_ofuncs },
112 	{}
113 };
114 
115 static struct nvkm_oclass
116 gt215_gr_sclass[] = {
117 	{ 0x0030, &nv50_gr_ofuncs },
118 	{ 0x502d, &nv50_gr_ofuncs },
119 	{ 0x5039, &nv50_gr_ofuncs },
120 	{ 0x50c0, &nv50_gr_ofuncs },
121 	{ 0x8597, &nv50_gr_ofuncs },
122 	{ 0x85c0, &nv50_gr_ofuncs },
123 	{}
124 };
125 
126 static struct nvkm_oclass
127 mcp89_gr_sclass[] = {
128 	{ 0x0030, &nv50_gr_ofuncs },
129 	{ 0x502d, &nv50_gr_ofuncs },
130 	{ 0x5039, &nv50_gr_ofuncs },
131 	{ 0x50c0, &nv50_gr_ofuncs },
132 	{ 0x85c0, &nv50_gr_ofuncs },
133 	{ 0x8697, &nv50_gr_ofuncs },
134 	{}
135 };
136 
137 /*******************************************************************************
138  * PGRAPH context
139  ******************************************************************************/
140 
141 static int
142 nv50_gr_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
143 		     struct nvkm_oclass *oclass, void *data, u32 size,
144 		     struct nvkm_object **pobject)
145 {
146 	struct nv50_gr_priv *priv = (void *)engine;
147 	struct nv50_gr_chan *chan;
148 	int ret;
149 
150 	ret = nvkm_gr_context_create(parent, engine, oclass, NULL, priv->size,
151 				     0, NVOBJ_FLAG_ZERO_ALLOC, &chan);
152 	*pobject = nv_object(chan);
153 	if (ret)
154 		return ret;
155 
156 	nv50_grctx_fill(nv_device(priv), nv_gpuobj(chan));
157 	return 0;
158 }
159 
160 static struct nvkm_oclass
161 nv50_gr_cclass = {
162 	.handle = NV_ENGCTX(GR, 0x50),
163 	.ofuncs = &(struct nvkm_ofuncs) {
164 		.ctor = nv50_gr_context_ctor,
165 		.dtor = _nvkm_gr_context_dtor,
166 		.init = _nvkm_gr_context_init,
167 		.fini = _nvkm_gr_context_fini,
168 		.rd32 = _nvkm_gr_context_rd32,
169 		.wr32 = _nvkm_gr_context_wr32,
170 	},
171 };
172 
173 /*******************************************************************************
174  * PGRAPH engine/subdev functions
175  ******************************************************************************/
176 
177 static const struct nvkm_bitfield nv50_pgr_status[] = {
178 	{ 0x00000001, "BUSY" }, /* set when any bit is set */
179 	{ 0x00000002, "DISPATCH" },
180 	{ 0x00000004, "UNK2" },
181 	{ 0x00000008, "UNK3" },
182 	{ 0x00000010, "UNK4" },
183 	{ 0x00000020, "UNK5" },
184 	{ 0x00000040, "M2MF" },
185 	{ 0x00000080, "UNK7" },
186 	{ 0x00000100, "CTXPROG" },
187 	{ 0x00000200, "VFETCH" },
188 	{ 0x00000400, "CCACHE_PREGEOM" },
189 	{ 0x00000800, "STRMOUT_VATTR_POSTGEOM" },
190 	{ 0x00001000, "VCLIP" },
191 	{ 0x00002000, "RATTR_APLANE" },
192 	{ 0x00004000, "TRAST" },
193 	{ 0x00008000, "CLIPID" },
194 	{ 0x00010000, "ZCULL" },
195 	{ 0x00020000, "ENG2D" },
196 	{ 0x00040000, "RMASK" },
197 	{ 0x00080000, "TPC_RAST" },
198 	{ 0x00100000, "TPC_PROP" },
199 	{ 0x00200000, "TPC_TEX" },
200 	{ 0x00400000, "TPC_GEOM" },
201 	{ 0x00800000, "TPC_MP" },
202 	{ 0x01000000, "ROP" },
203 	{}
204 };
205 
206 static const char *const nv50_pgr_vstatus_0[] = {
207 	"VFETCH", "CCACHE", "PREGEOM", "POSTGEOM", "VATTR", "STRMOUT", "VCLIP",
208 	NULL
209 };
210 
211 static const char *const nv50_pgr_vstatus_1[] = {
212 	"TPC_RAST", "TPC_PROP", "TPC_TEX", "TPC_GEOM", "TPC_MP", NULL
213 };
214 
215 static const char *const nv50_pgr_vstatus_2[] = {
216 	"RATTR", "APLANE", "TRAST", "CLIPID", "ZCULL", "ENG2D", "RMASK",
217 	"ROP", NULL
218 };
219 
220 static void
221 nvkm_pgr_vstatus_print(struct nv50_gr_priv *priv, int r,
222 		       const char *const units[], u32 status)
223 {
224 	int i;
225 
226 	nv_error(priv, "PGRAPH_VSTATUS%d: 0x%08x", r, status);
227 
228 	for (i = 0; units[i] && status; i++) {
229 		if ((status & 7) == 1)
230 			pr_cont(" %s", units[i]);
231 		status >>= 3;
232 	}
233 	if (status)
234 		pr_cont(" (invalid: 0x%x)", status);
235 	pr_cont("\n");
236 }
237 
238 static int
239 g84_gr_tlb_flush(struct nvkm_engine *engine)
240 {
241 	struct nvkm_timer *ptimer = nvkm_timer(engine);
242 	struct nv50_gr_priv *priv = (void *)engine;
243 	bool idle, timeout = false;
244 	unsigned long flags;
245 	u64 start;
246 	u32 tmp;
247 
248 	spin_lock_irqsave(&priv->lock, flags);
249 	nv_mask(priv, 0x400500, 0x00000001, 0x00000000);
250 
251 	start = ptimer->read(ptimer);
252 	do {
253 		idle = true;
254 
255 		for (tmp = nv_rd32(priv, 0x400380); tmp && idle; tmp >>= 3) {
256 			if ((tmp & 7) == 1)
257 				idle = false;
258 		}
259 
260 		for (tmp = nv_rd32(priv, 0x400384); tmp && idle; tmp >>= 3) {
261 			if ((tmp & 7) == 1)
262 				idle = false;
263 		}
264 
265 		for (tmp = nv_rd32(priv, 0x400388); tmp && idle; tmp >>= 3) {
266 			if ((tmp & 7) == 1)
267 				idle = false;
268 		}
269 	} while (!idle &&
270 		 !(timeout = ptimer->read(ptimer) - start > 2000000000));
271 
272 	if (timeout) {
273 		nv_error(priv, "PGRAPH TLB flush idle timeout fail\n");
274 
275 		tmp = nv_rd32(priv, 0x400700);
276 		nv_error(priv, "PGRAPH_STATUS  : 0x%08x", tmp);
277 		nvkm_bitfield_print(nv50_pgr_status, tmp);
278 		pr_cont("\n");
279 
280 		nvkm_pgr_vstatus_print(priv, 0, nv50_pgr_vstatus_0,
281 				       nv_rd32(priv, 0x400380));
282 		nvkm_pgr_vstatus_print(priv, 1, nv50_pgr_vstatus_1,
283 				       nv_rd32(priv, 0x400384));
284 		nvkm_pgr_vstatus_print(priv, 2, nv50_pgr_vstatus_2,
285 				       nv_rd32(priv, 0x400388));
286 	}
287 
288 
289 	nv_wr32(priv, 0x100c80, 0x00000001);
290 	if (!nv_wait(priv, 0x100c80, 0x00000001, 0x00000000))
291 		nv_error(priv, "vm flush timeout\n");
292 	nv_mask(priv, 0x400500, 0x00000001, 0x00000001);
293 	spin_unlock_irqrestore(&priv->lock, flags);
294 	return timeout ? -EBUSY : 0;
295 }
296 
297 static const struct nvkm_bitfield nv50_mp_exec_errors[] = {
298 	{ 0x01, "STACK_UNDERFLOW" },
299 	{ 0x02, "STACK_MISMATCH" },
300 	{ 0x04, "QUADON_ACTIVE" },
301 	{ 0x08, "TIMEOUT" },
302 	{ 0x10, "INVALID_OPCODE" },
303 	{ 0x20, "PM_OVERFLOW" },
304 	{ 0x40, "BREAKPOINT" },
305 	{}
306 };
307 
308 static const struct nvkm_bitfield nv50_mpc_traps[] = {
309 	{ 0x0000001, "LOCAL_LIMIT_READ" },
310 	{ 0x0000010, "LOCAL_LIMIT_WRITE" },
311 	{ 0x0000040, "STACK_LIMIT" },
312 	{ 0x0000100, "GLOBAL_LIMIT_READ" },
313 	{ 0x0001000, "GLOBAL_LIMIT_WRITE" },
314 	{ 0x0010000, "MP0" },
315 	{ 0x0020000, "MP1" },
316 	{ 0x0040000, "GLOBAL_LIMIT_RED" },
317 	{ 0x0400000, "GLOBAL_LIMIT_ATOM" },
318 	{ 0x4000000, "MP2" },
319 	{}
320 };
321 
322 static const struct nvkm_bitfield nv50_tex_traps[] = {
323 	{ 0x00000001, "" }, /* any bit set? */
324 	{ 0x00000002, "FAULT" },
325 	{ 0x00000004, "STORAGE_TYPE_MISMATCH" },
326 	{ 0x00000008, "LINEAR_MISMATCH" },
327 	{ 0x00000020, "WRONG_MEMTYPE" },
328 	{}
329 };
330 
331 static const struct nvkm_bitfield nv50_gr_trap_m2mf[] = {
332 	{ 0x00000001, "NOTIFY" },
333 	{ 0x00000002, "IN" },
334 	{ 0x00000004, "OUT" },
335 	{}
336 };
337 
338 static const struct nvkm_bitfield nv50_gr_trap_vfetch[] = {
339 	{ 0x00000001, "FAULT" },
340 	{}
341 };
342 
343 static const struct nvkm_bitfield nv50_gr_trap_strmout[] = {
344 	{ 0x00000001, "FAULT" },
345 	{}
346 };
347 
348 static const struct nvkm_bitfield nv50_gr_trap_ccache[] = {
349 	{ 0x00000001, "FAULT" },
350 	{}
351 };
352 
353 /* There must be a *lot* of these. Will take some time to gather them up. */
354 const struct nvkm_enum nv50_data_error_names[] = {
355 	{ 0x00000003, "INVALID_OPERATION", NULL },
356 	{ 0x00000004, "INVALID_VALUE", NULL },
357 	{ 0x00000005, "INVALID_ENUM", NULL },
358 	{ 0x00000008, "INVALID_OBJECT", NULL },
359 	{ 0x00000009, "READ_ONLY_OBJECT", NULL },
360 	{ 0x0000000a, "SUPERVISOR_OBJECT", NULL },
361 	{ 0x0000000b, "INVALID_ADDRESS_ALIGNMENT", NULL },
362 	{ 0x0000000c, "INVALID_BITFIELD", NULL },
363 	{ 0x0000000d, "BEGIN_END_ACTIVE", NULL },
364 	{ 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT", NULL },
365 	{ 0x0000000f, "VIEWPORT_ID_NEEDS_GP", NULL },
366 	{ 0x00000010, "RT_DOUBLE_BIND", NULL },
367 	{ 0x00000011, "RT_TYPES_MISMATCH", NULL },
368 	{ 0x00000012, "RT_LINEAR_WITH_ZETA", NULL },
369 	{ 0x00000015, "FP_TOO_FEW_REGS", NULL },
370 	{ 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH", NULL },
371 	{ 0x00000017, "RT_LINEAR_WITH_MSAA", NULL },
372 	{ 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT", NULL },
373 	{ 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT", NULL },
374 	{ 0x0000001a, "RT_INVALID_ALIGNMENT", NULL },
375 	{ 0x0000001b, "SAMPLER_OVER_LIMIT", NULL },
376 	{ 0x0000001c, "TEXTURE_OVER_LIMIT", NULL },
377 	{ 0x0000001e, "GP_TOO_MANY_OUTPUTS", NULL },
378 	{ 0x0000001f, "RT_BPP128_WITH_MS8", NULL },
379 	{ 0x00000021, "Z_OUT_OF_BOUNDS", NULL },
380 	{ 0x00000023, "XY_OUT_OF_BOUNDS", NULL },
381 	{ 0x00000024, "VP_ZERO_INPUTS", NULL },
382 	{ 0x00000027, "CP_MORE_PARAMS_THAN_SHARED", NULL },
383 	{ 0x00000028, "CP_NO_REG_SPACE_STRIPED", NULL },
384 	{ 0x00000029, "CP_NO_REG_SPACE_PACKED", NULL },
385 	{ 0x0000002a, "CP_NOT_ENOUGH_WARPS", NULL },
386 	{ 0x0000002b, "CP_BLOCK_SIZE_MISMATCH", NULL },
387 	{ 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS", NULL },
388 	{ 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS", NULL },
389 	{ 0x0000002e, "CP_NO_BLOCKDIM_LATCH", NULL },
390 	{ 0x00000031, "ENG2D_FORMAT_MISMATCH", NULL },
391 	{ 0x0000003f, "PRIMITIVE_ID_NEEDS_GP", NULL },
392 	{ 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT", NULL },
393 	{ 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT", NULL },
394 	{ 0x00000046, "LAYER_ID_NEEDS_GP", NULL },
395 	{ 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT", NULL },
396 	{ 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT", NULL },
397 	{}
398 };
399 
400 static const struct nvkm_bitfield nv50_gr_intr_name[] = {
401 	{ 0x00000001, "NOTIFY" },
402 	{ 0x00000002, "COMPUTE_QUERY" },
403 	{ 0x00000010, "ILLEGAL_MTHD" },
404 	{ 0x00000020, "ILLEGAL_CLASS" },
405 	{ 0x00000040, "DOUBLE_NOTIFY" },
406 	{ 0x00001000, "CONTEXT_SWITCH" },
407 	{ 0x00010000, "BUFFER_NOTIFY" },
408 	{ 0x00100000, "DATA_ERROR" },
409 	{ 0x00200000, "TRAP" },
410 	{ 0x01000000, "SINGLE_STEP" },
411 	{}
412 };
413 
414 static const struct nvkm_bitfield nv50_gr_trap_prop[] = {
415 	{ 0x00000004, "SURF_WIDTH_OVERRUN" },
416 	{ 0x00000008, "SURF_HEIGHT_OVERRUN" },
417 	{ 0x00000010, "DST2D_FAULT" },
418 	{ 0x00000020, "ZETA_FAULT" },
419 	{ 0x00000040, "RT_FAULT" },
420 	{ 0x00000080, "CUDA_FAULT" },
421 	{ 0x00000100, "DST2D_STORAGE_TYPE_MISMATCH" },
422 	{ 0x00000200, "ZETA_STORAGE_TYPE_MISMATCH" },
423 	{ 0x00000400, "RT_STORAGE_TYPE_MISMATCH" },
424 	{ 0x00000800, "DST2D_LINEAR_MISMATCH" },
425 	{ 0x00001000, "RT_LINEAR_MISMATCH" },
426 	{}
427 };
428 
429 static void
430 nv50_priv_prop_trap(struct nv50_gr_priv *priv,
431 		    u32 ustatus_addr, u32 ustatus, u32 tp)
432 {
433 	u32 e0c = nv_rd32(priv, ustatus_addr + 0x04);
434 	u32 e10 = nv_rd32(priv, ustatus_addr + 0x08);
435 	u32 e14 = nv_rd32(priv, ustatus_addr + 0x0c);
436 	u32 e18 = nv_rd32(priv, ustatus_addr + 0x10);
437 	u32 e1c = nv_rd32(priv, ustatus_addr + 0x14);
438 	u32 e20 = nv_rd32(priv, ustatus_addr + 0x18);
439 	u32 e24 = nv_rd32(priv, ustatus_addr + 0x1c);
440 
441 	/* CUDA memory: l[], g[] or stack. */
442 	if (ustatus & 0x00000080) {
443 		if (e18 & 0x80000000) {
444 			/* g[] read fault? */
445 			nv_error(priv, "TRAP_PROP - TP %d - CUDA_FAULT - Global read fault at address %02x%08x\n",
446 					 tp, e14, e10 | ((e18 >> 24) & 0x1f));
447 			e18 &= ~0x1f000000;
448 		} else if (e18 & 0xc) {
449 			/* g[] write fault? */
450 			nv_error(priv, "TRAP_PROP - TP %d - CUDA_FAULT - Global write fault at address %02x%08x\n",
451 				 tp, e14, e10 | ((e18 >> 7) & 0x1f));
452 			e18 &= ~0x00000f80;
453 		} else {
454 			nv_error(priv, "TRAP_PROP - TP %d - Unknown CUDA fault at address %02x%08x\n",
455 				 tp, e14, e10);
456 		}
457 		ustatus &= ~0x00000080;
458 	}
459 	if (ustatus) {
460 		nv_error(priv, "TRAP_PROP - TP %d -", tp);
461 		nvkm_bitfield_print(nv50_gr_trap_prop, ustatus);
462 		pr_cont(" - Address %02x%08x\n", e14, e10);
463 	}
464 	nv_error(priv, "TRAP_PROP - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
465 		 tp, e0c, e18, e1c, e20, e24);
466 }
467 
468 static void
469 nv50_priv_mp_trap(struct nv50_gr_priv *priv, int tpid, int display)
470 {
471 	u32 units = nv_rd32(priv, 0x1540);
472 	u32 addr, mp10, status, pc, oplow, ophigh;
473 	int i;
474 	int mps = 0;
475 	for (i = 0; i < 4; i++) {
476 		if (!(units & 1 << (i+24)))
477 			continue;
478 		if (nv_device(priv)->chipset < 0xa0)
479 			addr = 0x408200 + (tpid << 12) + (i << 7);
480 		else
481 			addr = 0x408100 + (tpid << 11) + (i << 7);
482 		mp10 = nv_rd32(priv, addr + 0x10);
483 		status = nv_rd32(priv, addr + 0x14);
484 		if (!status)
485 			continue;
486 		if (display) {
487 			nv_rd32(priv, addr + 0x20);
488 			pc = nv_rd32(priv, addr + 0x24);
489 			oplow = nv_rd32(priv, addr + 0x70);
490 			ophigh = nv_rd32(priv, addr + 0x74);
491 			nv_error(priv, "TRAP_MP_EXEC - "
492 					"TP %d MP %d:", tpid, i);
493 			nvkm_bitfield_print(nv50_mp_exec_errors, status);
494 			pr_cont(" at %06x warp %d, opcode %08x %08x\n",
495 					pc&0xffffff, pc >> 24,
496 					oplow, ophigh);
497 		}
498 		nv_wr32(priv, addr + 0x10, mp10);
499 		nv_wr32(priv, addr + 0x14, 0);
500 		mps++;
501 	}
502 	if (!mps && display)
503 		nv_error(priv, "TRAP_MP_EXEC - TP %d: "
504 				"No MPs claiming errors?\n", tpid);
505 }
506 
507 static void
508 nv50_priv_tp_trap(struct nv50_gr_priv *priv, int type, u32 ustatus_old,
509 		  u32 ustatus_new, int display, const char *name)
510 {
511 	int tps = 0;
512 	u32 units = nv_rd32(priv, 0x1540);
513 	int i, r;
514 	u32 ustatus_addr, ustatus;
515 	for (i = 0; i < 16; i++) {
516 		if (!(units & (1 << i)))
517 			continue;
518 		if (nv_device(priv)->chipset < 0xa0)
519 			ustatus_addr = ustatus_old + (i << 12);
520 		else
521 			ustatus_addr = ustatus_new + (i << 11);
522 		ustatus = nv_rd32(priv, ustatus_addr) & 0x7fffffff;
523 		if (!ustatus)
524 			continue;
525 		tps++;
526 		switch (type) {
527 		case 6: /* texture error... unknown for now */
528 			if (display) {
529 				nv_error(priv, "magic set %d:\n", i);
530 				for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
531 					nv_error(priv, "\t0x%08x: 0x%08x\n", r,
532 						nv_rd32(priv, r));
533 				if (ustatus) {
534 					nv_error(priv, "%s - TP%d:", name, i);
535 					nvkm_bitfield_print(nv50_tex_traps,
536 							       ustatus);
537 					pr_cont("\n");
538 					ustatus = 0;
539 				}
540 			}
541 			break;
542 		case 7: /* MP error */
543 			if (ustatus & 0x04030000) {
544 				nv50_priv_mp_trap(priv, i, display);
545 				ustatus &= ~0x04030000;
546 			}
547 			if (ustatus && display) {
548 				nv_error(priv, "%s - TP%d:", name, i);
549 				nvkm_bitfield_print(nv50_mpc_traps, ustatus);
550 				pr_cont("\n");
551 				ustatus = 0;
552 			}
553 			break;
554 		case 8: /* PROP error */
555 			if (display)
556 				nv50_priv_prop_trap(
557 						priv, ustatus_addr, ustatus, i);
558 			ustatus = 0;
559 			break;
560 		}
561 		if (ustatus) {
562 			if (display)
563 				nv_error(priv, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus);
564 		}
565 		nv_wr32(priv, ustatus_addr, 0xc0000000);
566 	}
567 
568 	if (!tps && display)
569 		nv_warn(priv, "%s - No TPs claiming errors?\n", name);
570 }
571 
572 static int
573 nv50_gr_trap_handler(struct nv50_gr_priv *priv, u32 display,
574 		     int chid, u64 inst, struct nvkm_object *engctx)
575 {
576 	u32 status = nv_rd32(priv, 0x400108);
577 	u32 ustatus;
578 
579 	if (!status && display) {
580 		nv_error(priv, "TRAP: no units reporting traps?\n");
581 		return 1;
582 	}
583 
584 	/* DISPATCH: Relays commands to other units and handles NOTIFY,
585 	 * COND, QUERY. If you get a trap from it, the command is still stuck
586 	 * in DISPATCH and you need to do something about it. */
587 	if (status & 0x001) {
588 		ustatus = nv_rd32(priv, 0x400804) & 0x7fffffff;
589 		if (!ustatus && display) {
590 			nv_error(priv, "TRAP_DISPATCH - no ustatus?\n");
591 		}
592 
593 		nv_wr32(priv, 0x400500, 0x00000000);
594 
595 		/* Known to be triggered by screwed up NOTIFY and COND... */
596 		if (ustatus & 0x00000001) {
597 			u32 addr = nv_rd32(priv, 0x400808);
598 			u32 subc = (addr & 0x00070000) >> 16;
599 			u32 mthd = (addr & 0x00001ffc);
600 			u32 datal = nv_rd32(priv, 0x40080c);
601 			u32 datah = nv_rd32(priv, 0x400810);
602 			u32 class = nv_rd32(priv, 0x400814);
603 			u32 r848 = nv_rd32(priv, 0x400848);
604 
605 			nv_error(priv, "TRAP DISPATCH_FAULT\n");
606 			if (display && (addr & 0x80000000)) {
607 				nv_error(priv,
608 					 "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 					 chid, inst,
610 					 nvkm_client_name(engctx), subc,
611 					 class, mthd, datah, datal, addr, r848);
612 			} else
613 			if (display) {
614 				nv_error(priv, "no stuck command?\n");
615 			}
616 
617 			nv_wr32(priv, 0x400808, 0);
618 			nv_wr32(priv, 0x4008e8, nv_rd32(priv, 0x4008e8) & 3);
619 			nv_wr32(priv, 0x400848, 0);
620 			ustatus &= ~0x00000001;
621 		}
622 
623 		if (ustatus & 0x00000002) {
624 			u32 addr = nv_rd32(priv, 0x40084c);
625 			u32 subc = (addr & 0x00070000) >> 16;
626 			u32 mthd = (addr & 0x00001ffc);
627 			u32 data = nv_rd32(priv, 0x40085c);
628 			u32 class = nv_rd32(priv, 0x400814);
629 
630 			nv_error(priv, "TRAP DISPATCH_QUERY\n");
631 			if (display && (addr & 0x80000000)) {
632 				nv_error(priv,
633 					 "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x 40084c 0x%08x\n",
634 					 chid, inst,
635 					 nvkm_client_name(engctx), subc,
636 					 class, mthd, data, addr);
637 			} else
638 			if (display) {
639 				nv_error(priv, "no stuck command?\n");
640 			}
641 
642 			nv_wr32(priv, 0x40084c, 0);
643 			ustatus &= ~0x00000002;
644 		}
645 
646 		if (ustatus && display) {
647 			nv_error(priv, "TRAP_DISPATCH (unknown "
648 				      "0x%08x)\n", ustatus);
649 		}
650 
651 		nv_wr32(priv, 0x400804, 0xc0000000);
652 		nv_wr32(priv, 0x400108, 0x001);
653 		status &= ~0x001;
654 		if (!status)
655 			return 0;
656 	}
657 
658 	/* M2MF: Memory to memory copy engine. */
659 	if (status & 0x002) {
660 		u32 ustatus = nv_rd32(priv, 0x406800) & 0x7fffffff;
661 		if (display) {
662 			nv_error(priv, "TRAP_M2MF");
663 			nvkm_bitfield_print(nv50_gr_trap_m2mf, ustatus);
664 			pr_cont("\n");
665 			nv_error(priv, "TRAP_M2MF %08x %08x %08x %08x\n",
666 				nv_rd32(priv, 0x406804), nv_rd32(priv, 0x406808),
667 				nv_rd32(priv, 0x40680c), nv_rd32(priv, 0x406810));
668 
669 		}
670 
671 		/* No sane way found yet -- just reset the bugger. */
672 		nv_wr32(priv, 0x400040, 2);
673 		nv_wr32(priv, 0x400040, 0);
674 		nv_wr32(priv, 0x406800, 0xc0000000);
675 		nv_wr32(priv, 0x400108, 0x002);
676 		status &= ~0x002;
677 	}
678 
679 	/* VFETCH: Fetches data from vertex buffers. */
680 	if (status & 0x004) {
681 		u32 ustatus = nv_rd32(priv, 0x400c04) & 0x7fffffff;
682 		if (display) {
683 			nv_error(priv, "TRAP_VFETCH");
684 			nvkm_bitfield_print(nv50_gr_trap_vfetch, ustatus);
685 			pr_cont("\n");
686 			nv_error(priv, "TRAP_VFETCH %08x %08x %08x %08x\n",
687 				nv_rd32(priv, 0x400c00), nv_rd32(priv, 0x400c08),
688 				nv_rd32(priv, 0x400c0c), nv_rd32(priv, 0x400c10));
689 		}
690 
691 		nv_wr32(priv, 0x400c04, 0xc0000000);
692 		nv_wr32(priv, 0x400108, 0x004);
693 		status &= ~0x004;
694 	}
695 
696 	/* STRMOUT: DirectX streamout / OpenGL transform feedback. */
697 	if (status & 0x008) {
698 		ustatus = nv_rd32(priv, 0x401800) & 0x7fffffff;
699 		if (display) {
700 			nv_error(priv, "TRAP_STRMOUT");
701 			nvkm_bitfield_print(nv50_gr_trap_strmout, ustatus);
702 			pr_cont("\n");
703 			nv_error(priv, "TRAP_STRMOUT %08x %08x %08x %08x\n",
704 				nv_rd32(priv, 0x401804), nv_rd32(priv, 0x401808),
705 				nv_rd32(priv, 0x40180c), nv_rd32(priv, 0x401810));
706 
707 		}
708 
709 		/* No sane way found yet -- just reset the bugger. */
710 		nv_wr32(priv, 0x400040, 0x80);
711 		nv_wr32(priv, 0x400040, 0);
712 		nv_wr32(priv, 0x401800, 0xc0000000);
713 		nv_wr32(priv, 0x400108, 0x008);
714 		status &= ~0x008;
715 	}
716 
717 	/* CCACHE: Handles code and c[] caches and fills them. */
718 	if (status & 0x010) {
719 		ustatus = nv_rd32(priv, 0x405018) & 0x7fffffff;
720 		if (display) {
721 			nv_error(priv, "TRAP_CCACHE");
722 			nvkm_bitfield_print(nv50_gr_trap_ccache, ustatus);
723 			pr_cont("\n");
724 			nv_error(priv, "TRAP_CCACHE %08x %08x %08x %08x"
725 				     " %08x %08x %08x\n",
726 				nv_rd32(priv, 0x405000), nv_rd32(priv, 0x405004),
727 				nv_rd32(priv, 0x405008), nv_rd32(priv, 0x40500c),
728 				nv_rd32(priv, 0x405010), nv_rd32(priv, 0x405014),
729 				nv_rd32(priv, 0x40501c));
730 
731 		}
732 
733 		nv_wr32(priv, 0x405018, 0xc0000000);
734 		nv_wr32(priv, 0x400108, 0x010);
735 		status &= ~0x010;
736 	}
737 
738 	/* Unknown, not seen yet... 0x402000 is the only trap status reg
739 	 * remaining, so try to handle it anyway. Perhaps related to that
740 	 * unknown DMA slot on tesla? */
741 	if (status & 0x20) {
742 		ustatus = nv_rd32(priv, 0x402000) & 0x7fffffff;
743 		if (display)
744 			nv_error(priv, "TRAP_UNKC04 0x%08x\n", ustatus);
745 		nv_wr32(priv, 0x402000, 0xc0000000);
746 		/* no status modifiction on purpose */
747 	}
748 
749 	/* TEXTURE: CUDA texturing units */
750 	if (status & 0x040) {
751 		nv50_priv_tp_trap(priv, 6, 0x408900, 0x408600, display,
752 				    "TRAP_TEXTURE");
753 		nv_wr32(priv, 0x400108, 0x040);
754 		status &= ~0x040;
755 	}
756 
757 	/* MP: CUDA execution engines. */
758 	if (status & 0x080) {
759 		nv50_priv_tp_trap(priv, 7, 0x408314, 0x40831c, display,
760 				    "TRAP_MP");
761 		nv_wr32(priv, 0x400108, 0x080);
762 		status &= ~0x080;
763 	}
764 
765 	/* PROP:  Handles TP-initiated uncached memory accesses:
766 	 * l[], g[], stack, 2d surfaces, render targets. */
767 	if (status & 0x100) {
768 		nv50_priv_tp_trap(priv, 8, 0x408e08, 0x408708, display,
769 				    "TRAP_PROP");
770 		nv_wr32(priv, 0x400108, 0x100);
771 		status &= ~0x100;
772 	}
773 
774 	if (status) {
775 		if (display)
776 			nv_error(priv, "TRAP: unknown 0x%08x\n", status);
777 		nv_wr32(priv, 0x400108, status);
778 	}
779 
780 	return 1;
781 }
782 
783 static void
784 nv50_gr_intr(struct nvkm_subdev *subdev)
785 {
786 	struct nvkm_fifo *pfifo = nvkm_fifo(subdev);
787 	struct nvkm_engine *engine = nv_engine(subdev);
788 	struct nvkm_object *engctx;
789 	struct nvkm_handle *handle = NULL;
790 	struct nv50_gr_priv *priv = (void *)subdev;
791 	u32 stat = nv_rd32(priv, 0x400100);
792 	u32 inst = nv_rd32(priv, 0x40032c) & 0x0fffffff;
793 	u32 addr = nv_rd32(priv, 0x400704);
794 	u32 subc = (addr & 0x00070000) >> 16;
795 	u32 mthd = (addr & 0x00001ffc);
796 	u32 data = nv_rd32(priv, 0x400708);
797 	u32 class = nv_rd32(priv, 0x400814);
798 	u32 show = stat, show_bitfield = stat;
799 	int chid;
800 
801 	engctx = nvkm_engctx_get(engine, inst);
802 	chid   = pfifo->chid(pfifo, engctx);
803 
804 	if (stat & 0x00000010) {
805 		handle = nvkm_handle_get_class(engctx, class);
806 		if (handle && !nv_call(handle->object, mthd, data))
807 			show &= ~0x00000010;
808 		nvkm_handle_put(handle);
809 	}
810 
811 	if (show & 0x00100000) {
812 		u32 ecode = nv_rd32(priv, 0x400110);
813 		nv_error(priv, "DATA_ERROR ");
814 		nvkm_enum_print(nv50_data_error_names, ecode);
815 		pr_cont("\n");
816 		show_bitfield &= ~0x00100000;
817 	}
818 
819 	if (stat & 0x00200000) {
820 		if (!nv50_gr_trap_handler(priv, show, chid, (u64)inst << 12,
821 					  engctx))
822 			show &= ~0x00200000;
823 		show_bitfield &= ~0x00200000;
824 	}
825 
826 	nv_wr32(priv, 0x400100, stat);
827 	nv_wr32(priv, 0x400500, 0x00010001);
828 
829 	if (show) {
830 		show &= show_bitfield;
831 		if (show) {
832 			nv_error(priv, "%s", "");
833 			nvkm_bitfield_print(nv50_gr_intr_name, show);
834 			pr_cont("\n");
835 		}
836 		nv_error(priv,
837 			 "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x\n",
838 			 chid, (u64)inst << 12, nvkm_client_name(engctx),
839 			 subc, class, mthd, data);
840 	}
841 
842 	if (nv_rd32(priv, 0x400824) & (1 << 31))
843 		nv_wr32(priv, 0x400824, nv_rd32(priv, 0x400824) & ~(1 << 31));
844 
845 	nvkm_engctx_put(engctx);
846 }
847 
848 static int
849 nv50_gr_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
850 	     struct nvkm_oclass *oclass, void *data, u32 size,
851 	     struct nvkm_object **pobject)
852 {
853 	struct nv50_gr_priv *priv;
854 	int ret;
855 
856 	ret = nvkm_gr_create(parent, engine, oclass, true, &priv);
857 	*pobject = nv_object(priv);
858 	if (ret)
859 		return ret;
860 
861 	nv_subdev(priv)->unit = 0x00201000;
862 	nv_subdev(priv)->intr = nv50_gr_intr;
863 	nv_engine(priv)->cclass = &nv50_gr_cclass;
864 
865 	priv->base.units = nv50_gr_units;
866 
867 	switch (nv_device(priv)->chipset) {
868 	case 0x50:
869 		nv_engine(priv)->sclass = nv50_gr_sclass;
870 		break;
871 	case 0x84:
872 	case 0x86:
873 	case 0x92:
874 	case 0x94:
875 	case 0x96:
876 	case 0x98:
877 		nv_engine(priv)->sclass = g84_gr_sclass;
878 		break;
879 	case 0xa0:
880 	case 0xaa:
881 	case 0xac:
882 		nv_engine(priv)->sclass = gt200_gr_sclass;
883 		break;
884 	case 0xa3:
885 	case 0xa5:
886 	case 0xa8:
887 		nv_engine(priv)->sclass = gt215_gr_sclass;
888 		break;
889 	case 0xaf:
890 		nv_engine(priv)->sclass = mcp89_gr_sclass;
891 		break;
892 
893 	}
894 
895 	/* unfortunate hw bug workaround... */
896 	if (nv_device(priv)->chipset != 0x50 &&
897 	    nv_device(priv)->chipset != 0xac)
898 		nv_engine(priv)->tlb_flush = g84_gr_tlb_flush;
899 
900 	spin_lock_init(&priv->lock);
901 	return 0;
902 }
903 
904 static int
905 nv50_gr_init(struct nvkm_object *object)
906 {
907 	struct nv50_gr_priv *priv = (void *)object;
908 	int ret, units, i;
909 
910 	ret = nvkm_gr_init(&priv->base);
911 	if (ret)
912 		return ret;
913 
914 	/* NV_PGRAPH_DEBUG_3_HW_CTX_SWITCH_ENABLED */
915 	nv_wr32(priv, 0x40008c, 0x00000004);
916 
917 	/* reset/enable traps and interrupts */
918 	nv_wr32(priv, 0x400804, 0xc0000000);
919 	nv_wr32(priv, 0x406800, 0xc0000000);
920 	nv_wr32(priv, 0x400c04, 0xc0000000);
921 	nv_wr32(priv, 0x401800, 0xc0000000);
922 	nv_wr32(priv, 0x405018, 0xc0000000);
923 	nv_wr32(priv, 0x402000, 0xc0000000);
924 
925 	units = nv_rd32(priv, 0x001540);
926 	for (i = 0; i < 16; i++) {
927 		if (!(units & (1 << i)))
928 			continue;
929 
930 		if (nv_device(priv)->chipset < 0xa0) {
931 			nv_wr32(priv, 0x408900 + (i << 12), 0xc0000000);
932 			nv_wr32(priv, 0x408e08 + (i << 12), 0xc0000000);
933 			nv_wr32(priv, 0x408314 + (i << 12), 0xc0000000);
934 		} else {
935 			nv_wr32(priv, 0x408600 + (i << 11), 0xc0000000);
936 			nv_wr32(priv, 0x408708 + (i << 11), 0xc0000000);
937 			nv_wr32(priv, 0x40831c + (i << 11), 0xc0000000);
938 		}
939 	}
940 
941 	nv_wr32(priv, 0x400108, 0xffffffff);
942 	nv_wr32(priv, 0x400138, 0xffffffff);
943 	nv_wr32(priv, 0x400100, 0xffffffff);
944 	nv_wr32(priv, 0x40013c, 0xffffffff);
945 	nv_wr32(priv, 0x400500, 0x00010001);
946 
947 	/* upload context program, initialise ctxctl defaults */
948 	ret = nv50_grctx_init(nv_device(priv), &priv->size);
949 	if (ret)
950 		return ret;
951 
952 	nv_wr32(priv, 0x400824, 0x00000000);
953 	nv_wr32(priv, 0x400828, 0x00000000);
954 	nv_wr32(priv, 0x40082c, 0x00000000);
955 	nv_wr32(priv, 0x400830, 0x00000000);
956 	nv_wr32(priv, 0x40032c, 0x00000000);
957 	nv_wr32(priv, 0x400330, 0x00000000);
958 
959 	/* some unknown zcull magic */
960 	switch (nv_device(priv)->chipset & 0xf0) {
961 	case 0x50:
962 	case 0x80:
963 	case 0x90:
964 		nv_wr32(priv, 0x402ca8, 0x00000800);
965 		break;
966 	case 0xa0:
967 	default:
968 		if (nv_device(priv)->chipset == 0xa0 ||
969 		    nv_device(priv)->chipset == 0xaa ||
970 		    nv_device(priv)->chipset == 0xac) {
971 			nv_wr32(priv, 0x402ca8, 0x00000802);
972 		} else {
973 			nv_wr32(priv, 0x402cc0, 0x00000000);
974 			nv_wr32(priv, 0x402ca8, 0x00000002);
975 		}
976 
977 		break;
978 	}
979 
980 	/* zero out zcull regions */
981 	for (i = 0; i < 8; i++) {
982 		nv_wr32(priv, 0x402c20 + (i * 0x10), 0x00000000);
983 		nv_wr32(priv, 0x402c24 + (i * 0x10), 0x00000000);
984 		nv_wr32(priv, 0x402c28 + (i * 0x10), 0x00000000);
985 		nv_wr32(priv, 0x402c2c + (i * 0x10), 0x00000000);
986 	}
987 	return 0;
988 }
989 
990 struct nvkm_oclass
991 nv50_gr_oclass = {
992 	.handle = NV_ENGINE(GR, 0x50),
993 	.ofuncs = &(struct nvkm_ofuncs) {
994 		.ctor = nv50_gr_ctor,
995 		.dtor = _nvkm_gr_dtor,
996 		.init = nv50_gr_init,
997 		.fini = _nvkm_gr_fini,
998 	},
999 };
1000