Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-btrfs-devel.git] / drivers / gpu / drm / nouveau / nv50_graph.c
blobd43c46caa76e19156c45a54cbb55850b2390fa9c
1 /*
2 * Copyright (C) 2007 Ben Skeggs.
3 * All Rights Reserved.
5 * Permission is hereby granted, free of charge, to any person obtaining
6 * a copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sublicense, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * The above copyright notice and this permission notice (including the
14 * next paragraph) shall be included in all copies or substantial
15 * portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
18 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
20 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
21 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
22 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
23 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 #include "drmP.h"
28 #include "drm.h"
29 #include "nouveau_drv.h"
30 #include "nouveau_ramht.h"
31 #include "nouveau_grctx.h"
32 #include "nouveau_dma.h"
33 #include "nouveau_vm.h"
34 #include "nv50_evo.h"
36 struct nv50_graph_engine {
37 struct nouveau_exec_engine base;
38 u32 ctxprog[512];
39 u32 ctxprog_size;
40 u32 grctx_size;
43 static void
44 nv50_graph_fifo_access(struct drm_device *dev, bool enabled)
46 const uint32_t mask = 0x00010001;
48 if (enabled)
49 nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) | mask);
50 else
51 nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) & ~mask);
54 static struct nouveau_channel *
55 nv50_graph_channel(struct drm_device *dev)
57 struct drm_nouveau_private *dev_priv = dev->dev_private;
58 uint32_t inst;
59 int i;
61 /* Be sure we're not in the middle of a context switch or bad things
62 * will happen, such as unloading the wrong pgraph context.
64 if (!nv_wait(dev, 0x400300, 0x00000001, 0x00000000))
65 NV_ERROR(dev, "Ctxprog is still running\n");
67 inst = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
68 if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
69 return NULL;
70 inst = (inst & NV50_PGRAPH_CTXCTL_CUR_INSTANCE) << 12;
72 for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
73 struct nouveau_channel *chan = dev_priv->channels.ptr[i];
75 if (chan && chan->ramin && chan->ramin->vinst == inst)
76 return chan;
79 return NULL;
82 static int
83 nv50_graph_do_load_context(struct drm_device *dev, uint32_t inst)
85 uint32_t fifo = nv_rd32(dev, 0x400500);
87 nv_wr32(dev, 0x400500, fifo & ~1);
88 nv_wr32(dev, 0x400784, inst);
89 nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x40);
90 nv_wr32(dev, 0x400320, nv_rd32(dev, 0x400320) | 0x11);
91 nv_wr32(dev, 0x400040, 0xffffffff);
92 (void)nv_rd32(dev, 0x400040);
93 nv_wr32(dev, 0x400040, 0x00000000);
94 nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 1);
96 if (nouveau_wait_for_idle(dev))
97 nv_wr32(dev, 0x40032c, inst | (1<<31));
98 nv_wr32(dev, 0x400500, fifo);
100 return 0;
103 static int
104 nv50_graph_unload_context(struct drm_device *dev)
106 uint32_t inst;
108 inst = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
109 if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
110 return 0;
111 inst &= NV50_PGRAPH_CTXCTL_CUR_INSTANCE;
113 nouveau_wait_for_idle(dev);
114 nv_wr32(dev, 0x400784, inst);
115 nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x20);
116 nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 0x01);
117 nouveau_wait_for_idle(dev);
119 nv_wr32(dev, NV50_PGRAPH_CTXCTL_CUR, inst);
120 return 0;
123 static void
124 nv50_graph_init_reset(struct drm_device *dev)
126 uint32_t pmc_e = NV_PMC_ENABLE_PGRAPH | (1 << 21);
127 NV_DEBUG(dev, "\n");
129 nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) & ~pmc_e);
130 nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) | pmc_e);
133 static void
134 nv50_graph_init_intr(struct drm_device *dev)
136 NV_DEBUG(dev, "\n");
138 nv_wr32(dev, NV03_PGRAPH_INTR, 0xffffffff);
139 nv_wr32(dev, 0x400138, 0xffffffff);
140 nv_wr32(dev, NV40_PGRAPH_INTR_EN, 0xffffffff);
143 static void
144 nv50_graph_init_regs__nv(struct drm_device *dev)
146 struct drm_nouveau_private *dev_priv = dev->dev_private;
147 uint32_t units = nv_rd32(dev, 0x1540);
148 int i;
150 NV_DEBUG(dev, "\n");
152 nv_wr32(dev, 0x400804, 0xc0000000);
153 nv_wr32(dev, 0x406800, 0xc0000000);
154 nv_wr32(dev, 0x400c04, 0xc0000000);
155 nv_wr32(dev, 0x401800, 0xc0000000);
156 nv_wr32(dev, 0x405018, 0xc0000000);
157 nv_wr32(dev, 0x402000, 0xc0000000);
159 for (i = 0; i < 16; i++) {
160 if (units & 1 << i) {
161 if (dev_priv->chipset < 0xa0) {
162 nv_wr32(dev, 0x408900 + (i << 12), 0xc0000000);
163 nv_wr32(dev, 0x408e08 + (i << 12), 0xc0000000);
164 nv_wr32(dev, 0x408314 + (i << 12), 0xc0000000);
165 } else {
166 nv_wr32(dev, 0x408600 + (i << 11), 0xc0000000);
167 nv_wr32(dev, 0x408708 + (i << 11), 0xc0000000);
168 nv_wr32(dev, 0x40831c + (i << 11), 0xc0000000);
173 nv_wr32(dev, 0x400108, 0xffffffff);
175 nv_wr32(dev, 0x400824, 0x00004000);
176 nv_wr32(dev, 0x400500, 0x00010001);
179 static void
180 nv50_graph_init_zcull(struct drm_device *dev)
182 struct drm_nouveau_private *dev_priv = dev->dev_private;
183 int i;
185 NV_DEBUG(dev, "\n");
187 switch (dev_priv->chipset & 0xf0) {
188 case 0x50:
189 case 0x80:
190 case 0x90:
191 nv_wr32(dev, 0x402ca8, 0x00000800);
192 break;
193 case 0xa0:
194 default:
195 nv_wr32(dev, 0x402cc0, 0x00000000);
196 if (dev_priv->chipset == 0xa0 ||
197 dev_priv->chipset == 0xaa ||
198 dev_priv->chipset == 0xac) {
199 nv_wr32(dev, 0x402ca8, 0x00000802);
200 } else {
201 nv_wr32(dev, 0x402cc0, 0x00000000);
202 nv_wr32(dev, 0x402ca8, 0x00000002);
205 break;
208 /* zero out zcull regions */
209 for (i = 0; i < 8; i++) {
210 nv_wr32(dev, 0x402c20 + (i * 8), 0x00000000);
211 nv_wr32(dev, 0x402c24 + (i * 8), 0x00000000);
212 nv_wr32(dev, 0x402c28 + (i * 8), 0x00000000);
213 nv_wr32(dev, 0x402c2c + (i * 8), 0x00000000);
217 static int
218 nv50_graph_init_ctxctl(struct drm_device *dev)
220 struct nv50_graph_engine *pgraph = nv_engine(dev, NVOBJ_ENGINE_GR);
221 int i;
223 NV_DEBUG(dev, "\n");
225 nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
226 for (i = 0; i < pgraph->ctxprog_size; i++)
227 nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_DATA, pgraph->ctxprog[i]);
229 nv_wr32(dev, 0x40008c, 0x00000004); /* HW_CTX_SWITCH_ENABLED */
230 nv_wr32(dev, 0x400320, 4);
231 nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, 0);
232 nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, 0);
233 return 0;
236 static int
237 nv50_graph_init(struct drm_device *dev, int engine)
239 int ret;
241 NV_DEBUG(dev, "\n");
243 nv50_graph_init_reset(dev);
244 nv50_graph_init_regs__nv(dev);
245 nv50_graph_init_zcull(dev);
247 ret = nv50_graph_init_ctxctl(dev);
248 if (ret)
249 return ret;
251 nv50_graph_init_intr(dev);
252 return 0;
255 static int
256 nv50_graph_fini(struct drm_device *dev, int engine, bool suspend)
258 nv_mask(dev, 0x400500, 0x00010001, 0x00000000);
259 if (!nv_wait(dev, 0x400700, ~0, 0) && suspend) {
260 nv_mask(dev, 0x400500, 0x00010001, 0x00010001);
261 return -EBUSY;
263 nv50_graph_unload_context(dev);
264 nv_wr32(dev, 0x40013c, 0x00000000);
265 return 0;
268 static int
269 nv50_graph_context_new(struct nouveau_channel *chan, int engine)
271 struct drm_device *dev = chan->dev;
272 struct drm_nouveau_private *dev_priv = dev->dev_private;
273 struct nouveau_gpuobj *ramin = chan->ramin;
274 struct nouveau_gpuobj *grctx = NULL;
275 struct nv50_graph_engine *pgraph = nv_engine(dev, engine);
276 struct nouveau_grctx ctx = {};
277 int hdr, ret;
279 NV_DEBUG(dev, "ch%d\n", chan->id);
281 ret = nouveau_gpuobj_new(dev, NULL, pgraph->grctx_size, 0,
282 NVOBJ_FLAG_ZERO_ALLOC |
283 NVOBJ_FLAG_ZERO_FREE, &grctx);
284 if (ret)
285 return ret;
287 hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
288 nv_wo32(ramin, hdr + 0x00, 0x00190002);
289 nv_wo32(ramin, hdr + 0x04, grctx->vinst + grctx->size - 1);
290 nv_wo32(ramin, hdr + 0x08, grctx->vinst);
291 nv_wo32(ramin, hdr + 0x0c, 0);
292 nv_wo32(ramin, hdr + 0x10, 0);
293 nv_wo32(ramin, hdr + 0x14, 0x00010000);
295 ctx.dev = chan->dev;
296 ctx.mode = NOUVEAU_GRCTX_VALS;
297 ctx.data = grctx;
298 nv50_grctx_init(&ctx);
300 nv_wo32(grctx, 0x00000, chan->ramin->vinst >> 12);
302 dev_priv->engine.instmem.flush(dev);
304 atomic_inc(&chan->vm->engref[NVOBJ_ENGINE_GR]);
305 chan->engctx[NVOBJ_ENGINE_GR] = grctx;
306 return 0;
309 static void
310 nv50_graph_context_del(struct nouveau_channel *chan, int engine)
312 struct nouveau_gpuobj *grctx = chan->engctx[engine];
313 struct drm_device *dev = chan->dev;
314 struct drm_nouveau_private *dev_priv = dev->dev_private;
315 struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
316 int i, hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
317 unsigned long flags;
319 NV_DEBUG(dev, "ch%d\n", chan->id);
321 if (!chan->ramin)
322 return;
324 spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
325 pfifo->reassign(dev, false);
326 nv50_graph_fifo_access(dev, false);
328 if (nv50_graph_channel(dev) == chan)
329 nv50_graph_unload_context(dev);
331 for (i = hdr; i < hdr + 24; i += 4)
332 nv_wo32(chan->ramin, i, 0);
333 dev_priv->engine.instmem.flush(dev);
335 nv50_graph_fifo_access(dev, true);
336 pfifo->reassign(dev, true);
337 spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
339 nouveau_gpuobj_ref(NULL, &grctx);
341 atomic_dec(&chan->vm->engref[engine]);
342 chan->engctx[engine] = NULL;
345 static int
346 nv50_graph_object_new(struct nouveau_channel *chan, int engine,
347 u32 handle, u16 class)
349 struct drm_device *dev = chan->dev;
350 struct drm_nouveau_private *dev_priv = dev->dev_private;
351 struct nouveau_gpuobj *obj = NULL;
352 int ret;
354 ret = nouveau_gpuobj_new(dev, chan, 16, 16, NVOBJ_FLAG_ZERO_FREE, &obj);
355 if (ret)
356 return ret;
357 obj->engine = 1;
358 obj->class = class;
360 nv_wo32(obj, 0x00, class);
361 nv_wo32(obj, 0x04, 0x00000000);
362 nv_wo32(obj, 0x08, 0x00000000);
363 nv_wo32(obj, 0x0c, 0x00000000);
364 dev_priv->engine.instmem.flush(dev);
366 ret = nouveau_ramht_insert(chan, handle, obj);
367 nouveau_gpuobj_ref(NULL, &obj);
368 return ret;
371 static void
372 nv50_graph_context_switch(struct drm_device *dev)
374 uint32_t inst;
376 nv50_graph_unload_context(dev);
378 inst = nv_rd32(dev, NV50_PGRAPH_CTXCTL_NEXT);
379 inst &= NV50_PGRAPH_CTXCTL_NEXT_INSTANCE;
380 nv50_graph_do_load_context(dev, inst);
382 nv_wr32(dev, NV40_PGRAPH_INTR_EN, nv_rd32(dev,
383 NV40_PGRAPH_INTR_EN) | NV_PGRAPH_INTR_CONTEXT_SWITCH);
386 static int
387 nv50_graph_nvsw_dma_vblsem(struct nouveau_channel *chan,
388 u32 class, u32 mthd, u32 data)
390 struct nouveau_gpuobj *gpuobj;
392 gpuobj = nouveau_ramht_find(chan, data);
393 if (!gpuobj)
394 return -ENOENT;
396 if (nouveau_notifier_offset(gpuobj, NULL))
397 return -EINVAL;
399 chan->nvsw.vblsem = gpuobj;
400 chan->nvsw.vblsem_offset = ~0;
401 return 0;
404 static int
405 nv50_graph_nvsw_vblsem_offset(struct nouveau_channel *chan,
406 u32 class, u32 mthd, u32 data)
408 if (nouveau_notifier_offset(chan->nvsw.vblsem, &data))
409 return -ERANGE;
411 chan->nvsw.vblsem_offset = data >> 2;
412 return 0;
415 static int
416 nv50_graph_nvsw_vblsem_release_val(struct nouveau_channel *chan,
417 u32 class, u32 mthd, u32 data)
419 chan->nvsw.vblsem_rval = data;
420 return 0;
423 static int
424 nv50_graph_nvsw_vblsem_release(struct nouveau_channel *chan,
425 u32 class, u32 mthd, u32 data)
427 struct drm_device *dev = chan->dev;
428 struct drm_nouveau_private *dev_priv = dev->dev_private;
430 if (!chan->nvsw.vblsem || chan->nvsw.vblsem_offset == ~0 || data > 1)
431 return -EINVAL;
433 drm_vblank_get(dev, data);
435 chan->nvsw.vblsem_head = data;
436 list_add(&chan->nvsw.vbl_wait, &dev_priv->vbl_waiting);
438 return 0;
441 static int
442 nv50_graph_nvsw_mthd_page_flip(struct nouveau_channel *chan,
443 u32 class, u32 mthd, u32 data)
445 nouveau_finish_page_flip(chan, NULL);
446 return 0;
450 static void
451 nv50_graph_tlb_flush(struct drm_device *dev, int engine)
453 nv50_vm_flush_engine(dev, 0);
456 static void
457 nv84_graph_tlb_flush(struct drm_device *dev, int engine)
459 struct drm_nouveau_private *dev_priv = dev->dev_private;
460 struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer;
461 bool idle, timeout = false;
462 unsigned long flags;
463 u64 start;
464 u32 tmp;
466 spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
467 nv_mask(dev, 0x400500, 0x00000001, 0x00000000);
469 start = ptimer->read(dev);
470 do {
471 idle = true;
473 for (tmp = nv_rd32(dev, 0x400380); tmp && idle; tmp >>= 3) {
474 if ((tmp & 7) == 1)
475 idle = false;
478 for (tmp = nv_rd32(dev, 0x400384); tmp && idle; tmp >>= 3) {
479 if ((tmp & 7) == 1)
480 idle = false;
483 for (tmp = nv_rd32(dev, 0x400388); tmp && idle; tmp >>= 3) {
484 if ((tmp & 7) == 1)
485 idle = false;
487 } while (!idle && !(timeout = ptimer->read(dev) - start > 2000000000));
489 if (timeout) {
490 NV_ERROR(dev, "PGRAPH TLB flush idle timeout fail: "
491 "0x%08x 0x%08x 0x%08x 0x%08x\n",
492 nv_rd32(dev, 0x400700), nv_rd32(dev, 0x400380),
493 nv_rd32(dev, 0x400384), nv_rd32(dev, 0x400388));
496 nv50_vm_flush_engine(dev, 0);
498 nv_mask(dev, 0x400500, 0x00000001, 0x00000001);
499 spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
502 static struct nouveau_enum nv50_mp_exec_error_names[] = {
503 { 3, "STACK_UNDERFLOW", NULL },
504 { 4, "QUADON_ACTIVE", NULL },
505 { 8, "TIMEOUT", NULL },
506 { 0x10, "INVALID_OPCODE", NULL },
507 { 0x40, "BREAKPOINT", NULL },
511 static struct nouveau_bitfield nv50_graph_trap_m2mf[] = {
512 { 0x00000001, "NOTIFY" },
513 { 0x00000002, "IN" },
514 { 0x00000004, "OUT" },
518 static struct nouveau_bitfield nv50_graph_trap_vfetch[] = {
519 { 0x00000001, "FAULT" },
523 static struct nouveau_bitfield nv50_graph_trap_strmout[] = {
524 { 0x00000001, "FAULT" },
528 static struct nouveau_bitfield nv50_graph_trap_ccache[] = {
529 { 0x00000001, "FAULT" },
533 /* There must be a *lot* of these. Will take some time to gather them up. */
534 struct nouveau_enum nv50_data_error_names[] = {
535 { 0x00000003, "INVALID_QUERY_OR_TEXTURE", NULL },
536 { 0x00000004, "INVALID_VALUE", NULL },
537 { 0x00000005, "INVALID_ENUM", NULL },
538 { 0x00000008, "INVALID_OBJECT", NULL },
539 { 0x00000009, "READ_ONLY_OBJECT", NULL },
540 { 0x0000000a, "SUPERVISOR_OBJECT", NULL },
541 { 0x0000000b, "INVALID_ADDRESS_ALIGNMENT", NULL },
542 { 0x0000000c, "INVALID_BITFIELD", NULL },
543 { 0x0000000d, "BEGIN_END_ACTIVE", NULL },
544 { 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT", NULL },
545 { 0x0000000f, "VIEWPORT_ID_NEEDS_GP", NULL },
546 { 0x00000010, "RT_DOUBLE_BIND", NULL },
547 { 0x00000011, "RT_TYPES_MISMATCH", NULL },
548 { 0x00000012, "RT_LINEAR_WITH_ZETA", NULL },
549 { 0x00000015, "FP_TOO_FEW_REGS", NULL },
550 { 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH", NULL },
551 { 0x00000017, "RT_LINEAR_WITH_MSAA", NULL },
552 { 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT", NULL },
553 { 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT", NULL },
554 { 0x0000001a, "RT_INVALID_ALIGNMENT", NULL },
555 { 0x0000001b, "SAMPLER_OVER_LIMIT", NULL },
556 { 0x0000001c, "TEXTURE_OVER_LIMIT", NULL },
557 { 0x0000001e, "GP_TOO_MANY_OUTPUTS", NULL },
558 { 0x0000001f, "RT_BPP128_WITH_MS8", NULL },
559 { 0x00000021, "Z_OUT_OF_BOUNDS", NULL },
560 { 0x00000023, "XY_OUT_OF_BOUNDS", NULL },
561 { 0x00000027, "CP_MORE_PARAMS_THAN_SHARED", NULL },
562 { 0x00000028, "CP_NO_REG_SPACE_STRIPED", NULL },
563 { 0x00000029, "CP_NO_REG_SPACE_PACKED", NULL },
564 { 0x0000002a, "CP_NOT_ENOUGH_WARPS", NULL },
565 { 0x0000002b, "CP_BLOCK_SIZE_MISMATCH", NULL },
566 { 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS", NULL },
567 { 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS", NULL },
568 { 0x0000002e, "CP_NO_BLOCKDIM_LATCH", NULL },
569 { 0x00000031, "ENG2D_FORMAT_MISMATCH", NULL },
570 { 0x0000003f, "PRIMITIVE_ID_NEEDS_GP", NULL },
571 { 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT", NULL },
572 { 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT", NULL },
573 { 0x00000046, "LAYER_ID_NEEDS_GP", NULL },
574 { 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT", NULL },
575 { 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT", NULL },
579 static struct nouveau_bitfield nv50_graph_intr[] = {
580 { 0x00000001, "NOTIFY" },
581 { 0x00000002, "COMPUTE_QUERY" },
582 { 0x00000010, "ILLEGAL_MTHD" },
583 { 0x00000020, "ILLEGAL_CLASS" },
584 { 0x00000040, "DOUBLE_NOTIFY" },
585 { 0x00001000, "CONTEXT_SWITCH" },
586 { 0x00010000, "BUFFER_NOTIFY" },
587 { 0x00100000, "DATA_ERROR" },
588 { 0x00200000, "TRAP" },
589 { 0x01000000, "SINGLE_STEP" },
593 static void
594 nv50_pgraph_mp_trap(struct drm_device *dev, int tpid, int display)
596 struct drm_nouveau_private *dev_priv = dev->dev_private;
597 uint32_t units = nv_rd32(dev, 0x1540);
598 uint32_t addr, mp10, status, pc, oplow, ophigh;
599 int i;
600 int mps = 0;
601 for (i = 0; i < 4; i++) {
602 if (!(units & 1 << (i+24)))
603 continue;
604 if (dev_priv->chipset < 0xa0)
605 addr = 0x408200 + (tpid << 12) + (i << 7);
606 else
607 addr = 0x408100 + (tpid << 11) + (i << 7);
608 mp10 = nv_rd32(dev, addr + 0x10);
609 status = nv_rd32(dev, addr + 0x14);
610 if (!status)
611 continue;
612 if (display) {
613 nv_rd32(dev, addr + 0x20);
614 pc = nv_rd32(dev, addr + 0x24);
615 oplow = nv_rd32(dev, addr + 0x70);
616 ophigh = nv_rd32(dev, addr + 0x74);
617 NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - "
618 "TP %d MP %d: ", tpid, i);
619 nouveau_enum_print(nv50_mp_exec_error_names, status);
620 printk(" at %06x warp %d, opcode %08x %08x\n",
621 pc&0xffffff, pc >> 24,
622 oplow, ophigh);
624 nv_wr32(dev, addr + 0x10, mp10);
625 nv_wr32(dev, addr + 0x14, 0);
626 mps++;
628 if (!mps && display)
629 NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - TP %d: "
630 "No MPs claiming errors?\n", tpid);
633 static void
634 nv50_pgraph_tp_trap(struct drm_device *dev, int type, uint32_t ustatus_old,
635 uint32_t ustatus_new, int display, const char *name)
637 struct drm_nouveau_private *dev_priv = dev->dev_private;
638 int tps = 0;
639 uint32_t units = nv_rd32(dev, 0x1540);
640 int i, r;
641 uint32_t ustatus_addr, ustatus;
642 for (i = 0; i < 16; i++) {
643 if (!(units & (1 << i)))
644 continue;
645 if (dev_priv->chipset < 0xa0)
646 ustatus_addr = ustatus_old + (i << 12);
647 else
648 ustatus_addr = ustatus_new + (i << 11);
649 ustatus = nv_rd32(dev, ustatus_addr) & 0x7fffffff;
650 if (!ustatus)
651 continue;
652 tps++;
653 switch (type) {
654 case 6: /* texture error... unknown for now */
655 if (display) {
656 NV_ERROR(dev, "magic set %d:\n", i);
657 for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
658 NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r,
659 nv_rd32(dev, r));
661 break;
662 case 7: /* MP error */
663 if (ustatus & 0x00010000) {
664 nv50_pgraph_mp_trap(dev, i, display);
665 ustatus &= ~0x00010000;
667 break;
668 case 8: /* TPDMA error */
670 uint32_t e0c = nv_rd32(dev, ustatus_addr + 4);
671 uint32_t e10 = nv_rd32(dev, ustatus_addr + 8);
672 uint32_t e14 = nv_rd32(dev, ustatus_addr + 0xc);
673 uint32_t e18 = nv_rd32(dev, ustatus_addr + 0x10);
674 uint32_t e1c = nv_rd32(dev, ustatus_addr + 0x14);
675 uint32_t e20 = nv_rd32(dev, ustatus_addr + 0x18);
676 uint32_t e24 = nv_rd32(dev, ustatus_addr + 0x1c);
677 /* 2d engine destination */
678 if (ustatus & 0x00000010) {
679 if (display) {
680 NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - Unknown fault at address %02x%08x\n",
681 i, e14, e10);
682 NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
683 i, e0c, e18, e1c, e20, e24);
685 ustatus &= ~0x00000010;
687 /* Render target */
688 if (ustatus & 0x00000040) {
689 if (display) {
690 NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - Unknown fault at address %02x%08x\n",
691 i, e14, e10);
692 NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
693 i, e0c, e18, e1c, e20, e24);
695 ustatus &= ~0x00000040;
697 /* CUDA memory: l[], g[] or stack. */
698 if (ustatus & 0x00000080) {
699 if (display) {
700 if (e18 & 0x80000000) {
701 /* g[] read fault? */
702 NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global read fault at address %02x%08x\n",
703 i, e14, e10 | ((e18 >> 24) & 0x1f));
704 e18 &= ~0x1f000000;
705 } else if (e18 & 0xc) {
706 /* g[] write fault? */
707 NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global write fault at address %02x%08x\n",
708 i, e14, e10 | ((e18 >> 7) & 0x1f));
709 e18 &= ~0x00000f80;
710 } else {
711 NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Unknown CUDA fault at address %02x%08x\n",
712 i, e14, e10);
714 NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
715 i, e0c, e18, e1c, e20, e24);
717 ustatus &= ~0x00000080;
720 break;
722 if (ustatus) {
723 if (display)
724 NV_INFO(dev, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus);
726 nv_wr32(dev, ustatus_addr, 0xc0000000);
729 if (!tps && display)
730 NV_INFO(dev, "%s - No TPs claiming errors?\n", name);
733 static int
734 nv50_pgraph_trap_handler(struct drm_device *dev, u32 display, u64 inst, u32 chid)
736 u32 status = nv_rd32(dev, 0x400108);
737 u32 ustatus;
739 if (!status && display) {
740 NV_INFO(dev, "PGRAPH - TRAP: no units reporting traps?\n");
741 return 1;
744 /* DISPATCH: Relays commands to other units and handles NOTIFY,
745 * COND, QUERY. If you get a trap from it, the command is still stuck
746 * in DISPATCH and you need to do something about it. */
747 if (status & 0x001) {
748 ustatus = nv_rd32(dev, 0x400804) & 0x7fffffff;
749 if (!ustatus && display) {
750 NV_INFO(dev, "PGRAPH_TRAP_DISPATCH - no ustatus?\n");
753 nv_wr32(dev, 0x400500, 0x00000000);
755 /* Known to be triggered by screwed up NOTIFY and COND... */
756 if (ustatus & 0x00000001) {
757 u32 addr = nv_rd32(dev, 0x400808);
758 u32 subc = (addr & 0x00070000) >> 16;
759 u32 mthd = (addr & 0x00001ffc);
760 u32 datal = nv_rd32(dev, 0x40080c);
761 u32 datah = nv_rd32(dev, 0x400810);
762 u32 class = nv_rd32(dev, 0x400814);
763 u32 r848 = nv_rd32(dev, 0x400848);
765 NV_INFO(dev, "PGRAPH - TRAP DISPATCH_FAULT\n");
766 if (display && (addr & 0x80000000)) {
767 NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) "
768 "subc %d class 0x%04x mthd 0x%04x "
769 "data 0x%08x%08x "
770 "400808 0x%08x 400848 0x%08x\n",
771 chid, inst, subc, class, mthd, datah,
772 datal, addr, r848);
773 } else
774 if (display) {
775 NV_INFO(dev, "PGRAPH - no stuck command?\n");
778 nv_wr32(dev, 0x400808, 0);
779 nv_wr32(dev, 0x4008e8, nv_rd32(dev, 0x4008e8) & 3);
780 nv_wr32(dev, 0x400848, 0);
781 ustatus &= ~0x00000001;
784 if (ustatus & 0x00000002) {
785 u32 addr = nv_rd32(dev, 0x40084c);
786 u32 subc = (addr & 0x00070000) >> 16;
787 u32 mthd = (addr & 0x00001ffc);
788 u32 data = nv_rd32(dev, 0x40085c);
789 u32 class = nv_rd32(dev, 0x400814);
791 NV_INFO(dev, "PGRAPH - TRAP DISPATCH_QUERY\n");
792 if (display && (addr & 0x80000000)) {
793 NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) "
794 "subc %d class 0x%04x mthd 0x%04x "
795 "data 0x%08x 40084c 0x%08x\n",
796 chid, inst, subc, class, mthd,
797 data, addr);
798 } else
799 if (display) {
800 NV_INFO(dev, "PGRAPH - no stuck command?\n");
803 nv_wr32(dev, 0x40084c, 0);
804 ustatus &= ~0x00000002;
807 if (ustatus && display) {
808 NV_INFO(dev, "PGRAPH - TRAP_DISPATCH (unknown "
809 "0x%08x)\n", ustatus);
812 nv_wr32(dev, 0x400804, 0xc0000000);
813 nv_wr32(dev, 0x400108, 0x001);
814 status &= ~0x001;
815 if (!status)
816 return 0;
819 /* M2MF: Memory to memory copy engine. */
820 if (status & 0x002) {
821 u32 ustatus = nv_rd32(dev, 0x406800) & 0x7fffffff;
822 if (display) {
823 NV_INFO(dev, "PGRAPH - TRAP_M2MF");
824 nouveau_bitfield_print(nv50_graph_trap_m2mf, ustatus);
825 printk("\n");
826 NV_INFO(dev, "PGRAPH - TRAP_M2MF %08x %08x %08x %08x\n",
827 nv_rd32(dev, 0x406804), nv_rd32(dev, 0x406808),
828 nv_rd32(dev, 0x40680c), nv_rd32(dev, 0x406810));
832 /* No sane way found yet -- just reset the bugger. */
833 nv_wr32(dev, 0x400040, 2);
834 nv_wr32(dev, 0x400040, 0);
835 nv_wr32(dev, 0x406800, 0xc0000000);
836 nv_wr32(dev, 0x400108, 0x002);
837 status &= ~0x002;
840 /* VFETCH: Fetches data from vertex buffers. */
841 if (status & 0x004) {
842 u32 ustatus = nv_rd32(dev, 0x400c04) & 0x7fffffff;
843 if (display) {
844 NV_INFO(dev, "PGRAPH - TRAP_VFETCH");
845 nouveau_bitfield_print(nv50_graph_trap_vfetch, ustatus);
846 printk("\n");
847 NV_INFO(dev, "PGRAPH - TRAP_VFETCH %08x %08x %08x %08x\n",
848 nv_rd32(dev, 0x400c00), nv_rd32(dev, 0x400c08),
849 nv_rd32(dev, 0x400c0c), nv_rd32(dev, 0x400c10));
852 nv_wr32(dev, 0x400c04, 0xc0000000);
853 nv_wr32(dev, 0x400108, 0x004);
854 status &= ~0x004;
857 /* STRMOUT: DirectX streamout / OpenGL transform feedback. */
858 if (status & 0x008) {
859 ustatus = nv_rd32(dev, 0x401800) & 0x7fffffff;
860 if (display) {
861 NV_INFO(dev, "PGRAPH - TRAP_STRMOUT");
862 nouveau_bitfield_print(nv50_graph_trap_strmout, ustatus);
863 printk("\n");
864 NV_INFO(dev, "PGRAPH - TRAP_STRMOUT %08x %08x %08x %08x\n",
865 nv_rd32(dev, 0x401804), nv_rd32(dev, 0x401808),
866 nv_rd32(dev, 0x40180c), nv_rd32(dev, 0x401810));
870 /* No sane way found yet -- just reset the bugger. */
871 nv_wr32(dev, 0x400040, 0x80);
872 nv_wr32(dev, 0x400040, 0);
873 nv_wr32(dev, 0x401800, 0xc0000000);
874 nv_wr32(dev, 0x400108, 0x008);
875 status &= ~0x008;
878 /* CCACHE: Handles code and c[] caches and fills them. */
879 if (status & 0x010) {
880 ustatus = nv_rd32(dev, 0x405018) & 0x7fffffff;
881 if (display) {
882 NV_INFO(dev, "PGRAPH - TRAP_CCACHE");
883 nouveau_bitfield_print(nv50_graph_trap_ccache, ustatus);
884 printk("\n");
885 NV_INFO(dev, "PGRAPH - TRAP_CCACHE %08x %08x %08x %08x"
886 " %08x %08x %08x\n",
887 nv_rd32(dev, 0x405000), nv_rd32(dev, 0x405004),
888 nv_rd32(dev, 0x405008), nv_rd32(dev, 0x40500c),
889 nv_rd32(dev, 0x405010), nv_rd32(dev, 0x405014),
890 nv_rd32(dev, 0x40501c));
894 nv_wr32(dev, 0x405018, 0xc0000000);
895 nv_wr32(dev, 0x400108, 0x010);
896 status &= ~0x010;
899 /* Unknown, not seen yet... 0x402000 is the only trap status reg
900 * remaining, so try to handle it anyway. Perhaps related to that
901 * unknown DMA slot on tesla? */
902 if (status & 0x20) {
903 ustatus = nv_rd32(dev, 0x402000) & 0x7fffffff;
904 if (display)
905 NV_INFO(dev, "PGRAPH - TRAP_UNKC04 0x%08x\n", ustatus);
906 nv_wr32(dev, 0x402000, 0xc0000000);
907 /* no status modifiction on purpose */
910 /* TEXTURE: CUDA texturing units */
911 if (status & 0x040) {
912 nv50_pgraph_tp_trap(dev, 6, 0x408900, 0x408600, display,
913 "PGRAPH - TRAP_TEXTURE");
914 nv_wr32(dev, 0x400108, 0x040);
915 status &= ~0x040;
918 /* MP: CUDA execution engines. */
919 if (status & 0x080) {
920 nv50_pgraph_tp_trap(dev, 7, 0x408314, 0x40831c, display,
921 "PGRAPH - TRAP_MP");
922 nv_wr32(dev, 0x400108, 0x080);
923 status &= ~0x080;
926 /* TPDMA: Handles TP-initiated uncached memory accesses:
927 * l[], g[], stack, 2d surfaces, render targets. */
928 if (status & 0x100) {
929 nv50_pgraph_tp_trap(dev, 8, 0x408e08, 0x408708, display,
930 "PGRAPH - TRAP_TPDMA");
931 nv_wr32(dev, 0x400108, 0x100);
932 status &= ~0x100;
935 if (status) {
936 if (display)
937 NV_INFO(dev, "PGRAPH - TRAP: unknown 0x%08x\n", status);
938 nv_wr32(dev, 0x400108, status);
941 return 1;
945 nv50_graph_isr_chid(struct drm_device *dev, u64 inst)
947 struct drm_nouveau_private *dev_priv = dev->dev_private;
948 struct nouveau_channel *chan;
949 unsigned long flags;
950 int i;
952 spin_lock_irqsave(&dev_priv->channels.lock, flags);
953 for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
954 chan = dev_priv->channels.ptr[i];
955 if (!chan || !chan->ramin)
956 continue;
958 if (inst == chan->ramin->vinst)
959 break;
961 spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
962 return i;
965 static void
966 nv50_graph_isr(struct drm_device *dev)
968 u32 stat;
970 while ((stat = nv_rd32(dev, 0x400100))) {
971 u64 inst = (u64)(nv_rd32(dev, 0x40032c) & 0x0fffffff) << 12;
972 u32 chid = nv50_graph_isr_chid(dev, inst);
973 u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
974 u32 subc = (addr & 0x00070000) >> 16;
975 u32 mthd = (addr & 0x00001ffc);
976 u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
977 u32 class = nv_rd32(dev, 0x400814);
978 u32 show = stat;
980 if (stat & 0x00000010) {
981 if (!nouveau_gpuobj_mthd_call2(dev, chid, class,
982 mthd, data))
983 show &= ~0x00000010;
986 if (stat & 0x00001000) {
987 nv_wr32(dev, 0x400500, 0x00000000);
988 nv_wr32(dev, 0x400100, 0x00001000);
989 nv_mask(dev, 0x40013c, 0x00001000, 0x00000000);
990 nv50_graph_context_switch(dev);
991 stat &= ~0x00001000;
992 show &= ~0x00001000;
995 show = (show && nouveau_ratelimit()) ? show : 0;
997 if (show & 0x00100000) {
998 u32 ecode = nv_rd32(dev, 0x400110);
999 NV_INFO(dev, "PGRAPH - DATA_ERROR ");
1000 nouveau_enum_print(nv50_data_error_names, ecode);
1001 printk("\n");
1004 if (stat & 0x00200000) {
1005 if (!nv50_pgraph_trap_handler(dev, show, inst, chid))
1006 show &= ~0x00200000;
1009 nv_wr32(dev, 0x400100, stat);
1010 nv_wr32(dev, 0x400500, 0x00010001);
1012 if (show) {
1013 NV_INFO(dev, "PGRAPH -");
1014 nouveau_bitfield_print(nv50_graph_intr, show);
1015 printk("\n");
1016 NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) subc %d "
1017 "class 0x%04x mthd 0x%04x data 0x%08x\n",
1018 chid, inst, subc, class, mthd, data);
1019 nv50_fb_vm_trap(dev, 1);
1023 if (nv_rd32(dev, 0x400824) & (1 << 31))
1024 nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) & ~(1 << 31));
1027 static void
1028 nv50_graph_destroy(struct drm_device *dev, int engine)
1030 struct nv50_graph_engine *pgraph = nv_engine(dev, engine);
1032 NVOBJ_ENGINE_DEL(dev, GR);
1034 nouveau_irq_unregister(dev, 12);
1035 kfree(pgraph);
1039 nv50_graph_create(struct drm_device *dev)
1041 struct drm_nouveau_private *dev_priv = dev->dev_private;
1042 struct nv50_graph_engine *pgraph;
1043 struct nouveau_grctx ctx = {};
1044 int ret;
1046 pgraph = kzalloc(sizeof(*pgraph),GFP_KERNEL);
1047 if (!pgraph)
1048 return -ENOMEM;
1050 ctx.dev = dev;
1051 ctx.mode = NOUVEAU_GRCTX_PROG;
1052 ctx.data = pgraph->ctxprog;
1053 ctx.ctxprog_max = ARRAY_SIZE(pgraph->ctxprog);
1055 ret = nv50_grctx_init(&ctx);
1056 if (ret) {
1057 NV_ERROR(dev, "PGRAPH: ctxprog build failed\n");
1058 kfree(pgraph);
1059 return 0;
1062 pgraph->grctx_size = ctx.ctxvals_pos * 4;
1063 pgraph->ctxprog_size = ctx.ctxprog_len;
1065 pgraph->base.destroy = nv50_graph_destroy;
1066 pgraph->base.init = nv50_graph_init;
1067 pgraph->base.fini = nv50_graph_fini;
1068 pgraph->base.context_new = nv50_graph_context_new;
1069 pgraph->base.context_del = nv50_graph_context_del;
1070 pgraph->base.object_new = nv50_graph_object_new;
1071 if (dev_priv->chipset == 0x50 || dev_priv->chipset == 0xac)
1072 pgraph->base.tlb_flush = nv50_graph_tlb_flush;
1073 else
1074 pgraph->base.tlb_flush = nv84_graph_tlb_flush;
1076 nouveau_irq_register(dev, 12, nv50_graph_isr);
1078 /* NVSW really doesn't live here... */
1079 NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
1080 NVOBJ_MTHD (dev, 0x506e, 0x018c, nv50_graph_nvsw_dma_vblsem);
1081 NVOBJ_MTHD (dev, 0x506e, 0x0400, nv50_graph_nvsw_vblsem_offset);
1082 NVOBJ_MTHD (dev, 0x506e, 0x0404, nv50_graph_nvsw_vblsem_release_val);
1083 NVOBJ_MTHD (dev, 0x506e, 0x0408, nv50_graph_nvsw_vblsem_release);
1084 NVOBJ_MTHD (dev, 0x506e, 0x0500, nv50_graph_nvsw_mthd_page_flip);
1086 NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
1087 NVOBJ_CLASS(dev, 0x0030, GR); /* null */
1088 NVOBJ_CLASS(dev, 0x5039, GR); /* m2mf */
1089 NVOBJ_CLASS(dev, 0x502d, GR); /* 2d */
1091 /* tesla */
1092 if (dev_priv->chipset == 0x50)
1093 NVOBJ_CLASS(dev, 0x5097, GR); /* tesla (nv50) */
1094 else
1095 if (dev_priv->chipset < 0xa0)
1096 NVOBJ_CLASS(dev, 0x8297, GR); /* tesla (nv8x/nv9x) */
1097 else {
1098 switch (dev_priv->chipset) {
1099 case 0xa0:
1100 case 0xaa:
1101 case 0xac:
1102 NVOBJ_CLASS(dev, 0x8397, GR);
1103 break;
1104 case 0xa3:
1105 case 0xa5:
1106 case 0xa8:
1107 NVOBJ_CLASS(dev, 0x8597, GR);
1108 break;
1109 case 0xaf:
1110 NVOBJ_CLASS(dev, 0x8697, GR);
1111 break;
1115 /* compute */
1116 NVOBJ_CLASS(dev, 0x50c0, GR);
1117 if (dev_priv->chipset > 0xa0 &&
1118 dev_priv->chipset != 0xaa &&
1119 dev_priv->chipset != 0xac)
1120 NVOBJ_CLASS(dev, 0x85c0, GR);
1122 return 0;