]> git.karo-electronics.de Git - mv-sheeva.git/blob - drivers/gpu/drm/nouveau/nv50_graph.c
drm/nv50: improve evo error handler when more than just channel 0 active
[mv-sheeva.git] / drivers / gpu / drm / nouveau / nv50_graph.c
1 /*
2  * Copyright (C) 2007 Ben Skeggs.
3  * All Rights Reserved.
4  *
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:
12  *
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.
16  *
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.
24  *
25  */
26
27 #include "drmP.h"
28 #include "drm.h"
29 #include "nouveau_drv.h"
30 #include "nouveau_ramht.h"
31 #include "nouveau_grctx.h"
32
33 static int nv50_graph_register(struct drm_device *);
34
35 static void
36 nv50_graph_init_reset(struct drm_device *dev)
37 {
38         uint32_t pmc_e = NV_PMC_ENABLE_PGRAPH | (1 << 21);
39
40         NV_DEBUG(dev, "\n");
41
42         nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) & ~pmc_e);
43         nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) |  pmc_e);
44 }
45
46 static void
47 nv50_graph_init_intr(struct drm_device *dev)
48 {
49         NV_DEBUG(dev, "\n");
50
51         nv_wr32(dev, NV03_PGRAPH_INTR, 0xffffffff);
52         nv_wr32(dev, 0x400138, 0xffffffff);
53         nv_wr32(dev, NV40_PGRAPH_INTR_EN, 0xffffffff);
54 }
55
56 static void
57 nv50_graph_init_regs__nv(struct drm_device *dev)
58 {
59         struct drm_nouveau_private *dev_priv = dev->dev_private;
60         uint32_t units = nv_rd32(dev, 0x1540);
61         int i;
62
63         NV_DEBUG(dev, "\n");
64
65         nv_wr32(dev, 0x400804, 0xc0000000);
66         nv_wr32(dev, 0x406800, 0xc0000000);
67         nv_wr32(dev, 0x400c04, 0xc0000000);
68         nv_wr32(dev, 0x401800, 0xc0000000);
69         nv_wr32(dev, 0x405018, 0xc0000000);
70         nv_wr32(dev, 0x402000, 0xc0000000);
71
72         for (i = 0; i < 16; i++) {
73                 if (units & 1 << i) {
74                         if (dev_priv->chipset < 0xa0) {
75                                 nv_wr32(dev, 0x408900 + (i << 12), 0xc0000000);
76                                 nv_wr32(dev, 0x408e08 + (i << 12), 0xc0000000);
77                                 nv_wr32(dev, 0x408314 + (i << 12), 0xc0000000);
78                         } else {
79                                 nv_wr32(dev, 0x408600 + (i << 11), 0xc0000000);
80                                 nv_wr32(dev, 0x408708 + (i << 11), 0xc0000000);
81                                 nv_wr32(dev, 0x40831c + (i << 11), 0xc0000000);
82                         }
83                 }
84         }
85
86         nv_wr32(dev, 0x400108, 0xffffffff);
87
88         nv_wr32(dev, 0x400824, 0x00004000);
89         nv_wr32(dev, 0x400500, 0x00010001);
90 }
91
92 static void
93 nv50_graph_init_regs(struct drm_device *dev)
94 {
95         NV_DEBUG(dev, "\n");
96
97         nv_wr32(dev, NV04_PGRAPH_DEBUG_3,
98                                 (1 << 2) /* HW_CONTEXT_SWITCH_ENABLED */);
99         nv_wr32(dev, 0x402ca8, 0x800);
100 }
101
102 static int
103 nv50_graph_init_ctxctl(struct drm_device *dev)
104 {
105         struct drm_nouveau_private *dev_priv = dev->dev_private;
106         struct nouveau_grctx ctx = {};
107         uint32_t *cp;
108         int i;
109
110         NV_DEBUG(dev, "\n");
111
112         cp = kmalloc(512 * 4, GFP_KERNEL);
113         if (!cp) {
114                 NV_ERROR(dev, "failed to allocate ctxprog\n");
115                 dev_priv->engine.graph.accel_blocked = true;
116                 return 0;
117         }
118
119         ctx.dev = dev;
120         ctx.mode = NOUVEAU_GRCTX_PROG;
121         ctx.data = cp;
122         ctx.ctxprog_max = 512;
123         if (!nv50_grctx_init(&ctx)) {
124                 dev_priv->engine.graph.grctx_size = ctx.ctxvals_pos * 4;
125
126                 nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
127                 for (i = 0; i < ctx.ctxprog_len; i++)
128                         nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_DATA, cp[i]);
129         } else {
130                 dev_priv->engine.graph.accel_blocked = true;
131         }
132         kfree(cp);
133
134         nv_wr32(dev, 0x400320, 4);
135         nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, 0);
136         nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, 0);
137         return 0;
138 }
139
140 int
141 nv50_graph_init(struct drm_device *dev)
142 {
143         int ret;
144
145         NV_DEBUG(dev, "\n");
146
147         nv50_graph_init_reset(dev);
148         nv50_graph_init_regs__nv(dev);
149         nv50_graph_init_regs(dev);
150
151         ret = nv50_graph_init_ctxctl(dev);
152         if (ret)
153                 return ret;
154
155         ret = nv50_graph_register(dev);
156         if (ret)
157                 return ret;
158         nv50_graph_init_intr(dev);
159         return 0;
160 }
161
162 void
163 nv50_graph_takedown(struct drm_device *dev)
164 {
165         NV_DEBUG(dev, "\n");
166 }
167
168 void
169 nv50_graph_fifo_access(struct drm_device *dev, bool enabled)
170 {
171         const uint32_t mask = 0x00010001;
172
173         if (enabled)
174                 nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) | mask);
175         else
176                 nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) & ~mask);
177 }
178
179 struct nouveau_channel *
180 nv50_graph_channel(struct drm_device *dev)
181 {
182         struct drm_nouveau_private *dev_priv = dev->dev_private;
183         uint32_t inst;
184         int i;
185
186         /* Be sure we're not in the middle of a context switch or bad things
187          * will happen, such as unloading the wrong pgraph context.
188          */
189         if (!nv_wait(dev, 0x400300, 0x00000001, 0x00000000))
190                 NV_ERROR(dev, "Ctxprog is still running\n");
191
192         inst = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
193         if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
194                 return NULL;
195         inst = (inst & NV50_PGRAPH_CTXCTL_CUR_INSTANCE) << 12;
196
197         for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
198                 struct nouveau_channel *chan = dev_priv->channels.ptr[i];
199
200                 if (chan && chan->ramin && chan->ramin->vinst == inst)
201                         return chan;
202         }
203
204         return NULL;
205 }
206
207 int
208 nv50_graph_create_context(struct nouveau_channel *chan)
209 {
210         struct drm_device *dev = chan->dev;
211         struct drm_nouveau_private *dev_priv = dev->dev_private;
212         struct nouveau_gpuobj *ramin = chan->ramin;
213         struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
214         struct nouveau_grctx ctx = {};
215         int hdr, ret;
216
217         NV_DEBUG(dev, "ch%d\n", chan->id);
218
219         ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 0,
220                                  NVOBJ_FLAG_ZERO_ALLOC |
221                                  NVOBJ_FLAG_ZERO_FREE, &chan->ramin_grctx);
222         if (ret)
223                 return ret;
224
225         hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
226         nv_wo32(ramin, hdr + 0x00, 0x00190002);
227         nv_wo32(ramin, hdr + 0x04, chan->ramin_grctx->vinst +
228                                    pgraph->grctx_size - 1);
229         nv_wo32(ramin, hdr + 0x08, chan->ramin_grctx->vinst);
230         nv_wo32(ramin, hdr + 0x0c, 0);
231         nv_wo32(ramin, hdr + 0x10, 0);
232         nv_wo32(ramin, hdr + 0x14, 0x00010000);
233
234         ctx.dev = chan->dev;
235         ctx.mode = NOUVEAU_GRCTX_VALS;
236         ctx.data = chan->ramin_grctx;
237         nv50_grctx_init(&ctx);
238
239         nv_wo32(chan->ramin_grctx, 0x00000, chan->ramin->vinst >> 12);
240
241         dev_priv->engine.instmem.flush(dev);
242         return 0;
243 }
244
245 void
246 nv50_graph_destroy_context(struct nouveau_channel *chan)
247 {
248         struct drm_device *dev = chan->dev;
249         struct drm_nouveau_private *dev_priv = dev->dev_private;
250         struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
251         int i, hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
252         unsigned long flags;
253
254         NV_DEBUG(dev, "ch%d\n", chan->id);
255
256         if (!chan->ramin)
257                 return;
258
259         spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
260         pgraph->fifo_access(dev, false);
261
262         if (pgraph->channel(dev) == chan)
263                 pgraph->unload_context(dev);
264
265         for (i = hdr; i < hdr + 24; i += 4)
266                 nv_wo32(chan->ramin, i, 0);
267         dev_priv->engine.instmem.flush(dev);
268
269         pgraph->fifo_access(dev, true);
270         spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
271
272         nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
273 }
274
275 static int
276 nv50_graph_do_load_context(struct drm_device *dev, uint32_t inst)
277 {
278         uint32_t fifo = nv_rd32(dev, 0x400500);
279
280         nv_wr32(dev, 0x400500, fifo & ~1);
281         nv_wr32(dev, 0x400784, inst);
282         nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x40);
283         nv_wr32(dev, 0x400320, nv_rd32(dev, 0x400320) | 0x11);
284         nv_wr32(dev, 0x400040, 0xffffffff);
285         (void)nv_rd32(dev, 0x400040);
286         nv_wr32(dev, 0x400040, 0x00000000);
287         nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 1);
288
289         if (nouveau_wait_for_idle(dev))
290                 nv_wr32(dev, 0x40032c, inst | (1<<31));
291         nv_wr32(dev, 0x400500, fifo);
292
293         return 0;
294 }
295
296 int
297 nv50_graph_load_context(struct nouveau_channel *chan)
298 {
299         uint32_t inst = chan->ramin->vinst >> 12;
300
301         NV_DEBUG(chan->dev, "ch%d\n", chan->id);
302         return nv50_graph_do_load_context(chan->dev, inst);
303 }
304
305 int
306 nv50_graph_unload_context(struct drm_device *dev)
307 {
308         uint32_t inst;
309
310         inst  = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
311         if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
312                 return 0;
313         inst &= NV50_PGRAPH_CTXCTL_CUR_INSTANCE;
314
315         nouveau_wait_for_idle(dev);
316         nv_wr32(dev, 0x400784, inst);
317         nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x20);
318         nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 0x01);
319         nouveau_wait_for_idle(dev);
320
321         nv_wr32(dev, NV50_PGRAPH_CTXCTL_CUR, inst);
322         return 0;
323 }
324
325 void
326 nv50_graph_context_switch(struct drm_device *dev)
327 {
328         uint32_t inst;
329
330         nv50_graph_unload_context(dev);
331
332         inst  = nv_rd32(dev, NV50_PGRAPH_CTXCTL_NEXT);
333         inst &= NV50_PGRAPH_CTXCTL_NEXT_INSTANCE;
334         nv50_graph_do_load_context(dev, inst);
335
336         nv_wr32(dev, NV40_PGRAPH_INTR_EN, nv_rd32(dev,
337                 NV40_PGRAPH_INTR_EN) | NV_PGRAPH_INTR_CONTEXT_SWITCH);
338 }
339
340 static int
341 nv50_graph_nvsw_dma_vblsem(struct nouveau_channel *chan,
342                            u32 class, u32 mthd, u32 data)
343 {
344         struct nouveau_gpuobj *gpuobj;
345
346         gpuobj = nouveau_ramht_find(chan, data);
347         if (!gpuobj)
348                 return -ENOENT;
349
350         if (nouveau_notifier_offset(gpuobj, NULL))
351                 return -EINVAL;
352
353         chan->nvsw.vblsem = gpuobj;
354         chan->nvsw.vblsem_offset = ~0;
355         return 0;
356 }
357
358 static int
359 nv50_graph_nvsw_vblsem_offset(struct nouveau_channel *chan,
360                               u32 class, u32 mthd, u32 data)
361 {
362         if (nouveau_notifier_offset(chan->nvsw.vblsem, &data))
363                 return -ERANGE;
364
365         chan->nvsw.vblsem_offset = data >> 2;
366         return 0;
367 }
368
369 static int
370 nv50_graph_nvsw_vblsem_release_val(struct nouveau_channel *chan,
371                                    u32 class, u32 mthd, u32 data)
372 {
373         chan->nvsw.vblsem_rval = data;
374         return 0;
375 }
376
377 static int
378 nv50_graph_nvsw_vblsem_release(struct nouveau_channel *chan,
379                                u32 class, u32 mthd, u32 data)
380 {
381         struct drm_device *dev = chan->dev;
382         struct drm_nouveau_private *dev_priv = dev->dev_private;
383
384         if (!chan->nvsw.vblsem || chan->nvsw.vblsem_offset == ~0 || data > 1)
385                 return -EINVAL;
386
387         if (!(nv_rd32(dev, NV50_PDISPLAY_INTR_EN_1) &
388                            NV50_PDISPLAY_INTR_EN_1_VBLANK_CRTC_(data))) {
389                 nv_wr32(dev, NV50_PDISPLAY_INTR_1,
390                              NV50_PDISPLAY_INTR_1_VBLANK_CRTC_(data));
391                 nv_mask(dev, NV50_PDISPLAY_INTR_EN_1, 0,
392                              NV50_PDISPLAY_INTR_EN_1_VBLANK_CRTC_(data));
393         }
394
395         list_add(&chan->nvsw.vbl_wait, &dev_priv->vbl_waiting);
396         return 0;
397 }
398
399 static int
400 nv50_graph_register(struct drm_device *dev)
401 {
402         struct drm_nouveau_private *dev_priv = dev->dev_private;
403
404         if (dev_priv->engine.graph.registered)
405                 return 0;
406
407         NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
408         NVOBJ_MTHD (dev, 0x506e, 0x018c, nv50_graph_nvsw_dma_vblsem);
409         NVOBJ_MTHD (dev, 0x506e, 0x0400, nv50_graph_nvsw_vblsem_offset);
410         NVOBJ_MTHD (dev, 0x506e, 0x0404, nv50_graph_nvsw_vblsem_release_val);
411         NVOBJ_MTHD (dev, 0x506e, 0x0408, nv50_graph_nvsw_vblsem_release);
412
413         NVOBJ_CLASS(dev, 0x0030, GR); /* null */
414         NVOBJ_CLASS(dev, 0x5039, GR); /* m2mf */
415         NVOBJ_CLASS(dev, 0x502d, GR); /* 2d */
416         NVOBJ_CLASS(dev, 0x50c0, GR); /* compute */
417         NVOBJ_CLASS(dev, 0x85c0, GR); /* compute (nva3, nva5, nva8) */
418
419         /* tesla */
420         if (dev_priv->chipset == 0x50)
421                 NVOBJ_CLASS(dev, 0x5097, GR); /* tesla (nv50) */
422         else
423         if (dev_priv->chipset < 0xa0)
424                 NVOBJ_CLASS(dev, 0x8297, GR); /* tesla (nv8x/nv9x) */
425         else {
426                 switch (dev_priv->chipset) {
427                 case 0xa0:
428                 case 0xaa:
429                 case 0xac:
430                         NVOBJ_CLASS(dev, 0x8397, GR);
431                         break;
432                 case 0xa3:
433                 case 0xa5:
434                 case 0xa8:
435                         NVOBJ_CLASS(dev, 0x8597, GR);
436                         break;
437                 case 0xaf:
438                         NVOBJ_CLASS(dev, 0x8697, GR);
439                         break;
440                 }
441         }
442
443         dev_priv->engine.graph.registered = true;
444         return 0;
445 }
446
447 void
448 nv50_graph_tlb_flush(struct drm_device *dev)
449 {
450         nv50_vm_flush(dev, 0);
451 }
452
453 void
454 nv86_graph_tlb_flush(struct drm_device *dev)
455 {
456         struct drm_nouveau_private *dev_priv = dev->dev_private;
457         struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer;
458         bool idle, timeout = false;
459         unsigned long flags;
460         u64 start;
461         u32 tmp;
462
463         spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
464         nv_mask(dev, 0x400500, 0x00000001, 0x00000000);
465
466         start = ptimer->read(dev);
467         do {
468                 idle = true;
469
470                 for (tmp = nv_rd32(dev, 0x400380); tmp && idle; tmp >>= 3) {
471                         if ((tmp & 7) == 1)
472                                 idle = false;
473                 }
474
475                 for (tmp = nv_rd32(dev, 0x400384); tmp && idle; tmp >>= 3) {
476                         if ((tmp & 7) == 1)
477                                 idle = false;
478                 }
479
480                 for (tmp = nv_rd32(dev, 0x400388); tmp && idle; tmp >>= 3) {
481                         if ((tmp & 7) == 1)
482                                 idle = false;
483                 }
484         } while (!idle && !(timeout = ptimer->read(dev) - start > 2000000000));
485
486         if (timeout) {
487                 NV_ERROR(dev, "PGRAPH TLB flush idle timeout fail: "
488                               "0x%08x 0x%08x 0x%08x 0x%08x\n",
489                          nv_rd32(dev, 0x400700), nv_rd32(dev, 0x400380),
490                          nv_rd32(dev, 0x400384), nv_rd32(dev, 0x400388));
491         }
492
493         nv50_vm_flush(dev, 0);
494
495         nv_mask(dev, 0x400500, 0x00000001, 0x00000001);
496         spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
497 }