Merge branches 'cma', 'misc', 'mlx4' and 'nes' into for-linus
[sfrench/cifs-2.6.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
31 #include "nouveau_grctx.h"
32
33 #define IS_G80 ((dev_priv->chipset & 0xf0) == 0x50)
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         NV_DEBUG(dev, "\n");
60
61         nv_wr32(dev, 0x400804, 0xc0000000);
62         nv_wr32(dev, 0x406800, 0xc0000000);
63         nv_wr32(dev, 0x400c04, 0xc0000000);
64         nv_wr32(dev, 0x401800, 0xc0000000);
65         nv_wr32(dev, 0x405018, 0xc0000000);
66         nv_wr32(dev, 0x402000, 0xc0000000);
67
68         nv_wr32(dev, 0x400108, 0xffffffff);
69
70         nv_wr32(dev, 0x400824, 0x00004000);
71         nv_wr32(dev, 0x400500, 0x00010001);
72 }
73
74 static void
75 nv50_graph_init_regs(struct drm_device *dev)
76 {
77         NV_DEBUG(dev, "\n");
78
79         nv_wr32(dev, NV04_PGRAPH_DEBUG_3,
80                                 (1 << 2) /* HW_CONTEXT_SWITCH_ENABLED */);
81         nv_wr32(dev, 0x402ca8, 0x800);
82 }
83
84 static int
85 nv50_graph_init_ctxctl(struct drm_device *dev)
86 {
87         struct drm_nouveau_private *dev_priv = dev->dev_private;
88
89         NV_DEBUG(dev, "\n");
90
91         if (nouveau_ctxfw) {
92                 nouveau_grctx_prog_load(dev);
93                 dev_priv->engine.graph.grctx_size = 0x70000;
94         }
95         if (!dev_priv->engine.graph.ctxprog) {
96                 struct nouveau_grctx ctx = {};
97                 uint32_t *cp = kmalloc(512 * 4, GFP_KERNEL);
98                 int i;
99                 if (!cp) {
100                         NV_ERROR(dev, "Couldn't alloc ctxprog! Disabling acceleration.\n");
101                         dev_priv->engine.graph.accel_blocked = true;
102                         return 0;
103                 }
104                 ctx.dev = dev;
105                 ctx.mode = NOUVEAU_GRCTX_PROG;
106                 ctx.data = cp;
107                 ctx.ctxprog_max = 512;
108                 if (!nv50_grctx_init(&ctx)) {
109                         dev_priv->engine.graph.grctx_size = ctx.ctxvals_pos * 4;
110
111                         nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
112                         for (i = 0; i < ctx.ctxprog_len; i++)
113                                 nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_DATA, cp[i]);
114                 } else {
115                         dev_priv->engine.graph.accel_blocked = true;
116                 }
117                 kfree(cp);
118         }
119
120         nv_wr32(dev, 0x400320, 4);
121         nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, 0);
122         nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, 0);
123         return 0;
124 }
125
126 int
127 nv50_graph_init(struct drm_device *dev)
128 {
129         int ret;
130
131         NV_DEBUG(dev, "\n");
132
133         nv50_graph_init_reset(dev);
134         nv50_graph_init_regs__nv(dev);
135         nv50_graph_init_regs(dev);
136         nv50_graph_init_intr(dev);
137
138         ret = nv50_graph_init_ctxctl(dev);
139         if (ret)
140                 return ret;
141
142         return 0;
143 }
144
145 void
146 nv50_graph_takedown(struct drm_device *dev)
147 {
148         NV_DEBUG(dev, "\n");
149         nouveau_grctx_fini(dev);
150 }
151
152 void
153 nv50_graph_fifo_access(struct drm_device *dev, bool enabled)
154 {
155         const uint32_t mask = 0x00010001;
156
157         if (enabled)
158                 nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) | mask);
159         else
160                 nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) & ~mask);
161 }
162
163 struct nouveau_channel *
164 nv50_graph_channel(struct drm_device *dev)
165 {
166         struct drm_nouveau_private *dev_priv = dev->dev_private;
167         uint32_t inst;
168         int i;
169
170         /* Be sure we're not in the middle of a context switch or bad things
171          * will happen, such as unloading the wrong pgraph context.
172          */
173         if (!nv_wait(0x400300, 0x00000001, 0x00000000))
174                 NV_ERROR(dev, "Ctxprog is still running\n");
175
176         inst = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
177         if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
178                 return NULL;
179         inst = (inst & NV50_PGRAPH_CTXCTL_CUR_INSTANCE) << 12;
180
181         for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
182                 struct nouveau_channel *chan = dev_priv->fifos[i];
183
184                 if (chan && chan->ramin && chan->ramin->instance == inst)
185                         return chan;
186         }
187
188         return NULL;
189 }
190
191 int
192 nv50_graph_create_context(struct nouveau_channel *chan)
193 {
194         struct drm_device *dev = chan->dev;
195         struct drm_nouveau_private *dev_priv = dev->dev_private;
196         struct nouveau_gpuobj *ramin = chan->ramin->gpuobj;
197         struct nouveau_gpuobj *ctx;
198         struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
199         int hdr, ret;
200
201         NV_DEBUG(dev, "ch%d\n", chan->id);
202
203         ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, pgraph->grctx_size,
204                                      0x1000, NVOBJ_FLAG_ZERO_ALLOC |
205                                      NVOBJ_FLAG_ZERO_FREE, &chan->ramin_grctx);
206         if (ret)
207                 return ret;
208         ctx = chan->ramin_grctx->gpuobj;
209
210         hdr = IS_G80 ? 0x200 : 0x20;
211         dev_priv->engine.instmem.prepare_access(dev, true);
212         nv_wo32(dev, ramin, (hdr + 0x00)/4, 0x00190002);
213         nv_wo32(dev, ramin, (hdr + 0x04)/4, chan->ramin_grctx->instance +
214                                            pgraph->grctx_size - 1);
215         nv_wo32(dev, ramin, (hdr + 0x08)/4, chan->ramin_grctx->instance);
216         nv_wo32(dev, ramin, (hdr + 0x0c)/4, 0);
217         nv_wo32(dev, ramin, (hdr + 0x10)/4, 0);
218         nv_wo32(dev, ramin, (hdr + 0x14)/4, 0x00010000);
219         dev_priv->engine.instmem.finish_access(dev);
220
221         dev_priv->engine.instmem.prepare_access(dev, true);
222         if (!pgraph->ctxprog) {
223                 struct nouveau_grctx ctx = {};
224                 ctx.dev = chan->dev;
225                 ctx.mode = NOUVEAU_GRCTX_VALS;
226                 ctx.data = chan->ramin_grctx->gpuobj;
227                 nv50_grctx_init(&ctx);
228         } else {
229                 nouveau_grctx_vals_load(dev, ctx);
230         }
231         nv_wo32(dev, ctx, 0x00000/4, chan->ramin->instance >> 12);
232         if ((dev_priv->chipset & 0xf0) == 0xa0)
233                 nv_wo32(dev, ctx, 0x00004/4, 0x00000000);
234         else
235                 nv_wo32(dev, ctx, 0x0011c/4, 0x00000000);
236         dev_priv->engine.instmem.finish_access(dev);
237
238         return 0;
239 }
240
241 void
242 nv50_graph_destroy_context(struct nouveau_channel *chan)
243 {
244         struct drm_device *dev = chan->dev;
245         struct drm_nouveau_private *dev_priv = dev->dev_private;
246         int i, hdr = IS_G80 ? 0x200 : 0x20;
247
248         NV_DEBUG(dev, "ch%d\n", chan->id);
249
250         if (!chan->ramin || !chan->ramin->gpuobj)
251                 return;
252
253         dev_priv->engine.instmem.prepare_access(dev, true);
254         for (i = hdr; i < hdr + 24; i += 4)
255                 nv_wo32(dev, chan->ramin->gpuobj, i/4, 0);
256         dev_priv->engine.instmem.finish_access(dev);
257
258         nouveau_gpuobj_ref_del(dev, &chan->ramin_grctx);
259 }
260
261 static int
262 nv50_graph_do_load_context(struct drm_device *dev, uint32_t inst)
263 {
264         uint32_t fifo = nv_rd32(dev, 0x400500);
265
266         nv_wr32(dev, 0x400500, fifo & ~1);
267         nv_wr32(dev, 0x400784, inst);
268         nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x40);
269         nv_wr32(dev, 0x400320, nv_rd32(dev, 0x400320) | 0x11);
270         nv_wr32(dev, 0x400040, 0xffffffff);
271         (void)nv_rd32(dev, 0x400040);
272         nv_wr32(dev, 0x400040, 0x00000000);
273         nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 1);
274
275         if (nouveau_wait_for_idle(dev))
276                 nv_wr32(dev, 0x40032c, inst | (1<<31));
277         nv_wr32(dev, 0x400500, fifo);
278
279         return 0;
280 }
281
282 int
283 nv50_graph_load_context(struct nouveau_channel *chan)
284 {
285         uint32_t inst = chan->ramin->instance >> 12;
286
287         NV_DEBUG(chan->dev, "ch%d\n", chan->id);
288         return nv50_graph_do_load_context(chan->dev, inst);
289 }
290
291 int
292 nv50_graph_unload_context(struct drm_device *dev)
293 {
294         uint32_t inst;
295
296         inst  = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
297         if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
298                 return 0;
299         inst &= NV50_PGRAPH_CTXCTL_CUR_INSTANCE;
300
301         nouveau_wait_for_idle(dev);
302         nv_wr32(dev, 0x400784, inst);
303         nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x20);
304         nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 0x01);
305         nouveau_wait_for_idle(dev);
306
307         nv_wr32(dev, NV50_PGRAPH_CTXCTL_CUR, inst);
308         return 0;
309 }
310
311 void
312 nv50_graph_context_switch(struct drm_device *dev)
313 {
314         uint32_t inst;
315
316         nv50_graph_unload_context(dev);
317
318         inst  = nv_rd32(dev, NV50_PGRAPH_CTXCTL_NEXT);
319         inst &= NV50_PGRAPH_CTXCTL_NEXT_INSTANCE;
320         nv50_graph_do_load_context(dev, inst);
321
322         nv_wr32(dev, NV40_PGRAPH_INTR_EN, nv_rd32(dev,
323                 NV40_PGRAPH_INTR_EN) | NV_PGRAPH_INTR_CONTEXT_SWITCH);
324 }
325
326 static int
327 nv50_graph_nvsw_dma_vblsem(struct nouveau_channel *chan, int grclass,
328                            int mthd, uint32_t data)
329 {
330         struct nouveau_gpuobj_ref *ref = NULL;
331
332         if (nouveau_gpuobj_ref_find(chan, data, &ref))
333                 return -ENOENT;
334
335         if (nouveau_notifier_offset(ref->gpuobj, NULL))
336                 return -EINVAL;
337
338         chan->nvsw.vblsem = ref->gpuobj;
339         chan->nvsw.vblsem_offset = ~0;
340         return 0;
341 }
342
343 static int
344 nv50_graph_nvsw_vblsem_offset(struct nouveau_channel *chan, int grclass,
345                               int mthd, uint32_t data)
346 {
347         if (nouveau_notifier_offset(chan->nvsw.vblsem, &data))
348                 return -ERANGE;
349
350         chan->nvsw.vblsem_offset = data >> 2;
351         return 0;
352 }
353
354 static int
355 nv50_graph_nvsw_vblsem_release_val(struct nouveau_channel *chan, int grclass,
356                                    int mthd, uint32_t data)
357 {
358         chan->nvsw.vblsem_rval = data;
359         return 0;
360 }
361
362 static int
363 nv50_graph_nvsw_vblsem_release(struct nouveau_channel *chan, int grclass,
364                                int mthd, uint32_t data)
365 {
366         struct drm_device *dev = chan->dev;
367         struct drm_nouveau_private *dev_priv = dev->dev_private;
368
369         if (!chan->nvsw.vblsem || chan->nvsw.vblsem_offset == ~0 || data > 1)
370                 return -EINVAL;
371
372         if (!(nv_rd32(dev, NV50_PDISPLAY_INTR_EN) &
373                       NV50_PDISPLAY_INTR_EN_VBLANK_CRTC_(data))) {
374                 nv_wr32(dev, NV50_PDISPLAY_INTR_1,
375                         NV50_PDISPLAY_INTR_1_VBLANK_CRTC_(data));
376                 nv_wr32(dev, NV50_PDISPLAY_INTR_EN, nv_rd32(dev,
377                         NV50_PDISPLAY_INTR_EN) |
378                         NV50_PDISPLAY_INTR_EN_VBLANK_CRTC_(data));
379         }
380
381         list_add(&chan->nvsw.vbl_wait, &dev_priv->vbl_waiting);
382         return 0;
383 }
384
385 static struct nouveau_pgraph_object_method nv50_graph_nvsw_methods[] = {
386         { 0x018c, nv50_graph_nvsw_dma_vblsem },
387         { 0x0400, nv50_graph_nvsw_vblsem_offset },
388         { 0x0404, nv50_graph_nvsw_vblsem_release_val },
389         { 0x0408, nv50_graph_nvsw_vblsem_release },
390         {}
391 };
392
393 struct nouveau_pgraph_object_class nv50_graph_grclass[] = {
394         { 0x506e, true, nv50_graph_nvsw_methods }, /* nvsw */
395         { 0x0030, false, NULL }, /* null */
396         { 0x5039, false, NULL }, /* m2mf */
397         { 0x502d, false, NULL }, /* 2d */
398         { 0x50c0, false, NULL }, /* compute */
399         { 0x5097, false, NULL }, /* tesla (nv50) */
400         { 0x8297, false, NULL }, /* tesla (nv80/nv90) */
401         { 0x8397, false, NULL }, /* tesla (nva0) */
402         { 0x8597, false, NULL }, /* tesla (nva8) */
403         {}
404 };