summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/svga/svga_resource_buffer_upload.c
blob: 3de5216a949324d1d4fc5b7fc66730643bc415b6 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
/**********************************************************
 * Copyright 2008-2009 VMware, Inc.  All rights reserved.
 *
 * Permission is hereby granted, free of charge, to any person
 * obtaining a copy of this software and associated documentation
 * files (the "Software"), to deal in the Software without
 * restriction, including without limitation the rights to use, copy,
 * modify, merge, publish, distribute, sublicense, and/or sell copies
 * of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be
 * included in all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 **********************************************************/

#include "svga_cmd.h"

#include "pipe/p_state.h"
#include "pipe/p_defines.h"
#include "util/u_inlines.h"
#include "os/os_thread.h"
#include "util/u_math.h"
#include "util/u_memory.h"

#include "svga_context.h"
#include "svga_screen.h"
#include "svga_resource_buffer.h"
#include "svga_resource_buffer_upload.h"
#include "svga_winsys.h"
#include "svga_debug.h"


/**
 * Allocate a winsys_buffer (ie. DMA, aka GMR memory).
 *
 * It will flush and retry in case the first attempt to create a DMA buffer
 * fails, so it should not be called from any function involved in flushing
 * to avoid recursion.
 */
struct svga_winsys_buffer *
svga_winsys_buffer_create( struct svga_context *svga,
                           unsigned alignment, 
                           unsigned usage,
                           unsigned size )
{
   struct svga_screen *svgascreen = svga_screen(svga->pipe.screen);
   struct svga_winsys_screen *sws = svgascreen->sws;
   struct svga_winsys_buffer *buf;
   
   /* Just try */
   buf = sws->buffer_create(sws, alignment, usage, size);
   if(!buf) {

      SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "flushing screen to find %d bytes GMR\n", 
               size); 
      
      /* Try flushing all pending DMAs */
      svga_context_flush(svga, NULL);
      buf = sws->buffer_create(sws, alignment, usage, size);
   }
   
   return buf;
}


void
svga_buffer_destroy_hw_storage(struct svga_screen *ss, struct svga_buffer *sbuf)
{
   struct svga_winsys_screen *sws = ss->sws;

   assert(!sbuf->map.count);
   assert(sbuf->hwbuf);
   if(sbuf->hwbuf) {
      sws->buffer_destroy(sws, sbuf->hwbuf);
      sbuf->hwbuf = NULL;
   }
}



/**
 * Allocate DMA'ble storage for the buffer. 
 * 
 * Called before mapping a buffer.
 */
enum pipe_error
svga_buffer_create_hw_storage(struct svga_screen *ss,
                              struct svga_buffer *sbuf)
{
   assert(!sbuf->user);

   if(!sbuf->hwbuf) {
      struct svga_winsys_screen *sws = ss->sws;
      unsigned alignment = 16;
      unsigned usage = 0;
      unsigned size = sbuf->b.b.width0;
      
      sbuf->hwbuf = sws->buffer_create(sws, alignment, usage, size);
      if(!sbuf->hwbuf)
         return PIPE_ERROR_OUT_OF_MEMORY;
      
      assert(!sbuf->dma.pending);
   }
   
   return PIPE_OK;
}



enum pipe_error
svga_buffer_create_host_surface(struct svga_screen *ss,
                                struct svga_buffer *sbuf)
{
   if(!sbuf->handle) {
      sbuf->key.flags = 0;
      
      sbuf->key.format = SVGA3D_BUFFER;
      if(sbuf->b.b.bind & PIPE_BIND_VERTEX_BUFFER)
         sbuf->key.flags |= SVGA3D_SURFACE_HINT_VERTEXBUFFER;
      if(sbuf->b.b.bind & PIPE_BIND_INDEX_BUFFER)
         sbuf->key.flags |= SVGA3D_SURFACE_HINT_INDEXBUFFER;
      
      sbuf->key.size.width = sbuf->b.b.width0;
      sbuf->key.size.height = 1;
      sbuf->key.size.depth = 1;
      
      sbuf->key.numFaces = 1;
      sbuf->key.numMipLevels = 1;
      sbuf->key.cachable = 1;
      
      SVGA_DBG(DEBUG_DMA, "surface_create for buffer sz %d\n", sbuf->b.b.width0);

      sbuf->handle = svga_screen_surface_create(ss, &sbuf->key);
      if(!sbuf->handle)
         return PIPE_ERROR_OUT_OF_MEMORY;
   
      /* Always set the discard flag on the first time the buffer is written
       * as svga_screen_surface_create might have passed a recycled host
       * buffer.
       */
      sbuf->dma.flags.discard = TRUE;

      SVGA_DBG(DEBUG_DMA, "   --> got sid %p sz %d (buffer)\n", sbuf->handle, sbuf->b.b.width0);
   }
   
   return PIPE_OK;
}   


void
svga_buffer_destroy_host_surface(struct svga_screen *ss,
                                 struct svga_buffer *sbuf)
{
   if(sbuf->handle) {
      SVGA_DBG(DEBUG_DMA, " ungrab sid %p sz %d\n", sbuf->handle, sbuf->b.b.width0);
      svga_screen_surface_destroy(ss, &sbuf->key, &sbuf->handle);
   }
}   


/**
 * Variant of SVGA3D_BufferDMA which leaves the copy box temporarily in blank.
 */
static enum pipe_error
svga_buffer_upload_command(struct svga_context *svga,
                           struct svga_buffer *sbuf)
{
   struct svga_winsys_context *swc = svga->swc;
   struct svga_winsys_buffer *guest = sbuf->hwbuf;
   struct svga_winsys_surface *host = sbuf->handle;
   SVGA3dTransferType transfer = SVGA3D_WRITE_HOST_VRAM;
   SVGA3dCmdSurfaceDMA *cmd;
   uint32 numBoxes = sbuf->map.num_ranges;
   SVGA3dCopyBox *boxes;
   SVGA3dCmdSurfaceDMASuffix *pSuffix;
   unsigned region_flags;
   unsigned surface_flags;
   struct pipe_resource *dummy;

   if(transfer == SVGA3D_WRITE_HOST_VRAM) {
      region_flags = SVGA_RELOC_READ;
      surface_flags = SVGA_RELOC_WRITE;
   }
   else if(transfer == SVGA3D_READ_HOST_VRAM) {
      region_flags = SVGA_RELOC_WRITE;
      surface_flags = SVGA_RELOC_READ;
   }
   else {
      assert(0);
      return PIPE_ERROR_BAD_INPUT;
   }

   assert(numBoxes);

   cmd = SVGA3D_FIFOReserve(swc,
                            SVGA_3D_CMD_SURFACE_DMA,
                            sizeof *cmd + numBoxes * sizeof *boxes + sizeof *pSuffix,
                            2);
   if(!cmd)
      return PIPE_ERROR_OUT_OF_MEMORY;

   swc->region_relocation(swc, &cmd->guest.ptr, guest, 0, region_flags);
   cmd->guest.pitch = 0;

   swc->surface_relocation(swc, &cmd->host.sid, host, surface_flags);
   cmd->host.face = 0;
   cmd->host.mipmap = 0;

   cmd->transfer = transfer;

   sbuf->dma.boxes = (SVGA3dCopyBox *)&cmd[1];
   sbuf->dma.svga = svga;

   /* Increment reference count */
   dummy = NULL;
   pipe_resource_reference(&dummy, &sbuf->b.b);

   pSuffix = (SVGA3dCmdSurfaceDMASuffix *)((uint8_t*)cmd + sizeof *cmd + numBoxes * sizeof *boxes);
   pSuffix->suffixSize = sizeof *pSuffix;
   pSuffix->maximumOffset = sbuf->b.b.width0;
   pSuffix->flags = sbuf->dma.flags;

   SVGA_FIFOCommitAll(swc);

   sbuf->dma.flags.discard = FALSE;

   return PIPE_OK;
}


/**
 * Patch up the upload DMA command reserved by svga_buffer_upload_command
 * with the final ranges.
 */
static void
svga_buffer_upload_flush(struct svga_context *svga,
                         struct svga_buffer *sbuf)
{
   SVGA3dCopyBox *boxes;
   unsigned i;

   assert(sbuf->handle); 
   assert(sbuf->hwbuf);
   assert(sbuf->map.num_ranges);
   assert(sbuf->dma.svga == svga);
   assert(sbuf->dma.boxes);
   
   /*
    * Patch the DMA command with the final copy box.
    */

   SVGA_DBG(DEBUG_DMA, "dma to sid %p\n", sbuf->handle);

   boxes = sbuf->dma.boxes;
   for(i = 0; i < sbuf->map.num_ranges; ++i) {
      SVGA_DBG(DEBUG_DMA, "  bytes %u - %u\n",
               sbuf->map.ranges[i].start, sbuf->map.ranges[i].end);

      boxes[i].x = sbuf->map.ranges[i].start;
      boxes[i].y = 0;
      boxes[i].z = 0;
      boxes[i].w = sbuf->map.ranges[i].end - sbuf->map.ranges[i].start;
      boxes[i].h = 1;
      boxes[i].d = 1;
      boxes[i].srcx = sbuf->map.ranges[i].start;
      boxes[i].srcy = 0;
      boxes[i].srcz = 0;
   }

   sbuf->map.num_ranges = 0;

   assert(sbuf->head.prev && sbuf->head.next);
   LIST_DEL(&sbuf->head);
#ifdef DEBUG
   sbuf->head.next = sbuf->head.prev = NULL; 
#endif
   sbuf->dma.pending = FALSE;

   sbuf->dma.svga = NULL;
   sbuf->dma.boxes = NULL;

   /* Decrement reference count */
   pipe_reference(&(sbuf->b.b.reference), NULL);
   sbuf = NULL;
}



/**
 * Note a dirty range.
 *
 * This function only notes the range down. It doesn't actually emit a DMA
 * upload command. That only happens when a context tries to refer to this
 * buffer, and the DMA upload command is added to that context's command buffer.
 * 
 * We try to lump as many contiguous DMA transfers together as possible.
 */
void
svga_buffer_add_range(struct svga_buffer *sbuf,
                      unsigned start,
                      unsigned end)
{
   unsigned i;
   unsigned nearest_range;
   unsigned nearest_dist;

   assert(end > start);
   
   if (sbuf->map.num_ranges < SVGA_BUFFER_MAX_RANGES) {
      nearest_range = sbuf->map.num_ranges;
      nearest_dist = ~0;
   } else {
      nearest_range = SVGA_BUFFER_MAX_RANGES - 1;
      nearest_dist = 0;
   }

   /*
    * Try to grow one of the ranges.
    *
    * Note that it is not this function task to care about overlapping ranges,
    * as the GMR was already given so it is too late to do anything. Situations
    * where overlapping ranges may pose a problem should be detected via
    * pipe_context::is_resource_referenced and the context that refers to the
    * buffer should be flushed.
    */

   for(i = 0; i < sbuf->map.num_ranges; ++i) {
      int left_dist;
      int right_dist;
      int dist;

      left_dist = start - sbuf->map.ranges[i].end;
      right_dist = sbuf->map.ranges[i].start - end;
      dist = MAX2(left_dist, right_dist);

      if (dist <= 0) {
         /*
          * Ranges are contiguous or overlapping -- extend this one and return.
          */

         sbuf->map.ranges[i].start = MIN2(sbuf->map.ranges[i].start, start);
         sbuf->map.ranges[i].end   = MAX2(sbuf->map.ranges[i].end,   end);
         return;
      }
      else {
         /*
          * Discontiguous ranges -- keep track of the nearest range.
          */

         if (dist < nearest_dist) {
            nearest_range = i;
            nearest_dist = dist;
         }
      }
   }

   /*
    * We cannot add a new range to an existing DMA command, so patch-up the
    * pending DMA upload and start clean.
    */

   if(sbuf->dma.pending)
      svga_buffer_upload_flush(sbuf->dma.svga, sbuf);

   assert(!sbuf->dma.pending);
   assert(!sbuf->dma.svga);
   assert(!sbuf->dma.boxes);

   if (sbuf->map.num_ranges < SVGA_BUFFER_MAX_RANGES) {
      /*
       * Add a new range.
       */

      sbuf->map.ranges[sbuf->map.num_ranges].start = start;
      sbuf->map.ranges[sbuf->map.num_ranges].end = end;
      ++sbuf->map.num_ranges;
   } else {
      /*
       * Everything else failed, so just extend the nearest range.
       *
       * It is OK to do this because we always keep a local copy of the
       * host buffer data, for SW TNL, and the host never modifies the buffer.
       */

      assert(nearest_range < SVGA_BUFFER_MAX_RANGES);
      assert(nearest_range < sbuf->map.num_ranges);
      sbuf->map.ranges[nearest_range].start = MIN2(sbuf->map.ranges[nearest_range].start, start);
      sbuf->map.ranges[nearest_range].end   = MAX2(sbuf->map.ranges[nearest_range].end,   end);
   }
}



/**
 * Copy the contents of the malloc buffer to a hardware buffer.
 */
static INLINE enum pipe_error
svga_buffer_update_hw(struct svga_screen *ss, struct svga_buffer *sbuf)
{
   assert(!sbuf->user);
   if(!sbuf->hwbuf) {
      enum pipe_error ret;
      void *map;
      
      assert(sbuf->swbuf);
      if(!sbuf->swbuf)
         return PIPE_ERROR;
      
      ret = svga_buffer_create_hw_storage(ss, sbuf);
      if(ret != PIPE_OK)
         return ret;

      pipe_mutex_lock(ss->swc_mutex);
      map = ss->sws->buffer_map(ss->sws, sbuf->hwbuf, PIPE_TRANSFER_WRITE);
      assert(map);
      if(!map) {
	 pipe_mutex_unlock(ss->swc_mutex);
         svga_buffer_destroy_hw_storage(ss, sbuf);
         return PIPE_ERROR;
      }

      memcpy(map, sbuf->swbuf, sbuf->b.b.width0);
      ss->sws->buffer_unmap(ss->sws, sbuf->hwbuf);

      /* This user/malloc buffer is now indistinguishable from a gpu buffer */
      assert(!sbuf->map.count);
      if(!sbuf->map.count) {
         if(sbuf->user)
            sbuf->user = FALSE;
         else
            align_free(sbuf->swbuf);
         sbuf->swbuf = NULL;
      }
      
      pipe_mutex_unlock(ss->swc_mutex);
   }
   
   return PIPE_OK;
}


/**
 * Upload the buffer to the host in a piecewise fashion.
 *
 * Used when the buffer is too big to fit in the GMR aperture.
 */
static INLINE enum pipe_error
svga_buffer_upload_piecewise(struct svga_screen *ss,
                             struct svga_context *svga,
                             struct svga_buffer *sbuf)
{
   struct svga_winsys_screen *sws = ss->sws;
   const unsigned alignment = sizeof(void *);
   const unsigned usage = 0;
   unsigned i;

   assert(sbuf->map.num_ranges);
   assert(!sbuf->dma.pending);

   SVGA_DBG(DEBUG_DMA, "dma to sid %p\n", sbuf->handle);

   for (i = 0; i < sbuf->map.num_ranges; ++i) {
      struct svga_buffer_range *range = &sbuf->map.ranges[i];
      unsigned offset = range->start;
      unsigned size = range->end - range->start;

      while (offset < range->end) {
         struct svga_winsys_buffer *hwbuf;
         uint8_t *map;
         enum pipe_error ret;

         if (offset + size > range->end)
            size = range->end - offset;

         hwbuf = sws->buffer_create(sws, alignment, usage, size);
         while (!hwbuf) {
            size /= 2;
            if (!size)
               return PIPE_ERROR_OUT_OF_MEMORY;
            hwbuf = sws->buffer_create(sws, alignment, usage, size);
         }

         SVGA_DBG(DEBUG_DMA, "  bytes %u - %u\n",
                  offset, offset + size);

         map = sws->buffer_map(sws, hwbuf,
                               PIPE_TRANSFER_WRITE |
                               PIPE_TRANSFER_DISCARD);
         assert(map);
         if (map) {
            memcpy(map, sbuf->swbuf, size);
            sws->buffer_unmap(sws, hwbuf);
         }

         ret = SVGA3D_BufferDMA(svga->swc,
                                hwbuf, sbuf->handle,
                                SVGA3D_WRITE_HOST_VRAM,
                                size, 0, offset, sbuf->dma.flags);
         if(ret != PIPE_OK) {
            svga_context_flush(svga, NULL);
            ret =  SVGA3D_BufferDMA(svga->swc,
                                    hwbuf, sbuf->handle,
                                    SVGA3D_WRITE_HOST_VRAM,
                                    size, 0, offset, sbuf->dma.flags);
            assert(ret == PIPE_OK);
         }

         sbuf->dma.flags.discard = FALSE;

         sws->buffer_destroy(sws, hwbuf);

         offset += size;
      }
   }

   sbuf->map.num_ranges = 0;

   return PIPE_OK;
}




/* Get (or create/upload) the winsys surface handle so that we can
 * refer to this buffer in fifo commands.
 */
struct svga_winsys_surface *
svga_buffer_handle(struct svga_context *svga,
                   struct pipe_resource *buf)
{
   struct pipe_screen *screen = svga->pipe.screen;
   struct svga_screen *ss = svga_screen(screen);
   struct svga_buffer *sbuf;
   enum pipe_error ret;

   if(!buf)
      return NULL;

   sbuf = svga_buffer(buf);
   
   assert(!sbuf->map.count);
   assert(!sbuf->user);
   
   if(!sbuf->handle) {
      ret = svga_buffer_create_host_surface(ss, sbuf);
      if(ret != PIPE_OK)
	 return NULL;
   }

   assert(sbuf->handle);

   if (sbuf->map.num_ranges) {
      if (!sbuf->dma.pending) {
         /*
          * No pending DMA upload yet, so insert a DMA upload command now.
          */

         /*
          * Migrate the data from swbuf -> hwbuf if necessary.
          */
         ret = svga_buffer_update_hw(ss, sbuf);
         if (ret == PIPE_OK) {
            /*
             * Queue a dma command.
             */

            ret = svga_buffer_upload_command(svga, sbuf);
            if (ret == PIPE_ERROR_OUT_OF_MEMORY) {
               svga_context_flush(svga, NULL);
               ret = svga_buffer_upload_command(svga, sbuf);
               assert(ret == PIPE_OK);
            }
            if (ret == PIPE_OK) {
               sbuf->dma.pending = TRUE;
               assert(!sbuf->head.prev && !sbuf->head.next);
               LIST_ADDTAIL(&sbuf->head, &svga->dirty_buffers);
            }
         }
         else if (ret == PIPE_ERROR_OUT_OF_MEMORY) {
            /*
             * The buffer is too big to fit in the GMR aperture, so break it in
             * smaller pieces.
             */
            ret = svga_buffer_upload_piecewise(ss, svga, sbuf);
         }

         if (ret != PIPE_OK) {
            /*
             * Something unexpected happened above. There is very little that
             * we can do other than proceeding while ignoring the dirty ranges.
             */
            assert(0);
            sbuf->map.num_ranges = 0;
         }
      }
      else {
         /*
          * There a pending dma already. Make sure it is from this context.
          */
         assert(sbuf->dma.svga == svga);
      }
   }

   assert(!sbuf->map.num_ranges || sbuf->dma.pending);

   return sbuf->handle;
}



void
svga_context_flush_buffers(struct svga_context *svga)
{
   struct list_head *curr, *next;
   struct svga_buffer *sbuf;

   curr = svga->dirty_buffers.next;
   next = curr->next;
   while(curr != &svga->dirty_buffers) {
      sbuf = LIST_ENTRY(struct svga_buffer, curr, head);

      assert(p_atomic_read(&sbuf->b.b.reference.count) != 0);
      assert(sbuf->dma.pending);
      
      svga_buffer_upload_flush(svga, sbuf);

      curr = next; 
      next = curr->next;
   }
}