Merge branch 'stable-3.2' into pandora-3.2
[pandora-kernel.git] / drivers / gpu / drm / vmwgfx / vmwgfx_fifo.c
1 /**************************************************************************
2  *
3  * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA
4  * All Rights Reserved.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the
8  * "Software"), to deal in the Software without restriction, including
9  * without limitation the rights to use, copy, modify, merge, publish,
10  * distribute, sub license, and/or sell copies of the Software, and to
11  * permit persons to whom the Software is furnished to do so, subject to
12  * the following conditions:
13  *
14  * The above copyright notice and this permission notice (including the
15  * next paragraph) shall be included in all copies or substantial portions
16  * of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
22  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24  * USE OR OTHER DEALINGS IN THE SOFTWARE.
25  *
26  **************************************************************************/
27
28 #include "vmwgfx_drv.h"
29 #include "drmP.h"
30 #include "ttm/ttm_placement.h"
31
32 bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
33 {
34         __le32 __iomem *fifo_mem = dev_priv->mmio_virt;
35         uint32_t fifo_min, hwversion;
36         const struct vmw_fifo_state *fifo = &dev_priv->fifo;
37
38         if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
39                 return false;
40
41         fifo_min = ioread32(fifo_mem  + SVGA_FIFO_MIN);
42         if (fifo_min <= SVGA_FIFO_3D_HWVERSION * sizeof(unsigned int))
43                 return false;
44
45         hwversion = ioread32(fifo_mem +
46                              ((fifo->capabilities &
47                                SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ?
48                               SVGA_FIFO_3D_HWVERSION_REVISED :
49                               SVGA_FIFO_3D_HWVERSION));
50
51         if (hwversion == 0)
52                 return false;
53
54         if (hwversion < SVGA3D_HWVERSION_WS8_B1)
55                 return false;
56
57         /* Non-Screen Object path does not support surfaces */
58         if (!dev_priv->sou_priv)
59                 return false;
60
61         return true;
62 }
63
64 bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv)
65 {
66         __le32 __iomem *fifo_mem = dev_priv->mmio_virt;
67         uint32_t caps;
68
69         if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
70                 return false;
71
72         caps = ioread32(fifo_mem + SVGA_FIFO_CAPABILITIES);
73         if (caps & SVGA_FIFO_CAP_PITCHLOCK)
74                 return true;
75
76         return false;
77 }
78
79 int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
80 {
81         __le32 __iomem *fifo_mem = dev_priv->mmio_virt;
82         uint32_t max;
83         uint32_t min;
84         uint32_t dummy;
85
86         fifo->static_buffer_size = VMWGFX_FIFO_STATIC_SIZE;
87         fifo->static_buffer = vmalloc(fifo->static_buffer_size);
88         if (unlikely(fifo->static_buffer == NULL))
89                 return -ENOMEM;
90
91         fifo->dynamic_buffer = NULL;
92         fifo->reserved_size = 0;
93         fifo->using_bounce_buffer = false;
94
95         mutex_init(&fifo->fifo_mutex);
96         init_rwsem(&fifo->rwsem);
97
98         /*
99          * Allow mapping the first page read-only to user-space.
100          */
101
102         DRM_INFO("width %d\n", vmw_read(dev_priv, SVGA_REG_WIDTH));
103         DRM_INFO("height %d\n", vmw_read(dev_priv, SVGA_REG_HEIGHT));
104         DRM_INFO("bpp %d\n", vmw_read(dev_priv, SVGA_REG_BITS_PER_PIXEL));
105
106         mutex_lock(&dev_priv->hw_mutex);
107         dev_priv->enable_state = vmw_read(dev_priv, SVGA_REG_ENABLE);
108         dev_priv->config_done_state = vmw_read(dev_priv, SVGA_REG_CONFIG_DONE);
109         dev_priv->traces_state = vmw_read(dev_priv, SVGA_REG_TRACES);
110         vmw_write(dev_priv, SVGA_REG_ENABLE, 1);
111
112         min = 4;
113         if (dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO)
114                 min = vmw_read(dev_priv, SVGA_REG_MEM_REGS);
115         min <<= 2;
116
117         if (min < PAGE_SIZE)
118                 min = PAGE_SIZE;
119
120         iowrite32(min, fifo_mem + SVGA_FIFO_MIN);
121         iowrite32(dev_priv->mmio_size, fifo_mem + SVGA_FIFO_MAX);
122         wmb();
123         iowrite32(min,  fifo_mem + SVGA_FIFO_NEXT_CMD);
124         iowrite32(min,  fifo_mem + SVGA_FIFO_STOP);
125         iowrite32(0, fifo_mem + SVGA_FIFO_BUSY);
126         mb();
127
128         vmw_write(dev_priv, SVGA_REG_CONFIG_DONE, 1);
129         mutex_unlock(&dev_priv->hw_mutex);
130
131         max = ioread32(fifo_mem + SVGA_FIFO_MAX);
132         min = ioread32(fifo_mem  + SVGA_FIFO_MIN);
133         fifo->capabilities = ioread32(fifo_mem + SVGA_FIFO_CAPABILITIES);
134
135         DRM_INFO("Fifo max 0x%08x min 0x%08x cap 0x%08x\n",
136                  (unsigned int) max,
137                  (unsigned int) min,
138                  (unsigned int) fifo->capabilities);
139
140         atomic_set(&dev_priv->marker_seq, dev_priv->last_read_seqno);
141         iowrite32(dev_priv->last_read_seqno, fifo_mem + SVGA_FIFO_FENCE);
142         vmw_marker_queue_init(&fifo->marker_queue);
143         return vmw_fifo_send_fence(dev_priv, &dummy);
144 }
145
146 void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason)
147 {
148         __le32 __iomem *fifo_mem = dev_priv->mmio_virt;
149
150         mutex_lock(&dev_priv->hw_mutex);
151
152         if (unlikely(ioread32(fifo_mem + SVGA_FIFO_BUSY) == 0)) {
153                 iowrite32(1, fifo_mem + SVGA_FIFO_BUSY);
154                 vmw_write(dev_priv, SVGA_REG_SYNC, reason);
155         }
156
157         mutex_unlock(&dev_priv->hw_mutex);
158 }
159
160 void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
161 {
162         __le32 __iomem *fifo_mem = dev_priv->mmio_virt;
163
164         mutex_lock(&dev_priv->hw_mutex);
165
166         vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC);
167         while (vmw_read(dev_priv, SVGA_REG_BUSY) != 0)
168                 ;
169
170         dev_priv->last_read_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
171
172         vmw_write(dev_priv, SVGA_REG_CONFIG_DONE,
173                   dev_priv->config_done_state);
174         vmw_write(dev_priv, SVGA_REG_ENABLE,
175                   dev_priv->enable_state);
176         vmw_write(dev_priv, SVGA_REG_TRACES,
177                   dev_priv->traces_state);
178
179         mutex_unlock(&dev_priv->hw_mutex);
180         vmw_marker_queue_takedown(&fifo->marker_queue);
181
182         if (likely(fifo->static_buffer != NULL)) {
183                 vfree(fifo->static_buffer);
184                 fifo->static_buffer = NULL;
185         }
186
187         if (likely(fifo->dynamic_buffer != NULL)) {
188                 vfree(fifo->dynamic_buffer);
189                 fifo->dynamic_buffer = NULL;
190         }
191 }
192
193 static bool vmw_fifo_is_full(struct vmw_private *dev_priv, uint32_t bytes)
194 {
195         __le32 __iomem *fifo_mem = dev_priv->mmio_virt;
196         uint32_t max = ioread32(fifo_mem + SVGA_FIFO_MAX);
197         uint32_t next_cmd = ioread32(fifo_mem + SVGA_FIFO_NEXT_CMD);
198         uint32_t min = ioread32(fifo_mem + SVGA_FIFO_MIN);
199         uint32_t stop = ioread32(fifo_mem + SVGA_FIFO_STOP);
200
201         return ((max - next_cmd) + (stop - min) <= bytes);
202 }
203
204 static int vmw_fifo_wait_noirq(struct vmw_private *dev_priv,
205                                uint32_t bytes, bool interruptible,
206                                unsigned long timeout)
207 {
208         int ret = 0;
209         unsigned long end_jiffies = jiffies + timeout;
210         DEFINE_WAIT(__wait);
211
212         DRM_INFO("Fifo wait noirq.\n");
213
214         for (;;) {
215                 prepare_to_wait(&dev_priv->fifo_queue, &__wait,
216                                 (interruptible) ?
217                                 TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE);
218                 if (!vmw_fifo_is_full(dev_priv, bytes))
219                         break;
220                 if (time_after_eq(jiffies, end_jiffies)) {
221                         ret = -EBUSY;
222                         DRM_ERROR("SVGA device lockup.\n");
223                         break;
224                 }
225                 schedule_timeout(1);
226                 if (interruptible && signal_pending(current)) {
227                         ret = -ERESTARTSYS;
228                         break;
229                 }
230         }
231         finish_wait(&dev_priv->fifo_queue, &__wait);
232         wake_up_all(&dev_priv->fifo_queue);
233         DRM_INFO("Fifo noirq exit.\n");
234         return ret;
235 }
236
237 static int vmw_fifo_wait(struct vmw_private *dev_priv,
238                          uint32_t bytes, bool interruptible,
239                          unsigned long timeout)
240 {
241         long ret = 1L;
242         unsigned long irq_flags;
243
244         if (likely(!vmw_fifo_is_full(dev_priv, bytes)))
245                 return 0;
246
247         vmw_fifo_ping_host(dev_priv, SVGA_SYNC_FIFOFULL);
248         if (!(dev_priv->capabilities & SVGA_CAP_IRQMASK))
249                 return vmw_fifo_wait_noirq(dev_priv, bytes,
250                                            interruptible, timeout);
251
252         mutex_lock(&dev_priv->hw_mutex);
253         if (atomic_add_return(1, &dev_priv->fifo_queue_waiters) > 0) {
254                 spin_lock_irqsave(&dev_priv->irq_lock, irq_flags);
255                 outl(SVGA_IRQFLAG_FIFO_PROGRESS,
256                      dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
257                 dev_priv->irq_mask |= SVGA_IRQFLAG_FIFO_PROGRESS;
258                 vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
259                 spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
260         }
261         mutex_unlock(&dev_priv->hw_mutex);
262
263         if (interruptible)
264                 ret = wait_event_interruptible_timeout
265                     (dev_priv->fifo_queue,
266                      !vmw_fifo_is_full(dev_priv, bytes), timeout);
267         else
268                 ret = wait_event_timeout
269                     (dev_priv->fifo_queue,
270                      !vmw_fifo_is_full(dev_priv, bytes), timeout);
271
272         if (unlikely(ret == 0))
273                 ret = -EBUSY;
274         else if (likely(ret > 0))
275                 ret = 0;
276
277         mutex_lock(&dev_priv->hw_mutex);
278         if (atomic_dec_and_test(&dev_priv->fifo_queue_waiters)) {
279                 spin_lock_irqsave(&dev_priv->irq_lock, irq_flags);
280                 dev_priv->irq_mask &= ~SVGA_IRQFLAG_FIFO_PROGRESS;
281                 vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
282                 spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
283         }
284         mutex_unlock(&dev_priv->hw_mutex);
285
286         return ret;
287 }
288
289 /**
290  * Reserve @bytes number of bytes in the fifo.
291  *
292  * This function will return NULL (error) on two conditions:
293  *  If it timeouts waiting for fifo space, or if @bytes is larger than the
294  *   available fifo space.
295  *
296  * Returns:
297  *   Pointer to the fifo, or null on error (possible hardware hang).
298  */
299 void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
300 {
301         struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
302         __le32 __iomem *fifo_mem = dev_priv->mmio_virt;
303         uint32_t max;
304         uint32_t min;
305         uint32_t next_cmd;
306         uint32_t reserveable = fifo_state->capabilities & SVGA_FIFO_CAP_RESERVE;
307         int ret;
308
309         mutex_lock(&fifo_state->fifo_mutex);
310         max = ioread32(fifo_mem + SVGA_FIFO_MAX);
311         min = ioread32(fifo_mem + SVGA_FIFO_MIN);
312         next_cmd = ioread32(fifo_mem + SVGA_FIFO_NEXT_CMD);
313
314         if (unlikely(bytes >= (max - min)))
315                 goto out_err;
316
317         BUG_ON(fifo_state->reserved_size != 0);
318         BUG_ON(fifo_state->dynamic_buffer != NULL);
319
320         fifo_state->reserved_size = bytes;
321
322         while (1) {
323                 uint32_t stop = ioread32(fifo_mem + SVGA_FIFO_STOP);
324                 bool need_bounce = false;
325                 bool reserve_in_place = false;
326
327                 if (next_cmd >= stop) {
328                         if (likely((next_cmd + bytes < max ||
329                                     (next_cmd + bytes == max && stop > min))))
330                                 reserve_in_place = true;
331
332                         else if (vmw_fifo_is_full(dev_priv, bytes)) {
333                                 ret = vmw_fifo_wait(dev_priv, bytes,
334                                                     false, 3 * HZ);
335                                 if (unlikely(ret != 0))
336                                         goto out_err;
337                         } else
338                                 need_bounce = true;
339
340                 } else {
341
342                         if (likely((next_cmd + bytes < stop)))
343                                 reserve_in_place = true;
344                         else {
345                                 ret = vmw_fifo_wait(dev_priv, bytes,
346                                                     false, 3 * HZ);
347                                 if (unlikely(ret != 0))
348                                         goto out_err;
349                         }
350                 }
351
352                 if (reserve_in_place) {
353                         if (reserveable || bytes <= sizeof(uint32_t)) {
354                                 fifo_state->using_bounce_buffer = false;
355
356                                 if (reserveable)
357                                         iowrite32(bytes, fifo_mem +
358                                                   SVGA_FIFO_RESERVED);
359                                 return fifo_mem + (next_cmd >> 2);
360                         } else {
361                                 need_bounce = true;
362                         }
363                 }
364
365                 if (need_bounce) {
366                         fifo_state->using_bounce_buffer = true;
367                         if (bytes < fifo_state->static_buffer_size)
368                                 return fifo_state->static_buffer;
369                         else {
370                                 fifo_state->dynamic_buffer = vmalloc(bytes);
371                                 if (!fifo_state->dynamic_buffer)
372                                         goto out_err;
373                                 return fifo_state->dynamic_buffer;
374                         }
375                 }
376         }
377 out_err:
378         fifo_state->reserved_size = 0;
379         mutex_unlock(&fifo_state->fifo_mutex);
380         return NULL;
381 }
382
383 static void vmw_fifo_res_copy(struct vmw_fifo_state *fifo_state,
384                               __le32 __iomem *fifo_mem,
385                               uint32_t next_cmd,
386                               uint32_t max, uint32_t min, uint32_t bytes)
387 {
388         uint32_t chunk_size = max - next_cmd;
389         uint32_t rest;
390         uint32_t *buffer = (fifo_state->dynamic_buffer != NULL) ?
391             fifo_state->dynamic_buffer : fifo_state->static_buffer;
392
393         if (bytes < chunk_size)
394                 chunk_size = bytes;
395
396         iowrite32(bytes, fifo_mem + SVGA_FIFO_RESERVED);
397         mb();
398         memcpy_toio(fifo_mem + (next_cmd >> 2), buffer, chunk_size);
399         rest = bytes - chunk_size;
400         if (rest)
401                 memcpy_toio(fifo_mem + (min >> 2), buffer + (chunk_size >> 2),
402                             rest);
403 }
404
405 static void vmw_fifo_slow_copy(struct vmw_fifo_state *fifo_state,
406                                __le32 __iomem *fifo_mem,
407                                uint32_t next_cmd,
408                                uint32_t max, uint32_t min, uint32_t bytes)
409 {
410         uint32_t *buffer = (fifo_state->dynamic_buffer != NULL) ?
411             fifo_state->dynamic_buffer : fifo_state->static_buffer;
412
413         while (bytes > 0) {
414                 iowrite32(*buffer++, fifo_mem + (next_cmd >> 2));
415                 next_cmd += sizeof(uint32_t);
416                 if (unlikely(next_cmd == max))
417                         next_cmd = min;
418                 mb();
419                 iowrite32(next_cmd, fifo_mem + SVGA_FIFO_NEXT_CMD);
420                 mb();
421                 bytes -= sizeof(uint32_t);
422         }
423 }
424
425 void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
426 {
427         struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
428         __le32 __iomem *fifo_mem = dev_priv->mmio_virt;
429         uint32_t next_cmd = ioread32(fifo_mem + SVGA_FIFO_NEXT_CMD);
430         uint32_t max = ioread32(fifo_mem + SVGA_FIFO_MAX);
431         uint32_t min = ioread32(fifo_mem + SVGA_FIFO_MIN);
432         bool reserveable = fifo_state->capabilities & SVGA_FIFO_CAP_RESERVE;
433
434         BUG_ON((bytes & 3) != 0);
435         BUG_ON(bytes > fifo_state->reserved_size);
436
437         fifo_state->reserved_size = 0;
438
439         if (fifo_state->using_bounce_buffer) {
440                 if (reserveable)
441                         vmw_fifo_res_copy(fifo_state, fifo_mem,
442                                           next_cmd, max, min, bytes);
443                 else
444                         vmw_fifo_slow_copy(fifo_state, fifo_mem,
445                                            next_cmd, max, min, bytes);
446
447                 if (fifo_state->dynamic_buffer) {
448                         vfree(fifo_state->dynamic_buffer);
449                         fifo_state->dynamic_buffer = NULL;
450                 }
451
452         }
453
454         down_write(&fifo_state->rwsem);
455         if (fifo_state->using_bounce_buffer || reserveable) {
456                 next_cmd += bytes;
457                 if (next_cmd >= max)
458                         next_cmd -= max - min;
459                 mb();
460                 iowrite32(next_cmd, fifo_mem + SVGA_FIFO_NEXT_CMD);
461         }
462
463         if (reserveable)
464                 iowrite32(0, fifo_mem + SVGA_FIFO_RESERVED);
465         mb();
466         up_write(&fifo_state->rwsem);
467         vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
468         mutex_unlock(&fifo_state->fifo_mutex);
469 }
470
471 int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
472 {
473         struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
474         struct svga_fifo_cmd_fence *cmd_fence;
475         void *fm;
476         int ret = 0;
477         uint32_t bytes = sizeof(__le32) + sizeof(*cmd_fence);
478
479         fm = vmw_fifo_reserve(dev_priv, bytes);
480         if (unlikely(fm == NULL)) {
481                 *seqno = atomic_read(&dev_priv->marker_seq);
482                 ret = -ENOMEM;
483                 (void)vmw_fallback_wait(dev_priv, false, true, *seqno,
484                                         false, 3*HZ);
485                 goto out_err;
486         }
487
488         do {
489                 *seqno = atomic_add_return(1, &dev_priv->marker_seq);
490         } while (*seqno == 0);
491
492         if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE)) {
493
494                 /*
495                  * Don't request hardware to send a fence. The
496                  * waiting code in vmwgfx_irq.c will emulate this.
497                  */
498
499                 vmw_fifo_commit(dev_priv, 0);
500                 return 0;
501         }
502
503         *(__le32 *) fm = cpu_to_le32(SVGA_CMD_FENCE);
504         cmd_fence = (struct svga_fifo_cmd_fence *)
505             ((unsigned long)fm + sizeof(__le32));
506
507         iowrite32(*seqno, &cmd_fence->fence);
508         vmw_fifo_commit(dev_priv, bytes);
509         (void) vmw_marker_push(&fifo_state->marker_queue, *seqno);
510         vmw_update_seqno(dev_priv, fifo_state);
511
512 out_err:
513         return ret;
514 }
515
516 /**
517  * vmw_fifo_emit_dummy_query - emits a dummy query to the fifo.
518  *
519  * @dev_priv: The device private structure.
520  * @cid: The hardware context id used for the query.
521  *
522  * This function is used to emit a dummy occlusion query with
523  * no primitives rendered between query begin and query end.
524  * It's used to provide a query barrier, in order to know that when
525  * this query is finished, all preceding queries are also finished.
526  *
527  * A Query results structure should have been initialized at the start
528  * of the dev_priv->dummy_query_bo buffer object. And that buffer object
529  * must also be either reserved or pinned when this function is called.
530  *
531  * Returns -ENOMEM on failure to reserve fifo space.
532  */
533 int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv,
534                               uint32_t cid)
535 {
536         /*
537          * A query wait without a preceding query end will
538          * actually finish all queries for this cid
539          * without writing to the query result structure.
540          */
541
542         struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
543         struct {
544                 SVGA3dCmdHeader header;
545                 SVGA3dCmdWaitForQuery body;
546         } *cmd;
547
548         cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
549
550         if (unlikely(cmd == NULL)) {
551                 DRM_ERROR("Out of fifo space for dummy query.\n");
552                 return -ENOMEM;
553         }
554
555         cmd->header.id = SVGA_3D_CMD_WAIT_FOR_QUERY;
556         cmd->header.size = sizeof(cmd->body);
557         cmd->body.cid = cid;
558         cmd->body.type = SVGA3D_QUERYTYPE_OCCLUSION;
559
560         if (bo->mem.mem_type == TTM_PL_VRAM) {
561                 cmd->body.guestResult.gmrId = SVGA_GMR_FRAMEBUFFER;
562                 cmd->body.guestResult.offset = bo->offset;
563         } else {
564                 cmd->body.guestResult.gmrId = bo->mem.start;
565                 cmd->body.guestResult.offset = 0;
566         }
567
568         vmw_fifo_commit(dev_priv, sizeof(*cmd));
569
570         return 0;
571 }