drm/i915/ringbuffer: Handle wrapping of the autoreported HEAD
[linux-2.6.git] / drivers / gpu / drm / i915 / intel_ringbuffer.c
1 /*
2  * Copyright © 2008-2010 Intel Corporation
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  *
23  * Authors:
24  *    Eric Anholt <eric@anholt.net>
25  *    Zou Nan hai <nanhai.zou@intel.com>
26  *    Xiang Hai hao<haihao.xiang@intel.com>
27  *
28  */
29
30 #include "drmP.h"
31 #include "drm.h"
32 #include "i915_drv.h"
33 #include "i915_drm.h"
34 #include "i915_trace.h"
35 #include "intel_drv.h"
36
37 static u32 i915_gem_get_seqno(struct drm_device *dev)
38 {
39         drm_i915_private_t *dev_priv = dev->dev_private;
40         u32 seqno;
41
42         seqno = dev_priv->next_seqno;
43
44         /* reserve 0 for non-seqno */
45         if (++dev_priv->next_seqno == 0)
46                 dev_priv->next_seqno = 1;
47
48         return seqno;
49 }
50
51 static void
52 render_ring_flush(struct drm_device *dev,
53                   struct intel_ring_buffer *ring,
54                   u32   invalidate_domains,
55                   u32   flush_domains)
56 {
57         drm_i915_private_t *dev_priv = dev->dev_private;
58         u32 cmd;
59
60 #if WATCH_EXEC
61         DRM_INFO("%s: invalidate %08x flush %08x\n", __func__,
62                   invalidate_domains, flush_domains);
63 #endif
64
65         trace_i915_gem_request_flush(dev, dev_priv->next_seqno,
66                                      invalidate_domains, flush_domains);
67
68         if ((invalidate_domains | flush_domains) & I915_GEM_GPU_DOMAINS) {
69                 /*
70                  * read/write caches:
71                  *
72                  * I915_GEM_DOMAIN_RENDER is always invalidated, but is
73                  * only flushed if MI_NO_WRITE_FLUSH is unset.  On 965, it is
74                  * also flushed at 2d versus 3d pipeline switches.
75                  *
76                  * read-only caches:
77                  *
78                  * I915_GEM_DOMAIN_SAMPLER is flushed on pre-965 if
79                  * MI_READ_FLUSH is set, and is always flushed on 965.
80                  *
81                  * I915_GEM_DOMAIN_COMMAND may not exist?
82                  *
83                  * I915_GEM_DOMAIN_INSTRUCTION, which exists on 965, is
84                  * invalidated when MI_EXE_FLUSH is set.
85                  *
86                  * I915_GEM_DOMAIN_VERTEX, which exists on 965, is
87                  * invalidated with every MI_FLUSH.
88                  *
89                  * TLBs:
90                  *
91                  * On 965, TLBs associated with I915_GEM_DOMAIN_COMMAND
92                  * and I915_GEM_DOMAIN_CPU in are invalidated at PTE write and
93                  * I915_GEM_DOMAIN_RENDER and I915_GEM_DOMAIN_SAMPLER
94                  * are flushed at any MI_FLUSH.
95                  */
96
97                 cmd = MI_FLUSH | MI_NO_WRITE_FLUSH;
98                 if ((invalidate_domains|flush_domains) &
99                     I915_GEM_DOMAIN_RENDER)
100                         cmd &= ~MI_NO_WRITE_FLUSH;
101                 if (INTEL_INFO(dev)->gen < 4) {
102                         /*
103                          * On the 965, the sampler cache always gets flushed
104                          * and this bit is reserved.
105                          */
106                         if (invalidate_domains & I915_GEM_DOMAIN_SAMPLER)
107                                 cmd |= MI_READ_FLUSH;
108                 }
109                 if (invalidate_domains & I915_GEM_DOMAIN_INSTRUCTION)
110                         cmd |= MI_EXE_FLUSH;
111
112 #if WATCH_EXEC
113                 DRM_INFO("%s: queue flush %08x to ring\n", __func__, cmd);
114 #endif
115                 intel_ring_begin(dev, ring, 2);
116                 intel_ring_emit(dev, ring, cmd);
117                 intel_ring_emit(dev, ring, MI_NOOP);
118                 intel_ring_advance(dev, ring);
119         }
120 }
121
122 static void ring_write_tail(struct drm_device *dev,
123                             struct intel_ring_buffer *ring,
124                             u32 value)
125 {
126         drm_i915_private_t *dev_priv = dev->dev_private;
127         I915_WRITE_TAIL(ring, value);
128 }
129
130 u32 intel_ring_get_active_head(struct drm_device *dev,
131                                struct intel_ring_buffer *ring)
132 {
133         drm_i915_private_t *dev_priv = dev->dev_private;
134         u32 acthd_reg = INTEL_INFO(dev)->gen >= 4 ?
135                         RING_ACTHD(ring->mmio_base) : ACTHD;
136
137         return I915_READ(acthd_reg);
138 }
139
140 static int init_ring_common(struct drm_device *dev,
141                             struct intel_ring_buffer *ring)
142 {
143         u32 head;
144         drm_i915_private_t *dev_priv = dev->dev_private;
145         struct drm_i915_gem_object *obj_priv;
146         obj_priv = to_intel_bo(ring->gem_object);
147
148         /* Stop the ring if it's running. */
149         I915_WRITE_CTL(ring, 0);
150         I915_WRITE_HEAD(ring, 0);
151         ring->write_tail(dev, ring, 0);
152
153         /* Initialize the ring. */
154         I915_WRITE_START(ring, obj_priv->gtt_offset);
155         head = I915_READ_HEAD(ring) & HEAD_ADDR;
156
157         /* G45 ring initialization fails to reset head to zero */
158         if (head != 0) {
159                 DRM_DEBUG_KMS("%s head not reset to zero "
160                               "ctl %08x head %08x tail %08x start %08x\n",
161                               ring->name,
162                               I915_READ_CTL(ring),
163                               I915_READ_HEAD(ring),
164                               I915_READ_TAIL(ring),
165                               I915_READ_START(ring));
166
167                 I915_WRITE_HEAD(ring, 0);
168
169                 if (I915_READ_HEAD(ring) & HEAD_ADDR) {
170                         DRM_ERROR("failed to set %s head to zero "
171                                   "ctl %08x head %08x tail %08x start %08x\n",
172                                   ring->name,
173                                   I915_READ_CTL(ring),
174                                   I915_READ_HEAD(ring),
175                                   I915_READ_TAIL(ring),
176                                   I915_READ_START(ring));
177                 }
178         }
179
180         I915_WRITE_CTL(ring,
181                         ((ring->gem_object->size - PAGE_SIZE) & RING_NR_PAGES)
182                         | RING_REPORT_64K | RING_VALID);
183
184         head = I915_READ_HEAD(ring) & HEAD_ADDR;
185         /* If the head is still not zero, the ring is dead */
186         if (head != 0) {
187                 DRM_ERROR("%s initialization failed "
188                                 "ctl %08x head %08x tail %08x start %08x\n",
189                                 ring->name,
190                                 I915_READ_CTL(ring),
191                                 I915_READ_HEAD(ring),
192                                 I915_READ_TAIL(ring),
193                                 I915_READ_START(ring));
194                 return -EIO;
195         }
196
197         if (!drm_core_check_feature(dev, DRIVER_MODESET))
198                 i915_kernel_lost_context(dev);
199         else {
200                 ring->head = I915_READ_HEAD(ring) & HEAD_ADDR;
201                 ring->tail = I915_READ_TAIL(ring) & TAIL_ADDR;
202                 ring->space = ring->head - (ring->tail + 8);
203                 if (ring->space < 0)
204                         ring->space += ring->size;
205         }
206         return 0;
207 }
208
209 static int init_render_ring(struct drm_device *dev,
210                             struct intel_ring_buffer *ring)
211 {
212         drm_i915_private_t *dev_priv = dev->dev_private;
213         int ret = init_ring_common(dev, ring);
214         int mode;
215
216         if (INTEL_INFO(dev)->gen > 3) {
217                 mode = VS_TIMER_DISPATCH << 16 | VS_TIMER_DISPATCH;
218                 if (IS_GEN6(dev))
219                         mode |= MI_FLUSH_ENABLE << 16 | MI_FLUSH_ENABLE;
220                 I915_WRITE(MI_MODE, mode);
221         }
222         return ret;
223 }
224
225 #define PIPE_CONTROL_FLUSH(addr)                                        \
226 do {                                                                    \
227         OUT_RING(GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |          \
228                  PIPE_CONTROL_DEPTH_STALL | 2);                         \
229         OUT_RING(addr | PIPE_CONTROL_GLOBAL_GTT);                       \
230         OUT_RING(0);                                                    \
231         OUT_RING(0);                                                    \
232 } while (0)
233
234 /**
235  * Creates a new sequence number, emitting a write of it to the status page
236  * plus an interrupt, which will trigger i915_user_interrupt_handler.
237  *
238  * Must be called with struct_lock held.
239  *
240  * Returned sequence numbers are nonzero on success.
241  */
242 static u32
243 render_ring_add_request(struct drm_device *dev,
244                         struct intel_ring_buffer *ring,
245                         u32 flush_domains)
246 {
247         drm_i915_private_t *dev_priv = dev->dev_private;
248         u32 seqno;
249
250         seqno = i915_gem_get_seqno(dev);
251
252         if (IS_GEN6(dev)) {
253                 BEGIN_LP_RING(6);
254                 OUT_RING(GFX_OP_PIPE_CONTROL | 3);
255                 OUT_RING(PIPE_CONTROL_QW_WRITE |
256                          PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_IS_FLUSH |
257                          PIPE_CONTROL_NOTIFY);
258                 OUT_RING(dev_priv->seqno_gfx_addr | PIPE_CONTROL_GLOBAL_GTT);
259                 OUT_RING(seqno);
260                 OUT_RING(0);
261                 OUT_RING(0);
262                 ADVANCE_LP_RING();
263         } else if (HAS_PIPE_CONTROL(dev)) {
264                 u32 scratch_addr = dev_priv->seqno_gfx_addr + 128;
265
266                 /*
267                  * Workaround qword write incoherence by flushing the
268                  * PIPE_NOTIFY buffers out to memory before requesting
269                  * an interrupt.
270                  */
271                 BEGIN_LP_RING(32);
272                 OUT_RING(GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |
273                          PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_TC_FLUSH);
274                 OUT_RING(dev_priv->seqno_gfx_addr | PIPE_CONTROL_GLOBAL_GTT);
275                 OUT_RING(seqno);
276                 OUT_RING(0);
277                 PIPE_CONTROL_FLUSH(scratch_addr);
278                 scratch_addr += 128; /* write to separate cachelines */
279                 PIPE_CONTROL_FLUSH(scratch_addr);
280                 scratch_addr += 128;
281                 PIPE_CONTROL_FLUSH(scratch_addr);
282                 scratch_addr += 128;
283                 PIPE_CONTROL_FLUSH(scratch_addr);
284                 scratch_addr += 128;
285                 PIPE_CONTROL_FLUSH(scratch_addr);
286                 scratch_addr += 128;
287                 PIPE_CONTROL_FLUSH(scratch_addr);
288                 OUT_RING(GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |
289                          PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_TC_FLUSH |
290                          PIPE_CONTROL_NOTIFY);
291                 OUT_RING(dev_priv->seqno_gfx_addr | PIPE_CONTROL_GLOBAL_GTT);
292                 OUT_RING(seqno);
293                 OUT_RING(0);
294                 ADVANCE_LP_RING();
295         } else {
296                 BEGIN_LP_RING(4);
297                 OUT_RING(MI_STORE_DWORD_INDEX);
298                 OUT_RING(I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
299                 OUT_RING(seqno);
300
301                 OUT_RING(MI_USER_INTERRUPT);
302                 ADVANCE_LP_RING();
303         }
304         return seqno;
305 }
306
307 static u32
308 render_ring_get_seqno(struct drm_device *dev,
309                       struct intel_ring_buffer *ring)
310 {
311         drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
312         if (HAS_PIPE_CONTROL(dev))
313                 return ((volatile u32 *)(dev_priv->seqno_page))[0];
314         else
315                 return intel_read_status_page(ring, I915_GEM_HWS_INDEX);
316 }
317
318 static void
319 render_ring_get_user_irq(struct drm_device *dev,
320                          struct intel_ring_buffer *ring)
321 {
322         drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
323         unsigned long irqflags;
324
325         spin_lock_irqsave(&dev_priv->user_irq_lock, irqflags);
326         if (dev->irq_enabled && (++ring->user_irq_refcount == 1)) {
327                 if (HAS_PCH_SPLIT(dev))
328                         ironlake_enable_graphics_irq(dev_priv, GT_PIPE_NOTIFY);
329                 else
330                         i915_enable_irq(dev_priv, I915_USER_INTERRUPT);
331         }
332         spin_unlock_irqrestore(&dev_priv->user_irq_lock, irqflags);
333 }
334
335 static void
336 render_ring_put_user_irq(struct drm_device *dev,
337                          struct intel_ring_buffer *ring)
338 {
339         drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
340         unsigned long irqflags;
341
342         spin_lock_irqsave(&dev_priv->user_irq_lock, irqflags);
343         BUG_ON(dev->irq_enabled && ring->user_irq_refcount <= 0);
344         if (dev->irq_enabled && (--ring->user_irq_refcount == 0)) {
345                 if (HAS_PCH_SPLIT(dev))
346                         ironlake_disable_graphics_irq(dev_priv, GT_PIPE_NOTIFY);
347                 else
348                         i915_disable_irq(dev_priv, I915_USER_INTERRUPT);
349         }
350         spin_unlock_irqrestore(&dev_priv->user_irq_lock, irqflags);
351 }
352
353 void intel_ring_setup_status_page(struct drm_device *dev,
354                                   struct intel_ring_buffer *ring)
355 {
356         drm_i915_private_t *dev_priv = dev->dev_private;
357         if (IS_GEN6(dev)) {
358                 I915_WRITE(RING_HWS_PGA_GEN6(ring->mmio_base),
359                            ring->status_page.gfx_addr);
360                 I915_READ(RING_HWS_PGA_GEN6(ring->mmio_base)); /* posting read */
361         } else {
362                 I915_WRITE(RING_HWS_PGA(ring->mmio_base),
363                            ring->status_page.gfx_addr);
364                 I915_READ(RING_HWS_PGA(ring->mmio_base)); /* posting read */
365         }
366
367 }
368
369 static void
370 bsd_ring_flush(struct drm_device *dev,
371                 struct intel_ring_buffer *ring,
372                 u32     invalidate_domains,
373                 u32     flush_domains)
374 {
375         intel_ring_begin(dev, ring, 2);
376         intel_ring_emit(dev, ring, MI_FLUSH);
377         intel_ring_emit(dev, ring, MI_NOOP);
378         intel_ring_advance(dev, ring);
379 }
380
381 static int init_bsd_ring(struct drm_device *dev,
382                          struct intel_ring_buffer *ring)
383 {
384         return init_ring_common(dev, ring);
385 }
386
387 static u32
388 ring_add_request(struct drm_device *dev,
389                  struct intel_ring_buffer *ring,
390                  u32 flush_domains)
391 {
392         u32 seqno;
393
394         seqno = i915_gem_get_seqno(dev);
395
396         intel_ring_begin(dev, ring, 4);
397         intel_ring_emit(dev, ring, MI_STORE_DWORD_INDEX);
398         intel_ring_emit(dev, ring,
399                         I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
400         intel_ring_emit(dev, ring, seqno);
401         intel_ring_emit(dev, ring, MI_USER_INTERRUPT);
402         intel_ring_advance(dev, ring);
403
404         DRM_DEBUG_DRIVER("%s %d\n", ring->name, seqno);
405
406         return seqno;
407 }
408
409 static void
410 bsd_ring_get_user_irq(struct drm_device *dev,
411                       struct intel_ring_buffer *ring)
412 {
413         /* do nothing */
414 }
415 static void
416 bsd_ring_put_user_irq(struct drm_device *dev,
417                       struct intel_ring_buffer *ring)
418 {
419         /* do nothing */
420 }
421
422 static u32
423 ring_status_page_get_seqno(struct drm_device *dev,
424                            struct intel_ring_buffer *ring)
425 {
426         return intel_read_status_page(ring, I915_GEM_HWS_INDEX);
427 }
428
429 static int
430 ring_dispatch_gem_execbuffer(struct drm_device *dev,
431                              struct intel_ring_buffer *ring,
432                              struct drm_i915_gem_execbuffer2 *exec,
433                              struct drm_clip_rect *cliprects,
434                              uint64_t exec_offset)
435 {
436         uint32_t exec_start;
437         exec_start = (uint32_t) exec_offset + exec->batch_start_offset;
438         intel_ring_begin(dev, ring, 2);
439         intel_ring_emit(dev, ring, MI_BATCH_BUFFER_START |
440                         (2 << 6) | MI_BATCH_NON_SECURE_I965);
441         intel_ring_emit(dev, ring, exec_start);
442         intel_ring_advance(dev, ring);
443         return 0;
444 }
445
446 static int
447 render_ring_dispatch_gem_execbuffer(struct drm_device *dev,
448                                     struct intel_ring_buffer *ring,
449                                     struct drm_i915_gem_execbuffer2 *exec,
450                                     struct drm_clip_rect *cliprects,
451                                     uint64_t exec_offset)
452 {
453         drm_i915_private_t *dev_priv = dev->dev_private;
454         int nbox = exec->num_cliprects;
455         int i = 0, count;
456         uint32_t exec_start, exec_len;
457         exec_start = (uint32_t) exec_offset + exec->batch_start_offset;
458         exec_len = (uint32_t) exec->batch_len;
459
460         trace_i915_gem_request_submit(dev, dev_priv->next_seqno + 1);
461
462         count = nbox ? nbox : 1;
463
464         for (i = 0; i < count; i++) {
465                 if (i < nbox) {
466                         int ret = i915_emit_box(dev, cliprects, i,
467                                                 exec->DR1, exec->DR4);
468                         if (ret)
469                                 return ret;
470                 }
471
472                 if (IS_I830(dev) || IS_845G(dev)) {
473                         intel_ring_begin(dev, ring, 4);
474                         intel_ring_emit(dev, ring, MI_BATCH_BUFFER);
475                         intel_ring_emit(dev, ring,
476                                         exec_start | MI_BATCH_NON_SECURE);
477                         intel_ring_emit(dev, ring, exec_start + exec_len - 4);
478                         intel_ring_emit(dev, ring, 0);
479                 } else {
480                         intel_ring_begin(dev, ring, 2);
481                         if (INTEL_INFO(dev)->gen >= 4) {
482                                 intel_ring_emit(dev, ring,
483                                                 MI_BATCH_BUFFER_START | (2 << 6)
484                                                 | MI_BATCH_NON_SECURE_I965);
485                                 intel_ring_emit(dev, ring, exec_start);
486                         } else {
487                                 intel_ring_emit(dev, ring, MI_BATCH_BUFFER_START
488                                                 | (2 << 6));
489                                 intel_ring_emit(dev, ring, exec_start |
490                                                 MI_BATCH_NON_SECURE);
491                         }
492                 }
493                 intel_ring_advance(dev, ring);
494         }
495
496         if (IS_G4X(dev) || IS_GEN5(dev)) {
497                 intel_ring_begin(dev, ring, 2);
498                 intel_ring_emit(dev, ring, MI_FLUSH |
499                                 MI_NO_WRITE_FLUSH |
500                                 MI_INVALIDATE_ISP );
501                 intel_ring_emit(dev, ring, MI_NOOP);
502                 intel_ring_advance(dev, ring);
503         }
504         /* XXX breadcrumb */
505
506         return 0;
507 }
508
509 static void cleanup_status_page(struct drm_device *dev,
510                                 struct intel_ring_buffer *ring)
511 {
512         drm_i915_private_t *dev_priv = dev->dev_private;
513         struct drm_gem_object *obj;
514         struct drm_i915_gem_object *obj_priv;
515
516         obj = ring->status_page.obj;
517         if (obj == NULL)
518                 return;
519         obj_priv = to_intel_bo(obj);
520
521         kunmap(obj_priv->pages[0]);
522         i915_gem_object_unpin(obj);
523         drm_gem_object_unreference(obj);
524         ring->status_page.obj = NULL;
525
526         memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map));
527 }
528
529 static int init_status_page(struct drm_device *dev,
530                             struct intel_ring_buffer *ring)
531 {
532         drm_i915_private_t *dev_priv = dev->dev_private;
533         struct drm_gem_object *obj;
534         struct drm_i915_gem_object *obj_priv;
535         int ret;
536
537         obj = i915_gem_alloc_object(dev, 4096);
538         if (obj == NULL) {
539                 DRM_ERROR("Failed to allocate status page\n");
540                 ret = -ENOMEM;
541                 goto err;
542         }
543         obj_priv = to_intel_bo(obj);
544         obj_priv->agp_type = AGP_USER_CACHED_MEMORY;
545
546         ret = i915_gem_object_pin(obj, 4096);
547         if (ret != 0) {
548                 goto err_unref;
549         }
550
551         ring->status_page.gfx_addr = obj_priv->gtt_offset;
552         ring->status_page.page_addr = kmap(obj_priv->pages[0]);
553         if (ring->status_page.page_addr == NULL) {
554                 memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map));
555                 goto err_unpin;
556         }
557         ring->status_page.obj = obj;
558         memset(ring->status_page.page_addr, 0, PAGE_SIZE);
559
560         intel_ring_setup_status_page(dev, ring);
561         DRM_DEBUG_DRIVER("%s hws offset: 0x%08x\n",
562                         ring->name, ring->status_page.gfx_addr);
563
564         return 0;
565
566 err_unpin:
567         i915_gem_object_unpin(obj);
568 err_unref:
569         drm_gem_object_unreference(obj);
570 err:
571         return ret;
572 }
573
574 int intel_init_ring_buffer(struct drm_device *dev,
575                            struct intel_ring_buffer *ring)
576 {
577         struct drm_i915_private *dev_priv = dev->dev_private;
578         struct drm_i915_gem_object *obj_priv;
579         struct drm_gem_object *obj;
580         int ret;
581
582         ring->dev = dev;
583         INIT_LIST_HEAD(&ring->active_list);
584         INIT_LIST_HEAD(&ring->request_list);
585         INIT_LIST_HEAD(&ring->gpu_write_list);
586
587         if (I915_NEED_GFX_HWS(dev)) {
588                 ret = init_status_page(dev, ring);
589                 if (ret)
590                         return ret;
591         }
592
593         obj = i915_gem_alloc_object(dev, ring->size);
594         if (obj == NULL) {
595                 DRM_ERROR("Failed to allocate ringbuffer\n");
596                 ret = -ENOMEM;
597                 goto err_hws;
598         }
599
600         ring->gem_object = obj;
601
602         ret = i915_gem_object_pin(obj, PAGE_SIZE);
603         if (ret)
604                 goto err_unref;
605
606         obj_priv = to_intel_bo(obj);
607         ring->map.size = ring->size;
608         ring->map.offset = dev->agp->base + obj_priv->gtt_offset;
609         ring->map.type = 0;
610         ring->map.flags = 0;
611         ring->map.mtrr = 0;
612
613         drm_core_ioremap_wc(&ring->map, dev);
614         if (ring->map.handle == NULL) {
615                 DRM_ERROR("Failed to map ringbuffer.\n");
616                 ret = -EINVAL;
617                 goto err_unpin;
618         }
619
620         ring->virtual_start = ring->map.handle;
621         ret = ring->init(dev, ring);
622         if (ret)
623                 goto err_unmap;
624
625         if (!drm_core_check_feature(dev, DRIVER_MODESET))
626                 i915_kernel_lost_context(dev);
627         else {
628                 ring->head = I915_READ_HEAD(ring) & HEAD_ADDR;
629                 ring->tail = I915_READ_TAIL(ring) & TAIL_ADDR;
630                 ring->space = ring->head - (ring->tail + 8);
631                 if (ring->space < 0)
632                         ring->space += ring->size;
633         }
634         return ret;
635
636 err_unmap:
637         drm_core_ioremapfree(&ring->map, dev);
638 err_unpin:
639         i915_gem_object_unpin(obj);
640 err_unref:
641         drm_gem_object_unreference(obj);
642         ring->gem_object = NULL;
643 err_hws:
644         cleanup_status_page(dev, ring);
645         return ret;
646 }
647
648 void intel_cleanup_ring_buffer(struct drm_device *dev,
649                                struct intel_ring_buffer *ring)
650 {
651         if (ring->gem_object == NULL)
652                 return;
653
654         drm_core_ioremapfree(&ring->map, dev);
655
656         i915_gem_object_unpin(ring->gem_object);
657         drm_gem_object_unreference(ring->gem_object);
658         ring->gem_object = NULL;
659
660         if (ring->cleanup)
661                 ring->cleanup(ring);
662
663         cleanup_status_page(dev, ring);
664 }
665
666 static int intel_wrap_ring_buffer(struct drm_device *dev,
667                                   struct intel_ring_buffer *ring)
668 {
669         unsigned int *virt;
670         int rem;
671         rem = ring->size - ring->tail;
672
673         if (ring->space < rem) {
674                 int ret = intel_wait_ring_buffer(dev, ring, rem);
675                 if (ret)
676                         return ret;
677         }
678
679         virt = (unsigned int *)(ring->virtual_start + ring->tail);
680         rem /= 8;
681         while (rem--) {
682                 *virt++ = MI_NOOP;
683                 *virt++ = MI_NOOP;
684         }
685
686         ring->tail = 0;
687         ring->space = ring->head - 8;
688
689         return 0;
690 }
691
692 int intel_wait_ring_buffer(struct drm_device *dev,
693                            struct intel_ring_buffer *ring, int n)
694 {
695         unsigned long end;
696         drm_i915_private_t *dev_priv = dev->dev_private;
697         u32 head;
698
699         trace_i915_ring_wait_begin (dev);
700         end = jiffies + 3 * HZ;
701         do {
702                 /* If the reported head position has wrapped or hasn't advanced,
703                  * fallback to the slow and accurate path.
704                  */
705                 head = intel_read_status_page(ring, 4);
706                 if (head < ring->actual_head)
707                         head = I915_READ_HEAD(ring);
708                 ring->actual_head = head;
709                 ring->head = head & HEAD_ADDR;
710                 ring->space = ring->head - (ring->tail + 8);
711                 if (ring->space < 0)
712                         ring->space += ring->size;
713                 if (ring->space >= n) {
714                         trace_i915_ring_wait_end (dev);
715                         return 0;
716                 }
717
718                 if (dev->primary->master) {
719                         struct drm_i915_master_private *master_priv = dev->primary->master->driver_priv;
720                         if (master_priv->sarea_priv)
721                                 master_priv->sarea_priv->perf_boxes |= I915_BOX_WAIT;
722                 }
723
724                 msleep(1);
725         } while (!time_after(jiffies, end));
726         trace_i915_ring_wait_end (dev);
727         return -EBUSY;
728 }
729
730 void intel_ring_begin(struct drm_device *dev,
731                       struct intel_ring_buffer *ring,
732                       int num_dwords)
733 {
734         int n = 4*num_dwords;
735         if (unlikely(ring->tail + n > ring->size))
736                 intel_wrap_ring_buffer(dev, ring);
737         if (unlikely(ring->space < n))
738                 intel_wait_ring_buffer(dev, ring, n);
739
740         ring->space -= n;
741 }
742
743 void intel_ring_advance(struct drm_device *dev,
744                         struct intel_ring_buffer *ring)
745 {
746         ring->tail &= ring->size - 1;
747         ring->write_tail(dev, ring, ring->tail);
748 }
749
750 static const struct intel_ring_buffer render_ring = {
751         .name                   = "render ring",
752         .id                     = RING_RENDER,
753         .mmio_base              = RENDER_RING_BASE,
754         .size                   = 32 * PAGE_SIZE,
755         .init                   = init_render_ring,
756         .write_tail             = ring_write_tail,
757         .flush                  = render_ring_flush,
758         .add_request            = render_ring_add_request,
759         .get_seqno              = render_ring_get_seqno,
760         .user_irq_get           = render_ring_get_user_irq,
761         .user_irq_put           = render_ring_put_user_irq,
762         .dispatch_gem_execbuffer = render_ring_dispatch_gem_execbuffer,
763 };
764
765 /* ring buffer for bit-stream decoder */
766
767 static const struct intel_ring_buffer bsd_ring = {
768         .name                   = "bsd ring",
769         .id                     = RING_BSD,
770         .mmio_base              = BSD_RING_BASE,
771         .size                   = 32 * PAGE_SIZE,
772         .init                   = init_bsd_ring,
773         .write_tail             = ring_write_tail,
774         .flush                  = bsd_ring_flush,
775         .add_request            = ring_add_request,
776         .get_seqno              = ring_status_page_get_seqno,
777         .user_irq_get           = bsd_ring_get_user_irq,
778         .user_irq_put           = bsd_ring_put_user_irq,
779         .dispatch_gem_execbuffer = ring_dispatch_gem_execbuffer,
780 };
781
782
783 static void gen6_bsd_ring_write_tail(struct drm_device *dev,
784                                      struct intel_ring_buffer *ring,
785                                      u32 value)
786 {
787        drm_i915_private_t *dev_priv = dev->dev_private;
788
789        /* Every tail move must follow the sequence below */
790        I915_WRITE(GEN6_BSD_SLEEP_PSMI_CONTROL,
791                GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_MODIFY_MASK |
792                GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_DISABLE);
793        I915_WRITE(GEN6_BSD_RNCID, 0x0);
794
795        if (wait_for((I915_READ(GEN6_BSD_SLEEP_PSMI_CONTROL) &
796                                GEN6_BSD_SLEEP_PSMI_CONTROL_IDLE_INDICATOR) == 0,
797                        50))
798                DRM_ERROR("timed out waiting for IDLE Indicator\n");
799
800        I915_WRITE_TAIL(ring, value);
801        I915_WRITE(GEN6_BSD_SLEEP_PSMI_CONTROL,
802                GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_MODIFY_MASK |
803                GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_ENABLE);
804 }
805
806 static void gen6_ring_flush(struct drm_device *dev,
807                             struct intel_ring_buffer *ring,
808                             u32 invalidate_domains,
809                             u32 flush_domains)
810 {
811        intel_ring_begin(dev, ring, 4);
812        intel_ring_emit(dev, ring, MI_FLUSH_DW);
813        intel_ring_emit(dev, ring, 0);
814        intel_ring_emit(dev, ring, 0);
815        intel_ring_emit(dev, ring, 0);
816        intel_ring_advance(dev, ring);
817 }
818
819 static int
820 gen6_ring_dispatch_gem_execbuffer(struct drm_device *dev,
821                                   struct intel_ring_buffer *ring,
822                                   struct drm_i915_gem_execbuffer2 *exec,
823                                   struct drm_clip_rect *cliprects,
824                                   uint64_t exec_offset)
825 {
826        uint32_t exec_start;
827
828        exec_start = (uint32_t) exec_offset + exec->batch_start_offset;
829
830        intel_ring_begin(dev, ring, 2);
831        intel_ring_emit(dev, ring,
832                        MI_BATCH_BUFFER_START | MI_BATCH_NON_SECURE_I965);
833        /* bit0-7 is the length on GEN6+ */
834        intel_ring_emit(dev, ring, exec_start);
835        intel_ring_advance(dev, ring);
836
837        return 0;
838 }
839
840 /* ring buffer for Video Codec for Gen6+ */
841 static const struct intel_ring_buffer gen6_bsd_ring = {
842        .name                    = "gen6 bsd ring",
843        .id                      = RING_BSD,
844        .mmio_base               = GEN6_BSD_RING_BASE,
845        .size                    = 32 * PAGE_SIZE,
846        .init                    = init_bsd_ring,
847        .write_tail              = gen6_bsd_ring_write_tail,
848        .flush                   = gen6_ring_flush,
849        .add_request             = ring_add_request,
850        .get_seqno               = ring_status_page_get_seqno,
851        .user_irq_get            = bsd_ring_get_user_irq,
852        .user_irq_put            = bsd_ring_put_user_irq,
853        .dispatch_gem_execbuffer = gen6_ring_dispatch_gem_execbuffer,
854 };
855
856 /* Blitter support (SandyBridge+) */
857
858 static void
859 blt_ring_get_user_irq(struct drm_device *dev,
860                       struct intel_ring_buffer *ring)
861 {
862         /* do nothing */
863 }
864 static void
865 blt_ring_put_user_irq(struct drm_device *dev,
866                       struct intel_ring_buffer *ring)
867 {
868         /* do nothing */
869 }
870
871
872 /* Workaround for some stepping of SNB,
873  * each time when BLT engine ring tail moved,
874  * the first command in the ring to be parsed
875  * should be MI_BATCH_BUFFER_START
876  */
877 #define NEED_BLT_WORKAROUND(dev) \
878         (IS_GEN6(dev) && (dev->pdev->revision < 8))
879
880 static inline struct drm_i915_gem_object *
881 to_blt_workaround(struct intel_ring_buffer *ring)
882 {
883         return ring->private;
884 }
885
886 static int blt_ring_init(struct drm_device *dev,
887                          struct intel_ring_buffer *ring)
888 {
889         if (NEED_BLT_WORKAROUND(dev)) {
890                 struct drm_i915_gem_object *obj;
891                 u32 __iomem *ptr;
892                 int ret;
893
894                 obj = to_intel_bo(i915_gem_alloc_object(dev, 4096));
895                 if (obj == NULL)
896                         return -ENOMEM;
897
898                 ret = i915_gem_object_pin(&obj->base, 4096);
899                 if (ret) {
900                         drm_gem_object_unreference(&obj->base);
901                         return ret;
902                 }
903
904                 ptr = kmap(obj->pages[0]);
905                 iowrite32(MI_BATCH_BUFFER_END, ptr);
906                 iowrite32(MI_NOOP, ptr+1);
907                 kunmap(obj->pages[0]);
908
909                 ret = i915_gem_object_set_to_gtt_domain(&obj->base, false);
910                 if (ret) {
911                         i915_gem_object_unpin(&obj->base);
912                         drm_gem_object_unreference(&obj->base);
913                         return ret;
914                 }
915
916                 ring->private = obj;
917         }
918
919         return init_ring_common(dev, ring);
920 }
921
922 static void blt_ring_begin(struct drm_device *dev,
923                            struct intel_ring_buffer *ring,
924                           int num_dwords)
925 {
926         if (ring->private) {
927                 intel_ring_begin(dev, ring, num_dwords+2);
928                 intel_ring_emit(dev, ring, MI_BATCH_BUFFER_START);
929                 intel_ring_emit(dev, ring, to_blt_workaround(ring)->gtt_offset);
930         } else
931                 intel_ring_begin(dev, ring, 4);
932 }
933
934 static void blt_ring_flush(struct drm_device *dev,
935                            struct intel_ring_buffer *ring,
936                            u32 invalidate_domains,
937                            u32 flush_domains)
938 {
939         blt_ring_begin(dev, ring, 4);
940         intel_ring_emit(dev, ring, MI_FLUSH_DW);
941         intel_ring_emit(dev, ring, 0);
942         intel_ring_emit(dev, ring, 0);
943         intel_ring_emit(dev, ring, 0);
944         intel_ring_advance(dev, ring);
945 }
946
947 static u32
948 blt_ring_add_request(struct drm_device *dev,
949                      struct intel_ring_buffer *ring,
950                      u32 flush_domains)
951 {
952         u32 seqno = i915_gem_get_seqno(dev);
953
954         blt_ring_begin(dev, ring, 4);
955         intel_ring_emit(dev, ring, MI_STORE_DWORD_INDEX);
956         intel_ring_emit(dev, ring,
957                         I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
958         intel_ring_emit(dev, ring, seqno);
959         intel_ring_emit(dev, ring, MI_USER_INTERRUPT);
960         intel_ring_advance(dev, ring);
961
962         DRM_DEBUG_DRIVER("%s %d\n", ring->name, seqno);
963         return seqno;
964 }
965
966 static void blt_ring_cleanup(struct intel_ring_buffer *ring)
967 {
968         if (!ring->private)
969                 return;
970
971         i915_gem_object_unpin(ring->private);
972         drm_gem_object_unreference(ring->private);
973         ring->private = NULL;
974 }
975
976 static const struct intel_ring_buffer gen6_blt_ring = {
977        .name                    = "blt ring",
978        .id                      = RING_BLT,
979        .mmio_base               = BLT_RING_BASE,
980        .size                    = 32 * PAGE_SIZE,
981        .init                    = blt_ring_init,
982        .write_tail              = ring_write_tail,
983        .flush                   = blt_ring_flush,
984        .add_request             = blt_ring_add_request,
985        .get_seqno               = ring_status_page_get_seqno,
986        .user_irq_get            = blt_ring_get_user_irq,
987        .user_irq_put            = blt_ring_put_user_irq,
988        .dispatch_gem_execbuffer = gen6_ring_dispatch_gem_execbuffer,
989        .cleanup                 = blt_ring_cleanup,
990 };
991
992 int intel_init_render_ring_buffer(struct drm_device *dev)
993 {
994         drm_i915_private_t *dev_priv = dev->dev_private;
995
996         dev_priv->render_ring = render_ring;
997
998         if (!I915_NEED_GFX_HWS(dev)) {
999                 dev_priv->render_ring.status_page.page_addr
1000                         = dev_priv->status_page_dmah->vaddr;
1001                 memset(dev_priv->render_ring.status_page.page_addr,
1002                                 0, PAGE_SIZE);
1003         }
1004
1005         return intel_init_ring_buffer(dev, &dev_priv->render_ring);
1006 }
1007
1008 int intel_init_bsd_ring_buffer(struct drm_device *dev)
1009 {
1010         drm_i915_private_t *dev_priv = dev->dev_private;
1011
1012         if (IS_GEN6(dev))
1013                 dev_priv->bsd_ring = gen6_bsd_ring;
1014         else
1015                 dev_priv->bsd_ring = bsd_ring;
1016
1017         return intel_init_ring_buffer(dev, &dev_priv->bsd_ring);
1018 }
1019
1020 int intel_init_blt_ring_buffer(struct drm_device *dev)
1021 {
1022         drm_i915_private_t *dev_priv = dev->dev_private;
1023
1024         dev_priv->blt_ring = gen6_blt_ring;
1025
1026         return intel_init_ring_buffer(dev, &dev_priv->blt_ring);
1027 }