2 * Mesa 3-D graphics library
4 * Copyright (C) 2012-2014 LunarG, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
25 * Chia-I Wu <olv@lunarg.com>
31 #define ETIME ETIMEDOUT
36 #include <intel_bufmgr.h>
38 #include "os/os_thread.h"
39 #include "state_tracker/drm_driver.h"
40 #include "pipe/p_state.h"
41 #include "util/u_inlines.h"
42 #include "util/u_memory.h"
43 #include "util/u_debug.h"
44 #include "../intel_winsys.h"
46 #define BATCH_SZ (8192 * sizeof(uint32_t))
50 drm_intel_bufmgr
*bufmgr
;
51 struct intel_winsys_info info
;
53 /* these are protected by the mutex */
55 drm_intel_context
*first_gem_ctx
;
56 struct drm_intel_decode
*decode
;
60 gem_bo(const struct intel_bo
*bo
)
62 return (drm_intel_bo
*) bo
;
66 get_param(struct intel_winsys
*winsys
, int param
, int *value
)
68 struct drm_i915_getparam gp
;
73 memset(&gp
, 0, sizeof(gp
));
77 err
= drmCommandWriteRead(winsys
->fd
, DRM_I915_GETPARAM
, &gp
, sizeof(gp
));
87 test_address_swizzling(struct intel_winsys
*winsys
)
90 uint32_t tiling
= I915_TILING_X
, swizzle
;
93 bo
= drm_intel_bo_alloc_tiled(winsys
->bufmgr
,
94 "address swizzling test", 64, 64, 4, &tiling
, &pitch
, 0);
96 drm_intel_bo_get_tiling(bo
, &tiling
, &swizzle
);
97 drm_intel_bo_unreference(bo
);
100 swizzle
= I915_BIT_6_SWIZZLE_NONE
;
103 return (swizzle
!= I915_BIT_6_SWIZZLE_NONE
);
107 test_reg_read(struct intel_winsys
*winsys
, uint32_t reg
)
111 return !drm_intel_reg_read(winsys
->bufmgr
, reg
, &dummy
);
115 probe_winsys(struct intel_winsys
*winsys
)
117 struct intel_winsys_info
*info
= &winsys
->info
;
121 * When we need the Nth vertex from a user vertex buffer, and the vertex is
122 * uploaded to, say, the beginning of a bo, we want the first vertex in the
123 * bo to be fetched. One way to do this is to set the base address of the
126 * bo->offset64 + (vb->buffer_offset - vb->stride * N).
128 * The second term may be negative, and we need kernel support to do that.
130 * This check is taken from the classic driver. u_vbuf_upload_buffers()
131 * guarantees the term is never negative, but it is good to require a
134 get_param(winsys
, I915_PARAM_HAS_RELAXED_DELTA
, &val
);
136 debug_error("kernel 2.6.39 required");
140 info
->devid
= drm_intel_bufmgr_gem_get_devid(winsys
->bufmgr
);
142 if (drm_intel_get_aperture_sizes(winsys
->fd
,
143 &info
->aperture_mappable
, &info
->aperture_total
)) {
144 debug_error("failed to query aperture sizes");
148 info
->max_batch_size
= BATCH_SZ
;
150 get_param(winsys
, I915_PARAM_HAS_LLC
, &val
);
152 info
->has_address_swizzling
= test_address_swizzling(winsys
);
154 winsys
->first_gem_ctx
= drm_intel_gem_context_create(winsys
->bufmgr
);
155 info
->has_logical_context
= (winsys
->first_gem_ctx
!= NULL
);
157 get_param(winsys
, I915_PARAM_HAS_ALIASING_PPGTT
, &val
);
158 info
->has_ppgtt
= val
;
160 /* test TIMESTAMP read */
161 info
->has_timestamp
= test_reg_read(winsys
, 0x2358);
163 get_param(winsys
, I915_PARAM_HAS_GEN7_SOL_RESET
, &val
);
164 info
->has_gen7_sol_reset
= val
;
169 struct intel_winsys
*
170 intel_winsys_create_for_fd(int fd
)
172 struct intel_winsys
*winsys
;
174 winsys
= CALLOC_STRUCT(intel_winsys
);
180 winsys
->bufmgr
= drm_intel_bufmgr_gem_init(winsys
->fd
, BATCH_SZ
);
181 if (!winsys
->bufmgr
) {
182 debug_error("failed to create GEM buffer manager");
187 pipe_mutex_init(winsys
->mutex
);
189 if (!probe_winsys(winsys
)) {
190 pipe_mutex_destroy(winsys
->mutex
);
191 drm_intel_bufmgr_destroy(winsys
->bufmgr
);
197 * No need to implicitly set up a fence register for each non-linear reloc
198 * entry. INTEL_RELOC_FENCE will be set on reloc entries that need them.
200 drm_intel_bufmgr_gem_enable_fenced_relocs(winsys
->bufmgr
);
202 drm_intel_bufmgr_gem_enable_reuse(winsys
->bufmgr
);
208 intel_winsys_destroy(struct intel_winsys
*winsys
)
211 drm_intel_decode_context_free(winsys
->decode
);
213 if (winsys
->first_gem_ctx
)
214 drm_intel_gem_context_destroy(winsys
->first_gem_ctx
);
216 pipe_mutex_destroy(winsys
->mutex
);
217 drm_intel_bufmgr_destroy(winsys
->bufmgr
);
221 const struct intel_winsys_info
*
222 intel_winsys_get_info(const struct intel_winsys
*winsys
)
224 return &winsys
->info
;
227 struct intel_context
*
228 intel_winsys_create_context(struct intel_winsys
*winsys
)
230 drm_intel_context
*gem_ctx
;
232 /* try the preallocated context first */
233 pipe_mutex_lock(winsys
->mutex
);
234 gem_ctx
= winsys
->first_gem_ctx
;
235 winsys
->first_gem_ctx
= NULL
;
236 pipe_mutex_unlock(winsys
->mutex
);
239 gem_ctx
= drm_intel_gem_context_create(winsys
->bufmgr
);
241 return (struct intel_context
*) gem_ctx
;
245 intel_winsys_destroy_context(struct intel_winsys
*winsys
,
246 struct intel_context
*ctx
)
248 drm_intel_gem_context_destroy((drm_intel_context
*) ctx
);
252 intel_winsys_read_reg(struct intel_winsys
*winsys
,
253 uint32_t reg
, uint64_t *val
)
255 return drm_intel_reg_read(winsys
->bufmgr
, reg
, val
);
259 intel_winsys_alloc_bo(struct intel_winsys
*winsys
,
261 enum intel_tiling_mode tiling
,
263 unsigned long height
,
266 const unsigned int alignment
= 4096; /* always page-aligned */
283 if (pitch
> ULONG_MAX
/ height
)
286 size
= pitch
* height
;
289 bo
= drm_intel_bo_alloc(winsys
->bufmgr
, name
, size
, alignment
);
292 bo
= drm_intel_bo_alloc_for_render(winsys
->bufmgr
,
293 name
, size
, alignment
);
296 if (bo
&& tiling
!= INTEL_TILING_NONE
) {
297 uint32_t real_tiling
= tiling
;
300 err
= drm_intel_bo_set_tiling(bo
, &real_tiling
, pitch
);
301 if (err
|| real_tiling
!= tiling
) {
302 assert(!"tiling mismatch");
303 drm_intel_bo_unreference(bo
);
308 return (struct intel_bo
*) bo
;
312 intel_winsys_import_handle(struct intel_winsys
*winsys
,
314 const struct winsys_handle
*handle
,
315 unsigned long height
,
316 enum intel_tiling_mode
*tiling
,
317 unsigned long *pitch
)
319 uint32_t real_tiling
, swizzle
;
323 switch (handle
->type
) {
324 case DRM_API_HANDLE_TYPE_SHARED
:
326 const uint32_t gem_name
= handle
->handle
;
327 bo
= drm_intel_bo_gem_create_from_name(winsys
->bufmgr
,
331 case DRM_API_HANDLE_TYPE_FD
:
333 const int fd
= (int) handle
->handle
;
334 bo
= drm_intel_bo_gem_create_from_prime(winsys
->bufmgr
,
335 fd
, height
* handle
->stride
);
346 err
= drm_intel_bo_get_tiling(bo
, &real_tiling
, &swizzle
);
348 drm_intel_bo_unreference(bo
);
352 *tiling
= real_tiling
;
353 *pitch
= handle
->stride
;
355 return (struct intel_bo
*) bo
;
359 intel_winsys_export_handle(struct intel_winsys
*winsys
,
361 enum intel_tiling_mode tiling
,
363 unsigned long height
,
364 struct winsys_handle
*handle
)
368 switch (handle
->type
) {
369 case DRM_API_HANDLE_TYPE_SHARED
:
373 err
= drm_intel_bo_flink(gem_bo(bo
), &name
);
375 handle
->handle
= name
;
378 case DRM_API_HANDLE_TYPE_KMS
:
379 handle
->handle
= gem_bo(bo
)->handle
;
381 case DRM_API_HANDLE_TYPE_FD
:
385 err
= drm_intel_bo_gem_export_to_prime(gem_bo(bo
), &fd
);
398 handle
->stride
= pitch
;
404 intel_winsys_can_submit_bo(struct intel_winsys
*winsys
,
405 struct intel_bo
**bo_array
,
408 return !drm_intel_bufmgr_check_aperture_space((drm_intel_bo
**) bo_array
,
413 intel_winsys_submit_bo(struct intel_winsys
*winsys
,
414 enum intel_ring_type ring
,
415 struct intel_bo
*bo
, int used
,
416 struct intel_context
*ctx
,
419 const unsigned long exec_flags
= (unsigned long) ring
| flags
;
421 /* logical contexts are only available for the render ring */
422 if (ring
!= INTEL_RING_RENDER
)
426 return drm_intel_gem_bo_context_exec(gem_bo(bo
),
427 (drm_intel_context
*) ctx
, used
, exec_flags
);
430 return drm_intel_bo_mrb_exec(gem_bo(bo
),
431 used
, NULL
, 0, 0, exec_flags
);
436 intel_winsys_decode_bo(struct intel_winsys
*winsys
,
437 struct intel_bo
*bo
, int used
)
441 ptr
= intel_bo_map(bo
, false);
443 debug_printf("failed to map buffer for decoding\n");
447 pipe_mutex_lock(winsys
->mutex
);
449 if (!winsys
->decode
) {
450 winsys
->decode
= drm_intel_decode_context_alloc(winsys
->info
.devid
);
451 if (!winsys
->decode
) {
452 pipe_mutex_unlock(winsys
->mutex
);
457 /* debug_printf()/debug_error() uses stderr by default */
458 drm_intel_decode_set_output_file(winsys
->decode
, stderr
);
464 drm_intel_decode_set_batch_pointer(winsys
->decode
,
465 ptr
, gem_bo(bo
)->offset64
, used
);
467 drm_intel_decode(winsys
->decode
);
469 pipe_mutex_unlock(winsys
->mutex
);
475 intel_bo_reference(struct intel_bo
*bo
)
477 drm_intel_bo_reference(gem_bo(bo
));
481 intel_bo_unreference(struct intel_bo
*bo
)
483 drm_intel_bo_unreference(gem_bo(bo
));
487 intel_bo_map(struct intel_bo
*bo
, bool write_enable
)
491 err
= drm_intel_bo_map(gem_bo(bo
), write_enable
);
493 debug_error("failed to map bo");
497 return gem_bo(bo
)->virtual;
501 intel_bo_map_gtt(struct intel_bo
*bo
)
505 err
= drm_intel_gem_bo_map_gtt(gem_bo(bo
));
507 debug_error("failed to map bo");
511 return gem_bo(bo
)->virtual;
515 intel_bo_map_unsynchronized(struct intel_bo
*bo
)
519 err
= drm_intel_gem_bo_map_unsynchronized(gem_bo(bo
));
521 debug_error("failed to map bo");
525 return gem_bo(bo
)->virtual;
529 intel_bo_unmap(struct intel_bo
*bo
)
533 err
= drm_intel_bo_unmap(gem_bo(bo
));
538 intel_bo_pwrite(struct intel_bo
*bo
, unsigned long offset
,
539 unsigned long size
, const void *data
)
541 return drm_intel_bo_subdata(gem_bo(bo
), offset
, size
, data
);
545 intel_bo_pread(struct intel_bo
*bo
, unsigned long offset
,
546 unsigned long size
, void *data
)
548 return drm_intel_bo_get_subdata(gem_bo(bo
), offset
, size
, data
);
552 intel_bo_add_reloc(struct intel_bo
*bo
, uint32_t offset
,
553 struct intel_bo
*target_bo
, uint32_t target_offset
,
554 uint32_t flags
, uint64_t *presumed_offset
)
556 uint32_t read_domains
, write_domain
;
559 if (flags
& INTEL_RELOC_WRITE
) {
561 * Because of the translation to domains, INTEL_RELOC_GGTT should only
562 * be set on GEN6 when the bo is written by MI_* or PIPE_CONTROL. The
563 * kernel will translate it back to INTEL_RELOC_GGTT.
565 write_domain
= (flags
& INTEL_RELOC_GGTT
) ?
566 I915_GEM_DOMAIN_INSTRUCTION
: I915_GEM_DOMAIN_RENDER
;
567 read_domains
= write_domain
;
570 read_domains
= I915_GEM_DOMAIN_RENDER
|
571 I915_GEM_DOMAIN_SAMPLER
|
572 I915_GEM_DOMAIN_INSTRUCTION
|
573 I915_GEM_DOMAIN_VERTEX
;
576 if (flags
& INTEL_RELOC_FENCE
) {
577 err
= drm_intel_bo_emit_reloc_fence(gem_bo(bo
), offset
,
578 gem_bo(target_bo
), target_offset
,
579 read_domains
, write_domain
);
581 err
= drm_intel_bo_emit_reloc(gem_bo(bo
), offset
,
582 gem_bo(target_bo
), target_offset
,
583 read_domains
, write_domain
);
586 *presumed_offset
= gem_bo(target_bo
)->offset64
+ target_offset
;
592 intel_bo_get_reloc_count(struct intel_bo
*bo
)
594 return drm_intel_gem_bo_get_reloc_count(gem_bo(bo
));
598 intel_bo_truncate_relocs(struct intel_bo
*bo
, int start
)
600 drm_intel_gem_bo_clear_relocs(gem_bo(bo
), start
);
604 intel_bo_has_reloc(struct intel_bo
*bo
, struct intel_bo
*target_bo
)
606 return drm_intel_bo_references(gem_bo(bo
), gem_bo(target_bo
));
610 intel_bo_wait(struct intel_bo
*bo
, int64_t timeout
)
614 err
= drm_intel_gem_bo_wait(gem_bo(bo
), timeout
);
615 /* consider the bo idle on errors */
616 if (err
&& err
!= -ETIME
)