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. When a fence register is needed for a reloc entry,
199 * drm_intel_bo_emit_reloc_fence() will be called explicitly.
201 * intel_bo_add_reloc() currently lacks "bool fenced" for this to work.
202 * But we never need a fence register on GEN4+ so we do not need to worry
205 drm_intel_bufmgr_gem_enable_fenced_relocs(winsys
->bufmgr
);
207 drm_intel_bufmgr_gem_enable_reuse(winsys
->bufmgr
);
213 intel_winsys_destroy(struct intel_winsys
*winsys
)
216 drm_intel_decode_context_free(winsys
->decode
);
218 if (winsys
->first_gem_ctx
)
219 drm_intel_gem_context_destroy(winsys
->first_gem_ctx
);
221 pipe_mutex_destroy(winsys
->mutex
);
222 drm_intel_bufmgr_destroy(winsys
->bufmgr
);
226 const struct intel_winsys_info
*
227 intel_winsys_get_info(const struct intel_winsys
*winsys
)
229 return &winsys
->info
;
232 struct intel_context
*
233 intel_winsys_create_context(struct intel_winsys
*winsys
)
235 drm_intel_context
*gem_ctx
;
237 /* try the preallocated context first */
238 pipe_mutex_lock(winsys
->mutex
);
239 gem_ctx
= winsys
->first_gem_ctx
;
240 winsys
->first_gem_ctx
= NULL
;
241 pipe_mutex_unlock(winsys
->mutex
);
244 gem_ctx
= drm_intel_gem_context_create(winsys
->bufmgr
);
246 return (struct intel_context
*) gem_ctx
;
250 intel_winsys_destroy_context(struct intel_winsys
*winsys
,
251 struct intel_context
*ctx
)
253 drm_intel_gem_context_destroy((drm_intel_context
*) ctx
);
257 intel_winsys_read_reg(struct intel_winsys
*winsys
,
258 uint32_t reg
, uint64_t *val
)
260 return drm_intel_reg_read(winsys
->bufmgr
, reg
, val
);
264 intel_winsys_alloc_bo(struct intel_winsys
*winsys
,
266 enum intel_tiling_mode tiling
,
268 unsigned long height
,
269 uint32_t initial_domain
)
271 const bool for_render
=
272 (initial_domain
& (INTEL_DOMAIN_RENDER
| INTEL_DOMAIN_INSTRUCTION
));
273 const unsigned int alignment
= 4096; /* always page-aligned */
290 if (pitch
> ULONG_MAX
/ height
)
293 size
= pitch
* height
;
296 bo
= drm_intel_bo_alloc_for_render(winsys
->bufmgr
,
297 name
, size
, alignment
);
300 bo
= drm_intel_bo_alloc(winsys
->bufmgr
, name
, size
, alignment
);
303 if (bo
&& tiling
!= INTEL_TILING_NONE
) {
304 uint32_t real_tiling
= tiling
;
307 err
= drm_intel_bo_set_tiling(bo
, &real_tiling
, pitch
);
308 if (err
|| real_tiling
!= tiling
) {
309 assert(!"tiling mismatch");
310 drm_intel_bo_unreference(bo
);
315 return (struct intel_bo
*) bo
;
319 intel_winsys_import_handle(struct intel_winsys
*winsys
,
321 const struct winsys_handle
*handle
,
322 unsigned long height
,
323 enum intel_tiling_mode
*tiling
,
324 unsigned long *pitch
)
326 uint32_t real_tiling
, swizzle
;
330 switch (handle
->type
) {
331 case DRM_API_HANDLE_TYPE_SHARED
:
333 const uint32_t gem_name
= handle
->handle
;
334 bo
= drm_intel_bo_gem_create_from_name(winsys
->bufmgr
,
338 case DRM_API_HANDLE_TYPE_FD
:
340 const int fd
= (int) handle
->handle
;
341 bo
= drm_intel_bo_gem_create_from_prime(winsys
->bufmgr
,
342 fd
, height
* handle
->stride
);
353 err
= drm_intel_bo_get_tiling(bo
, &real_tiling
, &swizzle
);
355 drm_intel_bo_unreference(bo
);
359 *tiling
= real_tiling
;
360 *pitch
= handle
->stride
;
362 return (struct intel_bo
*) bo
;
366 intel_winsys_export_handle(struct intel_winsys
*winsys
,
368 enum intel_tiling_mode tiling
,
370 unsigned long height
,
371 struct winsys_handle
*handle
)
375 switch (handle
->type
) {
376 case DRM_API_HANDLE_TYPE_SHARED
:
380 err
= drm_intel_bo_flink(gem_bo(bo
), &name
);
382 handle
->handle
= name
;
385 case DRM_API_HANDLE_TYPE_KMS
:
386 handle
->handle
= gem_bo(bo
)->handle
;
388 case DRM_API_HANDLE_TYPE_FD
:
392 err
= drm_intel_bo_gem_export_to_prime(gem_bo(bo
), &fd
);
405 handle
->stride
= pitch
;
411 intel_winsys_can_submit_bo(struct intel_winsys
*winsys
,
412 struct intel_bo
**bo_array
,
415 return !drm_intel_bufmgr_check_aperture_space((drm_intel_bo
**) bo_array
,
420 intel_winsys_submit_bo(struct intel_winsys
*winsys
,
421 enum intel_ring_type ring
,
422 struct intel_bo
*bo
, int used
,
423 struct intel_context
*ctx
,
426 const unsigned long exec_flags
= (unsigned long) ring
| flags
;
428 /* logical contexts are only available for the render ring */
429 if (ring
!= INTEL_RING_RENDER
)
433 return drm_intel_gem_bo_context_exec(gem_bo(bo
),
434 (drm_intel_context
*) ctx
, used
, exec_flags
);
437 return drm_intel_bo_mrb_exec(gem_bo(bo
),
438 used
, NULL
, 0, 0, exec_flags
);
443 intel_winsys_decode_bo(struct intel_winsys
*winsys
,
444 struct intel_bo
*bo
, int used
)
448 ptr
= intel_bo_map(bo
, false);
450 debug_printf("failed to map buffer for decoding\n");
454 pipe_mutex_lock(winsys
->mutex
);
456 if (!winsys
->decode
) {
457 winsys
->decode
= drm_intel_decode_context_alloc(winsys
->info
.devid
);
458 if (!winsys
->decode
) {
459 pipe_mutex_unlock(winsys
->mutex
);
464 /* debug_printf()/debug_error() uses stderr by default */
465 drm_intel_decode_set_output_file(winsys
->decode
, stderr
);
471 drm_intel_decode_set_batch_pointer(winsys
->decode
,
472 ptr
, gem_bo(bo
)->offset64
, used
);
474 drm_intel_decode(winsys
->decode
);
476 pipe_mutex_unlock(winsys
->mutex
);
482 intel_bo_reference(struct intel_bo
*bo
)
484 drm_intel_bo_reference(gem_bo(bo
));
488 intel_bo_unreference(struct intel_bo
*bo
)
490 drm_intel_bo_unreference(gem_bo(bo
));
494 intel_bo_map(struct intel_bo
*bo
, bool write_enable
)
498 err
= drm_intel_bo_map(gem_bo(bo
), write_enable
);
500 debug_error("failed to map bo");
504 return gem_bo(bo
)->virtual;
508 intel_bo_map_gtt(struct intel_bo
*bo
)
512 err
= drm_intel_gem_bo_map_gtt(gem_bo(bo
));
514 debug_error("failed to map bo");
518 return gem_bo(bo
)->virtual;
522 intel_bo_map_unsynchronized(struct intel_bo
*bo
)
526 err
= drm_intel_gem_bo_map_unsynchronized(gem_bo(bo
));
528 debug_error("failed to map bo");
532 return gem_bo(bo
)->virtual;
536 intel_bo_unmap(struct intel_bo
*bo
)
540 err
= drm_intel_bo_unmap(gem_bo(bo
));
545 intel_bo_pwrite(struct intel_bo
*bo
, unsigned long offset
,
546 unsigned long size
, const void *data
)
548 return drm_intel_bo_subdata(gem_bo(bo
), offset
, size
, data
);
552 intel_bo_pread(struct intel_bo
*bo
, unsigned long offset
,
553 unsigned long size
, void *data
)
555 return drm_intel_bo_get_subdata(gem_bo(bo
), offset
, size
, data
);
559 intel_bo_add_reloc(struct intel_bo
*bo
, uint32_t offset
,
560 struct intel_bo
*target_bo
, uint32_t target_offset
,
561 uint32_t read_domains
, uint32_t write_domain
,
562 uint64_t *presumed_offset
)
566 err
= drm_intel_bo_emit_reloc(gem_bo(bo
), offset
,
567 gem_bo(target_bo
), target_offset
,
568 read_domains
, write_domain
);
570 *presumed_offset
= gem_bo(target_bo
)->offset64
+ target_offset
;
576 intel_bo_get_reloc_count(struct intel_bo
*bo
)
578 return drm_intel_gem_bo_get_reloc_count(gem_bo(bo
));
582 intel_bo_truncate_relocs(struct intel_bo
*bo
, int start
)
584 drm_intel_gem_bo_clear_relocs(gem_bo(bo
), start
);
588 intel_bo_has_reloc(struct intel_bo
*bo
, struct intel_bo
*target_bo
)
590 return drm_intel_bo_references(gem_bo(bo
), gem_bo(target_bo
));
594 intel_bo_wait(struct intel_bo
*bo
, int64_t timeout
)
598 err
= drm_intel_gem_bo_wait(gem_bo(bo
), timeout
);
599 /* consider the bo idle on errors */
600 if (err
&& err
!= -ETIME
)