lima/ppir: add ppir_node to ppir_src
[mesa.git] / src / gallium / drivers / iris / iris_fence.c
1 /*
2 * Copyright © 2018 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 shall be included
12 * in all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
15 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
19 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
20 * DEALINGS IN THE SOFTWARE.
21 */
22
23 /**
24 * @file iris_fence.c
25 *
26 * Fences for driver and IPC serialisation, scheduling and synchronisation.
27 */
28
29 #include "util/u_inlines.h"
30 #include "intel/common/gen_gem.h"
31
32 #include "iris_batch.h"
33 #include "iris_bufmgr.h"
34 #include "iris_context.h"
35 #include "iris_fence.h"
36 #include "iris_screen.h"
37
38 static uint32_t
39 gem_syncobj_create(int fd, uint32_t flags)
40 {
41 struct drm_syncobj_create args = {
42 .flags = flags,
43 };
44
45 gen_ioctl(fd, DRM_IOCTL_SYNCOBJ_CREATE, &args);
46
47 return args.handle;
48 }
49
50 static void
51 gem_syncobj_destroy(int fd, uint32_t handle)
52 {
53 struct drm_syncobj_destroy args = {
54 .handle = handle,
55 };
56
57 gen_ioctl(fd, DRM_IOCTL_SYNCOBJ_DESTROY, &args);
58 }
59
60 /**
61 * Make a new sync-point.
62 */
63 struct iris_syncpt *
64 iris_create_syncpt(struct iris_screen *screen)
65 {
66 struct iris_syncpt *syncpt = malloc(sizeof(*syncpt));
67
68 if (!syncpt)
69 return NULL;
70
71 syncpt->handle = gem_syncobj_create(screen->fd, 0);
72 assert(syncpt->handle);
73
74 pipe_reference_init(&syncpt->ref, 1);
75
76 return syncpt;
77 }
78
79 void
80 iris_syncpt_destroy(struct iris_screen *screen, struct iris_syncpt *syncpt)
81 {
82 gem_syncobj_destroy(screen->fd, syncpt->handle);
83 free(syncpt);
84 }
85
86 /**
87 * Add a sync-point to the batch, with the given flags.
88 *
89 * \p flags One of I915_EXEC_FENCE_WAIT or I915_EXEC_FENCE_SIGNAL.
90 */
91 void
92 iris_batch_add_syncpt(struct iris_batch *batch,
93 struct iris_syncpt *syncpt,
94 unsigned flags)
95 {
96 struct drm_i915_gem_exec_fence *fence =
97 util_dynarray_grow(&batch->exec_fences, struct drm_i915_gem_exec_fence, 1);
98
99 *fence = (struct drm_i915_gem_exec_fence) {
100 .handle = syncpt->handle,
101 .flags = flags,
102 };
103
104 struct iris_syncpt **store =
105 util_dynarray_grow(&batch->syncpts, struct iris_syncpt *, 1);
106
107 *store = NULL;
108 iris_syncpt_reference(batch->screen, store, syncpt);
109 }
110
111 /* ------------------------------------------------------------------- */
112
113 struct pipe_fence_handle {
114 struct pipe_reference ref;
115 struct iris_syncpt *syncpt[IRIS_BATCH_COUNT];
116 unsigned count;
117 };
118
119 static void
120 iris_fence_destroy(struct pipe_screen *p_screen,
121 struct pipe_fence_handle *fence)
122 {
123 struct iris_screen *screen = (struct iris_screen *)p_screen;
124
125 for (unsigned i = 0; i < fence->count; i++)
126 iris_syncpt_reference(screen, &fence->syncpt[i], NULL);
127
128 free(fence);
129 }
130
131 static void
132 iris_fence_reference(struct pipe_screen *p_screen,
133 struct pipe_fence_handle **dst,
134 struct pipe_fence_handle *src)
135 {
136 if (pipe_reference(&(*dst)->ref, &src->ref))
137 iris_fence_destroy(p_screen, *dst);
138
139 *dst = src;
140 }
141
142 bool
143 iris_wait_syncpt(struct pipe_screen *p_screen,
144 struct iris_syncpt *syncpt,
145 int64_t timeout_nsec)
146 {
147 if (!syncpt)
148 return false;
149
150 struct iris_screen *screen = (struct iris_screen *)p_screen;
151 struct drm_syncobj_wait args = {
152 .handles = (uintptr_t)&syncpt->handle,
153 .count_handles = 1,
154 .timeout_nsec = timeout_nsec,
155 };
156 return gen_ioctl(screen->fd, DRM_IOCTL_SYNCOBJ_WAIT, &args);
157 }
158
159 static void
160 iris_fence_flush(struct pipe_context *ctx,
161 struct pipe_fence_handle **out_fence,
162 unsigned flags)
163 {
164 struct iris_screen *screen = (void *) ctx->screen;
165 struct iris_context *ice = (struct iris_context *)ctx;
166
167 /* XXX PIPE_FLUSH_DEFERRED */
168 for (unsigned i = 0; i < IRIS_BATCH_COUNT; i++)
169 iris_batch_flush(&ice->batches[i]);
170
171 if (!out_fence)
172 return;
173
174 struct pipe_fence_handle *fence = calloc(1, sizeof(*fence));
175 if (!fence)
176 return;
177
178 pipe_reference_init(&fence->ref, 1);
179
180 for (unsigned b = 0; b < IRIS_BATCH_COUNT; b++) {
181 if (!iris_wait_syncpt(ctx->screen, ice->batches[b].last_syncpt, 0))
182 continue;
183
184 iris_syncpt_reference(screen, &fence->syncpt[fence->count++],
185 ice->batches[b].last_syncpt);
186 }
187 *out_fence = fence;
188 }
189
190 static void
191 iris_fence_await(struct pipe_context *ctx,
192 struct pipe_fence_handle *fence)
193 {
194 struct iris_context *ice = (struct iris_context *)ctx;
195
196 for (unsigned b = 0; b < IRIS_BATCH_COUNT; b++) {
197 for (unsigned i = 0; i < fence->count; i++) {
198 iris_batch_add_syncpt(&ice->batches[b], fence->syncpt[i],
199 I915_EXEC_FENCE_WAIT);
200 }
201 }
202 }
203
204 #define NSEC_PER_SEC (1000 * USEC_PER_SEC)
205 #define USEC_PER_SEC (1000 * MSEC_PER_SEC)
206 #define MSEC_PER_SEC (1000)
207
208 static uint64_t
209 rel2abs(uint64_t timeout)
210 {
211 struct timespec ts;
212 uint64_t now;
213
214 if (!timeout)
215 return 0;
216
217 if (timeout == PIPE_TIMEOUT_INFINITE)
218 return INT64_MAX;
219
220 clock_gettime(CLOCK_MONOTONIC, &ts);
221 now = ts.tv_sec * NSEC_PER_SEC + ts.tv_nsec;
222
223 if (now > INT64_MAX - timeout)
224 return INT64_MAX;
225
226 return now + timeout;
227 }
228
229 static bool
230 iris_fence_finish(struct pipe_screen *p_screen,
231 struct pipe_context *ctx,
232 struct pipe_fence_handle *fence,
233 uint64_t timeout)
234 {
235 struct iris_screen *screen = (struct iris_screen *)p_screen;
236
237 if (!fence->count)
238 return true;
239
240 uint32_t handles[ARRAY_SIZE(fence->syncpt)];
241 for (unsigned i = 0; i < fence->count; i++)
242 handles[i] = fence->syncpt[i]->handle;
243
244 struct drm_syncobj_wait args = {
245 .handles = (uintptr_t)handles,
246 .count_handles = fence->count,
247 .timeout_nsec = rel2abs(timeout), /* XXX */
248 .flags = DRM_SYNCOBJ_WAIT_FLAGS_WAIT_ALL
249 };
250 return gen_ioctl(screen->fd, DRM_IOCTL_SYNCOBJ_WAIT, &args) == 0;
251 }
252
253 #ifndef SYNC_IOC_MAGIC
254 /* duplicated from linux/sync_file.h to avoid build-time dependency
255 * on new (v4.7) kernel headers. Once distro's are mostly using
256 * something newer than v4.7 drop this and #include <linux/sync_file.h>
257 * instead.
258 */
259 struct sync_merge_data {
260 char name[32];
261 __s32 fd2;
262 __s32 fence;
263 __u32 flags;
264 __u32 pad;
265 };
266
267 #define SYNC_IOC_MAGIC '>'
268 #define SYNC_IOC_MERGE _IOWR(SYNC_IOC_MAGIC, 3, struct sync_merge_data)
269 #endif
270
271 static int
272 sync_merge_fd(int sync_fd, int new_fd)
273 {
274 if (sync_fd == -1)
275 return new_fd;
276
277 if (new_fd == -1)
278 return sync_fd;
279
280 struct sync_merge_data args = {
281 .name = "iris fence",
282 .fd2 = new_fd,
283 .fence = -1,
284 };
285
286 gen_ioctl(sync_fd, SYNC_IOC_MERGE, &args);
287 close(new_fd);
288 close(sync_fd);
289
290 return args.fence;
291 }
292
293 static int
294 iris_fence_get_fd(struct pipe_screen *p_screen,
295 struct pipe_fence_handle *fence)
296 {
297 struct iris_screen *screen = (struct iris_screen *)p_screen;
298 int fd = -1;
299
300 for (unsigned i = 0; i < fence->count; i++) {
301 struct drm_syncobj_handle args = {
302 .handle = fence->syncpt[i]->handle,
303 .flags = DRM_SYNCOBJ_HANDLE_TO_FD_FLAGS_EXPORT_SYNC_FILE,
304 .fd = -1,
305 };
306
307 gen_ioctl(screen->fd, DRM_IOCTL_SYNCOBJ_HANDLE_TO_FD, &args);
308 fd = sync_merge_fd(fd, args.fd);
309 }
310
311 return fd;
312 }
313
314 static void
315 iris_fence_create_fd(struct pipe_context *ctx,
316 struct pipe_fence_handle **out,
317 int fd,
318 enum pipe_fd_type type)
319 {
320 assert(type == PIPE_FD_TYPE_NATIVE_SYNC);
321
322 struct iris_screen *screen = (struct iris_screen *)ctx->screen;
323 struct drm_syncobj_handle args = {
324 .flags = DRM_SYNCOBJ_FD_TO_HANDLE_FLAGS_IMPORT_SYNC_FILE,
325 .fd = fd,
326 };
327 gen_ioctl(screen->fd, DRM_IOCTL_SYNCOBJ_FD_TO_HANDLE, &args);
328
329 struct iris_syncpt *syncpt = malloc(sizeof(*syncpt));
330 syncpt->handle = args.handle;
331 pipe_reference_init(&syncpt->ref, 1);
332
333 struct pipe_fence_handle *fence = malloc(sizeof(*fence));
334 pipe_reference_init(&fence->ref, 1);
335 fence->syncpt[0] = syncpt;
336 fence->count = 1;
337
338 *out = fence;
339 }
340
341 void
342 iris_init_screen_fence_functions(struct pipe_screen *screen)
343 {
344 screen->fence_reference = iris_fence_reference;
345 screen->fence_finish = iris_fence_finish;
346 screen->fence_get_fd = iris_fence_get_fd;
347 }
348
349 void
350 iris_init_context_fence_functions(struct pipe_context *ctx)
351 {
352 ctx->flush = iris_fence_flush;
353 ctx->create_fence_fd = iris_fence_create_fd;
354 ctx->fence_server_sync = iris_fence_await;
355 }