diff options
author | Debarshi Dutta <ddutta@nvidia.com> | 2018-03-23 06:02:27 -0400 |
---|---|---|
committer | mobile promotions <svcmobile_promotions@nvidia.com> | 2018-04-23 00:04:48 -0400 |
commit | d0e4dfd6efd651abc431aba9cfae5907638f8172 (patch) | |
tree | 1a412eaa4636ff2e2862c1623bad3a5ea6883d4e /drivers/gpu/nvgpu/gk20a/fence_gk20a.c | |
parent | c918c42a4a3651f757c6966aead4b07eb4b56697 (diff) |
gpu: nvgpu: sync_framework cleanups
This patch deals with cleanups meant to make things simpler for the
upcoming os abstraction patches for the sync framework. This patch
causes some substantial changes which are listed out as follows.
1) sync_timeline is moved out of gk20a_fence into struct
nvgpu_channel_linux. New function pointers are created to facilitate os
independent methods for enabling/disabling timeline and are now named
as os_fence_framework. These function pointers are located in the struct
os_channel under struct gk20a.
2) construction of the channel_sync require nvgpu_finalize_poweron_linux()
to be invoked before invocations to nvgpu_init_mm_ce_context(). Hence,
these methods are now moved away from gk20a_finalize_poweron() and
invoked after nvgpu_finalize_poweron_linux().
3) sync_fence creation is now delinked from fence construction and move
to the channel_sync_gk20a's channel_incr methods. These sync_fences are
mainly associated with post_fences.
4) In case userspace requires the sync_fences to be constructed, we
try to obtain an fd before the gk20a_channel_submit_gpfifo() instead of
trying to do that later. This is used to avoid potential after effects
of duplicate work submission due to failure to obtain an unused fd.
JIRA NVGPU-66
Change-Id: I42a3e4e2e692a113b1b36d2b48ab107ae4444dfa
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1678400
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Diffstat (limited to 'drivers/gpu/nvgpu/gk20a/fence_gk20a.c')
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/fence_gk20a.c | 101 |
1 files changed, 33 insertions, 68 deletions
diff --git a/drivers/gpu/nvgpu/gk20a/fence_gk20a.c b/drivers/gpu/nvgpu/gk20a/fence_gk20a.c index f74afd6e..f0ad773f 100644 --- a/drivers/gpu/nvgpu/gk20a/fence_gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/fence_gk20a.c | |||
@@ -54,9 +54,10 @@ static void gk20a_fence_free(struct nvgpu_ref *ref) | |||
54 | struct gk20a *g = f->g; | 54 | struct gk20a *g = f->g; |
55 | 55 | ||
56 | #ifdef CONFIG_SYNC | 56 | #ifdef CONFIG_SYNC |
57 | if (f->sync_fence) | 57 | if (f->os_fence) |
58 | sync_fence_put(f->sync_fence); | 58 | sync_fence_put(f->os_fence); |
59 | #endif | 59 | #endif |
60 | |||
60 | if (f->semaphore) | 61 | if (f->semaphore) |
61 | nvgpu_semaphore_put(f->semaphore); | 62 | nvgpu_semaphore_put(f->semaphore); |
62 | 63 | ||
@@ -80,7 +81,7 @@ struct gk20a_fence *gk20a_fence_get(struct gk20a_fence *f) | |||
80 | return f; | 81 | return f; |
81 | } | 82 | } |
82 | 83 | ||
83 | static inline bool gk20a_fence_is_valid(struct gk20a_fence *f) | 84 | inline bool gk20a_fence_is_valid(struct gk20a_fence *f) |
84 | { | 85 | { |
85 | bool valid = f->valid; | 86 | bool valid = f->valid; |
86 | 87 | ||
@@ -88,6 +89,21 @@ static inline bool gk20a_fence_is_valid(struct gk20a_fence *f) | |||
88 | return valid; | 89 | return valid; |
89 | } | 90 | } |
90 | 91 | ||
92 | int gk20a_fence_install_fd(struct gk20a_fence *f, int fd) | ||
93 | { | ||
94 | #ifdef CONFIG_SYNC | ||
95 | if (!f || !gk20a_fence_is_valid(f) || !f->os_fence) | ||
96 | return -EINVAL; | ||
97 | |||
98 | sync_fence_get(f->os_fence); | ||
99 | sync_fence_install(f->os_fence, fd); | ||
100 | |||
101 | return 0; | ||
102 | #else | ||
103 | return -ENODEV; | ||
104 | #endif | ||
105 | } | ||
106 | |||
91 | int gk20a_fence_wait(struct gk20a *g, struct gk20a_fence *f, | 107 | int gk20a_fence_wait(struct gk20a *g, struct gk20a_fence *f, |
92 | unsigned long timeout) | 108 | unsigned long timeout) |
93 | { | 109 | { |
@@ -107,26 +123,6 @@ bool gk20a_fence_is_expired(struct gk20a_fence *f) | |||
107 | return true; | 123 | return true; |
108 | } | 124 | } |
109 | 125 | ||
110 | int gk20a_fence_install_fd(struct gk20a_fence *f) | ||
111 | { | ||
112 | #ifdef CONFIG_SYNC | ||
113 | int fd; | ||
114 | |||
115 | if (!f || !gk20a_fence_is_valid(f) || !f->sync_fence) | ||
116 | return -EINVAL; | ||
117 | |||
118 | fd = get_unused_fd_flags(O_RDWR); | ||
119 | if (fd < 0) | ||
120 | return fd; | ||
121 | |||
122 | sync_fence_get(f->sync_fence); | ||
123 | sync_fence_install(f->sync_fence, fd); | ||
124 | return fd; | ||
125 | #else | ||
126 | return -ENODEV; | ||
127 | #endif | ||
128 | } | ||
129 | |||
130 | int gk20a_alloc_fence_pool(struct channel_gk20a *c, unsigned int count) | 126 | int gk20a_alloc_fence_pool(struct channel_gk20a *c, unsigned int count) |
131 | { | 127 | { |
132 | int err; | 128 | int err; |
@@ -195,13 +191,14 @@ struct gk20a_fence *gk20a_alloc_fence(struct channel_gk20a *c) | |||
195 | 191 | ||
196 | void gk20a_init_fence(struct gk20a_fence *f, | 192 | void gk20a_init_fence(struct gk20a_fence *f, |
197 | const struct gk20a_fence_ops *ops, | 193 | const struct gk20a_fence_ops *ops, |
198 | struct sync_fence *sync_fence) | 194 | struct sync_fence *os_fence) |
199 | { | 195 | { |
200 | if (!f) | 196 | if (!f) |
201 | return; | 197 | return; |
202 | f->ops = ops; | 198 | f->ops = ops; |
203 | f->sync_fence = sync_fence; | ||
204 | f->syncpt_id = -1; | 199 | f->syncpt_id = -1; |
200 | f->semaphore = NULL; | ||
201 | f->os_fence = os_fence; | ||
205 | } | 202 | } |
206 | 203 | ||
207 | /* Fences that are backed by GPU semaphores: */ | 204 | /* Fences that are backed by GPU semaphores: */ |
@@ -227,36 +224,19 @@ static const struct gk20a_fence_ops nvgpu_semaphore_fence_ops = { | |||
227 | .is_expired = &nvgpu_semaphore_fence_is_expired, | 224 | .is_expired = &nvgpu_semaphore_fence_is_expired, |
228 | }; | 225 | }; |
229 | 226 | ||
230 | /* This function takes ownership of the semaphore */ | 227 | /* This function takes ownership of the semaphore as well as the os_fence */ |
231 | int gk20a_fence_from_semaphore( | 228 | int gk20a_fence_from_semaphore( |
232 | struct gk20a *g, | ||
233 | struct gk20a_fence *fence_out, | 229 | struct gk20a_fence *fence_out, |
234 | struct sync_timeline *timeline, | ||
235 | struct nvgpu_semaphore *semaphore, | 230 | struct nvgpu_semaphore *semaphore, |
236 | struct nvgpu_cond *semaphore_wq, | 231 | struct nvgpu_cond *semaphore_wq, |
237 | bool need_sync_fence) | 232 | struct sync_fence *os_fence) |
238 | { | 233 | { |
239 | struct gk20a_fence *f = fence_out; | 234 | struct gk20a_fence *f = fence_out; |
240 | struct sync_fence *sync_fence = NULL; | ||
241 | |||
242 | #ifdef CONFIG_SYNC | ||
243 | if (need_sync_fence) { | ||
244 | sync_fence = gk20a_sync_fence_create(g, timeline, semaphore, | ||
245 | "f-gk20a-0x%04x", | ||
246 | nvgpu_semaphore_gpu_ro_va(semaphore)); | ||
247 | if (!sync_fence) | ||
248 | return -ENOMEM; | ||
249 | } | ||
250 | #endif | ||
251 | 235 | ||
252 | gk20a_init_fence(f, &nvgpu_semaphore_fence_ops, sync_fence); | 236 | gk20a_init_fence(f, &nvgpu_semaphore_fence_ops, os_fence); |
253 | if (!f) { | 237 | if (!f) |
254 | #ifdef CONFIG_SYNC | ||
255 | if (sync_fence) | ||
256 | sync_fence_put(sync_fence); | ||
257 | #endif | ||
258 | return -EINVAL; | 238 | return -EINVAL; |
259 | } | 239 | |
260 | 240 | ||
261 | f->semaphore = semaphore; | 241 | f->semaphore = semaphore; |
262 | f->semaphore_wq = semaphore_wq; | 242 | f->semaphore_wq = semaphore_wq; |
@@ -306,32 +286,18 @@ static const struct gk20a_fence_ops gk20a_syncpt_fence_ops = { | |||
306 | .is_expired = &gk20a_syncpt_fence_is_expired, | 286 | .is_expired = &gk20a_syncpt_fence_is_expired, |
307 | }; | 287 | }; |
308 | 288 | ||
289 | /* This function takes the ownership of the os_fence */ | ||
309 | int gk20a_fence_from_syncpt( | 290 | int gk20a_fence_from_syncpt( |
310 | struct gk20a_fence *fence_out, | 291 | struct gk20a_fence *fence_out, |
311 | struct nvgpu_nvhost_dev *nvhost_dev, | 292 | struct nvgpu_nvhost_dev *nvhost_dev, |
312 | u32 id, u32 value, | 293 | u32 id, u32 value, struct sync_fence *os_fence) |
313 | bool need_sync_fence) | ||
314 | { | 294 | { |
315 | struct gk20a_fence *f = fence_out; | 295 | struct gk20a_fence *f = fence_out; |
316 | struct sync_fence *sync_fence = NULL; | ||
317 | |||
318 | #ifdef CONFIG_SYNC | ||
319 | if (need_sync_fence) { | ||
320 | sync_fence = nvgpu_nvhost_sync_create_fence(nvhost_dev, | ||
321 | id, value, "fence"); | ||
322 | if (IS_ERR(sync_fence)) | ||
323 | return PTR_ERR(sync_fence); | ||
324 | } | ||
325 | #endif | ||
326 | 296 | ||
327 | gk20a_init_fence(f, &gk20a_syncpt_fence_ops, sync_fence); | 297 | gk20a_init_fence(f, &gk20a_syncpt_fence_ops, os_fence); |
328 | if (!f) { | 298 | if (!f) |
329 | #ifdef CONFIG_SYNC | ||
330 | if (sync_fence) | ||
331 | sync_fence_put(sync_fence); | ||
332 | #endif | ||
333 | return -EINVAL; | 299 | return -EINVAL; |
334 | } | 300 | |
335 | f->nvhost_dev = nvhost_dev; | 301 | f->nvhost_dev = nvhost_dev; |
336 | f->syncpt_id = id; | 302 | f->syncpt_id = id; |
337 | f->syncpt_value = value; | 303 | f->syncpt_value = value; |
@@ -346,8 +312,7 @@ int gk20a_fence_from_syncpt( | |||
346 | int gk20a_fence_from_syncpt( | 312 | int gk20a_fence_from_syncpt( |
347 | struct gk20a_fence *fence_out, | 313 | struct gk20a_fence *fence_out, |
348 | struct nvgpu_nvhost_dev *nvhost_dev, | 314 | struct nvgpu_nvhost_dev *nvhost_dev, |
349 | u32 id, u32 value, | 315 | u32 id, u32 value, struct sync_fence *os_fence) |
350 | bool need_sync_fence) | ||
351 | { | 316 | { |
352 | return -EINVAL; | 317 | return -EINVAL; |
353 | } | 318 | } |