diff options
author | Joshua Bakita <bakitajoshua@gmail.com> | 2024-09-25 16:09:09 -0400 |
---|---|---|
committer | Joshua Bakita <bakitajoshua@gmail.com> | 2024-09-25 16:09:09 -0400 |
commit | f347fde22f1297e4f022600d201780d5ead78114 (patch) | |
tree | 76be305d6187003a1e0486ff6e91efb1062ae118 /include/os/linux/sched.c | |
parent | 8340d234d78a7d0f46c11a584de538148b78b7cb (diff) |
Delete no-longer-needed nvgpu headersHEADmasterjbakita-wip
The dependency on these was removed in commit 8340d234.
Diffstat (limited to 'include/os/linux/sched.c')
-rw-r--r-- | include/os/linux/sched.c | 666 |
1 files changed, 0 insertions, 666 deletions
diff --git a/include/os/linux/sched.c b/include/os/linux/sched.c deleted file mode 100644 index 30c58a1..0000000 --- a/include/os/linux/sched.c +++ /dev/null | |||
@@ -1,666 +0,0 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2016-2020, NVIDIA CORPORATION. All rights reserved. | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify it | ||
5 | * under the terms and conditions of the GNU General Public License, | ||
6 | * version 2, as published by the Free Software Foundation. | ||
7 | * | ||
8 | * This program is distributed in the hope it will be useful, but WITHOUT | ||
9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | ||
10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | ||
11 | * more details. | ||
12 | * | ||
13 | * You should have received a copy of the GNU General Public License | ||
14 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
15 | */ | ||
16 | #include <asm/barrier.h> | ||
17 | #include <linux/wait.h> | ||
18 | #include <linux/uaccess.h> | ||
19 | #include <linux/poll.h> | ||
20 | #include <uapi/linux/nvgpu.h> | ||
21 | |||
22 | #include <nvgpu/kmem.h> | ||
23 | #include <nvgpu/log.h> | ||
24 | #include <nvgpu/bug.h> | ||
25 | #include <nvgpu/barrier.h> | ||
26 | #include <nvgpu/gk20a.h> | ||
27 | |||
28 | #include "gk20a/gr_gk20a.h" | ||
29 | #include "sched.h" | ||
30 | #include "os_linux.h" | ||
31 | #include "ioctl_tsg.h" | ||
32 | |||
33 | #include <nvgpu/hw/gk20a/hw_ctxsw_prog_gk20a.h> | ||
34 | #include <nvgpu/hw/gk20a/hw_gr_gk20a.h> | ||
35 | |||
36 | ssize_t gk20a_sched_dev_read(struct file *filp, char __user *buf, | ||
37 | size_t size, loff_t *off) | ||
38 | { | ||
39 | struct gk20a *g = filp->private_data; | ||
40 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
41 | struct nvgpu_sched_event_arg event = { 0 }; | ||
42 | int err; | ||
43 | |||
44 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, | ||
45 | "filp=%p buf=%p size=%zu", filp, buf, size); | ||
46 | |||
47 | if (size < sizeof(event)) | ||
48 | return -EINVAL; | ||
49 | size = sizeof(event); | ||
50 | |||
51 | nvgpu_mutex_acquire(&sched->status_lock); | ||
52 | while (!sched->status) { | ||
53 | nvgpu_mutex_release(&sched->status_lock); | ||
54 | if (filp->f_flags & O_NONBLOCK) | ||
55 | return -EAGAIN; | ||
56 | err = NVGPU_COND_WAIT_INTERRUPTIBLE(&sched->readout_wq, | ||
57 | sched->status, 0); | ||
58 | if (err) | ||
59 | return err; | ||
60 | nvgpu_mutex_acquire(&sched->status_lock); | ||
61 | } | ||
62 | |||
63 | event.reserved = 0; | ||
64 | event.status = sched->status; | ||
65 | |||
66 | if (copy_to_user(buf, &event, size)) { | ||
67 | nvgpu_mutex_release(&sched->status_lock); | ||
68 | return -EFAULT; | ||
69 | } | ||
70 | |||
71 | sched->status = 0; | ||
72 | |||
73 | nvgpu_mutex_release(&sched->status_lock); | ||
74 | |||
75 | return size; | ||
76 | } | ||
77 | |||
78 | unsigned int gk20a_sched_dev_poll(struct file *filp, poll_table *wait) | ||
79 | { | ||
80 | struct gk20a *g = filp->private_data; | ||
81 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
82 | unsigned int mask = 0; | ||
83 | |||
84 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, " "); | ||
85 | |||
86 | nvgpu_mutex_acquire(&sched->status_lock); | ||
87 | poll_wait(filp, &sched->readout_wq.wq, wait); | ||
88 | if (sched->status) | ||
89 | mask |= POLLIN | POLLRDNORM; | ||
90 | nvgpu_mutex_release(&sched->status_lock); | ||
91 | |||
92 | return mask; | ||
93 | } | ||
94 | |||
95 | static int gk20a_sched_dev_ioctl_get_tsgs(struct gk20a *g, | ||
96 | struct nvgpu_sched_get_tsgs_args *arg) | ||
97 | { | ||
98 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
99 | |||
100 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "size=%u buffer=%llx", | ||
101 | arg->size, arg->buffer); | ||
102 | |||
103 | if ((arg->size < sched->bitmap_size) || (!arg->buffer)) { | ||
104 | arg->size = sched->bitmap_size; | ||
105 | return -ENOSPC; | ||
106 | } | ||
107 | |||
108 | nvgpu_mutex_acquire(&sched->status_lock); | ||
109 | if (copy_to_user((void __user *)(uintptr_t)arg->buffer, | ||
110 | sched->active_tsg_bitmap, sched->bitmap_size)) { | ||
111 | nvgpu_mutex_release(&sched->status_lock); | ||
112 | return -EFAULT; | ||
113 | } | ||
114 | nvgpu_mutex_release(&sched->status_lock); | ||
115 | |||
116 | return 0; | ||
117 | } | ||
118 | |||
119 | static int gk20a_sched_dev_ioctl_get_recent_tsgs(struct gk20a *g, | ||
120 | struct nvgpu_sched_get_tsgs_args *arg) | ||
121 | { | ||
122 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
123 | |||
124 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "size=%u buffer=%llx", | ||
125 | arg->size, arg->buffer); | ||
126 | |||
127 | if ((arg->size < sched->bitmap_size) || (!arg->buffer)) { | ||
128 | arg->size = sched->bitmap_size; | ||
129 | return -ENOSPC; | ||
130 | } | ||
131 | |||
132 | nvgpu_mutex_acquire(&sched->status_lock); | ||
133 | if (copy_to_user((void __user *)(uintptr_t)arg->buffer, | ||
134 | sched->recent_tsg_bitmap, sched->bitmap_size)) { | ||
135 | nvgpu_mutex_release(&sched->status_lock); | ||
136 | return -EFAULT; | ||
137 | } | ||
138 | |||
139 | memset(sched->recent_tsg_bitmap, 0, sched->bitmap_size); | ||
140 | nvgpu_mutex_release(&sched->status_lock); | ||
141 | |||
142 | return 0; | ||
143 | } | ||
144 | |||
145 | static int gk20a_sched_dev_ioctl_get_tsgs_by_pid(struct gk20a *g, | ||
146 | struct nvgpu_sched_get_tsgs_by_pid_args *arg) | ||
147 | { | ||
148 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
149 | struct fifo_gk20a *f = &g->fifo; | ||
150 | struct tsg_gk20a *tsg; | ||
151 | u64 *bitmap; | ||
152 | unsigned int tsgid; | ||
153 | /* pid at user level corresponds to kernel tgid */ | ||
154 | pid_t tgid = (pid_t)arg->pid; | ||
155 | int err = 0; | ||
156 | |||
157 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "pid=%d size=%u buffer=%llx", | ||
158 | (pid_t)arg->pid, arg->size, arg->buffer); | ||
159 | |||
160 | if ((arg->size < sched->bitmap_size) || (!arg->buffer)) { | ||
161 | arg->size = sched->bitmap_size; | ||
162 | return -ENOSPC; | ||
163 | } | ||
164 | |||
165 | bitmap = nvgpu_kzalloc(g, sched->bitmap_size); | ||
166 | if (!bitmap) | ||
167 | return -ENOMEM; | ||
168 | |||
169 | nvgpu_mutex_acquire(&sched->status_lock); | ||
170 | for (tsgid = 0; tsgid < f->num_channels; tsgid++) { | ||
171 | if (NVGPU_SCHED_ISSET(tsgid, sched->active_tsg_bitmap)) { | ||
172 | tsg = &f->tsg[tsgid]; | ||
173 | if (tsg->tgid == tgid) | ||
174 | NVGPU_SCHED_SET(tsgid, bitmap); | ||
175 | } | ||
176 | } | ||
177 | nvgpu_mutex_release(&sched->status_lock); | ||
178 | |||
179 | if (copy_to_user((void __user *)(uintptr_t)arg->buffer, | ||
180 | bitmap, sched->bitmap_size)) | ||
181 | err = -EFAULT; | ||
182 | |||
183 | nvgpu_kfree(g, bitmap); | ||
184 | |||
185 | return err; | ||
186 | } | ||
187 | |||
188 | static int gk20a_sched_dev_ioctl_get_params(struct gk20a *g, | ||
189 | struct nvgpu_sched_tsg_get_params_args *arg) | ||
190 | { | ||
191 | struct fifo_gk20a *f = &g->fifo; | ||
192 | struct tsg_gk20a *tsg; | ||
193 | u32 tsgid = arg->tsgid; | ||
194 | |||
195 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsgid); | ||
196 | |||
197 | if (tsgid >= f->num_channels) | ||
198 | return -EINVAL; | ||
199 | |||
200 | nvgpu_speculation_barrier(); | ||
201 | |||
202 | tsg = &f->tsg[tsgid]; | ||
203 | if (!nvgpu_ref_get_unless_zero(&tsg->refcount)) | ||
204 | return -ENXIO; | ||
205 | |||
206 | arg->pid = tsg->tgid; /* kernel tgid corresponds to user pid */ | ||
207 | arg->runlist_interleave = tsg->interleave_level; | ||
208 | arg->timeslice = gk20a_tsg_get_timeslice(tsg); | ||
209 | |||
210 | arg->graphics_preempt_mode = | ||
211 | tsg->gr_ctx.graphics_preempt_mode; | ||
212 | arg->compute_preempt_mode = | ||
213 | tsg->gr_ctx.compute_preempt_mode; | ||
214 | |||
215 | nvgpu_ref_put(&tsg->refcount, nvgpu_ioctl_tsg_release); | ||
216 | |||
217 | return 0; | ||
218 | } | ||
219 | |||
220 | static int gk20a_sched_dev_ioctl_tsg_set_timeslice( | ||
221 | struct gk20a *g, | ||
222 | struct nvgpu_sched_tsg_timeslice_args *arg) | ||
223 | { | ||
224 | struct fifo_gk20a *f = &g->fifo; | ||
225 | struct tsg_gk20a *tsg; | ||
226 | u32 tsgid = arg->tsgid; | ||
227 | int err; | ||
228 | |||
229 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsgid); | ||
230 | |||
231 | if (tsgid >= f->num_channels) | ||
232 | return -EINVAL; | ||
233 | |||
234 | nvgpu_speculation_barrier(); | ||
235 | |||
236 | tsg = &f->tsg[tsgid]; | ||
237 | if (!nvgpu_ref_get_unless_zero(&tsg->refcount)) | ||
238 | return -ENXIO; | ||
239 | |||
240 | err = gk20a_busy(g); | ||
241 | if (err) | ||
242 | goto done; | ||
243 | |||
244 | err = gk20a_tsg_set_timeslice(tsg, arg->timeslice); | ||
245 | |||
246 | gk20a_idle(g); | ||
247 | |||
248 | done: | ||
249 | nvgpu_ref_put(&tsg->refcount, nvgpu_ioctl_tsg_release); | ||
250 | |||
251 | return err; | ||
252 | } | ||
253 | |||
254 | static int gk20a_sched_dev_ioctl_tsg_set_runlist_interleave( | ||
255 | struct gk20a *g, | ||
256 | struct nvgpu_sched_tsg_runlist_interleave_args *arg) | ||
257 | { | ||
258 | struct fifo_gk20a *f = &g->fifo; | ||
259 | struct tsg_gk20a *tsg; | ||
260 | u32 tsgid = arg->tsgid; | ||
261 | int err; | ||
262 | |||
263 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsgid); | ||
264 | |||
265 | if (tsgid >= f->num_channels) | ||
266 | return -EINVAL; | ||
267 | |||
268 | nvgpu_speculation_barrier(); | ||
269 | |||
270 | tsg = &f->tsg[tsgid]; | ||
271 | if (!nvgpu_ref_get_unless_zero(&tsg->refcount)) | ||
272 | return -ENXIO; | ||
273 | |||
274 | err = gk20a_busy(g); | ||
275 | if (err) | ||
276 | goto done; | ||
277 | |||
278 | err = gk20a_tsg_set_runlist_interleave(tsg, arg->runlist_interleave); | ||
279 | |||
280 | gk20a_idle(g); | ||
281 | |||
282 | done: | ||
283 | nvgpu_ref_put(&tsg->refcount, nvgpu_ioctl_tsg_release); | ||
284 | |||
285 | return err; | ||
286 | } | ||
287 | |||
288 | static int gk20a_sched_dev_ioctl_lock_control(struct gk20a *g) | ||
289 | { | ||
290 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
291 | |||
292 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, " "); | ||
293 | |||
294 | nvgpu_mutex_acquire(&sched->control_lock); | ||
295 | sched->control_locked = true; | ||
296 | nvgpu_mutex_release(&sched->control_lock); | ||
297 | return 0; | ||
298 | } | ||
299 | |||
300 | static int gk20a_sched_dev_ioctl_unlock_control(struct gk20a *g) | ||
301 | { | ||
302 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
303 | |||
304 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, " "); | ||
305 | |||
306 | nvgpu_mutex_acquire(&sched->control_lock); | ||
307 | sched->control_locked = false; | ||
308 | nvgpu_mutex_release(&sched->control_lock); | ||
309 | return 0; | ||
310 | } | ||
311 | |||
312 | static int gk20a_sched_dev_ioctl_get_api_version(struct gk20a *g, | ||
313 | struct nvgpu_sched_api_version_args *args) | ||
314 | { | ||
315 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, " "); | ||
316 | |||
317 | args->version = NVGPU_SCHED_API_VERSION; | ||
318 | return 0; | ||
319 | } | ||
320 | |||
321 | static int gk20a_sched_dev_ioctl_get_tsg(struct gk20a *g, | ||
322 | struct nvgpu_sched_tsg_refcount_args *arg) | ||
323 | { | ||
324 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
325 | struct fifo_gk20a *f = &g->fifo; | ||
326 | struct tsg_gk20a *tsg; | ||
327 | u32 tsgid = arg->tsgid; | ||
328 | |||
329 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsgid); | ||
330 | |||
331 | if (tsgid >= f->num_channels) | ||
332 | return -EINVAL; | ||
333 | |||
334 | nvgpu_speculation_barrier(); | ||
335 | |||
336 | tsg = &f->tsg[tsgid]; | ||
337 | if (!nvgpu_ref_get_unless_zero(&tsg->refcount)) | ||
338 | return -ENXIO; | ||
339 | |||
340 | nvgpu_mutex_acquire(&sched->status_lock); | ||
341 | if (NVGPU_SCHED_ISSET(tsgid, sched->ref_tsg_bitmap)) { | ||
342 | nvgpu_warn(g, "tsgid=%d already referenced", tsgid); | ||
343 | /* unlock status_lock as nvgpu_ioctl_tsg_release locks it */ | ||
344 | nvgpu_mutex_release(&sched->status_lock); | ||
345 | nvgpu_ref_put(&tsg->refcount, nvgpu_ioctl_tsg_release); | ||
346 | return -ENXIO; | ||
347 | } | ||
348 | |||
349 | /* keep reference on TSG, will be released on | ||
350 | * NVGPU_SCHED_IOCTL_PUT_TSG ioctl, or close | ||
351 | */ | ||
352 | NVGPU_SCHED_SET(tsgid, sched->ref_tsg_bitmap); | ||
353 | nvgpu_mutex_release(&sched->status_lock); | ||
354 | |||
355 | return 0; | ||
356 | } | ||
357 | |||
358 | static int gk20a_sched_dev_ioctl_put_tsg(struct gk20a *g, | ||
359 | struct nvgpu_sched_tsg_refcount_args *arg) | ||
360 | { | ||
361 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
362 | struct fifo_gk20a *f = &g->fifo; | ||
363 | struct tsg_gk20a *tsg; | ||
364 | u32 tsgid = arg->tsgid; | ||
365 | |||
366 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsgid); | ||
367 | |||
368 | if (tsgid >= f->num_channels) | ||
369 | return -EINVAL; | ||
370 | |||
371 | nvgpu_speculation_barrier(); | ||
372 | |||
373 | nvgpu_mutex_acquire(&sched->status_lock); | ||
374 | if (!NVGPU_SCHED_ISSET(tsgid, sched->ref_tsg_bitmap)) { | ||
375 | nvgpu_mutex_release(&sched->status_lock); | ||
376 | nvgpu_warn(g, "tsgid=%d not previously referenced", tsgid); | ||
377 | return -ENXIO; | ||
378 | } | ||
379 | NVGPU_SCHED_CLR(tsgid, sched->ref_tsg_bitmap); | ||
380 | nvgpu_mutex_release(&sched->status_lock); | ||
381 | |||
382 | tsg = &f->tsg[tsgid]; | ||
383 | nvgpu_ref_put(&tsg->refcount, nvgpu_ioctl_tsg_release); | ||
384 | |||
385 | return 0; | ||
386 | } | ||
387 | |||
388 | int gk20a_sched_dev_open(struct inode *inode, struct file *filp) | ||
389 | { | ||
390 | struct nvgpu_os_linux *l = container_of(inode->i_cdev, | ||
391 | struct nvgpu_os_linux, sched.cdev); | ||
392 | struct gk20a *g; | ||
393 | struct nvgpu_sched_ctrl *sched; | ||
394 | int err = 0; | ||
395 | |||
396 | g = gk20a_get(&l->g); | ||
397 | if (!g) | ||
398 | return -ENODEV; | ||
399 | sched = &g->sched_ctrl; | ||
400 | |||
401 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "g=%p", g); | ||
402 | |||
403 | if (!sched->sw_ready) { | ||
404 | err = gk20a_busy(g); | ||
405 | if (err) | ||
406 | goto free_ref; | ||
407 | |||
408 | gk20a_idle(g); | ||
409 | } | ||
410 | |||
411 | if (!nvgpu_mutex_tryacquire(&sched->busy_lock)) { | ||
412 | err = -EBUSY; | ||
413 | goto free_ref; | ||
414 | } | ||
415 | |||
416 | memcpy(sched->recent_tsg_bitmap, sched->active_tsg_bitmap, | ||
417 | sched->bitmap_size); | ||
418 | memset(sched->ref_tsg_bitmap, 0, sched->bitmap_size); | ||
419 | |||
420 | filp->private_data = g; | ||
421 | nvgpu_log(g, gpu_dbg_sched, "filp=%p sched=%p", filp, sched); | ||
422 | |||
423 | free_ref: | ||
424 | if (err) | ||
425 | gk20a_put(g); | ||
426 | return err; | ||
427 | } | ||
428 | |||
429 | long gk20a_sched_dev_ioctl(struct file *filp, unsigned int cmd, | ||
430 | unsigned long arg) | ||
431 | { | ||
432 | struct gk20a *g = filp->private_data; | ||
433 | u8 buf[NVGPU_CTXSW_IOCTL_MAX_ARG_SIZE]; | ||
434 | int err = 0; | ||
435 | |||
436 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "nr=%d", _IOC_NR(cmd)); | ||
437 | |||
438 | if ((_IOC_TYPE(cmd) != NVGPU_SCHED_IOCTL_MAGIC) || | ||
439 | (_IOC_NR(cmd) == 0) || | ||
440 | (_IOC_NR(cmd) > NVGPU_SCHED_IOCTL_LAST) || | ||
441 | (_IOC_SIZE(cmd) > NVGPU_SCHED_IOCTL_MAX_ARG_SIZE)) | ||
442 | return -EINVAL; | ||
443 | |||
444 | memset(buf, 0, sizeof(buf)); | ||
445 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | ||
446 | if (copy_from_user(buf, (void __user *)arg, _IOC_SIZE(cmd))) | ||
447 | return -EFAULT; | ||
448 | } | ||
449 | |||
450 | nvgpu_speculation_barrier(); | ||
451 | switch (cmd) { | ||
452 | case NVGPU_SCHED_IOCTL_GET_TSGS: | ||
453 | err = gk20a_sched_dev_ioctl_get_tsgs(g, | ||
454 | (struct nvgpu_sched_get_tsgs_args *)buf); | ||
455 | break; | ||
456 | case NVGPU_SCHED_IOCTL_GET_RECENT_TSGS: | ||
457 | err = gk20a_sched_dev_ioctl_get_recent_tsgs(g, | ||
458 | (struct nvgpu_sched_get_tsgs_args *)buf); | ||
459 | break; | ||
460 | case NVGPU_SCHED_IOCTL_GET_TSGS_BY_PID: | ||
461 | err = gk20a_sched_dev_ioctl_get_tsgs_by_pid(g, | ||
462 | (struct nvgpu_sched_get_tsgs_by_pid_args *)buf); | ||
463 | break; | ||
464 | case NVGPU_SCHED_IOCTL_TSG_GET_PARAMS: | ||
465 | err = gk20a_sched_dev_ioctl_get_params(g, | ||
466 | (struct nvgpu_sched_tsg_get_params_args *)buf); | ||
467 | break; | ||
468 | case NVGPU_SCHED_IOCTL_TSG_SET_TIMESLICE: | ||
469 | err = gk20a_sched_dev_ioctl_tsg_set_timeslice(g, | ||
470 | (struct nvgpu_sched_tsg_timeslice_args *)buf); | ||
471 | break; | ||
472 | case NVGPU_SCHED_IOCTL_TSG_SET_RUNLIST_INTERLEAVE: | ||
473 | err = gk20a_sched_dev_ioctl_tsg_set_runlist_interleave(g, | ||
474 | (struct nvgpu_sched_tsg_runlist_interleave_args *)buf); | ||
475 | break; | ||
476 | case NVGPU_SCHED_IOCTL_LOCK_CONTROL: | ||
477 | err = gk20a_sched_dev_ioctl_lock_control(g); | ||
478 | break; | ||
479 | case NVGPU_SCHED_IOCTL_UNLOCK_CONTROL: | ||
480 | err = gk20a_sched_dev_ioctl_unlock_control(g); | ||
481 | break; | ||
482 | case NVGPU_SCHED_IOCTL_GET_API_VERSION: | ||
483 | err = gk20a_sched_dev_ioctl_get_api_version(g, | ||
484 | (struct nvgpu_sched_api_version_args *)buf); | ||
485 | break; | ||
486 | case NVGPU_SCHED_IOCTL_GET_TSG: | ||
487 | err = gk20a_sched_dev_ioctl_get_tsg(g, | ||
488 | (struct nvgpu_sched_tsg_refcount_args *)buf); | ||
489 | break; | ||
490 | case NVGPU_SCHED_IOCTL_PUT_TSG: | ||
491 | err = gk20a_sched_dev_ioctl_put_tsg(g, | ||
492 | (struct nvgpu_sched_tsg_refcount_args *)buf); | ||
493 | break; | ||
494 | default: | ||
495 | nvgpu_log_info(g, "unrecognized gpu ioctl cmd: 0x%x", cmd); | ||
496 | err = -ENOTTY; | ||
497 | } | ||
498 | |||
499 | /* Some ioctls like NVGPU_SCHED_IOCTL_GET_TSGS might be called on | ||
500 | * purpose with NULL buffer and/or zero size to discover TSG bitmap | ||
501 | * size. We need to update user arguments in this case too, even | ||
502 | * if we return an error. | ||
503 | */ | ||
504 | if ((!err || (err == -ENOSPC)) && (_IOC_DIR(cmd) & _IOC_READ)) { | ||
505 | if (copy_to_user((void __user *)arg, buf, _IOC_SIZE(cmd))) | ||
506 | err = -EFAULT; | ||
507 | } | ||
508 | |||
509 | return err; | ||
510 | } | ||
511 | |||
512 | int gk20a_sched_dev_release(struct inode *inode, struct file *filp) | ||
513 | { | ||
514 | struct gk20a *g = filp->private_data; | ||
515 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
516 | struct fifo_gk20a *f = &g->fifo; | ||
517 | struct tsg_gk20a *tsg; | ||
518 | unsigned int tsgid; | ||
519 | |||
520 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "sched: %p", sched); | ||
521 | |||
522 | /* release any reference to TSGs */ | ||
523 | for (tsgid = 0; tsgid < f->num_channels; tsgid++) { | ||
524 | if (NVGPU_SCHED_ISSET(tsgid, sched->ref_tsg_bitmap)) { | ||
525 | tsg = &f->tsg[tsgid]; | ||
526 | nvgpu_ref_put(&tsg->refcount, nvgpu_ioctl_tsg_release); | ||
527 | } | ||
528 | } | ||
529 | |||
530 | /* unlock control */ | ||
531 | nvgpu_mutex_acquire(&sched->control_lock); | ||
532 | sched->control_locked = false; | ||
533 | nvgpu_mutex_release(&sched->control_lock); | ||
534 | |||
535 | nvgpu_mutex_release(&sched->busy_lock); | ||
536 | gk20a_put(g); | ||
537 | return 0; | ||
538 | } | ||
539 | |||
540 | void gk20a_sched_ctrl_tsg_added(struct gk20a *g, struct tsg_gk20a *tsg) | ||
541 | { | ||
542 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
543 | int err; | ||
544 | |||
545 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsg->tsgid); | ||
546 | |||
547 | if (!sched->sw_ready) { | ||
548 | err = gk20a_busy(g); | ||
549 | if (err) { | ||
550 | WARN_ON(err); | ||
551 | return; | ||
552 | } | ||
553 | |||
554 | gk20a_idle(g); | ||
555 | } | ||
556 | |||
557 | nvgpu_mutex_acquire(&sched->status_lock); | ||
558 | NVGPU_SCHED_SET(tsg->tsgid, sched->active_tsg_bitmap); | ||
559 | NVGPU_SCHED_SET(tsg->tsgid, sched->recent_tsg_bitmap); | ||
560 | sched->status |= NVGPU_SCHED_STATUS_TSG_OPEN; | ||
561 | nvgpu_mutex_release(&sched->status_lock); | ||
562 | nvgpu_cond_signal_interruptible(&sched->readout_wq); | ||
563 | } | ||
564 | |||
565 | void gk20a_sched_ctrl_tsg_removed(struct gk20a *g, struct tsg_gk20a *tsg) | ||
566 | { | ||
567 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
568 | |||
569 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsg->tsgid); | ||
570 | |||
571 | nvgpu_mutex_acquire(&sched->status_lock); | ||
572 | NVGPU_SCHED_CLR(tsg->tsgid, sched->active_tsg_bitmap); | ||
573 | |||
574 | /* clear recent_tsg_bitmap as well: if app manager did not | ||
575 | * notice that TSG was previously added, no need to notify it | ||
576 | * if the TSG has been released in the meantime. If the | ||
577 | * TSG gets reallocated, app manager will be notified as usual. | ||
578 | */ | ||
579 | NVGPU_SCHED_CLR(tsg->tsgid, sched->recent_tsg_bitmap); | ||
580 | |||
581 | /* do not set event_pending, we only want to notify app manager | ||
582 | * when TSGs are added, so that it can apply sched params | ||
583 | */ | ||
584 | nvgpu_mutex_release(&sched->status_lock); | ||
585 | } | ||
586 | |||
587 | int gk20a_sched_ctrl_init(struct gk20a *g) | ||
588 | { | ||
589 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
590 | struct fifo_gk20a *f = &g->fifo; | ||
591 | int err; | ||
592 | |||
593 | if (sched->sw_ready) | ||
594 | return 0; | ||
595 | |||
596 | sched->bitmap_size = roundup(f->num_channels, 64) / 8; | ||
597 | sched->status = 0; | ||
598 | |||
599 | nvgpu_log(g, gpu_dbg_fn | gpu_dbg_sched, "g=%p sched=%p size=%zu", | ||
600 | g, sched, sched->bitmap_size); | ||
601 | |||
602 | sched->active_tsg_bitmap = nvgpu_kzalloc(g, sched->bitmap_size); | ||
603 | if (!sched->active_tsg_bitmap) | ||
604 | return -ENOMEM; | ||
605 | |||
606 | sched->recent_tsg_bitmap = nvgpu_kzalloc(g, sched->bitmap_size); | ||
607 | if (!sched->recent_tsg_bitmap) { | ||
608 | err = -ENOMEM; | ||
609 | goto free_active; | ||
610 | } | ||
611 | |||
612 | sched->ref_tsg_bitmap = nvgpu_kzalloc(g, sched->bitmap_size); | ||
613 | if (!sched->ref_tsg_bitmap) { | ||
614 | err = -ENOMEM; | ||
615 | goto free_recent; | ||
616 | } | ||
617 | |||
618 | nvgpu_cond_init(&sched->readout_wq); | ||
619 | |||
620 | err = nvgpu_mutex_init(&sched->status_lock); | ||
621 | if (err) | ||
622 | goto free_ref; | ||
623 | |||
624 | err = nvgpu_mutex_init(&sched->control_lock); | ||
625 | if (err) | ||
626 | goto free_status_lock; | ||
627 | |||
628 | err = nvgpu_mutex_init(&sched->busy_lock); | ||
629 | if (err) | ||
630 | goto free_control_lock; | ||
631 | |||
632 | sched->sw_ready = true; | ||
633 | |||
634 | return 0; | ||
635 | |||
636 | free_control_lock: | ||
637 | nvgpu_mutex_destroy(&sched->control_lock); | ||
638 | free_status_lock: | ||
639 | nvgpu_mutex_destroy(&sched->status_lock); | ||
640 | free_ref: | ||
641 | nvgpu_kfree(g, sched->ref_tsg_bitmap); | ||
642 | free_recent: | ||
643 | nvgpu_kfree(g, sched->recent_tsg_bitmap); | ||
644 | free_active: | ||
645 | nvgpu_kfree(g, sched->active_tsg_bitmap); | ||
646 | |||
647 | return err; | ||
648 | } | ||
649 | |||
650 | void gk20a_sched_ctrl_cleanup(struct gk20a *g) | ||
651 | { | ||
652 | struct nvgpu_sched_ctrl *sched = &g->sched_ctrl; | ||
653 | |||
654 | nvgpu_kfree(g, sched->active_tsg_bitmap); | ||
655 | nvgpu_kfree(g, sched->recent_tsg_bitmap); | ||
656 | nvgpu_kfree(g, sched->ref_tsg_bitmap); | ||
657 | sched->active_tsg_bitmap = NULL; | ||
658 | sched->recent_tsg_bitmap = NULL; | ||
659 | sched->ref_tsg_bitmap = NULL; | ||
660 | |||
661 | nvgpu_mutex_destroy(&sched->status_lock); | ||
662 | nvgpu_mutex_destroy(&sched->control_lock); | ||
663 | nvgpu_mutex_destroy(&sched->busy_lock); | ||
664 | |||
665 | sched->sw_ready = false; | ||
666 | } | ||