diff options
Diffstat (limited to 'drivers/gpu/nvgpu/common/linux/ioctl_channel.c')
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/ioctl_channel.c | 1452 |
1 files changed, 1452 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/common/linux/ioctl_channel.c b/drivers/gpu/nvgpu/common/linux/ioctl_channel.c new file mode 100644 index 00000000..31651795 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/ioctl_channel.c | |||
@@ -0,0 +1,1452 @@ | |||
1 | /* | ||
2 | * GK20A Graphics channel | ||
3 | * | ||
4 | * Copyright (c) 2011-2017, NVIDIA CORPORATION. All rights reserved. | ||
5 | * | ||
6 | * This program is free software; you can redistribute it and/or modify it | ||
7 | * under the terms and conditions of the GNU General Public License, | ||
8 | * version 2, as published by the Free Software Foundation. | ||
9 | * | ||
10 | * This program is distributed in the hope it will be useful, but WITHOUT | ||
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | ||
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | ||
13 | * more details. | ||
14 | * | ||
15 | * You should have received a copy of the GNU General Public License | ||
16 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
17 | */ | ||
18 | |||
19 | #include <trace/events/gk20a.h> | ||
20 | #include <linux/file.h> | ||
21 | #include <linux/anon_inodes.h> | ||
22 | #include <linux/dma-buf.h> | ||
23 | #include <linux/poll.h> | ||
24 | #include <uapi/linux/nvgpu.h> | ||
25 | #include <uapi/linux/nvgpu-t18x.h> | ||
26 | |||
27 | #include <nvgpu/semaphore.h> | ||
28 | #include <nvgpu/timers.h> | ||
29 | #include <nvgpu/kmem.h> | ||
30 | #include <nvgpu/log.h> | ||
31 | #include <nvgpu/list.h> | ||
32 | #include <nvgpu/debug.h> | ||
33 | #include <nvgpu/enabled.h> | ||
34 | |||
35 | #include "gk20a/gk20a.h" | ||
36 | #include "gk20a/dbg_gpu_gk20a.h" | ||
37 | #include "gk20a/fence_gk20a.h" | ||
38 | |||
39 | #include "platform_gk20a.h" | ||
40 | #include "ioctl_channel.h" | ||
41 | #include "channel.h" | ||
42 | #include "os_linux.h" | ||
43 | #include "ctxsw_trace.h" | ||
44 | |||
45 | static const char *gr_gk20a_graphics_preempt_mode_name(u32 graphics_preempt_mode) | ||
46 | { | ||
47 | switch (graphics_preempt_mode) { | ||
48 | case NVGPU_PREEMPTION_MODE_GRAPHICS_WFI: | ||
49 | return "WFI"; | ||
50 | default: | ||
51 | return "?"; | ||
52 | } | ||
53 | } | ||
54 | |||
55 | static const char *gr_gk20a_compute_preempt_mode_name(u32 compute_preempt_mode) | ||
56 | { | ||
57 | switch (compute_preempt_mode) { | ||
58 | case NVGPU_PREEMPTION_MODE_COMPUTE_WFI: | ||
59 | return "WFI"; | ||
60 | case NVGPU_PREEMPTION_MODE_COMPUTE_CTA: | ||
61 | return "CTA"; | ||
62 | default: | ||
63 | return "?"; | ||
64 | } | ||
65 | } | ||
66 | |||
67 | static void gk20a_channel_trace_sched_param( | ||
68 | void (*trace)(int chid, int tsgid, pid_t pid, u32 timeslice, | ||
69 | u32 timeout, const char *interleave, | ||
70 | const char *graphics_preempt_mode, | ||
71 | const char *compute_preempt_mode), | ||
72 | struct channel_gk20a *ch) | ||
73 | { | ||
74 | (trace)(ch->chid, ch->tsgid, ch->pid, | ||
75 | gk20a_is_channel_marked_as_tsg(ch) ? | ||
76 | tsg_gk20a_from_ch(ch)->timeslice_us : ch->timeslice_us, | ||
77 | ch->timeout_ms_max, | ||
78 | gk20a_fifo_interleave_level_name(ch->interleave_level), | ||
79 | gr_gk20a_graphics_preempt_mode_name(ch->ch_ctx.gr_ctx ? | ||
80 | ch->ch_ctx.gr_ctx->graphics_preempt_mode : 0), | ||
81 | gr_gk20a_compute_preempt_mode_name(ch->ch_ctx.gr_ctx ? | ||
82 | ch->ch_ctx.gr_ctx->compute_preempt_mode : 0)); | ||
83 | } | ||
84 | |||
85 | /* | ||
86 | * Although channels do have pointers back to the gk20a struct that they were | ||
87 | * created under in cases where the driver is killed that pointer can be bad. | ||
88 | * The channel memory can be freed before the release() function for a given | ||
89 | * channel is called. This happens when the driver dies and userspace doesn't | ||
90 | * get a chance to call release() until after the entire gk20a driver data is | ||
91 | * unloaded and freed. | ||
92 | */ | ||
93 | struct channel_priv { | ||
94 | struct gk20a *g; | ||
95 | struct channel_gk20a *c; | ||
96 | }; | ||
97 | |||
98 | #if defined(CONFIG_GK20A_CYCLE_STATS) | ||
99 | |||
100 | static int gk20a_channel_cycle_stats(struct channel_gk20a *ch, | ||
101 | struct nvgpu_cycle_stats_args *args) | ||
102 | { | ||
103 | struct dma_buf *dmabuf; | ||
104 | void *virtual_address; | ||
105 | |||
106 | /* is it allowed to handle calls for current GPU? */ | ||
107 | if (!nvgpu_is_enabled(ch->g, NVGPU_SUPPORT_CYCLE_STATS)) | ||
108 | return -ENOSYS; | ||
109 | |||
110 | if (args->dmabuf_fd && !ch->cyclestate.cyclestate_buffer_handler) { | ||
111 | |||
112 | /* set up new cyclestats buffer */ | ||
113 | dmabuf = dma_buf_get(args->dmabuf_fd); | ||
114 | if (IS_ERR(dmabuf)) | ||
115 | return PTR_ERR(dmabuf); | ||
116 | virtual_address = dma_buf_vmap(dmabuf); | ||
117 | if (!virtual_address) | ||
118 | return -ENOMEM; | ||
119 | |||
120 | ch->cyclestate.cyclestate_buffer_handler = dmabuf; | ||
121 | ch->cyclestate.cyclestate_buffer = virtual_address; | ||
122 | ch->cyclestate.cyclestate_buffer_size = dmabuf->size; | ||
123 | return 0; | ||
124 | |||
125 | } else if (!args->dmabuf_fd && | ||
126 | ch->cyclestate.cyclestate_buffer_handler) { | ||
127 | gk20a_channel_free_cycle_stats_buffer(ch); | ||
128 | return 0; | ||
129 | |||
130 | } else if (!args->dmabuf_fd && | ||
131 | !ch->cyclestate.cyclestate_buffer_handler) { | ||
132 | /* no requst from GL */ | ||
133 | return 0; | ||
134 | |||
135 | } else { | ||
136 | pr_err("channel already has cyclestats buffer\n"); | ||
137 | return -EINVAL; | ||
138 | } | ||
139 | } | ||
140 | |||
141 | static int gk20a_flush_cycle_stats_snapshot(struct channel_gk20a *ch) | ||
142 | { | ||
143 | int ret; | ||
144 | |||
145 | nvgpu_mutex_acquire(&ch->cs_client_mutex); | ||
146 | if (ch->cs_client) | ||
147 | ret = gr_gk20a_css_flush(ch, ch->cs_client); | ||
148 | else | ||
149 | ret = -EBADF; | ||
150 | nvgpu_mutex_release(&ch->cs_client_mutex); | ||
151 | |||
152 | return ret; | ||
153 | } | ||
154 | |||
155 | static int gk20a_attach_cycle_stats_snapshot(struct channel_gk20a *ch, | ||
156 | u32 dmabuf_fd, | ||
157 | u32 perfmon_id_count, | ||
158 | u32 *perfmon_id_start) | ||
159 | { | ||
160 | int ret; | ||
161 | |||
162 | nvgpu_mutex_acquire(&ch->cs_client_mutex); | ||
163 | if (ch->cs_client) { | ||
164 | ret = -EEXIST; | ||
165 | } else { | ||
166 | ret = gr_gk20a_css_attach(ch, | ||
167 | dmabuf_fd, | ||
168 | perfmon_id_count, | ||
169 | perfmon_id_start, | ||
170 | &ch->cs_client); | ||
171 | } | ||
172 | nvgpu_mutex_release(&ch->cs_client_mutex); | ||
173 | |||
174 | return ret; | ||
175 | } | ||
176 | |||
177 | static int gk20a_channel_cycle_stats_snapshot(struct channel_gk20a *ch, | ||
178 | struct nvgpu_cycle_stats_snapshot_args *args) | ||
179 | { | ||
180 | int ret; | ||
181 | |||
182 | /* is it allowed to handle calls for current GPU? */ | ||
183 | if (!nvgpu_is_enabled(ch->g, NVGPU_SUPPORT_CYCLE_STATS_SNAPSHOT)) | ||
184 | return -ENOSYS; | ||
185 | |||
186 | if (!args->dmabuf_fd) | ||
187 | return -EINVAL; | ||
188 | |||
189 | /* handle the command (most frequent cases first) */ | ||
190 | switch (args->cmd) { | ||
191 | case NVGPU_IOCTL_CHANNEL_CYCLE_STATS_SNAPSHOT_CMD_FLUSH: | ||
192 | ret = gk20a_flush_cycle_stats_snapshot(ch); | ||
193 | args->extra = 0; | ||
194 | break; | ||
195 | |||
196 | case NVGPU_IOCTL_CHANNEL_CYCLE_STATS_SNAPSHOT_CMD_ATTACH: | ||
197 | ret = gk20a_attach_cycle_stats_snapshot(ch, | ||
198 | args->dmabuf_fd, | ||
199 | args->extra, | ||
200 | &args->extra); | ||
201 | break; | ||
202 | |||
203 | case NVGPU_IOCTL_CHANNEL_CYCLE_STATS_SNAPSHOT_CMD_DETACH: | ||
204 | ret = gk20a_channel_free_cycle_stats_snapshot(ch); | ||
205 | args->extra = 0; | ||
206 | break; | ||
207 | |||
208 | default: | ||
209 | pr_err("cyclestats: unknown command %u\n", args->cmd); | ||
210 | ret = -EINVAL; | ||
211 | break; | ||
212 | } | ||
213 | |||
214 | return ret; | ||
215 | } | ||
216 | #endif | ||
217 | |||
218 | static int gk20a_channel_set_wdt_status(struct channel_gk20a *ch, | ||
219 | struct nvgpu_channel_wdt_args *args) | ||
220 | { | ||
221 | if (args->wdt_status == NVGPU_IOCTL_CHANNEL_DISABLE_WDT) | ||
222 | ch->wdt_enabled = false; | ||
223 | else if (args->wdt_status == NVGPU_IOCTL_CHANNEL_ENABLE_WDT) | ||
224 | ch->wdt_enabled = true; | ||
225 | |||
226 | return 0; | ||
227 | } | ||
228 | |||
229 | static void gk20a_channel_free_error_notifiers(struct channel_gk20a *ch) | ||
230 | { | ||
231 | nvgpu_mutex_acquire(&ch->error_notifier_mutex); | ||
232 | if (ch->error_notifier_ref) { | ||
233 | dma_buf_vunmap(ch->error_notifier_ref, ch->error_notifier_va); | ||
234 | dma_buf_put(ch->error_notifier_ref); | ||
235 | ch->error_notifier_ref = NULL; | ||
236 | ch->error_notifier = NULL; | ||
237 | ch->error_notifier_va = NULL; | ||
238 | } | ||
239 | nvgpu_mutex_release(&ch->error_notifier_mutex); | ||
240 | } | ||
241 | |||
242 | static int gk20a_init_error_notifier(struct channel_gk20a *ch, | ||
243 | struct nvgpu_set_error_notifier *args) | ||
244 | { | ||
245 | struct dma_buf *dmabuf; | ||
246 | void *va; | ||
247 | u64 end = args->offset + sizeof(struct nvgpu_notification); | ||
248 | |||
249 | if (!args->mem) { | ||
250 | pr_err("gk20a_init_error_notifier: invalid memory handle\n"); | ||
251 | return -EINVAL; | ||
252 | } | ||
253 | |||
254 | dmabuf = dma_buf_get(args->mem); | ||
255 | |||
256 | gk20a_channel_free_error_notifiers(ch); | ||
257 | |||
258 | if (IS_ERR(dmabuf)) { | ||
259 | pr_err("Invalid handle: %d\n", args->mem); | ||
260 | return -EINVAL; | ||
261 | } | ||
262 | |||
263 | if (end > dmabuf->size || end < sizeof(struct nvgpu_notification)) { | ||
264 | dma_buf_put(dmabuf); | ||
265 | nvgpu_err(ch->g, "gk20a_init_error_notifier: invalid offset"); | ||
266 | return -EINVAL; | ||
267 | } | ||
268 | |||
269 | /* map handle */ | ||
270 | va = dma_buf_vmap(dmabuf); | ||
271 | if (!va) { | ||
272 | dma_buf_put(dmabuf); | ||
273 | pr_err("Cannot map notifier handle\n"); | ||
274 | return -ENOMEM; | ||
275 | } | ||
276 | |||
277 | ch->error_notifier = va + args->offset; | ||
278 | ch->error_notifier_va = va; | ||
279 | memset(ch->error_notifier, 0, sizeof(struct nvgpu_notification)); | ||
280 | |||
281 | /* set channel notifiers pointer */ | ||
282 | nvgpu_mutex_acquire(&ch->error_notifier_mutex); | ||
283 | ch->error_notifier_ref = dmabuf; | ||
284 | nvgpu_mutex_release(&ch->error_notifier_mutex); | ||
285 | |||
286 | return 0; | ||
287 | } | ||
288 | |||
289 | /* | ||
290 | * This returns the channel with a reference. The caller must | ||
291 | * gk20a_channel_put() the ref back after use. | ||
292 | * | ||
293 | * NULL is returned if the channel was not found. | ||
294 | */ | ||
295 | struct channel_gk20a *gk20a_get_channel_from_file(int fd) | ||
296 | { | ||
297 | struct channel_gk20a *ch; | ||
298 | struct channel_priv *priv; | ||
299 | struct file *f = fget(fd); | ||
300 | |||
301 | if (!f) | ||
302 | return NULL; | ||
303 | |||
304 | if (f->f_op != &gk20a_channel_ops) { | ||
305 | fput(f); | ||
306 | return NULL; | ||
307 | } | ||
308 | |||
309 | priv = (struct channel_priv *)f->private_data; | ||
310 | ch = gk20a_channel_get(priv->c); | ||
311 | fput(f); | ||
312 | return ch; | ||
313 | } | ||
314 | |||
315 | int gk20a_channel_release(struct inode *inode, struct file *filp) | ||
316 | { | ||
317 | struct channel_priv *priv = filp->private_data; | ||
318 | struct channel_gk20a *ch; | ||
319 | struct gk20a *g; | ||
320 | |||
321 | int err; | ||
322 | |||
323 | /* We could still end up here even if the channel_open failed, e.g. | ||
324 | * if we ran out of hw channel IDs. | ||
325 | */ | ||
326 | if (!priv) | ||
327 | return 0; | ||
328 | |||
329 | ch = priv->c; | ||
330 | g = priv->g; | ||
331 | |||
332 | err = gk20a_busy(g); | ||
333 | if (err) { | ||
334 | nvgpu_err(g, "failed to release a channel!"); | ||
335 | goto channel_release; | ||
336 | } | ||
337 | |||
338 | trace_gk20a_channel_release(dev_name(dev_from_gk20a(g))); | ||
339 | |||
340 | gk20a_channel_close(ch); | ||
341 | gk20a_channel_free_error_notifiers(ch); | ||
342 | |||
343 | gk20a_idle(g); | ||
344 | |||
345 | channel_release: | ||
346 | gk20a_put(g); | ||
347 | nvgpu_kfree(g, filp->private_data); | ||
348 | filp->private_data = NULL; | ||
349 | return 0; | ||
350 | } | ||
351 | |||
352 | /* note: runlist_id -1 is synonym for the ENGINE_GR_GK20A runlist id */ | ||
353 | static int __gk20a_channel_open(struct gk20a *g, | ||
354 | struct file *filp, s32 runlist_id) | ||
355 | { | ||
356 | int err; | ||
357 | struct channel_gk20a *ch; | ||
358 | struct channel_priv *priv; | ||
359 | |||
360 | gk20a_dbg_fn(""); | ||
361 | |||
362 | g = gk20a_get(g); | ||
363 | if (!g) | ||
364 | return -ENODEV; | ||
365 | |||
366 | trace_gk20a_channel_open(dev_name(dev_from_gk20a(g))); | ||
367 | |||
368 | priv = nvgpu_kzalloc(g, sizeof(*priv)); | ||
369 | if (!priv) { | ||
370 | err = -ENOMEM; | ||
371 | goto free_ref; | ||
372 | } | ||
373 | |||
374 | err = gk20a_busy(g); | ||
375 | if (err) { | ||
376 | nvgpu_err(g, "failed to power on, %d", err); | ||
377 | goto fail_busy; | ||
378 | } | ||
379 | /* All the user space channel should be non privilege */ | ||
380 | ch = gk20a_open_new_channel(g, runlist_id, false); | ||
381 | gk20a_idle(g); | ||
382 | if (!ch) { | ||
383 | nvgpu_err(g, | ||
384 | "failed to get f"); | ||
385 | err = -ENOMEM; | ||
386 | goto fail_busy; | ||
387 | } | ||
388 | |||
389 | gk20a_channel_trace_sched_param( | ||
390 | trace_gk20a_channel_sched_defaults, ch); | ||
391 | |||
392 | priv->g = g; | ||
393 | priv->c = ch; | ||
394 | |||
395 | filp->private_data = priv; | ||
396 | return 0; | ||
397 | |||
398 | fail_busy: | ||
399 | nvgpu_kfree(g, priv); | ||
400 | free_ref: | ||
401 | gk20a_put(g); | ||
402 | return err; | ||
403 | } | ||
404 | |||
405 | int gk20a_channel_open(struct inode *inode, struct file *filp) | ||
406 | { | ||
407 | struct nvgpu_os_linux *l = container_of(inode->i_cdev, | ||
408 | struct nvgpu_os_linux, channel.cdev); | ||
409 | struct gk20a *g = &l->g; | ||
410 | int ret; | ||
411 | |||
412 | gk20a_dbg_fn("start"); | ||
413 | ret = __gk20a_channel_open(g, filp, -1); | ||
414 | |||
415 | gk20a_dbg_fn("end"); | ||
416 | return ret; | ||
417 | } | ||
418 | |||
419 | int gk20a_channel_open_ioctl(struct gk20a *g, | ||
420 | struct nvgpu_channel_open_args *args) | ||
421 | { | ||
422 | int err; | ||
423 | int fd; | ||
424 | struct file *file; | ||
425 | char name[64]; | ||
426 | s32 runlist_id = args->in.runlist_id; | ||
427 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
428 | |||
429 | err = get_unused_fd_flags(O_RDWR); | ||
430 | if (err < 0) | ||
431 | return err; | ||
432 | fd = err; | ||
433 | |||
434 | snprintf(name, sizeof(name), "nvhost-%s-fd%d", | ||
435 | dev_name(dev_from_gk20a(g)), fd); | ||
436 | |||
437 | file = anon_inode_getfile(name, l->channel.cdev.ops, NULL, O_RDWR); | ||
438 | if (IS_ERR(file)) { | ||
439 | err = PTR_ERR(file); | ||
440 | goto clean_up; | ||
441 | } | ||
442 | |||
443 | err = __gk20a_channel_open(g, file, runlist_id); | ||
444 | if (err) | ||
445 | goto clean_up_file; | ||
446 | |||
447 | fd_install(fd, file); | ||
448 | args->out.channel_fd = fd; | ||
449 | return 0; | ||
450 | |||
451 | clean_up_file: | ||
452 | fput(file); | ||
453 | clean_up: | ||
454 | put_unused_fd(fd); | ||
455 | return err; | ||
456 | } | ||
457 | |||
458 | static u32 nvgpu_gpfifo_user_flags_to_common_flags(u32 user_flags) | ||
459 | { | ||
460 | u32 flags = 0; | ||
461 | |||
462 | if (user_flags & NVGPU_ALLOC_GPFIFO_EX_FLAGS_VPR_ENABLED) | ||
463 | flags |= NVGPU_GPFIFO_FLAGS_SUPPORT_VPR; | ||
464 | |||
465 | if (user_flags & NVGPU_ALLOC_GPFIFO_EX_FLAGS_DETERMINISTIC) | ||
466 | flags |= NVGPU_GPFIFO_FLAGS_SUPPORT_DETERMINISTIC; | ||
467 | |||
468 | return flags; | ||
469 | } | ||
470 | |||
471 | static int nvgpu_channel_ioctl_alloc_gpfifo(struct channel_gk20a *c, | ||
472 | unsigned int num_entries, | ||
473 | unsigned int num_inflight_jobs, | ||
474 | u32 user_flags) | ||
475 | { | ||
476 | return gk20a_channel_alloc_gpfifo(c, num_entries, | ||
477 | num_inflight_jobs, | ||
478 | nvgpu_gpfifo_user_flags_to_common_flags(user_flags)); | ||
479 | } | ||
480 | |||
481 | |||
482 | static int gk20a_channel_wait_semaphore(struct channel_gk20a *ch, | ||
483 | ulong id, u32 offset, | ||
484 | u32 payload, u32 timeout) | ||
485 | { | ||
486 | struct dma_buf *dmabuf; | ||
487 | void *data; | ||
488 | u32 *semaphore; | ||
489 | int ret = 0; | ||
490 | |||
491 | /* do not wait if channel has timed out */ | ||
492 | if (ch->has_timedout) | ||
493 | return -ETIMEDOUT; | ||
494 | |||
495 | dmabuf = dma_buf_get(id); | ||
496 | if (IS_ERR(dmabuf)) { | ||
497 | nvgpu_err(ch->g, "invalid notifier nvmap handle 0x%lx", id); | ||
498 | return -EINVAL; | ||
499 | } | ||
500 | |||
501 | data = dma_buf_kmap(dmabuf, offset >> PAGE_SHIFT); | ||
502 | if (!data) { | ||
503 | nvgpu_err(ch->g, "failed to map notifier memory"); | ||
504 | ret = -EINVAL; | ||
505 | goto cleanup_put; | ||
506 | } | ||
507 | |||
508 | semaphore = data + (offset & ~PAGE_MASK); | ||
509 | |||
510 | ret = NVGPU_COND_WAIT_INTERRUPTIBLE( | ||
511 | &ch->semaphore_wq, | ||
512 | *semaphore == payload || ch->has_timedout, | ||
513 | timeout); | ||
514 | |||
515 | dma_buf_kunmap(dmabuf, offset >> PAGE_SHIFT, data); | ||
516 | cleanup_put: | ||
517 | dma_buf_put(dmabuf); | ||
518 | return ret; | ||
519 | } | ||
520 | |||
521 | static int gk20a_channel_wait(struct channel_gk20a *ch, | ||
522 | struct nvgpu_wait_args *args) | ||
523 | { | ||
524 | struct dma_buf *dmabuf; | ||
525 | struct gk20a *g = ch->g; | ||
526 | struct notification *notif; | ||
527 | struct timespec tv; | ||
528 | u64 jiffies; | ||
529 | ulong id; | ||
530 | u32 offset; | ||
531 | int remain, ret = 0; | ||
532 | u64 end; | ||
533 | |||
534 | gk20a_dbg_fn(""); | ||
535 | |||
536 | if (ch->has_timedout) | ||
537 | return -ETIMEDOUT; | ||
538 | |||
539 | switch (args->type) { | ||
540 | case NVGPU_WAIT_TYPE_NOTIFIER: | ||
541 | id = args->condition.notifier.dmabuf_fd; | ||
542 | offset = args->condition.notifier.offset; | ||
543 | end = offset + sizeof(struct notification); | ||
544 | |||
545 | dmabuf = dma_buf_get(id); | ||
546 | if (IS_ERR(dmabuf)) { | ||
547 | nvgpu_err(g, "invalid notifier nvmap handle 0x%lx", | ||
548 | id); | ||
549 | return -EINVAL; | ||
550 | } | ||
551 | |||
552 | if (end > dmabuf->size || end < sizeof(struct notification)) { | ||
553 | dma_buf_put(dmabuf); | ||
554 | nvgpu_err(g, "invalid notifier offset"); | ||
555 | return -EINVAL; | ||
556 | } | ||
557 | |||
558 | notif = dma_buf_vmap(dmabuf); | ||
559 | if (!notif) { | ||
560 | nvgpu_err(g, "failed to map notifier memory"); | ||
561 | return -ENOMEM; | ||
562 | } | ||
563 | |||
564 | notif = (struct notification *)((uintptr_t)notif + offset); | ||
565 | |||
566 | /* user should set status pending before | ||
567 | * calling this ioctl */ | ||
568 | remain = NVGPU_COND_WAIT_INTERRUPTIBLE( | ||
569 | &ch->notifier_wq, | ||
570 | notif->status == 0 || ch->has_timedout, | ||
571 | args->timeout); | ||
572 | |||
573 | if (remain == 0 && notif->status != 0) { | ||
574 | ret = -ETIMEDOUT; | ||
575 | goto notif_clean_up; | ||
576 | } else if (remain < 0) { | ||
577 | ret = -EINTR; | ||
578 | goto notif_clean_up; | ||
579 | } | ||
580 | |||
581 | /* TBD: fill in correct information */ | ||
582 | jiffies = get_jiffies_64(); | ||
583 | jiffies_to_timespec(jiffies, &tv); | ||
584 | notif->timestamp.nanoseconds[0] = tv.tv_nsec; | ||
585 | notif->timestamp.nanoseconds[1] = tv.tv_sec; | ||
586 | notif->info32 = 0xDEADBEEF; /* should be object name */ | ||
587 | notif->info16 = ch->chid; /* should be method offset */ | ||
588 | |||
589 | notif_clean_up: | ||
590 | dma_buf_vunmap(dmabuf, notif); | ||
591 | return ret; | ||
592 | |||
593 | case NVGPU_WAIT_TYPE_SEMAPHORE: | ||
594 | ret = gk20a_channel_wait_semaphore(ch, | ||
595 | args->condition.semaphore.dmabuf_fd, | ||
596 | args->condition.semaphore.offset, | ||
597 | args->condition.semaphore.payload, | ||
598 | args->timeout); | ||
599 | |||
600 | break; | ||
601 | |||
602 | default: | ||
603 | ret = -EINVAL; | ||
604 | break; | ||
605 | } | ||
606 | |||
607 | return ret; | ||
608 | } | ||
609 | |||
610 | static unsigned int gk20a_event_id_poll(struct file *filep, poll_table *wait) | ||
611 | { | ||
612 | unsigned int mask = 0; | ||
613 | struct gk20a_event_id_data *event_id_data = filep->private_data; | ||
614 | struct gk20a *g = event_id_data->g; | ||
615 | u32 event_id = event_id_data->event_id; | ||
616 | |||
617 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_info, ""); | ||
618 | |||
619 | poll_wait(filep, &event_id_data->event_id_wq.wq, wait); | ||
620 | |||
621 | nvgpu_mutex_acquire(&event_id_data->lock); | ||
622 | |||
623 | if (event_id_data->is_tsg) { | ||
624 | struct tsg_gk20a *tsg = g->fifo.tsg + event_id_data->id; | ||
625 | |||
626 | if (event_id_data->event_posted) { | ||
627 | gk20a_dbg_info( | ||
628 | "found pending event_id=%d on TSG=%d\n", | ||
629 | event_id, tsg->tsgid); | ||
630 | mask = (POLLPRI | POLLIN); | ||
631 | event_id_data->event_posted = false; | ||
632 | } | ||
633 | } else { | ||
634 | struct channel_gk20a *ch = g->fifo.channel | ||
635 | + event_id_data->id; | ||
636 | |||
637 | if (event_id_data->event_posted) { | ||
638 | gk20a_dbg_info( | ||
639 | "found pending event_id=%d on chid=%d\n", | ||
640 | event_id, ch->chid); | ||
641 | mask = (POLLPRI | POLLIN); | ||
642 | event_id_data->event_posted = false; | ||
643 | } | ||
644 | } | ||
645 | |||
646 | nvgpu_mutex_release(&event_id_data->lock); | ||
647 | |||
648 | return mask; | ||
649 | } | ||
650 | |||
651 | static int gk20a_event_id_release(struct inode *inode, struct file *filp) | ||
652 | { | ||
653 | struct gk20a_event_id_data *event_id_data = filp->private_data; | ||
654 | struct gk20a *g = event_id_data->g; | ||
655 | |||
656 | if (event_id_data->is_tsg) { | ||
657 | struct tsg_gk20a *tsg = g->fifo.tsg + event_id_data->id; | ||
658 | |||
659 | nvgpu_mutex_acquire(&tsg->event_id_list_lock); | ||
660 | nvgpu_list_del(&event_id_data->event_id_node); | ||
661 | nvgpu_mutex_release(&tsg->event_id_list_lock); | ||
662 | } else { | ||
663 | struct channel_gk20a *ch = g->fifo.channel + event_id_data->id; | ||
664 | |||
665 | nvgpu_mutex_acquire(&ch->event_id_list_lock); | ||
666 | nvgpu_list_del(&event_id_data->event_id_node); | ||
667 | nvgpu_mutex_release(&ch->event_id_list_lock); | ||
668 | } | ||
669 | |||
670 | nvgpu_mutex_destroy(&event_id_data->lock); | ||
671 | gk20a_put(g); | ||
672 | nvgpu_kfree(g, event_id_data); | ||
673 | filp->private_data = NULL; | ||
674 | |||
675 | return 0; | ||
676 | } | ||
677 | |||
678 | const struct file_operations gk20a_event_id_ops = { | ||
679 | .owner = THIS_MODULE, | ||
680 | .poll = gk20a_event_id_poll, | ||
681 | .release = gk20a_event_id_release, | ||
682 | }; | ||
683 | |||
684 | static int gk20a_channel_get_event_data_from_id(struct channel_gk20a *ch, | ||
685 | u32 event_id, | ||
686 | struct gk20a_event_id_data **event_id_data) | ||
687 | { | ||
688 | struct gk20a_event_id_data *local_event_id_data; | ||
689 | bool event_found = false; | ||
690 | |||
691 | nvgpu_mutex_acquire(&ch->event_id_list_lock); | ||
692 | list_for_each_entry(local_event_id_data, &ch->event_id_list, | ||
693 | event_id_node) { | ||
694 | if (local_event_id_data->event_id == event_id) { | ||
695 | event_found = true; | ||
696 | break; | ||
697 | } | ||
698 | } | ||
699 | nvgpu_mutex_release(&ch->event_id_list_lock); | ||
700 | |||
701 | if (event_found) { | ||
702 | *event_id_data = local_event_id_data; | ||
703 | return 0; | ||
704 | } else { | ||
705 | return -1; | ||
706 | } | ||
707 | } | ||
708 | |||
709 | /* | ||
710 | * Convert common event_id of the form NVGPU_EVENT_ID_* to Linux specific | ||
711 | * event_id of the form NVGPU_IOCTL_CHANNEL_EVENT_ID_* which is used in IOCTLs | ||
712 | */ | ||
713 | u32 nvgpu_event_id_to_ioctl_channel_event_id(u32 event_id) | ||
714 | { | ||
715 | switch (event_id) { | ||
716 | case NVGPU_EVENT_ID_BPT_INT: | ||
717 | return NVGPU_IOCTL_CHANNEL_EVENT_ID_BPT_INT; | ||
718 | case NVGPU_EVENT_ID_BPT_PAUSE: | ||
719 | return NVGPU_IOCTL_CHANNEL_EVENT_ID_BPT_PAUSE; | ||
720 | case NVGPU_EVENT_ID_BLOCKING_SYNC: | ||
721 | return NVGPU_IOCTL_CHANNEL_EVENT_ID_BLOCKING_SYNC; | ||
722 | case NVGPU_EVENT_ID_CILP_PREEMPTION_STARTED: | ||
723 | return NVGPU_IOCTL_CHANNEL_EVENT_ID_CILP_PREEMPTION_STARTED; | ||
724 | case NVGPU_EVENT_ID_CILP_PREEMPTION_COMPLETE: | ||
725 | return NVGPU_IOCTL_CHANNEL_EVENT_ID_CILP_PREEMPTION_COMPLETE; | ||
726 | case NVGPU_EVENT_ID_GR_SEMAPHORE_WRITE_AWAKEN: | ||
727 | return NVGPU_IOCTL_CHANNEL_EVENT_ID_GR_SEMAPHORE_WRITE_AWAKEN; | ||
728 | } | ||
729 | |||
730 | return NVGPU_IOCTL_CHANNEL_EVENT_ID_MAX; | ||
731 | } | ||
732 | |||
733 | void gk20a_channel_event_id_post_event(struct channel_gk20a *ch, | ||
734 | u32 __event_id) | ||
735 | { | ||
736 | struct gk20a_event_id_data *event_id_data; | ||
737 | u32 event_id; | ||
738 | int err = 0; | ||
739 | |||
740 | event_id = nvgpu_event_id_to_ioctl_channel_event_id(__event_id); | ||
741 | if (event_id >= NVGPU_IOCTL_CHANNEL_EVENT_ID_MAX) | ||
742 | return; | ||
743 | |||
744 | err = gk20a_channel_get_event_data_from_id(ch, event_id, | ||
745 | &event_id_data); | ||
746 | if (err) | ||
747 | return; | ||
748 | |||
749 | nvgpu_mutex_acquire(&event_id_data->lock); | ||
750 | |||
751 | gk20a_dbg_info( | ||
752 | "posting event for event_id=%d on ch=%d\n", | ||
753 | event_id, ch->chid); | ||
754 | event_id_data->event_posted = true; | ||
755 | |||
756 | nvgpu_cond_broadcast_interruptible(&event_id_data->event_id_wq); | ||
757 | |||
758 | nvgpu_mutex_release(&event_id_data->lock); | ||
759 | } | ||
760 | |||
761 | static int gk20a_channel_event_id_enable(struct channel_gk20a *ch, | ||
762 | int event_id, | ||
763 | int *fd) | ||
764 | { | ||
765 | struct gk20a *g; | ||
766 | int err = 0; | ||
767 | int local_fd; | ||
768 | struct file *file; | ||
769 | char name[64]; | ||
770 | struct gk20a_event_id_data *event_id_data; | ||
771 | |||
772 | g = gk20a_get(ch->g); | ||
773 | if (!g) | ||
774 | return -ENODEV; | ||
775 | |||
776 | err = gk20a_channel_get_event_data_from_id(ch, | ||
777 | event_id, &event_id_data); | ||
778 | if (err == 0) { | ||
779 | /* We already have event enabled */ | ||
780 | err = -EINVAL; | ||
781 | goto free_ref; | ||
782 | } | ||
783 | |||
784 | err = get_unused_fd_flags(O_RDWR); | ||
785 | if (err < 0) | ||
786 | goto free_ref; | ||
787 | local_fd = err; | ||
788 | |||
789 | snprintf(name, sizeof(name), "nvgpu-event%d-fd%d", | ||
790 | event_id, local_fd); | ||
791 | file = anon_inode_getfile(name, &gk20a_event_id_ops, | ||
792 | NULL, O_RDWR); | ||
793 | if (IS_ERR(file)) { | ||
794 | err = PTR_ERR(file); | ||
795 | goto clean_up; | ||
796 | } | ||
797 | |||
798 | event_id_data = nvgpu_kzalloc(ch->g, sizeof(*event_id_data)); | ||
799 | if (!event_id_data) { | ||
800 | err = -ENOMEM; | ||
801 | goto clean_up_file; | ||
802 | } | ||
803 | event_id_data->g = g; | ||
804 | event_id_data->id = ch->chid; | ||
805 | event_id_data->is_tsg = false; | ||
806 | event_id_data->event_id = event_id; | ||
807 | |||
808 | nvgpu_cond_init(&event_id_data->event_id_wq); | ||
809 | err = nvgpu_mutex_init(&event_id_data->lock); | ||
810 | if (err) | ||
811 | goto clean_up_free; | ||
812 | nvgpu_init_list_node(&event_id_data->event_id_node); | ||
813 | |||
814 | nvgpu_mutex_acquire(&ch->event_id_list_lock); | ||
815 | nvgpu_list_add_tail(&event_id_data->event_id_node, &ch->event_id_list); | ||
816 | nvgpu_mutex_release(&ch->event_id_list_lock); | ||
817 | |||
818 | fd_install(local_fd, file); | ||
819 | file->private_data = event_id_data; | ||
820 | |||
821 | *fd = local_fd; | ||
822 | |||
823 | return 0; | ||
824 | |||
825 | clean_up_free: | ||
826 | nvgpu_kfree(g, event_id_data); | ||
827 | clean_up_file: | ||
828 | fput(file); | ||
829 | clean_up: | ||
830 | put_unused_fd(local_fd); | ||
831 | free_ref: | ||
832 | gk20a_put(g); | ||
833 | return err; | ||
834 | } | ||
835 | |||
836 | static int gk20a_channel_event_id_ctrl(struct channel_gk20a *ch, | ||
837 | struct nvgpu_event_id_ctrl_args *args) | ||
838 | { | ||
839 | int err = 0; | ||
840 | int fd = -1; | ||
841 | |||
842 | if (args->event_id >= NVGPU_IOCTL_CHANNEL_EVENT_ID_MAX) | ||
843 | return -EINVAL; | ||
844 | |||
845 | if (gk20a_is_channel_marked_as_tsg(ch)) | ||
846 | return -EINVAL; | ||
847 | |||
848 | switch (args->cmd) { | ||
849 | case NVGPU_IOCTL_CHANNEL_EVENT_ID_CMD_ENABLE: | ||
850 | err = gk20a_channel_event_id_enable(ch, args->event_id, &fd); | ||
851 | if (!err) | ||
852 | args->event_fd = fd; | ||
853 | break; | ||
854 | |||
855 | default: | ||
856 | nvgpu_err(ch->g, | ||
857 | "unrecognized channel event id cmd: 0x%x", | ||
858 | args->cmd); | ||
859 | err = -EINVAL; | ||
860 | break; | ||
861 | } | ||
862 | |||
863 | return err; | ||
864 | } | ||
865 | |||
866 | static int gk20a_channel_zcull_bind(struct channel_gk20a *ch, | ||
867 | struct nvgpu_zcull_bind_args *args) | ||
868 | { | ||
869 | struct gk20a *g = ch->g; | ||
870 | struct gr_gk20a *gr = &g->gr; | ||
871 | |||
872 | gk20a_dbg_fn(""); | ||
873 | |||
874 | return g->ops.gr.bind_ctxsw_zcull(g, gr, ch, | ||
875 | args->gpu_va, args->mode); | ||
876 | } | ||
877 | |||
878 | static int gk20a_ioctl_channel_submit_gpfifo( | ||
879 | struct channel_gk20a *ch, | ||
880 | struct nvgpu_submit_gpfifo_args *args) | ||
881 | { | ||
882 | struct gk20a_fence *fence_out; | ||
883 | struct fifo_profile_gk20a *profile = NULL; | ||
884 | |||
885 | int ret = 0; | ||
886 | gk20a_dbg_fn(""); | ||
887 | |||
888 | #ifdef CONFIG_DEBUG_FS | ||
889 | profile = gk20a_fifo_profile_acquire(ch->g); | ||
890 | |||
891 | if (profile) | ||
892 | profile->timestamp[PROFILE_IOCTL_ENTRY] = sched_clock(); | ||
893 | #endif | ||
894 | if (ch->has_timedout) | ||
895 | return -ETIMEDOUT; | ||
896 | |||
897 | if ((NVGPU_SUBMIT_GPFIFO_FLAGS_RESCHEDULE_RUNLIST & args->flags) && | ||
898 | !capable(CAP_SYS_NICE)) | ||
899 | return -EPERM; | ||
900 | |||
901 | ret = gk20a_submit_channel_gpfifo(ch, NULL, args, args->num_entries, | ||
902 | args->flags, &args->fence, | ||
903 | &fence_out, false, profile); | ||
904 | |||
905 | if (ret) | ||
906 | goto clean_up; | ||
907 | |||
908 | /* Convert fence_out to something we can pass back to user space. */ | ||
909 | if (args->flags & NVGPU_SUBMIT_GPFIFO_FLAGS_FENCE_GET) { | ||
910 | if (args->flags & NVGPU_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE) { | ||
911 | int fd = gk20a_fence_install_fd(fence_out); | ||
912 | if (fd < 0) | ||
913 | ret = fd; | ||
914 | else | ||
915 | args->fence.id = fd; | ||
916 | } else { | ||
917 | args->fence.id = fence_out->syncpt_id; | ||
918 | args->fence.value = fence_out->syncpt_value; | ||
919 | } | ||
920 | } | ||
921 | gk20a_fence_put(fence_out); | ||
922 | #ifdef CONFIG_DEBUG_FS | ||
923 | if (profile) { | ||
924 | profile->timestamp[PROFILE_IOCTL_EXIT] = sched_clock(); | ||
925 | gk20a_fifo_profile_release(ch->g, profile); | ||
926 | } | ||
927 | #endif | ||
928 | clean_up: | ||
929 | return ret; | ||
930 | } | ||
931 | |||
932 | /* | ||
933 | * Convert linux specific runlist level of the form NVGPU_RUNLIST_INTERLEAVE_LEVEL_* | ||
934 | * to common runlist level of the form NVGPU_FIFO_RUNLIST_INTERLEAVE_LEVEL_* | ||
935 | */ | ||
936 | u32 nvgpu_get_common_runlist_level(u32 level) | ||
937 | { | ||
938 | switch (level) { | ||
939 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_LOW: | ||
940 | return NVGPU_FIFO_RUNLIST_INTERLEAVE_LEVEL_LOW; | ||
941 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_MEDIUM: | ||
942 | return NVGPU_FIFO_RUNLIST_INTERLEAVE_LEVEL_MEDIUM; | ||
943 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_HIGH: | ||
944 | return NVGPU_FIFO_RUNLIST_INTERLEAVE_LEVEL_HIGH; | ||
945 | default: | ||
946 | pr_err("%s: incorrect runlist level\n", __func__); | ||
947 | } | ||
948 | |||
949 | return level; | ||
950 | } | ||
951 | |||
952 | static int gk20a_ioctl_channel_set_runlist_interleave(struct channel_gk20a *ch, | ||
953 | u32 level) | ||
954 | { | ||
955 | int err = 0; | ||
956 | |||
957 | err = gk20a_busy(ch->g); | ||
958 | if (err) { | ||
959 | nvgpu_err(ch->g, "failed to power on, %d", err); | ||
960 | goto fail; | ||
961 | } | ||
962 | |||
963 | level = nvgpu_get_common_runlist_level(level); | ||
964 | err = gk20a_channel_set_runlist_interleave(ch, level); | ||
965 | |||
966 | gk20a_idle(ch->g); | ||
967 | gk20a_channel_trace_sched_param( | ||
968 | trace_gk20a_channel_set_runlist_interleave, ch); | ||
969 | |||
970 | fail: | ||
971 | return err; | ||
972 | } | ||
973 | |||
974 | static u32 nvgpu_obj_ctx_user_flags_to_common_flags(u32 user_flags) | ||
975 | { | ||
976 | u32 flags = 0; | ||
977 | |||
978 | if (user_flags & NVGPU_ALLOC_OBJ_FLAGS_GFXP) | ||
979 | flags |= NVGPU_OBJ_CTX_FLAGS_SUPPORT_GFXP; | ||
980 | |||
981 | if (user_flags & NVGPU_ALLOC_OBJ_FLAGS_CILP) | ||
982 | flags |= NVGPU_OBJ_CTX_FLAGS_SUPPORT_CILP; | ||
983 | |||
984 | return flags; | ||
985 | } | ||
986 | |||
987 | static int nvgpu_ioctl_channel_alloc_obj_ctx(struct channel_gk20a *ch, | ||
988 | u32 class_num, u32 user_flags) | ||
989 | { | ||
990 | return ch->g->ops.gr.alloc_obj_ctx(ch, class_num, | ||
991 | nvgpu_obj_ctx_user_flags_to_common_flags(user_flags)); | ||
992 | } | ||
993 | |||
994 | /* | ||
995 | * Convert common preemption mode flags of the form NVGPU_PREEMPTION_MODE_GRAPHICS_* | ||
996 | * into linux preemption mode flags of the form NVGPU_GRAPHICS_PREEMPTION_MODE_* | ||
997 | */ | ||
998 | u32 nvgpu_get_ioctl_graphics_preempt_mode_flags(u32 graphics_preempt_mode_flags) | ||
999 | { | ||
1000 | u32 flags = 0; | ||
1001 | |||
1002 | if (graphics_preempt_mode_flags & NVGPU_PREEMPTION_MODE_GRAPHICS_WFI) | ||
1003 | flags |= NVGPU_GRAPHICS_PREEMPTION_MODE_WFI; | ||
1004 | if (graphics_preempt_mode_flags & NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP) | ||
1005 | flags |= NVGPU_GRAPHICS_PREEMPTION_MODE_GFXP; | ||
1006 | |||
1007 | return flags; | ||
1008 | } | ||
1009 | |||
1010 | /* | ||
1011 | * Convert common preemption mode flags of the form NVGPU_PREEMPTION_MODE_COMPUTE_* | ||
1012 | * into linux preemption mode flags of the form NVGPU_COMPUTE_PREEMPTION_MODE_* | ||
1013 | */ | ||
1014 | u32 nvgpu_get_ioctl_compute_preempt_mode_flags(u32 compute_preempt_mode_flags) | ||
1015 | { | ||
1016 | u32 flags = 0; | ||
1017 | |||
1018 | if (compute_preempt_mode_flags & NVGPU_PREEMPTION_MODE_COMPUTE_WFI) | ||
1019 | flags |= NVGPU_COMPUTE_PREEMPTION_MODE_WFI; | ||
1020 | if (compute_preempt_mode_flags & NVGPU_PREEMPTION_MODE_COMPUTE_CTA) | ||
1021 | flags |= NVGPU_COMPUTE_PREEMPTION_MODE_CTA; | ||
1022 | if (compute_preempt_mode_flags & NVGPU_PREEMPTION_MODE_COMPUTE_CILP) | ||
1023 | flags |= NVGPU_COMPUTE_PREEMPTION_MODE_CILP; | ||
1024 | |||
1025 | return flags; | ||
1026 | } | ||
1027 | |||
1028 | /* | ||
1029 | * Convert common preemption modes of the form NVGPU_PREEMPTION_MODE_GRAPHICS_* | ||
1030 | * into linux preemption modes of the form NVGPU_GRAPHICS_PREEMPTION_MODE_* | ||
1031 | */ | ||
1032 | u32 nvgpu_get_ioctl_graphics_preempt_mode(u32 graphics_preempt_mode) | ||
1033 | { | ||
1034 | switch (graphics_preempt_mode) { | ||
1035 | case NVGPU_PREEMPTION_MODE_GRAPHICS_WFI: | ||
1036 | return NVGPU_GRAPHICS_PREEMPTION_MODE_WFI; | ||
1037 | case NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP: | ||
1038 | return NVGPU_GRAPHICS_PREEMPTION_MODE_GFXP; | ||
1039 | } | ||
1040 | |||
1041 | return graphics_preempt_mode; | ||
1042 | } | ||
1043 | |||
1044 | /* | ||
1045 | * Convert common preemption modes of the form NVGPU_PREEMPTION_MODE_COMPUTE_* | ||
1046 | * into linux preemption modes of the form NVGPU_COMPUTE_PREEMPTION_MODE_* | ||
1047 | */ | ||
1048 | u32 nvgpu_get_ioctl_compute_preempt_mode(u32 compute_preempt_mode) | ||
1049 | { | ||
1050 | switch (compute_preempt_mode) { | ||
1051 | case NVGPU_PREEMPTION_MODE_COMPUTE_WFI: | ||
1052 | return NVGPU_COMPUTE_PREEMPTION_MODE_WFI; | ||
1053 | case NVGPU_PREEMPTION_MODE_COMPUTE_CTA: | ||
1054 | return NVGPU_COMPUTE_PREEMPTION_MODE_CTA; | ||
1055 | case NVGPU_PREEMPTION_MODE_COMPUTE_CILP: | ||
1056 | return NVGPU_COMPUTE_PREEMPTION_MODE_CILP; | ||
1057 | } | ||
1058 | |||
1059 | return compute_preempt_mode; | ||
1060 | } | ||
1061 | |||
1062 | /* | ||
1063 | * Convert linux preemption modes of the form NVGPU_GRAPHICS_PREEMPTION_MODE_* | ||
1064 | * into common preemption modes of the form NVGPU_PREEMPTION_MODE_GRAPHICS_* | ||
1065 | */ | ||
1066 | static u32 nvgpu_get_common_graphics_preempt_mode(u32 graphics_preempt_mode) | ||
1067 | { | ||
1068 | switch (graphics_preempt_mode) { | ||
1069 | case NVGPU_GRAPHICS_PREEMPTION_MODE_WFI: | ||
1070 | return NVGPU_PREEMPTION_MODE_GRAPHICS_WFI; | ||
1071 | case NVGPU_GRAPHICS_PREEMPTION_MODE_GFXP: | ||
1072 | return NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP; | ||
1073 | } | ||
1074 | |||
1075 | return graphics_preempt_mode; | ||
1076 | } | ||
1077 | |||
1078 | /* | ||
1079 | * Convert linux preemption modes of the form NVGPU_COMPUTE_PREEMPTION_MODE_* | ||
1080 | * into common preemption modes of the form NVGPU_PREEMPTION_MODE_COMPUTE_* | ||
1081 | */ | ||
1082 | static u32 nvgpu_get_common_compute_preempt_mode(u32 compute_preempt_mode) | ||
1083 | { | ||
1084 | switch (compute_preempt_mode) { | ||
1085 | case NVGPU_COMPUTE_PREEMPTION_MODE_WFI: | ||
1086 | return NVGPU_PREEMPTION_MODE_COMPUTE_WFI; | ||
1087 | case NVGPU_COMPUTE_PREEMPTION_MODE_CTA: | ||
1088 | return NVGPU_PREEMPTION_MODE_COMPUTE_CTA; | ||
1089 | case NVGPU_COMPUTE_PREEMPTION_MODE_CILP: | ||
1090 | return NVGPU_PREEMPTION_MODE_COMPUTE_CILP; | ||
1091 | } | ||
1092 | |||
1093 | return compute_preempt_mode; | ||
1094 | } | ||
1095 | |||
1096 | static int nvgpu_ioctl_channel_set_preemption_mode(struct channel_gk20a *ch, | ||
1097 | u32 graphics_preempt_mode, u32 compute_preempt_mode) | ||
1098 | { | ||
1099 | int err; | ||
1100 | |||
1101 | if (ch->g->ops.gr.set_preemption_mode) { | ||
1102 | err = gk20a_busy(ch->g); | ||
1103 | if (err) { | ||
1104 | nvgpu_err(ch->g, "failed to power on, %d", err); | ||
1105 | return err; | ||
1106 | } | ||
1107 | err = ch->g->ops.gr.set_preemption_mode(ch, | ||
1108 | nvgpu_get_common_graphics_preempt_mode(graphics_preempt_mode), | ||
1109 | nvgpu_get_common_compute_preempt_mode(compute_preempt_mode)); | ||
1110 | gk20a_idle(ch->g); | ||
1111 | } else { | ||
1112 | err = -EINVAL; | ||
1113 | } | ||
1114 | |||
1115 | return err; | ||
1116 | } | ||
1117 | |||
1118 | long gk20a_channel_ioctl(struct file *filp, | ||
1119 | unsigned int cmd, unsigned long arg) | ||
1120 | { | ||
1121 | struct channel_priv *priv = filp->private_data; | ||
1122 | struct channel_gk20a *ch = priv->c; | ||
1123 | struct device *dev = dev_from_gk20a(ch->g); | ||
1124 | u8 buf[NVGPU_IOCTL_CHANNEL_MAX_ARG_SIZE] = {0}; | ||
1125 | int err = 0; | ||
1126 | |||
1127 | gk20a_dbg_fn("start %d", _IOC_NR(cmd)); | ||
1128 | |||
1129 | if ((_IOC_TYPE(cmd) != NVGPU_IOCTL_MAGIC) || | ||
1130 | (_IOC_NR(cmd) == 0) || | ||
1131 | (_IOC_NR(cmd) > NVGPU_IOCTL_CHANNEL_LAST) || | ||
1132 | (_IOC_SIZE(cmd) > NVGPU_IOCTL_CHANNEL_MAX_ARG_SIZE)) | ||
1133 | return -EINVAL; | ||
1134 | |||
1135 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | ||
1136 | if (copy_from_user(buf, (void __user *)arg, _IOC_SIZE(cmd))) | ||
1137 | return -EFAULT; | ||
1138 | } | ||
1139 | |||
1140 | /* take a ref or return timeout if channel refs can't be taken */ | ||
1141 | ch = gk20a_channel_get(ch); | ||
1142 | if (!ch) | ||
1143 | return -ETIMEDOUT; | ||
1144 | |||
1145 | /* protect our sanity for threaded userspace - most of the channel is | ||
1146 | * not thread safe */ | ||
1147 | nvgpu_mutex_acquire(&ch->ioctl_lock); | ||
1148 | |||
1149 | /* this ioctl call keeps a ref to the file which keeps a ref to the | ||
1150 | * channel */ | ||
1151 | |||
1152 | switch (cmd) { | ||
1153 | case NVGPU_IOCTL_CHANNEL_OPEN: | ||
1154 | err = gk20a_channel_open_ioctl(ch->g, | ||
1155 | (struct nvgpu_channel_open_args *)buf); | ||
1156 | break; | ||
1157 | case NVGPU_IOCTL_CHANNEL_SET_NVMAP_FD: | ||
1158 | break; | ||
1159 | case NVGPU_IOCTL_CHANNEL_ALLOC_OBJ_CTX: | ||
1160 | { | ||
1161 | struct nvgpu_alloc_obj_ctx_args *args = | ||
1162 | (struct nvgpu_alloc_obj_ctx_args *)buf; | ||
1163 | |||
1164 | err = gk20a_busy(ch->g); | ||
1165 | if (err) { | ||
1166 | dev_err(dev, | ||
1167 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1168 | __func__, cmd); | ||
1169 | break; | ||
1170 | } | ||
1171 | err = nvgpu_ioctl_channel_alloc_obj_ctx(ch, args->class_num, args->flags); | ||
1172 | gk20a_idle(ch->g); | ||
1173 | break; | ||
1174 | } | ||
1175 | case NVGPU_IOCTL_CHANNEL_ALLOC_GPFIFO_EX: | ||
1176 | { | ||
1177 | struct nvgpu_alloc_gpfifo_ex_args *alloc_gpfifo_ex_args = | ||
1178 | (struct nvgpu_alloc_gpfifo_ex_args *)buf; | ||
1179 | |||
1180 | err = gk20a_busy(ch->g); | ||
1181 | if (err) { | ||
1182 | dev_err(dev, | ||
1183 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1184 | __func__, cmd); | ||
1185 | break; | ||
1186 | } | ||
1187 | |||
1188 | if (!is_power_of_2(alloc_gpfifo_ex_args->num_entries)) { | ||
1189 | err = -EINVAL; | ||
1190 | gk20a_idle(ch->g); | ||
1191 | break; | ||
1192 | } | ||
1193 | err = nvgpu_channel_ioctl_alloc_gpfifo(ch, | ||
1194 | alloc_gpfifo_ex_args->num_entries, | ||
1195 | alloc_gpfifo_ex_args->num_inflight_jobs, | ||
1196 | alloc_gpfifo_ex_args->flags); | ||
1197 | gk20a_idle(ch->g); | ||
1198 | break; | ||
1199 | } | ||
1200 | case NVGPU_IOCTL_CHANNEL_ALLOC_GPFIFO: | ||
1201 | { | ||
1202 | struct nvgpu_alloc_gpfifo_args *alloc_gpfifo_args = | ||
1203 | (struct nvgpu_alloc_gpfifo_args *)buf; | ||
1204 | |||
1205 | err = gk20a_busy(ch->g); | ||
1206 | if (err) { | ||
1207 | dev_err(dev, | ||
1208 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1209 | __func__, cmd); | ||
1210 | break; | ||
1211 | } | ||
1212 | |||
1213 | /* | ||
1214 | * Kernel can insert one extra gpfifo entry before user | ||
1215 | * submitted gpfifos and another one after, for internal usage. | ||
1216 | * Triple the requested size. | ||
1217 | */ | ||
1218 | err = nvgpu_channel_ioctl_alloc_gpfifo(ch, | ||
1219 | alloc_gpfifo_args->num_entries * 3, | ||
1220 | 0, | ||
1221 | alloc_gpfifo_args->flags); | ||
1222 | gk20a_idle(ch->g); | ||
1223 | break; | ||
1224 | } | ||
1225 | case NVGPU_IOCTL_CHANNEL_SUBMIT_GPFIFO: | ||
1226 | err = gk20a_ioctl_channel_submit_gpfifo(ch, | ||
1227 | (struct nvgpu_submit_gpfifo_args *)buf); | ||
1228 | break; | ||
1229 | case NVGPU_IOCTL_CHANNEL_WAIT: | ||
1230 | err = gk20a_busy(ch->g); | ||
1231 | if (err) { | ||
1232 | dev_err(dev, | ||
1233 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1234 | __func__, cmd); | ||
1235 | break; | ||
1236 | } | ||
1237 | |||
1238 | /* waiting is thread-safe, not dropping this mutex could | ||
1239 | * deadlock in certain conditions */ | ||
1240 | nvgpu_mutex_release(&ch->ioctl_lock); | ||
1241 | |||
1242 | err = gk20a_channel_wait(ch, | ||
1243 | (struct nvgpu_wait_args *)buf); | ||
1244 | |||
1245 | nvgpu_mutex_acquire(&ch->ioctl_lock); | ||
1246 | |||
1247 | gk20a_idle(ch->g); | ||
1248 | break; | ||
1249 | case NVGPU_IOCTL_CHANNEL_ZCULL_BIND: | ||
1250 | err = gk20a_busy(ch->g); | ||
1251 | if (err) { | ||
1252 | dev_err(dev, | ||
1253 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1254 | __func__, cmd); | ||
1255 | break; | ||
1256 | } | ||
1257 | err = gk20a_channel_zcull_bind(ch, | ||
1258 | (struct nvgpu_zcull_bind_args *)buf); | ||
1259 | gk20a_idle(ch->g); | ||
1260 | break; | ||
1261 | case NVGPU_IOCTL_CHANNEL_SET_ERROR_NOTIFIER: | ||
1262 | err = gk20a_busy(ch->g); | ||
1263 | if (err) { | ||
1264 | dev_err(dev, | ||
1265 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1266 | __func__, cmd); | ||
1267 | break; | ||
1268 | } | ||
1269 | err = gk20a_init_error_notifier(ch, | ||
1270 | (struct nvgpu_set_error_notifier *)buf); | ||
1271 | gk20a_idle(ch->g); | ||
1272 | break; | ||
1273 | #ifdef CONFIG_GK20A_CYCLE_STATS | ||
1274 | case NVGPU_IOCTL_CHANNEL_CYCLE_STATS: | ||
1275 | err = gk20a_busy(ch->g); | ||
1276 | if (err) { | ||
1277 | dev_err(dev, | ||
1278 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1279 | __func__, cmd); | ||
1280 | break; | ||
1281 | } | ||
1282 | err = gk20a_channel_cycle_stats(ch, | ||
1283 | (struct nvgpu_cycle_stats_args *)buf); | ||
1284 | gk20a_idle(ch->g); | ||
1285 | break; | ||
1286 | #endif | ||
1287 | case NVGPU_IOCTL_CHANNEL_SET_TIMEOUT: | ||
1288 | { | ||
1289 | u32 timeout = | ||
1290 | (u32)((struct nvgpu_set_timeout_args *)buf)->timeout; | ||
1291 | gk20a_dbg(gpu_dbg_gpu_dbg, "setting timeout (%d ms) for chid %d", | ||
1292 | timeout, ch->chid); | ||
1293 | ch->timeout_ms_max = timeout; | ||
1294 | gk20a_channel_trace_sched_param( | ||
1295 | trace_gk20a_channel_set_timeout, ch); | ||
1296 | break; | ||
1297 | } | ||
1298 | case NVGPU_IOCTL_CHANNEL_SET_TIMEOUT_EX: | ||
1299 | { | ||
1300 | u32 timeout = | ||
1301 | (u32)((struct nvgpu_set_timeout_args *)buf)->timeout; | ||
1302 | bool timeout_debug_dump = !((u32) | ||
1303 | ((struct nvgpu_set_timeout_ex_args *)buf)->flags & | ||
1304 | (1 << NVGPU_TIMEOUT_FLAG_DISABLE_DUMP)); | ||
1305 | gk20a_dbg(gpu_dbg_gpu_dbg, "setting timeout (%d ms) for chid %d", | ||
1306 | timeout, ch->chid); | ||
1307 | ch->timeout_ms_max = timeout; | ||
1308 | ch->timeout_debug_dump = timeout_debug_dump; | ||
1309 | gk20a_channel_trace_sched_param( | ||
1310 | trace_gk20a_channel_set_timeout, ch); | ||
1311 | break; | ||
1312 | } | ||
1313 | case NVGPU_IOCTL_CHANNEL_GET_TIMEDOUT: | ||
1314 | ((struct nvgpu_get_param_args *)buf)->value = | ||
1315 | ch->has_timedout; | ||
1316 | break; | ||
1317 | case NVGPU_IOCTL_CHANNEL_ENABLE: | ||
1318 | err = gk20a_busy(ch->g); | ||
1319 | if (err) { | ||
1320 | dev_err(dev, | ||
1321 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1322 | __func__, cmd); | ||
1323 | break; | ||
1324 | } | ||
1325 | if (ch->g->ops.fifo.enable_channel) | ||
1326 | ch->g->ops.fifo.enable_channel(ch); | ||
1327 | else | ||
1328 | err = -ENOSYS; | ||
1329 | gk20a_idle(ch->g); | ||
1330 | break; | ||
1331 | case NVGPU_IOCTL_CHANNEL_DISABLE: | ||
1332 | err = gk20a_busy(ch->g); | ||
1333 | if (err) { | ||
1334 | dev_err(dev, | ||
1335 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1336 | __func__, cmd); | ||
1337 | break; | ||
1338 | } | ||
1339 | if (ch->g->ops.fifo.disable_channel) | ||
1340 | ch->g->ops.fifo.disable_channel(ch); | ||
1341 | else | ||
1342 | err = -ENOSYS; | ||
1343 | gk20a_idle(ch->g); | ||
1344 | break; | ||
1345 | case NVGPU_IOCTL_CHANNEL_PREEMPT: | ||
1346 | err = gk20a_busy(ch->g); | ||
1347 | if (err) { | ||
1348 | dev_err(dev, | ||
1349 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1350 | __func__, cmd); | ||
1351 | break; | ||
1352 | } | ||
1353 | err = gk20a_fifo_preempt(ch->g, ch); | ||
1354 | gk20a_idle(ch->g); | ||
1355 | break; | ||
1356 | case NVGPU_IOCTL_CHANNEL_FORCE_RESET: | ||
1357 | err = gk20a_busy(ch->g); | ||
1358 | if (err) { | ||
1359 | dev_err(dev, | ||
1360 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1361 | __func__, cmd); | ||
1362 | break; | ||
1363 | } | ||
1364 | err = ch->g->ops.fifo.force_reset_ch(ch, | ||
1365 | NVGPU_CHANNEL_RESETCHANNEL_VERIF_ERROR, true); | ||
1366 | gk20a_idle(ch->g); | ||
1367 | break; | ||
1368 | case NVGPU_IOCTL_CHANNEL_EVENT_ID_CTRL: | ||
1369 | err = gk20a_channel_event_id_ctrl(ch, | ||
1370 | (struct nvgpu_event_id_ctrl_args *)buf); | ||
1371 | break; | ||
1372 | #ifdef CONFIG_GK20A_CYCLE_STATS | ||
1373 | case NVGPU_IOCTL_CHANNEL_CYCLE_STATS_SNAPSHOT: | ||
1374 | err = gk20a_busy(ch->g); | ||
1375 | if (err) { | ||
1376 | dev_err(dev, | ||
1377 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1378 | __func__, cmd); | ||
1379 | break; | ||
1380 | } | ||
1381 | err = gk20a_channel_cycle_stats_snapshot(ch, | ||
1382 | (struct nvgpu_cycle_stats_snapshot_args *)buf); | ||
1383 | gk20a_idle(ch->g); | ||
1384 | break; | ||
1385 | #endif | ||
1386 | case NVGPU_IOCTL_CHANNEL_WDT: | ||
1387 | err = gk20a_channel_set_wdt_status(ch, | ||
1388 | (struct nvgpu_channel_wdt_args *)buf); | ||
1389 | break; | ||
1390 | case NVGPU_IOCTL_CHANNEL_SET_RUNLIST_INTERLEAVE: | ||
1391 | err = gk20a_ioctl_channel_set_runlist_interleave(ch, | ||
1392 | ((struct nvgpu_runlist_interleave_args *)buf)->level); | ||
1393 | break; | ||
1394 | case NVGPU_IOCTL_CHANNEL_SET_TIMESLICE: | ||
1395 | err = gk20a_busy(ch->g); | ||
1396 | if (err) { | ||
1397 | dev_err(dev, | ||
1398 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1399 | __func__, cmd); | ||
1400 | break; | ||
1401 | } | ||
1402 | err = ch->g->ops.fifo.channel_set_timeslice(ch, | ||
1403 | ((struct nvgpu_timeslice_args *)buf)->timeslice_us); | ||
1404 | |||
1405 | gk20a_idle(ch->g); | ||
1406 | gk20a_channel_trace_sched_param( | ||
1407 | trace_gk20a_channel_set_timeslice, ch); | ||
1408 | break; | ||
1409 | case NVGPU_IOCTL_CHANNEL_GET_TIMESLICE: | ||
1410 | ((struct nvgpu_timeslice_args *)buf)->timeslice_us = | ||
1411 | gk20a_channel_get_timeslice(ch); | ||
1412 | break; | ||
1413 | case NVGPU_IOCTL_CHANNEL_SET_PREEMPTION_MODE: | ||
1414 | err = nvgpu_ioctl_channel_set_preemption_mode(ch, | ||
1415 | ((struct nvgpu_preemption_mode_args *)buf)->graphics_preempt_mode, | ||
1416 | ((struct nvgpu_preemption_mode_args *)buf)->compute_preempt_mode); | ||
1417 | break; | ||
1418 | case NVGPU_IOCTL_CHANNEL_SET_BOOSTED_CTX: | ||
1419 | if (ch->g->ops.gr.set_boosted_ctx) { | ||
1420 | bool boost = | ||
1421 | ((struct nvgpu_boosted_ctx_args *)buf)->boost; | ||
1422 | |||
1423 | err = gk20a_busy(ch->g); | ||
1424 | if (err) { | ||
1425 | dev_err(dev, | ||
1426 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1427 | __func__, cmd); | ||
1428 | break; | ||
1429 | } | ||
1430 | err = ch->g->ops.gr.set_boosted_ctx(ch, boost); | ||
1431 | gk20a_idle(ch->g); | ||
1432 | } else { | ||
1433 | err = -EINVAL; | ||
1434 | } | ||
1435 | break; | ||
1436 | default: | ||
1437 | dev_dbg(dev, "unrecognized ioctl cmd: 0x%x", cmd); | ||
1438 | err = -ENOTTY; | ||
1439 | break; | ||
1440 | } | ||
1441 | |||
1442 | if ((err == 0) && (_IOC_DIR(cmd) & _IOC_READ)) | ||
1443 | err = copy_to_user((void __user *)arg, buf, _IOC_SIZE(cmd)); | ||
1444 | |||
1445 | nvgpu_mutex_release(&ch->ioctl_lock); | ||
1446 | |||
1447 | gk20a_channel_put(ch); | ||
1448 | |||
1449 | gk20a_dbg_fn("end"); | ||
1450 | |||
1451 | return err; | ||
1452 | } | ||