diff options
Diffstat (limited to 'include/os/linux/ioctl_channel.c')
-rw-r--r-- | include/os/linux/ioctl_channel.c | 1388 |
1 files changed, 0 insertions, 1388 deletions
diff --git a/include/os/linux/ioctl_channel.c b/include/os/linux/ioctl_channel.c deleted file mode 100644 index 0f39cc7..0000000 --- a/include/os/linux/ioctl_channel.c +++ /dev/null | |||
@@ -1,1388 +0,0 @@ | |||
1 | /* | ||
2 | * GK20A Graphics channel | ||
3 | * | ||
4 | * Copyright (c) 2011-2020, 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 | |||
26 | #include <nvgpu/semaphore.h> | ||
27 | #include <nvgpu/timers.h> | ||
28 | #include <nvgpu/kmem.h> | ||
29 | #include <nvgpu/log.h> | ||
30 | #include <nvgpu/list.h> | ||
31 | #include <nvgpu/debug.h> | ||
32 | #include <nvgpu/enabled.h> | ||
33 | #include <nvgpu/error_notifier.h> | ||
34 | #include <nvgpu/barrier.h> | ||
35 | #include <nvgpu/nvhost.h> | ||
36 | #include <nvgpu/os_sched.h> | ||
37 | #include <nvgpu/gk20a.h> | ||
38 | #include <nvgpu/channel.h> | ||
39 | #include <nvgpu/channel_sync.h> | ||
40 | |||
41 | #include "gk20a/dbg_gpu_gk20a.h" | ||
42 | #include "gk20a/fence_gk20a.h" | ||
43 | |||
44 | #include "platform_gk20a.h" | ||
45 | #include "ioctl_channel.h" | ||
46 | #include "channel.h" | ||
47 | #include "os_linux.h" | ||
48 | #include "ctxsw_trace.h" | ||
49 | |||
50 | /* the minimal size of client buffer */ | ||
51 | #define CSS_MIN_CLIENT_SNAPSHOT_SIZE \ | ||
52 | (sizeof(struct gk20a_cs_snapshot_fifo) + \ | ||
53 | sizeof(struct gk20a_cs_snapshot_fifo_entry) * 256) | ||
54 | |||
55 | static const char *gr_gk20a_graphics_preempt_mode_name(u32 graphics_preempt_mode) | ||
56 | { | ||
57 | switch (graphics_preempt_mode) { | ||
58 | case NVGPU_PREEMPTION_MODE_GRAPHICS_WFI: | ||
59 | return "WFI"; | ||
60 | default: | ||
61 | return "?"; | ||
62 | } | ||
63 | } | ||
64 | |||
65 | static const char *gr_gk20a_compute_preempt_mode_name(u32 compute_preempt_mode) | ||
66 | { | ||
67 | switch (compute_preempt_mode) { | ||
68 | case NVGPU_PREEMPTION_MODE_COMPUTE_WFI: | ||
69 | return "WFI"; | ||
70 | case NVGPU_PREEMPTION_MODE_COMPUTE_CTA: | ||
71 | return "CTA"; | ||
72 | default: | ||
73 | return "?"; | ||
74 | } | ||
75 | } | ||
76 | |||
77 | static void gk20a_channel_trace_sched_param( | ||
78 | void (*trace)(int chid, int tsgid, pid_t pid, u32 timeslice, | ||
79 | u32 timeout, const char *interleave, | ||
80 | const char *graphics_preempt_mode, | ||
81 | const char *compute_preempt_mode), | ||
82 | struct channel_gk20a *ch) | ||
83 | { | ||
84 | struct tsg_gk20a *tsg = tsg_gk20a_from_ch(ch); | ||
85 | |||
86 | if (!tsg) | ||
87 | return; | ||
88 | |||
89 | (trace)(ch->chid, ch->tsgid, ch->pid, | ||
90 | tsg_gk20a_from_ch(ch)->timeslice_us, | ||
91 | ch->timeout_ms_max, | ||
92 | gk20a_fifo_interleave_level_name(tsg->interleave_level), | ||
93 | gr_gk20a_graphics_preempt_mode_name( | ||
94 | tsg->gr_ctx.graphics_preempt_mode), | ||
95 | gr_gk20a_compute_preempt_mode_name( | ||
96 | tsg->gr_ctx.compute_preempt_mode)); | ||
97 | } | ||
98 | |||
99 | /* | ||
100 | * Although channels do have pointers back to the gk20a struct that they were | ||
101 | * created under in cases where the driver is killed that pointer can be bad. | ||
102 | * The channel memory can be freed before the release() function for a given | ||
103 | * channel is called. This happens when the driver dies and userspace doesn't | ||
104 | * get a chance to call release() until after the entire gk20a driver data is | ||
105 | * unloaded and freed. | ||
106 | */ | ||
107 | struct channel_priv { | ||
108 | struct gk20a *g; | ||
109 | struct channel_gk20a *c; | ||
110 | }; | ||
111 | |||
112 | #if defined(CONFIG_GK20A_CYCLE_STATS) | ||
113 | |||
114 | void gk20a_channel_free_cycle_stats_buffer(struct channel_gk20a *ch) | ||
115 | { | ||
116 | struct nvgpu_channel_linux *priv = ch->os_priv; | ||
117 | |||
118 | /* disable existing cyclestats buffer */ | ||
119 | nvgpu_mutex_acquire(&ch->cyclestate.cyclestate_buffer_mutex); | ||
120 | if (priv->cyclestate_buffer_handler) { | ||
121 | dma_buf_vunmap(priv->cyclestate_buffer_handler, | ||
122 | ch->cyclestate.cyclestate_buffer); | ||
123 | dma_buf_put(priv->cyclestate_buffer_handler); | ||
124 | priv->cyclestate_buffer_handler = NULL; | ||
125 | ch->cyclestate.cyclestate_buffer = NULL; | ||
126 | ch->cyclestate.cyclestate_buffer_size = 0; | ||
127 | } | ||
128 | nvgpu_mutex_release(&ch->cyclestate.cyclestate_buffer_mutex); | ||
129 | } | ||
130 | |||
131 | int gk20a_channel_cycle_stats(struct channel_gk20a *ch, int dmabuf_fd) | ||
132 | { | ||
133 | struct dma_buf *dmabuf; | ||
134 | void *virtual_address; | ||
135 | struct nvgpu_channel_linux *priv = ch->os_priv; | ||
136 | |||
137 | /* is it allowed to handle calls for current GPU? */ | ||
138 | if (!nvgpu_is_enabled(ch->g, NVGPU_SUPPORT_CYCLE_STATS)) | ||
139 | return -ENOSYS; | ||
140 | |||
141 | if (dmabuf_fd && !priv->cyclestate_buffer_handler) { | ||
142 | |||
143 | /* set up new cyclestats buffer */ | ||
144 | dmabuf = dma_buf_get(dmabuf_fd); | ||
145 | if (IS_ERR(dmabuf)) | ||
146 | return PTR_ERR(dmabuf); | ||
147 | virtual_address = dma_buf_vmap(dmabuf); | ||
148 | if (!virtual_address) | ||
149 | return -ENOMEM; | ||
150 | |||
151 | priv->cyclestate_buffer_handler = dmabuf; | ||
152 | ch->cyclestate.cyclestate_buffer = virtual_address; | ||
153 | ch->cyclestate.cyclestate_buffer_size = dmabuf->size; | ||
154 | return 0; | ||
155 | |||
156 | } else if (!dmabuf_fd && priv->cyclestate_buffer_handler) { | ||
157 | gk20a_channel_free_cycle_stats_buffer(ch); | ||
158 | return 0; | ||
159 | |||
160 | } else if (!dmabuf_fd && !priv->cyclestate_buffer_handler) { | ||
161 | /* no request from GL */ | ||
162 | return 0; | ||
163 | |||
164 | } else { | ||
165 | pr_err("channel already has cyclestats buffer\n"); | ||
166 | return -EINVAL; | ||
167 | } | ||
168 | } | ||
169 | |||
170 | int gk20a_flush_cycle_stats_snapshot(struct channel_gk20a *ch) | ||
171 | { | ||
172 | int ret; | ||
173 | |||
174 | nvgpu_mutex_acquire(&ch->cs_client_mutex); | ||
175 | if (ch->cs_client) | ||
176 | ret = gr_gk20a_css_flush(ch, ch->cs_client); | ||
177 | else | ||
178 | ret = -EBADF; | ||
179 | nvgpu_mutex_release(&ch->cs_client_mutex); | ||
180 | |||
181 | return ret; | ||
182 | } | ||
183 | |||
184 | int gk20a_attach_cycle_stats_snapshot(struct channel_gk20a *ch, | ||
185 | u32 dmabuf_fd, | ||
186 | u32 perfmon_id_count, | ||
187 | u32 *perfmon_id_start) | ||
188 | { | ||
189 | int ret = 0; | ||
190 | struct gk20a *g = ch->g; | ||
191 | struct gk20a_cs_snapshot_client_linux *client_linux; | ||
192 | struct gk20a_cs_snapshot_client *client; | ||
193 | |||
194 | nvgpu_mutex_acquire(&ch->cs_client_mutex); | ||
195 | if (ch->cs_client) { | ||
196 | nvgpu_mutex_release(&ch->cs_client_mutex); | ||
197 | return -EEXIST; | ||
198 | } | ||
199 | |||
200 | client_linux = nvgpu_kzalloc(g, sizeof(*client_linux)); | ||
201 | if (!client_linux) { | ||
202 | ret = -ENOMEM; | ||
203 | goto err; | ||
204 | } | ||
205 | |||
206 | client_linux->dmabuf_fd = dmabuf_fd; | ||
207 | client_linux->dma_handler = dma_buf_get(client_linux->dmabuf_fd); | ||
208 | if (IS_ERR(client_linux->dma_handler)) { | ||
209 | ret = PTR_ERR(client_linux->dma_handler); | ||
210 | client_linux->dma_handler = NULL; | ||
211 | goto err_free; | ||
212 | } | ||
213 | |||
214 | client = &client_linux->cs_client; | ||
215 | client->snapshot_size = client_linux->dma_handler->size; | ||
216 | if (client->snapshot_size < CSS_MIN_CLIENT_SNAPSHOT_SIZE) { | ||
217 | ret = -ENOMEM; | ||
218 | goto err_put; | ||
219 | } | ||
220 | |||
221 | client->snapshot = (struct gk20a_cs_snapshot_fifo *) | ||
222 | dma_buf_vmap(client_linux->dma_handler); | ||
223 | if (!client->snapshot) { | ||
224 | ret = -ENOMEM; | ||
225 | goto err_put; | ||
226 | } | ||
227 | |||
228 | ch->cs_client = client; | ||
229 | |||
230 | ret = gr_gk20a_css_attach(ch, | ||
231 | perfmon_id_count, | ||
232 | perfmon_id_start, | ||
233 | ch->cs_client); | ||
234 | |||
235 | nvgpu_mutex_release(&ch->cs_client_mutex); | ||
236 | |||
237 | return ret; | ||
238 | |||
239 | err_put: | ||
240 | dma_buf_put(client_linux->dma_handler); | ||
241 | err_free: | ||
242 | nvgpu_kfree(g, client_linux); | ||
243 | err: | ||
244 | nvgpu_mutex_release(&ch->cs_client_mutex); | ||
245 | return ret; | ||
246 | } | ||
247 | |||
248 | int gk20a_channel_free_cycle_stats_snapshot(struct channel_gk20a *ch) | ||
249 | { | ||
250 | int ret; | ||
251 | struct gk20a_cs_snapshot_client_linux *client_linux; | ||
252 | |||
253 | nvgpu_mutex_acquire(&ch->cs_client_mutex); | ||
254 | if (!ch->cs_client) { | ||
255 | nvgpu_mutex_release(&ch->cs_client_mutex); | ||
256 | return 0; | ||
257 | } | ||
258 | |||
259 | client_linux = container_of(ch->cs_client, | ||
260 | struct gk20a_cs_snapshot_client_linux, | ||
261 | cs_client); | ||
262 | |||
263 | ret = gr_gk20a_css_detach(ch, ch->cs_client); | ||
264 | |||
265 | if (client_linux->dma_handler) { | ||
266 | if (ch->cs_client->snapshot) | ||
267 | dma_buf_vunmap(client_linux->dma_handler, | ||
268 | ch->cs_client->snapshot); | ||
269 | dma_buf_put(client_linux->dma_handler); | ||
270 | } | ||
271 | |||
272 | ch->cs_client = NULL; | ||
273 | nvgpu_kfree(ch->g, client_linux); | ||
274 | |||
275 | nvgpu_mutex_release(&ch->cs_client_mutex); | ||
276 | |||
277 | return ret; | ||
278 | } | ||
279 | #endif | ||
280 | |||
281 | static int gk20a_channel_set_wdt_status(struct channel_gk20a *ch, | ||
282 | struct nvgpu_channel_wdt_args *args) | ||
283 | { | ||
284 | u32 status = args->wdt_status & (NVGPU_IOCTL_CHANNEL_DISABLE_WDT | | ||
285 | NVGPU_IOCTL_CHANNEL_ENABLE_WDT); | ||
286 | |||
287 | if (status == NVGPU_IOCTL_CHANNEL_DISABLE_WDT) | ||
288 | ch->timeout.enabled = false; | ||
289 | else if (status == NVGPU_IOCTL_CHANNEL_ENABLE_WDT) | ||
290 | ch->timeout.enabled = true; | ||
291 | else | ||
292 | return -EINVAL; | ||
293 | |||
294 | if (args->wdt_status & NVGPU_IOCTL_CHANNEL_WDT_FLAG_SET_TIMEOUT) | ||
295 | ch->timeout.limit_ms = args->timeout_ms; | ||
296 | |||
297 | ch->timeout.debug_dump = (args->wdt_status & | ||
298 | NVGPU_IOCTL_CHANNEL_WDT_FLAG_DISABLE_DUMP) == 0; | ||
299 | |||
300 | return 0; | ||
301 | } | ||
302 | |||
303 | static void gk20a_channel_free_error_notifiers(struct channel_gk20a *ch) | ||
304 | { | ||
305 | struct nvgpu_channel_linux *priv = ch->os_priv; | ||
306 | |||
307 | nvgpu_mutex_acquire(&priv->error_notifier.mutex); | ||
308 | if (priv->error_notifier.dmabuf) { | ||
309 | dma_buf_vunmap(priv->error_notifier.dmabuf, priv->error_notifier.vaddr); | ||
310 | dma_buf_put(priv->error_notifier.dmabuf); | ||
311 | priv->error_notifier.dmabuf = NULL; | ||
312 | priv->error_notifier.notification = NULL; | ||
313 | priv->error_notifier.vaddr = NULL; | ||
314 | } | ||
315 | nvgpu_mutex_release(&priv->error_notifier.mutex); | ||
316 | } | ||
317 | |||
318 | static int gk20a_init_error_notifier(struct channel_gk20a *ch, | ||
319 | struct nvgpu_set_error_notifier *args) | ||
320 | { | ||
321 | struct dma_buf *dmabuf; | ||
322 | void *va; | ||
323 | u64 end = args->offset + sizeof(struct nvgpu_notification); | ||
324 | struct nvgpu_channel_linux *priv = ch->os_priv; | ||
325 | |||
326 | if (!args->mem) { | ||
327 | pr_err("gk20a_init_error_notifier: invalid memory handle\n"); | ||
328 | return -EINVAL; | ||
329 | } | ||
330 | |||
331 | dmabuf = dma_buf_get(args->mem); | ||
332 | |||
333 | gk20a_channel_free_error_notifiers(ch); | ||
334 | |||
335 | if (IS_ERR(dmabuf)) { | ||
336 | pr_err("Invalid handle: %d\n", args->mem); | ||
337 | return -EINVAL; | ||
338 | } | ||
339 | |||
340 | if (end > dmabuf->size || end < sizeof(struct nvgpu_notification)) { | ||
341 | dma_buf_put(dmabuf); | ||
342 | nvgpu_err(ch->g, "gk20a_init_error_notifier: invalid offset"); | ||
343 | return -EINVAL; | ||
344 | } | ||
345 | |||
346 | nvgpu_speculation_barrier(); | ||
347 | |||
348 | /* map handle */ | ||
349 | va = dma_buf_vmap(dmabuf); | ||
350 | if (!va) { | ||
351 | dma_buf_put(dmabuf); | ||
352 | pr_err("Cannot map notifier handle\n"); | ||
353 | return -ENOMEM; | ||
354 | } | ||
355 | |||
356 | priv->error_notifier.notification = va + args->offset; | ||
357 | priv->error_notifier.vaddr = va; | ||
358 | memset(priv->error_notifier.notification, 0, | ||
359 | sizeof(struct nvgpu_notification)); | ||
360 | |||
361 | /* set channel notifiers pointer */ | ||
362 | nvgpu_mutex_acquire(&priv->error_notifier.mutex); | ||
363 | priv->error_notifier.dmabuf = dmabuf; | ||
364 | nvgpu_mutex_release(&priv->error_notifier.mutex); | ||
365 | |||
366 | return 0; | ||
367 | } | ||
368 | |||
369 | /* | ||
370 | * This returns the channel with a reference. The caller must | ||
371 | * gk20a_channel_put() the ref back after use. | ||
372 | * | ||
373 | * NULL is returned if the channel was not found. | ||
374 | */ | ||
375 | struct channel_gk20a *gk20a_get_channel_from_file(int fd) | ||
376 | { | ||
377 | struct channel_gk20a *ch; | ||
378 | struct channel_priv *priv; | ||
379 | struct file *f = fget(fd); | ||
380 | |||
381 | if (!f) | ||
382 | return NULL; | ||
383 | |||
384 | if (f->f_op != &gk20a_channel_ops) { | ||
385 | fput(f); | ||
386 | return NULL; | ||
387 | } | ||
388 | |||
389 | priv = (struct channel_priv *)f->private_data; | ||
390 | ch = gk20a_channel_get(priv->c); | ||
391 | fput(f); | ||
392 | return ch; | ||
393 | } | ||
394 | |||
395 | int gk20a_channel_release(struct inode *inode, struct file *filp) | ||
396 | { | ||
397 | struct channel_priv *priv = filp->private_data; | ||
398 | struct channel_gk20a *ch; | ||
399 | struct gk20a *g; | ||
400 | |||
401 | int err; | ||
402 | |||
403 | /* We could still end up here even if the channel_open failed, e.g. | ||
404 | * if we ran out of hw channel IDs. | ||
405 | */ | ||
406 | if (!priv) | ||
407 | return 0; | ||
408 | |||
409 | ch = priv->c; | ||
410 | g = priv->g; | ||
411 | |||
412 | err = gk20a_busy(g); | ||
413 | if (err) { | ||
414 | nvgpu_err(g, "failed to release a channel!"); | ||
415 | goto channel_release; | ||
416 | } | ||
417 | |||
418 | trace_gk20a_channel_release(dev_name(dev_from_gk20a(g))); | ||
419 | |||
420 | gk20a_channel_close(ch); | ||
421 | gk20a_channel_free_error_notifiers(ch); | ||
422 | |||
423 | gk20a_idle(g); | ||
424 | |||
425 | channel_release: | ||
426 | gk20a_put(g); | ||
427 | nvgpu_kfree(g, filp->private_data); | ||
428 | filp->private_data = NULL; | ||
429 | return 0; | ||
430 | } | ||
431 | |||
432 | /* note: runlist_id -1 is synonym for the ENGINE_GR_GK20A runlist id */ | ||
433 | static int __gk20a_channel_open(struct gk20a *g, | ||
434 | struct file *filp, s32 runlist_id) | ||
435 | { | ||
436 | int err; | ||
437 | struct channel_gk20a *ch; | ||
438 | struct channel_priv *priv; | ||
439 | |||
440 | nvgpu_log_fn(g, " "); | ||
441 | |||
442 | g = gk20a_get(g); | ||
443 | if (!g) | ||
444 | return -ENODEV; | ||
445 | |||
446 | trace_gk20a_channel_open(dev_name(dev_from_gk20a(g))); | ||
447 | |||
448 | priv = nvgpu_kzalloc(g, sizeof(*priv)); | ||
449 | if (!priv) { | ||
450 | err = -ENOMEM; | ||
451 | goto free_ref; | ||
452 | } | ||
453 | |||
454 | err = gk20a_busy(g); | ||
455 | if (err) { | ||
456 | nvgpu_err(g, "failed to power on, %d", err); | ||
457 | goto fail_busy; | ||
458 | } | ||
459 | /* All the user space channel should be non privilege */ | ||
460 | ch = gk20a_open_new_channel(g, runlist_id, false, | ||
461 | nvgpu_current_pid(g), nvgpu_current_tid(g)); | ||
462 | gk20a_idle(g); | ||
463 | if (!ch) { | ||
464 | nvgpu_err(g, | ||
465 | "failed to get f"); | ||
466 | err = -ENOMEM; | ||
467 | goto fail_busy; | ||
468 | } | ||
469 | |||
470 | gk20a_channel_trace_sched_param( | ||
471 | trace_gk20a_channel_sched_defaults, ch); | ||
472 | |||
473 | priv->g = g; | ||
474 | priv->c = ch; | ||
475 | |||
476 | filp->private_data = priv; | ||
477 | return 0; | ||
478 | |||
479 | fail_busy: | ||
480 | nvgpu_kfree(g, priv); | ||
481 | free_ref: | ||
482 | gk20a_put(g); | ||
483 | return err; | ||
484 | } | ||
485 | |||
486 | int gk20a_channel_open(struct inode *inode, struct file *filp) | ||
487 | { | ||
488 | struct nvgpu_os_linux *l = container_of(inode->i_cdev, | ||
489 | struct nvgpu_os_linux, channel.cdev); | ||
490 | struct gk20a *g = &l->g; | ||
491 | int ret; | ||
492 | |||
493 | nvgpu_log_fn(g, "start"); | ||
494 | ret = __gk20a_channel_open(g, filp, -1); | ||
495 | |||
496 | nvgpu_log_fn(g, "end"); | ||
497 | return ret; | ||
498 | } | ||
499 | |||
500 | int gk20a_channel_open_ioctl(struct gk20a *g, | ||
501 | struct nvgpu_channel_open_args *args) | ||
502 | { | ||
503 | int err; | ||
504 | int fd; | ||
505 | struct file *file; | ||
506 | char name[64]; | ||
507 | s32 runlist_id = args->in.runlist_id; | ||
508 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
509 | |||
510 | err = get_unused_fd_flags(O_RDWR); | ||
511 | if (err < 0) | ||
512 | return err; | ||
513 | fd = err; | ||
514 | |||
515 | snprintf(name, sizeof(name), "nvhost-%s-fd%d", | ||
516 | dev_name(dev_from_gk20a(g)), fd); | ||
517 | |||
518 | file = anon_inode_getfile(name, l->channel.cdev.ops, NULL, O_RDWR); | ||
519 | if (IS_ERR(file)) { | ||
520 | err = PTR_ERR(file); | ||
521 | goto clean_up; | ||
522 | } | ||
523 | |||
524 | err = __gk20a_channel_open(g, file, runlist_id); | ||
525 | if (err) | ||
526 | goto clean_up_file; | ||
527 | |||
528 | fd_install(fd, file); | ||
529 | args->out.channel_fd = fd; | ||
530 | return 0; | ||
531 | |||
532 | clean_up_file: | ||
533 | fput(file); | ||
534 | clean_up: | ||
535 | put_unused_fd(fd); | ||
536 | return err; | ||
537 | } | ||
538 | |||
539 | static u32 nvgpu_setup_bind_user_flags_to_common_flags(u32 user_flags) | ||
540 | { | ||
541 | u32 flags = 0; | ||
542 | |||
543 | if (user_flags & NVGPU_CHANNEL_SETUP_BIND_FLAGS_VPR_ENABLED) | ||
544 | flags |= NVGPU_SETUP_BIND_FLAGS_SUPPORT_VPR; | ||
545 | |||
546 | if (user_flags & NVGPU_CHANNEL_SETUP_BIND_FLAGS_DETERMINISTIC) | ||
547 | flags |= NVGPU_SETUP_BIND_FLAGS_SUPPORT_DETERMINISTIC; | ||
548 | |||
549 | if (user_flags & NVGPU_CHANNEL_SETUP_BIND_FLAGS_REPLAYABLE_FAULTS_ENABLE) | ||
550 | flags |= NVGPU_SETUP_BIND_FLAGS_REPLAYABLE_FAULTS_ENABLE; | ||
551 | |||
552 | if (user_flags & NVGPU_CHANNEL_SETUP_BIND_FLAGS_USERMODE_SUPPORT) | ||
553 | flags |= NVGPU_SETUP_BIND_FLAGS_USERMODE_SUPPORT; | ||
554 | |||
555 | return flags; | ||
556 | } | ||
557 | |||
558 | static void nvgpu_get_setup_bind_args( | ||
559 | struct nvgpu_channel_setup_bind_args *channel_setup_bind_args, | ||
560 | struct nvgpu_setup_bind_args *setup_bind_args) | ||
561 | { | ||
562 | setup_bind_args->num_gpfifo_entries = | ||
563 | channel_setup_bind_args->num_gpfifo_entries; | ||
564 | setup_bind_args->num_inflight_jobs = | ||
565 | channel_setup_bind_args->num_inflight_jobs; | ||
566 | setup_bind_args->userd_dmabuf_fd = | ||
567 | channel_setup_bind_args->userd_dmabuf_fd; | ||
568 | setup_bind_args->userd_dmabuf_offset = | ||
569 | channel_setup_bind_args->userd_dmabuf_offset; | ||
570 | setup_bind_args->gpfifo_dmabuf_fd = | ||
571 | channel_setup_bind_args->gpfifo_dmabuf_fd; | ||
572 | setup_bind_args->gpfifo_dmabuf_offset = | ||
573 | channel_setup_bind_args->gpfifo_dmabuf_offset; | ||
574 | setup_bind_args->flags = nvgpu_setup_bind_user_flags_to_common_flags( | ||
575 | channel_setup_bind_args->flags); | ||
576 | } | ||
577 | |||
578 | static void nvgpu_get_gpfifo_ex_args( | ||
579 | struct nvgpu_alloc_gpfifo_ex_args *alloc_gpfifo_ex_args, | ||
580 | struct nvgpu_setup_bind_args *setup_bind_args) | ||
581 | { | ||
582 | setup_bind_args->num_gpfifo_entries = alloc_gpfifo_ex_args->num_entries; | ||
583 | setup_bind_args->num_inflight_jobs = | ||
584 | alloc_gpfifo_ex_args->num_inflight_jobs; | ||
585 | setup_bind_args->flags = nvgpu_setup_bind_user_flags_to_common_flags( | ||
586 | alloc_gpfifo_ex_args->flags); | ||
587 | } | ||
588 | |||
589 | static void nvgpu_get_gpfifo_args( | ||
590 | struct nvgpu_alloc_gpfifo_args *alloc_gpfifo_args, | ||
591 | struct nvgpu_setup_bind_args *setup_bind_args) | ||
592 | { | ||
593 | /* | ||
594 | * Kernel can insert one extra gpfifo entry before user | ||
595 | * submitted gpfifos and another one after, for internal usage. | ||
596 | * Triple the requested size. | ||
597 | */ | ||
598 | setup_bind_args->num_gpfifo_entries = | ||
599 | alloc_gpfifo_args->num_entries * 3; | ||
600 | setup_bind_args->num_inflight_jobs = 0; | ||
601 | setup_bind_args->flags = nvgpu_setup_bind_user_flags_to_common_flags( | ||
602 | alloc_gpfifo_args->flags); | ||
603 | } | ||
604 | |||
605 | static void nvgpu_get_fence_args( | ||
606 | struct nvgpu_fence *fence_args_in, | ||
607 | struct nvgpu_channel_fence *fence_args_out) | ||
608 | { | ||
609 | fence_args_out->id = fence_args_in->id; | ||
610 | fence_args_out->value = fence_args_in->value; | ||
611 | } | ||
612 | |||
613 | static int gk20a_channel_wait_semaphore(struct channel_gk20a *ch, | ||
614 | ulong id, u32 offset, | ||
615 | u32 payload, u32 timeout) | ||
616 | { | ||
617 | struct dma_buf *dmabuf; | ||
618 | void *data; | ||
619 | u32 *semaphore; | ||
620 | int ret = 0; | ||
621 | |||
622 | /* do not wait if channel has timed out */ | ||
623 | if (gk20a_channel_check_timedout(ch)) { | ||
624 | return -ETIMEDOUT; | ||
625 | } | ||
626 | |||
627 | dmabuf = dma_buf_get(id); | ||
628 | if (IS_ERR(dmabuf)) { | ||
629 | nvgpu_err(ch->g, "invalid notifier nvmap handle 0x%lx", id); | ||
630 | return -EINVAL; | ||
631 | } | ||
632 | |||
633 | data = dma_buf_kmap(dmabuf, offset >> PAGE_SHIFT); | ||
634 | if (!data) { | ||
635 | nvgpu_err(ch->g, "failed to map notifier memory"); | ||
636 | ret = -EINVAL; | ||
637 | goto cleanup_put; | ||
638 | } | ||
639 | |||
640 | semaphore = data + (offset & ~PAGE_MASK); | ||
641 | |||
642 | ret = NVGPU_COND_WAIT_INTERRUPTIBLE( | ||
643 | &ch->semaphore_wq, | ||
644 | *semaphore == payload || | ||
645 | gk20a_channel_check_timedout(ch), | ||
646 | timeout); | ||
647 | |||
648 | dma_buf_kunmap(dmabuf, offset >> PAGE_SHIFT, data); | ||
649 | cleanup_put: | ||
650 | dma_buf_put(dmabuf); | ||
651 | return ret; | ||
652 | } | ||
653 | |||
654 | static int gk20a_channel_wait(struct channel_gk20a *ch, | ||
655 | struct nvgpu_wait_args *args) | ||
656 | { | ||
657 | struct dma_buf *dmabuf; | ||
658 | struct gk20a *g = ch->g; | ||
659 | struct notification *notif; | ||
660 | struct timespec tv; | ||
661 | u64 jiffies; | ||
662 | ulong id; | ||
663 | u32 offset; | ||
664 | int remain, ret = 0; | ||
665 | u64 end; | ||
666 | |||
667 | nvgpu_log_fn(g, " "); | ||
668 | |||
669 | if (gk20a_channel_check_timedout(ch)) { | ||
670 | return -ETIMEDOUT; | ||
671 | } | ||
672 | |||
673 | switch (args->type) { | ||
674 | case NVGPU_WAIT_TYPE_NOTIFIER: | ||
675 | id = args->condition.notifier.dmabuf_fd; | ||
676 | offset = args->condition.notifier.offset; | ||
677 | end = offset + sizeof(struct notification); | ||
678 | |||
679 | dmabuf = dma_buf_get(id); | ||
680 | if (IS_ERR(dmabuf)) { | ||
681 | nvgpu_err(g, "invalid notifier nvmap handle 0x%lx", | ||
682 | id); | ||
683 | return -EINVAL; | ||
684 | } | ||
685 | |||
686 | if (end > dmabuf->size || end < sizeof(struct notification)) { | ||
687 | dma_buf_put(dmabuf); | ||
688 | nvgpu_err(g, "invalid notifier offset"); | ||
689 | return -EINVAL; | ||
690 | } | ||
691 | |||
692 | nvgpu_speculation_barrier(); | ||
693 | |||
694 | notif = dma_buf_vmap(dmabuf); | ||
695 | if (!notif) { | ||
696 | nvgpu_err(g, "failed to map notifier memory"); | ||
697 | return -ENOMEM; | ||
698 | } | ||
699 | |||
700 | notif = (struct notification *)((uintptr_t)notif + offset); | ||
701 | |||
702 | /* user should set status pending before | ||
703 | * calling this ioctl */ | ||
704 | remain = NVGPU_COND_WAIT_INTERRUPTIBLE( | ||
705 | &ch->notifier_wq, | ||
706 | notif->status == 0 || | ||
707 | gk20a_channel_check_timedout(ch), | ||
708 | args->timeout); | ||
709 | |||
710 | if (remain == 0 && notif->status != 0) { | ||
711 | ret = -ETIMEDOUT; | ||
712 | goto notif_clean_up; | ||
713 | } else if (remain < 0) { | ||
714 | ret = -EINTR; | ||
715 | goto notif_clean_up; | ||
716 | } | ||
717 | |||
718 | /* TBD: fill in correct information */ | ||
719 | jiffies = get_jiffies_64(); | ||
720 | jiffies_to_timespec(jiffies, &tv); | ||
721 | notif->timestamp.nanoseconds[0] = tv.tv_nsec; | ||
722 | notif->timestamp.nanoseconds[1] = tv.tv_sec; | ||
723 | notif->info32 = 0xDEADBEEF; /* should be object name */ | ||
724 | notif->info16 = ch->chid; /* should be method offset */ | ||
725 | |||
726 | notif_clean_up: | ||
727 | dma_buf_vunmap(dmabuf, notif); | ||
728 | return ret; | ||
729 | |||
730 | case NVGPU_WAIT_TYPE_SEMAPHORE: | ||
731 | ret = gk20a_channel_wait_semaphore(ch, | ||
732 | args->condition.semaphore.dmabuf_fd, | ||
733 | args->condition.semaphore.offset, | ||
734 | args->condition.semaphore.payload, | ||
735 | args->timeout); | ||
736 | |||
737 | break; | ||
738 | |||
739 | default: | ||
740 | ret = -EINVAL; | ||
741 | break; | ||
742 | } | ||
743 | |||
744 | return ret; | ||
745 | } | ||
746 | |||
747 | static int gk20a_channel_zcull_bind(struct channel_gk20a *ch, | ||
748 | struct nvgpu_zcull_bind_args *args) | ||
749 | { | ||
750 | struct gk20a *g = ch->g; | ||
751 | struct gr_gk20a *gr = &g->gr; | ||
752 | |||
753 | nvgpu_log_fn(gr->g, " "); | ||
754 | |||
755 | return g->ops.gr.bind_ctxsw_zcull(g, gr, ch, | ||
756 | args->gpu_va, args->mode); | ||
757 | } | ||
758 | |||
759 | static int gk20a_ioctl_channel_submit_gpfifo( | ||
760 | struct channel_gk20a *ch, | ||
761 | struct nvgpu_submit_gpfifo_args *args) | ||
762 | { | ||
763 | struct nvgpu_channel_fence fence; | ||
764 | struct gk20a_fence *fence_out; | ||
765 | struct fifo_profile_gk20a *profile = NULL; | ||
766 | u32 submit_flags = 0; | ||
767 | int fd = -1; | ||
768 | struct gk20a *g = ch->g; | ||
769 | struct nvgpu_gpfifo_userdata userdata; | ||
770 | |||
771 | int ret = 0; | ||
772 | nvgpu_log_fn(g, " "); | ||
773 | |||
774 | profile = gk20a_fifo_profile_acquire(ch->g); | ||
775 | gk20a_fifo_profile_snapshot(profile, PROFILE_IOCTL_ENTRY); | ||
776 | |||
777 | if (gk20a_channel_check_timedout(ch)) { | ||
778 | return -ETIMEDOUT; | ||
779 | } | ||
780 | |||
781 | nvgpu_get_fence_args(&args->fence, &fence); | ||
782 | submit_flags = | ||
783 | nvgpu_submit_gpfifo_user_flags_to_common_flags(args->flags); | ||
784 | |||
785 | /* Try and allocate an fd here*/ | ||
786 | if ((args->flags & NVGPU_SUBMIT_GPFIFO_FLAGS_FENCE_GET) | ||
787 | && (args->flags & NVGPU_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE)) { | ||
788 | fd = get_unused_fd_flags(O_RDWR); | ||
789 | if (fd < 0) | ||
790 | return fd; | ||
791 | } | ||
792 | |||
793 | userdata.entries = (struct nvgpu_gpfifo_entry __user *) | ||
794 | (uintptr_t)args->gpfifo; | ||
795 | userdata.context = NULL; | ||
796 | |||
797 | ret = nvgpu_submit_channel_gpfifo_user(ch, | ||
798 | userdata, args->num_entries, | ||
799 | submit_flags, &fence, &fence_out, profile); | ||
800 | |||
801 | if (ret) { | ||
802 | if (fd != -1) | ||
803 | put_unused_fd(fd); | ||
804 | goto clean_up; | ||
805 | } | ||
806 | |||
807 | /* Convert fence_out to something we can pass back to user space. */ | ||
808 | if (args->flags & NVGPU_SUBMIT_GPFIFO_FLAGS_FENCE_GET) { | ||
809 | if (args->flags & NVGPU_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE) { | ||
810 | ret = gk20a_fence_install_fd(fence_out, fd); | ||
811 | if (ret) | ||
812 | put_unused_fd(fd); | ||
813 | else | ||
814 | args->fence.id = fd; | ||
815 | } else { | ||
816 | args->fence.id = fence_out->syncpt_id; | ||
817 | args->fence.value = fence_out->syncpt_value; | ||
818 | } | ||
819 | } | ||
820 | gk20a_fence_put(fence_out); | ||
821 | |||
822 | gk20a_fifo_profile_snapshot(profile, PROFILE_IOCTL_EXIT); | ||
823 | if (profile) | ||
824 | gk20a_fifo_profile_release(ch->g, profile); | ||
825 | |||
826 | clean_up: | ||
827 | return ret; | ||
828 | } | ||
829 | |||
830 | /* | ||
831 | * Convert linux specific runlist level of the form NVGPU_RUNLIST_INTERLEAVE_LEVEL_* | ||
832 | * to common runlist level of the form NVGPU_FIFO_RUNLIST_INTERLEAVE_LEVEL_* | ||
833 | */ | ||
834 | u32 nvgpu_get_common_runlist_level(u32 level) | ||
835 | { | ||
836 | nvgpu_speculation_barrier(); | ||
837 | switch (level) { | ||
838 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_LOW: | ||
839 | return NVGPU_FIFO_RUNLIST_INTERLEAVE_LEVEL_LOW; | ||
840 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_MEDIUM: | ||
841 | return NVGPU_FIFO_RUNLIST_INTERLEAVE_LEVEL_MEDIUM; | ||
842 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_HIGH: | ||
843 | return NVGPU_FIFO_RUNLIST_INTERLEAVE_LEVEL_HIGH; | ||
844 | default: | ||
845 | pr_err("%s: incorrect runlist level\n", __func__); | ||
846 | } | ||
847 | |||
848 | return level; | ||
849 | } | ||
850 | |||
851 | static u32 nvgpu_obj_ctx_user_flags_to_common_flags(u32 user_flags) | ||
852 | { | ||
853 | u32 flags = 0; | ||
854 | |||
855 | if (user_flags & NVGPU_ALLOC_OBJ_FLAGS_GFXP) | ||
856 | flags |= NVGPU_OBJ_CTX_FLAGS_SUPPORT_GFXP; | ||
857 | |||
858 | if (user_flags & NVGPU_ALLOC_OBJ_FLAGS_CILP) | ||
859 | flags |= NVGPU_OBJ_CTX_FLAGS_SUPPORT_CILP; | ||
860 | |||
861 | return flags; | ||
862 | } | ||
863 | |||
864 | static int nvgpu_ioctl_channel_alloc_obj_ctx(struct channel_gk20a *ch, | ||
865 | u32 class_num, u32 user_flags) | ||
866 | { | ||
867 | return ch->g->ops.gr.alloc_obj_ctx(ch, class_num, | ||
868 | nvgpu_obj_ctx_user_flags_to_common_flags(user_flags)); | ||
869 | } | ||
870 | |||
871 | /* | ||
872 | * Convert common preemption mode flags of the form NVGPU_PREEMPTION_MODE_GRAPHICS_* | ||
873 | * into linux preemption mode flags of the form NVGPU_GRAPHICS_PREEMPTION_MODE_* | ||
874 | */ | ||
875 | u32 nvgpu_get_ioctl_graphics_preempt_mode_flags(u32 graphics_preempt_mode_flags) | ||
876 | { | ||
877 | u32 flags = 0; | ||
878 | |||
879 | if (graphics_preempt_mode_flags & NVGPU_PREEMPTION_MODE_GRAPHICS_WFI) | ||
880 | flags |= NVGPU_GRAPHICS_PREEMPTION_MODE_WFI; | ||
881 | if (graphics_preempt_mode_flags & NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP) | ||
882 | flags |= NVGPU_GRAPHICS_PREEMPTION_MODE_GFXP; | ||
883 | |||
884 | return flags; | ||
885 | } | ||
886 | |||
887 | /* | ||
888 | * Convert common preemption mode flags of the form NVGPU_PREEMPTION_MODE_COMPUTE_* | ||
889 | * into linux preemption mode flags of the form NVGPU_COMPUTE_PREEMPTION_MODE_* | ||
890 | */ | ||
891 | u32 nvgpu_get_ioctl_compute_preempt_mode_flags(u32 compute_preempt_mode_flags) | ||
892 | { | ||
893 | u32 flags = 0; | ||
894 | |||
895 | if (compute_preempt_mode_flags & NVGPU_PREEMPTION_MODE_COMPUTE_WFI) | ||
896 | flags |= NVGPU_COMPUTE_PREEMPTION_MODE_WFI; | ||
897 | if (compute_preempt_mode_flags & NVGPU_PREEMPTION_MODE_COMPUTE_CTA) | ||
898 | flags |= NVGPU_COMPUTE_PREEMPTION_MODE_CTA; | ||
899 | if (compute_preempt_mode_flags & NVGPU_PREEMPTION_MODE_COMPUTE_CILP) | ||
900 | flags |= NVGPU_COMPUTE_PREEMPTION_MODE_CILP; | ||
901 | |||
902 | return flags; | ||
903 | } | ||
904 | |||
905 | /* | ||
906 | * Convert common preemption modes of the form NVGPU_PREEMPTION_MODE_GRAPHICS_* | ||
907 | * into linux preemption modes of the form NVGPU_GRAPHICS_PREEMPTION_MODE_* | ||
908 | */ | ||
909 | u32 nvgpu_get_ioctl_graphics_preempt_mode(u32 graphics_preempt_mode) | ||
910 | { | ||
911 | switch (graphics_preempt_mode) { | ||
912 | case NVGPU_PREEMPTION_MODE_GRAPHICS_WFI: | ||
913 | return NVGPU_GRAPHICS_PREEMPTION_MODE_WFI; | ||
914 | case NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP: | ||
915 | return NVGPU_GRAPHICS_PREEMPTION_MODE_GFXP; | ||
916 | } | ||
917 | |||
918 | return graphics_preempt_mode; | ||
919 | } | ||
920 | |||
921 | /* | ||
922 | * Convert common preemption modes of the form NVGPU_PREEMPTION_MODE_COMPUTE_* | ||
923 | * into linux preemption modes of the form NVGPU_COMPUTE_PREEMPTION_MODE_* | ||
924 | */ | ||
925 | u32 nvgpu_get_ioctl_compute_preempt_mode(u32 compute_preempt_mode) | ||
926 | { | ||
927 | switch (compute_preempt_mode) { | ||
928 | case NVGPU_PREEMPTION_MODE_COMPUTE_WFI: | ||
929 | return NVGPU_COMPUTE_PREEMPTION_MODE_WFI; | ||
930 | case NVGPU_PREEMPTION_MODE_COMPUTE_CTA: | ||
931 | return NVGPU_COMPUTE_PREEMPTION_MODE_CTA; | ||
932 | case NVGPU_PREEMPTION_MODE_COMPUTE_CILP: | ||
933 | return NVGPU_COMPUTE_PREEMPTION_MODE_CILP; | ||
934 | } | ||
935 | |||
936 | return compute_preempt_mode; | ||
937 | } | ||
938 | |||
939 | /* | ||
940 | * Convert linux preemption modes of the form NVGPU_GRAPHICS_PREEMPTION_MODE_* | ||
941 | * into common preemption modes of the form NVGPU_PREEMPTION_MODE_GRAPHICS_* | ||
942 | */ | ||
943 | static u32 nvgpu_get_common_graphics_preempt_mode(u32 graphics_preempt_mode) | ||
944 | { | ||
945 | nvgpu_speculation_barrier(); | ||
946 | switch (graphics_preempt_mode) { | ||
947 | case NVGPU_GRAPHICS_PREEMPTION_MODE_WFI: | ||
948 | return NVGPU_PREEMPTION_MODE_GRAPHICS_WFI; | ||
949 | case NVGPU_GRAPHICS_PREEMPTION_MODE_GFXP: | ||
950 | return NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP; | ||
951 | } | ||
952 | |||
953 | return graphics_preempt_mode; | ||
954 | } | ||
955 | |||
956 | /* | ||
957 | * Convert linux preemption modes of the form NVGPU_COMPUTE_PREEMPTION_MODE_* | ||
958 | * into common preemption modes of the form NVGPU_PREEMPTION_MODE_COMPUTE_* | ||
959 | */ | ||
960 | static u32 nvgpu_get_common_compute_preempt_mode(u32 compute_preempt_mode) | ||
961 | { | ||
962 | nvgpu_speculation_barrier(); | ||
963 | switch (compute_preempt_mode) { | ||
964 | case NVGPU_COMPUTE_PREEMPTION_MODE_WFI: | ||
965 | return NVGPU_PREEMPTION_MODE_COMPUTE_WFI; | ||
966 | case NVGPU_COMPUTE_PREEMPTION_MODE_CTA: | ||
967 | return NVGPU_PREEMPTION_MODE_COMPUTE_CTA; | ||
968 | case NVGPU_COMPUTE_PREEMPTION_MODE_CILP: | ||
969 | return NVGPU_PREEMPTION_MODE_COMPUTE_CILP; | ||
970 | } | ||
971 | |||
972 | return compute_preempt_mode; | ||
973 | } | ||
974 | |||
975 | static int nvgpu_ioctl_channel_set_preemption_mode(struct channel_gk20a *ch, | ||
976 | u32 graphics_preempt_mode, u32 compute_preempt_mode) | ||
977 | { | ||
978 | int err; | ||
979 | |||
980 | if (ch->g->ops.gr.set_preemption_mode) { | ||
981 | err = gk20a_busy(ch->g); | ||
982 | if (err) { | ||
983 | nvgpu_err(ch->g, "failed to power on, %d", err); | ||
984 | return err; | ||
985 | } | ||
986 | err = ch->g->ops.gr.set_preemption_mode(ch, | ||
987 | nvgpu_get_common_graphics_preempt_mode(graphics_preempt_mode), | ||
988 | nvgpu_get_common_compute_preempt_mode(compute_preempt_mode)); | ||
989 | gk20a_idle(ch->g); | ||
990 | } else { | ||
991 | err = -EINVAL; | ||
992 | } | ||
993 | |||
994 | return err; | ||
995 | } | ||
996 | |||
997 | static int nvgpu_ioctl_channel_get_user_syncpoint(struct channel_gk20a *ch, | ||
998 | struct nvgpu_get_user_syncpoint_args *args) | ||
999 | { | ||
1000 | #ifdef CONFIG_TEGRA_GK20A_NVHOST | ||
1001 | struct gk20a *g = ch->g; | ||
1002 | int err; | ||
1003 | |||
1004 | if (!nvgpu_is_enabled(g, NVGPU_SUPPORT_USER_SYNCPOINT)) { | ||
1005 | nvgpu_err(g, "user syncpoints not supported"); | ||
1006 | return -EINVAL; | ||
1007 | } | ||
1008 | |||
1009 | if (!nvgpu_has_syncpoints(g)) { | ||
1010 | nvgpu_err(g, "syncpoints not supported"); | ||
1011 | return -EINVAL; | ||
1012 | } | ||
1013 | |||
1014 | if (g->aggressive_sync_destroy_thresh) { | ||
1015 | nvgpu_err(g, "sufficient syncpoints not available"); | ||
1016 | return -EINVAL; | ||
1017 | } | ||
1018 | |||
1019 | nvgpu_mutex_acquire(&ch->sync_lock); | ||
1020 | if (ch->user_sync) { | ||
1021 | nvgpu_mutex_release(&ch->sync_lock); | ||
1022 | } else { | ||
1023 | ch->user_sync = nvgpu_channel_sync_create(ch, true); | ||
1024 | if (!ch->user_sync) { | ||
1025 | nvgpu_mutex_release(&ch->sync_lock); | ||
1026 | return -ENOMEM; | ||
1027 | } | ||
1028 | nvgpu_mutex_release(&ch->sync_lock); | ||
1029 | |||
1030 | if (g->ops.fifo.resetup_ramfc) { | ||
1031 | err = g->ops.fifo.resetup_ramfc(ch); | ||
1032 | if (err) | ||
1033 | return err; | ||
1034 | } | ||
1035 | } | ||
1036 | |||
1037 | args->syncpoint_id = ch->user_sync->syncpt_id(ch->user_sync); | ||
1038 | args->syncpoint_max = nvgpu_nvhost_syncpt_read_maxval(g->nvhost_dev, | ||
1039 | args->syncpoint_id); | ||
1040 | if (nvgpu_is_enabled(g, NVGPU_SUPPORT_SYNCPOINT_ADDRESS)) | ||
1041 | args->gpu_va = ch->user_sync->syncpt_address(ch->user_sync); | ||
1042 | else | ||
1043 | args->gpu_va = 0; | ||
1044 | |||
1045 | return 0; | ||
1046 | #else | ||
1047 | return -EINVAL; | ||
1048 | #endif | ||
1049 | } | ||
1050 | |||
1051 | long gk20a_channel_ioctl(struct file *filp, | ||
1052 | unsigned int cmd, unsigned long arg) | ||
1053 | { | ||
1054 | struct channel_priv *priv = filp->private_data; | ||
1055 | struct channel_gk20a *ch = priv->c; | ||
1056 | struct device *dev = dev_from_gk20a(ch->g); | ||
1057 | u8 buf[NVGPU_IOCTL_CHANNEL_MAX_ARG_SIZE] = {0}; | ||
1058 | int err = 0; | ||
1059 | struct gk20a *g = ch->g; | ||
1060 | |||
1061 | nvgpu_log_fn(g, "start %d", _IOC_NR(cmd)); | ||
1062 | |||
1063 | if ((_IOC_TYPE(cmd) != NVGPU_IOCTL_MAGIC) || | ||
1064 | (_IOC_NR(cmd) == 0) || | ||
1065 | (_IOC_NR(cmd) > NVGPU_IOCTL_CHANNEL_LAST) || | ||
1066 | (_IOC_SIZE(cmd) > NVGPU_IOCTL_CHANNEL_MAX_ARG_SIZE)) | ||
1067 | return -EINVAL; | ||
1068 | |||
1069 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | ||
1070 | if (copy_from_user(buf, (void __user *)arg, _IOC_SIZE(cmd))) | ||
1071 | return -EFAULT; | ||
1072 | } | ||
1073 | |||
1074 | /* take a ref or return timeout if channel refs can't be taken */ | ||
1075 | ch = gk20a_channel_get(ch); | ||
1076 | if (!ch) | ||
1077 | return -ETIMEDOUT; | ||
1078 | |||
1079 | /* protect our sanity for threaded userspace - most of the channel is | ||
1080 | * not thread safe */ | ||
1081 | nvgpu_mutex_acquire(&ch->ioctl_lock); | ||
1082 | |||
1083 | /* this ioctl call keeps a ref to the file which keeps a ref to the | ||
1084 | * channel */ | ||
1085 | |||
1086 | nvgpu_speculation_barrier(); | ||
1087 | switch (cmd) { | ||
1088 | case NVGPU_IOCTL_CHANNEL_OPEN: | ||
1089 | err = gk20a_channel_open_ioctl(ch->g, | ||
1090 | (struct nvgpu_channel_open_args *)buf); | ||
1091 | break; | ||
1092 | case NVGPU_IOCTL_CHANNEL_SET_NVMAP_FD: | ||
1093 | break; | ||
1094 | case NVGPU_IOCTL_CHANNEL_ALLOC_OBJ_CTX: | ||
1095 | { | ||
1096 | struct nvgpu_alloc_obj_ctx_args *args = | ||
1097 | (struct nvgpu_alloc_obj_ctx_args *)buf; | ||
1098 | |||
1099 | err = gk20a_busy(ch->g); | ||
1100 | if (err) { | ||
1101 | dev_err(dev, | ||
1102 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1103 | __func__, cmd); | ||
1104 | break; | ||
1105 | } | ||
1106 | err = nvgpu_ioctl_channel_alloc_obj_ctx(ch, args->class_num, args->flags); | ||
1107 | gk20a_idle(ch->g); | ||
1108 | break; | ||
1109 | } | ||
1110 | case NVGPU_IOCTL_CHANNEL_SETUP_BIND: | ||
1111 | { | ||
1112 | struct nvgpu_channel_setup_bind_args *channel_setup_bind_args = | ||
1113 | (struct nvgpu_channel_setup_bind_args *)buf; | ||
1114 | struct nvgpu_setup_bind_args setup_bind_args; | ||
1115 | |||
1116 | nvgpu_get_setup_bind_args(channel_setup_bind_args, | ||
1117 | &setup_bind_args); | ||
1118 | |||
1119 | err = gk20a_busy(ch->g); | ||
1120 | if (err) { | ||
1121 | dev_err(dev, | ||
1122 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1123 | __func__, cmd); | ||
1124 | break; | ||
1125 | } | ||
1126 | |||
1127 | if (!is_power_of_2(setup_bind_args.num_gpfifo_entries)) { | ||
1128 | err = -EINVAL; | ||
1129 | gk20a_idle(ch->g); | ||
1130 | break; | ||
1131 | } | ||
1132 | err = nvgpu_channel_setup_bind(ch, &setup_bind_args); | ||
1133 | channel_setup_bind_args->work_submit_token = | ||
1134 | setup_bind_args.work_submit_token; | ||
1135 | gk20a_idle(ch->g); | ||
1136 | break; | ||
1137 | } | ||
1138 | case NVGPU_IOCTL_CHANNEL_ALLOC_GPFIFO_EX: | ||
1139 | { | ||
1140 | struct nvgpu_alloc_gpfifo_ex_args *alloc_gpfifo_ex_args = | ||
1141 | (struct nvgpu_alloc_gpfifo_ex_args *)buf; | ||
1142 | struct nvgpu_setup_bind_args setup_bind_args; | ||
1143 | |||
1144 | nvgpu_get_gpfifo_ex_args(alloc_gpfifo_ex_args, &setup_bind_args); | ||
1145 | |||
1146 | err = gk20a_busy(ch->g); | ||
1147 | if (err) { | ||
1148 | dev_err(dev, | ||
1149 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1150 | __func__, cmd); | ||
1151 | break; | ||
1152 | } | ||
1153 | |||
1154 | if (!is_power_of_2(alloc_gpfifo_ex_args->num_entries)) { | ||
1155 | err = -EINVAL; | ||
1156 | gk20a_idle(ch->g); | ||
1157 | break; | ||
1158 | } | ||
1159 | err = nvgpu_channel_setup_bind(ch, &setup_bind_args); | ||
1160 | gk20a_idle(ch->g); | ||
1161 | break; | ||
1162 | } | ||
1163 | case NVGPU_IOCTL_CHANNEL_ALLOC_GPFIFO: | ||
1164 | { | ||
1165 | struct nvgpu_alloc_gpfifo_args *alloc_gpfifo_args = | ||
1166 | (struct nvgpu_alloc_gpfifo_args *)buf; | ||
1167 | struct nvgpu_setup_bind_args setup_bind_args; | ||
1168 | |||
1169 | nvgpu_get_gpfifo_args(alloc_gpfifo_args, &setup_bind_args); | ||
1170 | |||
1171 | err = gk20a_busy(ch->g); | ||
1172 | if (err) { | ||
1173 | dev_err(dev, | ||
1174 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1175 | __func__, cmd); | ||
1176 | break; | ||
1177 | } | ||
1178 | |||
1179 | err = nvgpu_channel_setup_bind(ch, &setup_bind_args); | ||
1180 | gk20a_idle(ch->g); | ||
1181 | break; | ||
1182 | } | ||
1183 | case NVGPU_IOCTL_CHANNEL_SUBMIT_GPFIFO: | ||
1184 | err = gk20a_ioctl_channel_submit_gpfifo(ch, | ||
1185 | (struct nvgpu_submit_gpfifo_args *)buf); | ||
1186 | break; | ||
1187 | case NVGPU_IOCTL_CHANNEL_WAIT: | ||
1188 | err = gk20a_busy(ch->g); | ||
1189 | if (err) { | ||
1190 | dev_err(dev, | ||
1191 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1192 | __func__, cmd); | ||
1193 | break; | ||
1194 | } | ||
1195 | |||
1196 | /* waiting is thread-safe, not dropping this mutex could | ||
1197 | * deadlock in certain conditions */ | ||
1198 | nvgpu_mutex_release(&ch->ioctl_lock); | ||
1199 | |||
1200 | err = gk20a_channel_wait(ch, | ||
1201 | (struct nvgpu_wait_args *)buf); | ||
1202 | |||
1203 | nvgpu_mutex_acquire(&ch->ioctl_lock); | ||
1204 | |||
1205 | gk20a_idle(ch->g); | ||
1206 | break; | ||
1207 | case NVGPU_IOCTL_CHANNEL_ZCULL_BIND: | ||
1208 | err = gk20a_busy(ch->g); | ||
1209 | if (err) { | ||
1210 | dev_err(dev, | ||
1211 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1212 | __func__, cmd); | ||
1213 | break; | ||
1214 | } | ||
1215 | err = gk20a_channel_zcull_bind(ch, | ||
1216 | (struct nvgpu_zcull_bind_args *)buf); | ||
1217 | gk20a_idle(ch->g); | ||
1218 | break; | ||
1219 | case NVGPU_IOCTL_CHANNEL_SET_ERROR_NOTIFIER: | ||
1220 | err = gk20a_busy(ch->g); | ||
1221 | if (err) { | ||
1222 | dev_err(dev, | ||
1223 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1224 | __func__, cmd); | ||
1225 | break; | ||
1226 | } | ||
1227 | err = gk20a_init_error_notifier(ch, | ||
1228 | (struct nvgpu_set_error_notifier *)buf); | ||
1229 | gk20a_idle(ch->g); | ||
1230 | break; | ||
1231 | case NVGPU_IOCTL_CHANNEL_SET_TIMEOUT: | ||
1232 | { | ||
1233 | u32 timeout = | ||
1234 | (u32)((struct nvgpu_set_timeout_args *)buf)->timeout; | ||
1235 | nvgpu_log(g, gpu_dbg_gpu_dbg, "setting timeout (%d ms) for chid %d", | ||
1236 | timeout, ch->chid); | ||
1237 | ch->timeout_ms_max = timeout; | ||
1238 | gk20a_channel_trace_sched_param( | ||
1239 | trace_gk20a_channel_set_timeout, ch); | ||
1240 | break; | ||
1241 | } | ||
1242 | case NVGPU_IOCTL_CHANNEL_SET_TIMEOUT_EX: | ||
1243 | { | ||
1244 | u32 timeout = | ||
1245 | (u32)((struct nvgpu_set_timeout_args *)buf)->timeout; | ||
1246 | bool timeout_debug_dump = !((u32) | ||
1247 | ((struct nvgpu_set_timeout_ex_args *)buf)->flags & | ||
1248 | (1 << NVGPU_TIMEOUT_FLAG_DISABLE_DUMP)); | ||
1249 | nvgpu_log(g, gpu_dbg_gpu_dbg, "setting timeout (%d ms) for chid %d", | ||
1250 | timeout, ch->chid); | ||
1251 | ch->timeout_ms_max = timeout; | ||
1252 | ch->timeout_debug_dump = timeout_debug_dump; | ||
1253 | gk20a_channel_trace_sched_param( | ||
1254 | trace_gk20a_channel_set_timeout, ch); | ||
1255 | break; | ||
1256 | } | ||
1257 | case NVGPU_IOCTL_CHANNEL_GET_TIMEDOUT: | ||
1258 | ((struct nvgpu_get_param_args *)buf)->value = | ||
1259 | gk20a_channel_check_timedout(ch); | ||
1260 | break; | ||
1261 | case NVGPU_IOCTL_CHANNEL_ENABLE: | ||
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 | if (ch->g->ops.fifo.enable_channel) | ||
1270 | ch->g->ops.fifo.enable_channel(ch); | ||
1271 | else | ||
1272 | err = -ENOSYS; | ||
1273 | gk20a_idle(ch->g); | ||
1274 | break; | ||
1275 | case NVGPU_IOCTL_CHANNEL_DISABLE: | ||
1276 | err = gk20a_busy(ch->g); | ||
1277 | if (err) { | ||
1278 | dev_err(dev, | ||
1279 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1280 | __func__, cmd); | ||
1281 | break; | ||
1282 | } | ||
1283 | if (ch->g->ops.fifo.disable_channel) | ||
1284 | ch->g->ops.fifo.disable_channel(ch); | ||
1285 | else | ||
1286 | err = -ENOSYS; | ||
1287 | gk20a_idle(ch->g); | ||
1288 | break; | ||
1289 | case NVGPU_IOCTL_CHANNEL_PREEMPT: | ||
1290 | err = gk20a_busy(ch->g); | ||
1291 | if (err) { | ||
1292 | dev_err(dev, | ||
1293 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1294 | __func__, cmd); | ||
1295 | break; | ||
1296 | } | ||
1297 | err = gk20a_fifo_preempt(ch->g, ch); | ||
1298 | gk20a_idle(ch->g); | ||
1299 | break; | ||
1300 | case NVGPU_IOCTL_CHANNEL_RESCHEDULE_RUNLIST: | ||
1301 | if (!capable(CAP_SYS_NICE)) { | ||
1302 | err = -EPERM; | ||
1303 | break; | ||
1304 | } | ||
1305 | if (!ch->g->ops.fifo.reschedule_runlist) { | ||
1306 | err = -ENOSYS; | ||
1307 | break; | ||
1308 | } | ||
1309 | err = gk20a_busy(ch->g); | ||
1310 | if (err) { | ||
1311 | dev_err(dev, | ||
1312 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1313 | __func__, cmd); | ||
1314 | break; | ||
1315 | } | ||
1316 | err = ch->g->ops.fifo.reschedule_runlist(ch, | ||
1317 | NVGPU_RESCHEDULE_RUNLIST_PREEMPT_NEXT & | ||
1318 | ((struct nvgpu_reschedule_runlist_args *)buf)->flags); | ||
1319 | gk20a_idle(ch->g); | ||
1320 | break; | ||
1321 | case NVGPU_IOCTL_CHANNEL_FORCE_RESET: | ||
1322 | err = gk20a_busy(ch->g); | ||
1323 | if (err) { | ||
1324 | dev_err(dev, | ||
1325 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1326 | __func__, cmd); | ||
1327 | break; | ||
1328 | } | ||
1329 | err = ch->g->ops.fifo.force_reset_ch(ch, | ||
1330 | NVGPU_ERR_NOTIFIER_RESETCHANNEL_VERIF_ERROR, true); | ||
1331 | gk20a_idle(ch->g); | ||
1332 | break; | ||
1333 | case NVGPU_IOCTL_CHANNEL_WDT: | ||
1334 | err = gk20a_channel_set_wdt_status(ch, | ||
1335 | (struct nvgpu_channel_wdt_args *)buf); | ||
1336 | break; | ||
1337 | case NVGPU_IOCTL_CHANNEL_SET_PREEMPTION_MODE: | ||
1338 | err = nvgpu_ioctl_channel_set_preemption_mode(ch, | ||
1339 | ((struct nvgpu_preemption_mode_args *)buf)->graphics_preempt_mode, | ||
1340 | ((struct nvgpu_preemption_mode_args *)buf)->compute_preempt_mode); | ||
1341 | break; | ||
1342 | case NVGPU_IOCTL_CHANNEL_SET_BOOSTED_CTX: | ||
1343 | if (ch->g->ops.gr.set_boosted_ctx) { | ||
1344 | bool boost = | ||
1345 | ((struct nvgpu_boosted_ctx_args *)buf)->boost; | ||
1346 | |||
1347 | err = gk20a_busy(ch->g); | ||
1348 | if (err) { | ||
1349 | dev_err(dev, | ||
1350 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1351 | __func__, cmd); | ||
1352 | break; | ||
1353 | } | ||
1354 | err = ch->g->ops.gr.set_boosted_ctx(ch, boost); | ||
1355 | gk20a_idle(ch->g); | ||
1356 | } else { | ||
1357 | err = -EINVAL; | ||
1358 | } | ||
1359 | break; | ||
1360 | case NVGPU_IOCTL_CHANNEL_GET_USER_SYNCPOINT: | ||
1361 | err = gk20a_busy(ch->g); | ||
1362 | if (err) { | ||
1363 | dev_err(dev, | ||
1364 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
1365 | __func__, cmd); | ||
1366 | break; | ||
1367 | } | ||
1368 | err = nvgpu_ioctl_channel_get_user_syncpoint(ch, | ||
1369 | (struct nvgpu_get_user_syncpoint_args *)buf); | ||
1370 | gk20a_idle(ch->g); | ||
1371 | break; | ||
1372 | default: | ||
1373 | dev_dbg(dev, "unrecognized ioctl cmd: 0x%x", cmd); | ||
1374 | err = -ENOTTY; | ||
1375 | break; | ||
1376 | } | ||
1377 | |||
1378 | if ((err == 0) && (_IOC_DIR(cmd) & _IOC_READ)) | ||
1379 | err = copy_to_user((void __user *)arg, buf, _IOC_SIZE(cmd)); | ||
1380 | |||
1381 | nvgpu_mutex_release(&ch->ioctl_lock); | ||
1382 | |||
1383 | gk20a_channel_put(ch); | ||
1384 | |||
1385 | nvgpu_log_fn(g, "end"); | ||
1386 | |||
1387 | return err; | ||
1388 | } | ||