diff options
author | Richard Zhao <rizhao@nvidia.com> | 2018-01-30 02:24:37 -0500 |
---|---|---|
committer | mobile promotions <svcmobile_promotions@nvidia.com> | 2018-02-27 17:30:52 -0500 |
commit | 6393eddfa996fba03464f897b85aa5ec79860fed (patch) | |
tree | 557ebe9be93e2b0464118e7d8ec019d9d5dbae5f /drivers/gpu/nvgpu/vgpu/fifo_vgpu.c | |
parent | 7932568b7fe9e16b2b83bc58b2b3686c0d5e52d4 (diff) |
gpu: nvgpu: vgpu: move common files out of linux folder
Most of files have been moved out of linux folder. More code could be
common as halifying going on.
Jira EVLR-2364
Change-Id: Ia9dbdbc82f45ceefe5c788eac7517000cd455d5e
Signed-off-by: Richard Zhao <rizhao@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1649947
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Diffstat (limited to 'drivers/gpu/nvgpu/vgpu/fifo_vgpu.c')
-rw-r--r-- | drivers/gpu/nvgpu/vgpu/fifo_vgpu.c | 770 |
1 files changed, 770 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/vgpu/fifo_vgpu.c b/drivers/gpu/nvgpu/vgpu/fifo_vgpu.c new file mode 100644 index 00000000..580bfb60 --- /dev/null +++ b/drivers/gpu/nvgpu/vgpu/fifo_vgpu.c | |||
@@ -0,0 +1,770 @@ | |||
1 | /* | ||
2 | * Virtualized GPU Fifo | ||
3 | * | ||
4 | * Copyright (c) 2014-2018, NVIDIA CORPORATION. All rights reserved. | ||
5 | * | ||
6 | * Permission is hereby granted, free of charge, to any person obtaining a | ||
7 | * copy of this software and associated documentation files (the "Software"), | ||
8 | * to deal in the Software without restriction, including without limitation | ||
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | ||
10 | * and/or sell copies of the Software, and to permit persons to whom the | ||
11 | * Software is furnished to do so, subject to the following conditions: | ||
12 | * | ||
13 | * The above copyright notice and this permission notice shall be included in | ||
14 | * all copies or substantial portions of the Software. | ||
15 | * | ||
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | ||
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | ||
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | ||
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | ||
21 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER | ||
22 | * DEALINGS IN THE SOFTWARE. | ||
23 | */ | ||
24 | |||
25 | #include <trace/events/gk20a.h> | ||
26 | |||
27 | #include <nvgpu/kmem.h> | ||
28 | #include <nvgpu/dma.h> | ||
29 | #include <nvgpu/atomic.h> | ||
30 | #include <nvgpu/bug.h> | ||
31 | #include <nvgpu/barrier.h> | ||
32 | #include <nvgpu/error_notifier.h> | ||
33 | #include <nvgpu/vgpu/vgpu_ivc.h> | ||
34 | #include <nvgpu/vgpu/vgpu.h> | ||
35 | |||
36 | #include "gk20a/gk20a.h" | ||
37 | #include "fifo_vgpu.h" | ||
38 | |||
39 | #include <nvgpu/hw/gk20a/hw_fifo_gk20a.h> | ||
40 | #include <nvgpu/hw/gk20a/hw_ram_gk20a.h> | ||
41 | |||
42 | void vgpu_channel_bind(struct channel_gk20a *ch) | ||
43 | { | ||
44 | struct tegra_vgpu_cmd_msg msg; | ||
45 | struct tegra_vgpu_channel_config_params *p = | ||
46 | &msg.params.channel_config; | ||
47 | int err; | ||
48 | |||
49 | gk20a_dbg_info("bind channel %d", ch->chid); | ||
50 | |||
51 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_BIND; | ||
52 | msg.handle = vgpu_get_handle(ch->g); | ||
53 | p->handle = ch->virt_ctx; | ||
54 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
55 | WARN_ON(err || msg.ret); | ||
56 | |||
57 | nvgpu_smp_wmb(); | ||
58 | nvgpu_atomic_set(&ch->bound, true); | ||
59 | } | ||
60 | |||
61 | void vgpu_channel_unbind(struct channel_gk20a *ch) | ||
62 | { | ||
63 | |||
64 | gk20a_dbg_fn(""); | ||
65 | |||
66 | if (nvgpu_atomic_cmpxchg(&ch->bound, true, false)) { | ||
67 | struct tegra_vgpu_cmd_msg msg; | ||
68 | struct tegra_vgpu_channel_config_params *p = | ||
69 | &msg.params.channel_config; | ||
70 | int err; | ||
71 | |||
72 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_UNBIND; | ||
73 | msg.handle = vgpu_get_handle(ch->g); | ||
74 | p->handle = ch->virt_ctx; | ||
75 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
76 | WARN_ON(err || msg.ret); | ||
77 | } | ||
78 | |||
79 | } | ||
80 | |||
81 | int vgpu_channel_alloc_inst(struct gk20a *g, struct channel_gk20a *ch) | ||
82 | { | ||
83 | struct tegra_vgpu_cmd_msg msg; | ||
84 | struct tegra_vgpu_channel_hwctx_params *p = &msg.params.channel_hwctx; | ||
85 | int err; | ||
86 | |||
87 | gk20a_dbg_fn(""); | ||
88 | |||
89 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_ALLOC_HWCTX; | ||
90 | msg.handle = vgpu_get_handle(g); | ||
91 | p->id = ch->chid; | ||
92 | p->pid = (u64)current->tgid; | ||
93 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
94 | if (err || msg.ret) { | ||
95 | nvgpu_err(g, "fail"); | ||
96 | return -ENOMEM; | ||
97 | } | ||
98 | |||
99 | ch->virt_ctx = p->handle; | ||
100 | gk20a_dbg_fn("done"); | ||
101 | return 0; | ||
102 | } | ||
103 | |||
104 | void vgpu_channel_free_inst(struct gk20a *g, struct channel_gk20a *ch) | ||
105 | { | ||
106 | struct tegra_vgpu_cmd_msg msg; | ||
107 | struct tegra_vgpu_channel_hwctx_params *p = &msg.params.channel_hwctx; | ||
108 | int err; | ||
109 | |||
110 | gk20a_dbg_fn(""); | ||
111 | |||
112 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_FREE_HWCTX; | ||
113 | msg.handle = vgpu_get_handle(g); | ||
114 | p->handle = ch->virt_ctx; | ||
115 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
116 | WARN_ON(err || msg.ret); | ||
117 | } | ||
118 | |||
119 | void vgpu_channel_enable(struct channel_gk20a *ch) | ||
120 | { | ||
121 | struct tegra_vgpu_cmd_msg msg; | ||
122 | struct tegra_vgpu_channel_config_params *p = | ||
123 | &msg.params.channel_config; | ||
124 | int err; | ||
125 | |||
126 | gk20a_dbg_fn(""); | ||
127 | |||
128 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_ENABLE; | ||
129 | msg.handle = vgpu_get_handle(ch->g); | ||
130 | p->handle = ch->virt_ctx; | ||
131 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
132 | WARN_ON(err || msg.ret); | ||
133 | } | ||
134 | |||
135 | void vgpu_channel_disable(struct channel_gk20a *ch) | ||
136 | { | ||
137 | struct tegra_vgpu_cmd_msg msg; | ||
138 | struct tegra_vgpu_channel_config_params *p = | ||
139 | &msg.params.channel_config; | ||
140 | int err; | ||
141 | |||
142 | gk20a_dbg_fn(""); | ||
143 | |||
144 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_DISABLE; | ||
145 | msg.handle = vgpu_get_handle(ch->g); | ||
146 | p->handle = ch->virt_ctx; | ||
147 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
148 | WARN_ON(err || msg.ret); | ||
149 | } | ||
150 | |||
151 | int vgpu_channel_setup_ramfc(struct channel_gk20a *ch, u64 gpfifo_base, | ||
152 | u32 gpfifo_entries, | ||
153 | unsigned long acquire_timeout, u32 flags) | ||
154 | { | ||
155 | struct tegra_vgpu_cmd_msg msg; | ||
156 | struct tegra_vgpu_ramfc_params *p = &msg.params.ramfc; | ||
157 | int err; | ||
158 | |||
159 | gk20a_dbg_fn(""); | ||
160 | |||
161 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_SETUP_RAMFC; | ||
162 | msg.handle = vgpu_get_handle(ch->g); | ||
163 | p->handle = ch->virt_ctx; | ||
164 | p->gpfifo_va = gpfifo_base; | ||
165 | p->num_entries = gpfifo_entries; | ||
166 | p->userd_addr = ch->userd_iova; | ||
167 | p->iova = 0; | ||
168 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
169 | |||
170 | return (err || msg.ret) ? -ENOMEM : 0; | ||
171 | } | ||
172 | |||
173 | int vgpu_fifo_init_engine_info(struct fifo_gk20a *f) | ||
174 | { | ||
175 | struct vgpu_priv_data *priv = vgpu_get_priv_data(f->g); | ||
176 | struct tegra_vgpu_engines_info *engines = &priv->constants.engines_info; | ||
177 | u32 i; | ||
178 | |||
179 | gk20a_dbg_fn(""); | ||
180 | |||
181 | if (engines->num_engines > TEGRA_VGPU_MAX_ENGINES) { | ||
182 | nvgpu_err(f->g, "num_engines %d larger than max %d", | ||
183 | engines->num_engines, TEGRA_VGPU_MAX_ENGINES); | ||
184 | return -EINVAL; | ||
185 | } | ||
186 | |||
187 | f->num_engines = engines->num_engines; | ||
188 | for (i = 0; i < f->num_engines; i++) { | ||
189 | struct fifo_engine_info_gk20a *info = | ||
190 | &f->engine_info[engines->info[i].engine_id]; | ||
191 | |||
192 | if (engines->info[i].engine_id >= f->max_engines) { | ||
193 | nvgpu_err(f->g, "engine id %d larger than max %d", | ||
194 | engines->info[i].engine_id, | ||
195 | f->max_engines); | ||
196 | return -EINVAL; | ||
197 | } | ||
198 | |||
199 | info->intr_mask = engines->info[i].intr_mask; | ||
200 | info->reset_mask = engines->info[i].reset_mask; | ||
201 | info->runlist_id = engines->info[i].runlist_id; | ||
202 | info->pbdma_id = engines->info[i].pbdma_id; | ||
203 | info->inst_id = engines->info[i].inst_id; | ||
204 | info->pri_base = engines->info[i].pri_base; | ||
205 | info->engine_enum = engines->info[i].engine_enum; | ||
206 | info->fault_id = engines->info[i].fault_id; | ||
207 | f->active_engines_list[i] = engines->info[i].engine_id; | ||
208 | } | ||
209 | |||
210 | gk20a_dbg_fn("done"); | ||
211 | |||
212 | return 0; | ||
213 | } | ||
214 | |||
215 | static int init_runlist(struct gk20a *g, struct fifo_gk20a *f) | ||
216 | { | ||
217 | struct fifo_runlist_info_gk20a *runlist; | ||
218 | unsigned int runlist_id = -1; | ||
219 | u32 i; | ||
220 | u64 runlist_size; | ||
221 | |||
222 | gk20a_dbg_fn(""); | ||
223 | |||
224 | f->max_runlists = g->ops.fifo.eng_runlist_base_size(); | ||
225 | f->runlist_info = nvgpu_kzalloc(g, | ||
226 | sizeof(struct fifo_runlist_info_gk20a) * | ||
227 | f->max_runlists); | ||
228 | if (!f->runlist_info) | ||
229 | goto clean_up_runlist; | ||
230 | |||
231 | memset(f->runlist_info, 0, (sizeof(struct fifo_runlist_info_gk20a) * | ||
232 | f->max_runlists)); | ||
233 | |||
234 | for (runlist_id = 0; runlist_id < f->max_runlists; runlist_id++) { | ||
235 | runlist = &f->runlist_info[runlist_id]; | ||
236 | |||
237 | runlist->active_channels = | ||
238 | nvgpu_kzalloc(g, DIV_ROUND_UP(f->num_channels, | ||
239 | BITS_PER_BYTE)); | ||
240 | if (!runlist->active_channels) | ||
241 | goto clean_up_runlist; | ||
242 | |||
243 | runlist_size = sizeof(u16) * f->num_channels; | ||
244 | for (i = 0; i < MAX_RUNLIST_BUFFERS; i++) { | ||
245 | int err = nvgpu_dma_alloc_sys(g, runlist_size, | ||
246 | &runlist->mem[i]); | ||
247 | if (err) { | ||
248 | nvgpu_err(g, "memory allocation failed"); | ||
249 | goto clean_up_runlist; | ||
250 | } | ||
251 | } | ||
252 | nvgpu_mutex_init(&runlist->mutex); | ||
253 | |||
254 | /* None of buffers is pinned if this value doesn't change. | ||
255 | Otherwise, one of them (cur_buffer) must have been pinned. */ | ||
256 | runlist->cur_buffer = MAX_RUNLIST_BUFFERS; | ||
257 | } | ||
258 | |||
259 | gk20a_dbg_fn("done"); | ||
260 | return 0; | ||
261 | |||
262 | clean_up_runlist: | ||
263 | gk20a_fifo_delete_runlist(f); | ||
264 | gk20a_dbg_fn("fail"); | ||
265 | return -ENOMEM; | ||
266 | } | ||
267 | |||
268 | static int vgpu_init_fifo_setup_sw(struct gk20a *g) | ||
269 | { | ||
270 | struct fifo_gk20a *f = &g->fifo; | ||
271 | struct vgpu_priv_data *priv = vgpu_get_priv_data(g); | ||
272 | unsigned int chid; | ||
273 | int err = 0; | ||
274 | |||
275 | gk20a_dbg_fn(""); | ||
276 | |||
277 | if (f->sw_ready) { | ||
278 | gk20a_dbg_fn("skip init"); | ||
279 | return 0; | ||
280 | } | ||
281 | |||
282 | f->g = g; | ||
283 | f->num_channels = priv->constants.num_channels; | ||
284 | f->max_engines = nvgpu_get_litter_value(g, GPU_LIT_HOST_NUM_ENGINES); | ||
285 | |||
286 | f->userd_entry_size = 1 << ram_userd_base_shift_v(); | ||
287 | |||
288 | err = nvgpu_dma_alloc_sys(g, f->userd_entry_size * f->num_channels, | ||
289 | &f->userd); | ||
290 | if (err) { | ||
291 | nvgpu_err(g, "memory allocation failed"); | ||
292 | goto clean_up; | ||
293 | } | ||
294 | |||
295 | /* bar1 va */ | ||
296 | if (g->ops.mm.is_bar1_supported(g)) { | ||
297 | f->userd.gpu_va = vgpu_bar1_map(g, &f->userd); | ||
298 | if (!f->userd.gpu_va) { | ||
299 | nvgpu_err(g, "gmmu mapping failed"); | ||
300 | goto clean_up; | ||
301 | } | ||
302 | /* if reduced BAR1 range is specified, use offset of 0 | ||
303 | * (server returns offset assuming full BAR1 range) | ||
304 | */ | ||
305 | if (vgpu_is_reduced_bar1(g)) | ||
306 | f->userd.gpu_va = 0; | ||
307 | } | ||
308 | |||
309 | gk20a_dbg(gpu_dbg_map_v, "userd bar1 va = 0x%llx", f->userd.gpu_va); | ||
310 | |||
311 | f->channel = nvgpu_vzalloc(g, f->num_channels * sizeof(*f->channel)); | ||
312 | f->tsg = nvgpu_vzalloc(g, f->num_channels * sizeof(*f->tsg)); | ||
313 | f->engine_info = nvgpu_kzalloc(g, f->max_engines * | ||
314 | sizeof(*f->engine_info)); | ||
315 | f->active_engines_list = nvgpu_kzalloc(g, f->max_engines * sizeof(u32)); | ||
316 | |||
317 | if (!(f->channel && f->tsg && f->engine_info && f->active_engines_list)) { | ||
318 | err = -ENOMEM; | ||
319 | goto clean_up; | ||
320 | } | ||
321 | memset(f->active_engines_list, 0xff, (f->max_engines * sizeof(u32))); | ||
322 | |||
323 | g->ops.fifo.init_engine_info(f); | ||
324 | |||
325 | init_runlist(g, f); | ||
326 | |||
327 | nvgpu_init_list_node(&f->free_chs); | ||
328 | nvgpu_mutex_init(&f->free_chs_mutex); | ||
329 | |||
330 | for (chid = 0; chid < f->num_channels; chid++) { | ||
331 | f->channel[chid].userd_iova = | ||
332 | nvgpu_mem_get_addr(g, &f->userd) + | ||
333 | chid * f->userd_entry_size; | ||
334 | f->channel[chid].userd_gpu_va = | ||
335 | f->userd.gpu_va + chid * f->userd_entry_size; | ||
336 | |||
337 | gk20a_init_channel_support(g, chid); | ||
338 | gk20a_init_tsg_support(g, chid); | ||
339 | } | ||
340 | nvgpu_mutex_init(&f->tsg_inuse_mutex); | ||
341 | |||
342 | err = nvgpu_channel_worker_init(g); | ||
343 | if (err) | ||
344 | goto clean_up; | ||
345 | |||
346 | f->deferred_reset_pending = false; | ||
347 | nvgpu_mutex_init(&f->deferred_reset_mutex); | ||
348 | |||
349 | f->channel_base = priv->constants.channel_base; | ||
350 | |||
351 | f->sw_ready = true; | ||
352 | |||
353 | gk20a_dbg_fn("done"); | ||
354 | return 0; | ||
355 | |||
356 | clean_up: | ||
357 | gk20a_dbg_fn("fail"); | ||
358 | /* FIXME: unmap from bar1 */ | ||
359 | nvgpu_dma_free(g, &f->userd); | ||
360 | |||
361 | memset(&f->userd, 0, sizeof(f->userd)); | ||
362 | |||
363 | nvgpu_vfree(g, f->channel); | ||
364 | f->channel = NULL; | ||
365 | nvgpu_vfree(g, f->tsg); | ||
366 | f->tsg = NULL; | ||
367 | nvgpu_kfree(g, f->engine_info); | ||
368 | f->engine_info = NULL; | ||
369 | nvgpu_kfree(g, f->active_engines_list); | ||
370 | f->active_engines_list = NULL; | ||
371 | |||
372 | return err; | ||
373 | } | ||
374 | |||
375 | int vgpu_init_fifo_setup_hw(struct gk20a *g) | ||
376 | { | ||
377 | gk20a_dbg_fn(""); | ||
378 | |||
379 | /* test write, read through bar1 @ userd region before | ||
380 | * turning on the snooping */ | ||
381 | { | ||
382 | struct fifo_gk20a *f = &g->fifo; | ||
383 | u32 v, v1 = 0x33, v2 = 0x55; | ||
384 | |||
385 | u32 bar1_vaddr = f->userd.gpu_va; | ||
386 | volatile u32 *cpu_vaddr = f->userd.cpu_va; | ||
387 | |||
388 | gk20a_dbg_info("test bar1 @ vaddr 0x%x", | ||
389 | bar1_vaddr); | ||
390 | |||
391 | v = gk20a_bar1_readl(g, bar1_vaddr); | ||
392 | |||
393 | *cpu_vaddr = v1; | ||
394 | nvgpu_mb(); | ||
395 | |||
396 | if (v1 != gk20a_bar1_readl(g, bar1_vaddr)) { | ||
397 | nvgpu_err(g, "bar1 broken @ gk20a!"); | ||
398 | return -EINVAL; | ||
399 | } | ||
400 | |||
401 | gk20a_bar1_writel(g, bar1_vaddr, v2); | ||
402 | |||
403 | if (v2 != gk20a_bar1_readl(g, bar1_vaddr)) { | ||
404 | nvgpu_err(g, "bar1 broken @ gk20a!"); | ||
405 | return -EINVAL; | ||
406 | } | ||
407 | |||
408 | /* is it visible to the cpu? */ | ||
409 | if (*cpu_vaddr != v2) { | ||
410 | nvgpu_err(g, "cpu didn't see bar1 write @ %p!", | ||
411 | cpu_vaddr); | ||
412 | } | ||
413 | |||
414 | /* put it back */ | ||
415 | gk20a_bar1_writel(g, bar1_vaddr, v); | ||
416 | } | ||
417 | |||
418 | gk20a_dbg_fn("done"); | ||
419 | |||
420 | return 0; | ||
421 | } | ||
422 | |||
423 | int vgpu_init_fifo_support(struct gk20a *g) | ||
424 | { | ||
425 | u32 err; | ||
426 | |||
427 | gk20a_dbg_fn(""); | ||
428 | |||
429 | err = vgpu_init_fifo_setup_sw(g); | ||
430 | if (err) | ||
431 | return err; | ||
432 | |||
433 | if (g->ops.fifo.init_fifo_setup_hw) | ||
434 | err = g->ops.fifo.init_fifo_setup_hw(g); | ||
435 | return err; | ||
436 | } | ||
437 | |||
438 | int vgpu_fifo_preempt_channel(struct gk20a *g, u32 chid) | ||
439 | { | ||
440 | struct fifo_gk20a *f = &g->fifo; | ||
441 | struct channel_gk20a *ch = &f->channel[chid]; | ||
442 | struct tegra_vgpu_cmd_msg msg; | ||
443 | struct tegra_vgpu_channel_config_params *p = | ||
444 | &msg.params.channel_config; | ||
445 | int err; | ||
446 | |||
447 | gk20a_dbg_fn(""); | ||
448 | |||
449 | if (!nvgpu_atomic_read(&ch->bound)) | ||
450 | return 0; | ||
451 | |||
452 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_PREEMPT; | ||
453 | msg.handle = vgpu_get_handle(g); | ||
454 | p->handle = ch->virt_ctx; | ||
455 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
456 | |||
457 | if (err || msg.ret) { | ||
458 | nvgpu_err(g, | ||
459 | "preempt channel %d failed", chid); | ||
460 | err = -ENOMEM; | ||
461 | } | ||
462 | |||
463 | return err; | ||
464 | } | ||
465 | |||
466 | int vgpu_fifo_preempt_tsg(struct gk20a *g, u32 tsgid) | ||
467 | { | ||
468 | struct tegra_vgpu_cmd_msg msg; | ||
469 | struct tegra_vgpu_tsg_preempt_params *p = | ||
470 | &msg.params.tsg_preempt; | ||
471 | int err; | ||
472 | |||
473 | gk20a_dbg_fn(""); | ||
474 | |||
475 | msg.cmd = TEGRA_VGPU_CMD_TSG_PREEMPT; | ||
476 | msg.handle = vgpu_get_handle(g); | ||
477 | p->tsg_id = tsgid; | ||
478 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
479 | err = err ? err : msg.ret; | ||
480 | |||
481 | if (err) { | ||
482 | nvgpu_err(g, | ||
483 | "preempt tsg %u failed", tsgid); | ||
484 | } | ||
485 | |||
486 | return err; | ||
487 | } | ||
488 | |||
489 | static int vgpu_submit_runlist(struct gk20a *g, u64 handle, u8 runlist_id, | ||
490 | u16 *runlist, u32 num_entries) | ||
491 | { | ||
492 | struct tegra_vgpu_cmd_msg msg; | ||
493 | struct tegra_vgpu_runlist_params *p; | ||
494 | int err; | ||
495 | void *oob_handle; | ||
496 | void *oob; | ||
497 | size_t size, oob_size; | ||
498 | |||
499 | oob_handle = vgpu_ivc_oob_get_ptr(vgpu_ivc_get_server_vmid(), | ||
500 | TEGRA_VGPU_QUEUE_CMD, | ||
501 | &oob, &oob_size); | ||
502 | if (!oob_handle) | ||
503 | return -EINVAL; | ||
504 | |||
505 | size = sizeof(*runlist) * num_entries; | ||
506 | if (oob_size < size) { | ||
507 | err = -ENOMEM; | ||
508 | goto done; | ||
509 | } | ||
510 | |||
511 | msg.cmd = TEGRA_VGPU_CMD_SUBMIT_RUNLIST; | ||
512 | msg.handle = handle; | ||
513 | p = &msg.params.runlist; | ||
514 | p->runlist_id = runlist_id; | ||
515 | p->num_entries = num_entries; | ||
516 | |||
517 | memcpy(oob, runlist, size); | ||
518 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
519 | |||
520 | err = (err || msg.ret) ? -1 : 0; | ||
521 | |||
522 | done: | ||
523 | vgpu_ivc_oob_put_ptr(oob_handle); | ||
524 | return err; | ||
525 | } | ||
526 | |||
527 | static int vgpu_fifo_update_runlist_locked(struct gk20a *g, u32 runlist_id, | ||
528 | u32 chid, bool add, | ||
529 | bool wait_for_finish) | ||
530 | { | ||
531 | struct fifo_gk20a *f = &g->fifo; | ||
532 | struct fifo_runlist_info_gk20a *runlist; | ||
533 | u16 *runlist_entry = NULL; | ||
534 | u32 count = 0; | ||
535 | |||
536 | gk20a_dbg_fn(""); | ||
537 | |||
538 | runlist = &f->runlist_info[runlist_id]; | ||
539 | |||
540 | /* valid channel, add/remove it from active list. | ||
541 | Otherwise, keep active list untouched for suspend/resume. */ | ||
542 | if (chid != (u32)~0) { | ||
543 | if (add) { | ||
544 | if (test_and_set_bit(chid, | ||
545 | runlist->active_channels) == 1) | ||
546 | return 0; | ||
547 | } else { | ||
548 | if (test_and_clear_bit(chid, | ||
549 | runlist->active_channels) == 0) | ||
550 | return 0; | ||
551 | } | ||
552 | } | ||
553 | |||
554 | if (chid != (u32)~0 || /* add/remove a valid channel */ | ||
555 | add /* resume to add all channels back */) { | ||
556 | u32 cid; | ||
557 | |||
558 | runlist_entry = runlist->mem[0].cpu_va; | ||
559 | for_each_set_bit(cid, | ||
560 | runlist->active_channels, f->num_channels) { | ||
561 | gk20a_dbg_info("add channel %d to runlist", cid); | ||
562 | runlist_entry[0] = cid; | ||
563 | runlist_entry++; | ||
564 | count++; | ||
565 | } | ||
566 | } else /* suspend to remove all channels */ | ||
567 | count = 0; | ||
568 | |||
569 | return vgpu_submit_runlist(g, vgpu_get_handle(g), runlist_id, | ||
570 | runlist->mem[0].cpu_va, count); | ||
571 | } | ||
572 | |||
573 | /* add/remove a channel from runlist | ||
574 | special cases below: runlist->active_channels will NOT be changed. | ||
575 | (chid == ~0 && !add) means remove all active channels from runlist. | ||
576 | (chid == ~0 && add) means restore all active channels on runlist. */ | ||
577 | int vgpu_fifo_update_runlist(struct gk20a *g, u32 runlist_id, | ||
578 | u32 chid, bool add, bool wait_for_finish) | ||
579 | { | ||
580 | struct fifo_runlist_info_gk20a *runlist = NULL; | ||
581 | struct fifo_gk20a *f = &g->fifo; | ||
582 | u32 ret = 0; | ||
583 | |||
584 | gk20a_dbg_fn(""); | ||
585 | |||
586 | runlist = &f->runlist_info[runlist_id]; | ||
587 | |||
588 | nvgpu_mutex_acquire(&runlist->mutex); | ||
589 | |||
590 | ret = vgpu_fifo_update_runlist_locked(g, runlist_id, chid, add, | ||
591 | wait_for_finish); | ||
592 | |||
593 | nvgpu_mutex_release(&runlist->mutex); | ||
594 | return ret; | ||
595 | } | ||
596 | |||
597 | int vgpu_fifo_wait_engine_idle(struct gk20a *g) | ||
598 | { | ||
599 | gk20a_dbg_fn(""); | ||
600 | |||
601 | return 0; | ||
602 | } | ||
603 | |||
604 | int vgpu_fifo_set_runlist_interleave(struct gk20a *g, | ||
605 | u32 id, | ||
606 | u32 runlist_id, | ||
607 | u32 new_level) | ||
608 | { | ||
609 | struct tegra_vgpu_cmd_msg msg = {0}; | ||
610 | struct tegra_vgpu_tsg_runlist_interleave_params *p = | ||
611 | &msg.params.tsg_interleave; | ||
612 | int err; | ||
613 | |||
614 | gk20a_dbg_fn(""); | ||
615 | |||
616 | msg.cmd = TEGRA_VGPU_CMD_TSG_SET_RUNLIST_INTERLEAVE; | ||
617 | msg.handle = vgpu_get_handle(g); | ||
618 | p->tsg_id = id; | ||
619 | p->level = new_level; | ||
620 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
621 | WARN_ON(err || msg.ret); | ||
622 | return err ? err : msg.ret; | ||
623 | } | ||
624 | |||
625 | int vgpu_fifo_force_reset_ch(struct channel_gk20a *ch, | ||
626 | u32 err_code, bool verbose) | ||
627 | { | ||
628 | struct tsg_gk20a *tsg = NULL; | ||
629 | struct channel_gk20a *ch_tsg = NULL; | ||
630 | struct gk20a *g = ch->g; | ||
631 | struct tegra_vgpu_cmd_msg msg = {0}; | ||
632 | struct tegra_vgpu_channel_config_params *p = | ||
633 | &msg.params.channel_config; | ||
634 | int err; | ||
635 | |||
636 | gk20a_dbg_fn(""); | ||
637 | |||
638 | if (gk20a_is_channel_marked_as_tsg(ch)) { | ||
639 | tsg = &g->fifo.tsg[ch->tsgid]; | ||
640 | |||
641 | nvgpu_rwsem_down_read(&tsg->ch_list_lock); | ||
642 | |||
643 | nvgpu_list_for_each_entry(ch_tsg, &tsg->ch_list, | ||
644 | channel_gk20a, ch_entry) { | ||
645 | if (gk20a_channel_get(ch_tsg)) { | ||
646 | nvgpu_set_error_notifier(ch_tsg, err_code); | ||
647 | ch_tsg->has_timedout = true; | ||
648 | gk20a_channel_put(ch_tsg); | ||
649 | } | ||
650 | } | ||
651 | |||
652 | nvgpu_rwsem_up_read(&tsg->ch_list_lock); | ||
653 | } else { | ||
654 | nvgpu_set_error_notifier(ch, err_code); | ||
655 | ch->has_timedout = true; | ||
656 | } | ||
657 | |||
658 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_FORCE_RESET; | ||
659 | msg.handle = vgpu_get_handle(ch->g); | ||
660 | p->handle = ch->virt_ctx; | ||
661 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
662 | WARN_ON(err || msg.ret); | ||
663 | if (!err) | ||
664 | gk20a_channel_abort(ch, false); | ||
665 | return err ? err : msg.ret; | ||
666 | } | ||
667 | |||
668 | static void vgpu_fifo_set_ctx_mmu_error_ch(struct gk20a *g, | ||
669 | struct channel_gk20a *ch) | ||
670 | { | ||
671 | /* | ||
672 | * If error code is already set, this mmu fault | ||
673 | * was triggered as part of recovery from other | ||
674 | * error condition. | ||
675 | * Don't overwrite error flag. | ||
676 | */ | ||
677 | nvgpu_set_error_notifier_if_empty(ch, | ||
678 | NVGPU_ERR_NOTIFIER_FIFO_ERROR_MMU_ERR_FLT); | ||
679 | |||
680 | /* mark channel as faulted */ | ||
681 | ch->has_timedout = true; | ||
682 | nvgpu_smp_wmb(); | ||
683 | /* unblock pending waits */ | ||
684 | nvgpu_cond_broadcast_interruptible(&ch->semaphore_wq); | ||
685 | nvgpu_cond_broadcast_interruptible(&ch->notifier_wq); | ||
686 | } | ||
687 | |||
688 | static void vgpu_fifo_set_ctx_mmu_error_ch_tsg(struct gk20a *g, | ||
689 | struct channel_gk20a *ch) | ||
690 | { | ||
691 | struct tsg_gk20a *tsg = NULL; | ||
692 | struct channel_gk20a *ch_tsg = NULL; | ||
693 | |||
694 | if (gk20a_is_channel_marked_as_tsg(ch)) { | ||
695 | tsg = &g->fifo.tsg[ch->tsgid]; | ||
696 | |||
697 | nvgpu_rwsem_down_read(&tsg->ch_list_lock); | ||
698 | |||
699 | nvgpu_list_for_each_entry(ch_tsg, &tsg->ch_list, | ||
700 | channel_gk20a, ch_entry) { | ||
701 | if (gk20a_channel_get(ch_tsg)) { | ||
702 | vgpu_fifo_set_ctx_mmu_error_ch(g, ch_tsg); | ||
703 | gk20a_channel_put(ch_tsg); | ||
704 | } | ||
705 | } | ||
706 | |||
707 | nvgpu_rwsem_up_read(&tsg->ch_list_lock); | ||
708 | } else { | ||
709 | vgpu_fifo_set_ctx_mmu_error_ch(g, ch); | ||
710 | } | ||
711 | } | ||
712 | |||
713 | int vgpu_fifo_isr(struct gk20a *g, struct tegra_vgpu_fifo_intr_info *info) | ||
714 | { | ||
715 | struct fifo_gk20a *f = &g->fifo; | ||
716 | struct channel_gk20a *ch = gk20a_channel_get(&f->channel[info->chid]); | ||
717 | |||
718 | gk20a_dbg_fn(""); | ||
719 | if (!ch) | ||
720 | return 0; | ||
721 | |||
722 | nvgpu_err(g, "fifo intr (%d) on ch %u", | ||
723 | info->type, info->chid); | ||
724 | |||
725 | trace_gk20a_channel_reset(ch->chid, ch->tsgid); | ||
726 | |||
727 | switch (info->type) { | ||
728 | case TEGRA_VGPU_FIFO_INTR_PBDMA: | ||
729 | nvgpu_set_error_notifier(ch, NVGPU_ERR_NOTIFIER_PBDMA_ERROR); | ||
730 | break; | ||
731 | case TEGRA_VGPU_FIFO_INTR_CTXSW_TIMEOUT: | ||
732 | nvgpu_set_error_notifier(ch, | ||
733 | NVGPU_ERR_NOTIFIER_FIFO_ERROR_IDLE_TIMEOUT); | ||
734 | break; | ||
735 | case TEGRA_VGPU_FIFO_INTR_MMU_FAULT: | ||
736 | vgpu_fifo_set_ctx_mmu_error_ch_tsg(g, ch); | ||
737 | gk20a_channel_abort(ch, false); | ||
738 | break; | ||
739 | default: | ||
740 | WARN_ON(1); | ||
741 | break; | ||
742 | } | ||
743 | |||
744 | gk20a_channel_put(ch); | ||
745 | return 0; | ||
746 | } | ||
747 | |||
748 | int vgpu_fifo_nonstall_isr(struct gk20a *g, | ||
749 | struct tegra_vgpu_fifo_nonstall_intr_info *info) | ||
750 | { | ||
751 | gk20a_dbg_fn(""); | ||
752 | |||
753 | switch (info->type) { | ||
754 | case TEGRA_VGPU_FIFO_NONSTALL_INTR_CHANNEL: | ||
755 | gk20a_channel_semaphore_wakeup(g, false); | ||
756 | break; | ||
757 | default: | ||
758 | WARN_ON(1); | ||
759 | break; | ||
760 | } | ||
761 | |||
762 | return 0; | ||
763 | } | ||
764 | |||
765 | u32 vgpu_fifo_default_timeslice_us(struct gk20a *g) | ||
766 | { | ||
767 | struct vgpu_priv_data *priv = vgpu_get_priv_data(g); | ||
768 | |||
769 | return priv->constants.default_timeslice_us; | ||
770 | } | ||