]>
Commit | Line | Data |
---|---|---|
6579324a TB |
1 | /* |
2 | * Tegra host1x Job | |
3 | * | |
4 | * Copyright (c) 2010-2013, NVIDIA Corporation. | |
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 <linux/dma-mapping.h> | |
20 | #include <linux/err.h> | |
21 | #include <linux/kref.h> | |
22 | #include <linux/module.h> | |
23 | #include <linux/scatterlist.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/vmalloc.h> | |
26 | #include <trace/events/host1x.h> | |
27 | ||
28 | #include "channel.h" | |
29 | #include "dev.h" | |
30 | #include "host1x_bo.h" | |
31 | #include "job.h" | |
32 | #include "syncpt.h" | |
33 | ||
34 | struct host1x_job *host1x_job_alloc(struct host1x_channel *ch, | |
35 | u32 num_cmdbufs, u32 num_relocs, | |
36 | u32 num_waitchks) | |
37 | { | |
38 | struct host1x_job *job = NULL; | |
39 | unsigned int num_unpins = num_cmdbufs + num_relocs; | |
40 | u64 total; | |
41 | void *mem; | |
42 | ||
43 | /* Check that we're not going to overflow */ | |
44 | total = sizeof(struct host1x_job) + | |
f5fda676 DC |
45 | (u64)num_relocs * sizeof(struct host1x_reloc) + |
46 | (u64)num_unpins * sizeof(struct host1x_job_unpin_data) + | |
47 | (u64)num_waitchks * sizeof(struct host1x_waitchk) + | |
48 | (u64)num_cmdbufs * sizeof(struct host1x_job_gather) + | |
49 | (u64)num_unpins * sizeof(dma_addr_t) + | |
50 | (u64)num_unpins * sizeof(u32 *); | |
6579324a TB |
51 | if (total > ULONG_MAX) |
52 | return NULL; | |
53 | ||
54 | mem = job = kzalloc(total, GFP_KERNEL); | |
55 | if (!job) | |
56 | return NULL; | |
57 | ||
58 | kref_init(&job->ref); | |
59 | job->channel = ch; | |
60 | ||
61 | /* Redistribute memory to the structs */ | |
62 | mem += sizeof(struct host1x_job); | |
63 | job->relocarray = num_relocs ? mem : NULL; | |
64 | mem += num_relocs * sizeof(struct host1x_reloc); | |
65 | job->unpins = num_unpins ? mem : NULL; | |
66 | mem += num_unpins * sizeof(struct host1x_job_unpin_data); | |
67 | job->waitchk = num_waitchks ? mem : NULL; | |
68 | mem += num_waitchks * sizeof(struct host1x_waitchk); | |
69 | job->gathers = num_cmdbufs ? mem : NULL; | |
70 | mem += num_cmdbufs * sizeof(struct host1x_job_gather); | |
71 | job->addr_phys = num_unpins ? mem : NULL; | |
72 | ||
73 | job->reloc_addr_phys = job->addr_phys; | |
74 | job->gather_addr_phys = &job->addr_phys[num_relocs]; | |
75 | ||
76 | return job; | |
77 | } | |
78 | ||
79 | struct host1x_job *host1x_job_get(struct host1x_job *job) | |
80 | { | |
81 | kref_get(&job->ref); | |
82 | return job; | |
83 | } | |
84 | ||
85 | static void job_free(struct kref *ref) | |
86 | { | |
87 | struct host1x_job *job = container_of(ref, struct host1x_job, ref); | |
88 | ||
89 | kfree(job); | |
90 | } | |
91 | ||
92 | void host1x_job_put(struct host1x_job *job) | |
93 | { | |
94 | kref_put(&job->ref, job_free); | |
95 | } | |
96 | ||
97 | void host1x_job_add_gather(struct host1x_job *job, struct host1x_bo *bo, | |
98 | u32 words, u32 offset) | |
99 | { | |
100 | struct host1x_job_gather *cur_gather = &job->gathers[job->num_gathers]; | |
101 | ||
102 | cur_gather->words = words; | |
103 | cur_gather->bo = bo; | |
104 | cur_gather->offset = offset; | |
105 | job->num_gathers++; | |
106 | } | |
107 | ||
108 | /* | |
109 | * NULL an already satisfied WAIT_SYNCPT host method, by patching its | |
110 | * args in the command stream. The method data is changed to reference | |
111 | * a reserved (never given out or incr) HOST1X_SYNCPT_RESERVED syncpt | |
112 | * with a matching threshold value of 0, so is guaranteed to be popped | |
113 | * by the host HW. | |
114 | */ | |
115 | static void host1x_syncpt_patch_offset(struct host1x_syncpt *sp, | |
116 | struct host1x_bo *h, u32 offset) | |
117 | { | |
118 | void *patch_addr = NULL; | |
119 | ||
120 | /* patch the wait */ | |
121 | patch_addr = host1x_bo_kmap(h, offset >> PAGE_SHIFT); | |
122 | if (patch_addr) { | |
123 | host1x_syncpt_patch_wait(sp, | |
124 | patch_addr + (offset & ~PAGE_MASK)); | |
125 | host1x_bo_kunmap(h, offset >> PAGE_SHIFT, patch_addr); | |
126 | } else | |
127 | pr_err("Could not map cmdbuf for wait check\n"); | |
128 | } | |
129 | ||
130 | /* | |
131 | * Check driver supplied waitchk structs for syncpt thresholds | |
132 | * that have already been satisfied and NULL the comparison (to | |
133 | * avoid a wrap condition in the HW). | |
134 | */ | |
135 | static int do_waitchks(struct host1x_job *job, struct host1x *host, | |
136 | struct host1x_bo *patch) | |
137 | { | |
138 | int i; | |
139 | ||
140 | /* compare syncpt vs wait threshold */ | |
141 | for (i = 0; i < job->num_waitchk; i++) { | |
142 | struct host1x_waitchk *wait = &job->waitchk[i]; | |
143 | struct host1x_syncpt *sp = | |
144 | host1x_syncpt_get(host, wait->syncpt_id); | |
145 | ||
146 | /* validate syncpt id */ | |
147 | if (wait->syncpt_id > host1x_syncpt_nb_pts(host)) | |
148 | continue; | |
149 | ||
150 | /* skip all other gathers */ | |
151 | if (patch != wait->bo) | |
152 | continue; | |
153 | ||
154 | trace_host1x_syncpt_wait_check(wait->bo, wait->offset, | |
155 | wait->syncpt_id, wait->thresh, | |
156 | host1x_syncpt_read_min(sp)); | |
157 | ||
158 | if (host1x_syncpt_is_expired(sp, wait->thresh)) { | |
159 | dev_dbg(host->dev, | |
160 | "drop WAIT id %d (%s) thresh 0x%x, min 0x%x\n", | |
161 | wait->syncpt_id, sp->name, wait->thresh, | |
162 | host1x_syncpt_read_min(sp)); | |
163 | ||
164 | host1x_syncpt_patch_offset(sp, patch, wait->offset); | |
165 | } | |
166 | ||
167 | wait->bo = NULL; | |
168 | } | |
169 | ||
170 | return 0; | |
171 | } | |
172 | ||
173 | static unsigned int pin_job(struct host1x_job *job) | |
174 | { | |
175 | unsigned int i; | |
176 | ||
177 | job->num_unpins = 0; | |
178 | ||
179 | for (i = 0; i < job->num_relocs; i++) { | |
180 | struct host1x_reloc *reloc = &job->relocarray[i]; | |
181 | struct sg_table *sgt; | |
182 | dma_addr_t phys_addr; | |
183 | ||
184 | reloc->target = host1x_bo_get(reloc->target); | |
185 | if (!reloc->target) | |
186 | goto unpin; | |
187 | ||
188 | phys_addr = host1x_bo_pin(reloc->target, &sgt); | |
189 | if (!phys_addr) | |
190 | goto unpin; | |
191 | ||
192 | job->addr_phys[job->num_unpins] = phys_addr; | |
193 | job->unpins[job->num_unpins].bo = reloc->target; | |
194 | job->unpins[job->num_unpins].sgt = sgt; | |
195 | job->num_unpins++; | |
196 | } | |
197 | ||
198 | for (i = 0; i < job->num_gathers; i++) { | |
199 | struct host1x_job_gather *g = &job->gathers[i]; | |
200 | struct sg_table *sgt; | |
201 | dma_addr_t phys_addr; | |
202 | ||
203 | g->bo = host1x_bo_get(g->bo); | |
204 | if (!g->bo) | |
205 | goto unpin; | |
206 | ||
207 | phys_addr = host1x_bo_pin(g->bo, &sgt); | |
208 | if (!phys_addr) | |
209 | goto unpin; | |
210 | ||
211 | job->addr_phys[job->num_unpins] = phys_addr; | |
212 | job->unpins[job->num_unpins].bo = g->bo; | |
213 | job->unpins[job->num_unpins].sgt = sgt; | |
214 | job->num_unpins++; | |
215 | } | |
216 | ||
217 | return job->num_unpins; | |
218 | ||
219 | unpin: | |
220 | host1x_job_unpin(job); | |
221 | return 0; | |
222 | } | |
223 | ||
224 | static unsigned int do_relocs(struct host1x_job *job, struct host1x_bo *cmdbuf) | |
225 | { | |
226 | int i = 0; | |
227 | u32 last_page = ~0; | |
228 | void *cmdbuf_page_addr = NULL; | |
229 | ||
230 | /* pin & patch the relocs for one gather */ | |
3364cd28 | 231 | for (i = 0; i < job->num_relocs; i++) { |
6579324a TB |
232 | struct host1x_reloc *reloc = &job->relocarray[i]; |
233 | u32 reloc_addr = (job->reloc_addr_phys[i] + | |
234 | reloc->target_offset) >> reloc->shift; | |
235 | u32 *target; | |
236 | ||
237 | /* skip all other gathers */ | |
3364cd28 | 238 | if (cmdbuf != reloc->cmdbuf) |
6579324a | 239 | continue; |
6579324a TB |
240 | |
241 | if (last_page != reloc->cmdbuf_offset >> PAGE_SHIFT) { | |
242 | if (cmdbuf_page_addr) | |
243 | host1x_bo_kunmap(cmdbuf, last_page, | |
244 | cmdbuf_page_addr); | |
245 | ||
246 | cmdbuf_page_addr = host1x_bo_kmap(cmdbuf, | |
247 | reloc->cmdbuf_offset >> PAGE_SHIFT); | |
248 | last_page = reloc->cmdbuf_offset >> PAGE_SHIFT; | |
249 | ||
250 | if (unlikely(!cmdbuf_page_addr)) { | |
251 | pr_err("Could not map cmdbuf for relocation\n"); | |
252 | return -ENOMEM; | |
253 | } | |
254 | } | |
255 | ||
256 | target = cmdbuf_page_addr + (reloc->cmdbuf_offset & ~PAGE_MASK); | |
257 | *target = reloc_addr; | |
6579324a TB |
258 | } |
259 | ||
260 | if (cmdbuf_page_addr) | |
261 | host1x_bo_kunmap(cmdbuf, last_page, cmdbuf_page_addr); | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
5060d8ec | 266 | static bool check_reloc(struct host1x_reloc *reloc, struct host1x_bo *cmdbuf, |
6579324a TB |
267 | unsigned int offset) |
268 | { | |
269 | offset *= sizeof(u32); | |
270 | ||
271 | if (reloc->cmdbuf != cmdbuf || reloc->cmdbuf_offset != offset) | |
5060d8ec | 272 | return false; |
6579324a | 273 | |
5060d8ec | 274 | return true; |
6579324a TB |
275 | } |
276 | ||
277 | struct host1x_firewall { | |
278 | struct host1x_job *job; | |
279 | struct device *dev; | |
280 | ||
281 | unsigned int num_relocs; | |
282 | struct host1x_reloc *reloc; | |
283 | ||
284 | struct host1x_bo *cmdbuf_id; | |
285 | unsigned int offset; | |
286 | ||
287 | u32 words; | |
288 | u32 class; | |
289 | u32 reg; | |
290 | u32 mask; | |
291 | u32 count; | |
292 | }; | |
293 | ||
294 | static int check_mask(struct host1x_firewall *fw) | |
295 | { | |
296 | u32 mask = fw->mask; | |
297 | u32 reg = fw->reg; | |
298 | ||
299 | while (mask) { | |
300 | if (fw->words == 0) | |
301 | return -EINVAL; | |
302 | ||
303 | if (mask & 1) { | |
304 | if (fw->job->is_addr_reg(fw->dev, fw->class, reg)) { | |
5060d8ec AM |
305 | if (!fw->num_relocs) |
306 | return -EINVAL; | |
307 | if (!check_reloc(fw->reloc, fw->cmdbuf_id, | |
308 | fw->offset)) | |
6579324a TB |
309 | return -EINVAL; |
310 | fw->reloc++; | |
311 | fw->num_relocs--; | |
312 | } | |
313 | fw->words--; | |
314 | fw->offset++; | |
315 | } | |
316 | mask >>= 1; | |
317 | reg++; | |
318 | } | |
319 | ||
320 | return 0; | |
321 | } | |
322 | ||
323 | static int check_incr(struct host1x_firewall *fw) | |
324 | { | |
325 | u32 count = fw->count; | |
326 | u32 reg = fw->reg; | |
327 | ||
64c173d3 | 328 | while (count) { |
6579324a TB |
329 | if (fw->words == 0) |
330 | return -EINVAL; | |
331 | ||
332 | if (fw->job->is_addr_reg(fw->dev, fw->class, reg)) { | |
5060d8ec AM |
333 | if (!fw->num_relocs) |
334 | return -EINVAL; | |
335 | if (!check_reloc(fw->reloc, fw->cmdbuf_id, fw->offset)) | |
6579324a TB |
336 | return -EINVAL; |
337 | fw->reloc++; | |
338 | fw->num_relocs--; | |
339 | } | |
340 | reg++; | |
341 | fw->words--; | |
342 | fw->offset++; | |
343 | count--; | |
344 | } | |
345 | ||
346 | return 0; | |
347 | } | |
348 | ||
349 | static int check_nonincr(struct host1x_firewall *fw) | |
350 | { | |
351 | int is_addr_reg = fw->job->is_addr_reg(fw->dev, fw->class, fw->reg); | |
352 | u32 count = fw->count; | |
353 | ||
354 | while (count) { | |
355 | if (fw->words == 0) | |
356 | return -EINVAL; | |
357 | ||
358 | if (is_addr_reg) { | |
5060d8ec AM |
359 | if (!fw->num_relocs) |
360 | return -EINVAL; | |
361 | if (!check_reloc(fw->reloc, fw->cmdbuf_id, fw->offset)) | |
6579324a TB |
362 | return -EINVAL; |
363 | fw->reloc++; | |
364 | fw->num_relocs--; | |
365 | } | |
366 | fw->words--; | |
367 | fw->offset++; | |
368 | count--; | |
369 | } | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
afac0e43 | 374 | static int validate(struct host1x_firewall *fw, struct host1x_job_gather *g) |
6579324a | 375 | { |
3364cd28 AM |
376 | u32 *cmdbuf_base = (u32 *)fw->job->gather_copy_mapped + |
377 | (g->offset / sizeof(u32)); | |
6579324a | 378 | int err = 0; |
6579324a | 379 | |
afac0e43 | 380 | if (!fw->job->is_addr_reg) |
6579324a TB |
381 | return 0; |
382 | ||
afac0e43 TB |
383 | fw->words = g->words; |
384 | fw->cmdbuf_id = g->bo; | |
385 | fw->offset = 0; | |
6579324a | 386 | |
afac0e43 TB |
387 | while (fw->words && !err) { |
388 | u32 word = cmdbuf_base[fw->offset]; | |
6579324a TB |
389 | u32 opcode = (word & 0xf0000000) >> 28; |
390 | ||
afac0e43 TB |
391 | fw->mask = 0; |
392 | fw->reg = 0; | |
393 | fw->count = 0; | |
394 | fw->words--; | |
395 | fw->offset++; | |
6579324a TB |
396 | |
397 | switch (opcode) { | |
398 | case 0: | |
afac0e43 TB |
399 | fw->class = word >> 6 & 0x3ff; |
400 | fw->mask = word & 0x3f; | |
401 | fw->reg = word >> 16 & 0xfff; | |
402 | err = check_mask(fw); | |
6579324a TB |
403 | if (err) |
404 | goto out; | |
405 | break; | |
406 | case 1: | |
afac0e43 TB |
407 | fw->reg = word >> 16 & 0xfff; |
408 | fw->count = word & 0xffff; | |
409 | err = check_incr(fw); | |
6579324a TB |
410 | if (err) |
411 | goto out; | |
412 | break; | |
413 | ||
414 | case 2: | |
afac0e43 TB |
415 | fw->reg = word >> 16 & 0xfff; |
416 | fw->count = word & 0xffff; | |
417 | err = check_nonincr(fw); | |
6579324a TB |
418 | if (err) |
419 | goto out; | |
420 | break; | |
421 | ||
422 | case 3: | |
afac0e43 TB |
423 | fw->mask = word & 0xffff; |
424 | fw->reg = word >> 16 & 0xfff; | |
425 | err = check_mask(fw); | |
6579324a TB |
426 | if (err) |
427 | goto out; | |
428 | break; | |
429 | case 4: | |
430 | case 5: | |
431 | case 14: | |
432 | break; | |
433 | default: | |
434 | err = -EINVAL; | |
435 | break; | |
436 | } | |
437 | } | |
438 | ||
439 | /* No relocs should remain at this point */ | |
afac0e43 | 440 | if (fw->num_relocs) |
6579324a TB |
441 | err = -EINVAL; |
442 | ||
443 | out: | |
6579324a TB |
444 | return err; |
445 | } | |
446 | ||
447 | static inline int copy_gathers(struct host1x_job *job, struct device *dev) | |
448 | { | |
3364cd28 | 449 | struct host1x_firewall fw; |
6579324a TB |
450 | size_t size = 0; |
451 | size_t offset = 0; | |
452 | int i; | |
453 | ||
3364cd28 AM |
454 | fw.job = job; |
455 | fw.dev = dev; | |
456 | fw.reloc = job->relocarray; | |
457 | fw.num_relocs = job->num_relocs; | |
458 | fw.class = 0; | |
459 | ||
6579324a TB |
460 | for (i = 0; i < job->num_gathers; i++) { |
461 | struct host1x_job_gather *g = &job->gathers[i]; | |
462 | size += g->words * sizeof(u32); | |
463 | } | |
464 | ||
465 | job->gather_copy_mapped = dma_alloc_writecombine(dev, size, | |
466 | &job->gather_copy, | |
467 | GFP_KERNEL); | |
468 | if (!job->gather_copy_mapped) { | |
469 | int err = PTR_ERR(job->gather_copy_mapped); | |
470 | job->gather_copy_mapped = NULL; | |
471 | return err; | |
472 | } | |
473 | ||
474 | job->gather_copy_size = size; | |
475 | ||
476 | for (i = 0; i < job->num_gathers; i++) { | |
477 | struct host1x_job_gather *g = &job->gathers[i]; | |
478 | void *gather; | |
479 | ||
3364cd28 | 480 | /* Copy the gather */ |
6579324a TB |
481 | gather = host1x_bo_mmap(g->bo); |
482 | memcpy(job->gather_copy_mapped + offset, gather + g->offset, | |
483 | g->words * sizeof(u32)); | |
484 | host1x_bo_munmap(g->bo, gather); | |
485 | ||
3364cd28 | 486 | /* Store the location in the buffer */ |
6579324a TB |
487 | g->base = job->gather_copy; |
488 | g->offset = offset; | |
3364cd28 AM |
489 | |
490 | /* Validate the job */ | |
491 | if (validate(&fw, g)) | |
492 | return -EINVAL; | |
6579324a TB |
493 | |
494 | offset += g->words * sizeof(u32); | |
495 | } | |
496 | ||
497 | return 0; | |
498 | } | |
499 | ||
500 | int host1x_job_pin(struct host1x_job *job, struct device *dev) | |
501 | { | |
502 | int err; | |
503 | unsigned int i, j; | |
504 | struct host1x *host = dev_get_drvdata(dev->parent); | |
505 | DECLARE_BITMAP(waitchk_mask, host1x_syncpt_nb_pts(host)); | |
506 | ||
507 | bitmap_zero(waitchk_mask, host1x_syncpt_nb_pts(host)); | |
508 | for (i = 0; i < job->num_waitchk; i++) { | |
509 | u32 syncpt_id = job->waitchk[i].syncpt_id; | |
510 | if (syncpt_id < host1x_syncpt_nb_pts(host)) | |
511 | set_bit(syncpt_id, waitchk_mask); | |
512 | } | |
513 | ||
514 | /* get current syncpt values for waitchk */ | |
515 | for_each_set_bit(i, waitchk_mask, host1x_syncpt_nb_pts(host)) | |
516 | host1x_syncpt_load(host->syncpt + i); | |
517 | ||
518 | /* pin memory */ | |
519 | err = pin_job(job); | |
520 | if (!err) | |
521 | goto out; | |
522 | ||
523 | /* patch gathers */ | |
524 | for (i = 0; i < job->num_gathers; i++) { | |
525 | struct host1x_job_gather *g = &job->gathers[i]; | |
526 | ||
527 | /* process each gather mem only once */ | |
528 | if (g->handled) | |
529 | continue; | |
530 | ||
531 | g->base = job->gather_addr_phys[i]; | |
532 | ||
533 | for (j = 0; j < job->num_gathers; j++) | |
534 | if (job->gathers[j].bo == g->bo) | |
535 | job->gathers[j].handled = true; | |
536 | ||
3364cd28 | 537 | err = do_relocs(job, g->bo); |
6579324a | 538 | if (err) |
3364cd28 | 539 | break; |
6579324a | 540 | |
3364cd28 | 541 | err = do_waitchks(job, host, g->bo); |
6579324a TB |
542 | if (err) |
543 | break; | |
544 | } | |
545 | ||
546 | if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) && !err) { | |
547 | err = copy_gathers(job, dev); | |
548 | if (err) { | |
549 | host1x_job_unpin(job); | |
550 | return err; | |
551 | } | |
552 | } | |
553 | ||
554 | out: | |
555 | wmb(); | |
556 | ||
557 | return err; | |
558 | } | |
559 | ||
560 | void host1x_job_unpin(struct host1x_job *job) | |
561 | { | |
562 | unsigned int i; | |
563 | ||
564 | for (i = 0; i < job->num_unpins; i++) { | |
565 | struct host1x_job_unpin_data *unpin = &job->unpins[i]; | |
566 | host1x_bo_unpin(unpin->bo, unpin->sgt); | |
567 | host1x_bo_put(unpin->bo); | |
568 | } | |
569 | job->num_unpins = 0; | |
570 | ||
571 | if (job->gather_copy_size) | |
572 | dma_free_writecombine(job->channel->dev, job->gather_copy_size, | |
573 | job->gather_copy_mapped, | |
574 | job->gather_copy); | |
575 | } | |
576 | ||
577 | /* | |
578 | * Debug routine used to dump job entries | |
579 | */ | |
580 | void host1x_job_dump(struct device *dev, struct host1x_job *job) | |
581 | { | |
582 | dev_dbg(dev, " SYNCPT_ID %d\n", job->syncpt_id); | |
583 | dev_dbg(dev, " SYNCPT_VAL %d\n", job->syncpt_end); | |
584 | dev_dbg(dev, " FIRST_GET 0x%x\n", job->first_get); | |
585 | dev_dbg(dev, " TIMEOUT %d\n", job->timeout); | |
586 | dev_dbg(dev, " NUM_SLOTS %d\n", job->num_slots); | |
587 | dev_dbg(dev, " NUM_HANDLES %d\n", job->num_unpins); | |
588 | } |