Merge remote branch 'main/master' into radeon-rewrite
[mesa.git] / src / mesa / drivers / dri / radeon / radeon_cs_legacy.c
1 /*
2 * Copyright © 2008 Nicolai Haehnle
3 * Copyright © 2008 Jérôme Glisse
4 * 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
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
18 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
19 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
20 * USE OR OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * The above copyright notice and this permission notice (including the
23 * next paragraph) shall be included in all copies or substantial portions
24 * of the Software.
25 */
26 /*
27 * Authors:
28 * Aapo Tahkola <aet@rasterburn.org>
29 * Nicolai Haehnle <prefect_@gmx.net>
30 * Jérôme Glisse <glisse@freedesktop.org>
31 */
32 #include <errno.h>
33
34 #include "radeon_bocs_wrapper.h"
35
36 struct cs_manager_legacy {
37 struct radeon_cs_manager base;
38 struct radeon_context *ctx;
39 /* hack for scratch stuff */
40 uint32_t pending_age;
41 uint32_t pending_count;
42
43
44 };
45
46 struct cs_reloc_legacy {
47 struct radeon_cs_reloc base;
48 uint32_t cindices;
49 uint32_t *indices;
50 };
51
52
53 static struct radeon_cs *cs_create(struct radeon_cs_manager *csm,
54 uint32_t ndw)
55 {
56 struct radeon_cs *cs;
57
58 cs = (struct radeon_cs*)calloc(1, sizeof(struct radeon_cs));
59 if (cs == NULL) {
60 return NULL;
61 }
62 cs->csm = csm;
63 cs->ndw = (ndw + 0x3FF) & (~0x3FF);
64 cs->packets = (uint32_t*)malloc(4*cs->ndw);
65 if (cs->packets == NULL) {
66 free(cs);
67 return NULL;
68 }
69 cs->relocs_total_size = 0;
70 return cs;
71 }
72
73 static int cs_write_reloc(struct radeon_cs *cs,
74 struct radeon_bo *bo,
75 uint32_t read_domain,
76 uint32_t write_domain,
77 uint32_t flags)
78 {
79 struct cs_reloc_legacy *relocs;
80 int i;
81
82 relocs = (struct cs_reloc_legacy *)cs->relocs;
83 /* check domains */
84 if ((read_domain && write_domain) || (!read_domain && !write_domain)) {
85 /* in one CS a bo can only be in read or write domain but not
86 * in read & write domain at the same sime
87 */
88 return -EINVAL;
89 }
90 if (read_domain == RADEON_GEM_DOMAIN_CPU) {
91 return -EINVAL;
92 }
93 if (write_domain == RADEON_GEM_DOMAIN_CPU) {
94 return -EINVAL;
95 }
96 /* check if bo is already referenced */
97 for(i = 0; i < cs->crelocs; i++) {
98 uint32_t *indices;
99
100 if (relocs[i].base.bo->handle == bo->handle) {
101 /* Check domains must be in read or write. As we check already
102 * checked that in argument one of the read or write domain was
103 * set we only need to check that if previous reloc as the read
104 * domain set then the read_domain should also be set for this
105 * new relocation.
106 */
107 if (relocs[i].base.read_domain && !read_domain) {
108 return -EINVAL;
109 }
110 if (relocs[i].base.write_domain && !write_domain) {
111 return -EINVAL;
112 }
113 relocs[i].base.read_domain |= read_domain;
114 relocs[i].base.write_domain |= write_domain;
115 /* save indice */
116 relocs[i].cindices++;
117 indices = (uint32_t*)realloc(relocs[i].indices,
118 relocs[i].cindices * 4);
119 if (indices == NULL) {
120 relocs[i].cindices -= 1;
121 return -ENOMEM;
122 }
123 relocs[i].indices = indices;
124 relocs[i].indices[relocs[i].cindices - 1] = cs->cdw - 1;
125 return 0;
126 }
127 }
128 /* add bo to reloc */
129 relocs = (struct cs_reloc_legacy*)
130 realloc(cs->relocs,
131 sizeof(struct cs_reloc_legacy) * (cs->crelocs + 1));
132 if (relocs == NULL) {
133 return -ENOMEM;
134 }
135 cs->relocs = relocs;
136 relocs[cs->crelocs].base.bo = bo;
137 relocs[cs->crelocs].base.read_domain = read_domain;
138 relocs[cs->crelocs].base.write_domain = write_domain;
139 relocs[cs->crelocs].base.flags = flags;
140 relocs[cs->crelocs].indices = (uint32_t*)malloc(4);
141 if (relocs[cs->crelocs].indices == NULL) {
142 return -ENOMEM;
143 }
144 relocs[cs->crelocs].indices[0] = cs->cdw - 1;
145 relocs[cs->crelocs].cindices = 1;
146 cs->relocs_total_size += radeon_bo_legacy_relocs_size(bo);
147 cs->crelocs++;
148 radeon_bo_ref(bo);
149 return 0;
150 }
151
152 static int cs_begin(struct radeon_cs *cs,
153 uint32_t ndw,
154 const char *file,
155 const char *func,
156 int line)
157 {
158 if (cs->section) {
159 fprintf(stderr, "CS already in a section(%s,%s,%d)\n",
160 cs->section_file, cs->section_func, cs->section_line);
161 fprintf(stderr, "CS can't start section(%s,%s,%d)\n",
162 file, func, line);
163 return -EPIPE;
164 }
165 cs->section = 1;
166 cs->section_ndw = ndw;
167 cs->section_cdw = 0;
168 cs->section_file = file;
169 cs->section_func = func;
170 cs->section_line = line;
171
172
173 if (cs->cdw + ndw > cs->ndw) {
174 uint32_t tmp, *ptr;
175 int num = (ndw > 0x3FF) ? ndw : 0x3FF;
176
177 tmp = (cs->cdw + 1 + num) & (~num);
178 ptr = (uint32_t*)realloc(cs->packets, 4 * tmp);
179 if (ptr == NULL) {
180 return -ENOMEM;
181 }
182 cs->packets = ptr;
183 cs->ndw = tmp;
184 }
185
186 return 0;
187 }
188
189 static int cs_end(struct radeon_cs *cs,
190 const char *file,
191 const char *func,
192 int line)
193
194 {
195 if (!cs->section) {
196 fprintf(stderr, "CS no section to end at (%s,%s,%d)\n",
197 file, func, line);
198 return -EPIPE;
199 }
200 cs->section = 0;
201 if (cs->section_ndw != cs->section_cdw) {
202 fprintf(stderr, "CS section size missmatch start at (%s,%s,%d) %d vs %d\n",
203 cs->section_file, cs->section_func, cs->section_line, cs->section_ndw, cs->section_cdw);
204 fprintf(stderr, "CS section end at (%s,%s,%d)\n",
205 file, func, line);
206 return -EPIPE;
207 }
208 return 0;
209 }
210
211 static int cs_process_relocs(struct radeon_cs *cs)
212 {
213 struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm;
214 struct cs_reloc_legacy *relocs;
215 int i, j, r;
216
217 csm = (struct cs_manager_legacy*)cs->csm;
218 relocs = (struct cs_reloc_legacy *)cs->relocs;
219 restart:
220 for (i = 0; i < cs->crelocs; i++) {
221 for (j = 0; j < relocs[i].cindices; j++) {
222 uint32_t soffset, eoffset;
223
224 r = radeon_bo_legacy_validate(relocs[i].base.bo,
225 &soffset, &eoffset);
226 if (r == -EAGAIN)
227 goto restart;
228 if (r) {
229 fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n",
230 relocs[i].base.bo, soffset, eoffset);
231 return r;
232 }
233 cs->packets[relocs[i].indices[j]] += soffset;
234 if (cs->packets[relocs[i].indices[j]] >= eoffset) {
235 /* radeon_bo_debug(relocs[i].base.bo, 12); */
236 fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n",
237 relocs[i].base.bo, soffset, eoffset);
238 fprintf(stderr, "above end: %p 0x%08X 0x%08X\n",
239 relocs[i].base.bo,
240 cs->packets[relocs[i].indices[j]],
241 eoffset);
242 exit(0);
243 return -EINVAL;
244 }
245 }
246 }
247 return 0;
248 }
249
250 static int cs_set_age(struct radeon_cs *cs)
251 {
252 struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm;
253 struct cs_reloc_legacy *relocs;
254 int i;
255
256 relocs = (struct cs_reloc_legacy *)cs->relocs;
257 for (i = 0; i < cs->crelocs; i++) {
258 radeon_bo_legacy_pending(relocs[i].base.bo, csm->pending_age);
259 radeon_bo_unref(relocs[i].base.bo);
260 }
261 return 0;
262 }
263
264 static void dump_cmdbuf(struct radeon_cs *cs)
265 {
266 int i;
267 for (i = 0; i < cs->cdw; i++){
268 fprintf(stderr,"%x: %08x\n", i, cs->packets[i]);
269 }
270
271 }
272 static int cs_emit(struct radeon_cs *cs)
273 {
274 struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm;
275 drm_radeon_cmd_buffer_t cmd;
276 drm_r300_cmd_header_t age;
277 uint64_t ull;
278 int r;
279
280 csm->ctx->vtbl.emit_cs_header(cs, csm->ctx);
281
282 /* append buffer age */
283 if (IS_R300_CLASS(csm->ctx->radeonScreen)) {
284 age.scratch.cmd_type = R300_CMD_SCRATCH;
285 /* Scratch register 2 corresponds to what radeonGetAge polls */
286 csm->pending_age = 0;
287 csm->pending_count = 1;
288 ull = (uint64_t) (intptr_t) &csm->pending_age;
289 age.scratch.reg = 2;
290 age.scratch.n_bufs = 1;
291 age.scratch.flags = 0;
292 radeon_cs_write_dword(cs, age.u);
293 radeon_cs_write_qword(cs, ull);
294 radeon_cs_write_dword(cs, 0);
295 }
296
297 r = cs_process_relocs(cs);
298 if (r) {
299 return 0;
300 }
301
302 cmd.buf = (char *)cs->packets;
303 cmd.bufsz = cs->cdw * 4;
304 if (csm->ctx->state.scissor.enabled) {
305 cmd.nbox = csm->ctx->state.scissor.numClipRects;
306 cmd.boxes = (drm_clip_rect_t *) csm->ctx->state.scissor.pClipRects;
307 } else {
308 cmd.nbox = csm->ctx->numClipRects;
309 cmd.boxes = (drm_clip_rect_t *) csm->ctx->pClipRects;
310 }
311
312 //dump_cmdbuf(cs);
313
314 r = drmCommandWrite(cs->csm->fd, DRM_RADEON_CMDBUF, &cmd, sizeof(cmd));
315 if (r) {
316 return r;
317 }
318 if (!IS_R300_CLASS(csm->ctx->radeonScreen)) {
319 drm_radeon_irq_emit_t emit_cmd;
320 emit_cmd.irq_seq = &csm->pending_age;
321 r = drmCommandWrite(cs->csm->fd, DRM_RADEON_IRQ_EMIT, &emit_cmd, sizeof(emit_cmd));
322 if (r) {
323 return r;
324 }
325 }
326 cs_set_age(cs);
327
328 cs->csm->read_used = 0;
329 cs->csm->vram_write_used = 0;
330 cs->csm->gart_write_used = 0;
331 return 0;
332 }
333
334 static void inline cs_free_reloc(void *relocs_p, int crelocs)
335 {
336 struct cs_reloc_legacy *relocs = relocs_p;
337 int i;
338 if (!relocs_p)
339 return;
340 for (i = 0; i < crelocs; i++)
341 free(relocs[i].indices);
342 }
343
344 static int cs_destroy(struct radeon_cs *cs)
345 {
346 cs_free_reloc(cs->relocs, cs->crelocs);
347 free(cs->relocs);
348 free(cs->packets);
349 free(cs);
350 return 0;
351 }
352
353 static int cs_erase(struct radeon_cs *cs)
354 {
355 cs_free_reloc(cs->relocs, cs->crelocs);
356 free(cs->relocs);
357 cs->relocs_total_size = 0;
358 cs->relocs = NULL;
359 cs->crelocs = 0;
360 cs->cdw = 0;
361 cs->section = 0;
362 return 0;
363 }
364
365 static int cs_need_flush(struct radeon_cs *cs)
366 {
367 /* this function used to flush when the BO usage got to
368 * a certain size, now the higher levels handle this better */
369 return 0;
370 }
371
372 static void cs_print(struct radeon_cs *cs, FILE *file)
373 {
374 }
375
376 static int cs_check_space(struct radeon_cs *cs, struct radeon_cs_space_check *bos, int num_bo)
377 {
378 struct radeon_cs_manager *csm = cs->csm;
379 int this_op_read = 0, this_op_gart_write = 0, this_op_vram_write = 0;
380 uint32_t read_domains, write_domain;
381 int i;
382 struct radeon_bo *bo;
383
384 /* check the totals for this operation */
385
386 if (num_bo == 0)
387 return 0;
388
389 /* prepare */
390 for (i = 0; i < num_bo; i++) {
391 bo = bos[i].bo;
392
393 bos[i].new_accounted = 0;
394 read_domains = bos[i].read_domains;
395 write_domain = bos[i].write_domain;
396
397 /* pinned bos don't count */
398 if (radeon_legacy_bo_is_static(bo))
399 continue;
400
401 /* already accounted this bo */
402 if (write_domain && (write_domain == bo->space_accounted))
403 continue;
404
405 if (read_domains && ((read_domains << 16) == bo->space_accounted))
406 continue;
407
408 if (bo->space_accounted == 0) {
409 if (write_domain == RADEON_GEM_DOMAIN_VRAM)
410 this_op_vram_write += bo->size;
411 else if (write_domain == RADEON_GEM_DOMAIN_GTT)
412 this_op_gart_write += bo->size;
413 else
414 this_op_read += bo->size;
415 bos[i].new_accounted = (read_domains << 16) | write_domain;
416 } else {
417 uint16_t old_read, old_write;
418
419 old_read = bo->space_accounted >> 16;
420 old_write = bo->space_accounted & 0xffff;
421
422 if (write_domain && (old_read & write_domain)) {
423 bos[i].new_accounted = write_domain;
424 /* moving from read to a write domain */
425 if (write_domain == RADEON_GEM_DOMAIN_VRAM) {
426 this_op_read -= bo->size;
427 this_op_vram_write += bo->size;
428 } else if (write_domain == RADEON_GEM_DOMAIN_VRAM) {
429 this_op_read -= bo->size;
430 this_op_gart_write += bo->size;
431 }
432 } else if (read_domains & old_write) {
433 bos[i].new_accounted = bo->space_accounted & 0xffff;
434 } else {
435 /* rewrite the domains */
436 if (write_domain != old_write)
437 fprintf(stderr,"WRITE DOMAIN RELOC FAILURE 0x%x %d %d\n", bo->handle, write_domain, old_write);
438 if (read_domains != old_read)
439 fprintf(stderr,"READ DOMAIN RELOC FAILURE 0x%x %d %d\n", bo->handle, read_domains, old_read);
440 return RADEON_CS_SPACE_FLUSH;
441 }
442 }
443 }
444
445 if (this_op_read < 0)
446 this_op_read = 0;
447
448 /* check sizes - operation first */
449 if ((this_op_read + this_op_gart_write > csm->gart_limit) ||
450 (this_op_vram_write > csm->vram_limit)) {
451 return RADEON_CS_SPACE_OP_TO_BIG;
452 }
453
454 if (((csm->vram_write_used + this_op_vram_write) > csm->vram_limit) ||
455 ((csm->read_used + csm->gart_write_used + this_op_gart_write + this_op_read) > csm->gart_limit)) {
456 return RADEON_CS_SPACE_FLUSH;
457 }
458
459 csm->gart_write_used += this_op_gart_write;
460 csm->vram_write_used += this_op_vram_write;
461 csm->read_used += this_op_read;
462 /* commit */
463 for (i = 0; i < num_bo; i++) {
464 bo = bos[i].bo;
465 bo->space_accounted = bos[i].new_accounted;
466 }
467
468 return RADEON_CS_SPACE_OK;
469 }
470
471 static struct radeon_cs_funcs radeon_cs_legacy_funcs = {
472 cs_create,
473 cs_write_reloc,
474 cs_begin,
475 cs_end,
476 cs_emit,
477 cs_destroy,
478 cs_erase,
479 cs_need_flush,
480 cs_print,
481 cs_check_space
482 };
483
484 struct radeon_cs_manager *radeon_cs_manager_legacy_ctor(struct radeon_context *ctx)
485 {
486 struct cs_manager_legacy *csm;
487
488 csm = (struct cs_manager_legacy*)
489 calloc(1, sizeof(struct cs_manager_legacy));
490 if (csm == NULL) {
491 return NULL;
492 }
493 csm->base.funcs = &radeon_cs_legacy_funcs;
494 csm->base.fd = ctx->dri.fd;
495 csm->ctx = ctx;
496 csm->pending_age = 1;
497 return (struct radeon_cs_manager*)csm;
498 }
499
500 void radeon_cs_manager_legacy_dtor(struct radeon_cs_manager *csm)
501 {
502 free(csm);
503 }
504