1 /*
2 * Copyright © 2016 Broadcom
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include <stdio.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include "drm-uapi/v3d_drm.h"
28 #include "clif_dump.h"
29 #include "clif_private.h"
30 #include "util/list.h"
31 #include "util/ralloc.h"
32
33 #include "broadcom/cle/v3d_decoder.h"
34
35 struct reloc_worklist_entry *
clif_dump_add_address_to_worklist(struct clif_dump * clif,enum reloc_worklist_type type,uint32_t addr)36 clif_dump_add_address_to_worklist(struct clif_dump *clif,
37 enum reloc_worklist_type type,
38 uint32_t addr)
39 {
40 struct reloc_worklist_entry *entry =
41 rzalloc(clif, struct reloc_worklist_entry);
42 if (!entry)
43 return NULL;
44
45 entry->type = type;
46 entry->addr = addr;
47
48 list_addtail(&entry->link, &clif->worklist);
49
50 return entry;
51 }
52
53 struct clif_dump *
clif_dump_init(const struct v3d_device_info * devinfo,FILE * out,bool pretty,bool nobin)54 clif_dump_init(const struct v3d_device_info *devinfo,
55 FILE *out, bool pretty, bool nobin)
56 {
57 struct clif_dump *clif = rzalloc(NULL, struct clif_dump);
58
59 clif->devinfo = devinfo;
60 clif->out = out;
61 clif->spec = v3d_spec_load(devinfo);
62 clif->pretty = pretty;
63 clif->nobin = nobin;
64
65 list_inithead(&clif->worklist);
66
67 return clif;
68 }
69
70 void
clif_dump_destroy(struct clif_dump * clif)71 clif_dump_destroy(struct clif_dump *clif)
72 {
73 ralloc_free(clif);
74 }
75
76 struct clif_bo *
clif_lookup_bo(struct clif_dump * clif,uint32_t addr)77 clif_lookup_bo(struct clif_dump *clif, uint32_t addr)
78 {
79 for (int i = 0; i < clif->bo_count; i++) {
80 struct clif_bo *bo = &clif->bo[i];
81
82 if (addr >= bo->offset &&
83 addr < bo->offset + bo->size) {
84 return bo;
85 }
86 }
87
88 return NULL;
89 }
90
91 static bool
clif_lookup_vaddr(struct clif_dump * clif,uint32_t addr,void ** vaddr)92 clif_lookup_vaddr(struct clif_dump *clif, uint32_t addr, void **vaddr)
93 {
94 struct clif_bo *bo = clif_lookup_bo(clif, addr);
95 if (!bo)
96 return false;
97
98 *vaddr = bo->vaddr + addr - bo->offset;
99 return true;
100 }
101
102 #define out_uint(_clif, field) out(_clif, " /* %s = */ %u\n", \
103 #field, values-> field);
104
105 static bool
clif_dump_packet(struct clif_dump * clif,uint32_t offset,const uint8_t * cl,uint32_t * size,bool reloc_mode)106 clif_dump_packet(struct clif_dump *clif, uint32_t offset, const uint8_t *cl,
107 uint32_t *size, bool reloc_mode)
108 {
109
110 switch (clif->devinfo->ver) {
111 case 42:
112 return v3d42_clif_dump_packet(clif, offset, cl, size, reloc_mode);
113 case 71:
114 return v3d71_clif_dump_packet(clif, offset, cl, size, reloc_mode);
115 default:
116 break;
117 };
118 unreachable("Unknown HW version");
119 }
120
121 static uint32_t
clif_dump_cl(struct clif_dump * clif,uint32_t start,uint32_t end,bool reloc_mode)122 clif_dump_cl(struct clif_dump *clif, uint32_t start, uint32_t end,
123 bool reloc_mode)
124 {
125 struct clif_bo *bo = clif_lookup_bo(clif, start);
126 if (!bo) {
127 out(clif, "Failed to look up address 0x%08x\n",
128 start);
129 return 0;
130 }
131
132 void *start_vaddr = bo->vaddr + start - bo->offset;
133
134 /* The end address is optional (for example, a BRANCH instruction
135 * won't set an end), but is used for BCL/RCL termination.
136 */
137 void *end_vaddr = NULL;
138 if (end && !clif_lookup_vaddr(clif, end, &end_vaddr)) {
139 out(clif, "Failed to look up address 0x%08x\n",
140 end);
141 return 0;
142 }
143
144 if (!reloc_mode)
145 out(clif, "@format ctrllist /* [%s+0x%08x] */\n",
146 bo->name, start - bo->offset);
147
148 uint32_t size;
149 uint8_t *cl = start_vaddr;
150 while (clif_dump_packet(clif, start, cl, &size, reloc_mode)) {
151 cl += size;
152 start += size;
153
154 if (cl == end_vaddr)
155 break;
156 }
157
158 return (void *)cl - bo->vaddr;
159 }
160
161 /* Walks the worklist, parsing the relocs for any memory regions that might
162 * themselves have additional relocations.
163 */
164 static uint32_t
clif_dump_gl_shader_state_record(struct clif_dump * clif,struct reloc_worklist_entry * reloc,void * vaddr,bool including_gs)165 clif_dump_gl_shader_state_record(struct clif_dump *clif,
166 struct reloc_worklist_entry *reloc,
167 void *vaddr,
168 bool including_gs)
169 {
170 struct v3d_group *state = v3d_spec_find_struct(clif->spec,
171 "GL Shader State Record");
172 struct v3d_group *attr = v3d_spec_find_struct(clif->spec,
173 "GL Shader State Attribute Record");
174 assert(state);
175 assert(attr);
176 uint32_t offset = 0;
177
178 if (including_gs) {
179 struct v3d_group *gs_state = v3d_spec_find_struct(clif->spec,
180 "Geometry Shader State Record");
181 assert(gs_state);
182 out(clif, "@format shadrec_gl_geom\n");
183 v3d_print_group(clif, gs_state, 0, vaddr + offset);
184 offset += v3d_group_get_length(gs_state);
185 /* Extra pad when geometry/tessellation shader is present */
186 offset += 20;
187 }
188 out(clif, "@format shadrec_gl_main\n");
189 v3d_print_group(clif, state, 0, vaddr + offset);
190 offset += v3d_group_get_length(state);
191
192 for (int i = 0; i < reloc->shader_state.num_attrs; i++) {
193 out(clif, "@format shadrec_gl_attr /* %d */\n", i);
194 v3d_print_group(clif, attr, 0, vaddr + offset);
195 offset += v3d_group_get_length(attr);
196 }
197
198 return offset;
199 }
200
201 static void
clif_process_worklist(struct clif_dump * clif)202 clif_process_worklist(struct clif_dump *clif)
203 {
204 list_for_each_entry_safe(struct reloc_worklist_entry, reloc,
205 &clif->worklist, link) {
206 void *vaddr;
207 if (!clif_lookup_vaddr(clif, reloc->addr, &vaddr)) {
208 out(clif, "Failed to look up address 0x%08x\n",
209 reloc->addr);
210 continue;
211 }
212
213 switch (reloc->type) {
214 case reloc_cl:
215 clif_dump_cl(clif, reloc->addr, reloc->cl.end, true);
216 break;
217
218 case reloc_gl_shader_state:
219 case reloc_gl_including_gs_shader_state:
220 break;
221 case reloc_generic_tile_list:
222 clif_dump_cl(clif, reloc->addr,
223 reloc->generic_tile_list.end, true);
224 break;
225 }
226 }
227 }
228
229 static int
worklist_entry_compare(const void * a,const void * b)230 worklist_entry_compare(const void *a, const void *b)
231 {
232 return ((*(struct reloc_worklist_entry **)a)->addr -
233 (*(struct reloc_worklist_entry **)b)->addr);
234 }
235
236 static bool
clif_dump_if_blank(struct clif_dump * clif,struct clif_bo * bo,uint32_t start,uint32_t end)237 clif_dump_if_blank(struct clif_dump *clif, struct clif_bo *bo,
238 uint32_t start, uint32_t end)
239 {
240 for (int i = start; i < end; i++) {
241 if (((uint8_t *)bo->vaddr)[i] != 0)
242 return false;
243 }
244
245 out(clif, "\n");
246 out(clif, "@format blank %d /* [%s+0x%08x..0x%08x] */\n", end - start,
247 bo->name, start, end - 1);
248 return true;
249 }
250
251 /* Dumps the binary data in the BO from start to end (relative to the start of
252 * the BO).
253 */
254 static void
clif_dump_binary(struct clif_dump * clif,struct clif_bo * bo,uint32_t start,uint32_t end)255 clif_dump_binary(struct clif_dump *clif, struct clif_bo *bo,
256 uint32_t start, uint32_t end)
257 {
258 if (clif->pretty && clif->nobin)
259 return;
260
261 if (start == end)
262 return;
263
264 if (clif_dump_if_blank(clif, bo, start, end))
265 return;
266
267 out(clif, "@format binary /* [%s+0x%08x] */\n",
268 bo->name, start);
269
270 uint32_t offset = start;
271 int dumped_in_line = 0;
272 while (offset < end) {
273 if (clif_dump_if_blank(clif, bo, offset, end))
274 return;
275
276 if (end - offset >= 4) {
277 out(clif, "0x%08x ", *(uint32_t *)(bo->vaddr + offset));
278 offset += 4;
279 } else {
280 out(clif, "0x%02x ", *(uint8_t *)(bo->vaddr + offset));
281 offset++;
282 }
283
284 if (++dumped_in_line == 8) {
285 out(clif, "\n");
286 dumped_in_line = 0;
287 }
288 }
289 if (dumped_in_line)
290 out(clif, "\n");
291 }
292
293 /* Walks the list of relocations, dumping each buffer's contents (using our
294 * codegenned dump routines for pretty printing, and most importantly proper
295 * address references so that the CLIF parser can relocate buffers).
296 */
297 static void
clif_dump_buffers(struct clif_dump * clif)298 clif_dump_buffers(struct clif_dump *clif)
299 {
300 int num_relocs = 0;
301 list_for_each_entry(struct reloc_worklist_entry, reloc,
302 &clif->worklist, link) {
303 num_relocs++;
304 }
305 struct reloc_worklist_entry **relocs =
306 ralloc_array(clif, struct reloc_worklist_entry *, num_relocs);
307 int i = 0;
308 list_for_each_entry(struct reloc_worklist_entry, reloc,
309 &clif->worklist, link) {
310 relocs[i++] = reloc;
311 }
312 qsort(relocs, num_relocs, sizeof(*relocs), worklist_entry_compare);
313
314 struct clif_bo *bo = NULL;
315 uint32_t offset = 0;
316
317 for (i = 0; i < num_relocs; i++) {
318 struct reloc_worklist_entry *reloc = relocs[i];
319 struct clif_bo *new_bo = clif_lookup_bo(clif, reloc->addr);
320
321 if (!new_bo) {
322 out(clif, "Failed to look up address 0x%08x\n",
323 reloc->addr);
324 continue;
325 }
326
327 if (new_bo != bo) {
328 if (bo) {
329 /* Finish out the last of the last BO. */
330 clif_dump_binary(clif, bo,
331 offset,
332 bo->size);
333 }
334
335 out(clif, "\n");
336 out(clif, "@buffer %s\n", new_bo->name);
337 bo = new_bo;
338 offset = 0;
339 bo->dumped = true;
340 }
341
342 int reloc_offset = reloc->addr - bo->offset;
343 if (offset != reloc_offset)
344 clif_dump_binary(clif, bo, offset, reloc_offset);
345 offset = reloc_offset;
346
347 switch (reloc->type) {
348 case reloc_cl:
349 offset = clif_dump_cl(clif, reloc->addr, reloc->cl.end,
350 false);
351 out(clif, "\n");
352 break;
353
354 case reloc_gl_shader_state:
355 case reloc_gl_including_gs_shader_state:
356 offset += clif_dump_gl_shader_state_record(clif,
357 reloc,
358 bo->vaddr +
359 offset,
360 reloc->type == reloc_gl_including_gs_shader_state);
361 break;
362 case reloc_generic_tile_list:
363 offset = clif_dump_cl(clif, reloc->addr,
364 reloc->generic_tile_list.end,
365 false);
366 break;
367 }
368 out(clif, "\n");
369 }
370
371 if (bo) {
372 clif_dump_binary(clif, bo, offset, bo->size);
373 }
374
375 /* For any BOs that didn't have relocations, just dump them raw. */
376 for (int i = 0; i < clif->bo_count; i++) {
377 bo = &clif->bo[i];
378 if (bo->dumped)
379 continue;
380 out(clif, "@buffer %s\n", bo->name);
381 clif_dump_binary(clif, bo, 0, bo->size);
382 out(clif, "\n");
383 }
384 }
385
386 void
clif_dump_add_cl(struct clif_dump * clif,uint32_t start,uint32_t end)387 clif_dump_add_cl(struct clif_dump *clif, uint32_t start, uint32_t end)
388 {
389 struct reloc_worklist_entry *entry =
390 clif_dump_add_address_to_worklist(clif, reloc_cl, start);
391
392 entry->cl.end = end;
393 }
394
395 static int
clif_bo_offset_compare(const void * a,const void * b)396 clif_bo_offset_compare(const void *a, const void *b)
397 {
398 return ((struct clif_bo *)a)->offset - ((struct clif_bo *)b)->offset;
399 }
400
401 void
clif_dump(struct clif_dump * clif,const struct drm_v3d_submit_cl * submit)402 clif_dump(struct clif_dump *clif, const struct drm_v3d_submit_cl *submit)
403 {
404 clif_dump_add_cl(clif, submit->bcl_start, submit->bcl_end);
405 clif_dump_add_cl(clif, submit->rcl_start, submit->rcl_end);
406
407 qsort(clif->bo, clif->bo_count, sizeof(clif->bo[0]),
408 clif_bo_offset_compare);
409
410 /* A buffer needs to be defined before we can emit a CLIF address
411 * referencing it, so emit them all now.
412 */
413 for (int i = 0; i < clif->bo_count; i++) {
414 out(clif, "@createbuf_aligned 4096 %s\n", clif->bo[i].name);
415 }
416
417 /* Walk the worklist figuring out the locations of structs based on
418 * the CL contents.
419 */
420 clif_process_worklist(clif);
421
422 /* Dump the contents of the buffers using the relocations we found to
423 * pretty-print structures.
424 */
425 clif_dump_buffers(clif);
426
427 out(clif, "@add_bin 0\n ");
428 out_address(clif, submit->bcl_start);
429 out(clif, "\n ");
430 out_address(clif, submit->bcl_end);
431 out(clif, "\n ");
432 out_address(clif, submit->qma);
433 out(clif, "\n %d\n ", submit->qms);
434 out_address(clif, submit->qts);
435 out(clif, "\n");
436 out(clif, "@wait_bin_all_cores\n");
437
438 out(clif, "@add_render 0\n ");
439 out_address(clif, submit->rcl_start);
440 out(clif, "\n ");
441 out_address(clif, submit->rcl_end);
442 out(clif, "\n ");
443 out_address(clif, submit->qma);
444 out(clif, "\n");
445 out(clif, "@wait_render_all_cores\n");
446 }
447
448 void
clif_dump_add_bo(struct clif_dump * clif,const char * name,uint32_t offset,uint32_t size,void * vaddr)449 clif_dump_add_bo(struct clif_dump *clif, const char *name,
450 uint32_t offset, uint32_t size, void *vaddr)
451 {
452 if (clif->bo_count >= clif->bo_array_size) {
453 clif->bo_array_size = MAX2(4, clif->bo_array_size * 2);
454 clif->bo = reralloc(clif, clif->bo, struct clif_bo,
455 clif->bo_array_size);
456 }
457
458 /* CLIF relocs use the buffer name, so make sure they're unique. */
459 for (int i = 0; i < clif->bo_count; i++)
460 assert(strcmp(clif->bo[i].name, name) != 0);
461
462 clif->bo[clif->bo_count].name = ralloc_strdup(clif, name);
463 clif->bo[clif->bo_count].offset = offset;
464 clif->bo[clif->bo_count].size = size;
465 clif->bo[clif->bo_count].vaddr = vaddr;
466 clif->bo[clif->bo_count].dumped = false;
467 clif->bo_count++;
468 }
469