pan/decode: Resolve crash with NULL attr/varyings
[mesa.git] / src / panfrost / pandecode / decode.c
1 /*
2 * Copyright (C) 2017-2019 Alyssa Rosenzweig
3 * Copyright (C) 2017-2019 Connor Abbott
4 * Copyright (C) 2019 Collabora, Ltd.
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 (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * SOFTWARE.
24 */
25
26 #include <panfrost-job.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <memory.h>
30 #include <stdbool.h>
31 #include <stdarg.h>
32 #include "decode.h"
33 #include "util/u_math.h"
34
35 #include "pan_pretty_print.h"
36 #include "midgard/disassemble.h"
37 #include "bifrost/disassemble.h"
38
39 int pandecode_jc(mali_ptr jc_gpu_va, bool bifrost);
40
41 #define MEMORY_PROP(obj, p) {\
42 if (obj->p) { \
43 char *a = pointer_as_memory_reference(obj->p); \
44 pandecode_prop("%s = %s", #p, a); \
45 free(a); \
46 } \
47 }
48
49 #define MEMORY_PROP_DIR(obj, p) {\
50 if (obj.p) { \
51 char *a = pointer_as_memory_reference(obj.p); \
52 pandecode_prop("%s = %s", #p, a); \
53 free(a); \
54 } \
55 }
56
57 #define DYN_MEMORY_PROP(obj, no, p) { \
58 if (obj->p) \
59 pandecode_prop("%s = %s_%d_p", #p, #p, no); \
60 }
61
62 /* Semantic logging type.
63 *
64 * Raw: for raw messages to be printed as is.
65 * Message: for helpful information to be commented out in replays.
66 * Property: for properties of a struct
67 *
68 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
69 */
70
71 enum pandecode_log_type {
72 PANDECODE_RAW,
73 PANDECODE_MESSAGE,
74 PANDECODE_PROPERTY
75 };
76
77 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
78 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
79 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
80
81 unsigned pandecode_indent = 0;
82
83 static void
84 pandecode_make_indent(void)
85 {
86 for (unsigned i = 0; i < pandecode_indent; ++i)
87 printf(" ");
88 }
89
90 static void
91 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
92 {
93 va_list ap;
94
95 pandecode_make_indent();
96
97 if (type == PANDECODE_MESSAGE)
98 printf("// ");
99 else if (type == PANDECODE_PROPERTY)
100 printf(".");
101
102 va_start(ap, format);
103 vprintf(format, ap);
104 va_end(ap);
105
106 if (type == PANDECODE_PROPERTY)
107 printf(",\n");
108 }
109
110 static void
111 pandecode_log_cont(const char *format, ...)
112 {
113 va_list ap;
114
115 va_start(ap, format);
116 vprintf(format, ap);
117 va_end(ap);
118 }
119
120 struct pandecode_flag_info {
121 u64 flag;
122 const char *name;
123 };
124
125 static void
126 pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
127 u64 flags)
128 {
129 bool decodable_flags_found = false;
130
131 for (int i = 0; flag_info[i].name; i++) {
132 if ((flags & flag_info[i].flag) != flag_info[i].flag)
133 continue;
134
135 if (!decodable_flags_found) {
136 decodable_flags_found = true;
137 } else {
138 pandecode_log_cont(" | ");
139 }
140
141 pandecode_log_cont("%s", flag_info[i].name);
142
143 flags &= ~flag_info[i].flag;
144 }
145
146 if (decodable_flags_found) {
147 if (flags)
148 pandecode_log_cont(" | 0x%" PRIx64, flags);
149 } else {
150 pandecode_log_cont("0x%" PRIx64, flags);
151 }
152 }
153
154 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
155 static const struct pandecode_flag_info gl_enable_flag_info[] = {
156 FLAG_INFO(OCCLUSION_QUERY),
157 FLAG_INFO(OCCLUSION_PRECISE),
158 FLAG_INFO(FRONT_CCW_TOP),
159 FLAG_INFO(CULL_FACE_FRONT),
160 FLAG_INFO(CULL_FACE_BACK),
161 {}
162 };
163 #undef FLAG_INFO
164
165 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
166 static const struct pandecode_flag_info clear_flag_info[] = {
167 FLAG_INFO(FAST),
168 FLAG_INFO(SLOW),
169 FLAG_INFO(SLOW_STENCIL),
170 {}
171 };
172 #undef FLAG_INFO
173
174 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
175 static const struct pandecode_flag_info mask_flag_info[] = {
176 FLAG_INFO(R),
177 FLAG_INFO(G),
178 FLAG_INFO(B),
179 FLAG_INFO(A),
180 {}
181 };
182 #undef FLAG_INFO
183
184 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
185 static const struct pandecode_flag_info u3_flag_info[] = {
186 FLAG_INFO(HAS_MSAA),
187 FLAG_INFO(CAN_DISCARD),
188 FLAG_INFO(HAS_BLEND_SHADER),
189 FLAG_INFO(DEPTH_TEST),
190 {}
191 };
192
193 static const struct pandecode_flag_info u4_flag_info[] = {
194 FLAG_INFO(NO_MSAA),
195 FLAG_INFO(NO_DITHER),
196 FLAG_INFO(DEPTH_RANGE_A),
197 FLAG_INFO(DEPTH_RANGE_B),
198 FLAG_INFO(STENCIL_TEST),
199 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER),
200 {}
201 };
202 #undef FLAG_INFO
203
204 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
205 static const struct pandecode_flag_info fb_fmt_flag_info[] = {
206 FLAG_INFO(MSAA_A),
207 FLAG_INFO(MSAA_B),
208 FLAG_INFO(MSAA_8),
209 {}
210 };
211 #undef FLAG_INFO
212
213 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
214 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
215 FLAG_INFO(MSAA),
216 FLAG_INFO(SRGB),
217 {}
218 };
219 #undef FLAG_INFO
220
221 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
222 static const struct pandecode_flag_info mfbd_extra_flag_info[] = {
223 FLAG_INFO(PRESENT),
224 FLAG_INFO(AFBC),
225 FLAG_INFO(ZS),
226 {}
227 };
228 #undef FLAG_INFO
229
230 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
231 static const struct pandecode_flag_info shader_midgard1_flag_info [] = {
232 FLAG_INFO(EARLY_Z),
233 FLAG_INFO(HELPER_INVOCATIONS),
234 FLAG_INFO(READS_TILEBUFFER),
235 FLAG_INFO(READS_ZS),
236 {}
237 };
238 #undef FLAG_INFO
239
240 #define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
241 static const struct pandecode_flag_info mfbd_flag_info [] = {
242 FLAG_INFO(DEPTH_WRITE),
243 FLAG_INFO(EXTRA),
244 {}
245 };
246 #undef FLAG_INFO
247
248 #define FLAG_INFO(flag) { MALI_SAMP_##flag, "MALI_SAMP_" #flag }
249 static const struct pandecode_flag_info sampler_flag_info [] = {
250 FLAG_INFO(MAG_NEAREST),
251 FLAG_INFO(MIN_NEAREST),
252 FLAG_INFO(MIP_LINEAR_1),
253 FLAG_INFO(MIP_LINEAR_2),
254 FLAG_INFO(NORM_COORDS),
255 {}
256 };
257 #undef FLAG_INFO
258
259 extern char *replace_fragment;
260 extern char *replace_vertex;
261
262 static char *
263 pandecode_job_type(enum mali_job_type type)
264 {
265 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
266
267 switch (type) {
268 DEFINE_CASE(NULL);
269 DEFINE_CASE(SET_VALUE);
270 DEFINE_CASE(CACHE_FLUSH);
271 DEFINE_CASE(COMPUTE);
272 DEFINE_CASE(VERTEX);
273 DEFINE_CASE(TILER);
274 DEFINE_CASE(FUSED);
275 DEFINE_CASE(FRAGMENT);
276
277 case JOB_NOT_STARTED:
278 return "NOT_STARTED";
279
280 default:
281 pandecode_log("Warning! Unknown job type %x\n", type);
282 return "!?!?!?";
283 }
284
285 #undef DEFINE_CASE
286 }
287
288 static char *
289 pandecode_draw_mode(enum mali_draw_mode mode)
290 {
291 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
292
293 switch (mode) {
294 DEFINE_CASE(DRAW_NONE);
295 DEFINE_CASE(POINTS);
296 DEFINE_CASE(LINES);
297 DEFINE_CASE(TRIANGLES);
298 DEFINE_CASE(TRIANGLE_STRIP);
299 DEFINE_CASE(TRIANGLE_FAN);
300 DEFINE_CASE(LINE_STRIP);
301 DEFINE_CASE(LINE_LOOP);
302 DEFINE_CASE(POLYGON);
303 DEFINE_CASE(QUADS);
304 DEFINE_CASE(QUAD_STRIP);
305
306 default:
307 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
308 }
309
310 #undef DEFINE_CASE
311 }
312
313 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
314 static char *
315 pandecode_func(enum mali_func mode)
316 {
317 switch (mode) {
318 DEFINE_CASE(NEVER);
319 DEFINE_CASE(LESS);
320 DEFINE_CASE(EQUAL);
321 DEFINE_CASE(LEQUAL);
322 DEFINE_CASE(GREATER);
323 DEFINE_CASE(NOTEQUAL);
324 DEFINE_CASE(GEQUAL);
325 DEFINE_CASE(ALWAYS);
326
327 default:
328 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
329 }
330 }
331 #undef DEFINE_CASE
332
333 /* Why is this duplicated? Who knows... */
334 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
335 static char *
336 pandecode_alt_func(enum mali_alt_func mode)
337 {
338 switch (mode) {
339 DEFINE_CASE(NEVER);
340 DEFINE_CASE(LESS);
341 DEFINE_CASE(EQUAL);
342 DEFINE_CASE(LEQUAL);
343 DEFINE_CASE(GREATER);
344 DEFINE_CASE(NOTEQUAL);
345 DEFINE_CASE(GEQUAL);
346 DEFINE_CASE(ALWAYS);
347
348 default:
349 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
350 }
351 }
352 #undef DEFINE_CASE
353
354 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
355 static char *
356 pandecode_stencil_op(enum mali_stencil_op op)
357 {
358 switch (op) {
359 DEFINE_CASE(KEEP);
360 DEFINE_CASE(REPLACE);
361 DEFINE_CASE(ZERO);
362 DEFINE_CASE(INVERT);
363 DEFINE_CASE(INCR_WRAP);
364 DEFINE_CASE(DECR_WRAP);
365 DEFINE_CASE(INCR);
366 DEFINE_CASE(DECR);
367
368 default:
369 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
370 }
371 }
372
373 #undef DEFINE_CASE
374
375 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
376 static char *pandecode_attr_mode(enum mali_attr_mode mode)
377 {
378 switch(mode) {
379 DEFINE_CASE(UNUSED);
380 DEFINE_CASE(LINEAR);
381 DEFINE_CASE(POT_DIVIDE);
382 DEFINE_CASE(MODULO);
383 DEFINE_CASE(NPOT_DIVIDE);
384 DEFINE_CASE(IMAGE);
385 DEFINE_CASE(INTERNAL);
386 default:
387 return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
388 }
389 }
390
391 #undef DEFINE_CASE
392
393 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
394 static char *
395 pandecode_channel(enum mali_channel channel)
396 {
397 switch (channel) {
398 DEFINE_CASE(RED);
399 DEFINE_CASE(GREEN);
400 DEFINE_CASE(BLUE);
401 DEFINE_CASE(ALPHA);
402 DEFINE_CASE(ZERO);
403 DEFINE_CASE(ONE);
404 DEFINE_CASE(RESERVED_0);
405 DEFINE_CASE(RESERVED_1);
406
407 default:
408 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
409 }
410 }
411 #undef DEFINE_CASE
412
413 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
414 static char *
415 pandecode_wrap_mode(enum mali_wrap_mode op)
416 {
417 switch (op) {
418 DEFINE_CASE(REPEAT);
419 DEFINE_CASE(CLAMP_TO_EDGE);
420 DEFINE_CASE(CLAMP_TO_BORDER);
421 DEFINE_CASE(MIRRORED_REPEAT);
422
423 default:
424 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
425 }
426 }
427 #undef DEFINE_CASE
428
429 #define DEFINE_CASE(name) case MALI_TEX_## name: return "MALI_TEX_" #name
430 static char *
431 pandecode_texture_type(enum mali_texture_type type)
432 {
433 switch (type) {
434 DEFINE_CASE(1D);
435 DEFINE_CASE(2D);
436 DEFINE_CASE(3D);
437 DEFINE_CASE(CUBE);
438
439 default:
440 unreachable("Unknown case");
441 }
442 }
443 #undef DEFINE_CASE
444
445 #define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
446 static char *
447 pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt)
448 {
449 switch (fmt) {
450 DEFINE_CASE(TILED);
451 DEFINE_CASE(UNKNOWN);
452 DEFINE_CASE(LINEAR);
453 DEFINE_CASE(AFBC);
454
455 default:
456 unreachable("Invalid case");
457 }
458 }
459 #undef DEFINE_CASE
460
461 /* Midgard's tiler descriptor is embedded within the
462 * larger FBD */
463
464 static void
465 pandecode_midgard_tiler_descriptor(const struct midgard_tiler_descriptor *t)
466 {
467 pandecode_log(".tiler = {\n");
468 pandecode_indent++;
469
470 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
471 pandecode_prop("flags = 0x%" PRIx16, t->flags);
472 pandecode_prop("polygon_list_size = 0x%x", t->polygon_list_size);
473
474 MEMORY_PROP(t, polygon_list);
475 MEMORY_PROP(t, polygon_list_body);
476
477 MEMORY_PROP(t, heap_start);
478
479 if (t->heap_start == t->heap_end) {
480 /* Print identically to show symmetry for empty tiler heaps */
481 MEMORY_PROP(t, heap_start);
482 } else {
483 /* Points to the end of a buffer */
484 char *a = pointer_as_memory_reference(t->heap_end - 1);
485 pandecode_prop("heap_end = %s + 1", a);
486 free(a);
487 }
488
489 bool nonzero_weights = false;
490
491 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
492 nonzero_weights |= t->weights[w] != 0x0;
493 }
494
495 if (nonzero_weights) {
496 pandecode_log(".weights = {");
497
498 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
499 pandecode_log("%d, ", t->weights[w]);
500 }
501
502 pandecode_log("},");
503 }
504
505 pandecode_indent--;
506 pandecode_log("}\n");
507 }
508
509 static void
510 pandecode_sfbd(uint64_t gpu_va, int job_no)
511 {
512 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
513 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
514
515 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
516 pandecode_indent++;
517
518 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
519 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
520
521 pandecode_log(".format = ");
522 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
523 pandecode_log_cont(",\n");
524
525 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
526 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
527
528 MEMORY_PROP(s, framebuffer);
529 pandecode_prop("stride = %d", s->stride);
530
531 /* Earlier in the actual commandstream -- right before width -- but we
532 * delay to flow nicer */
533
534 pandecode_log(".clear_flags = ");
535 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
536 pandecode_log_cont(",\n");
537
538 if (s->depth_buffer | s->depth_buffer_enable) {
539 MEMORY_PROP(s, depth_buffer);
540 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
541 }
542
543 if (s->stencil_buffer | s->stencil_buffer_enable) {
544 MEMORY_PROP(s, stencil_buffer);
545 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
546 }
547
548 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
549 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
550 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
551 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
552 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
553 }
554
555 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
556 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
557 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
558 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
559 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
560 }
561
562 if (s->clear_stencil) {
563 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
564 }
565
566 MEMORY_PROP(s, unknown_address_0);
567 const struct midgard_tiler_descriptor t = s->tiler;
568 pandecode_midgard_tiler_descriptor(&t);
569
570 pandecode_indent--;
571 pandecode_log("};\n");
572
573 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
574 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
575 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
576 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
577
578 printf(".zero3 = {");
579
580 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
581 printf("%X, ", s->zero3[i]);
582
583 printf("},\n");
584
585 printf(".zero6 = {");
586
587 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
588 printf("%X, ", s->zero6[i]);
589
590 printf("},\n");
591 }
592
593 static void
594 pandecode_u32_slide(unsigned name, const u32 *slide, unsigned count)
595 {
596 pandecode_log(".unknown%d = {", name);
597
598 for (int i = 0; i < count; ++i)
599 printf("%X, ", slide[i]);
600
601 pandecode_log("},\n");
602 }
603
604 #define SHORT_SLIDE(num) \
605 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
606
607 static void
608 pandecode_compute_fbd(uint64_t gpu_va, int job_no)
609 {
610 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
611 const struct mali_compute_fbd *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
612
613 pandecode_log("struct mali_compute_fbd framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
614 pandecode_indent++;
615
616 SHORT_SLIDE(1);
617
618 pandecode_indent--;
619 printf("},\n");
620 }
621
622 static void
623 pandecode_swizzle(unsigned swizzle)
624 {
625 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
626 pandecode_channel((swizzle >> 0) & 0x7),
627 pandecode_channel((swizzle >> 3) & 0x7),
628 pandecode_channel((swizzle >> 6) & 0x7),
629 pandecode_channel((swizzle >> 9) & 0x7));
630 }
631
632 static void
633 pandecode_rt_format(struct mali_rt_format format)
634 {
635 pandecode_log(".format = {\n");
636 pandecode_indent++;
637
638 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
639 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
640 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
641
642 pandecode_prop("block = %s",
643 pandecode_mfbd_block_format(format.block));
644
645 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
646 MALI_NEGATIVE(format.nr_channels));
647
648 pandecode_log(".flags = ");
649 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
650 pandecode_log_cont(",\n");
651
652 pandecode_swizzle(format.swizzle);
653
654 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
655
656 pandecode_indent--;
657 pandecode_log("},\n");
658 }
659
660 static void
661 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
662 {
663 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
664 pandecode_indent++;
665
666 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
667 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
668 struct pandecode_mapped_memory *mem =
669 pandecode_find_mapped_gpu_mem_containing(rt_va);
670 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
671
672 pandecode_log("{\n");
673 pandecode_indent++;
674
675 pandecode_rt_format(rt->format);
676
677 if (rt->format.block == MALI_MFBD_BLOCK_AFBC) {
678 pandecode_log(".afbc = {\n");
679 pandecode_indent++;
680
681 char *a = pointer_as_memory_reference(rt->afbc.metadata);
682 pandecode_prop("metadata = %s", a);
683 free(a);
684
685 pandecode_prop("stride = %d", rt->afbc.stride);
686 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
687
688 pandecode_indent--;
689 pandecode_log("},\n");
690 } else {
691 pandecode_log(".chunknown = {\n");
692 pandecode_indent++;
693
694 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
695
696 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
697 pandecode_prop("pointer = %s", a);
698 free(a);
699
700 pandecode_indent--;
701 pandecode_log("},\n");
702 }
703
704 MEMORY_PROP(rt, framebuffer);
705 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
706
707 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
708 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
709 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
710 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
711 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
712 }
713
714 if (rt->zero1 || rt->zero2 || rt->zero3) {
715 pandecode_msg("render target zeros tripped\n");
716 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
717 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
718 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
719 }
720
721 pandecode_indent--;
722 pandecode_log("},\n");
723 }
724
725 pandecode_indent--;
726 pandecode_log("};\n");
727 }
728
729 static unsigned
730 pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
731 {
732 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
733 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
734
735 if (fb->sample_locations) {
736 /* The blob stores all possible sample locations in a single buffer
737 * allocated on startup, and just switches the pointer when switching
738 * MSAA state. For now, we just put the data into the cmdstream, but we
739 * should do something like what the blob does with a real driver.
740 *
741 * There seem to be 32 slots for sample locations, followed by another
742 * 16. The second 16 is just the center location followed by 15 zeros
743 * in all the cases I've identified (maybe shader vs. depth/color
744 * samples?).
745 */
746
747 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
748
749 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
750
751 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
752 pandecode_indent++;
753
754 for (int i = 0; i < 32 + 16; i++) {
755 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
756 }
757
758 pandecode_indent--;
759 pandecode_log("};\n");
760 }
761
762 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
763 pandecode_indent++;
764
765 pandecode_prop("unk0 = 0x%x", fb->unk0);
766
767 if (fb->sample_locations)
768 pandecode_prop("sample_locations = sample_locations_%d", job_no);
769
770 /* Assume that unknown1 was emitted in the last job for
771 * now */
772 MEMORY_PROP(fb, unknown1);
773
774 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
775 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
776 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
777 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
778
779 pandecode_prop("unk1 = 0x%x", fb->unk1);
780 pandecode_prop("unk2 = 0x%x", fb->unk2);
781 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
782 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
783
784 pandecode_log(".mfbd_flags = ");
785 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
786 pandecode_log_cont(",\n");
787
788 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
789 pandecode_prop("clear_depth = %f", fb->clear_depth);
790
791 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
792 MEMORY_PROP(fb, scratchpad);
793 const struct midgard_tiler_descriptor t = fb->tiler;
794 pandecode_midgard_tiler_descriptor(&t);
795
796 if (fb->zero3 || fb->zero4) {
797 pandecode_msg("framebuffer zeros tripped\n");
798 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
799 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
800 }
801
802 pandecode_indent--;
803 pandecode_log("};\n");
804
805 gpu_va += sizeof(struct bifrost_framebuffer);
806
807 if ((fb->mfbd_flags & MALI_MFBD_EXTRA) && with_render_targets) {
808 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
809 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
810
811 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
812 pandecode_indent++;
813
814 MEMORY_PROP(fbx, checksum);
815
816 if (fbx->checksum_stride)
817 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
818
819 pandecode_log(".flags = ");
820 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
821 pandecode_log_cont(",\n");
822
823 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
824 pandecode_log(".ds_afbc = {\n");
825 pandecode_indent++;
826
827 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil_afbc_metadata);
828 pandecode_prop("depth_stencil_afbc_stride = %d",
829 fbx->ds_afbc.depth_stencil_afbc_stride);
830 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil);
831
832 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
833 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
834 pandecode_prop("zero1 = 0x%" PRIx32,
835 fbx->ds_afbc.zero1);
836 pandecode_prop("padding = 0x%" PRIx64,
837 fbx->ds_afbc.padding);
838 }
839
840 pandecode_indent--;
841 pandecode_log("},\n");
842 } else {
843 pandecode_log(".ds_linear = {\n");
844 pandecode_indent++;
845
846 if (fbx->ds_linear.depth) {
847 MEMORY_PROP_DIR(fbx->ds_linear, depth);
848 pandecode_prop("depth_stride = %d",
849 fbx->ds_linear.depth_stride);
850 }
851
852 if (fbx->ds_linear.stencil) {
853 MEMORY_PROP_DIR(fbx->ds_linear, stencil);
854 pandecode_prop("stencil_stride = %d",
855 fbx->ds_linear.stencil_stride);
856 }
857
858 if (fbx->ds_linear.depth_stride_zero ||
859 fbx->ds_linear.stencil_stride_zero ||
860 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
861 pandecode_msg("Depth/stencil zeros tripped\n");
862 pandecode_prop("depth_stride_zero = 0x%x",
863 fbx->ds_linear.depth_stride_zero);
864 pandecode_prop("stencil_stride_zero = 0x%x",
865 fbx->ds_linear.stencil_stride_zero);
866 pandecode_prop("zero1 = 0x%" PRIx32,
867 fbx->ds_linear.zero1);
868 pandecode_prop("zero2 = 0x%" PRIx32,
869 fbx->ds_linear.zero2);
870 }
871
872 pandecode_indent--;
873 pandecode_log("},\n");
874 }
875
876 if (fbx->zero3 || fbx->zero4) {
877 pandecode_msg("fb_extra zeros tripped\n");
878 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
879 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
880 }
881
882 pandecode_indent--;
883 pandecode_log("};\n");
884
885 gpu_va += sizeof(struct bifrost_fb_extra);
886 }
887
888 if (with_render_targets)
889 pandecode_render_target(gpu_va, job_no, fb);
890
891 /* Passback the render target count */
892 return MALI_NEGATIVE(fb->rt_count_1);
893 }
894
895 /* Just add a comment decoding the shift/odd fields forming the padded vertices
896 * count */
897
898 static void
899 pandecode_padded_vertices(unsigned shift, unsigned k)
900 {
901 unsigned odd = 2*k + 1;
902 unsigned pot = 1 << shift;
903 pandecode_msg("padded_num_vertices = %d\n", odd * pot);
904 }
905
906 /* Given a magic divisor, recover what we were trying to divide by.
907 *
908 * Let m represent the magic divisor. By definition, m is an element on Z, whre
909 * 0 <= m < 2^N, for N bits in m.
910 *
911 * Let q represent the number we would like to divide by.
912 *
913 * By definition of a magic divisor for N-bit unsigned integers (a number you
914 * multiply by to magically get division), m is a number such that:
915 *
916 * (m * x) & (2^N - 1) = floor(x/q).
917 * for all x on Z where 0 <= x < 2^N
918 *
919 * Ignore the case where any of the above values equals zero; it is irrelevant
920 * for our purposes (instanced arrays).
921 *
922 * Choose x = q. Then:
923 *
924 * (m * x) & (2^N - 1) = floor(x/q).
925 * (m * q) & (2^N - 1) = floor(q/q).
926 *
927 * floor(q/q) = floor(1) = 1, therefore:
928 *
929 * (m * q) & (2^N - 1) = 1
930 *
931 * Recall the identity that the bitwise AND of one less than a power-of-two
932 * equals the modulo with that power of two, i.e. for all x:
933 *
934 * x & (2^N - 1) = x % N
935 *
936 * Therefore:
937 *
938 * mq % (2^N) = 1
939 *
940 * By definition, a modular multiplicative inverse of a number m is the number
941 * q such that with respect to a modulos M:
942 *
943 * mq % M = 1
944 *
945 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
946 *
947 */
948
949 static void
950 pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
951 {
952 #if 0
953 /* Compute the modular inverse of `magic` with respect to 2^(32 -
954 * shift) the most lame way possible... just repeatedly add.
955 * Asymptoptically slow but nobody cares in practice, unless you have
956 * massive numbers of vertices or high divisors. */
957
958 unsigned inverse = 0;
959
960 /* Magic implicitly has the highest bit set */
961 magic |= (1 << 31);
962
963 /* Depending on rounding direction */
964 if (extra)
965 magic++;
966
967 for (;;) {
968 uint32_t product = magic * inverse;
969
970 if (shift) {
971 product >>= shift;
972 }
973
974 if (product == 1)
975 break;
976
977 ++inverse;
978 }
979
980 pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
981
982 /* Recall we're supposed to divide by (gl_level_divisor *
983 * padded_num_vertices) */
984
985 unsigned padded_num_vertices = inverse / orig_divisor;
986
987 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
988 #endif
989 }
990
991 static void
992 pandecode_attributes(const struct pandecode_mapped_memory *mem,
993 mali_ptr addr, int job_no, char *suffix,
994 int count, bool varying)
995 {
996 char *prefix = varying ? "varyings" : "attributes";
997
998 if (!addr) {
999 pandecode_msg("no %s\n", prefix);
1000 return;
1001 }
1002
1003 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
1004
1005 char base[128];
1006 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
1007
1008 for (int i = 0; i < count; ++i) {
1009 enum mali_attr_mode mode = attr[i].elements & 7;
1010
1011 if (mode == MALI_ATTR_UNUSED)
1012 continue;
1013
1014 mali_ptr raw_elements = attr[i].elements & ~7;
1015
1016 /* TODO: Do we maybe want to dump the attribute values
1017 * themselves given the specified format? Or is that too hard?
1018 * */
1019
1020 char *a = pointer_as_memory_reference(raw_elements);
1021 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
1022 free(a);
1023 }
1024
1025 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
1026 pandecode_indent++;
1027
1028 for (int i = 0; i < count; ++i) {
1029 pandecode_log("{\n");
1030 pandecode_indent++;
1031
1032 unsigned mode = attr[i].elements & 7;
1033 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode(mode));
1034 pandecode_prop("shift = %d", attr[i].shift);
1035 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
1036 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
1037 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
1038
1039 /* Decode further where possible */
1040
1041 if (mode == MALI_ATTR_MODULO) {
1042 pandecode_padded_vertices(
1043 attr[i].shift,
1044 attr[i].extra_flags);
1045 }
1046
1047 pandecode_indent--;
1048 pandecode_log("}, \n");
1049
1050 if (mode == MALI_ATTR_NPOT_DIVIDE) {
1051 i++;
1052 pandecode_log("{\n");
1053 pandecode_indent++;
1054 pandecode_prop("unk = 0x%x", attr[i].unk);
1055 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
1056 if (attr[i].zero != 0)
1057 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
1058 pandecode_prop("divisor = %d", attr[i].divisor);
1059 pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
1060 pandecode_indent--;
1061 pandecode_log("}, \n");
1062 }
1063
1064 }
1065
1066 pandecode_indent--;
1067 pandecode_log("};\n");
1068 }
1069
1070 static mali_ptr
1071 pandecode_shader_address(const char *name, mali_ptr ptr)
1072 {
1073 /* TODO: Decode flags */
1074 mali_ptr shader_ptr = ptr & ~15;
1075
1076 char *a = pointer_as_memory_reference(shader_ptr);
1077 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
1078 free(a);
1079
1080 return shader_ptr;
1081 }
1082
1083 static bool
1084 all_zero(unsigned *buffer, unsigned count)
1085 {
1086 for (unsigned i = 0; i < count; ++i) {
1087 if (buffer[i])
1088 return false;
1089 }
1090
1091 return true;
1092 }
1093
1094 static void
1095 pandecode_stencil(const char *name, const struct mali_stencil_test *stencil)
1096 {
1097 if (all_zero((unsigned *) stencil, sizeof(stencil) / sizeof(unsigned)))
1098 return;
1099
1100 const char *func = pandecode_func(stencil->func);
1101 const char *sfail = pandecode_stencil_op(stencil->sfail);
1102 const char *dpfail = pandecode_stencil_op(stencil->dpfail);
1103 const char *dppass = pandecode_stencil_op(stencil->dppass);
1104
1105 if (stencil->zero)
1106 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
1107
1108 pandecode_log(".stencil_%s = {\n", name);
1109 pandecode_indent++;
1110 pandecode_prop("ref = %d", stencil->ref);
1111 pandecode_prop("mask = 0x%02X", stencil->mask);
1112 pandecode_prop("func = %s", func);
1113 pandecode_prop("sfail = %s", sfail);
1114 pandecode_prop("dpfail = %s", dpfail);
1115 pandecode_prop("dppass = %s", dppass);
1116 pandecode_indent--;
1117 pandecode_log("},\n");
1118 }
1119
1120 static void
1121 pandecode_blend_equation(const struct mali_blend_equation *blend)
1122 {
1123 if (blend->zero1)
1124 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
1125
1126 pandecode_log(".equation = {\n");
1127 pandecode_indent++;
1128
1129 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
1130 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
1131
1132 pandecode_log(".color_mask = ");
1133 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
1134 pandecode_log_cont(",\n");
1135
1136 pandecode_indent--;
1137 pandecode_log("},\n");
1138 }
1139
1140 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1141
1142 static unsigned
1143 decode_bifrost_constant(u16 constant)
1144 {
1145 float lo = (float) (constant & 0xFF);
1146 float hi = (float) (constant >> 8);
1147
1148 return (hi / 255.0) + (lo / 65535.0);
1149 }
1150
1151 static mali_ptr
1152 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
1153 {
1154 struct bifrost_blend_rt *b =
1155 ((struct bifrost_blend_rt *) descs) + rt_no;
1156
1157 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1158 pandecode_indent++;
1159
1160 pandecode_prop("flags = 0x%" PRIx16, b->flags);
1161 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
1162 b->constant, decode_bifrost_constant(b->constant));
1163
1164 /* TODO figure out blend shader enable bit */
1165 pandecode_blend_equation(&b->equation);
1166 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1167 pandecode_prop("index = 0x%" PRIx16, b->index);
1168 pandecode_prop("shader = 0x%" PRIx32, b->shader);
1169
1170 pandecode_indent--;
1171 pandecode_log("},\n");
1172
1173 return 0;
1174 }
1175
1176 static mali_ptr
1177 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
1178 {
1179 if (all_zero((unsigned *) blend, sizeof(blend) / sizeof(unsigned)))
1180 return 0;
1181
1182 pandecode_log(".blend = {\n");
1183 pandecode_indent++;
1184
1185 if (is_shader) {
1186 pandecode_shader_address("shader", blend->shader);
1187 } else {
1188 pandecode_blend_equation(&blend->equation);
1189 pandecode_prop("constant = %f", blend->constant);
1190 }
1191
1192 pandecode_indent--;
1193 pandecode_log("},\n");
1194
1195 /* Return blend shader to disassemble if present */
1196 return is_shader ? (blend->shader & ~0xF) : 0;
1197 }
1198
1199 static mali_ptr
1200 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1201 {
1202 struct midgard_blend_rt *b =
1203 ((struct midgard_blend_rt *) descs) + rt_no;
1204
1205 /* Flags determine presence of blend shader */
1206 bool is_shader = (b->flags & 0xF) >= 0x2;
1207
1208 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1209 pandecode_indent++;
1210
1211 pandecode_prop("flags = 0x%" PRIx64, b->flags);
1212
1213 union midgard_blend blend = b->blend;
1214 mali_ptr shader = pandecode_midgard_blend(&blend, is_shader);
1215
1216 pandecode_indent--;
1217 pandecode_log("};\n");
1218
1219 return shader;
1220 }
1221
1222 static int
1223 pandecode_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1224 {
1225 char base[128];
1226 char *prefix = varying ? "varying" : "attribute";
1227 unsigned max_index = 0;
1228 snprintf(base, sizeof(base), "%s_meta", prefix);
1229
1230 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
1231 pandecode_indent++;
1232
1233 struct mali_attr_meta *attr_meta;
1234 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
1235
1236 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
1237
1238 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
1239 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1240 sizeof(*attr_mem));
1241
1242 pandecode_log("{\n");
1243 pandecode_indent++;
1244 pandecode_prop("index = %d", attr_meta->index);
1245
1246 if (attr_meta->index > max_index)
1247 max_index = attr_meta->index;
1248 pandecode_swizzle(attr_meta->swizzle);
1249 pandecode_prop("format = %s", pandecode_format(attr_meta->format));
1250
1251 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1252 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1253 pandecode_prop("src_offset = %d", attr_meta->src_offset);
1254 pandecode_indent--;
1255 pandecode_log("},\n");
1256
1257 }
1258
1259 pandecode_indent--;
1260 pandecode_log("};\n");
1261
1262 return count ? (max_index + 1) : 0;
1263 }
1264
1265 static void
1266 pandecode_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1267 {
1268 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1269
1270 if (imem) {
1271 /* Indices are literally just a u32 array :) */
1272
1273 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1274
1275 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1276 pandecode_indent++;
1277
1278 for (unsigned i = 0; i < (index_count + 1); i += 3)
1279 pandecode_log("%d, %d, %d,\n",
1280 indices[i],
1281 indices[i + 1],
1282 indices[i + 2]);
1283
1284 pandecode_indent--;
1285 pandecode_log("};\n");
1286 }
1287 }
1288
1289 /* return bits [lo, hi) of word */
1290 static u32
1291 bits(u32 word, u32 lo, u32 hi)
1292 {
1293 if (hi - lo >= 32)
1294 return word; // avoid undefined behavior with the shift
1295
1296 return (word >> lo) & ((1 << (hi - lo)) - 1);
1297 }
1298
1299 static void
1300 pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
1301 {
1302 pandecode_log_cont("{\n");
1303 pandecode_indent++;
1304
1305 pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
1306 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1307 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1308 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1309 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1310 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1311 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
1312
1313 /* Decode invocation_count. See the comment before the definition of
1314 * invocation_count for an explanation.
1315 */
1316 pandecode_msg("size: (%d, %d, %d)\n",
1317 bits(p->invocation_count, 0, p->size_y_shift) + 1,
1318 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
1319 bits(p->invocation_count, p->size_z_shift,
1320 p->workgroups_x_shift) + 1);
1321 pandecode_msg("workgroups: (%d, %d, %d)\n",
1322 bits(p->invocation_count, p->workgroups_x_shift,
1323 p->workgroups_y_shift) + 1,
1324 bits(p->invocation_count, p->workgroups_y_shift,
1325 p->workgroups_z_shift) + 1,
1326 bits(p->invocation_count, p->workgroups_z_shift,
1327 32) + 1);
1328
1329 /* TODO: Decode */
1330 if (p->unknown_draw)
1331 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1332
1333 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1334
1335 pandecode_prop("draw_mode = %s", pandecode_draw_mode(p->draw_mode));
1336
1337 /* Index count only exists for tiler jobs anyway */
1338
1339 if (p->index_count)
1340 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1341
1342 if (p->offset_bias_correction)
1343 pandecode_prop("offset_bias_correction = %d", p->offset_bias_correction);
1344
1345 DYN_MEMORY_PROP(p, job_no, indices);
1346
1347 if (p->zero1) {
1348 pandecode_msg("Zero tripped\n");
1349 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1350 }
1351
1352 pandecode_indent--;
1353 pandecode_log("},\n");
1354 }
1355
1356 static void
1357 pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1358 {
1359 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1360
1361 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1362
1363 for (int i = 0; i < ubufs_count; i++) {
1364 mali_ptr ptr = ubufs[i].ptr << 2;
1365 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1366 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1367 char name[50];
1368 snprintf(name, sizeof(name), "ubuf_%d", i);
1369 /* The blob uses ubuf 0 to upload internal stuff and
1370 * uniforms that won't fit/are accessed indirectly, so
1371 * it puts it in the batchbuffer.
1372 */
1373 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1374 pandecode_indent++;
1375
1376 for (int j = 0; j <= ubufs[i].size; j++) {
1377 for (int k = 0; k < 4; k++) {
1378 if (k == 0)
1379 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1380 else
1381 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1382
1383 }
1384
1385 pandecode_log_cont("\n");
1386 }
1387
1388 pandecode_indent--;
1389 pandecode_log("};\n");
1390 }
1391
1392 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%"PRIx64"_%d[] = {\n",
1393 pubufs, job_no);
1394 pandecode_indent++;
1395
1396 for (int i = 0; i < ubufs_count; i++) {
1397 pandecode_log("{\n");
1398 pandecode_indent++;
1399 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1400 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1401 pandecode_indent--;
1402 pandecode_log("},\n");
1403 }
1404
1405 pandecode_indent--;
1406 pandecode_log("};\n");
1407 }
1408
1409 static void
1410 pandecode_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1411 {
1412
1413 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1414
1415 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1416
1417 if (scratchpad->zero)
1418 pandecode_msg("XXX scratchpad zero tripped");
1419
1420 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64"_%d%s = {\n", pscratchpad, job_no, suffix);
1421 pandecode_indent++;
1422
1423 pandecode_prop("flags = 0x%x", scratchpad->flags);
1424 MEMORY_PROP(scratchpad, gpu_scratchpad);
1425
1426 pandecode_indent--;
1427 pandecode_log("};\n");
1428 }
1429
1430 static void
1431 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1432 bool is_bifrost)
1433 {
1434 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1435 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1436
1437 /* Compute maximum possible size */
1438 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1439
1440 /* Print some boilerplate to clearly denote the assembly (which doesn't
1441 * obey indentation rules), and actually do the disassembly! */
1442
1443 printf("\n\n");
1444
1445 if (is_bifrost) {
1446 disassemble_bifrost(code, sz, false);
1447 } else {
1448 disassemble_midgard(code, sz);
1449 }
1450
1451 printf("\n\n");
1452 }
1453
1454 static void
1455 pandecode_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1456 int job_no, enum mali_job_type job_type,
1457 char *suffix, bool is_bifrost)
1458 {
1459 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1460 struct pandecode_mapped_memory *attr_mem;
1461
1462 unsigned rt_count = 1;
1463
1464 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1465 * are the only things actually needed from the FBD, vertex/tiler jobs
1466 * no longer reference the FBD -- instead, this field points to some
1467 * info about the scratchpad.
1468 */
1469 if (is_bifrost)
1470 pandecode_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1471 else if (p->framebuffer & MALI_MFBD)
1472 rt_count = pandecode_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1473 else if (job_type == JOB_TYPE_COMPUTE)
1474 pandecode_compute_fbd((u64) (uintptr_t) p->framebuffer, job_no);
1475 else
1476 pandecode_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1477
1478 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1479 int texture_count = 0, sampler_count = 0;
1480
1481 if (shader_meta_ptr) {
1482 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1483 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1484
1485 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", shader_meta_ptr, job_no, suffix);
1486 pandecode_indent++;
1487
1488 /* Save for dumps */
1489 attribute_count = s->attribute_count;
1490 varying_count = s->varying_count;
1491 texture_count = s->texture_count;
1492 sampler_count = s->sampler_count;
1493
1494 if (is_bifrost) {
1495 uniform_count = s->bifrost2.uniform_count;
1496 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1497 } else {
1498 uniform_count = s->midgard1.uniform_count;
1499 uniform_buffer_count = s->midgard1.uniform_buffer_count;
1500 }
1501
1502 mali_ptr shader_ptr = pandecode_shader_address("shader", s->shader);
1503
1504 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1505 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1506 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1507 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1508
1509 if (is_bifrost) {
1510 pandecode_log(".bifrost1 = {\n");
1511 pandecode_indent++;
1512
1513 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1514 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1515
1516 pandecode_indent--;
1517 pandecode_log("},\n");
1518 } else {
1519 pandecode_log(".midgard1 = {\n");
1520 pandecode_indent++;
1521
1522 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1523 pandecode_prop("uniform_buffer_count = %" PRId16, s->midgard1.uniform_buffer_count);
1524 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1525
1526 pandecode_log(".flags = ");
1527 pandecode_log_decoded_flags(shader_midgard1_flag_info, s->midgard1.flags);
1528 pandecode_log_cont(",\n");
1529
1530 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1531
1532 pandecode_indent--;
1533 pandecode_log("},\n");
1534 }
1535
1536 if (s->depth_units || s->depth_factor) {
1537 pandecode_prop("depth_factor = %f", s->depth_factor);
1538 pandecode_prop("depth_units = %f", s->depth_units);
1539 }
1540
1541 if (s->alpha_coverage) {
1542 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1543 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1544
1545 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1546 invert_alpha_coverage ? "~" : "",
1547 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1548 }
1549
1550 if (s->unknown2_3 || s->unknown2_4) {
1551 pandecode_log(".unknown2_3 = ");
1552
1553 int unknown2_3 = s->unknown2_3;
1554 int unknown2_4 = s->unknown2_4;
1555
1556 /* We're not quite sure what these flags mean without the depth test, if anything */
1557
1558 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1559 const char *func = pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3));
1560 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1561
1562 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1563 }
1564
1565 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1566 pandecode_log_cont(",\n");
1567
1568 pandecode_log(".unknown2_4 = ");
1569 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1570 pandecode_log_cont(",\n");
1571 }
1572
1573 if (s->stencil_mask_front || s->stencil_mask_back) {
1574 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1575 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1576 }
1577
1578 pandecode_stencil("front", &s->stencil_front);
1579 pandecode_stencil("back", &s->stencil_back);
1580
1581 if (is_bifrost) {
1582 pandecode_log(".bifrost2 = {\n");
1583 pandecode_indent++;
1584
1585 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1586 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1587 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1588 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1589
1590 pandecode_indent--;
1591 pandecode_log("},\n");
1592 } else if (s->midgard2.unknown2_7) {
1593 pandecode_log(".midgard2 = {\n");
1594 pandecode_indent++;
1595
1596 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1597 pandecode_indent--;
1598 pandecode_log("},\n");
1599 }
1600
1601 if (s->unknown2_8)
1602 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1603
1604 if (!is_bifrost) {
1605 /* TODO: Blend shaders routing/disasm */
1606
1607 union midgard_blend blend = s->blend;
1608 pandecode_midgard_blend(&blend, false);
1609 }
1610
1611 pandecode_indent--;
1612 pandecode_log("};\n");
1613
1614 /* MRT blend fields are used whenever MFBD is used, with
1615 * per-RT descriptors */
1616
1617 if (job_type == JOB_TYPE_TILER) {
1618 void* blend_base = (void *) (s + 1);
1619
1620 for (unsigned i = 0; i < rt_count; i++) {
1621 mali_ptr shader = 0;
1622
1623 if (is_bifrost)
1624 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1625 else
1626 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1627
1628 if (shader & ~0xF)
1629 pandecode_shader_disassemble(shader, job_no, job_type, false);
1630 }
1631 }
1632
1633 if (shader_ptr & ~0xF)
1634 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost);
1635 } else
1636 pandecode_msg("<no shader>\n");
1637
1638 if (p->viewport) {
1639 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1640 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1641
1642 pandecode_log("struct mali_viewport viewport_%"PRIx64"_%d%s = {\n", p->viewport, job_no, suffix);
1643 pandecode_indent++;
1644
1645 pandecode_prop("clip_minx = %f", f->clip_minx);
1646 pandecode_prop("clip_miny = %f", f->clip_miny);
1647 pandecode_prop("clip_minz = %f", f->clip_minz);
1648 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1649 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1650 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1651
1652 /* Only the higher coordinates are MALI_POSITIVE scaled */
1653
1654 pandecode_prop("viewport0 = { %d, %d }",
1655 f->viewport0[0], f->viewport0[1]);
1656
1657 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1658 f->viewport1[0] + 1, f->viewport1[1] + 1);
1659
1660 pandecode_indent--;
1661 pandecode_log("};\n");
1662 }
1663
1664 if (p->attribute_meta) {
1665 unsigned max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
1666
1667 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1668 pandecode_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1669 }
1670
1671 /* Varyings are encoded like attributes but not actually sent; we just
1672 * pass a zero buffer with the right stride/size set, (or whatever)
1673 * since the GPU will write to it itself */
1674
1675 if (p->varying_meta) {
1676 varying_count = pandecode_attribute_meta(job_no, varying_count, p, true, suffix);
1677 }
1678
1679 if (p->varyings) {
1680 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1681
1682 /* Number of descriptors depends on whether there are
1683 * non-internal varyings */
1684
1685 pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true);
1686 }
1687
1688 bool is_compute = job_type == JOB_TYPE_COMPUTE;
1689
1690 if (p->uniforms && !is_compute) {
1691 int rows = uniform_count, width = 4;
1692 size_t sz = rows * width * sizeof(float);
1693
1694 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1695 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1696 u32 *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1697
1698 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no, suffix);
1699
1700 pandecode_indent++;
1701
1702 for (int row = 0; row < rows; row++) {
1703 for (int i = 0; i < width; i++) {
1704 u32 v = uniforms[i];
1705 float f;
1706 memcpy(&f, &v, sizeof(v));
1707 pandecode_log_cont("%X /* %f */, ", v, f);
1708 }
1709
1710 pandecode_log_cont("\n");
1711
1712 uniforms += width;
1713 }
1714
1715 pandecode_indent--;
1716 pandecode_log("};\n");
1717 } else if (p->uniforms) {
1718 int rows = uniform_count * 2;
1719 size_t sz = rows * sizeof(mali_ptr);
1720
1721 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1722 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1723 mali_ptr *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1724
1725 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no, suffix);
1726
1727 pandecode_indent++;
1728
1729 for (int row = 0; row < rows; row++) {
1730 char *a = pointer_as_memory_reference(uniforms[row]);
1731 pandecode_log("%s,\n", a);
1732 free(a);
1733 }
1734
1735 pandecode_indent--;
1736 pandecode_log("};\n");
1737
1738 }
1739
1740 if (p->uniform_buffers) {
1741 pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1742 }
1743
1744 if (p->texture_trampoline) {
1745 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1746
1747 if (mmem) {
1748 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1749
1750 pandecode_log("uint64_t texture_trampoline_%"PRIx64"_%d[] = {\n", p->texture_trampoline, job_no);
1751 pandecode_indent++;
1752
1753 for (int tex = 0; tex < texture_count; ++tex) {
1754 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1755 char *a = pointer_as_memory_reference(*u);
1756 pandecode_log("%s,\n", a);
1757 free(a);
1758 }
1759
1760 pandecode_indent--;
1761 pandecode_log("};\n");
1762
1763 /* Now, finally, descend down into the texture descriptor */
1764 for (int tex = 0; tex < texture_count; ++tex) {
1765 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1766 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1767
1768 if (tmem) {
1769 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1770
1771 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", *u, job_no, tex);
1772 pandecode_indent++;
1773
1774 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1775 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1776 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1777 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16 ")", t->array_size + 1);
1778 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1779 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1780 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1781
1782 struct mali_texture_format f = t->format;
1783
1784 pandecode_log(".format = {\n");
1785 pandecode_indent++;
1786
1787 pandecode_swizzle(f.swizzle);
1788 pandecode_prop("format = %s", pandecode_format(f.format));
1789 pandecode_prop("type = %s", pandecode_texture_type(f.type));
1790 pandecode_prop("srgb = %" PRId32, f.srgb);
1791 pandecode_prop("unknown1 = %" PRId32, f.unknown1);
1792 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1793
1794 pandecode_indent--;
1795 pandecode_log("},\n");
1796
1797 pandecode_swizzle(t->swizzle);
1798
1799 if (t->swizzle_zero) {
1800 /* Shouldn't happen */
1801 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1802 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1803 }
1804
1805 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1806
1807 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1808 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1809 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1810
1811 pandecode_log(".payload = {\n");
1812 pandecode_indent++;
1813
1814 /* A bunch of bitmap pointers follow.
1815 * We work out the correct number,
1816 * based on the mipmap/cubemap
1817 * properties, but dump extra
1818 * possibilities to futureproof */
1819
1820 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1821 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1822
1823 /* Miptree for each face */
1824 if (f.type == MALI_TEX_CUBE)
1825 bitmap_count *= 6;
1826
1827 /* Array of textures */
1828 bitmap_count *= MALI_NEGATIVE(t->array_size);
1829
1830 /* Stride for each element */
1831 if (manual_stride)
1832 bitmap_count *= 2;
1833
1834 /* Sanity check the size */
1835 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1836 assert (bitmap_count <= max_count);
1837
1838 /* Dump more to be safe, but not _that_ much more */
1839 int safe_count = MIN2(bitmap_count * 2, max_count);
1840
1841 for (int i = 0; i < safe_count; ++i) {
1842 char *prefix = (i >= bitmap_count) ? "// " : "";
1843
1844 /* How we dump depends if this is a stride or a pointer */
1845
1846 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1847 /* signed 32-bit snuck in as a 64-bit pointer */
1848 uint64_t stride_set = t->payload[i];
1849 uint32_t clamped_stride = stride_set;
1850 int32_t stride = clamped_stride;
1851 assert(stride_set == clamped_stride);
1852 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1853 } else {
1854 char *a = pointer_as_memory_reference(t->payload[i]);
1855 pandecode_log("%s%s, \n", prefix, a);
1856 free(a);
1857 }
1858 }
1859
1860 pandecode_indent--;
1861 pandecode_log("},\n");
1862
1863 pandecode_indent--;
1864 pandecode_log("};\n");
1865 }
1866 }
1867 }
1868 }
1869
1870 if (p->sampler_descriptor) {
1871 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1872
1873 if (smem) {
1874 struct mali_sampler_descriptor *s;
1875
1876 mali_ptr d = p->sampler_descriptor;
1877
1878 for (int i = 0; i < sampler_count; ++i) {
1879 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1880
1881 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", d + sizeof(*s) * i, job_no, i);
1882 pandecode_indent++;
1883
1884 pandecode_log(".filter_mode = ");
1885 pandecode_log_decoded_flags(sampler_flag_info, s->filter_mode);
1886 pandecode_log_cont(",\n");
1887
1888 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1889 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1890
1891 pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
1892 pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
1893 pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
1894
1895 pandecode_prop("compare_func = %s", pandecode_alt_func(s->compare_func));
1896
1897 if (s->zero || s->zero2) {
1898 pandecode_msg("Zero tripped\n");
1899 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1900 }
1901
1902 pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
1903
1904 pandecode_prop("border_color = { %f, %f, %f, %f }",
1905 s->border_color[0],
1906 s->border_color[1],
1907 s->border_color[2],
1908 s->border_color[3]);
1909
1910 pandecode_indent--;
1911 pandecode_log("};\n");
1912 }
1913 }
1914 }
1915 }
1916
1917 static void
1918 pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1919 {
1920 pandecode_log_cont("{\n");
1921 pandecode_indent++;
1922
1923 MEMORY_PROP(p, position_varying);
1924 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1925 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1926 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1927 DYN_MEMORY_PROP(p, job_no, uniforms);
1928 DYN_MEMORY_PROP(p, job_no, attributes);
1929 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1930 DYN_MEMORY_PROP(p, job_no, varyings);
1931 DYN_MEMORY_PROP(p, job_no, varying_meta);
1932 DYN_MEMORY_PROP(p, job_no, viewport);
1933 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1934
1935 if (is_bifrost)
1936 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1937 else
1938 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1939
1940 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1941 pandecode_prop("flags = %d", p->flags);
1942
1943 pandecode_indent--;
1944 pandecode_log("},\n");
1945 }
1946
1947 static void
1948 pandecode_vertex_only_bfr(struct bifrost_vertex_only *v)
1949 {
1950 pandecode_log_cont("{\n");
1951 pandecode_indent++;
1952
1953 pandecode_prop("unk2 = 0x%x", v->unk2);
1954
1955 if (v->zero0 || v->zero1) {
1956 pandecode_msg("vertex only zero tripped");
1957 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1958 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1959 }
1960
1961 pandecode_indent--;
1962 pandecode_log("}\n");
1963 }
1964
1965 static void
1966 pandecode_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1967 {
1968
1969 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1970 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1971
1972 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1973 pandecode_indent++;
1974
1975 if (h->zero) {
1976 pandecode_msg("tiler heap zero tripped\n");
1977 pandecode_prop("zero = 0x%x", h->zero);
1978 }
1979
1980 for (int i = 0; i < 12; i++) {
1981 if (h->zeros[i] != 0) {
1982 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1983 i, h->zeros[i]);
1984 }
1985 }
1986
1987 pandecode_prop("heap_size = 0x%x", h->heap_size);
1988 MEMORY_PROP(h, tiler_heap_start);
1989 MEMORY_PROP(h, tiler_heap_free);
1990
1991 /* this might point to the beginning of another buffer, when it's
1992 * really the end of the tiler heap buffer, so we have to be careful
1993 * here. but for zero length, we need the same pointer.
1994 */
1995
1996 if (h->tiler_heap_end == h->tiler_heap_start) {
1997 MEMORY_PROP(h, tiler_heap_start);
1998 } else {
1999 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
2000 pandecode_prop("tiler_heap_end = %s + 1", a);
2001 free(a);
2002 }
2003
2004 pandecode_indent--;
2005 pandecode_log("};\n");
2006 }
2007
2008 static void
2009 pandecode_tiler_meta(mali_ptr gpu_va, int job_no)
2010 {
2011 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
2012 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
2013
2014 pandecode_tiler_heap_meta(t->tiler_heap_meta, job_no);
2015
2016 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
2017 pandecode_indent++;
2018
2019 if (t->zero0 || t->zero1) {
2020 pandecode_msg("tiler meta zero tripped");
2021 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
2022 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2023 }
2024
2025 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
2026 pandecode_prop("flags = 0x%" PRIx16, t->flags);
2027
2028 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
2029 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
2030 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
2031
2032 for (int i = 0; i < 12; i++) {
2033 if (t->zeros[i] != 0) {
2034 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
2035 i, t->zeros[i]);
2036 }
2037 }
2038
2039 pandecode_indent--;
2040 pandecode_log("};\n");
2041 }
2042
2043 static void
2044 pandecode_gl_enables(uint32_t gl_enables, int job_type)
2045 {
2046 pandecode_log(".gl_enables = ");
2047
2048 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
2049
2050 pandecode_log_cont(",\n");
2051 }
2052
2053 static void
2054 pandecode_primitive_size(union midgard_primitive_size u, bool constant)
2055 {
2056 if (u.pointer == 0x0)
2057 return;
2058
2059 pandecode_log(".primitive_size = {\n");
2060 pandecode_indent++;
2061
2062 if (constant) {
2063 pandecode_prop("constant = %f", u.constant);
2064 } else {
2065 MEMORY_PROP((&u), pointer);
2066 }
2067
2068 pandecode_indent--;
2069 pandecode_log("},\n");
2070 }
2071
2072 static void
2073 pandecode_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
2074 {
2075 pandecode_log_cont("{\n");
2076 pandecode_indent++;
2077
2078 /* TODO: gl_PointSize on Bifrost */
2079 pandecode_primitive_size(t->primitive_size, true);
2080
2081 DYN_MEMORY_PROP(t, job_no, tiler_meta);
2082 pandecode_gl_enables(t->gl_enables, JOB_TYPE_TILER);
2083
2084 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
2085 || t->zero6 || t->zero7 || t->zero8) {
2086 pandecode_msg("tiler only zero tripped");
2087 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2088 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
2089 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
2090 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
2091 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
2092 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
2093 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
2094 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
2095 }
2096
2097 pandecode_indent--;
2098 pandecode_log("},\n");
2099 }
2100
2101 static int
2102 pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
2103 const struct pandecode_mapped_memory *mem,
2104 mali_ptr payload, int job_no)
2105 {
2106 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
2107
2108 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
2109
2110 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
2111 pandecode_indent++;
2112
2113 pandecode_log(".prefix = ");
2114 pandecode_vertex_tiler_prefix(&v->prefix, job_no);
2115
2116 pandecode_log(".vertex = ");
2117 pandecode_vertex_only_bfr(&v->vertex);
2118
2119 pandecode_log(".postfix = ");
2120 pandecode_vertex_tiler_postfix(&v->postfix, job_no, true);
2121
2122 pandecode_indent--;
2123 pandecode_log("};\n");
2124
2125 return sizeof(*v);
2126 }
2127
2128 static int
2129 pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
2130 const struct pandecode_mapped_memory *mem,
2131 mali_ptr payload, int job_no)
2132 {
2133 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
2134
2135 pandecode_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
2136
2137 pandecode_indices(t->prefix.indices, t->prefix.index_count, job_no);
2138 pandecode_tiler_meta(t->tiler.tiler_meta, job_no);
2139
2140 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
2141 pandecode_indent++;
2142
2143 pandecode_log(".prefix = ");
2144 pandecode_vertex_tiler_prefix(&t->prefix, job_no);
2145
2146 pandecode_log(".tiler = ");
2147 pandecode_tiler_only_bfr(&t->tiler, job_no);
2148
2149 pandecode_log(".postfix = ");
2150 pandecode_vertex_tiler_postfix(&t->postfix, job_no, true);
2151
2152 pandecode_indent--;
2153 pandecode_log("};\n");
2154
2155 return sizeof(*t);
2156 }
2157
2158 static int
2159 pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
2160 const struct pandecode_mapped_memory *mem,
2161 mali_ptr payload, int job_no)
2162 {
2163 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
2164
2165 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
2166
2167 pandecode_indices(v->prefix.indices, v->prefix.index_count, job_no);
2168
2169 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
2170 pandecode_indent++;
2171
2172 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
2173 pandecode_primitive_size(v->primitive_size, !has_primitive_pointer);
2174
2175 pandecode_log(".prefix = ");
2176 pandecode_vertex_tiler_prefix(&v->prefix, job_no);
2177
2178 pandecode_gl_enables(v->gl_enables, h->job_type);
2179
2180 if (v->instance_shift || v->instance_odd) {
2181 pandecode_prop("instance_shift = 0x%d /* %d */",
2182 v->instance_shift, 1 << v->instance_shift);
2183 pandecode_prop("instance_odd = 0x%X /* %d */",
2184 v->instance_odd, (2 * v->instance_odd) + 1);
2185
2186 pandecode_padded_vertices(v->instance_shift, v->instance_odd);
2187 }
2188
2189 if (v->offset_start)
2190 pandecode_prop("offset_start = %d", v->offset_start);
2191
2192 if (v->zero5) {
2193 pandecode_msg("Zero tripped\n");
2194 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
2195 }
2196
2197 pandecode_log(".postfix = ");
2198 pandecode_vertex_tiler_postfix(&v->postfix, job_no, false);
2199
2200 pandecode_indent--;
2201 pandecode_log("};\n");
2202
2203 return sizeof(*v);
2204 }
2205
2206 static int
2207 pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
2208 mali_ptr payload, int job_no,
2209 bool is_bifrost)
2210 {
2211 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
2212
2213 bool fbd_dumped = false;
2214
2215 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
2216 /* Only SFBDs are understood, not MFBDs. We're speculating,
2217 * based on the versioning, kernel code, etc, that the
2218 * difference is between Single FrameBuffer Descriptor and
2219 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2220 * up with multi-framebuffer support being added (T7xx onwards,
2221 * including Gxx). In any event, there's some field shuffling
2222 * that we haven't looked into yet. */
2223
2224 pandecode_sfbd(s->framebuffer & FBD_MASK, job_no);
2225 fbd_dumped = true;
2226 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
2227 /* We don't know if Bifrost supports SFBD's at all, since the
2228 * driver never uses them. And the format is different from
2229 * Midgard anyways, due to the tiler heap and scratchpad being
2230 * moved out into separate structures, so it's not clear what a
2231 * Bifrost SFBD would even look like without getting an actual
2232 * trace, which appears impossible.
2233 */
2234
2235 pandecode_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
2236 fbd_dumped = true;
2237 }
2238
2239 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
2240 pandecode_log("struct mali_payload_fragment payload_%"PRIx64"_%d = {\n", payload, job_no);
2241 pandecode_indent++;
2242
2243 /* See the comments by the macro definitions for mathematical context
2244 * on why this is so weird */
2245
2246 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
2247 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2248
2249 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2250 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
2251 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
2252
2253 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2254 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
2255 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
2256
2257 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2258 * we have to fallback on offsets. */
2259
2260 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
2261
2262 if (fbd_dumped)
2263 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
2264 else
2265 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
2266
2267 pandecode_indent--;
2268 pandecode_log("};\n");
2269
2270 return sizeof(*s);
2271 }
2272
2273 static int job_descriptor_number = 0;
2274
2275 int
2276 pandecode_jc(mali_ptr jc_gpu_va, bool bifrost)
2277 {
2278 struct mali_job_descriptor_header *h;
2279
2280 int start_number = 0;
2281
2282 bool first = true;
2283 bool last_size;
2284
2285 do {
2286 struct pandecode_mapped_memory *mem =
2287 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2288
2289 void *payload;
2290
2291 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2292
2293 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2294 * high 32-bits of the 64-bit pointer are reused to store
2295 * something else.
2296 */
2297 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2298 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2299 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2300
2301 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2302 MALI_PAYLOAD_SIZE);
2303
2304 int job_no = job_descriptor_number++;
2305
2306 if (first)
2307 start_number = job_no;
2308
2309 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
2310 pandecode_indent++;
2311
2312 pandecode_prop("job_type = %s", pandecode_job_type(h->job_type));
2313
2314 /* Save for next job fixing */
2315 last_size = h->job_descriptor_size;
2316
2317 if (h->job_descriptor_size)
2318 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2319
2320 if (h->exception_status != 0x1)
2321 pandecode_prop("exception_status = %x (source ID: 0x%x access: 0x%x exception: 0x%x)",
2322 h->exception_status,
2323 (h->exception_status >> 16) & 0xFFFF,
2324 (h->exception_status >> 8) & 0x3,
2325 h->exception_status & 0xFF);
2326
2327 if (h->first_incomplete_task)
2328 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2329
2330 if (h->fault_pointer)
2331 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2332
2333 if (h->job_barrier)
2334 pandecode_prop("job_barrier = %d", h->job_barrier);
2335
2336 pandecode_prop("job_index = %d", h->job_index);
2337
2338 if (h->unknown_flags)
2339 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2340
2341 if (h->job_dependency_index_1)
2342 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2343
2344 if (h->job_dependency_index_2)
2345 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2346
2347 pandecode_indent--;
2348 pandecode_log("};\n");
2349
2350 /* Do not touch the field yet -- decode the payload first, and
2351 * don't touch that either. This is essential for the uploads
2352 * to occur in sequence and therefore be dynamically allocated
2353 * correctly. Do note the size, however, for that related
2354 * reason. */
2355
2356 switch (h->job_type) {
2357 case JOB_TYPE_SET_VALUE: {
2358 struct mali_payload_set_value *s = payload;
2359 pandecode_log("struct mali_payload_set_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
2360 pandecode_indent++;
2361 MEMORY_PROP(s, out);
2362 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2363 pandecode_indent--;
2364 pandecode_log("};\n");
2365
2366 break;
2367 }
2368
2369 case JOB_TYPE_TILER:
2370 case JOB_TYPE_VERTEX:
2371 case JOB_TYPE_COMPUTE:
2372 if (bifrost) {
2373 if (h->job_type == JOB_TYPE_TILER)
2374 pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no);
2375 else
2376 pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no);
2377 } else
2378 pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2379
2380 break;
2381
2382 case JOB_TYPE_FRAGMENT:
2383 pandecode_fragment_job(mem, payload_ptr, job_no, bifrost);
2384 break;
2385
2386 default:
2387 break;
2388 }
2389
2390 /* Handle linkage */
2391
2392 if (!first) {
2393 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2394
2395 if (last_size)
2396 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2397 else
2398 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2399 }
2400
2401 first = false;
2402
2403 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2404
2405 return start_number;
2406 }