projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
panfrost: Remove midgard1 bitfield
[mesa.git] / src / panfrost / lib / 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 <midgard_pack.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <memory.h>
30 #include <stdbool.h>
31 #include <stdarg.h>
32 #include <ctype.h>
33 #include "decode.h"
34 #include "util/macros.h"
35 #include "util/u_math.h"
36
37 #include "midgard/disassemble.h"
38 #include "bifrost/disassemble.h"
39
40 #include "pan_encoder.h"
41
42 static void pandecode_swizzle(unsigned swizzle, enum mali_format format);
43
44 #define MEMORY_PROP(obj, p) {\
45 if (obj->p) { \
46 char *a = pointer_as_memory_reference(obj->p); \
47 pandecode_prop("%s = %s", #p, a); \
48 free(a); \
49 } \
50 }
51
52 #define MEMORY_PROP_DIR(obj, p) {\
53 if (obj.p) { \
54 char *a = pointer_as_memory_reference(obj.p); \
55 pandecode_prop("%s = %s", #p, a); \
56 free(a); \
57 } \
58 }
59
60 #define DUMP_CL(title, T, cl, indent) {\
61 fprintf(pandecode_dump_stream, "%s\n", title); \
62 struct MALI_ ## T temp; \
63 MALI_ ## T ## _unpack((const uint8_t *) cl, &temp); \
64 MALI_ ## T ## _print(pandecode_dump_stream, &temp, indent * 2); \
65 }
66
67 #define MAP_ADDR(T, addr, cl) \
68 const uint8_t *cl = 0; \
69 { \
70 struct pandecode_mapped_memory *mapped_mem = pandecode_find_mapped_gpu_mem_containing(addr); \
71 cl = pandecode_fetch_gpu_mem(mapped_mem, addr, MALI_ ## T ## _LENGTH); \
72 }
73
74 #define DUMP_ADDR(title, T, addr, indent) {\
75 MAP_ADDR(T, addr, cl) \
76 DUMP_CL(title, T, cl, indent); \
77 }
78
79 FILE *pandecode_dump_stream;
80
81 /* Semantic logging type.
82 *
83 * Raw: for raw messages to be printed as is.
84 * Message: for helpful information to be commented out in replays.
85 * Property: for properties of a struct
86 *
87 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
88 */
89
90 enum pandecode_log_type {
91 PANDECODE_RAW,
92 PANDECODE_MESSAGE,
93 PANDECODE_PROPERTY
94 };
95
96 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
97 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
98 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
99
100 unsigned pandecode_indent = 0;
101
102 static void
103 pandecode_make_indent(void)
104 {
105 for (unsigned i = 0; i < pandecode_indent; ++i)
106 fprintf(pandecode_dump_stream, " ");
107 }
108
109 static void
110 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
111 {
112 va_list ap;
113
114 pandecode_make_indent();
115
116 if (type == PANDECODE_MESSAGE)
117 fprintf(pandecode_dump_stream, "// ");
118 else if (type == PANDECODE_PROPERTY)
119 fprintf(pandecode_dump_stream, ".");
120
121 va_start(ap, format);
122 vfprintf(pandecode_dump_stream, format, ap);
123 va_end(ap);
124
125 if (type == PANDECODE_PROPERTY)
126 fprintf(pandecode_dump_stream, ",\n");
127 }
128
129 static void
130 pandecode_log_cont(const char *format, ...)
131 {
132 va_list ap;
133
134 va_start(ap, format);
135 vfprintf(pandecode_dump_stream, format, ap);
136 va_end(ap);
137 }
138
139 /* To check for memory safety issues, validates that the given pointer in GPU
140 * memory is valid, containing at least sz bytes. The goal is to eliminate
141 * GPU-side memory bugs (NULL pointer dereferences, buffer overflows, or buffer
142 * overruns) by statically validating pointers.
143 */
144
145 static void
146 pandecode_validate_buffer(mali_ptr addr, size_t sz)
147 {
148 if (!addr) {
149 pandecode_msg("XXX: null pointer deref");
150 return;
151 }
152
153 /* Find a BO */
154
155 struct pandecode_mapped_memory *bo =
156 pandecode_find_mapped_gpu_mem_containing(addr);
157
158 if (!bo) {
159 pandecode_msg("XXX: invalid memory dereference\n");
160 return;
161 }
162
163 /* Bounds check */
164
165 unsigned offset = addr - bo->gpu_va;
166 unsigned total = offset + sz;
167
168 if (total > bo->length) {
169 pandecode_msg("XXX: buffer overrun. "
170 "Chunk of size %zu at offset %d in buffer of size %zu. "
171 "Overrun by %zu bytes. \n",
172 sz, offset, bo->length, total - bo->length);
173 return;
174 }
175 }
176
177 struct pandecode_flag_info {
178 u64 flag;
179 const char *name;
180 };
181
182 static void
183 pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
184 u64 flags)
185 {
186 bool decodable_flags_found = false;
187
188 for (int i = 0; flag_info[i].name; i++) {
189 if ((flags & flag_info[i].flag) != flag_info[i].flag)
190 continue;
191
192 if (!decodable_flags_found) {
193 decodable_flags_found = true;
194 } else {
195 pandecode_log_cont(" | ");
196 }
197
198 pandecode_log_cont("%s", flag_info[i].name);
199
200 flags &= ~flag_info[i].flag;
201 }
202
203 if (decodable_flags_found) {
204 if (flags)
205 pandecode_log_cont(" | 0x%" PRIx64, flags);
206 } else {
207 pandecode_log_cont("0x%" PRIx64, flags);
208 }
209 }
210
211 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
212 static const struct pandecode_flag_info gl_enable_flag_info[] = {
213 FLAG_INFO(OCCLUSION_QUERY),
214 FLAG_INFO(OCCLUSION_PRECISE),
215 FLAG_INFO(FRONT_CCW_TOP),
216 FLAG_INFO(CULL_FACE_FRONT),
217 FLAG_INFO(CULL_FACE_BACK),
218 {}
219 };
220 #undef FLAG_INFO
221
222 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
223 static const struct pandecode_flag_info clear_flag_info[] = {
224 FLAG_INFO(FAST),
225 FLAG_INFO(SLOW),
226 FLAG_INFO(SLOW_STENCIL),
227 {}
228 };
229 #undef FLAG_INFO
230
231 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
232 static const struct pandecode_flag_info u3_flag_info[] = {
233 FLAG_INFO(HAS_MSAA),
234 FLAG_INFO(PER_SAMPLE),
235 FLAG_INFO(CAN_DISCARD),
236 FLAG_INFO(HAS_BLEND_SHADER),
237 FLAG_INFO(DEPTH_WRITEMASK),
238 FLAG_INFO(DEPTH_CLIP_NEAR),
239 FLAG_INFO(DEPTH_CLIP_FAR),
240 {}
241 };
242
243 static const struct pandecode_flag_info u4_flag_info[] = {
244 FLAG_INFO(NO_MSAA),
245 FLAG_INFO(NO_DITHER),
246 FLAG_INFO(DEPTH_RANGE_A),
247 FLAG_INFO(DEPTH_RANGE_B),
248 FLAG_INFO(STENCIL_TEST),
249 FLAG_INFO(ALPHA_TO_COVERAGE),
250 {}
251 };
252 #undef FLAG_INFO
253
254 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
255 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
256 FLAG_INFO(SRGB),
257 {}
258 };
259 #undef FLAG_INFO
260
261 #define FLAG_INFO(flag) { MALI_AFBC_##flag, "MALI_AFBC_" #flag }
262 static const struct pandecode_flag_info afbc_fmt_flag_info[] = {
263 FLAG_INFO(YTR),
264 {}
265 };
266 #undef FLAG_INFO
267
268 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
269 static const struct pandecode_flag_info mfbd_extra_flag_hi_info[] = {
270 FLAG_INFO(PRESENT),
271 {}
272 };
273 #undef FLAG_INFO
274
275 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
276 static const struct pandecode_flag_info mfbd_extra_flag_lo_info[] = {
277 FLAG_INFO(ZS),
278 {}
279 };
280 #undef FLAG_INFO
281
282 #define FLAG_INFO(flag) { MALI_BIFROST_##flag, "MALI_BIFROST_" #flag }
283 static const struct pandecode_flag_info shader_bifrost_info [] = {
284 FLAG_INFO(FULL_THREAD),
285 FLAG_INFO(EARLY_Z),
286 FLAG_INFO(FIRST_ATEST),
287 {}
288 };
289
290 #undef FLAG_INFO
291
292 #define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
293 static const struct pandecode_flag_info mfbd_flag_info [] = {
294 FLAG_INFO(DEPTH_WRITE),
295 FLAG_INFO(EXTRA),
296 {}
297 };
298 #undef FLAG_INFO
299
300 #define FLAG_INFO(flag) { MALI_SFBD_FORMAT_##flag, "MALI_SFBD_FORMAT_" #flag }
301 static const struct pandecode_flag_info sfbd_unk1_info [] = {
302 FLAG_INFO(MSAA_8),
303 FLAG_INFO(MSAA_A),
304 {}
305 };
306 #undef FLAG_INFO
307
308 #define FLAG_INFO(flag) { MALI_SFBD_FORMAT_##flag, "MALI_SFBD_FORMAT_" #flag }
309 static const struct pandecode_flag_info sfbd_unk2_info [] = {
310 FLAG_INFO(MSAA_B),
311 FLAG_INFO(SRGB),
312 {}
313 };
314 #undef FLAG_INFO
315
316 /* Midgard's tiler descriptor is embedded within the
317 * larger FBD */
318
319 static void
320 pandecode_midgard_tiler_descriptor(
321 const struct midgard_tiler_descriptor *t,
322 unsigned width,
323 unsigned height,
324 bool is_fragment,
325 bool has_hierarchy)
326 {
327 pandecode_log(".tiler = {\n");
328 pandecode_indent++;
329
330 if (t->hierarchy_mask == MALI_TILER_DISABLED)
331 pandecode_prop("hierarchy_mask = MALI_TILER_DISABLED");
332 else
333 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
334
335 /* We know this name from the kernel, but we never see it nonzero */
336
337 if (t->flags)
338 pandecode_msg("XXX: unexpected tiler flags 0x%" PRIx16, t->flags);
339
340 MEMORY_PROP(t, polygon_list);
341
342 /* The body is offset from the base of the polygon list */
343 //assert(t->polygon_list_body > t->polygon_list);
344 unsigned body_offset = t->polygon_list_body - t->polygon_list;
345
346 /* It needs to fit inside the reported size */
347 //assert(t->polygon_list_size >= body_offset);
348
349 /* Now that we've sanity checked, we'll try to calculate the sizes
350 * ourselves for comparison */
351
352 unsigned ref_header = panfrost_tiler_header_size(width, height, t->hierarchy_mask, has_hierarchy);
353 unsigned ref_size = panfrost_tiler_full_size(width, height, t->hierarchy_mask, has_hierarchy);
354
355 if (!((ref_header == body_offset) && (ref_size == t->polygon_list_size))) {
356 pandecode_msg("XXX: bad polygon list size (expected %d / 0x%x)\n",
357 ref_header, ref_size);
358 pandecode_prop("polygon_list_size = 0x%x", t->polygon_list_size);
359 pandecode_msg("body offset %d\n", body_offset);
360 }
361
362 /* The tiler heap has a start and end specified -- it should be
363 * identical to what we have in the BO. The exception is if tiling is
364 * disabled. */
365
366 MEMORY_PROP(t, heap_start);
367 assert(t->heap_end >= t->heap_start);
368
369 unsigned heap_size = t->heap_end - t->heap_start;
370
371 /* Tiling is enabled with a special flag */
372 unsigned hierarchy_mask = t->hierarchy_mask & MALI_HIERARCHY_MASK;
373 unsigned tiler_flags = t->hierarchy_mask ^ hierarchy_mask;
374
375 bool tiling_enabled = hierarchy_mask;
376
377 if (tiling_enabled) {
378 /* We should also have no other flags */
379 if (tiler_flags)
380 pandecode_msg("XXX: unexpected tiler %X\n", tiler_flags);
381 } else {
382 /* When tiling is disabled, we should have that flag and no others */
383
384 if (tiler_flags != MALI_TILER_DISABLED) {
385 pandecode_msg("XXX: unexpected tiler flag %X, expected MALI_TILER_DISABLED\n",
386 tiler_flags);
387 }
388
389 /* We should also have an empty heap */
390 if (heap_size) {
391 pandecode_msg("XXX: tiler heap size %d given, expected empty\n",
392 heap_size);
393 }
394
395 /* Disabled tiling is used only for clear-only jobs, which are
396 * purely FRAGMENT, so we should never see this for
397 * non-FRAGMENT descriptors. */
398
399 if (!is_fragment)
400 pandecode_msg("XXX: tiler disabled for non-FRAGMENT job\n");
401 }
402
403 /* We've never seen weights used in practice, but we know from the
404 * kernel these fields is there */
405
406 bool nonzero_weights = false;
407
408 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
409 nonzero_weights |= t->weights[w] != 0x0;
410 }
411
412 if (nonzero_weights) {
413 pandecode_log(".weights = { ");
414
415 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
416 pandecode_log_cont("%d, ", t->weights[w]);
417 }
418
419 pandecode_log("},");
420 }
421
422 pandecode_indent--;
423 pandecode_log("}\n");
424 }
425
426 /* TODO: The Bifrost tiler is not understood at all yet */
427
428 static void
429 pandecode_bifrost_tiler_descriptor(const struct mali_framebuffer *fb)
430 {
431 pandecode_log(".tiler = {\n");
432 pandecode_indent++;
433
434 MEMORY_PROP(fb, tiler_meta);
435
436 for (int i = 0; i < 16; i++) {
437 if (fb->zeros[i] != 0) {
438 pandecode_msg("XXX: tiler descriptor zero %d tripped, value %x\n",
439 i, fb->zeros[i]);
440 }
441 }
442
443 pandecode_log("},\n");
444
445 pandecode_indent--;
446 pandecode_log("}\n");
447
448 }
449
450 /* Information about the framebuffer passed back for
451 * additional analysis */
452
453 struct pandecode_fbd {
454 unsigned width;
455 unsigned height;
456 unsigned rt_count;
457 bool has_extra;
458 };
459
460 static void
461 pandecode_sfbd_format(struct mali_sfbd_format format)
462 {
463 pandecode_log(".format = {\n");
464 pandecode_indent++;
465
466 pandecode_log(".unk1 = ");
467 pandecode_log_decoded_flags(sfbd_unk1_info, format.unk1);
468 pandecode_log_cont(",\n");
469
470 /* TODO: Map formats so we can check swizzles and print nicely */
471 pandecode_log("swizzle");
472 pandecode_swizzle(format.swizzle, MALI_RGBA8_UNORM);
473 pandecode_log_cont(",\n");
474
475 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
476 (format.nr_channels + 1));
477
478 pandecode_log(".unk2 = ");
479 pandecode_log_decoded_flags(sfbd_unk2_info, format.unk2);
480 pandecode_log_cont(",\n");
481
482 pandecode_prop("block = %s", mali_block_format_as_str(format.block));
483
484 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
485
486 pandecode_indent--;
487 pandecode_log("},\n");
488 }
489
490 static void
491 pandecode_shared_memory(const struct mali_shared_memory *desc, bool is_compute)
492 {
493 pandecode_prop("stack_shift = 0x%x", desc->stack_shift);
494
495 if (desc->unk0)
496 pandecode_prop("unk0 = 0x%x", desc->unk0);
497
498 if (desc->shared_workgroup_count != 0x1F) {
499 pandecode_prop("shared_workgroup_count = %d", desc->shared_workgroup_count);
500 if (!is_compute)
501 pandecode_msg("XXX: wrong workgroup count for noncompute\n");
502 }
503
504 if (desc->shared_unk1 || desc->shared_shift) {
505 pandecode_prop("shared_unk1 = %X", desc->shared_unk1);
506 pandecode_prop("shared_shift = %X", desc->shared_shift);
507
508 if (!is_compute)
509 pandecode_msg("XXX: shared memory configured in noncompute shader");
510 }
511
512 if (desc->shared_zero) {
513 pandecode_msg("XXX: shared memory zero tripped\n");
514 pandecode_prop("shared_zero = 0x%" PRIx32, desc->shared_zero);
515 }
516
517 if (desc->shared_memory && !is_compute)
518 pandecode_msg("XXX: shared memory used in noncompute shader\n");
519
520 MEMORY_PROP(desc, scratchpad);
521 MEMORY_PROP(desc, shared_memory);
522 MEMORY_PROP(desc, unknown1);
523
524 if (desc->scratchpad) {
525 struct pandecode_mapped_memory *smem =
526 pandecode_find_mapped_gpu_mem_containing(desc->scratchpad);
527
528 pandecode_msg("scratchpad size %u\n", smem->length);
529 }
530
531 }
532
533 static struct pandecode_fbd
534 pandecode_sfbd(uint64_t gpu_va, int job_no, bool is_fragment, unsigned gpu_id)
535 {
536 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
537 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
538
539 struct pandecode_fbd info = {
540 .has_extra = false,
541 .rt_count = 1
542 };
543
544 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
545 pandecode_indent++;
546
547 pandecode_log(".shared_memory = {\n");
548 pandecode_indent++;
549 pandecode_shared_memory(&s->shared_memory, false);
550 pandecode_indent--;
551 pandecode_log("},\n");
552
553 pandecode_sfbd_format(s->format);
554
555 info.width = s->width + 1;
556 info.height = s->height + 1;
557
558 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", info.width);
559 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", info.height);
560
561 MEMORY_PROP(s, checksum);
562
563 if (s->checksum_stride)
564 pandecode_prop("checksum_stride = %d", s->checksum_stride);
565
566 MEMORY_PROP(s, framebuffer);
567 pandecode_prop("stride = %d", s->stride);
568
569 /* Earlier in the actual commandstream -- right before width -- but we
570 * delay to flow nicer */
571
572 pandecode_log(".clear_flags = ");
573 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
574 pandecode_log_cont(",\n");
575
576 if (s->depth_buffer) {
577 MEMORY_PROP(s, depth_buffer);
578 pandecode_prop("depth_stride = %d", s->depth_stride);
579 }
580
581 if (s->stencil_buffer) {
582 MEMORY_PROP(s, stencil_buffer);
583 pandecode_prop("stencil_stride = %d", s->stencil_stride);
584 }
585
586 if (s->depth_stride_zero ||
587 s->stencil_stride_zero ||
588 s->zero7 || s->zero8) {
589 pandecode_msg("XXX: Depth/stencil zeros tripped\n");
590 pandecode_prop("depth_stride_zero = 0x%x",
591 s->depth_stride_zero);
592 pandecode_prop("stencil_stride_zero = 0x%x",
593 s->stencil_stride_zero);
594 pandecode_prop("zero7 = 0x%" PRIx32,
595 s->zero7);
596 pandecode_prop("zero8 = 0x%" PRIx32,
597 s->zero8);
598 }
599
600 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
601 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
602 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
603 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
604 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
605 }
606
607 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
608 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
609 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
610 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
611 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
612 }
613
614 if (s->clear_stencil) {
615 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
616 }
617
618 const struct midgard_tiler_descriptor t = s->tiler;
619
620 bool has_hierarchy = !(gpu_id == 0x0720 || gpu_id == 0x0820 || gpu_id == 0x0830);
621 pandecode_midgard_tiler_descriptor(&t, s->width + 1, s->height + 1, is_fragment, has_hierarchy);
622
623 pandecode_indent--;
624 pandecode_log("};\n");
625
626 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
627 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
628 pandecode_prop("zero5 = 0x%" PRIx32, s->zero5);
629
630 pandecode_log_cont(".zero3 = {");
631
632 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
633 pandecode_log_cont("%X, ", s->zero3[i]);
634
635 pandecode_log_cont("},\n");
636
637 pandecode_log_cont(".zero6 = {");
638
639 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
640 pandecode_log_cont("%X, ", s->zero6[i]);
641
642 pandecode_log_cont("},\n");
643
644 return info;
645 }
646
647 static void
648 pandecode_compute_fbd(uint64_t gpu_va, int job_no)
649 {
650 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
651 const struct mali_shared_memory *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
652
653 pandecode_log("struct mali_shared_memory shared_%"PRIx64"_%d = {\n", gpu_va, job_no);
654 pandecode_indent++;
655 pandecode_shared_memory(s, true);
656 pandecode_indent--;
657 pandecode_log("},\n");
658 }
659
660 /* Extracts the number of components associated with a Mali format */
661
662 static unsigned
663 pandecode_format_component_count(enum mali_format fmt)
664 {
665 /* Mask out the format class */
666 unsigned top = fmt & 0b11100000;
667
668 switch (top) {
669 case MALI_FORMAT_SNORM:
670 case MALI_FORMAT_UINT:
671 case MALI_FORMAT_UNORM:
672 case MALI_FORMAT_SINT:
673 return ((fmt >> 3) & 3) + 1;
674 default:
675 /* TODO: Validate */
676 return 4;
677 }
678 }
679
680 /* Extracts a mask of accessed components from a 12-bit Mali swizzle */
681
682 static unsigned
683 pandecode_access_mask_from_channel_swizzle(unsigned swizzle)
684 {
685 unsigned mask = 0;
686 assert(MALI_CHANNEL_R == 0);
687
688 for (unsigned c = 0; c < 4; ++c) {
689 enum mali_channel chan = (swizzle >> (3*c)) & 0x7;
690
691 if (chan <= MALI_CHANNEL_A)
692 mask |= (1 << chan);
693 }
694
695 return mask;
696 }
697
698 /* Validates that a (format, swizzle) pair is valid, in the sense that the
699 * swizzle doesn't access any components that are undefined in the format.
700 * Returns whether the swizzle is trivial (doesn't do any swizzling) and can be
701 * omitted */
702
703 static bool
704 pandecode_validate_format_swizzle(enum mali_format fmt, unsigned swizzle)
705 {
706 unsigned nr_comp = pandecode_format_component_count(fmt);
707 unsigned access_mask = pandecode_access_mask_from_channel_swizzle(swizzle);
708 unsigned valid_mask = (1 << nr_comp) - 1;
709 unsigned invalid_mask = ~valid_mask;
710
711 if (access_mask & invalid_mask) {
712 pandecode_msg("XXX: invalid components accessed\n");
713 return false;
714 }
715
716 /* Check for the default non-swizzling swizzle so we can suppress
717 * useless printing for the defaults */
718
719 unsigned default_swizzles[4] = {
720 MALI_CHANNEL_R | (MALI_CHANNEL_0 << 3) | (MALI_CHANNEL_0 << 6) | (MALI_CHANNEL_1 << 9),
721 MALI_CHANNEL_R | (MALI_CHANNEL_G << 3) | (MALI_CHANNEL_0 << 6) | (MALI_CHANNEL_1 << 9),
722 MALI_CHANNEL_R | (MALI_CHANNEL_G << 3) | (MALI_CHANNEL_B << 6) | (MALI_CHANNEL_1 << 9),
723 MALI_CHANNEL_R | (MALI_CHANNEL_G << 3) | (MALI_CHANNEL_B << 6) | (MALI_CHANNEL_A << 9)
724 };
725
726 return (swizzle == default_swizzles[nr_comp - 1]);
727 }
728
729 static void
730 pandecode_swizzle(unsigned swizzle, enum mali_format format)
731 {
732 /* First, do some validation */
733 bool trivial_swizzle = pandecode_validate_format_swizzle(
734 format, swizzle);
735
736 if (trivial_swizzle)
737 return;
738
739 /* Next, print the swizzle */
740 pandecode_log_cont(".");
741
742 static const char components[] = "rgba01";
743
744 for (unsigned c = 0; c < 4; ++c) {
745 enum mali_channel chan = (swizzle >> (3 * c)) & 0x7;
746
747 if (chan > MALI_CHANNEL_1) {
748 pandecode_log("XXX: invalid swizzle channel %d\n", chan);
749 continue;
750 }
751 pandecode_log_cont("%c", components[chan]);
752 }
753 }
754
755 static void
756 pandecode_rt_format(struct mali_rt_format format)
757 {
758 pandecode_log(".format = {\n");
759 pandecode_indent++;
760
761 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
762 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
763 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
764 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
765
766 pandecode_prop("block = %s", mali_block_format_as_str(format.block));
767
768 /* TODO: Map formats so we can check swizzles and print nicely */
769 pandecode_log("swizzle");
770 pandecode_swizzle(format.swizzle, MALI_RGBA8_UNORM);
771 pandecode_log_cont(",\n");
772
773 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
774 (format.nr_channels + 1));
775
776 pandecode_log(".flags = ");
777 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
778 pandecode_log_cont(",\n");
779
780 pandecode_prop("msaa = %s", mali_msaa_as_str(format.msaa));
781
782 /* In theory, the no_preload bit can be cleared to enable MFBD preload,
783 * which is a faster hardware-based alternative to the wallpaper method
784 * to preserve framebuffer contents across frames. In practice, MFBD
785 * preload is buggy on Midgard, and so this is a chicken bit. If this
786 * bit isn't set, most likely something broke unrelated to preload */
787
788 if (!format.no_preload) {
789 pandecode_msg("XXX: buggy MFBD preload enabled - chicken bit should be clear\n");
790 pandecode_prop("no_preload = 0x%" PRIx32, format.no_preload);
791 }
792
793 if (format.zero)
794 pandecode_prop("zero = 0x%" PRIx32, format.zero);
795
796 pandecode_indent--;
797 pandecode_log("},\n");
798 }
799
800 static void
801 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct mali_framebuffer *fb)
802 {
803 pandecode_log("struct mali_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
804 pandecode_indent++;
805
806 for (int i = 0; i < (fb->rt_count_1 + 1); i++) {
807 mali_ptr rt_va = gpu_va + i * sizeof(struct mali_render_target);
808 struct pandecode_mapped_memory *mem =
809 pandecode_find_mapped_gpu_mem_containing(rt_va);
810 const struct mali_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
811
812 pandecode_log("{\n");
813 pandecode_indent++;
814
815 pandecode_rt_format(rt->format);
816
817 if (rt->format.block == MALI_BLOCK_FORMAT_AFBC) {
818 pandecode_log(".afbc = {\n");
819 pandecode_indent++;
820
821 char *a = pointer_as_memory_reference(rt->afbc.metadata);
822 pandecode_prop("metadata = %s", a);
823 free(a);
824
825 pandecode_prop("stride = %d", rt->afbc.stride);
826
827 pandecode_log(".flags = ");
828 pandecode_log_decoded_flags(afbc_fmt_flag_info, rt->afbc.flags);
829 pandecode_log_cont(",\n");
830
831 pandecode_indent--;
832 pandecode_log("},\n");
833 } else if (rt->afbc.metadata || rt->afbc.stride || rt->afbc.flags) {
834 pandecode_msg("XXX: AFBC disabled but AFBC field set (0x%lX, 0x%x, 0x%x)\n",
835 rt->afbc.metadata,
836 rt->afbc.stride,
837 rt->afbc.flags);
838 }
839
840 MEMORY_PROP(rt, framebuffer);
841 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
842
843 if (rt->layer_stride)
844 pandecode_prop("layer_stride = %d", rt->layer_stride);
845
846 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
847 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
848 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
849 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
850 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
851 }
852
853 if (rt->zero1 || rt->zero2) {
854 pandecode_msg("XXX: render target zeros tripped\n");
855 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
856 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
857 }
858
859 pandecode_indent--;
860 pandecode_log("},\n");
861 }
862
863 pandecode_indent--;
864 pandecode_log("};\n");
865 }
866
867 static struct pandecode_fbd
868 pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool is_fragment, bool is_compute, bool is_bifrost)
869 {
870 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
871 const struct mali_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
872
873 struct pandecode_fbd info;
874
875 if (is_bifrost && fb->msaa.sample_locations) {
876 /* The blob stores all possible sample locations in a single buffer
877 * allocated on startup, and just switches the pointer when switching
878 * MSAA state. For now, we just put the data into the cmdstream, but we
879 * should do something like what the blob does with a real driver.
880 *
881 * There seem to be 32 slots for sample locations, followed by another
882 * 16. The second 16 is just the center location followed by 15 zeros
883 * in all the cases I've identified (maybe shader vs. depth/color
884 * samples?).
885 */
886
887 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->msaa.sample_locations);
888
889 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->msaa.sample_locations);
890
891 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
892 pandecode_indent++;
893
894 for (int i = 0; i < 32 + 16; i++) {
895 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
896 }
897
898 pandecode_indent--;
899 pandecode_log("};\n");
900 }
901
902 pandecode_log("struct mali_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
903 pandecode_indent++;
904
905 if (is_bifrost) {
906 pandecode_log(".msaa = {\n");
907 pandecode_indent++;
908
909 if (fb->msaa.sample_locations)
910 pandecode_prop("sample_locations = sample_locations_%d", job_no);
911 else
912 pandecode_msg("XXX: sample_locations missing\n");
913
914 if (fb->msaa.zero1 || fb->msaa.zero2 || fb->msaa.zero4) {
915 pandecode_msg("XXX: multisampling zero tripped\n");
916 pandecode_prop("zero1 = %" PRIx64, fb->msaa.zero1);
917 pandecode_prop("zero2 = %" PRIx64, fb->msaa.zero2);
918 pandecode_prop("zero4 = %" PRIx64, fb->msaa.zero4);
919 }
920
921 pandecode_indent--;
922 pandecode_log("},\n");
923 } else {
924 pandecode_log(".shared_memory = {\n");
925 pandecode_indent++;
926 pandecode_shared_memory(&fb->shared_memory, is_compute);
927 pandecode_indent--;
928 pandecode_log("},\n");
929 }
930
931 info.width = fb->width1 + 1;
932 info.height = fb->height1 + 1;
933 info.rt_count = fb->rt_count_1 + 1;
934
935 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
936 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
937 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
938 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
939
940 pandecode_prop("unk1 = 0x%x", fb->unk1);
941 pandecode_prop("unk2 = 0x%x", fb->unk2);
942 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
943 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
944
945 pandecode_log(".mfbd_flags = ");
946 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
947 pandecode_log_cont(",\n");
948
949 if (fb->clear_stencil)
950 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
951
952 if (fb->clear_depth)
953 pandecode_prop("clear_depth = %f", fb->clear_depth);
954
955 if (!is_compute)
956 if (is_bifrost)
957 pandecode_bifrost_tiler_descriptor(fb);
958 else {
959 const struct midgard_tiler_descriptor t = fb->tiler;
960 pandecode_midgard_tiler_descriptor(&t, fb->width1 + 1, fb->height1 + 1, is_fragment, true);
961 }
962 else
963 pandecode_msg("XXX: skipping compute MFBD, fixme\n");
964
965 if (fb->zero3 || fb->zero4) {
966 pandecode_msg("XXX: framebuffer zeros tripped\n");
967 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
968 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
969 }
970
971 pandecode_indent--;
972 pandecode_log("};\n");
973
974 gpu_va += sizeof(struct mali_framebuffer);
975
976 info.has_extra = (fb->mfbd_flags & MALI_MFBD_EXTRA) && is_fragment;
977
978 if (info.has_extra) {
979 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
980 const struct mali_framebuffer_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
981
982 pandecode_log("struct mali_framebuffer_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
983 pandecode_indent++;
984
985 MEMORY_PROP(fbx, checksum);
986
987 if (fbx->checksum_stride)
988 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
989
990 pandecode_log(".flags_hi = ");
991 pandecode_log_decoded_flags(mfbd_extra_flag_hi_info, fbx->flags_hi);
992 pandecode_log_cont(",\n");
993
994 pandecode_log(".flags_lo = ");
995 pandecode_log_decoded_flags(mfbd_extra_flag_lo_info, fbx->flags_lo);
996 pandecode_log_cont(",\n");
997
998 pandecode_prop("zs_block = %s", mali_block_format_as_str(fbx->zs_block));
999 pandecode_prop("zs_samples = MALI_POSITIVE(%u)", fbx->zs_samples + 1);
1000
1001 if (fbx->zs_block == MALI_BLOCK_FORMAT_AFBC) {
1002 pandecode_log(".ds_afbc = {\n");
1003 pandecode_indent++;
1004
1005 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil_afbc_metadata);
1006 pandecode_prop("depth_stencil_afbc_stride = %d",
1007 fbx->ds_afbc.depth_stencil_afbc_stride);
1008 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil);
1009
1010 pandecode_log(".flags = ");
1011 pandecode_log_decoded_flags(afbc_fmt_flag_info, fbx->ds_afbc.flags);
1012 pandecode_log_cont(",\n");
1013
1014 if (fbx->ds_afbc.padding) {
1015 pandecode_msg("XXX: Depth/stencil AFBC zeros tripped\n");
1016 pandecode_prop("padding = 0x%" PRIx64, fbx->ds_afbc.padding);
1017 }
1018
1019 pandecode_indent--;
1020 pandecode_log("},\n");
1021 } else {
1022 pandecode_log(".ds_linear = {\n");
1023 pandecode_indent++;
1024
1025 if (fbx->ds_linear.depth) {
1026 MEMORY_PROP_DIR(fbx->ds_linear, depth);
1027 pandecode_prop("depth_stride = %d",
1028 fbx->ds_linear.depth_stride);
1029 pandecode_prop("depth_layer_stride = %d",
1030 fbx->ds_linear.depth_layer_stride);
1031 } else if (fbx->ds_linear.depth_stride || fbx->ds_linear.depth_layer_stride) {
1032 pandecode_msg("XXX: depth stride zero tripped %d %d\n", fbx->ds_linear.depth_stride, fbx->ds_linear.depth_layer_stride);
1033 }
1034
1035 if (fbx->ds_linear.stencil) {
1036 MEMORY_PROP_DIR(fbx->ds_linear, stencil);
1037 pandecode_prop("stencil_stride = %d",
1038 fbx->ds_linear.stencil_stride);
1039 pandecode_prop("stencil_layer_stride = %d",
1040 fbx->ds_linear.stencil_layer_stride);
1041 } else if (fbx->ds_linear.stencil_stride || fbx->ds_linear.stencil_layer_stride) {
1042 pandecode_msg("XXX: stencil stride zero tripped %d %d\n", fbx->ds_linear.stencil_stride, fbx->ds_linear.stencil_layer_stride);
1043 }
1044
1045 if (fbx->ds_linear.depth_stride_zero ||
1046 fbx->ds_linear.stencil_stride_zero) {
1047 pandecode_msg("XXX: Depth/stencil zeros tripped\n");
1048 pandecode_prop("depth_stride_zero = 0x%x",
1049 fbx->ds_linear.depth_stride_zero);
1050 pandecode_prop("stencil_stride_zero = 0x%x",
1051 fbx->ds_linear.stencil_stride_zero);
1052 }
1053
1054 pandecode_indent--;
1055 pandecode_log("},\n");
1056 }
1057
1058 if (fbx->clear_color_1 | fbx->clear_color_2) {
1059 pandecode_prop("clear_color_1 = 0x%" PRIx32, fbx->clear_color_1);
1060 pandecode_prop("clear_color_2 = 0x%" PRIx32, fbx->clear_color_2);
1061 }
1062
1063 if (fbx->zero3) {
1064 pandecode_msg("XXX: fb_extra zeros tripped\n");
1065 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
1066 }
1067
1068 pandecode_indent--;
1069 pandecode_log("};\n");
1070
1071 gpu_va += sizeof(struct mali_framebuffer_extra);
1072 }
1073
1074 if (is_fragment)
1075 pandecode_render_target(gpu_va, job_no, fb);
1076
1077 return info;
1078 }
1079
1080 static void
1081 pandecode_attributes(const struct pandecode_mapped_memory *mem,
1082 mali_ptr addr, int job_no, char *suffix,
1083 int count, bool varying, enum mali_job_type job_type)
1084 {
1085 char *prefix = varying ? "Varying" : "Attribute";
1086 assert(addr);
1087
1088 if (!count) {
1089 pandecode_msg("warn: No %s records\n", prefix);
1090 return;
1091 }
1092
1093 MAP_ADDR(ATTRIBUTE_BUFFER, addr, cl);
1094
1095 for (int i = 0; i < count; ++i) {
1096 fprintf(pandecode_dump_stream, "%s\n", prefix);
1097
1098 struct MALI_ATTRIBUTE_BUFFER temp;
1099 MALI_ATTRIBUTE_BUFFER_unpack(cl + i * MALI_ATTRIBUTE_BUFFER_LENGTH, &temp);
1100 MALI_ATTRIBUTE_BUFFER_print(pandecode_dump_stream, &temp, 2);
1101
1102 if (temp.type == MALI_ATTRIBUTE_TYPE_1D_NPOT_DIVISOR) {
1103 struct MALI_ATTRIBUTE_BUFFER_CONTINUATION_NPOT temp2;
1104 MALI_ATTRIBUTE_BUFFER_CONTINUATION_NPOT_unpack(cl + (i + 1) * MALI_ATTRIBUTE_BUFFER_LENGTH, &temp2);
1105 MALI_ATTRIBUTE_BUFFER_CONTINUATION_NPOT_print(pandecode_dump_stream, &temp2, 2);
1106 }
1107 }
1108 }
1109
1110 static mali_ptr
1111 pandecode_shader_address(const char *name, mali_ptr ptr)
1112 {
1113 /* TODO: Decode flags */
1114 mali_ptr shader_ptr = ptr & ~15;
1115
1116 char *a = pointer_as_memory_reference(shader_ptr);
1117 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
1118 free(a);
1119
1120 return shader_ptr;
1121 }
1122
1123 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1124
1125 static unsigned
1126 decode_bifrost_constant(u16 constant)
1127 {
1128 float lo = (float) (constant & 0xFF);
1129 float hi = (float) (constant >> 8);
1130
1131 return (hi / 255.0) + (lo / 65535.0);
1132 }
1133
1134 static mali_ptr
1135 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
1136 {
1137 struct bifrost_blend_rt *b =
1138 ((struct bifrost_blend_rt *) descs) + rt_no;
1139
1140 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1141 pandecode_indent++;
1142
1143 pandecode_prop("flags = 0x%" PRIx16, b->flags);
1144 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
1145 b->constant, decode_bifrost_constant(b->constant));
1146
1147 /* TODO figure out blend shader enable bit */
1148 DUMP_CL("Equation", BLEND_EQUATION, &b->equation, 2);
1149
1150 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1151 pandecode_prop("index = 0x%" PRIx16, b->index);
1152
1153 pandecode_log(".format = %s", mali_format_as_str(b->format));
1154 pandecode_swizzle(b->swizzle, b->format);
1155 pandecode_log_cont(",\n");
1156
1157 pandecode_prop("swizzle = 0x%" PRIx32, b->swizzle);
1158 pandecode_prop("format = 0x%" PRIx32, b->format);
1159
1160 if (b->zero1) {
1161 pandecode_msg("XXX: pandecode_bifrost_blend zero1 tripped\n");
1162 pandecode_prop("zero1 = 0x%" PRIx32, b->zero1);
1163 }
1164
1165 pandecode_log(".shader_type = ");
1166 switch(b->shader_type) {
1167 case BIFROST_BLEND_F16:
1168 pandecode_log_cont("BIFROST_BLEND_F16");
1169 break;
1170 case BIFROST_BLEND_F32:
1171 pandecode_log_cont("BIFROST_BLEND_F32");
1172 break;
1173 case BIFROST_BLEND_I32:
1174 pandecode_log_cont("BIFROST_BLEND_I32");
1175 break;
1176 case BIFROST_BLEND_U32:
1177 pandecode_log_cont("BIFROST_BLEND_U32");
1178 break;
1179 case BIFROST_BLEND_I16:
1180 pandecode_log_cont("BIFROST_BLEND_I16");
1181 break;
1182 case BIFROST_BLEND_U16:
1183 pandecode_log_cont("BIFROST_BLEND_U16");
1184 break;
1185 }
1186 pandecode_log_cont(",\n");
1187
1188 if (b->zero2) {
1189 pandecode_msg("XXX: pandecode_bifrost_blend zero2 tripped\n");
1190 pandecode_prop("zero2 = 0x%" PRIx32, b->zero2);
1191 }
1192
1193 pandecode_prop("shader = 0x%" PRIx32, b->shader);
1194
1195 pandecode_indent--;
1196 pandecode_log("},\n");
1197
1198 return 0;
1199 }
1200
1201 static mali_ptr
1202 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
1203 {
1204 /* constant/equation is in a union */
1205 if (!blend->shader)
1206 return 0;
1207
1208 pandecode_log(".blend = {\n");
1209 pandecode_indent++;
1210
1211 if (is_shader) {
1212 pandecode_shader_address("shader", blend->shader);
1213 } else {
1214 DUMP_CL("Equation", BLEND_EQUATION, &blend->equation, 2);
1215 pandecode_prop("constant = %f", blend->constant);
1216 }
1217
1218 pandecode_indent--;
1219 pandecode_log("},\n");
1220
1221 /* Return blend shader to disassemble if present */
1222 return is_shader ? (blend->shader & ~0xF) : 0;
1223 }
1224
1225 static mali_ptr
1226 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1227 {
1228 struct midgard_blend_rt *b =
1229 ((struct midgard_blend_rt *) descs) + rt_no;
1230
1231 /* Flags determine presence of blend shader */
1232 bool is_shader = b->flags.opaque[0] & 0x2;
1233
1234 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1235 pandecode_indent++;
1236
1237 DUMP_CL("Flags", BLEND_FLAGS, &b->flags, 2);
1238
1239 union midgard_blend blend = b->blend;
1240 mali_ptr shader = pandecode_midgard_blend(&blend, is_shader);
1241
1242 pandecode_indent--;
1243 pandecode_log("};\n");
1244
1245 return shader;
1246 }
1247
1248 /* Attributes and varyings have descriptor records, which contain information
1249 * about their format and ordering with the attribute/varying buffers. We'll
1250 * want to validate that the combinations specified are self-consistent.
1251 */
1252
1253 static int
1254 pandecode_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1255 {
1256 const char *prefix = varying ? "Varying" : "Attribute";
1257 mali_ptr p = varying ? v->varying_meta : v->attribute_meta;
1258
1259 for (int i = 0; i < count; ++i, p += MALI_ATTRIBUTE_LENGTH)
1260 DUMP_ADDR(prefix, ATTRIBUTE, p, 1);
1261
1262 return count;
1263 }
1264
1265 /* return bits [lo, hi) of word */
1266 static u32
1267 bits(u32 word, u32 lo, u32 hi)
1268 {
1269 if (hi - lo >= 32)
1270 return word; // avoid undefined behavior with the shift
1271
1272 return (word >> lo) & ((1 << (hi - lo)) - 1);
1273 }
1274
1275 static void
1276 pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no, bool graphics)
1277 {
1278 pandecode_log(".prefix = {\n");
1279 pandecode_indent++;
1280
1281 /* Decode invocation_count. See the comment before the definition of
1282 * invocation_count for an explanation.
1283 */
1284
1285 unsigned size_y_shift = bits(p->invocation_shifts, 0, 5);
1286 unsigned size_z_shift = bits(p->invocation_shifts, 5, 10);
1287 unsigned workgroups_x_shift = bits(p->invocation_shifts, 10, 16);
1288 unsigned workgroups_y_shift = bits(p->invocation_shifts, 16, 22);
1289 unsigned workgroups_z_shift = bits(p->invocation_shifts, 22, 28);
1290 unsigned workgroups_x_shift_2 = bits(p->invocation_shifts, 28, 32);
1291
1292 unsigned size_x = bits(p->invocation_count, 0, size_y_shift) + 1;
1293 unsigned size_y = bits(p->invocation_count, size_y_shift, size_z_shift) + 1;
1294 unsigned size_z = bits(p->invocation_count, size_z_shift, workgroups_x_shift) + 1;
1295
1296 unsigned groups_x = bits(p->invocation_count, workgroups_x_shift, workgroups_y_shift) + 1;
1297 unsigned groups_y = bits(p->invocation_count, workgroups_y_shift, workgroups_z_shift) + 1;
1298 unsigned groups_z = bits(p->invocation_count, workgroups_z_shift, 32) + 1;
1299
1300 /* Even though we have this decoded, we want to ensure that the
1301 * representation is "unique" so we don't lose anything by printing only
1302 * the final result. More specifically, we need to check that we were
1303 * passed something in canonical form, since the definition per the
1304 * hardware is inherently not unique. How? Well, take the resulting
1305 * decode and pack it ourselves! If it is bit exact with what we
1306 * decoded, we're good to go. */
1307
1308 struct mali_vertex_tiler_prefix ref;
1309 panfrost_pack_work_groups_compute(&ref, groups_x, groups_y, groups_z, size_x, size_y, size_z, graphics);
1310
1311 bool canonical =
1312 (p->invocation_count == ref.invocation_count) &&
1313 (p->invocation_shifts == ref.invocation_shifts);
1314
1315 if (!canonical) {
1316 pandecode_msg("XXX: non-canonical workgroups packing\n");
1317 pandecode_msg("expected: %X, %X",
1318 ref.invocation_count,
1319 ref.invocation_shifts);
1320
1321 pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
1322 pandecode_prop("size_y_shift = %d", size_y_shift);
1323 pandecode_prop("size_z_shift = %d", size_z_shift);
1324 pandecode_prop("workgroups_x_shift = %d", workgroups_x_shift);
1325 pandecode_prop("workgroups_y_shift = %d", workgroups_y_shift);
1326 pandecode_prop("workgroups_z_shift = %d", workgroups_z_shift);
1327 pandecode_prop("workgroups_x_shift_2 = %d", workgroups_x_shift_2);
1328 }
1329
1330 /* Regardless, print the decode */
1331 pandecode_msg("size (%d, %d, %d), count (%d, %d, %d)\n",
1332 size_x, size_y, size_z,
1333 groups_x, groups_y, groups_z);
1334
1335 /* TODO: Decode */
1336 if (p->unknown_draw)
1337 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1338
1339 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1340
1341 if (p->draw_mode != MALI_DRAW_MODE_NONE)
1342 pandecode_prop("draw_mode = %s", mali_draw_mode_as_str(p->draw_mode));
1343
1344 /* Index count only exists for tiler jobs anyway */
1345
1346 if (p->index_count)
1347 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1348
1349
1350 unsigned index_raw_size = (p->unknown_draw & MALI_DRAW_INDEXED_SIZE);
1351 index_raw_size >>= MALI_DRAW_INDEXED_SHIFT;
1352
1353 /* Validate an index buffer is present if we need one. TODO: verify
1354 * relationship between invocation_count and index_count */
1355
1356 if (p->indices) {
1357 unsigned count = p->index_count;
1358
1359 /* Grab the size */
1360 unsigned size = (index_raw_size == 0x3) ? 4 : index_raw_size;
1361
1362 /* Ensure we got a size, and if so, validate the index buffer
1363 * is large enough to hold a full set of indices of the given
1364 * size */
1365
1366 if (!index_raw_size)
1367 pandecode_msg("XXX: index size missing\n");
1368 else
1369 pandecode_validate_buffer(p->indices, count * size);
1370 } else if (index_raw_size)
1371 pandecode_msg("XXX: unexpected index size %u\n", index_raw_size);
1372
1373 if (p->offset_bias_correction)
1374 pandecode_prop("offset_bias_correction = %d", p->offset_bias_correction);
1375
1376 /* TODO: Figure out what this is. It's not zero */
1377 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1378
1379 pandecode_indent--;
1380 pandecode_log("},\n");
1381 }
1382
1383 static void
1384 pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1385 {
1386 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1387 uint64_t *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1388
1389 for (int i = 0; i < ubufs_count; i++) {
1390 unsigned size = (ubufs[i] & ((1 << 10) - 1)) * 16;
1391 mali_ptr addr = (ubufs[i] >> 10) << 2;
1392
1393 pandecode_validate_buffer(addr, size);
1394
1395 char *ptr = pointer_as_memory_reference(addr);
1396 pandecode_log("ubuf_%d[%u] = %s;\n", i, size, ptr);
1397 free(ptr);
1398 }
1399
1400 pandecode_log("\n");
1401 }
1402
1403 static void
1404 pandecode_uniforms(mali_ptr uniforms, unsigned uniform_count)
1405 {
1406 pandecode_validate_buffer(uniforms, uniform_count * 16);
1407
1408 char *ptr = pointer_as_memory_reference(uniforms);
1409 pandecode_log("vec4 uniforms[%u] = %s;\n", uniform_count, ptr);
1410 free(ptr);
1411 }
1412
1413 static const char *
1414 shader_type_for_job(unsigned type)
1415 {
1416 switch (type) {
1417 case MALI_JOB_TYPE_VERTEX: return "VERTEX";
1418 case MALI_JOB_TYPE_TILER: return "FRAGMENT";
1419 case MALI_JOB_TYPE_COMPUTE: return "COMPUTE";
1420 default:
1421 return "UNKNOWN";
1422 }
1423 }
1424
1425 static unsigned shader_id = 0;
1426
1427 static struct midgard_disasm_stats
1428 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1429 bool is_bifrost, unsigned gpu_id)
1430 {
1431 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1432 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1433
1434 /* Compute maximum possible size */
1435 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1436
1437 /* Print some boilerplate to clearly denote the assembly (which doesn't
1438 * obey indentation rules), and actually do the disassembly! */
1439
1440 pandecode_log_cont("\n\n");
1441
1442 struct midgard_disasm_stats stats;
1443
1444 if (is_bifrost) {
1445 disassemble_bifrost(pandecode_dump_stream, code, sz, true);
1446
1447 /* TODO: Extend stats to Bifrost */
1448 stats.texture_count = -128;
1449 stats.sampler_count = -128;
1450 stats.attribute_count = -128;
1451 stats.varying_count = -128;
1452 stats.uniform_count = -128;
1453 stats.uniform_buffer_count = -128;
1454 stats.work_count = -128;
1455
1456 stats.instruction_count = 0;
1457 stats.bundle_count = 0;
1458 stats.quadword_count = 0;
1459 stats.helper_invocations = false;
1460 } else {
1461 stats = disassemble_midgard(pandecode_dump_stream,
1462 code, sz, gpu_id,
1463 type == MALI_JOB_TYPE_TILER ?
1464 MESA_SHADER_FRAGMENT : MESA_SHADER_VERTEX);
1465 }
1466
1467 /* Print shader-db stats. Skip COMPUTE jobs since they are used for
1468 * driver-internal purposes with the blob and interfere */
1469
1470 bool should_shaderdb = type != MALI_JOB_TYPE_COMPUTE;
1471
1472 if (should_shaderdb) {
1473 unsigned nr_threads =
1474 (stats.work_count <= 4) ? 4 :
1475 (stats.work_count <= 8) ? 2 :
1476 1;
1477
1478 pandecode_log_cont("shader%d - MESA_SHADER_%s shader: "
1479 "%u inst, %u bundles, %u quadwords, "
1480 "%u registers, %u threads, 0 loops, 0:0 spills:fills\n\n\n",
1481 shader_id++,
1482 shader_type_for_job(type),
1483 stats.instruction_count, stats.bundle_count, stats.quadword_count,
1484 stats.work_count, nr_threads);
1485 }
1486
1487
1488 return stats;
1489 }
1490
1491 static void
1492 pandecode_texture_payload(mali_ptr payload,
1493 enum mali_texture_dimension dim,
1494 enum mali_texture_layout layout,
1495 bool manual_stride,
1496 uint8_t levels,
1497 uint16_t depth,
1498 uint16_t array_size,
1499 struct pandecode_mapped_memory *tmem)
1500 {
1501 pandecode_log(".payload = {\n");
1502 pandecode_indent++;
1503
1504 /* A bunch of bitmap pointers follow.
1505 * We work out the correct number,
1506 * based on the mipmap/cubemap
1507 * properties, but dump extra
1508 * possibilities to futureproof */
1509
1510 int bitmap_count = levels + 1;
1511
1512 /* Miptree for each face */
1513 if (dim == MALI_TEXTURE_DIMENSION_CUBE)
1514 bitmap_count *= 6;
1515
1516 /* Array of layers */
1517 bitmap_count *= depth;
1518
1519 /* Array of textures */
1520 bitmap_count *= array_size;
1521
1522 /* Stride for each element */
1523 if (manual_stride)
1524 bitmap_count *= 2;
1525
1526 mali_ptr *pointers_and_strides = pandecode_fetch_gpu_mem(tmem,
1527 payload, sizeof(mali_ptr) * bitmap_count);
1528 for (int i = 0; i < bitmap_count; ++i) {
1529 /* How we dump depends if this is a stride or a pointer */
1530
1531 if (manual_stride && (i & 1)) {
1532 /* signed 32-bit snuck in as a 64-bit pointer */
1533 uint64_t stride_set = pointers_and_strides[i];
1534 uint32_t clamped_stride = stride_set;
1535 int32_t stride = clamped_stride;
1536 assert(stride_set == clamped_stride);
1537 pandecode_log("(mali_ptr) %d /* stride */, \n", stride);
1538 } else {
1539 char *a = pointer_as_memory_reference(pointers_and_strides[i]);
1540 pandecode_log("%s, \n", a);
1541 free(a);
1542 }
1543 }
1544
1545 pandecode_indent--;
1546 pandecode_log("},\n");
1547 }
1548
1549 static void
1550 pandecode_texture(mali_ptr u,
1551 struct pandecode_mapped_memory *tmem,
1552 unsigned job_no, unsigned tex)
1553 {
1554 struct pandecode_mapped_memory *mapped_mem = pandecode_find_mapped_gpu_mem_containing(u);
1555 const uint8_t *cl = pandecode_fetch_gpu_mem(mapped_mem, u, MALI_MIDGARD_TEXTURE_LENGTH);
1556
1557 struct MALI_MIDGARD_TEXTURE temp;
1558 MALI_MIDGARD_TEXTURE_unpack(cl, &temp);
1559 MALI_MIDGARD_TEXTURE_print(pandecode_dump_stream, &temp, 2);
1560
1561 pandecode_texture_payload(u + MALI_MIDGARD_TEXTURE_LENGTH,
1562 temp.dimension, temp.texel_ordering, temp.manual_stride,
1563 temp.levels, temp.depth, temp.array_size, mapped_mem);
1564 }
1565
1566 static void
1567 pandecode_bifrost_texture(
1568 const void *cl,
1569 unsigned job_no,
1570 unsigned tex)
1571 {
1572 struct MALI_BIFROST_TEXTURE temp;
1573 MALI_BIFROST_TEXTURE_unpack(cl, &temp);
1574 MALI_BIFROST_TEXTURE_print(pandecode_dump_stream, &temp, 2);
1575
1576 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(temp.surfaces);
1577 pandecode_texture_payload(temp.surfaces, temp.dimension, temp.texel_ordering,
1578 true, temp.levels, 1, 1, tmem);
1579 }
1580
1581 /* For shader properties like texture_count, we have a claimed property in the shader_meta, and the actual Truth from static analysis (this may just be an upper limit). We validate accordingly */
1582
1583 static void
1584 pandecode_shader_prop(const char *name, unsigned claim, signed truth, bool fuzzy)
1585 {
1586 /* Nothing to do */
1587 if (claim == truth)
1588 return;
1589
1590 if (fuzzy && (truth < 0))
1591 pandecode_msg("XXX: fuzzy %s, claimed %d, expected %d\n", name, claim, truth);
1592
1593 if ((truth >= 0) && !fuzzy) {
1594 pandecode_msg("%s: expected %s = %d, claimed %u\n",
1595 (truth < claim) ? "warn" : "XXX",
1596 name, truth, claim);
1597 } else if ((claim > -truth) && !fuzzy) {
1598 pandecode_msg("XXX: expected %s <= %u, claimed %u\n",
1599 name, -truth, claim);
1600 } else if (fuzzy && (claim < truth))
1601 pandecode_msg("XXX: expected %s >= %u, claimed %u\n",
1602 name, truth, claim);
1603
1604 pandecode_log(".%s = %" PRId16, name, claim);
1605
1606 if (fuzzy)
1607 pandecode_log_cont(" /* %u used */", truth);
1608
1609 pandecode_log_cont(",\n");
1610 }
1611
1612 static void
1613 pandecode_blend_shader_disassemble(mali_ptr shader, int job_no, int job_type,
1614 bool is_bifrost, unsigned gpu_id)
1615 {
1616 struct midgard_disasm_stats stats =
1617 pandecode_shader_disassemble(shader, job_no, job_type, is_bifrost, gpu_id);
1618
1619 bool has_texture = (stats.texture_count > 0);
1620 bool has_sampler = (stats.sampler_count > 0);
1621 bool has_attribute = (stats.attribute_count > 0);
1622 bool has_varying = (stats.varying_count > 0);
1623 bool has_uniform = (stats.uniform_count > 0);
1624 bool has_ubo = (stats.uniform_buffer_count > 0);
1625
1626 if (has_texture || has_sampler)
1627 pandecode_msg("XXX: blend shader accessing textures\n");
1628
1629 if (has_attribute || has_varying)
1630 pandecode_msg("XXX: blend shader accessing interstage\n");
1631
1632 if (has_uniform || has_ubo)
1633 pandecode_msg("XXX: blend shader accessing uniforms\n");
1634 }
1635
1636 static void
1637 pandecode_textures(mali_ptr textures, unsigned texture_count, int job_no, bool is_bifrost)
1638 {
1639 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(textures);
1640
1641 if (!mmem)
1642 return;
1643
1644 pandecode_log("Textures (%"PRIx64"):\n", textures);
1645
1646 if (is_bifrost) {
1647 const void *cl = pandecode_fetch_gpu_mem(mmem,
1648 textures, MALI_BIFROST_TEXTURE_LENGTH *
1649 texture_count);
1650
1651 for (unsigned tex = 0; tex < texture_count; ++tex) {
1652 pandecode_bifrost_texture(cl +
1653 MALI_BIFROST_TEXTURE_LENGTH * tex,
1654 job_no, tex);
1655 }
1656 } else {
1657 mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures);
1658
1659 for (int tex = 0; tex < texture_count; ++tex) {
1660 mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr));
1661 char *a = pointer_as_memory_reference(*u);
1662 pandecode_log("%s,\n", a);
1663 free(a);
1664 }
1665
1666 /* Now, finally, descend down into the texture descriptor */
1667 for (unsigned tex = 0; tex < texture_count; ++tex) {
1668 mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr));
1669 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1670 if (tmem)
1671 pandecode_texture(*u, tmem, job_no, tex);
1672 }
1673 }
1674 }
1675
1676 static void
1677 pandecode_samplers(mali_ptr samplers, unsigned sampler_count, int job_no, bool is_bifrost)
1678 {
1679 for (int i = 0; i < sampler_count; ++i) {
1680 if (is_bifrost) {
1681 DUMP_ADDR("Sampler", BIFROST_SAMPLER, samplers + (MALI_BIFROST_SAMPLER_LENGTH * i), 1);
1682 } else {
1683 DUMP_ADDR("Sampler", MIDGARD_SAMPLER, samplers + (MALI_MIDGARD_SAMPLER_LENGTH * i), 1);
1684 }
1685 }
1686 }
1687
1688 static void
1689 pandecode_vertex_tiler_postfix_pre(
1690 const struct mali_vertex_tiler_postfix *p,
1691 int job_no, enum mali_job_type job_type,
1692 char *suffix, bool is_bifrost, unsigned gpu_id)
1693 {
1694 struct pandecode_mapped_memory *attr_mem;
1695
1696 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1697 * are the only things actually needed from the FBD, vertex/tiler jobs
1698 * no longer reference the FBD -- instead, this field points to some
1699 * info about the scratchpad.
1700 */
1701
1702 struct pandecode_fbd fbd_info = {
1703 /* Default for Bifrost */
1704 .rt_count = 1
1705 };
1706
1707 if (is_bifrost) {
1708 pandecode_log_cont("\t/* %X %/\n", p->shared_memory & 1);
1709 pandecode_compute_fbd(p->shared_memory & ~1, job_no);
1710 } else if (p->shared_memory & MALI_MFBD)
1711 fbd_info = pandecode_mfbd_bfr((u64) ((uintptr_t) p->shared_memory) & FBD_MASK, job_no, false, job_type == MALI_JOB_TYPE_COMPUTE, false);
1712 else if (job_type == MALI_JOB_TYPE_COMPUTE)
1713 pandecode_compute_fbd((u64) (uintptr_t) p->shared_memory, job_no);
1714 else
1715 fbd_info = pandecode_sfbd((u64) (uintptr_t) p->shared_memory, job_no, false, gpu_id);
1716
1717 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1718 int texture_count = 0, sampler_count = 0;
1719
1720 if (p->shader) {
1721 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->shader);
1722 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, p->shader);
1723
1724 /* Disassemble ahead-of-time to get stats. Initialize with
1725 * stats for the missing-shader case so we get validation
1726 * there, too */
1727
1728 struct midgard_disasm_stats info = {
1729 .texture_count = 0,
1730 .sampler_count = 0,
1731 .attribute_count = 0,
1732 .varying_count = 0,
1733 .work_count = 1,
1734
1735 .uniform_count = -128,
1736 .uniform_buffer_count = 0
1737 };
1738
1739 if (s->shader & ~0xF)
1740 info = pandecode_shader_disassemble(s->shader & ~0xF, job_no, job_type, is_bifrost, gpu_id);
1741
1742 struct MALI_MIDGARD_PROPERTIES midg_props;
1743
1744 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", p->shader, job_no, suffix);
1745 pandecode_indent++;
1746
1747 /* Save for dumps */
1748 attribute_count = s->attribute_count;
1749 varying_count = s->varying_count;
1750 texture_count = s->texture_count;
1751 sampler_count = s->sampler_count;
1752
1753 if (is_bifrost) {
1754 uniform_count = s->bifrost2.uniform_count;
1755 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1756 } else {
1757 uint32_t opaque = s->midgard_props.opaque[0];
1758 MALI_MIDGARD_PROPERTIES_unpack((const uint8_t *) &opaque, &midg_props);
1759
1760 uniform_count = midg_props.uniform_count;
1761 uniform_buffer_count = midg_props.uniform_buffer_count;
1762 }
1763
1764 pandecode_shader_address("shader", s->shader);
1765
1766 pandecode_shader_prop("texture_count", s->texture_count, info.texture_count, false);
1767 pandecode_shader_prop("sampler_count", s->sampler_count, info.sampler_count, false);
1768 pandecode_shader_prop("attribute_count", s->attribute_count, info.attribute_count, false);
1769 pandecode_shader_prop("varying_count", s->varying_count, info.varying_count, false);
1770
1771 if (is_bifrost) {
1772 pandecode_log("bifrost1.unk1 = ");
1773 pandecode_log_decoded_flags(shader_bifrost_info, s->bifrost1.unk1);
1774 pandecode_log_cont(",\n");
1775 } else {
1776 MALI_MIDGARD_PROPERTIES_print(pandecode_dump_stream, &midg_props, 2);
1777 }
1778
1779 if (s->depth_units || s->depth_factor) {
1780 pandecode_prop("depth_factor = %f", s->depth_factor);
1781 pandecode_prop("depth_units = %f", s->depth_units);
1782 }
1783
1784 if (s->coverage_mask)
1785 pandecode_prop("coverage_mask = 0x%X", s->coverage_mask);
1786
1787 if (s->unknown2_2)
1788 pandecode_prop(".unknown2_2 = %X", s->unknown2_2);
1789
1790 if (s->unknown2_3 || s->unknown2_4) {
1791 pandecode_log(".unknown2_3 = ");
1792
1793 int unknown2_3 = s->unknown2_3;
1794 int unknown2_4 = s->unknown2_4;
1795
1796 /* We're not quite sure what these flags mean without the depth test, if anything */
1797
1798 if (unknown2_3 & (MALI_DEPTH_WRITEMASK | MALI_DEPTH_FUNC_MASK)) {
1799 const char *func = mali_func_as_str(MALI_GET_DEPTH_FUNC(unknown2_3));
1800 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1801
1802 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1803 }
1804
1805 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1806 pandecode_log_cont(",\n");
1807
1808 pandecode_log(".unknown2_4 = ");
1809 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1810 pandecode_log_cont(",\n");
1811 }
1812
1813 if (s->stencil_mask_front || s->stencil_mask_back) {
1814 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1815 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1816 }
1817
1818 DUMP_CL("Stencil front", STENCIL, &s->stencil_front, 1);
1819 DUMP_CL("Stencil back", STENCIL, &s->stencil_back, 1);
1820
1821 if (is_bifrost) {
1822 pandecode_log(".bifrost2 = {\n");
1823 pandecode_indent++;
1824
1825 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1826 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1827 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1828 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1829
1830 pandecode_indent--;
1831 pandecode_log("},\n");
1832 } else if (s->midgard2.unknown2_7) {
1833 pandecode_log(".midgard2 = {\n");
1834 pandecode_indent++;
1835
1836 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1837 pandecode_indent--;
1838 pandecode_log("},\n");
1839 }
1840
1841 if (s->padding) {
1842 pandecode_msg("XXX: shader padding tripped\n");
1843 pandecode_prop("padding = 0x%" PRIx32, s->padding);
1844 }
1845
1846 if (!is_bifrost) {
1847 /* TODO: Blend shaders routing/disasm */
1848 union midgard_blend blend = s->blend;
1849 mali_ptr shader = pandecode_midgard_blend(&blend, s->unknown2_3 & MALI_HAS_BLEND_SHADER);
1850 if (shader & ~0xF)
1851 pandecode_blend_shader_disassemble(shader, job_no, job_type, false, gpu_id);
1852 } else {
1853 pandecode_msg("mdg_blend = %" PRIx64 "\n", s->blend.shader);
1854 }
1855
1856 pandecode_indent--;
1857 pandecode_log("};\n");
1858
1859 /* MRT blend fields are used whenever MFBD is used, with
1860 * per-RT descriptors */
1861
1862 if (job_type == MALI_JOB_TYPE_TILER && (is_bifrost || p->shared_memory & MALI_MFBD)) {
1863 void* blend_base = (void *) (s + 1);
1864
1865 for (unsigned i = 0; i < fbd_info.rt_count; i++) {
1866 mali_ptr shader = 0;
1867
1868 if (is_bifrost)
1869 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1870 else
1871 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1872
1873 if (shader & ~0xF)
1874 pandecode_blend_shader_disassemble(shader, job_no, job_type, false, gpu_id);
1875
1876 }
1877 }
1878 } else
1879 pandecode_msg("XXX: missing shader descriptor\n");
1880
1881 if (p->viewport)
1882 DUMP_ADDR("Viewport", VIEWPORT, p->viewport, 1);
1883
1884 unsigned max_attr_index = 0;
1885
1886 if (p->attribute_meta)
1887 max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
1888
1889 if (p->attributes) {
1890 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1891 pandecode_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index, false, job_type);
1892 }
1893
1894 /* Varyings are encoded like attributes but not actually sent; we just
1895 * pass a zero buffer with the right stride/size set, (or whatever)
1896 * since the GPU will write to it itself */
1897
1898 if (p->varying_meta) {
1899 varying_count = pandecode_attribute_meta(job_no, varying_count, p, true, suffix);
1900 }
1901
1902 if (p->varyings) {
1903 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1904
1905 /* Number of descriptors depends on whether there are
1906 * non-internal varyings */
1907
1908 pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true, job_type);
1909 }
1910
1911 if (p->uniform_buffers) {
1912 if (uniform_buffer_count)
1913 pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1914 else
1915 pandecode_msg("warn: UBOs specified but not referenced\n");
1916 } else if (uniform_buffer_count)
1917 pandecode_msg("XXX: UBOs referenced but not specified\n");
1918
1919 /* We don't want to actually dump uniforms, but we do need to validate
1920 * that the counts we were given are sane */
1921
1922 if (p->uniforms) {
1923 if (uniform_count)
1924 pandecode_uniforms(p->uniforms, uniform_count);
1925 else
1926 pandecode_msg("warn: Uniforms specified but not referenced\n");
1927 } else if (uniform_count)
1928 pandecode_msg("XXX: Uniforms referenced but not specified\n");
1929
1930 if (p->textures)
1931 pandecode_textures(p->textures, texture_count, job_no, is_bifrost);
1932
1933 if (p->sampler_descriptor)
1934 pandecode_samplers(p->sampler_descriptor, sampler_count, job_no, is_bifrost);
1935 }
1936
1937 static void
1938 pandecode_gl_enables(uint32_t gl_enables, int job_type)
1939 {
1940 pandecode_log(".gl_enables = ");
1941
1942 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
1943
1944 pandecode_log_cont(",\n");
1945 }
1946
1947 static void
1948 pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1949 {
1950 if (p->shader & 0xF)
1951 pandecode_msg("warn: shader tagged %X\n", (unsigned) (p->shader & 0xF));
1952
1953 pandecode_log(".postfix = {\n");
1954 pandecode_indent++;
1955
1956 pandecode_gl_enables(p->gl_enables, MALI_JOB_TYPE_TILER);
1957 pandecode_prop("instance_shift = 0x%x", p->instance_shift);
1958 pandecode_prop("instance_odd = 0x%x", p->instance_odd);
1959
1960 if (p->zero4) {
1961 pandecode_msg("XXX: vertex only zero tripped");
1962 pandecode_prop("zero4 = 0x%" PRIx32, p->zero4);
1963 }
1964
1965 pandecode_prop("offset_start = 0x%x", p->offset_start);
1966
1967 if (p->zero5) {
1968 pandecode_msg("XXX: vertex only zero tripped");
1969 pandecode_prop("zero5 = 0x%" PRIx64, p->zero5);
1970 }
1971
1972 MEMORY_PROP(p, position_varying);
1973 MEMORY_PROP(p, occlusion_counter);
1974
1975 pandecode_indent--;
1976 pandecode_log("},\n");
1977 }
1978
1979 static void
1980 pandecode_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1981 {
1982 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1983 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1984
1985 pandecode_log("struct bifrost_tiler_heap_meta tiler_heap_meta_%"PRIx64"_%d = {\n", gpu_va, job_no);
1986 pandecode_indent++;
1987
1988 if (h->zero) {
1989 pandecode_msg("XXX: tiler heap zero tripped\n");
1990 pandecode_prop("zero = 0x%x", h->zero);
1991 }
1992
1993 pandecode_prop("heap_size = 0x%x", h->heap_size);
1994 MEMORY_PROP(h, tiler_heap_start);
1995 MEMORY_PROP(h, tiler_heap_free);
1996
1997 /* this might point to the beginning of another buffer, when it's
1998 * really the end of the tiler heap buffer, so we have to be careful
1999 * here. but for zero length, we need the same pointer.
2000 */
2001
2002 if (h->tiler_heap_end == h->tiler_heap_start) {
2003 MEMORY_PROP(h, tiler_heap_start);
2004 } else {
2005 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
2006 pandecode_prop("tiler_heap_end = %s + 1", a);
2007 free(a);
2008 }
2009
2010 for (int i = 0; i < 10; i++) {
2011 if (h->zeros[i] != 0) {
2012 pandecode_msg("XXX: tiler heap zero %d tripped, value %x\n",
2013 i, h->zeros[i]);
2014 }
2015 }
2016
2017 if (h->unk1 != 0x1) {
2018 pandecode_msg("XXX: tiler heap unk1 tripped\n");
2019 pandecode_prop("unk1 = 0x%x", h->unk1);
2020 }
2021
2022 if (h->unk7e007e != 0x7e007e) {
2023 pandecode_msg("XXX: tiler heap unk7e007e tripped\n");
2024 pandecode_prop("unk7e007e = 0x%x", h->unk7e007e);
2025 }
2026
2027 pandecode_indent--;
2028 pandecode_log("};\n");
2029 }
2030
2031 static void
2032 pandecode_tiler_meta(mali_ptr gpu_va, int job_no)
2033 {
2034 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
2035 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
2036
2037 pandecode_tiler_heap_meta(t->tiler_heap_meta, job_no);
2038
2039 pandecode_log("struct bifrost_tiler_meta tiler_meta_%"PRIx64"_%d = {\n", gpu_va, job_no);
2040 pandecode_indent++;
2041
2042 pandecode_prop("tiler_heap_next_start = 0x%" PRIx32, t->tiler_heap_next_start);
2043 pandecode_prop("used_hierarchy_mask = 0x%" PRIx32, t->used_hierarchy_mask);
2044
2045 if (t->hierarchy_mask != 0xa &&
2046 t->hierarchy_mask != 0x14 &&
2047 t->hierarchy_mask != 0x28 &&
2048 t->hierarchy_mask != 0x50 &&
2049 t->hierarchy_mask != 0xa0)
2050 pandecode_prop("XXX: Unexpected hierarchy_mask (not 0xa, 0x14, 0x28, 0x50 or 0xa0)!");
2051
2052 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
2053
2054 pandecode_prop("flags = 0x%" PRIx16, t->flags);
2055
2056 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
2057 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
2058
2059 if (t->zero0) {
2060 pandecode_msg("XXX: tiler meta zero tripped\n");
2061 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
2062 }
2063
2064 for (int i = 0; i < 12; i++) {
2065 if (t->zeros[i] != 0) {
2066 pandecode_msg("XXX: tiler heap zero %d tripped, value %" PRIx64 "\n",
2067 i, t->zeros[i]);
2068 }
2069 }
2070
2071 pandecode_indent--;
2072 pandecode_log("};\n");
2073 }
2074
2075 static void
2076 pandecode_primitive_size(union midgard_primitive_size u, bool constant)
2077 {
2078 if (u.pointer == 0x0)
2079 return;
2080
2081 pandecode_log(".primitive_size = {\n");
2082 pandecode_indent++;
2083
2084 if (constant) {
2085 pandecode_prop("constant = %f", u.constant);
2086 } else {
2087 MEMORY_PROP((&u), pointer);
2088 }
2089
2090 pandecode_indent--;
2091 pandecode_log("},\n");
2092 }
2093
2094 static void
2095 pandecode_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
2096 {
2097 pandecode_log_cont("{\n");
2098 pandecode_indent++;
2099
2100 /* TODO: gl_PointSize on Bifrost */
2101 pandecode_primitive_size(t->primitive_size, true);
2102
2103 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
2104 || t->zero6) {
2105 pandecode_msg("XXX: tiler only zero tripped\n");
2106 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2107 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
2108 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
2109 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
2110 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
2111 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
2112 }
2113
2114 pandecode_indent--;
2115 pandecode_log("},\n");
2116 }
2117
2118 static int
2119 pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
2120 const struct pandecode_mapped_memory *mem,
2121 mali_ptr payload, int job_no, unsigned gpu_id)
2122 {
2123 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
2124
2125 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true, gpu_id);
2126
2127 pandecode_log("struct bifrost_payload_vertex payload_%"PRIx64"_%d = {\n", payload, job_no);
2128 pandecode_indent++;
2129
2130 pandecode_vertex_tiler_prefix(&v->prefix, job_no, false);
2131 pandecode_vertex_tiler_postfix(&v->postfix, job_no, true);
2132
2133 pandecode_indent--;
2134 pandecode_log("};\n");
2135
2136 return sizeof(*v);
2137 }
2138
2139 static int
2140 pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
2141 const struct pandecode_mapped_memory *mem,
2142 mali_ptr payload, int job_no, unsigned gpu_id)
2143 {
2144 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
2145
2146 pandecode_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true, gpu_id);
2147 pandecode_tiler_meta(t->tiler.tiler_meta, job_no);
2148
2149 pandecode_log("struct bifrost_payload_tiler payload_%"PRIx64"_%d = {\n", payload, job_no);
2150 pandecode_indent++;
2151
2152 pandecode_vertex_tiler_prefix(&t->prefix, job_no, false);
2153
2154 pandecode_log(".tiler = ");
2155 pandecode_tiler_only_bfr(&t->tiler, job_no);
2156
2157 pandecode_vertex_tiler_postfix(&t->postfix, job_no, true);
2158
2159 pandecode_indent--;
2160 pandecode_log("};\n");
2161
2162 return sizeof(*t);
2163 }
2164
2165 static int
2166 pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
2167 const struct pandecode_mapped_memory *mem,
2168 mali_ptr payload, int job_no, unsigned gpu_id)
2169 {
2170 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
2171 bool is_graphics = (h->job_type == MALI_JOB_TYPE_VERTEX) || (h->job_type == MALI_JOB_TYPE_TILER);
2172
2173 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false, gpu_id);
2174
2175 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
2176 pandecode_indent++;
2177
2178 pandecode_vertex_tiler_prefix(&v->prefix, job_no, is_graphics);
2179 pandecode_vertex_tiler_postfix(&v->postfix, job_no, false);
2180
2181 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
2182 pandecode_primitive_size(v->primitive_size, !has_primitive_pointer);
2183
2184 pandecode_indent--;
2185 pandecode_log("};\n");
2186
2187 return sizeof(*v);
2188 }
2189
2190 static int
2191 pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
2192 mali_ptr payload, int job_no,
2193 bool is_bifrost, unsigned gpu_id)
2194 {
2195 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
2196
2197 bool is_mfbd = s->framebuffer & MALI_MFBD;
2198
2199 if (!is_mfbd && is_bifrost)
2200 pandecode_msg("XXX: Bifrost fragment must use MFBD\n");
2201
2202 struct pandecode_fbd info;
2203
2204 if (is_mfbd)
2205 info = pandecode_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true, false, is_bifrost);
2206 else
2207 info = pandecode_sfbd(s->framebuffer & FBD_MASK, job_no, true, gpu_id);
2208
2209 /* Compute the tag for the tagged pointer. This contains the type of
2210 * FBD (MFBD/SFBD), and in the case of an MFBD, information about which
2211 * additional structures follow the MFBD header (an extra payload or
2212 * not, as well as a count of render targets) */
2213
2214 unsigned expected_tag = is_mfbd ? MALI_MFBD : 0;
2215
2216 if (is_mfbd) {
2217 if (info.has_extra)
2218 expected_tag |= MALI_MFBD_TAG_EXTRA;
2219
2220 expected_tag |= (MALI_POSITIVE(info.rt_count) << 2);
2221 }
2222
2223 if ((s->min_tile_coord | s->max_tile_coord) & ~(MALI_X_COORD_MASK | MALI_Y_COORD_MASK)) {
2224 pandecode_msg("XXX: unexpected tile coordinate bits\n");
2225 pandecode_prop("min_tile_coord = 0x%X\n", s->min_tile_coord);
2226 pandecode_prop("max_tile_coord = 0x%X\n", s->max_tile_coord);
2227 }
2228
2229 /* Extract tile coordinates */
2230
2231 unsigned min_x = MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT;
2232 unsigned min_y = MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT;
2233
2234 unsigned max_x = (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT;
2235 unsigned max_y = (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT;
2236
2237 /* For the max, we also want the floored (rather than ceiled) version for checking */
2238
2239 unsigned max_x_f = (MALI_TILE_COORD_X(s->max_tile_coord)) << MALI_TILE_SHIFT;
2240 unsigned max_y_f = (MALI_TILE_COORD_Y(s->max_tile_coord)) << MALI_TILE_SHIFT;
2241
2242 /* Validate the coordinates are well-ordered */
2243
2244 if (min_x == max_x)
2245 pandecode_msg("XXX: empty X coordinates (%u = %u)\n", min_x, max_x);
2246 else if (min_x > max_x)
2247 pandecode_msg("XXX: misordered X coordinates (%u > %u)\n", min_x, max_x);
2248
2249 if (min_y == max_y)
2250 pandecode_msg("XXX: empty X coordinates (%u = %u)\n", min_x, max_x);
2251 else if (min_y > max_y)
2252 pandecode_msg("XXX: misordered X coordinates (%u > %u)\n", min_x, max_x);
2253
2254 /* Validate the coordinates fit inside the framebuffer. We use floor,
2255 * rather than ceil, for the max coordinates, since the tile
2256 * coordinates for something like an 800x600 framebuffer will actually
2257 * resolve to 800x608, which would otherwise trigger a Y-overflow */
2258
2259 if ((min_x > info.width) || (max_x_f > info.width))
2260 pandecode_msg("XXX: tile coordinates overflow in X direction\n");
2261
2262 if ((min_y > info.height) || (max_y_f > info.height))
2263 pandecode_msg("XXX: tile coordinates overflow in Y direction\n");
2264
2265 /* After validation, we print */
2266
2267 pandecode_log("fragment (%u, %u) ... (%u, %u)\n\n", min_x, min_y, max_x, max_y);
2268
2269 /* The FBD is a tagged pointer */
2270
2271 unsigned tag = (s->framebuffer & ~FBD_MASK);
2272
2273 if (tag != expected_tag)
2274 pandecode_msg("XXX: expected FBD tag %X but got %X\n", expected_tag, tag);
2275
2276 return sizeof(*s);
2277 }
2278
2279 /* Entrypoint to start tracing. jc_gpu_va is the GPU address for the first job
2280 * in the chain; later jobs are found by walking the chain. Bifrost is, well,
2281 * if it's bifrost or not. GPU ID is the more finegrained ID (at some point, we
2282 * might wish to combine this with the bifrost parameter) because some details
2283 * are model-specific even within a particular architecture. Minimal traces
2284 * *only* examine the job descriptors, skipping printing entirely if there is
2285 * no faults, and only descends into the payload if there are faults. This is
2286 * useful for looking for faults without the overhead of invasive traces. */
2287
2288 void
2289 pandecode_jc(mali_ptr jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal)
2290 {
2291 pandecode_dump_file_open();
2292
2293 struct mali_job_descriptor_header *h;
2294 unsigned job_descriptor_number = 0;
2295
2296 do {
2297 struct pandecode_mapped_memory *mem =
2298 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2299
2300 void *payload;
2301
2302 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2303
2304 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h);
2305 payload = pandecode_fetch_gpu_mem(mem, payload_ptr, 256);
2306
2307 int job_no = job_descriptor_number++;
2308
2309 /* If the job is good to go, skip it in minimal mode */
2310 if (minimal && (h->exception_status == 0x0 || h->exception_status == 0x1))
2311 continue;
2312
2313 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
2314 pandecode_indent++;
2315
2316 pandecode_prop("job_type = %s", mali_job_type_as_str(h->job_type));
2317
2318 if (h->job_descriptor_size)
2319 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2320
2321 if (h->exception_status && h->exception_status != 0x1)
2322 pandecode_prop("exception_status = %x (source ID: 0x%x access: %s exception: 0x%x)",
2323 h->exception_status,
2324 (h->exception_status >> 16) & 0xFFFF,
2325 mali_exception_access_as_str((h->exception_status >> 8) & 0x3),
2326 h->exception_status & 0xFF);
2327
2328 if (h->first_incomplete_task)
2329 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2330
2331 if (h->fault_pointer)
2332 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2333
2334 if (h->job_barrier)
2335 pandecode_prop("job_barrier = %d", h->job_barrier);
2336
2337 pandecode_prop("job_index = %d", h->job_index);
2338
2339 if (h->unknown_flags)
2340 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2341
2342 if (h->job_dependency_index_1)
2343 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2344
2345 if (h->job_dependency_index_2)
2346 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2347
2348 pandecode_indent--;
2349 pandecode_log("};\n");
2350
2351 switch (h->job_type) {
2352 case MALI_JOB_TYPE_WRITE_VALUE: {
2353 struct mali_payload_write_value *s = payload;
2354 pandecode_log("struct mali_payload_write_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
2355 pandecode_indent++;
2356 MEMORY_PROP(s, address);
2357
2358 if (s->value_descriptor != MALI_WRITE_VALUE_ZERO) {
2359 pandecode_msg("XXX: unknown value descriptor\n");
2360 pandecode_prop("value_descriptor = 0x%" PRIX32, s->value_descriptor);
2361 }
2362
2363 if (s->reserved) {
2364 pandecode_msg("XXX: set value tripped\n");
2365 pandecode_prop("reserved = 0x%" PRIX32, s->reserved);
2366 }
2367
2368 pandecode_prop("immediate = 0x%" PRIX64, s->immediate);
2369 pandecode_indent--;
2370 pandecode_log("};\n");
2371
2372 break;
2373 }
2374
2375 case MALI_JOB_TYPE_TILER:
2376 case MALI_JOB_TYPE_VERTEX:
2377 case MALI_JOB_TYPE_COMPUTE:
2378 if (bifrost) {
2379 if (h->job_type == MALI_JOB_TYPE_TILER)
2380 pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
2381 else
2382 pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
2383 } else
2384 pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no, gpu_id);
2385
2386 break;
2387
2388 case MALI_JOB_TYPE_FRAGMENT:
2389 pandecode_fragment_job(mem, payload_ptr, job_no, bifrost, gpu_id);
2390 break;
2391
2392 default:
2393 break;
2394 }
2395 } while ((jc_gpu_va = h->next_job));
2396
2397 pandecode_map_read_write();
2398 }