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 <genxml/gen_macros.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <memory.h>
30 #include <stdbool.h>
31 #include <stdarg.h>
32 #include <errno.h>
33 #include <ctype.h>
34 #include "decode.h"
35
36 #include "midgard/disassemble.h"
37 #include "bifrost/disassemble.h"
38 #include "bifrost/valhall/disassemble.h"
39
40 #define DUMP_UNPACKED(T, var, ...) { \
41 pandecode_log(__VA_ARGS__); \
42 pan_print(pandecode_dump_stream, T, var, (pandecode_indent + 1) * 2); \
43 }
44
45 #define DUMP_CL(T, cl, ...) {\
46 pan_unpack(cl, T, temp); \
47 DUMP_UNPACKED(T, temp, __VA_ARGS__); \
48 }
49
50 #define DUMP_SECTION(A, S, cl, ...) { \
51 pan_section_unpack(cl, A, S, temp); \
52 pandecode_log(__VA_ARGS__); \
53 pan_section_print(pandecode_dump_stream, A, S, temp, (pandecode_indent + 1) * 2); \
54 }
55
56 #define MAP_ADDR(T, addr, cl) \
57 const uint8_t *cl = 0; \
58 { \
59 struct pandecode_mapped_memory *mapped_mem = pandecode_find_mapped_gpu_mem_containing(addr); \
60 cl = pandecode_fetch_gpu_mem(mapped_mem, addr, pan_size(T)); \
61 }
62
63 #define DUMP_ADDR(T, addr, ...) {\
64 MAP_ADDR(T, addr, cl) \
65 DUMP_CL(T, cl, __VA_ARGS__); \
66 }
67
68 /* Semantic logging type.
69 *
70 * Raw: for raw messages to be printed as is.
71 * Message: for helpful information to be commented out in replays.
72 *
73 * Use one of pandecode_log or pandecode_msg as syntax sugar.
74 */
75
76 enum pandecode_log_type {
77 PANDECODE_RAW,
78 PANDECODE_MESSAGE,
79 };
80
81 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
82 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
83
84 static unsigned pandecode_indent = 0;
85
86 static void
pandecode_make_indent(void)87 pandecode_make_indent(void)
88 {
89 for (unsigned i = 0; i < pandecode_indent; ++i)
90 fprintf(pandecode_dump_stream, " ");
91 }
92
93 static void PRINTFLIKE(2, 3)
pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)94 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
95 {
96 va_list ap;
97
98 pandecode_make_indent();
99
100 if (type == PANDECODE_MESSAGE)
101 fprintf(pandecode_dump_stream, "// ");
102
103 va_start(ap, format);
104 vfprintf(pandecode_dump_stream, format, ap);
105 va_end(ap);
106 }
107
108 static void
pandecode_log_cont(const char *format, ...)109 pandecode_log_cont(const char *format, ...)
110 {
111 va_list ap;
112
113 va_start(ap, format);
114 vfprintf(pandecode_dump_stream, format, ap);
115 va_end(ap);
116 }
117
118 /* To check for memory safety issues, validates that the given pointer in GPU
119 * memory is valid, containing at least sz bytes. The goal is to eliminate
120 * GPU-side memory bugs (NULL pointer dereferences, buffer overflows, or buffer
121 * overruns) by statically validating pointers.
122 */
123
124 static void
pandecode_validate_buffer(mali_ptr addr, size_t sz)125 pandecode_validate_buffer(mali_ptr addr, size_t sz)
126 {
127 if (!addr) {
128 pandecode_msg("XXX: null pointer deref\n");
129 return;
130 }
131
132 /* Find a BO */
133
134 struct pandecode_mapped_memory *bo =
135 pandecode_find_mapped_gpu_mem_containing(addr);
136
137 if (!bo) {
138 pandecode_msg("XXX: invalid memory dereference\n");
139 return;
140 }
141
142 /* Bounds check */
143
144 unsigned offset = addr - bo->gpu_va;
145 unsigned total = offset + sz;
146
147 if (total > bo->length) {
148 pandecode_msg("XXX: buffer overrun. "
149 "Chunk of size %zu at offset %d in buffer of size %zu. "
150 "Overrun by %zu bytes. \n",
151 sz, offset, bo->length, total - bo->length);
152 return;
153 }
154 }
155
156 #if PAN_ARCH <= 5
157 /* Midgard's tiler descriptor is embedded within the
158 * larger FBD */
159
160 static void
pandecode_midgard_tiler_descriptor( const struct mali_tiler_context_packed *tp, const struct mali_tiler_weights_packed *wp)161 pandecode_midgard_tiler_descriptor(
162 const struct mali_tiler_context_packed *tp,
163 const struct mali_tiler_weights_packed *wp)
164 {
165 pan_unpack(tp, TILER_CONTEXT, t);
166 DUMP_UNPACKED(TILER_CONTEXT, t, "Tiler:\n");
167
168 /* We've never seen weights used in practice, but they exist */
169 pan_unpack(wp, TILER_WEIGHTS, w);
170 bool nonzero_weights = false;
171
172 nonzero_weights |= w.weight0 != 0x0;
173 nonzero_weights |= w.weight1 != 0x0;
174 nonzero_weights |= w.weight2 != 0x0;
175 nonzero_weights |= w.weight3 != 0x0;
176 nonzero_weights |= w.weight4 != 0x0;
177 nonzero_weights |= w.weight5 != 0x0;
178 nonzero_weights |= w.weight6 != 0x0;
179 nonzero_weights |= w.weight7 != 0x0;
180
181 if (nonzero_weights)
182 DUMP_UNPACKED(TILER_WEIGHTS, w, "Tiler Weights:\n");
183 }
184 #endif
185
186 /* Information about the framebuffer passed back for
187 * additional analysis */
188
189 struct pandecode_fbd {
190 unsigned width;
191 unsigned height;
192 unsigned rt_count;
193 bool has_extra;
194 };
195
196 #if PAN_ARCH == 4
197 static struct pandecode_fbd
pandecode_sfbd(uint64_t gpu_va, int job_no, bool is_fragment, unsigned gpu_id)198 pandecode_sfbd(uint64_t gpu_va, int job_no, bool is_fragment, unsigned gpu_id)
199 {
200 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
201 const void *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
202
203 struct pandecode_fbd info = {
204 .has_extra = false,
205 .rt_count = 1
206 };
207
208 pandecode_log("Framebuffer:\n");
209 pandecode_indent++;
210
211 DUMP_SECTION(FRAMEBUFFER, LOCAL_STORAGE, s, "Local Storage:\n");
212 pan_section_unpack(s, FRAMEBUFFER, PARAMETERS, p);
213 DUMP_UNPACKED(FRAMEBUFFER_PARAMETERS, p, "Parameters:\n");
214
215 const void *t = pan_section_ptr(s, FRAMEBUFFER, TILER);
216 const void *w = pan_section_ptr(s, FRAMEBUFFER, TILER_WEIGHTS);
217
218 pandecode_midgard_tiler_descriptor(t, w);
219
220 pandecode_indent--;
221
222 /* Dummy unpack of the padding section to make sure all words are 0.
223 * No need to call print here since the section is supposed to be empty.
224 */
225 pan_section_unpack(s, FRAMEBUFFER, PADDING_1, padding1);
226 pan_section_unpack(s, FRAMEBUFFER, PADDING_2, padding2);
227 pandecode_log("\n");
228
229 return info;
230 }
231 #endif
232
233 #if PAN_ARCH >= 5
234 static void
pandecode_local_storage(uint64_t gpu_va, int job_no)235 pandecode_local_storage(uint64_t gpu_va, int job_no)
236 {
237 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
238 const struct mali_local_storage_packed *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
239 DUMP_CL(LOCAL_STORAGE, s, "Local Storage:\n");
240 }
241
242 static void
pandecode_render_target(uint64_t gpu_va, unsigned job_no, unsigned gpu_id, const struct MALI_FRAMEBUFFER_PARAMETERS *fb)243 pandecode_render_target(uint64_t gpu_va, unsigned job_no, unsigned gpu_id,
244 const struct MALI_FRAMEBUFFER_PARAMETERS *fb)
245 {
246 pandecode_log("Color Render Targets:\n");
247 pandecode_indent++;
248
249 for (int i = 0; i < (fb->render_target_count); i++) {
250 mali_ptr rt_va = gpu_va + i * pan_size(RENDER_TARGET);
251 struct pandecode_mapped_memory *mem =
252 pandecode_find_mapped_gpu_mem_containing(rt_va);
253 const struct mali_render_target_packed *PANDECODE_PTR_VAR(rtp, mem, (mali_ptr) rt_va);
254 DUMP_CL(RENDER_TARGET, rtp, "Color Render Target %d:\n", i);
255 }
256
257 pandecode_indent--;
258 pandecode_log("\n");
259 }
260 #endif
261
262 #if PAN_ARCH >= 6
263 static void
pandecode_sample_locations(const void *fb, int job_no)264 pandecode_sample_locations(const void *fb, int job_no)
265 {
266 pan_section_unpack(fb, FRAMEBUFFER, PARAMETERS, params);
267
268 struct pandecode_mapped_memory *smem =
269 pandecode_find_mapped_gpu_mem_containing(params.sample_locations);
270
271 const u16 *PANDECODE_PTR_VAR(samples, smem, params.sample_locations);
272
273 pandecode_log("Sample locations:\n");
274 for (int i = 0; i < 33; i++) {
275 pandecode_log(" (%d, %d),\n",
276 samples[2 * i] - 128,
277 samples[2 * i + 1] - 128);
278 }
279 }
280 #endif
281
282 static void
283 pandecode_dcd(const struct MALI_DRAW *p,
284 int job_no, enum mali_job_type job_type,
285 char *suffix, unsigned gpu_id);
286
287 #if PAN_ARCH >= 5
288 static struct pandecode_fbd
pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool is_fragment, unsigned gpu_id)289 pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool is_fragment, unsigned gpu_id)
290 {
291 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
292 const void *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
293 pan_section_unpack(fb, FRAMEBUFFER, PARAMETERS, params);
294
295 struct pandecode_fbd info;
296
297 #if PAN_ARCH >= 6
298 pandecode_sample_locations(fb, job_no);
299
300 pan_section_unpack(fb, FRAMEBUFFER, PARAMETERS, bparams);
301 unsigned dcd_size = pan_size(DRAW);
302 struct pandecode_mapped_memory *dcdmem =
303 pandecode_find_mapped_gpu_mem_containing(bparams.frame_shader_dcds);
304
305 if (bparams.pre_frame_0 != MALI_PRE_POST_FRAME_SHADER_MODE_NEVER) {
306 const void *PANDECODE_PTR_VAR(dcd, dcdmem, bparams.frame_shader_dcds + (0 * dcd_size));
307 pan_unpack(dcd, DRAW, draw);
308 pandecode_log("Pre frame 0:\n");
309 pandecode_dcd(&draw, job_no, MALI_JOB_TYPE_FRAGMENT, "", gpu_id);
310 }
311
312 if (bparams.pre_frame_1 != MALI_PRE_POST_FRAME_SHADER_MODE_NEVER) {
313 const void *PANDECODE_PTR_VAR(dcd, dcdmem, bparams.frame_shader_dcds + (1 * dcd_size));
314 pan_unpack(dcd, DRAW, draw);
315 pandecode_log("Pre frame 1:\n");
316 pandecode_dcd(&draw, job_no, MALI_JOB_TYPE_FRAGMENT, "", gpu_id);
317 }
318
319 if (bparams.post_frame != MALI_PRE_POST_FRAME_SHADER_MODE_NEVER) {
320 const void *PANDECODE_PTR_VAR(dcd, dcdmem, bparams.frame_shader_dcds + (2 * dcd_size));
321 pan_unpack(dcd, DRAW, draw);
322 pandecode_log("Post frame:\n");
323 pandecode_dcd(&draw, job_no, MALI_JOB_TYPE_FRAGMENT, "", gpu_id);
324 }
325 #endif
326
327 pandecode_log("Multi-Target Framebuffer:\n");
328 pandecode_indent++;
329
330 #if PAN_ARCH <= 5
331 DUMP_SECTION(FRAMEBUFFER, LOCAL_STORAGE, fb, "Local Storage:\n");
332 #endif
333
334 info.width = params.width;
335 info.height = params.height;
336 info.rt_count = params.render_target_count;
337 DUMP_UNPACKED(FRAMEBUFFER_PARAMETERS, params, "Parameters:\n");
338
339 #if PAN_ARCH <= 5
340 const void *t = pan_section_ptr(fb, FRAMEBUFFER, TILER);
341 const void *w = pan_section_ptr(fb, FRAMEBUFFER, TILER_WEIGHTS);
342 pandecode_midgard_tiler_descriptor(t, w);
343 #endif
344
345 pandecode_indent--;
346 pandecode_log("\n");
347
348 gpu_va += pan_size(FRAMEBUFFER);
349
350 info.has_extra = params.has_zs_crc_extension;
351
352 if (info.has_extra) {
353 struct pandecode_mapped_memory *mem =
354 pandecode_find_mapped_gpu_mem_containing(gpu_va);
355 const struct mali_zs_crc_extension_packed *PANDECODE_PTR_VAR(zs_crc, mem, (mali_ptr)gpu_va);
356 DUMP_CL(ZS_CRC_EXTENSION, zs_crc, "ZS CRC Extension:\n");
357 pandecode_log("\n");
358
359 gpu_va += pan_size(ZS_CRC_EXTENSION);
360 }
361
362 if (is_fragment)
363 pandecode_render_target(gpu_va, job_no, gpu_id, ¶ms);
364
365 return info;
366 }
367 #endif
368
369 #if PAN_ARCH <= 7
370 static void
pandecode_attributes(const struct pandecode_mapped_memory *mem, mali_ptr addr, int job_no, char *suffix, int count, bool varying, enum mali_job_type job_type)371 pandecode_attributes(const struct pandecode_mapped_memory *mem,
372 mali_ptr addr, int job_no, char *suffix,
373 int count, bool varying, enum mali_job_type job_type)
374 {
375 char *prefix = varying ? "Varying" : "Attribute";
376 assert(addr);
377
378 if (!count) {
379 pandecode_msg("warn: No %s records\n", prefix);
380 return;
381 }
382
383 MAP_ADDR(ATTRIBUTE_BUFFER, addr, cl);
384
385 for (int i = 0; i < count; ++i) {
386 pan_unpack(cl + i * pan_size(ATTRIBUTE_BUFFER), ATTRIBUTE_BUFFER, temp);
387 DUMP_UNPACKED(ATTRIBUTE_BUFFER, temp, "%s:\n", prefix);
388
389 switch (temp.type) {
390 case MALI_ATTRIBUTE_TYPE_1D_NPOT_DIVISOR_WRITE_REDUCTION:
391 case MALI_ATTRIBUTE_TYPE_1D_NPOT_DIVISOR: {
392 pan_unpack(cl + (i + 1) * pan_size(ATTRIBUTE_BUFFER),
393 ATTRIBUTE_BUFFER_CONTINUATION_NPOT, temp2);
394 pan_print(pandecode_dump_stream, ATTRIBUTE_BUFFER_CONTINUATION_NPOT,
395 temp2, (pandecode_indent + 1) * 2);
396 i++;
397 break;
398 }
399 case MALI_ATTRIBUTE_TYPE_3D_LINEAR:
400 case MALI_ATTRIBUTE_TYPE_3D_INTERLEAVED: {
401 pan_unpack(cl + (i + 1) * pan_size(ATTRIBUTE_BUFFER_CONTINUATION_3D),
402 ATTRIBUTE_BUFFER_CONTINUATION_3D, temp2);
403 pan_print(pandecode_dump_stream, ATTRIBUTE_BUFFER_CONTINUATION_3D,
404 temp2, (pandecode_indent + 1) * 2);
405 i++;
406 break;
407 }
408 default:
409 break;
410 }
411 }
412 pandecode_log("\n");
413 }
414 #endif
415
416 #if PAN_ARCH >= 6
417 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
418
419 static mali_ptr
pandecode_bifrost_blend(void *descs, int job_no, int rt_no, mali_ptr frag_shader)420 pandecode_bifrost_blend(void *descs, int job_no, int rt_no, mali_ptr frag_shader)
421 {
422 pan_unpack(descs + (rt_no * pan_size(BLEND)), BLEND, b);
423 DUMP_UNPACKED(BLEND, b, "Blend RT %d:\n", rt_no);
424 if (b.internal.mode != MALI_BLEND_MODE_SHADER)
425 return 0;
426
427 return (frag_shader & 0xFFFFFFFF00000000ULL) | b.internal.shader.pc;
428 }
429 #elif PAN_ARCH == 5
430 static mali_ptr
pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)431 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
432 {
433 pan_unpack(descs + (rt_no * pan_size(BLEND)), BLEND, b);
434 DUMP_UNPACKED(BLEND, b, "Blend RT %d:\n", rt_no);
435 return b.blend_shader ? (b.shader_pc & ~0xf) : 0;
436 }
437 #endif
438
439 #if PAN_ARCH <= 7
440 static unsigned
pandecode_attribute_meta(int count, mali_ptr attribute, bool varying)441 pandecode_attribute_meta(int count, mali_ptr attribute, bool varying)
442 {
443 unsigned max = 0;
444
445 for (int i = 0; i < count; ++i, attribute += pan_size(ATTRIBUTE)) {
446 MAP_ADDR(ATTRIBUTE, attribute, cl);
447 pan_unpack(cl, ATTRIBUTE, a);
448 DUMP_UNPACKED(ATTRIBUTE, a, "%s:\n", varying ? "Varying" : "Attribute");
449 max = MAX2(max, a.buffer_index);
450 }
451
452 pandecode_log("\n");
453 return MIN2(max + 1, 256);
454 }
455
456 /* return bits [lo, hi) of word */
457 static u32
bits(u32 word, u32 lo, u32 hi)458 bits(u32 word, u32 lo, u32 hi)
459 {
460 if (hi - lo >= 32)
461 return word; // avoid undefined behavior with the shift
462
463 if (lo >= 32)
464 return 0;
465
466 return (word >> lo) & ((1 << (hi - lo)) - 1);
467 }
468
469 static void
pandecode_invocation(const void *i)470 pandecode_invocation(const void *i)
471 {
472 /* Decode invocation_count. See the comment before the definition of
473 * invocation_count for an explanation.
474 */
475 pan_unpack(i, INVOCATION, invocation);
476
477 unsigned size_x = bits(invocation.invocations, 0, invocation.size_y_shift) + 1;
478 unsigned size_y = bits(invocation.invocations, invocation.size_y_shift, invocation.size_z_shift) + 1;
479 unsigned size_z = bits(invocation.invocations, invocation.size_z_shift, invocation.workgroups_x_shift) + 1;
480
481 unsigned groups_x = bits(invocation.invocations, invocation.workgroups_x_shift, invocation.workgroups_y_shift) + 1;
482 unsigned groups_y = bits(invocation.invocations, invocation.workgroups_y_shift, invocation.workgroups_z_shift) + 1;
483 unsigned groups_z = bits(invocation.invocations, invocation.workgroups_z_shift, 32) + 1;
484
485 pandecode_log("Invocation (%d, %d, %d) x (%d, %d, %d)\n",
486 size_x, size_y, size_z,
487 groups_x, groups_y, groups_z);
488
489 DUMP_UNPACKED(INVOCATION, invocation, "Invocation:\n")
490 }
491 #endif
492
493 static void
pandecode_primitive(const void *p)494 pandecode_primitive(const void *p)
495 {
496 pan_unpack(p, PRIMITIVE, primitive);
497 DUMP_UNPACKED(PRIMITIVE, primitive, "Primitive:\n");
498
499 #if PAN_ARCH <= 7
500 /* Validate an index buffer is present if we need one. TODO: verify
501 * relationship between invocation_count and index_count */
502
503 if (primitive.indices) {
504 /* Grab the size */
505 unsigned size = (primitive.index_type == MALI_INDEX_TYPE_UINT32) ?
506 sizeof(uint32_t) : primitive.index_type;
507
508 /* Ensure we got a size, and if so, validate the index buffer
509 * is large enough to hold a full set of indices of the given
510 * size */
511
512 if (!size)
513 pandecode_msg("XXX: index size missing\n");
514 else
515 pandecode_validate_buffer(primitive.indices, primitive.index_count * size);
516 } else if (primitive.index_type)
517 pandecode_msg("XXX: unexpected index size\n");
518 #endif
519 }
520
521 static void
pandecode_primitive_size(const void *s, bool constant)522 pandecode_primitive_size(const void *s, bool constant)
523 {
524 pan_unpack(s, PRIMITIVE_SIZE, ps);
525 if (ps.size_array == 0x0)
526 return;
527
528 DUMP_UNPACKED(PRIMITIVE_SIZE, ps, "Primitive Size:\n")
529 }
530
531 #if PAN_ARCH <= 7
532 static void
pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)533 pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
534 {
535 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
536 uint64_t *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
537
538 for (int i = 0; i < ubufs_count; i++) {
539 mali_ptr addr = (ubufs[i] >> 10) << 2;
540 unsigned size = addr ? (((ubufs[i] & ((1 << 10) - 1)) + 1) * 16) : 0;
541
542 pandecode_validate_buffer(addr, size);
543
544 char *ptr = pointer_as_memory_reference(addr);
545 pandecode_log("ubuf_%d[%u] = %s;\n", i, size, ptr);
546 free(ptr);
547 }
548
549 pandecode_log("\n");
550 }
551
552 static void
pandecode_uniforms(mali_ptr uniforms, unsigned uniform_count)553 pandecode_uniforms(mali_ptr uniforms, unsigned uniform_count)
554 {
555 pandecode_validate_buffer(uniforms, uniform_count * 16);
556
557 char *ptr = pointer_as_memory_reference(uniforms);
558 pandecode_log("vec4 uniforms[%u] = %s;\n", uniform_count, ptr);
559 free(ptr);
560 pandecode_log("\n");
561 }
562 #endif
563
564 static const char *
shader_type_for_job(unsigned type)565 shader_type_for_job(unsigned type)
566 {
567 switch (type) {
568 #if PAN_ARCH <= 7
569 case MALI_JOB_TYPE_VERTEX: return "VERTEX";
570 #endif
571 case MALI_JOB_TYPE_TILER: return "FRAGMENT";
572 case MALI_JOB_TYPE_FRAGMENT: return "FRAGMENT";
573 case MALI_JOB_TYPE_COMPUTE: return "COMPUTE";
574 default: return "UNKNOWN";
575 }
576 }
577
578 static unsigned shader_id = 0;
579
580 static struct midgard_disasm_stats
pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type, unsigned gpu_id)581 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
582 unsigned gpu_id)
583 {
584 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
585 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
586
587 /* Compute maximum possible size */
588 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
589
590 /* Print some boilerplate to clearly denote the assembly (which doesn't
591 * obey indentation rules), and actually do the disassembly! */
592
593 pandecode_log_cont("\n\n");
594
595 struct midgard_disasm_stats stats = { 0 };
596
597 #if PAN_ARCH >= 9
598 disassemble_valhall(pandecode_dump_stream, (const uint64_t *) code, sz, true);
599 #elif PAN_ARCH >= 6 && PAN_ARCH <= 7
600 disassemble_bifrost(pandecode_dump_stream, code, sz, false);
601 #else
602 stats = disassemble_midgard(pandecode_dump_stream,
603 code, sz, gpu_id, true);
604 #endif
605
606 unsigned nr_threads =
607 (stats.work_count <= 4) ? 4 :
608 (stats.work_count <= 8) ? 2 :
609 1;
610
611 pandecode_log_cont("shader%d - MESA_SHADER_%s shader: "
612 "%u inst, %u bundles, %u quadwords, "
613 "%u registers, %u threads, 0 loops, 0:0 spills:fills\n\n\n",
614 shader_id++,
615 shader_type_for_job(type),
616 stats.instruction_count, stats.bundle_count, stats.quadword_count,
617 stats.work_count, nr_threads);
618
619 return stats;
620 }
621
622 #if PAN_ARCH <= 7
623 static void
pandecode_texture_payload(mali_ptr payload, enum mali_texture_dimension dim, enum mali_texture_layout layout, bool manual_stride, uint8_t levels, uint16_t nr_samples, uint16_t array_size, struct pandecode_mapped_memory *tmem)624 pandecode_texture_payload(mali_ptr payload,
625 enum mali_texture_dimension dim,
626 enum mali_texture_layout layout,
627 bool manual_stride,
628 uint8_t levels,
629 uint16_t nr_samples,
630 uint16_t array_size,
631 struct pandecode_mapped_memory *tmem)
632 {
633 pandecode_log(".payload = {\n");
634 pandecode_indent++;
635
636 /* A bunch of bitmap pointers follow.
637 * We work out the correct number,
638 * based on the mipmap/cubemap
639 * properties, but dump extra
640 * possibilities to futureproof */
641
642 int bitmap_count = levels;
643
644 /* Miptree for each face */
645 if (dim == MALI_TEXTURE_DIMENSION_CUBE)
646 bitmap_count *= 6;
647
648 /* Array of layers */
649 bitmap_count *= nr_samples;
650
651 /* Array of textures */
652 bitmap_count *= array_size;
653
654 /* Stride for each element */
655 if (manual_stride)
656 bitmap_count *= 2;
657
658 mali_ptr *pointers_and_strides = pandecode_fetch_gpu_mem(tmem,
659 payload, sizeof(mali_ptr) * bitmap_count);
660 for (int i = 0; i < bitmap_count; ++i) {
661 /* How we dump depends if this is a stride or a pointer */
662
663 if (manual_stride && (i & 1)) {
664 /* signed 32-bit snuck in as a 64-bit pointer */
665 uint64_t stride_set = pointers_and_strides[i];
666 int32_t row_stride = stride_set;
667 int32_t surface_stride = stride_set >> 32;
668 pandecode_log("(mali_ptr) %d /* surface stride */ %d /* row stride */, \n",
669 surface_stride, row_stride);
670 } else {
671 char *a = pointer_as_memory_reference(pointers_and_strides[i]);
672 pandecode_log("%s, \n", a);
673 free(a);
674 }
675 }
676
677 pandecode_indent--;
678 pandecode_log("},\n");
679 }
680 #endif
681
682 #if PAN_ARCH <= 5
683 static void
pandecode_texture(mali_ptr u, struct pandecode_mapped_memory *tmem, unsigned job_no, unsigned tex)684 pandecode_texture(mali_ptr u,
685 struct pandecode_mapped_memory *tmem,
686 unsigned job_no, unsigned tex)
687 {
688 struct pandecode_mapped_memory *mapped_mem = pandecode_find_mapped_gpu_mem_containing(u);
689 const uint8_t *cl = pandecode_fetch_gpu_mem(mapped_mem, u, pan_size(TEXTURE));
690
691 pan_unpack(cl, TEXTURE, temp);
692 DUMP_UNPACKED(TEXTURE, temp, "Texture:\n")
693
694 pandecode_indent++;
695 unsigned nr_samples = temp.dimension == MALI_TEXTURE_DIMENSION_3D ?
696 1 : temp.sample_count;
697 pandecode_texture_payload(u + pan_size(TEXTURE),
698 temp.dimension, temp.texel_ordering, temp.manual_stride,
699 temp.levels, nr_samples, temp.array_size, mapped_mem);
700 pandecode_indent--;
701 }
702 #else
703 static void
pandecode_bifrost_texture( const void *cl, unsigned job_no, unsigned tex)704 pandecode_bifrost_texture(
705 const void *cl,
706 unsigned job_no,
707 unsigned tex)
708 {
709 pan_unpack(cl, TEXTURE, temp);
710 DUMP_UNPACKED(TEXTURE, temp, "Texture:\n")
711
712 pandecode_indent++;
713
714 #if PAN_ARCH >= 9
715 int plane_count = temp.levels * temp.array_size;
716
717 /* Miptree for each face */
718 if (temp.dimension == MALI_TEXTURE_DIMENSION_CUBE)
719 plane_count *= 6;
720
721 for (unsigned i = 0; i < plane_count; ++i)
722 DUMP_ADDR(PLANE, temp.surfaces + i * pan_size(PLANE), "Plane %u:\n", i);
723 #else
724 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(temp.surfaces);
725 unsigned nr_samples = temp.dimension == MALI_TEXTURE_DIMENSION_3D ?
726 1 : temp.sample_count;
727
728 pandecode_texture_payload(temp.surfaces, temp.dimension, temp.texel_ordering,
729 true, temp.levels, nr_samples, temp.array_size, tmem);
730 #endif
731 pandecode_indent--;
732 }
733 #endif
734
735 #if PAN_ARCH <= 7
736 static void
pandecode_blend_shader_disassemble(mali_ptr shader, int job_no, int job_type, unsigned gpu_id)737 pandecode_blend_shader_disassemble(mali_ptr shader, int job_no, int job_type,
738 unsigned gpu_id)
739 {
740 struct midgard_disasm_stats stats =
741 pandecode_shader_disassemble(shader, job_no, job_type, gpu_id);
742
743 bool has_texture = (stats.texture_count > 0);
744 bool has_sampler = (stats.sampler_count > 0);
745 bool has_attribute = (stats.attribute_count > 0);
746 bool has_varying = (stats.varying_count > 0);
747 bool has_uniform = (stats.uniform_count > 0);
748 bool has_ubo = (stats.uniform_buffer_count > 0);
749
750 if (has_texture || has_sampler)
751 pandecode_msg("XXX: blend shader accessing textures\n");
752
753 if (has_attribute || has_varying)
754 pandecode_msg("XXX: blend shader accessing interstage\n");
755
756 if (has_uniform || has_ubo)
757 pandecode_msg("XXX: blend shader accessing uniforms\n");
758 }
759
760 static void
pandecode_textures(mali_ptr textures, unsigned texture_count, int job_no)761 pandecode_textures(mali_ptr textures, unsigned texture_count, int job_no)
762 {
763 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(textures);
764
765 if (!mmem)
766 return;
767
768 pandecode_log("Textures %"PRIx64"_%d:\n", textures, job_no);
769 pandecode_indent++;
770
771 #if PAN_ARCH >= 6
772 const void *cl =
773 pandecode_fetch_gpu_mem(mmem,
774 textures,
775 pan_size(TEXTURE) *
776 texture_count);
777
778 for (unsigned tex = 0; tex < texture_count; ++tex) {
779 pandecode_bifrost_texture(cl + pan_size(TEXTURE) * tex,
780 job_no, tex);
781 }
782 #else
783 mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures);
784
785 for (int tex = 0; tex < texture_count; ++tex) {
786 mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr));
787 char *a = pointer_as_memory_reference(*u);
788 pandecode_log("%s,\n", a);
789 free(a);
790 }
791
792 /* Now, finally, descend down into the texture descriptor */
793 for (unsigned tex = 0; tex < texture_count; ++tex) {
794 mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr));
795 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
796 if (tmem)
797 pandecode_texture(*u, tmem, job_no, tex);
798 }
799 #endif
800 pandecode_indent--;
801 pandecode_log("\n");
802 }
803
804 static void
pandecode_samplers(mali_ptr samplers, unsigned sampler_count, int job_no)805 pandecode_samplers(mali_ptr samplers, unsigned sampler_count, int job_no)
806 {
807 pandecode_log("Samplers %"PRIx64"_%d:\n", samplers, job_no);
808 pandecode_indent++;
809
810 for (int i = 0; i < sampler_count; ++i)
811 DUMP_ADDR(SAMPLER, samplers + (pan_size(SAMPLER) * i), "Sampler %d:\n", i);
812
813 pandecode_indent--;
814 pandecode_log("\n");
815 }
816
817 static void
pandecode_dcd(const struct MALI_DRAW *p, int job_no, enum mali_job_type job_type, char *suffix, unsigned gpu_id)818 pandecode_dcd(const struct MALI_DRAW *p,
819 int job_no, enum mali_job_type job_type,
820 char *suffix, unsigned gpu_id)
821 {
822 struct pandecode_mapped_memory *attr_mem;
823
824 #if PAN_ARCH >= 5
825 struct pandecode_fbd fbd_info = {
826 /* Default for Bifrost */
827 .rt_count = 1
828 };
829 #endif
830
831 #if PAN_ARCH >= 6
832 pandecode_local_storage(p->thread_storage & ~1, job_no);
833 #elif PAN_ARCH == 5
834 if (job_type != MALI_JOB_TYPE_TILER) {
835 pandecode_local_storage(p->thread_storage & ~1, job_no);
836 } else {
837 assert(p->fbd & MALI_FBD_TAG_IS_MFBD);
838 fbd_info = pandecode_mfbd_bfr((u64) ((uintptr_t) p->fbd) & ~MALI_FBD_TAG_MASK,
839 job_no, false, gpu_id);
840 }
841 #else
842 pandecode_sfbd((u64) (uintptr_t) p->fbd, job_no, false, gpu_id);
843 #endif
844
845 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
846 int texture_count = 0, sampler_count = 0;
847
848 if (p->state) {
849 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->state);
850 uint32_t *cl = pandecode_fetch_gpu_mem(smem, p->state, pan_size(RENDERER_STATE));
851
852 pan_unpack(cl, RENDERER_STATE, state);
853
854 if (state.shader.shader & ~0xF)
855 pandecode_shader_disassemble(state.shader.shader & ~0xF, job_no, job_type, gpu_id);
856
857 #if PAN_ARCH >= 6
858 bool idvs = (job_type == MALI_JOB_TYPE_INDEXED_VERTEX);
859
860 if (idvs && state.secondary_shader)
861 pandecode_shader_disassemble(state.secondary_shader, job_no, job_type, gpu_id);
862 #endif
863 DUMP_UNPACKED(RENDERER_STATE, state, "State:\n");
864 pandecode_indent++;
865
866 /* Save for dumps */
867 attribute_count = state.shader.attribute_count;
868 varying_count = state.shader.varying_count;
869 texture_count = state.shader.texture_count;
870 sampler_count = state.shader.sampler_count;
871 uniform_buffer_count = state.properties.uniform_buffer_count;
872
873 #if PAN_ARCH >= 6
874 uniform_count = state.preload.uniform_count;
875 #else
876 uniform_count = state.properties.uniform_count;
877 #endif
878
879 #if PAN_ARCH == 4
880 mali_ptr shader = state.blend_shader & ~0xF;
881 if (state.multisample_misc.blend_shader && shader)
882 pandecode_blend_shader_disassemble(shader, job_no, job_type, gpu_id);
883 #endif
884 pandecode_indent--;
885 pandecode_log("\n");
886
887 /* MRT blend fields are used whenever MFBD is used, with
888 * per-RT descriptors */
889
890 #if PAN_ARCH >= 5
891 if ((job_type == MALI_JOB_TYPE_TILER || job_type == MALI_JOB_TYPE_FRAGMENT) &&
892 (PAN_ARCH >= 6 || p->thread_storage & MALI_FBD_TAG_IS_MFBD)) {
893 void* blend_base = ((void *) cl) + pan_size(RENDERER_STATE);
894
895 for (unsigned i = 0; i < fbd_info.rt_count; i++) {
896 mali_ptr shader = 0;
897
898 #if PAN_ARCH >= 6
899 shader = pandecode_bifrost_blend(blend_base, job_no, i,
900 state.shader.shader);
901 #else
902 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
903 #endif
904 if (shader & ~0xF)
905 pandecode_blend_shader_disassemble(shader, job_no, job_type,
906 gpu_id);
907 }
908 }
909 #endif
910 } else
911 pandecode_msg("XXX: missing shader descriptor\n");
912
913 if (p->viewport) {
914 DUMP_ADDR(VIEWPORT, p->viewport, "Viewport:\n");
915 pandecode_log("\n");
916 }
917
918 unsigned max_attr_index = 0;
919
920 if (p->attributes)
921 max_attr_index = pandecode_attribute_meta(attribute_count, p->attributes, false);
922
923 if (p->attribute_buffers) {
924 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attribute_buffers);
925 pandecode_attributes(attr_mem, p->attribute_buffers, job_no, suffix, max_attr_index, false, job_type);
926 }
927
928 if (p->varyings) {
929 varying_count = pandecode_attribute_meta(varying_count, p->varyings, true);
930 }
931
932 if (p->varying_buffers) {
933 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varying_buffers);
934 pandecode_attributes(attr_mem, p->varying_buffers, job_no, suffix, varying_count, true, job_type);
935 }
936
937 if (p->uniform_buffers) {
938 if (uniform_buffer_count)
939 pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
940 else
941 pandecode_msg("warn: UBOs specified but not referenced\n");
942 } else if (uniform_buffer_count)
943 pandecode_msg("XXX: UBOs referenced but not specified\n");
944
945 /* We don't want to actually dump uniforms, but we do need to validate
946 * that the counts we were given are sane */
947
948 if (p->push_uniforms) {
949 if (uniform_count)
950 pandecode_uniforms(p->push_uniforms, uniform_count);
951 else
952 pandecode_msg("warn: Uniforms specified but not referenced\n");
953 } else if (uniform_count)
954 pandecode_msg("XXX: Uniforms referenced but not specified\n");
955
956 if (p->textures)
957 pandecode_textures(p->textures, texture_count, job_no);
958
959 if (p->samplers)
960 pandecode_samplers(p->samplers, sampler_count, job_no);
961 }
962
963 static void
pandecode_vertex_compute_geometry_job(const struct MALI_JOB_HEADER *h, const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no, unsigned gpu_id)964 pandecode_vertex_compute_geometry_job(const struct MALI_JOB_HEADER *h,
965 const struct pandecode_mapped_memory *mem,
966 mali_ptr job, int job_no, unsigned gpu_id)
967 {
968 struct mali_compute_job_packed *PANDECODE_PTR_VAR(p, mem, job);
969 pan_section_unpack(p, COMPUTE_JOB, DRAW, draw);
970 pandecode_dcd(&draw, job_no, h->type, "", gpu_id);
971
972 pandecode_log("Vertex Job Payload:\n");
973 pandecode_indent++;
974 pandecode_invocation(pan_section_ptr(p, COMPUTE_JOB, INVOCATION));
975 DUMP_SECTION(COMPUTE_JOB, PARAMETERS, p, "Vertex Job Parameters:\n");
976 DUMP_UNPACKED(DRAW, draw, "Draw:\n");
977 pandecode_indent--;
978 pandecode_log("\n");
979 }
980 #endif
981
982 #if PAN_ARCH >= 6
983 static void
pandecode_bifrost_tiler_heap(mali_ptr gpu_va, int job_no)984 pandecode_bifrost_tiler_heap(mali_ptr gpu_va, int job_no)
985 {
986 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
987 pan_unpack(PANDECODE_PTR(mem, gpu_va, void), TILER_HEAP, h);
988 DUMP_UNPACKED(TILER_HEAP, h, "Bifrost Tiler Heap:\n");
989 }
990
991 static void
pandecode_bifrost_tiler(mali_ptr gpu_va, int job_no)992 pandecode_bifrost_tiler(mali_ptr gpu_va, int job_no)
993 {
994 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
995 pan_unpack(PANDECODE_PTR(mem, gpu_va, void), TILER_CONTEXT, t);
996
997 if (t.heap)
998 pandecode_bifrost_tiler_heap(t.heap, job_no);
999
1000 DUMP_UNPACKED(TILER_CONTEXT, t, "Bifrost Tiler:\n");
1001 }
1002
1003 #if PAN_ARCH <= 7
1004 static void
pandecode_indexed_vertex_job(const struct MALI_JOB_HEADER *h, const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no, unsigned gpu_id)1005 pandecode_indexed_vertex_job(const struct MALI_JOB_HEADER *h,
1006 const struct pandecode_mapped_memory *mem,
1007 mali_ptr job, int job_no, unsigned gpu_id)
1008 {
1009 struct mali_indexed_vertex_job_packed *PANDECODE_PTR_VAR(p, mem, job);
1010
1011 pandecode_log("Vertex:\n");
1012 pan_section_unpack(p, INDEXED_VERTEX_JOB, VERTEX_DRAW, vert_draw);
1013 pandecode_dcd(&vert_draw, job_no, h->type, "", gpu_id);
1014 DUMP_UNPACKED(DRAW, vert_draw, "Vertex Draw:\n");
1015
1016 pandecode_log("Fragment:\n");
1017 pan_section_unpack(p, INDEXED_VERTEX_JOB, FRAGMENT_DRAW, frag_draw);
1018 pandecode_dcd(&frag_draw, job_no, MALI_JOB_TYPE_FRAGMENT, "", gpu_id);
1019 DUMP_UNPACKED(DRAW, frag_draw, "Fragment Draw:\n");
1020
1021 pan_section_unpack(p, INDEXED_VERTEX_JOB, TILER, tiler_ptr);
1022 pandecode_log("Tiler Job Payload:\n");
1023 pandecode_indent++;
1024 pandecode_bifrost_tiler(tiler_ptr.address, job_no);
1025 pandecode_indent--;
1026
1027 pandecode_invocation(pan_section_ptr(p, INDEXED_VERTEX_JOB, INVOCATION));
1028 pandecode_primitive(pan_section_ptr(p, INDEXED_VERTEX_JOB, PRIMITIVE));
1029
1030 /* TODO: gl_PointSize on Bifrost */
1031 pandecode_primitive_size(pan_section_ptr(p, INDEXED_VERTEX_JOB, PRIMITIVE_SIZE), true);
1032
1033 pan_section_unpack(p, INDEXED_VERTEX_JOB, PADDING, padding);
1034 }
1035 #endif
1036 #endif
1037
1038 static void
pandecode_tiler_job(const struct MALI_JOB_HEADER *h, const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no, unsigned gpu_id)1039 pandecode_tiler_job(const struct MALI_JOB_HEADER *h,
1040 const struct pandecode_mapped_memory *mem,
1041 mali_ptr job, int job_no, unsigned gpu_id)
1042 {
1043 struct mali_tiler_job_packed *PANDECODE_PTR_VAR(p, mem, job);
1044 pan_section_unpack(p, TILER_JOB, DRAW, draw);
1045 pandecode_dcd(&draw, job_no, h->type, "", gpu_id);
1046 pandecode_log("Tiler Job Payload:\n");
1047 pandecode_indent++;
1048
1049 #if PAN_ARCH <= 7
1050 pandecode_invocation(pan_section_ptr(p, TILER_JOB, INVOCATION));
1051 #endif
1052
1053 pandecode_primitive(pan_section_ptr(p, TILER_JOB, PRIMITIVE));
1054 DUMP_UNPACKED(DRAW, draw, "Draw:\n");
1055
1056 #if PAN_ARCH >= 6
1057 pan_section_unpack(p, TILER_JOB, TILER, tiler_ptr);
1058 pandecode_bifrost_tiler(tiler_ptr.address, job_no);
1059
1060 /* TODO: gl_PointSize on Bifrost */
1061 pandecode_primitive_size(pan_section_ptr(p, TILER_JOB, PRIMITIVE_SIZE), true);
1062
1063 #if PAN_ARCH >= 9
1064 DUMP_SECTION(TILER_JOB, INSTANCE_COUNT, p, "Instance count:\n");
1065 DUMP_SECTION(TILER_JOB, VERTEX_COUNT, p, "Vertex count:\n");
1066 DUMP_SECTION(TILER_JOB, SCISSOR, p, "Scissor:\n");
1067 DUMP_SECTION(TILER_JOB, INDICES, p, "Indices:\n");
1068 #else
1069 pan_section_unpack(p, TILER_JOB, PADDING, padding);
1070 #endif
1071
1072 #else
1073 pan_section_unpack(p, TILER_JOB, PRIMITIVE, primitive);
1074 pandecode_primitive_size(pan_section_ptr(p, TILER_JOB, PRIMITIVE_SIZE),
1075 primitive.point_size_array_format == MALI_POINT_SIZE_ARRAY_FORMAT_NONE);
1076 #endif
1077 pandecode_indent--;
1078 pandecode_log("\n");
1079 }
1080
1081 static void
pandecode_fragment_job(const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no, unsigned gpu_id)1082 pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
1083 mali_ptr job, int job_no, unsigned gpu_id)
1084 {
1085 struct mali_fragment_job_packed *PANDECODE_PTR_VAR(p, mem, job);
1086 pan_section_unpack(p, FRAGMENT_JOB, PAYLOAD, s);
1087
1088
1089 #if PAN_ARCH == 4
1090 pandecode_sfbd(s.framebuffer, job_no, true, gpu_id);
1091 #else
1092 assert(s.framebuffer & MALI_FBD_TAG_IS_MFBD);
1093
1094 struct pandecode_fbd info;
1095
1096 info = pandecode_mfbd_bfr(s.framebuffer & ~MALI_FBD_TAG_MASK, job_no,
1097 true, gpu_id);
1098 #endif
1099
1100 #if PAN_ARCH >= 5
1101 unsigned expected_tag = 0;
1102
1103 /* Compute the tag for the tagged pointer. This contains the type of
1104 * FBD (MFBD/SFBD), and in the case of an MFBD, information about which
1105 * additional structures follow the MFBD header (an extra payload or
1106 * not, as well as a count of render targets) */
1107
1108 expected_tag = MALI_FBD_TAG_IS_MFBD;
1109 if (info.has_extra)
1110 expected_tag |= MALI_FBD_TAG_HAS_ZS_RT;
1111
1112 expected_tag |= MALI_FBD_TAG_IS_MFBD | (MALI_POSITIVE(info.rt_count) << 2);
1113 #endif
1114
1115 DUMP_UNPACKED(FRAGMENT_JOB_PAYLOAD, s, "Fragment Job Payload:\n");
1116
1117 #if PAN_ARCH >= 5
1118 /* The FBD is a tagged pointer */
1119
1120 unsigned tag = (s.framebuffer & MALI_FBD_TAG_MASK);
1121
1122 if (tag != expected_tag)
1123 pandecode_msg("XXX: expected FBD tag %X but got %X\n", expected_tag, tag);
1124 #endif
1125
1126 pandecode_log("\n");
1127 }
1128
1129 static void
pandecode_write_value_job(const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no)1130 pandecode_write_value_job(const struct pandecode_mapped_memory *mem,
1131 mali_ptr job, int job_no)
1132 {
1133 struct mali_write_value_job_packed *PANDECODE_PTR_VAR(p, mem, job);
1134 pan_section_unpack(p, WRITE_VALUE_JOB, PAYLOAD, u);
1135 DUMP_SECTION(WRITE_VALUE_JOB, PAYLOAD, p, "Write Value Payload:\n");
1136 pandecode_log("\n");
1137 }
1138
1139 static void
pandecode_cache_flush_job(const struct pandecode_mapped_memory *mem, mali_ptr job, int job_no)1140 pandecode_cache_flush_job(const struct pandecode_mapped_memory *mem,
1141 mali_ptr job, int job_no)
1142 {
1143 struct mali_cache_flush_job_packed *PANDECODE_PTR_VAR(p, mem, job);
1144 pan_section_unpack(p, CACHE_FLUSH_JOB, PAYLOAD, u);
1145 DUMP_SECTION(CACHE_FLUSH_JOB, PAYLOAD, p, "Cache Flush Payload:\n");
1146 pandecode_log("\n");
1147 }
1148
1149 #if PAN_ARCH >= 9
1150 static void
dump_fau(mali_ptr addr, unsigned count, const char *name)1151 dump_fau(mali_ptr addr, unsigned count, const char *name)
1152 {
1153 struct pandecode_mapped_memory *mem =
1154 pandecode_find_mapped_gpu_mem_containing(addr);
1155 const uint32_t *PANDECODE_PTR_VAR(raw, mem, addr);
1156
1157 pandecode_validate_buffer(addr, count * 8);
1158
1159 fprintf(pandecode_dump_stream, "%s:\n", name);
1160 for (unsigned i = 0; i < count; ++i) {
1161 fprintf(pandecode_dump_stream, " %08X %08X\n",
1162 raw[2*i], raw[2*i + 1]);
1163 }
1164 fprintf(pandecode_dump_stream, "\n");
1165 }
1166
1167 static mali_ptr
pandecode_shader(mali_ptr addr, const char *label, unsigned gpu_id)1168 pandecode_shader(mali_ptr addr, const char *label, unsigned gpu_id)
1169 {
1170 MAP_ADDR(SHADER_PROGRAM, addr, cl);
1171 pan_unpack(cl, SHADER_PROGRAM, desc);
1172
1173 assert(desc.type == 8);
1174
1175 DUMP_UNPACKED(SHADER_PROGRAM, desc, "%s Shader:\n", label);
1176 pandecode_shader_disassemble(desc.binary, 0, 0, gpu_id);
1177 return desc.binary;
1178 }
1179
1180 static void
pandecode_resources(mali_ptr addr, unsigned size)1181 pandecode_resources(mali_ptr addr, unsigned size)
1182 {
1183 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(addr);
1184 const uint8_t *cl = pandecode_fetch_gpu_mem(mem, addr, size);
1185 assert((size % 0x20) == 0);
1186
1187 for (unsigned i = 0; i < size; i += 0x20) {
1188 unsigned type = (cl[i] & 0xF);
1189
1190 switch (type) {
1191 case MALI_DESCRIPTOR_TYPE_SAMPLER:
1192 DUMP_CL(SAMPLER, cl + i, "Sampler:\n");
1193 break;
1194 case MALI_DESCRIPTOR_TYPE_TEXTURE:
1195 pandecode_bifrost_texture(cl + i, 0, i);
1196 break;
1197 case MALI_DESCRIPTOR_TYPE_ATTRIBUTE:
1198 DUMP_CL(ATTRIBUTE, cl + i, "Attribute:\n");
1199 break;
1200 case MALI_DESCRIPTOR_TYPE_BUFFER:
1201 DUMP_CL(BUFFER, cl + i, "Buffer:\n");
1202 break;
1203 default:
1204 fprintf(pandecode_dump_stream, "Unknown descriptor type %X\n", type);
1205 break;
1206 }
1207 }
1208 }
1209
1210 static void
pandecode_resource_tables(mali_ptr addr, const char *label)1211 pandecode_resource_tables(mali_ptr addr, const char *label)
1212 {
1213 unsigned count = addr & 0x3F;
1214 addr = addr & ~0x3F;
1215
1216 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(addr);
1217 const uint8_t *cl = pandecode_fetch_gpu_mem(mem, addr, MALI_RESOURCE_LENGTH * count);
1218
1219 for (unsigned i = 0; i < count; ++i) {
1220 pan_unpack(cl + i * MALI_RESOURCE_LENGTH, RESOURCE, entry);
1221 DUMP_UNPACKED(RESOURCE, entry, "Entry %u:\n", i);
1222
1223 pandecode_indent += 2;
1224 if (entry.address)
1225 pandecode_resources(entry.address, entry.size);
1226 pandecode_indent -= 2;
1227 }
1228 }
1229
1230 static void
pandecode_depth_stencil(mali_ptr addr)1231 pandecode_depth_stencil(mali_ptr addr)
1232 {
1233 MAP_ADDR(DEPTH_STENCIL, addr, cl);
1234 pan_unpack(cl, DEPTH_STENCIL, desc);
1235 DUMP_UNPACKED(DEPTH_STENCIL, desc, "Depth/stencil");
1236 }
1237
1238 static void
pandecode_shader_environment(const struct MALI_SHADER_ENVIRONMENT *p, unsigned gpu_id)1239 pandecode_shader_environment(const struct MALI_SHADER_ENVIRONMENT *p,
1240 unsigned gpu_id)
1241 {
1242 if (p->shader)
1243 pandecode_shader(p->shader, "Shader", gpu_id);
1244
1245 if (p->resources)
1246 pandecode_resource_tables(p->resources, "Resources");
1247
1248 if (p->thread_storage)
1249 pandecode_local_storage(p->thread_storage, 0);
1250
1251 if (p->fau)
1252 dump_fau(p->fau, p->fau_count, "FAU");
1253 }
1254
1255 static void
pandecode_dcd(const struct MALI_DRAW *p, int job_no, enum mali_job_type job_type, char *suffix, unsigned gpu_id)1256 pandecode_dcd(const struct MALI_DRAW *p,
1257 int job_no, enum mali_job_type job_type,
1258 char *suffix, unsigned gpu_id)
1259 {
1260 mali_ptr frag_shader = 0;
1261
1262 pandecode_depth_stencil(p->depth_stencil);
1263
1264 for (unsigned i = 0; i < p->blend_count; ++i) {
1265 struct pandecode_mapped_memory *blend_mem =
1266 pandecode_find_mapped_gpu_mem_containing(p->blend);
1267
1268 struct mali_blend_packed *PANDECODE_PTR_VAR(blend_descs, blend_mem, p->blend);
1269
1270 mali_ptr blend_shader = pandecode_bifrost_blend(blend_descs, 0, i, frag_shader);
1271 if (blend_shader) {
1272 fprintf(pandecode_dump_stream, "Blend shader %u", i);
1273 pandecode_shader_disassemble(blend_shader, 0, 0, gpu_id);
1274 }
1275 }
1276
1277 pandecode_shader_environment(&p->shader, gpu_id);
1278 DUMP_UNPACKED(DRAW, *p, "Draw:\n");
1279 }
1280
1281 static void
pandecode_malloc_vertex_job(const struct pandecode_mapped_memory *mem, mali_ptr job, unsigned gpu_id)1282 pandecode_malloc_vertex_job(const struct pandecode_mapped_memory *mem,
1283 mali_ptr job, unsigned gpu_id)
1284 {
1285 struct mali_malloc_vertex_job_packed *PANDECODE_PTR_VAR(p, mem, job);
1286
1287 DUMP_SECTION(MALLOC_VERTEX_JOB, PRIMITIVE, p, "Primitive:\n");
1288 DUMP_SECTION(MALLOC_VERTEX_JOB, INSTANCE_COUNT, p, "Instance count:\n");
1289 DUMP_SECTION(MALLOC_VERTEX_JOB, ALLOCATION, p, "Allocation:\n");
1290 DUMP_SECTION(MALLOC_VERTEX_JOB, TILER, p, "Tiler:\n");
1291 DUMP_SECTION(MALLOC_VERTEX_JOB, SCISSOR, p, "Scissor:\n");
1292 DUMP_SECTION(MALLOC_VERTEX_JOB, PRIMITIVE_SIZE, p, "Primitive Size:\n");
1293 DUMP_SECTION(MALLOC_VERTEX_JOB, INDICES, p, "Indices:\n");
1294
1295 pan_section_unpack(p, MALLOC_VERTEX_JOB, DRAW, dcd);
1296
1297 pan_section_unpack(p, MALLOC_VERTEX_JOB, TILER, tiler_ptr);
1298 pandecode_log("Tiler Job Payload:\n");
1299 pandecode_indent++;
1300 if (tiler_ptr.address)
1301 pandecode_bifrost_tiler(tiler_ptr.address, 0);
1302 else
1303 pandecode_log("<omitted>\n");
1304 pandecode_indent--;
1305
1306 pandecode_dcd(&dcd, 0, 0, NULL, gpu_id);
1307
1308 pan_section_unpack(p, MALLOC_VERTEX_JOB, POSITION, position);
1309 pan_section_unpack(p, MALLOC_VERTEX_JOB, VARYING, varying);
1310 pandecode_shader_environment(&position, gpu_id);
1311 pandecode_shader_environment(&varying, gpu_id);
1312 }
1313
1314 static void
pandecode_compute_job(const struct pandecode_mapped_memory *mem, mali_ptr job, unsigned gpu_id)1315 pandecode_compute_job(const struct pandecode_mapped_memory *mem, mali_ptr job, unsigned gpu_id)
1316 {
1317 struct mali_compute_job_packed *PANDECODE_PTR_VAR(p, mem, job);
1318 pan_section_unpack(p, COMPUTE_JOB, PAYLOAD, payload);
1319
1320 pandecode_shader(payload.compute.shader, "Shader", gpu_id);
1321 if (payload.compute.thread_storage)
1322 pandecode_local_storage(payload.compute.thread_storage, 0);
1323 if (payload.compute.fau)
1324 dump_fau(payload.compute.fau, payload.compute.fau_count, "FAU");
1325 if (payload.compute.resources)
1326 pandecode_resource_tables(payload.compute.resources, "Resources");
1327
1328 DUMP_UNPACKED(COMPUTE_PAYLOAD, payload, "Compute:\n");
1329 }
1330 #endif
1331
1332 /* Entrypoint to start tracing. jc_gpu_va is the GPU address for the first job
1333 * in the chain; later jobs are found by walking the chain. Bifrost is, well,
1334 * if it's bifrost or not. GPU ID is the more finegrained ID (at some point, we
1335 * might wish to combine this with the bifrost parameter) because some details
1336 * are model-specific even within a particular architecture. */
1337
1338 void
pandecode_jc(mali_ptr jc_gpu_va, unsigned gpu_id)1339 GENX(pandecode_jc)(mali_ptr jc_gpu_va, unsigned gpu_id)
1340 {
1341 pandecode_dump_file_open();
1342
1343 unsigned job_descriptor_number = 0;
1344 mali_ptr next_job = 0;
1345
1346 do {
1347 struct pandecode_mapped_memory *mem =
1348 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
1349
1350 pan_unpack(PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_header_packed),
1351 JOB_HEADER, h);
1352 next_job = h.next;
1353
1354 int job_no = job_descriptor_number++;
1355
1356 DUMP_UNPACKED(JOB_HEADER, h, "Job Header (%" PRIx64 "):\n", jc_gpu_va);
1357 pandecode_log("\n");
1358
1359 switch (h.type) {
1360 case MALI_JOB_TYPE_WRITE_VALUE:
1361 pandecode_write_value_job(mem, jc_gpu_va, job_no);
1362 break;
1363
1364 case MALI_JOB_TYPE_CACHE_FLUSH:
1365 pandecode_cache_flush_job(mem, jc_gpu_va, job_no);
1366 break;
1367
1368 case MALI_JOB_TYPE_TILER:
1369 pandecode_tiler_job(&h, mem, jc_gpu_va, job_no, gpu_id);
1370 break;
1371
1372 #if PAN_ARCH <= 7
1373 case MALI_JOB_TYPE_VERTEX:
1374 case MALI_JOB_TYPE_COMPUTE:
1375 pandecode_vertex_compute_geometry_job(&h, mem, jc_gpu_va, job_no, gpu_id);
1376 break;
1377
1378 #if PAN_ARCH >= 6
1379 case MALI_JOB_TYPE_INDEXED_VERTEX:
1380 pandecode_indexed_vertex_job(&h, mem, jc_gpu_va, job_no, gpu_id);
1381 break;
1382 #endif
1383 #else
1384 case MALI_JOB_TYPE_COMPUTE:
1385 pandecode_compute_job(mem, jc_gpu_va, gpu_id);
1386 break;
1387
1388 case MALI_JOB_TYPE_MALLOC_VERTEX:
1389 pandecode_malloc_vertex_job(mem, jc_gpu_va, gpu_id);
1390 break;
1391 #endif
1392
1393 case MALI_JOB_TYPE_FRAGMENT:
1394 pandecode_fragment_job(mem, jc_gpu_va, job_no, gpu_id);
1395 break;
1396
1397 default:
1398 break;
1399 }
1400 } while ((jc_gpu_va = next_job));
1401
1402 fflush(pandecode_dump_stream);
1403 pandecode_map_read_write();
1404 }
1405
1406 void
pandecode_abort_on_fault(mali_ptr jc_gpu_va)1407 GENX(pandecode_abort_on_fault)(mali_ptr jc_gpu_va)
1408 {
1409 mali_ptr next_job = 0;
1410
1411 do {
1412 struct pandecode_mapped_memory *mem =
1413 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
1414
1415 pan_unpack(PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_header_packed),
1416 JOB_HEADER, h);
1417 next_job = h.next;
1418
1419 /* Ensure the job is marked COMPLETE */
1420 if (h.exception_status != 0x1) {
1421 fprintf(stderr, "Incomplete job or timeout\n");
1422 fflush(NULL);
1423 abort();
1424 }
1425 } while ((jc_gpu_va = next_job));
1426
1427 pandecode_map_read_write();
1428 }
1429