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 "decode.h"
27 #include <ctype.h>
28 #include <errno.h>
29 #include <memory.h>
30 #include <stdarg.h>
31 #include <stdbool.h>
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <genxml/gen_macros.h>
35 #include <sys/mman.h>
36
37 #include "compiler/bifrost/disassemble.h"
38 #include "compiler/valhall/disassemble.h"
39 #include "midgard/disassemble.h"
40 #include "util/set.h"
41 #include "pan_format.h"
42
43 #if PAN_ARCH <= 5
44 /* Midgard's tiler descriptor is embedded within the
45 * larger FBD */
46
47 static void
pandecode_midgard_tiler_descriptor(struct pandecode_context * ctx,const struct mali_tiler_context_packed * tp,const struct mali_tiler_weights_packed * wp)48 pandecode_midgard_tiler_descriptor(struct pandecode_context *ctx,
49 const struct mali_tiler_context_packed *tp,
50 const struct mali_tiler_weights_packed *wp)
51 {
52 pan_unpack(tp, TILER_CONTEXT, t);
53 DUMP_UNPACKED(ctx, TILER_CONTEXT, t, "Tiler:\n");
54
55 /* We've never seen weights used in practice, but they exist */
56 pan_unpack(wp, TILER_WEIGHTS, w);
57 bool nonzero_weights = false;
58
59 nonzero_weights |= w.weight0 != 0x0;
60 nonzero_weights |= w.weight1 != 0x0;
61 nonzero_weights |= w.weight2 != 0x0;
62 nonzero_weights |= w.weight3 != 0x0;
63 nonzero_weights |= w.weight4 != 0x0;
64 nonzero_weights |= w.weight5 != 0x0;
65 nonzero_weights |= w.weight6 != 0x0;
66 nonzero_weights |= w.weight7 != 0x0;
67
68 if (nonzero_weights)
69 DUMP_UNPACKED(ctx, TILER_WEIGHTS, w, "Tiler Weights:\n");
70 }
71 #endif
72
73 #if PAN_ARCH >= 5
74 static void
pandecode_render_target(struct pandecode_context * ctx,uint64_t gpu_va,unsigned gpu_id,const struct MALI_FRAMEBUFFER_PARAMETERS * fb)75 pandecode_render_target(struct pandecode_context *ctx, uint64_t gpu_va,
76 unsigned gpu_id,
77 const struct MALI_FRAMEBUFFER_PARAMETERS *fb)
78 {
79 pandecode_log(ctx, "Color Render Targets @%" PRIx64 ":\n", gpu_va);
80 ctx->indent++;
81
82 for (int i = 0; i < (fb->render_target_count); i++) {
83 mali_ptr rt_va = gpu_va + i * pan_size(RENDER_TARGET);
84 const struct mali_render_target_packed *PANDECODE_PTR_VAR(
85 ctx, rtp, (mali_ptr)rt_va);
86 DUMP_CL(ctx, RENDER_TARGET, rtp, "Color Render Target %d:\n", i);
87 }
88
89 ctx->indent--;
90 pandecode_log(ctx, "\n");
91 }
92 #endif
93
94 #if PAN_ARCH >= 6
95 static void
pandecode_sample_locations(struct pandecode_context * ctx,const void * fb)96 pandecode_sample_locations(struct pandecode_context *ctx, const void *fb)
97 {
98 pan_section_unpack(fb, FRAMEBUFFER, PARAMETERS, params);
99
100 const u16 *PANDECODE_PTR_VAR(ctx, samples, params.sample_locations);
101
102 pandecode_log(ctx, "Sample locations @%" PRIx64 ":\n",
103 params.sample_locations);
104 for (int i = 0; i < 33; i++) {
105 pandecode_log(ctx, " (%d, %d),\n", samples[2 * i] - 128,
106 samples[2 * i + 1] - 128);
107 }
108 }
109 #endif
110
111 struct pandecode_fbd
GENX(pandecode_fbd)112 GENX(pandecode_fbd)(struct pandecode_context *ctx, uint64_t gpu_va,
113 bool is_fragment, unsigned gpu_id)
114 {
115 const void *PANDECODE_PTR_VAR(ctx, fb, (mali_ptr)gpu_va);
116 pan_section_unpack(fb, FRAMEBUFFER, PARAMETERS, params);
117 DUMP_UNPACKED(ctx, FRAMEBUFFER_PARAMETERS, params, "Parameters:\n");
118
119 #if PAN_ARCH >= 6
120 pandecode_sample_locations(ctx, fb);
121
122 unsigned dcd_size = pan_size(DRAW);
123 unsigned job_type_param = 0;
124
125 #if PAN_ARCH <= 9
126 job_type_param = MALI_JOB_TYPE_FRAGMENT;
127 #endif
128
129 if (params.pre_frame_0 != MALI_PRE_POST_FRAME_SHADER_MODE_NEVER) {
130 const void *PANDECODE_PTR_VAR(ctx, dcd,
131 params.frame_shader_dcds + (0 * dcd_size));
132 pan_unpack(dcd, DRAW, draw);
133 pandecode_log(ctx, "Pre frame 0 @%" PRIx64 " (mode=%d):\n",
134 params.frame_shader_dcds, params.pre_frame_0);
135 GENX(pandecode_dcd)(ctx, &draw, job_type_param, gpu_id);
136 }
137
138 if (params.pre_frame_1 != MALI_PRE_POST_FRAME_SHADER_MODE_NEVER) {
139 const void *PANDECODE_PTR_VAR(ctx, dcd,
140 params.frame_shader_dcds + (1 * dcd_size));
141 pan_unpack(dcd, DRAW, draw);
142 pandecode_log(ctx, "Pre frame 1 @%" PRIx64 ":\n",
143 params.frame_shader_dcds + (1 * dcd_size));
144 GENX(pandecode_dcd)(ctx, &draw, job_type_param, gpu_id);
145 }
146
147 if (params.post_frame != MALI_PRE_POST_FRAME_SHADER_MODE_NEVER) {
148 const void *PANDECODE_PTR_VAR(ctx, dcd,
149 params.frame_shader_dcds + (2 * dcd_size));
150 pan_unpack(dcd, DRAW, draw);
151 pandecode_log(ctx, "Post frame:\n");
152 GENX(pandecode_dcd)(ctx, &draw, job_type_param, gpu_id);
153 }
154 #else
155 DUMP_SECTION(ctx, FRAMEBUFFER, LOCAL_STORAGE, fb, "Local Storage:\n");
156
157 const void *t = pan_section_ptr(fb, FRAMEBUFFER, TILER);
158 const void *w = pan_section_ptr(fb, FRAMEBUFFER, TILER_WEIGHTS);
159 pandecode_midgard_tiler_descriptor(ctx, t, w);
160 #endif
161
162 pandecode_log(ctx, "Framebuffer @%" PRIx64 ":\n", gpu_va);
163 ctx->indent++;
164
165 DUMP_UNPACKED(ctx, FRAMEBUFFER_PARAMETERS, params, "Parameters:\n");
166 #if PAN_ARCH >= 6
167 if (params.tiler)
168 GENX(pandecode_tiler)(ctx, params.tiler, gpu_id);
169 #endif
170
171 ctx->indent--;
172 pandecode_log(ctx, "\n");
173
174 #if PAN_ARCH >= 5
175 gpu_va += pan_size(FRAMEBUFFER);
176
177 if (params.has_zs_crc_extension) {
178 const struct mali_zs_crc_extension_packed *PANDECODE_PTR_VAR(
179 ctx, zs_crc, (mali_ptr)gpu_va);
180 DUMP_CL(ctx, ZS_CRC_EXTENSION, zs_crc, "ZS CRC Extension:\n");
181 pandecode_log(ctx, "\n");
182
183 gpu_va += pan_size(ZS_CRC_EXTENSION);
184 }
185
186 if (is_fragment)
187 pandecode_render_target(ctx, gpu_va, gpu_id, ¶ms);
188
189 return (struct pandecode_fbd){
190 .rt_count = params.render_target_count,
191 .has_extra = params.has_zs_crc_extension,
192 };
193 #else
194 /* Dummy unpack of the padding section to make sure all words are 0.
195 * No need to call print here since the section is supposed to be empty.
196 */
197 pan_section_unpack(fb, FRAMEBUFFER, PADDING_1, padding1);
198 pan_section_unpack(fb, FRAMEBUFFER, PADDING_2, padding2);
199
200 return (struct pandecode_fbd){
201 .rt_count = 1,
202 };
203 #endif
204 }
205
206 #if PAN_ARCH >= 5
207 mali_ptr
GENX(pandecode_blend)208 GENX(pandecode_blend)(struct pandecode_context *ctx, void *descs, int rt_no,
209 mali_ptr frag_shader)
210 {
211 pan_unpack(descs + (rt_no * pan_size(BLEND)), BLEND, b);
212 DUMP_UNPACKED(ctx, BLEND, b, "Blend RT %d:\n", rt_no);
213 #if PAN_ARCH >= 6
214 if (b.internal.mode != MALI_BLEND_MODE_SHADER)
215 return 0;
216 /* If we don't have a frag shader, we can't extract the LSB of the blend
217 * shader so return NULL in that case. It doesn't matter, because the
218 * blend shader won't be executed anyway, so disassembling is not
219 * super useful. */
220 if (!frag_shader)
221 return 0;
222
223 return (frag_shader & 0xFFFFFFFF00000000ULL) | b.internal.shader.pc;
224 #else
225 return b.blend_shader ? (b.shader_pc & ~0xf) : 0;
226 #endif
227 }
228 #endif
229
230 #if PAN_ARCH <= 7
231 static bool
panfrost_is_yuv_format(uint32_t packed)232 panfrost_is_yuv_format(uint32_t packed)
233 {
234 #if PAN_ARCH == 7
235 enum mali_format mali_fmt = packed >> 12;
236 return mali_fmt >= MALI_YUV8 && mali_fmt <= MALI_CUSTOM_YUV_5;
237 #else
238 /* Currently only supported by panfrost on v7 */
239 assert(0);
240 return false;
241 #endif
242 }
243
244 static void
pandecode_texture_payload(struct pandecode_context * ctx,mali_ptr payload,const struct MALI_TEXTURE * tex)245 pandecode_texture_payload(struct pandecode_context *ctx, mali_ptr payload,
246 const struct MALI_TEXTURE *tex)
247 {
248 unsigned nr_samples =
249 tex->dimension == MALI_TEXTURE_DIMENSION_3D ? 1 : tex->sample_count;
250
251 if (!payload)
252 return;
253
254 /* A bunch of bitmap pointers follow.
255 * We work out the correct number,
256 * based on the mipmap/cubemap
257 * properties, but dump extra
258 * possibilities to futureproof */
259
260 int bitmap_count = tex->levels;
261
262 /* Miptree for each face */
263 if (tex->dimension == MALI_TEXTURE_DIMENSION_CUBE)
264 bitmap_count *= 6;
265
266 /* Array of layers */
267 bitmap_count *= nr_samples;
268
269 /* Array of textures */
270 bitmap_count *= tex->array_size;
271
272 #define PANDECODE_EMIT_TEX_PAYLOAD_DESC(T, msg) \
273 for (int i = 0; i < bitmap_count; ++i) { \
274 uint64_t addr = payload + pan_size(T) * i; \
275 pan_unpack(PANDECODE_PTR(ctx, addr, void), T, s); \
276 DUMP_UNPACKED(ctx, T, s, msg " @%" PRIx64 ":\n", addr) \
277 }
278
279 #if PAN_ARCH <= 5
280 switch (tex->surface_type) {
281 case MALI_SURFACE_TYPE_32:
282 PANDECODE_EMIT_TEX_PAYLOAD_DESC(SURFACE_32, "Surface 32");
283 break;
284 case MALI_SURFACE_TYPE_64:
285 PANDECODE_EMIT_TEX_PAYLOAD_DESC(SURFACE, "Surface");
286 break;
287 case MALI_SURFACE_TYPE_32_WITH_ROW_STRIDE:
288 PANDECODE_EMIT_TEX_PAYLOAD_DESC(SURFACE_32, "Surface 32 With Row Stride");
289 break;
290 case MALI_SURFACE_TYPE_64_WITH_STRIDES:
291 PANDECODE_EMIT_TEX_PAYLOAD_DESC(SURFACE_WITH_STRIDE,
292 "Surface With Stride");
293 break;
294 default:
295 fprintf(ctx->dump_stream, "Unknown surface descriptor type %X\n",
296 tex->surface_type);
297 break;
298 }
299 #elif PAN_ARCH == 6
300 PANDECODE_EMIT_TEX_PAYLOAD_DESC(SURFACE_WITH_STRIDE, "Surface With Stride");
301 #else
302 STATIC_ASSERT(PAN_ARCH == 7);
303 if (panfrost_is_yuv_format(tex->format)) {
304 PANDECODE_EMIT_TEX_PAYLOAD_DESC(MULTIPLANAR_SURFACE, "Surface YUV");
305 } else {
306 PANDECODE_EMIT_TEX_PAYLOAD_DESC(SURFACE_WITH_STRIDE,
307 "Surface With Stride");
308 }
309 #endif
310
311 #undef PANDECODE_EMIT_TEX_PAYLOAD_DESC
312 }
313 #endif
314
315 #if PAN_ARCH <= 5
316 void
GENX(pandecode_texture)317 GENX(pandecode_texture)(struct pandecode_context *ctx, mali_ptr u, unsigned tex)
318 {
319 const uint8_t *cl = pandecode_fetch_gpu_mem(ctx, u, pan_size(TEXTURE));
320
321 pan_unpack(cl, TEXTURE, temp);
322 DUMP_UNPACKED(ctx, TEXTURE, temp, "Texture:\n")
323
324 ctx->indent++;
325 pandecode_texture_payload(ctx, u + pan_size(TEXTURE), &temp);
326 ctx->indent--;
327 }
328 #else
329 void
GENX(pandecode_texture)330 GENX(pandecode_texture)(struct pandecode_context *ctx, const void *cl,
331 unsigned tex)
332 {
333 pan_unpack(cl, TEXTURE, temp);
334 DUMP_UNPACKED(ctx, TEXTURE, temp, "Texture:\n")
335
336 ctx->indent++;
337
338 #if PAN_ARCH >= 9
339 int plane_count = temp.levels * temp.array_size;
340
341 /* Miptree for each face */
342 if (temp.dimension == MALI_TEXTURE_DIMENSION_CUBE)
343 plane_count *= 6;
344
345 for (unsigned i = 0; i < plane_count; ++i)
346 DUMP_ADDR(ctx, PLANE, temp.surfaces + i * pan_size(PLANE), "Plane %u:\n",
347 i);
348 #else
349 pandecode_texture_payload(ctx, temp.surfaces, &temp);
350 #endif
351 ctx->indent--;
352 }
353 #endif
354
355 #if PAN_ARCH >= 6
356 void
GENX(pandecode_tiler)357 GENX(pandecode_tiler)(struct pandecode_context *ctx, mali_ptr gpu_va,
358 unsigned gpu_id)
359 {
360 pan_unpack(PANDECODE_PTR(ctx, gpu_va, void), TILER_CONTEXT, t);
361
362 if (t.heap) {
363 pan_unpack(PANDECODE_PTR(ctx, t.heap, void), TILER_HEAP, h);
364 DUMP_UNPACKED(ctx, TILER_HEAP, h, "Tiler Heap:\n");
365 }
366
367 DUMP_UNPACKED(ctx, TILER_CONTEXT, t, "Tiler Context @%" PRIx64 ":\n",
368 gpu_va);
369 }
370 #endif
371
372 #if PAN_ARCH >= 9
373 void
GENX(pandecode_fau)374 GENX(pandecode_fau)(struct pandecode_context *ctx, mali_ptr addr,
375 unsigned count, const char *name)
376 {
377 if (count == 0)
378 return;
379
380 const uint32_t *PANDECODE_PTR_VAR(ctx, raw, addr);
381
382 pandecode_validate_buffer(ctx, addr, count * 8);
383
384 fprintf(ctx->dump_stream, "%s @%" PRIx64 ":\n", name, addr);
385 for (unsigned i = 0; i < count; ++i) {
386 fprintf(ctx->dump_stream, " %08X %08X\n", raw[2 * i], raw[2 * i + 1]);
387 }
388 fprintf(ctx->dump_stream, "\n");
389 }
390
391 mali_ptr
GENX(pandecode_shader)392 GENX(pandecode_shader)(struct pandecode_context *ctx, mali_ptr addr,
393 const char *label, unsigned gpu_id)
394 {
395 MAP_ADDR(ctx, SHADER_PROGRAM, addr, cl);
396 pan_unpack(cl, SHADER_PROGRAM, desc);
397
398 assert(desc.type == 8);
399
400 DUMP_UNPACKED(ctx, SHADER_PROGRAM, desc, "%s Shader @%" PRIx64 ":\n", label,
401 addr);
402 pandecode_shader_disassemble(ctx, desc.binary, gpu_id);
403 return desc.binary;
404 }
405
406 static void
pandecode_resources(struct pandecode_context * ctx,mali_ptr addr,unsigned size)407 pandecode_resources(struct pandecode_context *ctx, mali_ptr addr, unsigned size)
408 {
409 const uint8_t *cl = pandecode_fetch_gpu_mem(ctx, addr, size);
410 assert((size % 0x20) == 0);
411
412 for (unsigned i = 0; i < size; i += 0x20) {
413 unsigned type = (cl[i] & 0xF);
414
415 switch (type) {
416 case MALI_DESCRIPTOR_TYPE_SAMPLER:
417 DUMP_CL(ctx, SAMPLER, cl + i, "Sampler @%" PRIx64 ":\n", addr + i);
418 break;
419 case MALI_DESCRIPTOR_TYPE_TEXTURE:
420 pandecode_log(ctx, "Texture @%" PRIx64 "\n", addr + i);
421 GENX(pandecode_texture)(ctx, cl + i, i);
422 break;
423 case MALI_DESCRIPTOR_TYPE_ATTRIBUTE:
424 DUMP_CL(ctx, ATTRIBUTE, cl + i, "Attribute @%" PRIx64 ":\n", addr + i);
425 break;
426 case MALI_DESCRIPTOR_TYPE_BUFFER:
427 DUMP_CL(ctx, BUFFER, cl + i, "Buffer @%" PRIx64 ":\n", addr + i);
428 break;
429 default:
430 fprintf(ctx->dump_stream, "Unknown descriptor type %X\n", type);
431 break;
432 }
433 }
434 }
435
436 void
GENX(pandecode_resource_tables)437 GENX(pandecode_resource_tables)(struct pandecode_context *ctx, mali_ptr addr,
438 const char *label)
439 {
440 unsigned count = addr & 0x3F;
441 addr = addr & ~0x3F;
442
443 const uint8_t *cl =
444 pandecode_fetch_gpu_mem(ctx, addr, MALI_RESOURCE_LENGTH * count);
445
446 pandecode_log(ctx, "%s resource table @%" PRIx64 "\n", label, addr);
447 ctx->indent += 2;
448 for (unsigned i = 0; i < count; ++i) {
449 pan_unpack(cl + i * MALI_RESOURCE_LENGTH, RESOURCE, entry);
450 DUMP_UNPACKED(ctx, RESOURCE, entry, "Entry %u @%" PRIx64 ":\n", i,
451 addr + i * MALI_RESOURCE_LENGTH);
452
453 ctx->indent += 2;
454 if (entry.address)
455 pandecode_resources(ctx, entry.address, entry.size);
456 ctx->indent -= 2;
457 }
458 ctx->indent -= 2;
459 }
460
461 void
GENX(pandecode_depth_stencil)462 GENX(pandecode_depth_stencil)(struct pandecode_context *ctx, mali_ptr addr)
463 {
464 MAP_ADDR(ctx, DEPTH_STENCIL, addr, cl);
465 pan_unpack(cl, DEPTH_STENCIL, desc);
466 DUMP_UNPACKED(ctx, DEPTH_STENCIL, desc, "Depth/stencil");
467 }
468
469 void
GENX(pandecode_shader_environment)470 GENX(pandecode_shader_environment)(struct pandecode_context *ctx,
471 const struct MALI_SHADER_ENVIRONMENT *p,
472 unsigned gpu_id)
473 {
474 if (p->shader)
475 GENX(pandecode_shader)(ctx, p->shader, "Shader", gpu_id);
476
477 if (p->resources)
478 GENX(pandecode_resource_tables)(ctx, p->resources, "Resources");
479
480 if (p->thread_storage)
481 DUMP_ADDR(ctx, LOCAL_STORAGE, p->thread_storage, "Local Storage:\n");
482
483 if (p->fau)
484 GENX(pandecode_fau)(ctx, p->fau, p->fau_count, "FAU");
485 }
486
487 void
GENX(pandecode_blend_descs)488 GENX(pandecode_blend_descs)(struct pandecode_context *ctx, mali_ptr blend,
489 unsigned count, mali_ptr frag_shader,
490 unsigned gpu_id)
491 {
492 for (unsigned i = 0; i < count; ++i) {
493 struct mali_blend_packed *PANDECODE_PTR_VAR(ctx, blend_descs, blend);
494
495 mali_ptr blend_shader =
496 GENX(pandecode_blend)(ctx, blend_descs, i, frag_shader);
497 if (blend_shader) {
498 fprintf(ctx->dump_stream, "Blend shader %u @%" PRIx64 "", i,
499 blend_shader);
500 pandecode_shader_disassemble(ctx, blend_shader, gpu_id);
501 }
502 }
503 }
504
505 void
GENX(pandecode_dcd)506 GENX(pandecode_dcd)(struct pandecode_context *ctx, const struct MALI_DRAW *p,
507 unsigned unused, unsigned gpu_id)
508 {
509 mali_ptr frag_shader = 0;
510
511 GENX(pandecode_depth_stencil)(ctx, p->depth_stencil);
512 GENX(pandecode_blend_descs)
513 (ctx, p->blend, p->blend_count, frag_shader, gpu_id);
514 GENX(pandecode_shader_environment)(ctx, &p->shader, gpu_id);
515 DUMP_UNPACKED(ctx, DRAW, *p, "Draw:\n");
516 }
517 #endif
518