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
76enum 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
84static unsigned pandecode_indent = 0;
85
86static void
87pandecode_make_indent(void)
88{
89        for (unsigned i = 0; i < pandecode_indent; ++i)
90                fprintf(pandecode_dump_stream, "  ");
91}
92
93static void PRINTFLIKE(2, 3)
94pandecode_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
108static void
109pandecode_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
124static void
125pandecode_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
160static void
161pandecode_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
189struct pandecode_fbd {
190        unsigned width;
191        unsigned height;
192        unsigned rt_count;
193        bool has_extra;
194};
195
196#if PAN_ARCH == 4
197static struct pandecode_fbd
198pandecode_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
234static void
235pandecode_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
242static void
243pandecode_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
263static void
264pandecode_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
282static void
283pandecode_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
288static struct pandecode_fbd
289pandecode_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, &params);
364
365        return info;
366}
367#endif
368
369#if PAN_ARCH <= 7
370static void
371pandecode_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
419static mali_ptr
420pandecode_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
430static mali_ptr
431pandecode_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
440static unsigned
441pandecode_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 */
457static u32
458bits(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
469static void
470pandecode_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
493static void
494pandecode_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
521static void
522pandecode_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
532static void
533pandecode_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
552static void
553pandecode_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
564static const char *
565shader_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
578static unsigned shader_id = 0;
579
580static struct midgard_disasm_stats
581pandecode_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
623static void
624pandecode_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
683static void
684pandecode_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
703static void
704pandecode_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
736static void
737pandecode_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
760static void
761pandecode_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
804static void
805pandecode_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
817static void
818pandecode_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
963static void
964pandecode_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
983static void
984pandecode_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
991static void
992pandecode_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
1004static void
1005pandecode_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
1038static void
1039pandecode_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
1081static void
1082pandecode_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
1129static void
1130pandecode_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
1139static void
1140pandecode_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
1150static void
1151dump_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
1167static mali_ptr
1168pandecode_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
1180static void
1181pandecode_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
1210static void
1211pandecode_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
1230static void
1231pandecode_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
1238static void
1239pandecode_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
1255static void
1256pandecode_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
1281static void
1282pandecode_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
1314static void
1315pandecode_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
1338void
1339GENX(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
1406void
1407GENX(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