2 * Copyright © 2016 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
34 #include <sys/types.h>
39 #include "util/macros.h"
42 #include "intel_aub.h"
43 #include "gen_disasm.h"
45 /* Below is the only command missing from intel_aub.h in libdrm
46 * So, reuse intel_aub.h from libdrm and #define the
47 * AUB_MI_BATCH_BUFFER_END as below
49 #define AUB_MI_BATCH_BUFFER_END (0x0500 << 16)
52 #define HEADER CSI "37;44m"
53 #define NORMAL CSI "0m"
57 static bool option_full_decode
= true;
58 static bool option_print_offsets
= true;
59 static enum { COLOR_AUTO
, COLOR_ALWAYS
, COLOR_NEVER
} option_color
;
63 struct gen_disasm
*disasm
;
65 uint64_t gtt_size
, gtt_end
;
67 uint64_t general_state_base
;
68 uint64_t surface_state_base
;
69 uint64_t dynamic_state_base
;
70 uint64_t instruction_base
;
71 uint64_t instruction_bound
;
73 static inline uint32_t
74 field(uint32_t value
, int start
, int end
)
78 mask
= ~0U >> (31 - end
+ start
);
80 return (value
>> start
) & mask
;
83 struct brw_instruction
;
86 valid_offset(uint32_t offset
)
88 return offset
< gtt_end
;
92 print_dword_val(struct gen_field_iterator
*iter
, uint64_t offset
,
97 f
= iter
->group
->fields
[iter
->i
- 1];
98 const int dword
= f
->start
/ 32;
100 if (*dword_num
!= dword
) {
101 printf("0x%08"PRIx64
": 0x%08x : Dword %d\n",
102 offset
+ 4 * dword
, iter
->p
[dword
], dword
);
108 print_iterator_values(struct gen_field_iterator
*iter
, int *idx
)
111 if (strstr(iter
->value
, "struct") == NULL
) {
112 printf(" %s: %s\n", iter
->name
, iter
->value
);
114 token
= strtok(iter
->value
, " ");
116 token
= strtok(NULL
, " ");
117 *idx
= atoi(strtok(NULL
, ">"));
121 printf(" %s:<struct %s>\n", iter
->name
, token
);
127 decode_structure(struct gen_spec
*spec
, struct gen_group
*strct
,
130 struct gen_field_iterator iter
;
132 int idx
= 0, dword_num
= 0;
135 if (option_print_offsets
)
136 offset
= (void *) p
- gtt
;
140 gen_field_iterator_init(&iter
, strct
, p
);
141 while (gen_field_iterator_next(&iter
)) {
143 print_dword_val(&iter
, offset
, &dword_num
);
144 token
= print_iterator_values(&iter
, &idx
);
146 struct gen_group
*struct_val
= gen_spec_find_struct(spec
, token
);
147 decode_structure(spec
, struct_val
, &p
[idx
]);
154 handle_struct_decode(struct gen_spec
*spec
, char *struct_name
, uint32_t *p
)
156 if (struct_name
== NULL
)
158 struct gen_group
*struct_val
= gen_spec_find_struct(spec
, struct_name
);
159 decode_structure(spec
, struct_val
, p
);
163 dump_binding_table(struct gen_spec
*spec
, uint32_t offset
)
165 uint32_t *pointers
, i
;
167 struct gen_group
*surface_state
;
169 surface_state
= gen_spec_find_struct(spec
, "RENDER_SURFACE_STATE");
170 if (surface_state
== NULL
) {
171 printf("did not find RENDER_SURFACE_STATE info\n");
175 start
= surface_state_base
+ offset
;
176 pointers
= gtt
+ start
;
177 for (i
= 0; i
< 16; i
++) {
178 if (pointers
[i
] == 0)
180 start
= pointers
[i
] + surface_state_base
;
181 if (!valid_offset(start
)) {
182 printf("pointer %u: %08x <not valid>\n",
186 printf("pointer %u: %08x\n", i
, pointers
[i
]);
189 decode_structure(spec
, surface_state
, gtt
+ start
);
194 handle_3dstate_index_buffer(struct gen_spec
*spec
, uint32_t *p
)
197 uint32_t length
, i
, type
, size
;
200 type
= (p
[1] >> 8) & 3;
202 length
= p
[4] / size
;
208 for (i
= 0; i
< length
; i
++) {
211 printf("%3d ", ((uint8_t *)start
)[i
]);
214 printf("%3d ", ((uint16_t *)start
)[i
]);
217 printf("%3d ", ((uint32_t *)start
)[i
]);
221 if (length
< p
[4] / size
)
227 static inline uint64_t
228 get_qword(uint32_t *p
)
230 return ((uint64_t) p
[1] << 32) | p
[0];
234 handle_state_base_address(struct gen_spec
*spec
, uint32_t *p
)
236 uint64_t mask
= ~((1 << 12) - 1);
238 if (gen_spec_get_gen(spec
) >= gen_make_gen(8,0)) {
240 general_state_base
= get_qword(&p
[1]) & mask
;
242 surface_state_base
= get_qword(&p
[4]) & mask
;
244 dynamic_state_base
= get_qword(&p
[6]) & mask
;
246 instruction_base
= get_qword(&p
[10]) & mask
;
248 instruction_bound
= p
[15] & mask
;
251 surface_state_base
= p
[2] & mask
;
253 dynamic_state_base
= p
[3] & mask
;
255 instruction_base
= p
[5] & mask
;
257 instruction_bound
= p
[9] & mask
;
262 dump_samplers(struct gen_spec
*spec
, uint32_t offset
)
266 struct gen_group
*sampler_state
;
268 sampler_state
= gen_spec_find_struct(spec
, "SAMPLER_STATE");
270 start
= dynamic_state_base
+ offset
;
271 for (i
= 0; i
< 4; i
++) {
272 printf("sampler state %d\n", i
);
273 decode_structure(spec
, sampler_state
, gtt
+ start
+ i
* 16);
278 handle_media_interface_descriptor_load(struct gen_spec
*spec
, uint32_t *p
)
280 int i
, length
= p
[2] / 32;
281 struct gen_group
*descriptor_structure
;
282 uint32_t *descriptors
;
284 struct brw_instruction
*insns
;
286 descriptor_structure
=
287 gen_spec_find_struct(spec
, "INTERFACE_DESCRIPTOR_DATA");
288 if (descriptor_structure
== NULL
) {
289 printf("did not find INTERFACE_DESCRIPTOR_DATA info\n");
293 start
= dynamic_state_base
+ p
[3];
294 descriptors
= gtt
+ start
;
295 for (i
= 0; i
< length
; i
++, descriptors
+= 8) {
296 printf("descriptor %u: %08x\n", i
, *descriptors
);
297 decode_structure(spec
, descriptor_structure
, descriptors
);
299 start
= instruction_base
+ descriptors
[0];
300 if (!valid_offset(start
)) {
301 printf("kernel: %08"PRIx64
" <not valid>\n", start
);
304 printf("kernel: %08"PRIx64
"\n", start
);
307 insns
= (struct brw_instruction
*) (gtt
+ start
);
308 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
310 dump_samplers(spec
, descriptors
[3] & ~0x1f);
311 dump_binding_table(spec
, descriptors
[4] & ~0x1f);
315 /* Heuristic to determine whether a uint32_t is probably actually a float
316 * (http://stackoverflow.com/a/2953466)
320 probably_float(uint32_t bits
)
322 int exp
= ((bits
& 0x7f800000U
) >> 23) - 127;
323 uint32_t mant
= bits
& 0x007fffff;
326 if (exp
== -127 && mant
== 0)
329 /* +- 1 billionth to 1 billion */
330 if (-30 <= exp
&& exp
<= 30)
333 /* some value with only a few binary digits */
334 if ((mant
& 0x0000ffff) == 0)
341 handle_3dstate_vertex_buffers(struct gen_spec
*spec
, uint32_t *p
)
343 uint32_t *end
, *s
, *dw
, *dwend
;
345 int n
, i
, count
, stride
;
347 end
= (p
[0] & 0xff) + p
+ 2;
348 for (s
= &p
[1], n
= 0; s
< end
; s
+= 4, n
++) {
349 if (gen_spec_get_gen(spec
) >= gen_make_gen(8, 0)) {
350 offset
= *(uint64_t *) &s
[1];
351 dwend
= gtt
+ offset
+ s
[3];
354 dwend
= gtt
+ s
[2] + 1;
357 stride
= field(s
[0], 0, 11);
359 printf("vertex buffer %d, size %d\n", n
, s
[3]);
360 for (dw
= gtt
+ offset
, i
= 0; dw
< dwend
&& i
< 256; dw
++) {
361 if (count
== 0 && count
% (8 * 4) == 0)
364 if (probably_float(*dw
))
365 printf(" %8.2f", *(float *) dw
);
367 printf(" 0x%08x", *dw
);
372 if (count
== stride
) {
375 } else if (count
% (8 * 4) == 0) {
381 if (count
> 0 && count
% (8 * 4) != 0)
387 handle_3dstate_vs(struct gen_spec
*spec
, uint32_t *p
)
390 struct brw_instruction
*insns
;
393 if (gen_spec_get_gen(spec
) >= gen_make_gen(8, 0)) {
394 start
= get_qword(&p
[1]);
395 vs_enable
= p
[7] & 1;
398 vs_enable
= p
[5] & 1;
402 printf("instruction_base %08"PRIx64
", start %08"PRIx64
"\n",
403 instruction_base
, start
);
405 insns
= (struct brw_instruction
*) (gtt
+ instruction_base
+ start
);
406 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
411 handle_3dstate_hs(struct gen_spec
*spec
, uint32_t *p
)
414 struct brw_instruction
*insns
;
417 if (gen_spec_get_gen(spec
) >= gen_make_gen(8, 0)) {
418 start
= get_qword(&p
[4]);
423 hs_enable
= p
[2] & 0x80000000;
426 printf("instruction_base %08"PRIx64
", start %08"PRIx64
"\n",
427 instruction_base
, start
);
429 insns
= (struct brw_instruction
*) (gtt
+ instruction_base
+ start
);
430 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
435 handle_3dstate_constant(struct gen_spec
*spec
, uint32_t *p
)
441 for (i
= 0; i
< 4; i
++) {
442 length
= (p
[1 + i
/ 2] >> (i
& 1) * 16) & 0xffff;
443 f
= (float *) (gtt
+ p
[3 + i
* 2] + dynamic_state_base
);
445 for (j
= 0; j
< length
* 8; j
++) {
446 if (probably_float(dw
[j
]))
447 printf(" %04.3f", f
[j
]);
449 printf(" 0x%08x", dw
[j
]);
458 handle_3dstate_ps(struct gen_spec
*spec
, uint32_t *p
)
460 uint32_t mask
= ~((1 << 6) - 1);
462 struct brw_instruction
*insns
;
463 static const char unused
[] = "unused";
464 static const char *pixel_type
[3] = {"8 pixel", "16 pixel", "32 pixel"};
465 const char *k0
, *k1
, *k2
;
466 uint32_t k_mask
, k1_offset
, k2_offset
;
468 if (gen_spec_get_gen(spec
) >= gen_make_gen(8, 0)) {
479 #define DISPATCH_16 2
480 #define DISPATCH_32 4
493 case DISPATCH_8
| DISPATCH_16
:
503 case DISPATCH_16
| DISPATCH_32
:
508 case DISPATCH_8
| DISPATCH_16
| DISPATCH_32
:
520 start
= instruction_base
+ (p
[1] & mask
);
521 printf(" Kernel[0] %s\n", k0
);
523 insns
= (struct brw_instruction
*) (gtt
+ start
);
524 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
527 start
= instruction_base
+ (p
[k1_offset
] & mask
);
528 printf(" Kernel[1] %s\n", k1
);
530 insns
= (struct brw_instruction
*) (gtt
+ start
);
531 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
534 start
= instruction_base
+ (p
[k2_offset
] & mask
);
535 printf(" Kernel[2] %s\n", k2
);
537 insns
= (struct brw_instruction
*) (gtt
+ start
);
538 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
543 handle_3dstate_binding_table_pointers(struct gen_spec
*spec
, uint32_t *p
)
545 dump_binding_table(spec
, p
[1]);
549 handle_3dstate_sampler_state_pointers(struct gen_spec
*spec
, uint32_t *p
)
551 dump_samplers(spec
, p
[1]);
555 handle_3dstate_viewport_state_pointers_cc(struct gen_spec
*spec
, uint32_t *p
)
558 struct gen_group
*cc_viewport
;
560 cc_viewport
= gen_spec_find_struct(spec
, "CC_VIEWPORT");
562 start
= dynamic_state_base
+ (p
[1] & ~0x1fu
);
563 for (uint32_t i
= 0; i
< 4; i
++) {
564 printf("viewport %d\n", i
);
565 decode_structure(spec
, cc_viewport
, gtt
+ start
+ i
* 8);
570 handle_3dstate_viewport_state_pointers_sf_clip(struct gen_spec
*spec
,
574 struct gen_group
*sf_clip_viewport
;
576 sf_clip_viewport
= gen_spec_find_struct(spec
, "SF_CLIP_VIEWPORT");
578 start
= dynamic_state_base
+ (p
[1] & ~0x3fu
);
579 for (uint32_t i
= 0; i
< 4; i
++) {
580 printf("viewport %d\n", i
);
581 decode_structure(spec
, sf_clip_viewport
, gtt
+ start
+ i
* 64);
586 handle_3dstate_blend_state_pointers(struct gen_spec
*spec
, uint32_t *p
)
589 struct gen_group
*blend_state
;
591 blend_state
= gen_spec_find_struct(spec
, "BLEND_STATE");
593 start
= dynamic_state_base
+ (p
[1] & ~0x3fu
);
594 decode_structure(spec
, blend_state
, gtt
+ start
);
598 handle_3dstate_cc_state_pointers(struct gen_spec
*spec
, uint32_t *p
)
601 struct gen_group
*cc_state
;
603 cc_state
= gen_spec_find_struct(spec
, "COLOR_CALC_STATE");
605 start
= dynamic_state_base
+ (p
[1] & ~0x3fu
);
606 decode_structure(spec
, cc_state
, gtt
+ start
);
610 handle_3dstate_scissor_state_pointers(struct gen_spec
*spec
, uint32_t *p
)
613 struct gen_group
*scissor_rect
;
615 scissor_rect
= gen_spec_find_struct(spec
, "SCISSOR_RECT");
617 start
= dynamic_state_base
+ (p
[1] & ~0x1fu
);
618 decode_structure(spec
, scissor_rect
, gtt
+ start
);
622 handle_load_register_imm(struct gen_spec
*spec
, uint32_t *p
)
624 struct gen_group
*reg
= gen_spec_find_register(spec
, p
[1]);
627 printf("register %s (0x%x): 0x%x\n",
628 reg
->name
, reg
->register_offset
, p
[2]);
629 decode_structure(spec
, reg
, &p
[2]);
633 #define ARRAY_LENGTH(a) (sizeof (a) / sizeof (a)[0])
635 #define STATE_BASE_ADDRESS 0x61010000
637 #define MEDIA_INTERFACE_DESCRIPTOR_LOAD 0x70020000
639 #define _3DSTATE_INDEX_BUFFER 0x780a0000
640 #define _3DSTATE_VERTEX_BUFFERS 0x78080000
642 #define _3DSTATE_VS 0x78100000
643 #define _3DSTATE_GS 0x78110000
644 #define _3DSTATE_HS 0x781b0000
645 #define _3DSTATE_DS 0x781d0000
647 #define _3DSTATE_CONSTANT_VS 0x78150000
648 #define _3DSTATE_CONSTANT_GS 0x78160000
649 #define _3DSTATE_CONSTANT_PS 0x78170000
650 #define _3DSTATE_CONSTANT_HS 0x78190000
651 #define _3DSTATE_CONSTANT_DS 0x781A0000
653 #define _3DSTATE_PS 0x78200000
655 #define _3DSTATE_BINDING_TABLE_POINTERS_VS 0x78260000
656 #define _3DSTATE_BINDING_TABLE_POINTERS_HS 0x78270000
657 #define _3DSTATE_BINDING_TABLE_POINTERS_DS 0x78280000
658 #define _3DSTATE_BINDING_TABLE_POINTERS_GS 0x78290000
659 #define _3DSTATE_BINDING_TABLE_POINTERS_PS 0x782a0000
661 #define _3DSTATE_SAMPLER_STATE_POINTERS_VS 0x782b0000
662 #define _3DSTATE_SAMPLER_STATE_POINTERS_GS 0x782e0000
663 #define _3DSTATE_SAMPLER_STATE_POINTERS_PS 0x782f0000
665 #define _3DSTATE_VIEWPORT_STATE_POINTERS_CC 0x78230000
666 #define _3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP 0x78210000
667 #define _3DSTATE_BLEND_STATE_POINTERS 0x78240000
668 #define _3DSTATE_CC_STATE_POINTERS 0x780e0000
669 #define _3DSTATE_SCISSOR_STATE_POINTERS 0x780f0000
671 #define _MI_LOAD_REGISTER_IMM 0x11000000
673 struct custom_handler
{
675 void (*handle
)(struct gen_spec
*spec
, uint32_t *p
);
676 } custom_handlers
[] = {
677 { STATE_BASE_ADDRESS
, handle_state_base_address
},
678 { MEDIA_INTERFACE_DESCRIPTOR_LOAD
, handle_media_interface_descriptor_load
},
679 { _3DSTATE_VERTEX_BUFFERS
, handle_3dstate_vertex_buffers
},
680 { _3DSTATE_INDEX_BUFFER
, handle_3dstate_index_buffer
},
681 { _3DSTATE_VS
, handle_3dstate_vs
},
682 { _3DSTATE_GS
, handle_3dstate_vs
},
683 { _3DSTATE_DS
, handle_3dstate_vs
},
684 { _3DSTATE_HS
, handle_3dstate_hs
},
685 { _3DSTATE_CONSTANT_VS
, handle_3dstate_constant
},
686 { _3DSTATE_CONSTANT_GS
, handle_3dstate_constant
},
687 { _3DSTATE_CONSTANT_PS
, handle_3dstate_constant
},
688 { _3DSTATE_CONSTANT_HS
, handle_3dstate_constant
},
689 { _3DSTATE_CONSTANT_DS
, handle_3dstate_constant
},
690 { _3DSTATE_PS
, handle_3dstate_ps
},
692 { _3DSTATE_BINDING_TABLE_POINTERS_VS
, handle_3dstate_binding_table_pointers
},
693 { _3DSTATE_BINDING_TABLE_POINTERS_HS
, handle_3dstate_binding_table_pointers
},
694 { _3DSTATE_BINDING_TABLE_POINTERS_DS
, handle_3dstate_binding_table_pointers
},
695 { _3DSTATE_BINDING_TABLE_POINTERS_GS
, handle_3dstate_binding_table_pointers
},
696 { _3DSTATE_BINDING_TABLE_POINTERS_PS
, handle_3dstate_binding_table_pointers
},
698 { _3DSTATE_SAMPLER_STATE_POINTERS_VS
, handle_3dstate_sampler_state_pointers
},
699 { _3DSTATE_SAMPLER_STATE_POINTERS_GS
, handle_3dstate_sampler_state_pointers
},
700 { _3DSTATE_SAMPLER_STATE_POINTERS_PS
, handle_3dstate_sampler_state_pointers
},
702 { _3DSTATE_VIEWPORT_STATE_POINTERS_CC
, handle_3dstate_viewport_state_pointers_cc
},
703 { _3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP
, handle_3dstate_viewport_state_pointers_sf_clip
},
704 { _3DSTATE_BLEND_STATE_POINTERS
, handle_3dstate_blend_state_pointers
},
705 { _3DSTATE_CC_STATE_POINTERS
, handle_3dstate_cc_state_pointers
},
706 { _3DSTATE_SCISSOR_STATE_POINTERS
, handle_3dstate_scissor_state_pointers
},
707 { _MI_LOAD_REGISTER_IMM
, handle_load_register_imm
}
711 parse_commands(struct gen_spec
*spec
, uint32_t *cmds
, int size
, int engine
)
713 uint32_t *p
, *end
= cmds
+ size
/ 4;
714 unsigned int length
, i
;
715 struct gen_group
*inst
;
717 for (p
= cmds
; p
< end
; p
+= length
) {
718 inst
= gen_spec_find_instruction(spec
, p
);
720 printf("unknown instruction %08x\n", p
[0]);
721 length
= (p
[0] & 0xff) + 2;
724 length
= gen_group_get_length(inst
, p
);
726 const char *color
, *reset_color
= NORMAL
;
729 if (option_full_decode
)
734 if (option_color
== COLOR_NEVER
) {
739 if (option_print_offsets
)
740 offset
= (void *) p
- gtt
;
744 printf("%s0x%08"PRIx64
": 0x%08x: %-80s%s\n",
746 gen_group_get_name(inst
), reset_color
);
748 if (option_full_decode
) {
749 struct gen_field_iterator iter
;
751 int idx
= 0, dword_num
= 0;
752 gen_field_iterator_init(&iter
, inst
, p
);
753 while (gen_field_iterator_next(&iter
)) {
755 print_dword_val(&iter
, offset
, &dword_num
);
757 token
= print_iterator_values(&iter
, &idx
);
759 printf("0x%08"PRIx64
": 0x%08x : Dword %d\n",
760 offset
+ 4 * idx
, p
[idx
], idx
);
761 handle_struct_decode(spec
,token
, &p
[idx
]);
766 for (i
= 0; i
< ARRAY_LENGTH(custom_handlers
); i
++) {
767 if (gen_group_get_opcode(inst
) ==
768 custom_handlers
[i
].opcode
)
769 custom_handlers
[i
].handle(spec
, p
);
773 if ((p
[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_START
) {
775 if (gen_spec_get_gen(spec
) >= gen_make_gen(8,0))
776 start
= get_qword(&p
[1]);
780 parse_commands(spec
, gtt
+ start
, 1 << 20, engine
);
781 } else if ((p
[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_END
) {
787 #define GEN_ENGINE_RENDER 1
788 #define GEN_ENGINE_BLITTER 2
791 handle_trace_block(struct gen_spec
*spec
, uint32_t *p
)
793 int operation
= p
[1] & AUB_TRACE_OPERATION_MASK
;
794 int type
= p
[1] & AUB_TRACE_TYPE_MASK
;
795 int address_space
= p
[1] & AUB_TRACE_ADDRESS_SPACE_MASK
;
796 uint64_t offset
= p
[3];
797 uint32_t size
= p
[4];
798 int header_length
= p
[0] & 0xffff;
799 uint32_t *data
= p
+ header_length
+ 2;
800 int engine
= GEN_ENGINE_RENDER
;
802 if (gen_spec_get_gen(spec
) >= gen_make_gen(8,0))
803 offset
+= (uint64_t) p
[5] << 32;
806 case AUB_TRACE_OP_DATA_WRITE
:
807 if (address_space
!= AUB_TRACE_MEMTYPE_GTT
)
809 if (gtt_size
< offset
+ size
) {
810 fprintf(stderr
, "overflow gtt space: %s\n", strerror(errno
));
813 memcpy((char *) gtt
+ offset
, data
, size
);
814 if (gtt_end
< offset
+ size
)
815 gtt_end
= offset
+ size
;
817 case AUB_TRACE_OP_COMMAND_WRITE
:
819 case AUB_TRACE_TYPE_RING_PRB0
:
820 engine
= GEN_ENGINE_RENDER
;
822 case AUB_TRACE_TYPE_RING_PRB2
:
823 engine
= GEN_ENGINE_BLITTER
;
826 printf("command write to unknown ring %d\n", type
);
830 parse_commands(spec
, data
, size
, engine
);
840 uint32_t *map
, *end
, *cursor
;
843 static struct aub_file
*
844 aub_file_open(const char *filename
)
846 struct aub_file
*file
;
848 file
= malloc(sizeof *file
);
849 file
->filename
= strdup(filename
);
850 file
->fd
= open(file
->filename
, O_RDONLY
);
851 if (file
->fd
== -1) {
852 fprintf(stderr
, "open %s failed: %s\n", file
->filename
, strerror(errno
));
856 if (fstat(file
->fd
, &file
->sb
) == -1) {
857 fprintf(stderr
, "stat failed: %s\n", strerror(errno
));
861 file
->map
= mmap(NULL
, file
->sb
.st_size
,
862 PROT_READ
, MAP_SHARED
, file
->fd
, 0);
863 if (file
->map
== MAP_FAILED
) {
864 fprintf(stderr
, "mmap failed: %s\n", strerror(errno
));
868 file
->cursor
= file
->map
;
869 file
->end
= file
->map
+ file
->sb
.st_size
/ 4;
871 /* mmap a terabyte for our gtt space. */
872 gtt_size
= 1ul << 40;
873 gtt
= mmap(NULL
, gtt_size
, PROT_READ
| PROT_WRITE
,
874 MAP_PRIVATE
| MAP_ANONYMOUS
| MAP_NORESERVE
, -1, 0);
875 if (gtt
== MAP_FAILED
) {
876 fprintf(stderr
, "failed to alloc gtt space: %s\n", strerror(errno
));
883 #define TYPE(dw) (((dw) >> 29) & 7)
884 #define OPCODE(dw) (((dw) >> 23) & 0x3f)
885 #define SUBOPCODE(dw) (((dw) >> 16) & 0x7f)
887 #define MAKE_HEADER(type, opcode, subopcode) \
888 (((type) << 29) | ((opcode) << 23) | ((subopcode) << 16))
892 /* Classic AUB opcodes */
893 #define OPCODE_AUB 0x01
894 #define SUBOPCODE_HEADER 0x05
895 #define SUBOPCODE_BLOCK 0x41
896 #define SUBOPCODE_BMP 0x1e
898 /* Newer version AUB opcode */
899 #define OPCODE_NEW_AUB 0x2e
900 #define SUBOPCODE_VERSION 0x00
901 #define SUBOPCODE_REG_WRITE 0x03
902 #define SUBOPCODE_MEM_POLL 0x05
903 #define SUBOPCODE_MEM_WRITE 0x06
905 #define MAKE_GEN(major, minor) ( ((major) << 8) | (minor) )
911 { "bwr", MAKE_GEN(4, 0) },
912 { "cln", MAKE_GEN(4, 0) },
913 { "blc", MAKE_GEN(4, 0) },
914 { "ctg", MAKE_GEN(4, 0) },
915 { "el", MAKE_GEN(4, 0) },
916 { "il", MAKE_GEN(4, 0) },
917 { "sbr", MAKE_GEN(6, 0) },
918 { "ivb", MAKE_GEN(7, 0) },
919 { "lrb2", MAKE_GEN(0, 0) },
920 { "hsw", MAKE_GEN(7, 5) },
921 { "vlv", MAKE_GEN(7, 0) },
922 { "bdw", MAKE_GEN(8, 0) },
923 { "skl", MAKE_GEN(9, 0) },
924 { "chv", MAKE_GEN(8, 0) },
925 { "bxt", MAKE_GEN(9, 0) }
929 aub_file_decode_batch(struct aub_file
*file
, struct gen_spec
*spec
)
931 uint32_t *p
, h
, device
, data_type
;
932 int header_length
, payload_size
, bias
;
936 header_length
= h
& 0xffff;
946 printf("unknown opcode %d at %td/%td\n",
947 OPCODE(h
), file
->cursor
- file
->map
,
948 file
->end
- file
->map
);
949 file
->cursor
= file
->end
;
954 switch (h
& 0xffff0000) {
955 case MAKE_HEADER(TYPE_AUB
, OPCODE_AUB
, SUBOPCODE_HEADER
):
956 payload_size
= p
[12];
958 case MAKE_HEADER(TYPE_AUB
, OPCODE_AUB
, SUBOPCODE_BLOCK
):
960 handle_trace_block(spec
, p
);
962 case MAKE_HEADER(TYPE_AUB
, OPCODE_AUB
, SUBOPCODE_BMP
):
965 case MAKE_HEADER(TYPE_AUB
, OPCODE_NEW_AUB
, SUBOPCODE_VERSION
):
966 printf("version block: dw1 %08x\n", p
[1]);
967 device
= (p
[1] >> 8) & 0xff;
968 printf(" device %s\n", device_map
[device
].name
);
970 case MAKE_HEADER(TYPE_AUB
, OPCODE_NEW_AUB
, SUBOPCODE_REG_WRITE
):
971 printf("register write block: (dwords %d)\n", h
& 0xffff);
972 printf(" reg 0x%x, data 0x%x\n", p
[1], p
[5]);
974 case MAKE_HEADER(TYPE_AUB
, OPCODE_NEW_AUB
, SUBOPCODE_MEM_WRITE
):
975 printf("memory write block (dwords %d):\n", h
& 0xffff);
976 printf(" address 0x%"PRIx64
"\n", *(uint64_t *) &p
[1]);
977 data_type
= (p
[3] >> 20) & 0xff;
979 printf(" data type 0x%x\n", data_type
);
980 printf(" address space 0x%x\n", (p
[3] >> 28) & 0xf);
982 case MAKE_HEADER(TYPE_AUB
, OPCODE_NEW_AUB
, SUBOPCODE_MEM_POLL
):
983 printf("memory poll block (dwords %d):\n", h
& 0xffff);
986 printf("unknown block type=0x%x, opcode=0x%x, "
987 "subopcode=0x%x (%08x)\n", TYPE(h
), OPCODE(h
), SUBOPCODE(h
), h
);
990 file
->cursor
= p
+ header_length
+ bias
+ payload_size
/ 4;
994 aub_file_more_stuff(struct aub_file
*file
)
996 return file
->cursor
< file
->end
;
1008 if (pipe(fds
) == -1)
1018 execlp("less", "less", "-FRSi", NULL
);
1027 print_help(const char *progname
, FILE *file
)
1030 "Usage: %s [OPTION]... FILE\n"
1031 "Decode aub file contents.\n\n"
1032 "A valid --gen option must be provided.\n\n"
1033 " --help display this help and exit\n"
1034 " --gen=platform decode for given platform (ivb, byt, hsw, bdw, chv, skl, kbl or bxt)\n"
1035 " --headers decode only command headers\n"
1036 " --color[=WHEN] colorize the output; WHEN can be 'auto' (default\n"
1037 " if omitted), 'always', or 'never'\n"
1038 " --no-pager don't launch pager\n"
1039 " --no-offsets don't print instruction offsets\n",
1044 is_prefix(const char *arg
, const char *prefix
, const char **value
)
1046 int l
= strlen(prefix
);
1048 if (strncmp(arg
, prefix
, l
) == 0 && (arg
[l
] == '\0' || arg
[l
] == '=')) {
1050 *value
= arg
+ l
+ 1;
1060 int main(int argc
, char *argv
[])
1062 struct gen_spec
*spec
;
1063 struct aub_file
*file
;
1065 bool found_arg_gen
= false, pager
= true;
1066 const char *value
, *input_file
;
1067 char gen_file
[256], gen_val
[24];
1074 { "ivb", 0x0166, 7, 0 }, /* Intel(R) Ivybridge Mobile GT2 */
1075 { "hsw", 0x0416, 7, 5 }, /* Intel(R) Haswell Mobile GT2 */
1076 { "byt", 0x0155, 7, 5 }, /* Intel(R) Bay Trail */
1077 { "bdw", 0x1616, 8, 0 }, /* Intel(R) HD Graphics 5500 (Broadwell GT2) */
1078 { "chv", 0x22B3, 8, 0 }, /* Intel(R) HD Graphics (Cherryview) */
1079 { "skl", 0x1912, 9, 0 }, /* Intel(R) HD Graphics 530 (Skylake GT2) */
1080 { "kbl", 0x591D, 9, 0 }, /* Intel(R) Kabylake GT2 */
1081 { "bxt", 0x0A84, 9, 0 } /* Intel(R) HD Graphics (Broxton) */
1085 print_help(argv
[0], stderr
);
1089 for (i
= 1; i
< argc
; ++i
) {
1090 if (strcmp(argv
[i
], "--no-pager") == 0) {
1092 } else if (strcmp(argv
[i
], "--no-offsets") == 0) {
1093 option_print_offsets
= false;
1094 } else if (is_prefix(argv
[i
], "--gen", &value
)) {
1095 if (value
== NULL
) {
1096 fprintf(stderr
, "option '--gen' requires an argument\n");
1099 found_arg_gen
= true;
1100 snprintf(gen_val
, sizeof(gen_val
), "%s", value
);
1101 } else if (strcmp(argv
[i
], "--headers") == 0) {
1102 option_full_decode
= false;
1103 } else if (is_prefix(argv
[i
], "--color", &value
)) {
1104 if (value
== NULL
|| strcmp(value
, "always") == 0)
1105 option_color
= COLOR_ALWAYS
;
1106 else if (strcmp(value
, "never") == 0)
1107 option_color
= COLOR_NEVER
;
1108 else if (strcmp(value
, "auto") == 0)
1109 option_color
= COLOR_AUTO
;
1111 fprintf(stderr
, "invalid value for --color: %s", value
);
1114 } else if (strcmp(argv
[i
], "--help") == 0) {
1115 print_help(argv
[0], stdout
);
1118 if (argv
[i
][0] == '-') {
1119 fprintf(stderr
, "unknown option %s\n", argv
[i
]);
1122 input_file
= argv
[i
];
1127 if (!found_arg_gen
) {
1128 fprintf(stderr
, "argument --gen is required\n");
1132 for (i
= 0; i
< ARRAY_SIZE(gens
); i
++) {
1133 if (!strcmp(gen_val
, gens
[i
].name
)) {
1140 fprintf(stderr
, "can't parse gen: %s, expected ivb, byt, hsw, "
1141 "bdw, chv, skl, kbl or bxt\n", gen_val
);
1145 /* Do this before we redirect stdout to pager. */
1146 if (option_color
== COLOR_AUTO
)
1147 option_color
= isatty(1) ? COLOR_ALWAYS
: COLOR_NEVER
;
1149 if (isatty(1) && pager
)
1152 if (gen
->minor
> 0) {
1153 snprintf(gen_file
, sizeof(gen_file
), "../genxml/gen%d%d.xml",
1154 gen
->major
, gen
->minor
);
1156 snprintf(gen_file
, sizeof(gen_file
), "../genxml/gen%d.xml", gen
->major
);
1159 spec
= gen_spec_load(gen_file
);
1160 disasm
= gen_disasm_create(gen
->pci_id
);
1162 if (input_file
== NULL
) {
1163 print_help(input_file
, stderr
);
1166 file
= aub_file_open(input_file
);
1169 while (aub_file_more_stuff(file
))
1170 aub_file_decode_batch(file
, spec
);
1173 /* close the stdout which is opened to write the output */
1178 return EXIT_SUCCESS
;