2 * Copyright 2015 Advanced Micro Devices, Inc.
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
31 #define VG(x) ((void)0)
35 #include "sid_tables.h"
36 #include "util/u_math.h"
37 #include "util/u_memory.h"
38 #include "util/u_string.h"
43 /* Parsed IBs are difficult to read without colors. Use "less -R file" to
44 * read them, or use "aha -b -f file" to convert them to html.
46 #define COLOR_RESET "\033[0m"
47 #define COLOR_RED "\033[31m"
48 #define COLOR_GREEN "\033[1;32m"
49 #define COLOR_YELLOW "\033[1;33m"
50 #define COLOR_CYAN "\033[1;36m"
59 unsigned trace_id_count
;
60 enum chip_class chip_class
;
61 ac_debug_addr_callback addr_callback
;
62 void *addr_callback_data
;
67 static void ac_do_parse_ib(FILE *f
, struct ac_ib_parser
*ib
);
69 static void print_spaces(FILE *f
, unsigned num
)
71 fprintf(f
, "%*s", num
, "");
74 static void print_value(FILE *file
, uint32_t value
, int bits
)
76 /* Guess if it's int or float */
77 if (value
<= (1 << 15)) {
79 fprintf(file
, "%u\n", value
);
81 fprintf(file
, "%u (0x%0*x)\n", value
, bits
/ 4, value
);
85 if (fabs(f
) < 100000 && f
* 10 == floor(f
* 10))
86 fprintf(file
, "%.1ff (0x%0*x)\n", f
, bits
/ 4, value
);
88 /* Don't print more leading zeros than there are bits. */
89 fprintf(file
, "0x%0*x\n", bits
/ 4, value
);
93 static void print_named_value(FILE *file
, const char *name
, uint32_t value
, int bits
)
95 print_spaces(file
, INDENT_PKT
);
96 fprintf(file
, COLOR_YELLOW
"%s" COLOR_RESET
" <- ", name
);
97 print_value(file
, value
, bits
);
100 static const struct si_reg
*find_register(enum chip_class chip_class
, unsigned offset
)
102 const struct si_reg
*table
;
105 switch (chip_class
) {
108 table
= gfx10_reg_table
;
109 table_size
= ARRAY_SIZE(gfx10_reg_table
);
112 table
= gfx9_reg_table
;
113 table_size
= ARRAY_SIZE(gfx9_reg_table
);
116 table
= gfx8_reg_table
;
117 table_size
= ARRAY_SIZE(gfx8_reg_table
);
120 table
= gfx7_reg_table
;
121 table_size
= ARRAY_SIZE(gfx7_reg_table
);
124 table
= gfx6_reg_table
;
125 table_size
= ARRAY_SIZE(gfx6_reg_table
);
131 for (unsigned i
= 0; i
< table_size
; i
++) {
132 const struct si_reg
*reg
= &table
[i
];
134 if (reg
->offset
== offset
)
141 const char *ac_get_register_name(enum chip_class chip_class
, unsigned offset
)
143 const struct si_reg
*reg
= find_register(chip_class
, offset
);
145 return reg
? sid_strings
+ reg
->name_offset
: "(no name)";
148 void ac_dump_reg(FILE *file
, enum chip_class chip_class
, unsigned offset
, uint32_t value
,
151 const struct si_reg
*reg
= find_register(chip_class
, offset
);
154 const char *reg_name
= sid_strings
+ reg
->name_offset
;
155 bool first_field
= true;
157 print_spaces(file
, INDENT_PKT
);
158 fprintf(file
, COLOR_YELLOW
"%s" COLOR_RESET
" <- ", reg_name
);
160 if (!reg
->num_fields
) {
161 print_value(file
, value
, 32);
165 for (unsigned f
= 0; f
< reg
->num_fields
; f
++) {
166 const struct si_field
*field
= sid_fields_table
+ reg
->fields_offset
+ f
;
167 const int *values_offsets
= sid_strings_offsets
+ field
->values_offset
;
168 uint32_t val
= (value
& field
->mask
) >> (ffs(field
->mask
) - 1);
170 if (!(field
->mask
& field_mask
))
173 /* Indent the field. */
175 print_spaces(file
, INDENT_PKT
+ strlen(reg_name
) + 4);
177 /* Print the field. */
178 fprintf(file
, "%s = ", sid_strings
+ field
->name_offset
);
180 if (val
< field
->num_values
&& values_offsets
[val
] >= 0)
181 fprintf(file
, "%s\n", sid_strings
+ values_offsets
[val
]);
183 print_value(file
, val
, util_bitcount(field
->mask
));
190 print_spaces(file
, INDENT_PKT
);
191 fprintf(file
, COLOR_YELLOW
"0x%05x" COLOR_RESET
" <- 0x%08x\n", offset
, value
);
194 static uint32_t ac_ib_get(struct ac_ib_parser
*ib
)
198 if (ib
->cur_dw
< ib
->num_dw
) {
199 v
= ib
->ib
[ib
->cur_dw
];
201 /* Help figure out where garbage data is written to IBs.
203 * Arguably we should do this already when the IBs are written,
204 * see RADEON_VALGRIND. The problem is that client-requests to
205 * Valgrind have an overhead even when Valgrind isn't running,
206 * and radeon_emit is performance sensitive...
208 if (VALGRIND_CHECK_VALUE_IS_DEFINED(v
))
209 fprintf(ib
->f
, COLOR_RED
"Valgrind: The next DWORD is garbage" COLOR_RESET
"\n");
211 fprintf(ib
->f
, "\n\035#%08x ", v
);
213 fprintf(ib
->f
, "\n\035#???????? ");
220 static void ac_parse_set_reg_packet(FILE *f
, unsigned count
, unsigned reg_offset
,
221 struct ac_ib_parser
*ib
)
223 unsigned reg_dw
= ac_ib_get(ib
);
224 unsigned reg
= ((reg_dw
& 0xFFFF) << 2) + reg_offset
;
225 unsigned index
= reg_dw
>> 28;
229 print_spaces(f
, INDENT_PKT
);
230 fprintf(f
, "INDEX = %u\n", index
);
233 for (i
= 0; i
< count
; i
++)
234 ac_dump_reg(f
, ib
->chip_class
, reg
+ i
* 4, ac_ib_get(ib
), ~0);
237 static void ac_parse_packet3(FILE *f
, uint32_t header
, struct ac_ib_parser
*ib
,
238 int *current_trace_id
)
240 unsigned first_dw
= ib
->cur_dw
;
241 int count
= PKT_COUNT_G(header
);
242 unsigned op
= PKT3_IT_OPCODE_G(header
);
243 const char *predicate
= PKT3_PREDICATE(header
) ? "(predicate)" : "";
246 /* Print the name first. */
247 for (i
= 0; i
< ARRAY_SIZE(packet3_table
); i
++)
248 if (packet3_table
[i
].op
== op
)
251 if (i
< ARRAY_SIZE(packet3_table
)) {
252 const char *name
= sid_strings
+ packet3_table
[i
].name_offset
;
254 if (op
== PKT3_SET_CONTEXT_REG
|| op
== PKT3_SET_CONFIG_REG
|| op
== PKT3_SET_UCONFIG_REG
||
255 op
== PKT3_SET_UCONFIG_REG_INDEX
|| op
== PKT3_SET_SH_REG
)
256 fprintf(f
, COLOR_CYAN
"%s%s" COLOR_CYAN
":\n", name
, predicate
);
258 fprintf(f
, COLOR_GREEN
"%s%s" COLOR_RESET
":\n", name
, predicate
);
260 fprintf(f
, COLOR_RED
"PKT3_UNKNOWN 0x%x%s" COLOR_RESET
":\n", op
, predicate
);
262 /* Print the contents. */
264 case PKT3_SET_CONTEXT_REG
:
265 ac_parse_set_reg_packet(f
, count
, SI_CONTEXT_REG_OFFSET
, ib
);
267 case PKT3_SET_CONFIG_REG
:
268 ac_parse_set_reg_packet(f
, count
, SI_CONFIG_REG_OFFSET
, ib
);
270 case PKT3_SET_UCONFIG_REG
:
271 case PKT3_SET_UCONFIG_REG_INDEX
:
272 ac_parse_set_reg_packet(f
, count
, CIK_UCONFIG_REG_OFFSET
, ib
);
274 case PKT3_SET_SH_REG
:
275 ac_parse_set_reg_packet(f
, count
, SI_SH_REG_OFFSET
, ib
);
277 case PKT3_ACQUIRE_MEM
:
278 ac_dump_reg(f
, ib
->chip_class
, R_0301F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
279 ac_dump_reg(f
, ib
->chip_class
, R_0301F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
280 ac_dump_reg(f
, ib
->chip_class
, R_030230_CP_COHER_SIZE_HI
, ac_ib_get(ib
), ~0);
281 ac_dump_reg(f
, ib
->chip_class
, R_0301F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
282 ac_dump_reg(f
, ib
->chip_class
, R_0301E4_CP_COHER_BASE_HI
, ac_ib_get(ib
), ~0);
283 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
284 if (ib
->chip_class
>= GFX10
)
285 ac_dump_reg(f
, ib
->chip_class
, R_586_GCR_CNTL
, ac_ib_get(ib
), ~0);
287 case PKT3_SURFACE_SYNC
:
288 if (ib
->chip_class
>= GFX7
) {
289 ac_dump_reg(f
, ib
->chip_class
, R_0301F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
290 ac_dump_reg(f
, ib
->chip_class
, R_0301F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
291 ac_dump_reg(f
, ib
->chip_class
, R_0301F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
293 ac_dump_reg(f
, ib
->chip_class
, R_0085F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
294 ac_dump_reg(f
, ib
->chip_class
, R_0085F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
295 ac_dump_reg(f
, ib
->chip_class
, R_0085F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
297 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
299 case PKT3_EVENT_WRITE
: {
300 uint32_t event_dw
= ac_ib_get(ib
);
301 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
302 S_028A90_EVENT_TYPE(~0));
303 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
304 print_named_value(f
, "INV_L2", (event_dw
>> 20) & 0x1, 1);
306 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
307 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 16);
311 case PKT3_EVENT_WRITE_EOP
: {
312 uint32_t event_dw
= ac_ib_get(ib
);
313 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
314 S_028A90_EVENT_TYPE(~0));
315 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
316 print_named_value(f
, "TCL1_VOL_ACTION_ENA", (event_dw
>> 12) & 0x1, 1);
317 print_named_value(f
, "TC_VOL_ACTION_ENA", (event_dw
>> 13) & 0x1, 1);
318 print_named_value(f
, "TC_WB_ACTION_ENA", (event_dw
>> 15) & 0x1, 1);
319 print_named_value(f
, "TCL1_ACTION_ENA", (event_dw
>> 16) & 0x1, 1);
320 print_named_value(f
, "TC_ACTION_ENA", (event_dw
>> 17) & 0x1, 1);
321 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
322 uint32_t addr_hi_dw
= ac_ib_get(ib
);
323 print_named_value(f
, "ADDRESS_HI", addr_hi_dw
, 16);
324 print_named_value(f
, "DST_SEL", (addr_hi_dw
>> 16) & 0x3, 2);
325 print_named_value(f
, "INT_SEL", (addr_hi_dw
>> 24) & 0x7, 3);
326 print_named_value(f
, "DATA_SEL", addr_hi_dw
>> 29, 3);
327 print_named_value(f
, "DATA_LO", ac_ib_get(ib
), 32);
328 print_named_value(f
, "DATA_HI", ac_ib_get(ib
), 32);
331 case PKT3_RELEASE_MEM
: {
332 uint32_t event_dw
= ac_ib_get(ib
);
333 if (ib
->chip_class
>= GFX10
) {
334 ac_dump_reg(f
, ib
->chip_class
, R_490_RELEASE_MEM_OP
, event_dw
, ~0u);
336 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
337 S_028A90_EVENT_TYPE(~0));
338 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
339 print_named_value(f
, "TCL1_VOL_ACTION_ENA", (event_dw
>> 12) & 0x1, 1);
340 print_named_value(f
, "TC_VOL_ACTION_ENA", (event_dw
>> 13) & 0x1, 1);
341 print_named_value(f
, "TC_WB_ACTION_ENA", (event_dw
>> 15) & 0x1, 1);
342 print_named_value(f
, "TCL1_ACTION_ENA", (event_dw
>> 16) & 0x1, 1);
343 print_named_value(f
, "TC_ACTION_ENA", (event_dw
>> 17) & 0x1, 1);
344 print_named_value(f
, "TC_NC_ACTION_ENA", (event_dw
>> 19) & 0x1, 1);
345 print_named_value(f
, "TC_WC_ACTION_ENA", (event_dw
>> 20) & 0x1, 1);
346 print_named_value(f
, "TC_MD_ACTION_ENA", (event_dw
>> 21) & 0x1, 1);
348 uint32_t sel_dw
= ac_ib_get(ib
);
349 print_named_value(f
, "DST_SEL", (sel_dw
>> 16) & 0x3, 2);
350 print_named_value(f
, "INT_SEL", (sel_dw
>> 24) & 0x7, 3);
351 print_named_value(f
, "DATA_SEL", sel_dw
>> 29, 3);
352 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
353 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 32);
354 print_named_value(f
, "DATA_LO", ac_ib_get(ib
), 32);
355 print_named_value(f
, "DATA_HI", ac_ib_get(ib
), 32);
356 print_named_value(f
, "CTXID", ac_ib_get(ib
), 32);
359 case PKT3_WAIT_REG_MEM
:
360 print_named_value(f
, "OP", ac_ib_get(ib
), 32);
361 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
362 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 32);
363 print_named_value(f
, "REF", ac_ib_get(ib
), 32);
364 print_named_value(f
, "MASK", ac_ib_get(ib
), 32);
365 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
367 case PKT3_DRAW_INDEX_AUTO
:
368 ac_dump_reg(f
, ib
->chip_class
, R_030930_VGT_NUM_INDICES
, ac_ib_get(ib
), ~0);
369 ac_dump_reg(f
, ib
->chip_class
, R_0287F0_VGT_DRAW_INITIATOR
, ac_ib_get(ib
), ~0);
371 case PKT3_DRAW_INDEX_2
:
372 ac_dump_reg(f
, ib
->chip_class
, R_028A78_VGT_DMA_MAX_SIZE
, ac_ib_get(ib
), ~0);
373 ac_dump_reg(f
, ib
->chip_class
, R_0287E8_VGT_DMA_BASE
, ac_ib_get(ib
), ~0);
374 ac_dump_reg(f
, ib
->chip_class
, R_0287E4_VGT_DMA_BASE_HI
, ac_ib_get(ib
), ~0);
375 ac_dump_reg(f
, ib
->chip_class
, R_030930_VGT_NUM_INDICES
, ac_ib_get(ib
), ~0);
376 ac_dump_reg(f
, ib
->chip_class
, R_0287F0_VGT_DRAW_INITIATOR
, ac_ib_get(ib
), ~0);
378 case PKT3_INDEX_TYPE
:
379 ac_dump_reg(f
, ib
->chip_class
, R_028A7C_VGT_DMA_INDEX_TYPE
, ac_ib_get(ib
), ~0);
381 case PKT3_NUM_INSTANCES
:
382 ac_dump_reg(f
, ib
->chip_class
, R_030934_VGT_NUM_INSTANCES
, ac_ib_get(ib
), ~0);
384 case PKT3_WRITE_DATA
:
385 ac_dump_reg(f
, ib
->chip_class
, R_370_CONTROL
, ac_ib_get(ib
), ~0);
386 ac_dump_reg(f
, ib
->chip_class
, R_371_DST_ADDR_LO
, ac_ib_get(ib
), ~0);
387 ac_dump_reg(f
, ib
->chip_class
, R_372_DST_ADDR_HI
, ac_ib_get(ib
), ~0);
388 /* The payload is written automatically */
391 ac_dump_reg(f
, ib
->chip_class
, R_410_CP_DMA_WORD0
, ac_ib_get(ib
), ~0);
392 ac_dump_reg(f
, ib
->chip_class
, R_411_CP_DMA_WORD1
, ac_ib_get(ib
), ~0);
393 ac_dump_reg(f
, ib
->chip_class
, R_412_CP_DMA_WORD2
, ac_ib_get(ib
), ~0);
394 ac_dump_reg(f
, ib
->chip_class
, R_413_CP_DMA_WORD3
, ac_ib_get(ib
), ~0);
395 ac_dump_reg(f
, ib
->chip_class
, R_414_COMMAND
, ac_ib_get(ib
), ~0);
398 ac_dump_reg(f
, ib
->chip_class
, R_500_DMA_DATA_WORD0
, ac_ib_get(ib
), ~0);
399 ac_dump_reg(f
, ib
->chip_class
, R_501_SRC_ADDR_LO
, ac_ib_get(ib
), ~0);
400 ac_dump_reg(f
, ib
->chip_class
, R_502_SRC_ADDR_HI
, ac_ib_get(ib
), ~0);
401 ac_dump_reg(f
, ib
->chip_class
, R_503_DST_ADDR_LO
, ac_ib_get(ib
), ~0);
402 ac_dump_reg(f
, ib
->chip_class
, R_504_DST_ADDR_HI
, ac_ib_get(ib
), ~0);
403 ac_dump_reg(f
, ib
->chip_class
, R_414_COMMAND
, ac_ib_get(ib
), ~0);
405 case PKT3_INDIRECT_BUFFER_SI
:
406 case PKT3_INDIRECT_BUFFER_CONST
:
407 case PKT3_INDIRECT_BUFFER_CIK
: {
408 uint32_t base_lo_dw
= ac_ib_get(ib
);
409 ac_dump_reg(f
, ib
->chip_class
, R_3F0_IB_BASE_LO
, base_lo_dw
, ~0);
410 uint32_t base_hi_dw
= ac_ib_get(ib
);
411 ac_dump_reg(f
, ib
->chip_class
, R_3F1_IB_BASE_HI
, base_hi_dw
, ~0);
412 uint32_t control_dw
= ac_ib_get(ib
);
413 ac_dump_reg(f
, ib
->chip_class
, R_3F2_IB_CONTROL
, control_dw
, ~0);
415 if (!ib
->addr_callback
)
418 uint64_t addr
= ((uint64_t)base_hi_dw
<< 32) | base_lo_dw
;
419 void *data
= ib
->addr_callback(ib
->addr_callback_data
, addr
);
423 if (G_3F2_CHAIN(control_dw
)) {
425 ib
->num_dw
= G_3F2_IB_SIZE(control_dw
);
430 struct ac_ib_parser ib_recurse
;
431 memcpy(&ib_recurse
, ib
, sizeof(ib_recurse
));
432 ib_recurse
.ib
= data
;
433 ib_recurse
.num_dw
= G_3F2_IB_SIZE(control_dw
);
434 ib_recurse
.cur_dw
= 0;
435 if (ib_recurse
.trace_id_count
) {
436 if (*current_trace_id
== *ib
->trace_ids
) {
437 ++ib_recurse
.trace_ids
;
438 --ib_recurse
.trace_id_count
;
440 ib_recurse
.trace_id_count
= 0;
444 fprintf(f
, "\n\035>------------------ nested begin ------------------\n");
445 ac_do_parse_ib(f
, &ib_recurse
);
446 fprintf(f
, "\n\035<------------------- nested end -------------------\n");
449 case PKT3_CLEAR_STATE
:
450 case PKT3_INCREMENT_DE_COUNTER
:
451 case PKT3_PFP_SYNC_ME
:
454 if (header
== PKT3_NOP_PAD
) {
455 count
= -1; /* One dword NOP. */
456 } else if (count
== 0 && ib
->cur_dw
< ib
->num_dw
&& AC_IS_TRACE_POINT(ib
->ib
[ib
->cur_dw
])) {
457 unsigned packet_id
= AC_GET_TRACE_POINT_ID(ib
->ib
[ib
->cur_dw
]);
459 print_spaces(f
, INDENT_PKT
);
460 fprintf(f
, COLOR_RED
"Trace point ID: %u\n", packet_id
);
462 if (!ib
->trace_id_count
)
463 break; /* tracing was disabled */
465 *current_trace_id
= packet_id
;
467 print_spaces(f
, INDENT_PKT
);
468 if (packet_id
< *ib
->trace_ids
)
469 fprintf(f
, COLOR_RED
"This trace point was reached by the CP." COLOR_RESET
"\n");
470 else if (packet_id
== *ib
->trace_ids
)
471 fprintf(f
, COLOR_RED
"!!!!! This is the last trace point that "
472 "was reached by the CP !!!!!" COLOR_RESET
"\n");
473 else if (packet_id
+ 1 == *ib
->trace_ids
)
474 fprintf(f
, COLOR_RED
"!!!!! This is the first trace point that "
475 "was NOT been reached by the CP !!!!!" COLOR_RESET
"\n");
477 fprintf(f
, COLOR_RED
"!!!!! This trace point was NOT reached "
478 "by the CP !!!!!" COLOR_RESET
"\n");
484 /* print additional dwords */
485 while (ib
->cur_dw
<= first_dw
+ count
)
488 if (ib
->cur_dw
> first_dw
+ count
+ 1)
489 fprintf(f
, COLOR_RED
"\n!!!!! count in header too low !!!!!" COLOR_RESET
"\n");
493 * Parse and print an IB into a file.
495 static void ac_do_parse_ib(FILE *f
, struct ac_ib_parser
*ib
)
497 int current_trace_id
= -1;
499 while (ib
->cur_dw
< ib
->num_dw
) {
500 uint32_t header
= ac_ib_get(ib
);
501 unsigned type
= PKT_TYPE_G(header
);
505 ac_parse_packet3(f
, header
, ib
, ¤t_trace_id
);
509 if (header
== 0x80000000) {
510 fprintf(f
, COLOR_GREEN
"NOP (type 2)" COLOR_RESET
"\n");
515 fprintf(f
, "Unknown packet type %i\n", type
);
521 static void format_ib_output(FILE *f
, char *out
)
528 if (out
[0] == '\n' && out
[1] == '\035')
530 if (out
[0] == '\035') {
538 unsigned indent
= 4 * depth
;
543 print_spaces(f
, indent
);
545 char *end
= strchrnul(out
, '\n');
546 fwrite(out
, end
- out
, 1, f
);
547 fputc('\n', f
); /* always end with a new line */
559 * Parse and print an IB into a file.
563 * \param num_dw size of the IB
564 * \param chip_class chip class
565 * \param trace_ids the last trace IDs that are known to have been reached
566 * and executed by the CP, typically read from a buffer
567 * \param trace_id_count The number of entries in the trace_ids array.
568 * \param addr_callback Get a mapped pointer of the IB at a given address. Can
570 * \param addr_callback_data user data for addr_callback
572 void ac_parse_ib_chunk(FILE *f
, uint32_t *ib_ptr
, int num_dw
, const int *trace_ids
,
573 unsigned trace_id_count
, enum chip_class chip_class
,
574 ac_debug_addr_callback addr_callback
, void *addr_callback_data
)
576 struct ac_ib_parser ib
= {};
579 ib
.trace_ids
= trace_ids
;
580 ib
.trace_id_count
= trace_id_count
;
581 ib
.chip_class
= chip_class
;
582 ib
.addr_callback
= addr_callback
;
583 ib
.addr_callback_data
= addr_callback_data
;
587 FILE *memf
= open_memstream(&out
, &outsize
);
589 ac_do_parse_ib(memf
, &ib
);
593 format_ib_output(f
, out
);
597 if (ib
.cur_dw
> ib
.num_dw
) {
598 printf("\nPacket ends after the end of IB.\n");
604 * Parse and print an IB into a file.
608 * \param num_dw size of the IB
609 * \param chip_class chip class
610 * \param trace_ids the last trace IDs that are known to have been reached
611 * and executed by the CP, typically read from a buffer
612 * \param trace_id_count The number of entries in the trace_ids array.
613 * \param addr_callback Get a mapped pointer of the IB at a given address. Can
615 * \param addr_callback_data user data for addr_callback
617 void ac_parse_ib(FILE *f
, uint32_t *ib
, int num_dw
, const int *trace_ids
, unsigned trace_id_count
,
618 const char *name
, enum chip_class chip_class
, ac_debug_addr_callback addr_callback
,
619 void *addr_callback_data
)
621 fprintf(f
, "------------------ %s begin ------------------\n", name
);
623 ac_parse_ib_chunk(f
, ib
, num_dw
, trace_ids
, trace_id_count
, chip_class
, addr_callback
,
626 fprintf(f
, "------------------- %s end -------------------\n\n", name
);
630 * Parse dmesg and return TRUE if a VM fault has been detected.
632 * \param chip_class chip class
633 * \param old_dmesg_timestamp previous dmesg timestamp parsed at init time
634 * \param out_addr detected VM fault addr
636 bool ac_vm_fault_occured(enum chip_class chip_class
, uint64_t *old_dmesg_timestamp
,
642 uint64_t dmesg_timestamp
= 0;
645 FILE *p
= popen("dmesg", "r");
649 while (fgets(line
, sizeof(line
), p
)) {
652 if (!line
[0] || line
[0] == '\n')
655 /* Get the timestamp. */
656 if (sscanf(line
, "[%u.%u]", &sec
, &usec
) != 2) {
657 static bool hit
= false;
659 fprintf(stderr
, "%s: failed to parse line '%s'\n", __func__
, line
);
664 dmesg_timestamp
= sec
* 1000000ull + usec
;
666 /* If just updating the timestamp. */
670 /* Process messages only if the timestamp is newer. */
671 if (dmesg_timestamp
<= *old_dmesg_timestamp
)
674 /* Only process the first VM fault. */
678 /* Remove trailing \n */
680 if (len
&& line
[len
- 1] == '\n')
683 /* Get the message part. */
684 msg
= strchr(line
, ']');
689 const char *header_line
, *addr_line_prefix
, *addr_line_format
;
691 if (chip_class
>= GFX9
) {
693 * ..: [gfxhub] VMC page fault (src_id:0 ring:158 vm_id:2 pas_id:0)
694 * ..: at page 0x0000000219f8f000 from 27
695 * ..: VM_L2_PROTECTION_FAULT_STATUS:0x0020113C
697 header_line
= "VMC page fault";
698 addr_line_prefix
= " at page";
699 addr_line_format
= "%" PRIx64
;
701 header_line
= "GPU fault detected:";
702 addr_line_prefix
= "VM_CONTEXT1_PROTECTION_FAULT_ADDR";
703 addr_line_format
= "%" PRIX64
;
708 if (strstr(msg
, header_line
))
712 msg
= strstr(msg
, addr_line_prefix
);
714 msg
= strstr(msg
, "0x");
717 if (sscanf(msg
, addr_line_format
, out_addr
) == 1)
729 if (dmesg_timestamp
> *old_dmesg_timestamp
)
730 *old_dmesg_timestamp
= dmesg_timestamp
;
735 static int compare_wave(const void *p1
, const void *p2
)
737 struct ac_wave_info
*w1
= (struct ac_wave_info
*)p1
;
738 struct ac_wave_info
*w2
= (struct ac_wave_info
*)p2
;
740 /* Sort waves according to PC and then SE, SH, CU, etc. */
757 if (w1
->simd
< w2
->simd
)
759 if (w1
->simd
> w2
->simd
)
761 if (w1
->wave
< w2
->wave
)
763 if (w1
->wave
> w2
->wave
)
769 /* Return wave information. "waves" should be a large enough array. */
770 unsigned ac_get_wave_info(enum chip_class chip_class
,
771 struct ac_wave_info waves
[AC_MAX_WAVES_PER_CHIP
])
773 char line
[2000], cmd
[128];
774 unsigned num_waves
= 0;
776 sprintf(cmd
, "umr -O halt_waves -wa %s", chip_class
>= GFX10
? "gfx_0.0.0" : "gfx");
778 FILE *p
= popen(cmd
, "r");
782 if (!fgets(line
, sizeof(line
), p
) || strncmp(line
, "SE", 2) != 0) {
787 while (fgets(line
, sizeof(line
), p
)) {
788 struct ac_wave_info
*w
;
789 uint32_t pc_hi
, pc_lo
, exec_hi
, exec_lo
;
791 assert(num_waves
< AC_MAX_WAVES_PER_CHIP
);
792 w
= &waves
[num_waves
];
794 if (sscanf(line
, "%u %u %u %u %u %x %x %x %x %x %x %x", &w
->se
, &w
->sh
, &w
->cu
, &w
->simd
,
795 &w
->wave
, &w
->status
, &pc_hi
, &pc_lo
, &w
->inst_dw0
, &w
->inst_dw1
, &exec_hi
,
797 w
->pc
= ((uint64_t)pc_hi
<< 32) | pc_lo
;
798 w
->exec
= ((uint64_t)exec_hi
<< 32) | exec_lo
;
804 qsort(waves
, num_waves
, sizeof(struct ac_wave_info
), compare_wave
);