Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
drm/amdgpu: make convert_ras_err_addr visible outside UMC block
And change some UMC v12 specific functions to generic version, so the
code can be shared.
Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
authored by
Tao Zhou
and committed by
Alex Deucher
4e7812e2
3d60a30c
+78
-58
+63
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
+63
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
···
444
444
445
445
return ret;
446
446
}
447
+
448
+
int amdgpu_umc_lookup_bad_pages_in_a_row(struct amdgpu_device *adev,
449
+
uint64_t pa_addr, uint64_t *pfns, int len)
450
+
{
451
+
uint32_t i, ret = 0, pos = 0;
452
+
struct ta_ras_query_address_output addr_out;
453
+
struct ras_err_data err_data;
454
+
455
+
err_data.err_addr =
456
+
kcalloc(adev->umc.max_ras_err_cnt_per_query,
457
+
sizeof(struct eeprom_table_record), GFP_KERNEL);
458
+
if (!err_data.err_addr) {
459
+
dev_warn(adev->dev, "Failed to alloc memory in bad page lookup!\n");
460
+
return 0;
461
+
}
462
+
463
+
addr_out.pa.pa = pa_addr;
464
+
465
+
if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr)
466
+
adev->umc.ras->convert_ras_err_addr(adev, &err_data, NULL,
467
+
&addr_out, false);
468
+
else
469
+
goto out;
470
+
471
+
for (i = 0; i < adev->umc.max_ras_err_cnt_per_query; i++) {
472
+
if (pos >= len)
473
+
goto out;
474
+
475
+
pfns[pos] = err_data.err_addr[pos].retired_page;
476
+
pos++;
477
+
}
478
+
ret = pos;
479
+
480
+
out:
481
+
kfree(err_data.err_addr);
482
+
return ret;
483
+
}
484
+
485
+
int amdgpu_umc_mca_to_addr(struct amdgpu_device *adev,
486
+
uint64_t err_addr, uint32_t ch, uint32_t umc,
487
+
uint32_t node, uint32_t socket,
488
+
uint64_t *addr, bool dump_addr)
489
+
{
490
+
struct ta_ras_query_address_input addr_in;
491
+
struct ta_ras_query_address_output addr_out;
492
+
493
+
memset(&addr_in, 0, sizeof(addr_in));
494
+
addr_in.ma.err_addr = err_addr;
495
+
addr_in.ma.ch_inst = ch;
496
+
addr_in.ma.umc_inst = umc;
497
+
addr_in.ma.node_inst = node;
498
+
addr_in.ma.socket_id = socket;
499
+
500
+
if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr)
501
+
adev->umc.ras->convert_ras_err_addr(adev, NULL, &addr_in,
502
+
&addr_out, dump_addr);
503
+
else
504
+
return 0;
505
+
506
+
*addr = addr_out.pa.pa;
507
+
508
+
return 0;
509
+
}
+11
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
+11
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
···
70
70
enum amdgpu_mca_error_type type, void *ras_error_status);
71
71
int (*update_ecc_status)(struct amdgpu_device *adev,
72
72
uint64_t status, uint64_t ipid, uint64_t addr);
73
+
void (*convert_ras_err_addr)(struct amdgpu_device *adev,
74
+
struct ras_err_data *err_data,
75
+
struct ta_ras_query_address_input *addr_in,
76
+
struct ta_ras_query_address_output *addr_out,
77
+
bool dump_addr);
73
78
};
74
79
75
80
struct amdgpu_umc_funcs {
···
139
134
140
135
void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
141
136
void *ras_error_status);
137
+
int amdgpu_umc_lookup_bad_pages_in_a_row(struct amdgpu_device *adev,
138
+
uint64_t pa_addr, uint64_t *pfns, int len);
139
+
int amdgpu_umc_mca_to_addr(struct amdgpu_device *adev,
140
+
uint64_t err_addr, uint32_t ch, uint32_t umc,
141
+
uint32_t node, uint32_t socket,
142
+
uint64_t *addr, bool dump_addr);
142
143
#endif
+4
-58
drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
+4
-58
drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
···
243
243
}
244
244
}
245
245
246
-
static int umc_v12_0_lookup_bad_pages_in_a_row(struct amdgpu_device *adev,
247
-
uint64_t pa_addr, uint64_t *pfns, int len)
248
-
{
249
-
uint32_t i, ret = 0, pos = 0;
250
-
struct ta_ras_query_address_output addr_out;
251
-
struct ras_err_data err_data;
252
-
253
-
err_data.err_addr =
254
-
kcalloc(adev->umc.max_ras_err_cnt_per_query,
255
-
sizeof(struct eeprom_table_record), GFP_KERNEL);
256
-
if (!err_data.err_addr) {
257
-
dev_warn(adev->dev, "Failed to alloc memory in bad page lookup!\n");
258
-
return 0;
259
-
}
260
-
261
-
addr_out.pa.pa = pa_addr;
262
-
umc_v12_0_convert_error_address(adev, &err_data, NULL, &addr_out, false);
263
-
264
-
for (i = 0; i < adev->umc.max_ras_err_cnt_per_query; i++) {
265
-
if (pos >= len)
266
-
goto out;
267
-
268
-
pfns[pos] = err_data.err_addr[pos].retired_page;
269
-
pos++;
270
-
}
271
-
ret = pos;
272
-
273
-
out:
274
-
kfree(err_data.err_addr);
275
-
return ret;
276
-
}
277
-
278
-
static int umc_v12_0_convert_mca_to_addr(struct amdgpu_device *adev,
279
-
uint64_t err_addr, uint32_t ch, uint32_t umc,
280
-
uint32_t node, uint32_t socket,
281
-
uint64_t *addr, bool dump_addr)
282
-
{
283
-
struct ta_ras_query_address_input addr_in;
284
-
struct ta_ras_query_address_output addr_out;
285
-
286
-
memset(&addr_in, 0, sizeof(addr_in));
287
-
addr_in.ma.err_addr = err_addr;
288
-
addr_in.ma.ch_inst = ch;
289
-
addr_in.ma.umc_inst = umc;
290
-
addr_in.ma.node_inst = node;
291
-
addr_in.ma.socket_id = socket;
292
-
addr_in.addr_type = TA_RAS_MCA_TO_PA;
293
-
294
-
umc_v12_0_convert_error_address(adev, NULL, &addr_in, &addr_out, dump_addr);
295
-
296
-
*addr = addr_out.pa.pa;
297
-
298
-
return 0;
299
-
}
300
-
301
246
static int umc_v12_0_query_error_address(struct amdgpu_device *adev,
302
247
uint32_t node_inst, uint32_t umc_inst,
303
248
uint32_t ch_inst, void *data)
···
470
525
MCA_IPID_2_UMC_CH(ipid),
471
526
err_addr);
472
527
473
-
ret = umc_v12_0_convert_mca_to_addr(adev,
528
+
ret = amdgpu_umc_mca_to_addr(adev,
474
529
err_addr, MCA_IPID_2_UMC_CH(ipid),
475
530
MCA_IPID_2_UMC_INST(ipid), MCA_IPID_2_DIE_ID(ipid),
476
531
MCA_IPID_2_SOCKET_ID(ipid), &pa_addr, true);
···
504
559
con->umc_ecc_log.de_queried_count++;
505
560
506
561
memset(page_pfn, 0, sizeof(page_pfn));
507
-
count = umc_v12_0_lookup_bad_pages_in_a_row(adev,
562
+
count = amdgpu_umc_lookup_bad_pages_in_a_row(adev,
508
563
pa_addr,
509
564
page_pfn, ARRAY_SIZE(page_pfn));
510
565
if (count <= 0) {
···
547
602
return -EINVAL;
548
603
549
604
memset(page_pfn, 0, sizeof(page_pfn));
550
-
count = umc_v12_0_lookup_bad_pages_in_a_row(adev,
605
+
count = amdgpu_umc_lookup_bad_pages_in_a_row(adev,
551
606
ecc_err->pa_pfn << AMDGPU_GPU_PAGE_SHIFT,
552
607
page_pfn, ARRAY_SIZE(page_pfn));
553
608
···
604
659
.ecc_info_query_ras_error_address = umc_v12_0_query_ras_ecc_err_addr,
605
660
.check_ecc_err_status = umc_v12_0_check_ecc_err_status,
606
661
.update_ecc_status = umc_v12_0_update_ecc_status,
662
+
.convert_ras_err_addr = umc_v12_0_convert_error_address,
607
663
};
608
664