Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'edac_urgent_for_v6.15_rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras
Pull EDAC fixes from Borislav Petkov:
"Two fixes to the AMD translation library for the MI300 side of things:
- Use the row[13] bit when calculating the memory row to retire
- Mask the physical row address in order to avoid creating duplicate
error records"
* tag 'edac_urgent_for_v6.15_rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras:
RAS/AMD/FMPM: Get masked address
RAS/AMD/ATL: Include row[13] bit in row retirement
+28
-3
+3
drivers/ras/amd/atl/internal.h
+3
drivers/ras/amd/atl/internal.h
+17
-2
drivers/ras/amd/atl/umc.c
+17
-2
drivers/ras/amd/atl/umc.c
···
229
229
* Additionally, the PC and Bank bits may be hashed. This must be accounted for before
230
230
* reconstructing the normalized address.
231
231
*/
232
-
#define MI300_UMC_MCA_COL GENMASK(5, 1)
233
232
#define MI300_UMC_MCA_BANK GENMASK(9, 6)
234
233
#define MI300_UMC_MCA_ROW GENMASK(24, 10)
235
234
#define MI300_UMC_MCA_PC BIT(25)
···
319
320
* See amd_atl::convert_dram_to_norm_addr_mi300() for MI300 address formats.
320
321
*/
321
322
#define MI300_NUM_COL BIT(HWEIGHT(MI300_UMC_MCA_COL))
322
-
static void retire_row_mi300(struct atl_err *a_err)
323
+
static void _retire_row_mi300(struct atl_err *a_err)
323
324
{
324
325
unsigned long addr;
325
326
struct page *p;
···
348
349
349
350
memory_failure(addr, 0);
350
351
}
352
+
}
353
+
354
+
/*
355
+
* In addition to the column bits, the row[13] bit should also be included when
356
+
* calculating addresses affected by a physical row.
357
+
*
358
+
* Instead of running through another loop over a single bit, just run through
359
+
* the column bits twice and flip the row[13] bit in-between.
360
+
*
361
+
* See MI300_UMC_MCA_ROW for the row bits in MCA_ADDR_UMC value.
362
+
*/
363
+
static void retire_row_mi300(struct atl_err *a_err)
364
+
{
365
+
_retire_row_mi300(a_err);
366
+
a_err->addr ^= MI300_UMC_MCA_ROW13;
367
+
_retire_row_mi300(a_err);
351
368
}
352
369
353
370
void amd_retire_dram_row(struct atl_err *a_err)
+8
-1
drivers/ras/amd/fmpm.c
+8
-1
drivers/ras/amd/fmpm.c
···
250
250
return true;
251
251
}
252
252
253
+
/*
254
+
* Row retirement is done on MI300 systems, and some bits are 'don't
255
+
* care' for comparing addresses with unique physical rows. This
256
+
* includes all column bits and the row[13] bit.
257
+
*/
258
+
#define MASK_ADDR(addr) ((addr) & ~(MI300_UMC_MCA_ROW13 | MI300_UMC_MCA_COL))
259
+
253
260
static bool fpds_equal(struct cper_fru_poison_desc *old, struct cper_fru_poison_desc *new)
254
261
{
255
262
/*
···
265
258
*
266
259
* Also, order the checks from most->least likely to fail to shortcut the code.
267
260
*/
268
-
if (old->addr != new->addr)
261
+
if (MASK_ADDR(old->addr) != MASK_ADDR(new->addr))
269
262
return false;
270
263
271
264
if (old->hw_id != new->hw_id)