When a MCE happens that is to be logged onto bank 4 of AMD multi-node
processors, they are reported only to corresponding node base core of
the cpu on which the error occurred.
Refer D18F3x44[NbMcaToMstCpuEn] on BKDGs of Fam10h and later for
clarifications on the reporting of MC4 errors only to NBC MSRs.
We don't have the exception handler wired up to handle this currently.
As a consequence, do_machine_check only runs on the core on which error
occurred and (since according to the BKDGs, reads to MC4_STATUS MSR of
non-NBC will simply RAZ) the exception is ignored for the core.
This is a problem as now we have dropped MCEs.
I tested this on Fam10h and Fam15h using mce_amd_inj and by triggering
a real HW MCE using Boris' new interface; And can confirm the behavior.
This patch set fixes the issue by looking at the NBC MSRs when bank 4
errors happen on AMD multi node processors.
Patch 1: Refactor AMD cpu topology functions so that we can get some
relevant info that we need to use in EDAC, MC handler routines
Patch 2: The fix to our problem
Patch 3: Modify mce_amd_inj interfaces to write to only NBC for bank 4
errors. Only then will they be picked up for error handling.
Aravind Gopalakrishnan (3):
x86,amd: Refactor amd cpu topology functions for multi-node processors
x86, mce: Handle AMD MCE on bank4 on NBC for multi-node processors
edac, mce_amd_inj: Inject errors only on NBC for bank 4 errors
arch/x86/include/asm/processor.h | 1 +
arch/x86/kernel/cpu/amd.c | 78 ++++++++++++++----
arch/x86/kernel/cpu/mcheck/mce.c | 167 +++++++++++++++++++++++++++++++++++----
drivers/edac/mce_amd_inj.c | 21 ++++-
4 files changed, 235 insertions(+), 32 deletions(-)
--
2.0.2
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
