Index: sys/x86/include/mca.h =================================================================== --- sys/x86/include/mca.h +++ sys/x86/include/mca.h @@ -46,6 +46,31 @@ int mr_cpu; }; +enum mca_stat_types { + MCA_T_NONE = 0, + MCA_T_UNCLASSIFIED, + MCA_T_UCODE_ROM_PARITY, + MCA_T_EXTERNAL, + MCA_T_FRC, + MCA_T_INTERNAL_PARITY, + MCA_T_SMM_HANDLER, + MCA_T_INTERNAL_TIMER, + MCA_T_GENERIC_IO, + MCA_T_INTERNAL, + MCA_T_MEMORY, + MCA_T_TLB, + MCA_T_MEMCONTROLLER_GEN, + MCA_T_MEMCONTROLLER_RD, + MCA_T_MEMCONTROLLER_WR, + MCA_T_MEMCONTROLLER_AC, + MCA_T_MEMCONTROLLER_MS, + MCA_T_MEMCONTROLLER_OTHER, + MCA_T_CACHE, + MCA_T_BUS, + MCA_T_UNKNOWN, + MCA_T_COUNT /* Must stay last */ +}; + #ifdef _KERNEL void cmc_intr(void); Index: sys/x86/x86/mca.c =================================================================== --- sys/x86/x86/mca.c +++ sys/x86/x86/mca.c @@ -132,6 +132,10 @@ static struct timeout_task mca_scan_task; static struct mtx mca_lock; +static uint64_t mca_stats[MCA_T_COUNT]; +SYSCTL_OPAQUE(_hw_mca, OID_AUTO, stats, CTLFLAG_RD | CTLFLAG_SKIP, mca_stats, + MCA_T_COUNT * sizeof(mca_stats[0]), "S", "Array of MCA events by type"); + static unsigned int mca_ia32_ctl_reg(int bank) { @@ -354,21 +358,27 @@ } static const char * -mca_error_mmtype(uint16_t mca_error) +mca_error_mmtype(uint16_t mca_error, enum mca_stat_types *event_type) { switch ((mca_error & 0x70) >> 4) { case 0x0: + *event_type = MCA_T_MEMCONTROLLER_GEN; return ("GEN"); case 0x1: + *event_type = MCA_T_MEMCONTROLLER_RD; return ("RD"); case 0x2: + *event_type = MCA_T_MEMCONTROLLER_WR; return ("WR"); case 0x3: + *event_type = MCA_T_MEMCONTROLLER_AC; return ("AC"); case 0x4: + *event_type = MCA_T_MEMCONTROLLER_MS; return ("MS"); } + *event_type = MCA_T_MEMCONTROLLER_OTHER; return ("???"); } @@ -424,6 +434,7 @@ mca_log(const struct mca_record *rec) { uint16_t mca_error; + enum mca_stat_types event_type; if (mca_mute(rec)) return; @@ -464,31 +475,40 @@ if (rec->mr_status & MC_STATUS_OVER) printf("OVER "); mca_error = rec->mr_status & MC_STATUS_MCA_ERROR; + event_type = MCA_T_COUNT; switch (mca_error) { /* Simple error codes. */ case 0x0000: printf("no error"); + event_type = MCA_T_NONE; break; case 0x0001: printf("unclassified error"); + event_type = MCA_T_UNCLASSIFIED; break; case 0x0002: printf("ucode ROM parity error"); + event_type = MCA_T_UCODE_ROM_PARITY; break; case 0x0003: printf("external error"); + event_type = MCA_T_EXTERNAL; break; case 0x0004: printf("FRC error"); + event_type = MCA_T_FRC; break; case 0x0005: printf("internal parity error"); + event_type = MCA_T_INTERNAL_PARITY; break; case 0x0006: printf("SMM handler code access violation"); + event_type = MCA_T_SMM_HANDLER; break; case 0x0400: printf("internal timer error"); + event_type = MCA_T_INTERNAL_TIMER; break; case 0x0e0b: printf("generic I/O error"); @@ -500,10 +520,12 @@ (int)((rec->mr_misc & MC_MISC_PCIE_SLOT) >> 19), (int)((rec->mr_misc & MC_MISC_PCIE_FUNC) >> 16)); } + event_type = MCA_T_GENERIC_IO; break; default: if ((mca_error & 0xfc00) == 0x0400) { printf("internal error %x", mca_error & 0x03ff); + event_type = MCA_T_INTERNAL; break; } @@ -512,6 +534,7 @@ /* Memory hierarchy error. */ if ((mca_error & 0xeffc) == 0x000c) { printf("%s memory error", mca_error_level(mca_error)); + event_type = MCA_T_MEMORY; break; } @@ -519,12 +542,14 @@ if ((mca_error & 0xeff0) == 0x0010) { printf("%sTLB %s error", mca_error_ttype(mca_error), mca_error_level(mca_error)); + event_type = MCA_T_TLB; break; } /* Memory controller error. */ if ((mca_error & 0xef80) == 0x0080) { - printf("%s channel ", mca_error_mmtype(mca_error)); + printf("%s channel ", mca_error_mmtype(mca_error, + &event_type)); if ((mca_error & 0x000f) != 0x000f) printf("%d", mca_error & 0x000f); else @@ -539,12 +564,14 @@ mca_error_ttype(mca_error), mca_error_level(mca_error), mca_error_request(mca_error)); + event_type = MCA_T_CACHE; break; } /* Extended memory error. */ if ((mca_error & 0xef80) == 0x0280) { - printf("%s channel ", mca_error_mmtype(mca_error)); + printf("%s channel ", mca_error_mmtype(mca_error, + &event_type)); if ((mca_error & 0x000f) != 0x000f) printf("%d", mca_error & 0x000f); else @@ -556,6 +583,7 @@ /* Bus and/or Interconnect error. */ if ((mca_error & 0xe800) == 0x0800) { printf("BUS%s ", mca_error_level(mca_error)); + event_type = MCA_T_BUS; switch ((mca_error & 0x0600) >> 9) { case 0: printf("Source"); @@ -591,6 +619,7 @@ } printf("unknown error %x", mca_error); + event_type = MCA_T_UNKNOWN; break; } printf("\n"); @@ -606,6 +635,11 @@ } if (rec->mr_status & MC_STATUS_MISCV) printf("MCA: Misc 0x%llx\n", (long long)rec->mr_misc); + KASSERT(event_type >= 0 && event_type < MCA_T_COUNT, + ("%s: invalid event type (%d)", __func__, event_type)); + if (event_type < 0 || event_type >= MCA_T_COUNT) + event_type = MCA_T_UNKNOWN; + atomic_add_64(&mca_stats[event_type], 1); } static bool