commit 2a194a220d4a377f2b25e308a01ffe9675b93c31 Author: Krishna Dhulipala Date: Tue Jun 27 12:09:45 2023 -0700 Unified SEL logging of AER events diff --git a/Makefile.am b/Makefile.am index fabca78..1ea3356 100644 --- a/Makefile.am +++ b/Makefile.am @@ -63,13 +63,17 @@ endif if WITH_AMP_NS_DECODE rasdaemon_SOURCES += non-standard-ampere.c endif +if WITH_OPENBMC_UNIFIED_SEL + rasdaemon_SOURCES += unified-sel.c +endif rasdaemon_LDADD = -lpthread $(SQLITE3_LIBS) libtrace/libtrace.a include_HEADERS = config.h ras-events.h ras-logger.h ras-mc-handler.h \ ras-aer-handler.h ras-mce-handler.h ras-record.h bitfield.h ras-report.h \ ras-extlog-handler.h ras-arm-handler.h ras-non-standard-handler.h \ ras-devlink-handler.h ras-diskerror-handler.h rbtree.h ras-page-isolation.h \ - non-standard-hisilicon.h non-standard-ampere.h ras-memory-failure-handler.h + non-standard-hisilicon.h non-standard-ampere.h ras-memory-failure-handler.h \ + unified-sel.h # This rule can't be called with more than one Makefile job (like make -j8) # I can't figure out a way to fix that diff --git a/configure.ac b/configure.ac index f7d1947..4a534b7 100644 --- a/configure.ac +++ b/configure.ac @@ -161,6 +161,16 @@ AS_IF([test "x$enable_amp_ns_decode" = "xyes" || test "x$enable_all" == "xyes"], AM_CONDITIONAL([WITH_AMP_NS_DECODE], [test x$enable_amp_ns_decode = xyes || test x$enable_all == xyes]) AM_COND_IF([WITH_AMP_NS_DECODE], [USE_AMP_NS_DECODE="yes"], [USE_AMP_NS_DECODE="no"]) +AC_ARG_ENABLE([openbmc_unified_sel], + AS_HELP_STRING([--enable-openbmc-unified-sel], [enable OPENBMC_UNIFIED_SEL events (currently exprimental)])) + +AS_IF([test "x$enable_openbmc_unified_sel" = "xyes" || test "x$enable_all" = "xyes"], [ + AC_DEFINE(HAVE_OPENBMC_UNIFIED_SEL,1,"have OpenBMC unified SEL") + AC_SUBST([WITH_OPENBMC_UNIFIED_SEL]) +]) +AM_CONDITIONAL([WITH_OPENBMC_UNIFIED_SEL], [test x$enable_openbmc_unified_sel = xyes || test x$enabl_all = xyes]) +AM_COND_IF([WITH_OPENBMC_UNIFIED_SEL], [USE_OPENBMC_UNIFIED_SEL="yes"], [USE_OPENBMC_UNIFIED_SEL="no"]) + test "$sysconfdir" = '${prefix}/etc' && sysconfdir=/etc CFLAGS="$CFLAGS -Wall -Wmissing-prototypes -Wstrict-prototypes" @@ -201,4 +211,5 @@ compile time options summary Memory Failure : $USE_MEMORY_FAILURE Memory CE PFA : $USE_MEMORY_CE_PFA AMP RAS errors : $USE_AMP_NS_DECODE + OpenBMC unified : $USE_OPENBMC_UNIFIED_SEL EOF diff --git a/ras-aer-handler.c b/ras-aer-handler.c index 8ddd439..c03f6ec 100644 --- a/ras-aer-handler.c +++ b/ras-aer-handler.c @@ -25,6 +25,7 @@ #include "ras-logger.h" #include "bitfield.h" #include "ras-report.h" +#include "unified-sel.h" /* bit field meaning for correctable error */ static const char *aer_cor_errors[32] = { @@ -35,12 +36,15 @@ static const char *aer_cor_errors[32] = { [8] = "RELAY_NUM Rollover", [12] = "Replay Timer Timeout", [13] = "Advisory Non-Fatal", + [14] = "Corrected Internal", + [15] = "Header Log Overflow", }; /* bit field meaning for uncorrectable error */ static const char *aer_uncor_errors[32] = { /* Uncorrectable errors */ [4] = "Data Link Protocol", + [5] = "Surprise Link Down", [12] = "Poisoned TLP", [13] = "Flow Control Protocol", [14] = "Completion Timeout", @@ -50,6 +54,12 @@ static const char *aer_uncor_errors[32] = { [18] = "Malformed TLP", [19] = "ECRC", [20] = "Unsupported Request", + [21] = "ACS Violation", + [22] = "Uncorrected Internal", + [23] = "MC Blocked TLP", + [24] = "AtomicOp Egress Blocked", + [25] = "TLP Prefix Blocked", + [26] = "Poisoned TLP Egrees Blocked", }; #define BUF_LEN 1024 @@ -151,5 +161,10 @@ int ras_aer_event_handler(struct trace_seq *s, ras_report_aer_event(ras, &ev); #endif +#ifdef HAVE_OPENBMC_UNIFIED_SEL + if (openbmc_unified_sel_log(severity_val, ev.dev_name, status_val) < 0) + return -1; +#endif + return 0; } diff --git a/unified-sel.c b/unified-sel.c new file mode 100644 index 0000000..287bb4f --- /dev/null +++ b/unified-sel.c @@ -0,0 +1,114 @@ +/* + * Copyright (c) 2023, Meta Platforms Inc. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + */ + +#include +#include +#include +#include +#include "ras-record.h" +#include "ras-logger.h" +#include "ras-report.h" +#include "unified-sel.h" + +/* CPU Root Port Error ID corresponding to each status bit set */ +static const char *cor_error_ids[32] = { + /* Correctable errors */ + [0] = "0x00", /* Receiver Error */ + [6] = "0x01", /* Bad TLP */ + [7] = "0x02", /* Bad DLLP */ + [8] = "0x04", /* RELAY_NUM Rollover */ + [12] = "0x03", /* Replay Timer Timeout */ + [13] = "0x05", /* Advisory Non-Fatal */ + [14] = "0x06", /* Corrected Internal */ + [15] = "0x07", /* Header Log Overflow */ +}; + +static const char *uncor_error_ids[32] = { + /* Uncorrectable errors */ + [4] = "0x20", /* Data Link Protocol */ + [5] = "0x21", /* Surprise Link Down */ + [12] = "0x22", /* Poisoned TLP */ + [13] = "0x23", /* Flow Control Protocol */ + [14] = "0x24", /* Completion Timeout */ + [15] = "0x25", /* Completer Abort */ + [16] = "0x26", /* Unexpected Completion */ + [17] = "0x27", /* Receiver Overflow */ + [18] = "0x29", /* Malformed TLP */ + [19] = "0x29", /* ECRC */ + [20] = "0x2A", /* Unsupported Request */ + [21] = "0x2B", /* ACS Violation */ + [22] = "0x2C", /* Uncorrected Internal */ + [23] = "0x2D", /* MC Blocked TLP */ + [24] = "0x2E", /* AtomicOp Egress Blocked */ + [25] = "0x2F", /* TLP Prefix Blocked */ + [26] = "0x30", /* Poisoned TLP Egrees Blocked */ +}; + +static int verify_id_log_sel(uint64_t status, + const char **idarray, + unsigned bus, + unsigned dev_fn) +{ + int i; + char openbmc_ipmi_add_sel[105]; + + /* + * Get PCIe AER error source bus/dev/fn and save it to the BMC SEL + * as a OpenBMC unified SEL record type. + * The IPMI command and record fields are defined in IPMI Specification v2.0 (IPMI Spec) + * ipmitool raw 0x0a 0x44 is "Add SEL Entry Command" defined in IPMI spec chapter 31.6 + * The 16 byte that follow form the SEL Record + * defined in IPMI spec chapter 32.1 "SEL Event Records" + * Byte 1~2 are Record ID = 0x00 0x00, unused + * Byte 3 is Record Type = 0xFB, OEM non-timestamped record type for OpenBMC unified SEL + * Byte 4~16 are OEM defined + * Byte 11: + * Byte11[7:3] Device# + * Byte11[2:0] Function# + * Byte 12: Bus number + * Byte 13-15: Reserved + * Byte 16: ID of the error detected on the PCle device that triggered this SEL record + */ + + /* Potentially all error status bits could be set for a given PCIe device. + * Therefore, iterate over all 32 bits each of cor and uncor errors + */ + for (i = 0; i < 32; i++) { + if ((status & (1 << i)) && idarray[i]) { + sprintf(openbmc_ipmi_add_sel, + "ipmitool raw 0x0a 0x44 0x00 0x00 0xFB 0x20 0x00 0x00 0x00 0x00 0x01 0x00 0x%02x 0x%02x 0x01 0x00 0xff %s", + dev_fn, bus, idarray[i]); + if (system(openbmc_ipmi_add_sel) != 0) + return -1; + } + } + return 0; +} + +int openbmc_unified_sel_log(uint64_t severity, const char *dev_name, uint64_t status) +{ + int bus, dev, dev_fn, fn; + + sscanf(dev_name, "%*x:%x:%x.%x", &bus, &dev, &fn); + dev_fn = (((dev & 0x1f) << 3) | (fn & 0x7)); + + /* Use the appropriate correctable or uncorrectable error status ID + * for a gien severity level + */ + if (severity == HW_EVENT_AER_CORRECTED) { + if (verify_id_log_sel(status, cor_error_ids, bus, dev_fn) < 0) + return -1; + } + else { + if (verify_id_log_sel(status, uncor_error_ids, bus, dev_fn) < 0) + return -1; + } + return 0; +} diff --git a/unified-sel.h b/unified-sel.h new file mode 100644 index 0000000..17458a5 --- /dev/null +++ b/unified-sel.h @@ -0,0 +1,17 @@ +/* + * Copyright (c) 2023, Meta Platforms Inc. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + */ + + +#ifndef _UNIFIED_SEL_H +#define _UNIFIED_SEL_H + +int openbmc_unified_sel_log(uint64_t severity, const char *dev_name, uint64_t status); + +#endif