149 lines
5.7 KiB
Diff
149 lines
5.7 KiB
Diff
commit b497a3d6a39d402c41065e9284d49114b97e3bfe
|
|
Author: Shiju Jose <shiju.jose@huawei.com>
|
|
Date: Mon Mar 8 16:57:28 2021 +0000
|
|
|
|
rasdaemon: ras-mc-ctl: Add memory failure events
|
|
|
|
Add supporting memory failure errors (memory_failure_event)
|
|
to the ras-mc-ctl tool.
|
|
|
|
Sample Log,
|
|
ras-mc-ctl --summary
|
|
...
|
|
Memory failure events summary:
|
|
Delayed errors: 4
|
|
Failed errors: 1
|
|
...
|
|
|
|
ras-mc-ctl --errors
|
|
...
|
|
Memory failure events:
|
|
1 2020-10-28 23:20:41 -0800 error: pfn=0x204000000, page_type=free buddy page, action_result=Delayed
|
|
2 2020-10-28 23:31:38 -0800 error: pfn=0x204000000, page_type=free buddy page, action_result=Delayed
|
|
3 2020-10-28 23:54:54 -0800 error: pfn=0x205000000, page_type=free buddy page, action_result=Delayed
|
|
4 2020-10-29 00:12:25 -0800 error: pfn=0x204000000, page_type=free buddy page, action_result=Delayed
|
|
5 2020-10-29 00:26:36 -0800 error: pfn=0x204000000, page_type=free buddy page, action_result=Failed
|
|
|
|
Signed-off-by: Shiju Jose <shiju.jose@huawei.com>
|
|
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
|
|
|
|
---
|
|
configure.ac | 11 +++++++++++
|
|
util/ras-mc-ctl.in | 46 +++++++++++++++++++++++++++++++++++++++++++---
|
|
2 files changed, 54 insertions(+), 3 deletions(-)
|
|
|
|
--- a/util/ras-mc-ctl.in 2021-10-13 13:51:00.887292563 -0400
|
|
+++ b/util/ras-mc-ctl.in 2021-10-13 13:51:27.536061894 -0400
|
|
@@ -44,11 +44,13 @@ my $modprobe = find_prog ("modprobe")
|
|
my $has_aer = 0;
|
|
my $has_arm = 0;
|
|
my $has_extlog = 0;
|
|
+my $has_mem_failure = 0;
|
|
my $has_mce = 0;
|
|
|
|
@WITH_AER_TRUE@$has_aer = 1;
|
|
@WITH_ARM_TRUE@$has_arm = 1;
|
|
@WITH_EXTLOG_TRUE@$has_extlog = 1;
|
|
+@WITH_MEMORY_FAILURE_TRUE@$has_mem_failure = 1;
|
|
@WITH_MCE_TRUE@$has_mce = 1;
|
|
|
|
my %conf = ();
|
|
@@ -1132,7 +1134,7 @@ sub summary
|
|
{
|
|
require DBI;
|
|
my ($query, $query_handle, $out);
|
|
- my ($err_type, $label, $mc, $top, $mid, $low, $count, $msg);
|
|
+ my ($err_type, $label, $mc, $top, $mid, $low, $count, $msg, $action_result);
|
|
my ($etype, $severity, $etype_string, $severity_string);
|
|
my ($affinity, $mpidr);
|
|
|
|
@@ -1203,9 +1205,27 @@ sub summary
|
|
$out .= "\t$count $etype_string $severity_string errors\n";
|
|
}
|
|
if ($out ne "") {
|
|
- print "Extlog records summary:\n$out";
|
|
+ print "Extlog records summary:\n$out\n";
|
|
} else {
|
|
- print "No Extlog errors.\n";
|
|
+ print "No Extlog errors.\n\n";
|
|
+ }
|
|
+ $query_handle->finish;
|
|
+ }
|
|
+
|
|
+ # Memory failure errors
|
|
+ if ($has_mem_failure == 1) {
|
|
+ $query = "select action_result, count(*) from memory_failure_event group by action_result";
|
|
+ $query_handle = $dbh->prepare($query);
|
|
+ $query_handle->execute();
|
|
+ $query_handle->bind_columns(\($action_result, $count));
|
|
+ $out = "";
|
|
+ while($query_handle->fetch()) {
|
|
+ $out .= "\t$action_result errors: $count\n";
|
|
+ }
|
|
+ if ($out ne "") {
|
|
+ print "Memory failure events summary:\n$out\n";
|
|
+ } else {
|
|
+ print "No Memory failure errors.\n\n";
|
|
}
|
|
$query_handle->finish;
|
|
}
|
|
@@ -1238,6 +1258,7 @@ sub errors
|
|
my ($mcgcap,$mcgstatus, $status, $misc, $ip, $tsc, $walltime, $cpu, $cpuid, $apicid, $socketid, $cs, $bank, $cpuvendor, $bank_name, $mcgstatus_msg, $mcistatus_msg, $user_action, $mc_location);
|
|
my ($timestamp, $etype, $severity, $etype_string, $severity_string, $fru_id, $fru_text, $cper_data);
|
|
my ($error_count, $affinity, $mpidr, $r_state, $psci_state);
|
|
+ my ($pfn, $page_type, $action_result);
|
|
|
|
my $dbh = DBI->connect("dbi:SQLite:dbname=$dbname", "", "", {});
|
|
|
|
@@ -1329,6 +1350,25 @@ $out .= sprintf "address=0x%08x, ", $add
|
|
}
|
|
$query_handle->finish;
|
|
}
|
|
+
|
|
+ # Memory failure errors
|
|
+ if ($has_mem_failure == 1) {
|
|
+ $query = "select id, timestamp, pfn, page_type, action_result from memory_failure_event order by id";
|
|
+ $query_handle = $dbh->prepare($query);
|
|
+ $query_handle->execute();
|
|
+ $query_handle->bind_columns(\($id, $timestamp, $pfn, $page_type, $action_result));
|
|
+ $out = "";
|
|
+ while($query_handle->fetch()) {
|
|
+ $out .= "$id $timestamp error: ";
|
|
+ $out .= "pfn=$pfn, page_type=$page_type, action_result=$action_result\n";
|
|
+ }
|
|
+ if ($out ne "") {
|
|
+ print "Memory failure events:\n$out\n";
|
|
+ } else {
|
|
+ print "No Memory failure errors.\n\n";
|
|
+ }
|
|
+ $query_handle->finish;
|
|
+ }
|
|
|
|
# MCE mce_record errors
|
|
if ($has_mce == 1) {
|
|
--- a/configure.ac 2018-04-25 06:28:51.000000000 -0400
|
|
+++ b/configure.ac 2021-10-13 13:51:00.916292312 -0400
|
|
@@ -80,6 +80,16 @@ AS_IF([test "x$enable_extlog" = "xyes"],
|
|
])
|
|
AM_CONDITIONAL([WITH_EXTLOG], [test x$enable_extlog = xyes])
|
|
|
|
+AC_ARG_ENABLE([memory_failure],
|
|
+ AS_HELP_STRING([--enable-memory-failure], [enable memory failure events (currently experimental)]))
|
|
+
|
|
+AS_IF([test "x$enable_memory_failure" = "xyes" || test "x$enable_all" == "xyes"], [
|
|
+ AC_DEFINE(HAVE_MEMORY_FAILURE,1,"have memory failure events collect")
|
|
+ AC_SUBST([WITH_MEMORY_FAILURE])
|
|
+])
|
|
+AM_CONDITIONAL([WITH_MEMORY_FAILURE], [test x$enable_memory_failure = xyes || test x$enable_all == xyes])
|
|
+AM_COND_IF([WITH_MEMORY_FAILURE], [USE_MEMORY_FAILURE="yes"], [USE_MEMORY_FAILURE="no"])
|
|
+
|
|
AC_ARG_ENABLE([abrt_report],
|
|
AS_HELP_STRING([--enable-abrt-report], [enable report event to ABRT (currently experimental)]))
|
|
|
|
@@ -127,4 +137,5 @@ compile time options summary
|
|
ABRT report : $enable_abrt_report
|
|
HIP07 SAS HW errors : $enable_hisi_ns_decode
|
|
ARM events : $enable_arm
|
|
+ Memory Failure : $USE_MEMORY_FAILURE
|
|
EOF
|