1 #!/bin/sh 1 #!/bin/sh
2 # SPDX-License-Identifier: GPL-2.0-only 2 # SPDX-License-Identifier: GPL-2.0-only
3 # 3 #
4 # Copyright 2015, Daniel Axtens, IBM Corporati 4 # Copyright 2015, Daniel Axtens, IBM Corporation
5 # 5 #
6 6
7 7
8 # do we have ./getscom, ./putscom? 8 # do we have ./getscom, ./putscom?
9 if [ -x ./getscom ] && [ -x ./putscom ]; then 9 if [ -x ./getscom ] && [ -x ./putscom ]; then
10 GETSCOM=./getscom 10 GETSCOM=./getscom
11 PUTSCOM=./putscom 11 PUTSCOM=./putscom
12 elif which getscom > /dev/null; then 12 elif which getscom > /dev/null; then
13 GETSCOM=$(which getscom) 13 GETSCOM=$(which getscom)
14 PUTSCOM=$(which putscom) 14 PUTSCOM=$(which putscom)
15 else 15 else
16 cat <<EOF 16 cat <<EOF
17 Can't find getscom/putscom in . or \$PATH. 17 Can't find getscom/putscom in . or \$PATH.
18 See https://github.com/open-power/skiboot. 18 See https://github.com/open-power/skiboot.
19 The tool is in external/xscom-utils 19 The tool is in external/xscom-utils
20 EOF 20 EOF
21 exit 1 21 exit 1
22 fi 22 fi
23 23
24 # We will get 8 HMI events per injection 24 # We will get 8 HMI events per injection
25 # todo: deal with things being offline 25 # todo: deal with things being offline
26 expected_hmis=8 26 expected_hmis=8
27 COUNT_HMIS() { 27 COUNT_HMIS() {
28 dmesg | grep -c 'Harmless Hypervisor Maint 28 dmesg | grep -c 'Harmless Hypervisor Maintenance interrupt'
29 } 29 }
30 30
31 # massively expand snooze delay, allowing inje 31 # massively expand snooze delay, allowing injection on all cores
32 ppc64_cpu --smt-snooze-delay=1000000000 32 ppc64_cpu --smt-snooze-delay=1000000000
33 33
34 # when we exit, restore it 34 # when we exit, restore it
35 trap "ppc64_cpu --smt-snooze-delay=100" 0 1 35 trap "ppc64_cpu --smt-snooze-delay=100" 0 1
36 36
37 # for each chip+core combination 37 # for each chip+core combination
38 # todo - less fragile parsing 38 # todo - less fragile parsing
39 grep -E -o 'OCC: Chip [0-9a-f]+ Core [0-9a-f]' !! 39 egrep -o 'OCC: Chip [0-9a-f]+ Core [0-9a-f]' < /sys/firmware/opal/msglog |
40 while read chipcore; do 40 while read chipcore; do
41 chip=$(echo "$chipcore"|awk '{print $3 41 chip=$(echo "$chipcore"|awk '{print $3}')
42 core=$(echo "$chipcore"|awk '{print $5 42 core=$(echo "$chipcore"|awk '{print $5}')
43 fir="0x1${core}013100" 43 fir="0x1${core}013100"
44 44
45 # verify that Core FIR is zero as expe 45 # verify that Core FIR is zero as expected
46 if [ "$($GETSCOM -c 0x${chip} $fir)" ! 46 if [ "$($GETSCOM -c 0x${chip} $fir)" != 0 ]; then
47 echo "FIR was not zero before 47 echo "FIR was not zero before injection for chip $chip, core $core. Aborting!"
48 echo "Result of $GETSCOM -c 0x 48 echo "Result of $GETSCOM -c 0x${chip} $fir:"
49 $GETSCOM -c 0x${chip} $fir 49 $GETSCOM -c 0x${chip} $fir
50 echo "If you get a -5 error, t 50 echo "If you get a -5 error, the core may be in idle state. Try stress-ng."
51 echo "Otherwise, try $PUTSCOM 51 echo "Otherwise, try $PUTSCOM -c 0x${chip} $fir 0"
52 exit 1 52 exit 1
53 fi 53 fi
54 54
55 # keep track of the number of HMIs han 55 # keep track of the number of HMIs handled
56 old_hmis=$(COUNT_HMIS) 56 old_hmis=$(COUNT_HMIS)
57 57
58 # do injection, adding a marker to dme 58 # do injection, adding a marker to dmesg for clarity
59 echo "Injecting HMI on core $core, chi 59 echo "Injecting HMI on core $core, chip $chip" | tee /dev/kmsg
60 # inject a RegFile recoverable error 60 # inject a RegFile recoverable error
61 if ! $PUTSCOM -c 0x${chip} $fir 200000 61 if ! $PUTSCOM -c 0x${chip} $fir 2000000000000000 > /dev/null; then
62 echo "Error injecting. Abortin 62 echo "Error injecting. Aborting!"
63 exit 1 63 exit 1
64 fi 64 fi
65 65
66 # now we want to wait for all the HMIs 66 # now we want to wait for all the HMIs to be processed
67 # we expect one per thread on the core 67 # we expect one per thread on the core
68 i=0; 68 i=0;
69 new_hmis=$(COUNT_HMIS) 69 new_hmis=$(COUNT_HMIS)
70 while [ $new_hmis -lt $((old_hmis + ex 70 while [ $new_hmis -lt $((old_hmis + expected_hmis)) ] && [ $i -lt 12 ]; do
71 echo "Seen $((new_hmis - old_hmis) 71 echo "Seen $((new_hmis - old_hmis)) HMI(s) out of $expected_hmis expected, sleeping"
72 sleep 5; 72 sleep 5;
73 i=$((i + 1)) 73 i=$((i + 1))
74 new_hmis=$(COUNT_HMIS) 74 new_hmis=$(COUNT_HMIS)
75 done 75 done
76 if [ $i = 12 ]; then 76 if [ $i = 12 ]; then
77 echo "Haven't seen expected $expec 77 echo "Haven't seen expected $expected_hmis recoveries after 1 min. Aborting."
78 exit 1 78 exit 1
79 fi 79 fi
80 echo "Processed $expected_hmis events; 80 echo "Processed $expected_hmis events; presumed success. Check dmesg."
81 echo "" 81 echo ""
82 done 82 done
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.