irqbalance01 reports broken - RHEL9 about ltp HOT 12 CLOSED

@wangli5665 any change you could have look into it?

from ltp.

@mpw5421 Any chance running this under gdb and getting a backtrace? See: https://github.com/linux-test-project/ltp/#debugging-with-gdb

from ltp.

@metan-ucw I read the link you provided, but wasn't exactly sure how to execute that.

I ran the test manually under strace since I knew how to do that from troubleshooting other LTP testcases I opened bugs for against RedHat. The output is attached.

strace.ltp.irqbalance.txt

from ltp.

Doesn't reproduce for me on RHEL9.3. Could you try running it under valgrind? (e.g. valgrind ./irqbalance01)

from ltp.

Also, is this x86_64 or ppc64le?

from ltp.

Also, is this x86_64 or ppc64le?

It's actually s390x on LPAR. Sorry for not including that in the original comment.

Sure, valgrind output attached:

valgrind.ltp.irqbalance.txt

from ltp.

@mpw5421 Thanks, that looks like we have a bug in test, which fails to parse /proc/interrupts. Could you please attach also output of "cat /proc/interrupts"?

from ltp.

It appears to be enough to move named interrupts to begging of file to trigger it on x86 as well. "row" variable keeps incrementing and eventually there's a write out-of-bounds:

irqbalance01.c:129: TINFO: Found 8 CPUS, 45 IRQs
realloc(): invalid next size
tst_test.c:1653: TBROK: Test killed by SIGIOT/SIGABRT!

from ltp.

Output from cat /proc/interrupts:

interrupts.txt

from ltp.

Thanks, that confirms my suspicion.
It's a bit too late today, so this could be wrong, but I was thinking:

@@ -154,7 +154,6 @@ static void collect_irq_info(void)
                        if (acc != -1)
                                tst_brk(TBROK, "Unexpected EOL");
                        col = 0;
-                       row++;
                        break;
                case '0' ... '9':
                        if (acc == -1)
@@ -167,6 +166,7 @@ static void collect_irq_info(void)
                        if (acc == -1 || col != 0)
                                tst_brk(TBROK, "Unexpected ':'");
                        irq_ids[row] = acc;
+                       row++;
                        acc = -1;
                        break;
                default:

CC @richiejp

from ltp.

To be honest I can't remember how it works and I'm not working on LTP at the moment. However it sounds plausible because the parser made a lot of assumptions about the file format.

from ltp.

v1 patch posted to ML: https://lists.linux.it/pipermail/ltp/2024-January/036666.html

from ltp.