runtime/debug: TestPanicOnFault consistently failing on netbsd-arm-bsiegert

[bcmills]

allocating objects
starting gc
starting dump
done dump
--- FAIL: TestPanicOnFault (0.00s)
    panic_test.go:45: fault does not contain address
FAIL
FAIL	runtime/debug	0.166s

2021-03-15T10:23:23-6ccb5c4/netbsd-arm-bsiegert
2021-03-14T21:50:39-d0d38f0/netbsd-arm-bsiegert
2021-03-14T21:10:45-3cdd5c3/netbsd-arm-bsiegert
2021-03-14T19:58:14-a8b59fe/netbsd-arm-bsiegert
2021-03-14T17:56:50-061a690/netbsd-arm-bsiegert
2021-03-14T13:09:57-88b8a16/netbsd-arm-bsiegert
2021-03-13T19:03:29-7936efe/netbsd-arm-bsiegert
2021-03-13T18:35:01-1767d2c/netbsd-arm-bsiegert
2021-03-13T18:09:48-4bd4dfe/netbsd-arm-bsiegert
2021-03-13T18:03:52-8336c31/netbsd-arm-bsiegert
2021-03-13T18:02:24-3224990/netbsd-arm-bsiegert
2021-03-13T17:07:07-fedb494/netbsd-arm-bsiegert
2021-03-13T11:56:59-289d34a/netbsd-arm-bsiegert
2021-03-13T11:26:35-83e79c7/netbsd-arm-bsiegert
2021-03-13T11:26:16-59e0129/netbsd-arm-bsiegert
2021-03-13T04:53:32-a8a8528/netbsd-arm-bsiegert
2021-03-13T03:38:42-8e725f8/netbsd-arm-bsiegert
2021-03-13T03:16:55-73eb27b/netbsd-arm-bsiegert
2021-03-13T00:39:23-7588ef0/netbsd-arm-bsiegert
2021-03-13T00:38:58-16ad1ea/netbsd-arm-bsiegert
2021-03-13T00:32:00-7b47f9a/netbsd-arm-bsiegert
2021-03-12T22:45:44-3eebc26/netbsd-arm-bsiegert

This builder has apparently been offline for a while, so it's not clear to me when this regression started.

CC @bsiegert @coypoop @golang/release

[bsiegert]

I meant to write about this, you beat me to it :)

While the builder was offline, I updated it to the latest NetBSD-current (9.99.80), which is when this failure started appearing. (Before that, it was running 8.99.24, from 2018.)

I'll try to see if it is reproducible with the latest stable release (9.1).

[bsiegert]

The builder is running NetBSD-9.1 now. Unfortunately, the error message is still the same: https://build.golang.org/log/4978e756d9fde52031a84fb8c0dd112f535cf01a

[randall77]

Try writing a C program that faults, and grab the faulting address in the C signal handler. See if you can make that work.

[thanm]

@bsiegert this same problem has been happening for a couple of months now. Any update? Seems that this builder is not really providing a good "signal" if it has been failing for so long. It is not really a good use of developer's time to have to check each CL to make sure that it's just the same "known" failure as opposed to something new.

[bsiegert]

Sorry for the lack of updates.

I looked into this just yesterday -- what is happening is that we do not get the fault address from the kernel any more, even though SetPanicOnFault has been called. However, I could not find a reason why that would be the case. I suspect a regression in the OS, since it started happening after a kernel update.

I would like to send a CL to skip the test for now, so that the builder can once again provide meaningful signal. Would that be okay?

[thanm]

SGTM

[gopherbot]

Change https://golang.org/cl/335909 mentions this issue: runtime/debug: skip TestPanicOnFault on netbsd/arm