On 27.12.2017 12:55, Martin Husemann wrote:
> On Tue, Dec 26, 2017 at 02:35:45PM -0500, Christos Zoulas wrote:
>> So now it is really only one test case "signal3".
>
> Yes.
>
>> How about this (restoring the test to what it was before):
>>
>> -#if defined(__sparc__)
>> +#if defined(__sparc__) && !defined(__sparc64__)
>> atf_tc_expect_timeout("PR kern/52167");
>
> No, this is not the issue, and it is not sparc64 only.
>
> I think PTRACE_BREAKPOINT_ASM is wrong on arm (will double check), but
> I don't understand what is failing on sparc64, see ktrace with debug
> enabled below.
>
> Martin
>
There were recently changes for golang and signal handling there.. it
looks to me that they affected stability of this test.
I'm still focused on LLVM bits, so I restrict myself temporarily to
keeping the number of unexpected failures 0 and keep this clean for releng.
Unless someone can see a hotfix, I propose to disable it for now -
similar to the suspend* ones.
The original sparc and sparc64 case, was that the kernel was not moving
the instruction pointer post the software trap instruction.
I will work on the ptrace(2) failures when I will be done with LLVM and
that should be in close time. I have a list of known to me bugs that are
not included in the ATF testsuite... so this will be a long debugging
session.
Attachment:
signature.asc
Description: OpenPGP digital signature