Re: [PATCH v2] riscv: add support for SECCOMP and SECCOMP_FILTER

On Wed, Aug 28, 2019 at 10:36 AM Kees Cook [off-list ref] wrote:

On Fri, Aug 23, 2019 at 05:30:53PM -0700, Paul Walmsley wrote:

quoted

On Thu, 22 Aug 2019, David Abdurachmanov wrote:

quoted

There is one failing kernel selftest: global.user_notification_signal

Is this the only failing test?  Or are the rest of the selftests skipped
when this test fails, and no further tests are run, as seems to be shown
here:

  https://lore.kernel.org/linux-riscv/CADnnUqcmDMRe1f+3jG8SPR6jRrnBsY8VVD70VbKEm0NqYeoicA@mail.gmail.com/ (local)

For example, looking at the source, I'd naively expect to see the
user_notification_closed_listener test result -- which follows right
after the failing test in the selftest source.  But there aren't any
results?

Also - could you follow up with the author of this failing test to see if
we can get some more clarity about what might be going wrong here?  It
appears that the failing test was added in commit 6a21cc50f0c7f ("seccomp:
add a return code to trap to userspace") by Tycho Andersen
[off-list ref].

So, the original email says the riscv series is tested on top of 5.2-rc7,
but just for fun, can you confirm that you're building a tree that includes
9dd3fcb0ab73 ("selftests/seccomp: Handle namespace failures gracefully")? I
assume it does, but I suspect something similar is happening, where the
environment is slightly different than expected and the test stalls.

Does it behave the same way under emulation (i.e. can I hope to
reproduce this myself?)

This was tested in 5.2-rc7 and later in 5.3-rc with the same behavior.
Also VM or physical HW doesn't matter, same result.

--
Kees Cook