[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CANA+-vCPOSWq5o6c3kT=jdLmyyJW9P9_-ZORPus6aFbFoBNUkA@mail.gmail.com>
Date: Mon, 11 Mar 2019 21:57:42 -0700
From: Tri Vo <trong@...roid.com>
To: Peter Oberparleiter <oberpar@...ux.ibm.com>
Cc: Greg Hackmann <ghackmann@...roid.com>,
Nick Desaulniers <ndesaulniers@...gle.com>,
LKML <linux-kernel@...r.kernel.org>, kernel-team@...roid.com,
Masahiro Yamada <yamada.masahiro@...ionext.com>,
Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [PATCH v3 0/3] gcov: add Clang support
On Tue, Mar 5, 2019 at 6:29 AM Peter Oberparleiter
<oberpar@...ux.ibm.com> wrote:
>
> On 23.01.2019 00:37, Tri Vo wrote:
> > This patch series adds Clang supoprt for gcov.
> >
> > Patch 1 refactors existing code in preparation for Clang support.
> > Patch 2 implements necessary LLVM runtime hooks and gcov kernel interfaces.
> > Patch 3 updates documentation.
>
> Thanks for the updates! I've provided suggestions for some minor
> improvements in my other review e-mails.
>
> With those changes applied, the patch set is in my opinion ready for
> inclusion into the mainline kernel. My suggestion would be to re-post
> the resulting version while putting Andrew Morton on CC as gcov changes
> are typically integrated via his tree.
>
> Also I've successfully re-tested this patch set version on s390 using
> GCC 7.3.0 to ensure that the existing GCC support is still working.
> Unfortunately I wasn't able to test the Clang version due to some
> compile problems on s390 (unrelated to this patch set).
>
> If you haven't done so, I would like to suggest to run the following
> tests with the Clang gcov-kernel version that should catch problems in
> some corner cases:
>
> 1. Unload a module, then use llvm-cov on the associated coverage file.
>
> Expectation: correct llvm-cov output including coverage of module
> exit code.
>
> 2. Unload a module, modify its source, re-compile it and load it again,
> then use llvm-cov on the associated coverage file.
>
> Expectation: kernel message "discarding saved data", correct llvm-cov
> output with no coverage of module exit code.
>
> 3. Unload a module, then reset all coverage data by writing to
> /sys/kernel/debug/gcov/reset.
>
> Expectation: all coverage files associated with the module are
> removed from debugfs.
Thanks for the suggested test cases! The current patchset doesn't seem
to handle module loading/unloading correctly. I'll fix that in a
follow-up.
Powered by blists - more mailing lists