Re: [PATCH 6.6.y 0/6] Backported patches to fix selftest tpdir2

[Greg KH <gregkh@linuxfoundation.org>]

On Fri, Apr 04, 2025 at 03:58:36PM +0800, Kang Wenlin wrote:
> Hi Greg
> 
> Thanks for your response.
> 
> 
> On 4/3/2025 22:52, Greg KH wrote:
> > CAUTION: This email comes from a non Wind River email account!
> > Do not click links or open attachments unless you recognize the sender and know the content is safe.
> > 
> > On Wed, Apr 02, 2025 at 04:26:50PM +0800, Kang Wenlin wrote:
> > > From: Wenlin Kang <wenlin.kang@windriver.com>
> > > 
> > > The selftest tpdir2 terminated with a 'Segmentation fault' during loading.
> > > 
> > > root@localhost:~# cd linux-kenel/tools/testing/selftests/arm64/abi && make
> > > root@localhost:~/linux-kernel/tools/testing/selftests/arm64/abi# ./tpidr2
> > > Segmentation fault
> > > 
> > > The cause of this is the __arch_clear_user() failure.
> > > 
> > > load_elf_binary() [fs/binfmt_elf.c]
> > >    -> if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bes)))
> > >      -> padzero()
> > >        -> clear_user() [arch/arm64/include/asm/uaccess.h]
> > >          -> __arch_clear_user() [arch/arm64/lib/clear_user.S]
> > > 
> > > For more details, please see:
> > > https://lore.kernel.org/lkml/1d0342f3-0474-482b-b6db-81ca7820a462@t-8ch.de/T/
> > This is just a userspace issue (i.e. don't do that, and if you do want
> > to do that, use a new kernel!)
> > 
> > Why do these changes need to be backported, do you have real users that
> > are crashing in this way to require these changes?
> 
> 
> This issue was identified during our internal testing, and I found
> similar cases discussed in the link above. Upon reviewing the kernel
> code, I noticed that a patch series already accepted into mainline
> addresses this problem. Since these patches are already upstream
> and effectively resolve the issue, I decided to backport them.
> We believe this provides a more robust and maintainable solution
> compared to relying on users to avoid the triggering behavior.

Fixing something just to get the selftests to pass is fine, but do you
actually know of a real-world case where this is a problem that needs to
be resolved?  That's what I'm asking here, do you have users that have
run into this issue?  I ask as it's not a regression from what I can
determine, but rather a new "feature".

thanks,

greg k-h