Re: [PING][PATCH,v5][AArch64] MTE corefile support

[Catalin Marinas via Gdb-patches <gdb-patches@sourceware.org>]

On Wed, Jul 28, 2021 at 10:26:58PM -0400, Simon Marchi wrote:
> On 2021-07-14 9:07 a.m., Catalin Marinas wrote:
> > On Sun, Jul 11, 2021 at 07:22:00AM -0700, Joel Brobecker wrote:
> >> Luis is trying to have this new feature ready for the GDB 11.1
> >> release. As I understand it, the AArch64-specific part has already
> >> been approved, so he's waiting for someone to review the rest of
> >> the changes. Is there a Global Maintainer who would feel comfortable
> >> doing so and who would have the time?
> > 
> > I'm a Linux kernel maintainer, so not much for me to review on the gdb
> > front. But feedback from the gdb maintainers would be appreciated.
> > 
> > W.r.t. upstreaming the feature in gdb, I'd wait until the Linux
> > counterpart is merged. We agreed in principle on the ABI but I wouldn't
> > consider it stable until the feature hits a kernel release.
> 
> I reviewed a few of the previous iterations, and there were only really
> some nits remaining, so I am confident that this version is OK, from a
> GDB point of view.
> 
> ABI-wise, I tend to agree with Catalin: as long as it's not in a
> released kernel, the format *could* change.  But is it a problem in
> practice?  Let's say this happens:
> 
>  - We merge support for MTE core files in GDB 11
>  - The format gets changed, merged and released in a kernel
>  - We fix GDB 12 based on the new format
> 
> Then GDB 11 users end up with a GDB that can't read MTE notes. But had
> we not merged the support for it, they would also end up with a GDB that
> can't read MTE notes.  In both cases, we'll tell them "upgrade to GDB
> 12" (once it's released).

I'm more worried about GDB 11 reading the MTE notes incorrectly rather
than not reading them at all.

> If we merge support for MTE in core files now and if the format
> doesn't change, everybody is happy.  I guess it depends on how likely it
> is that the format is going to change.

That's hard to say before we go through at least one Linux kernel
mailing review.

-- 
Catalin