Re: [PATCH 09/11] tests/tcg: disable pauth for aarch64 gdb tests

On 3/17/23 16:37, Peter Maydell wrote:
> On Wed, 15 Mar 2023 at 09:51, Luis Machado <[email protected]> wrote:
>> On 3/13/23 11:44, Luis Machado wrote:
>>> On 3/13/23 11:22, Peter Maydell via Gdb wrote:
>>>> Luis and I came up with two options:
>>>>
>>>> (1) leave QEMU outputting the pauth xml as-is, and tell people
>>>> whose gdb 12 crashes that they should upgrade to a newer gdb
>>>>
>>>> (2) make QEMU output the pauth info under a different XML namespace,
>>>> and tell people who need backtraces when pauth is enabled
>>>> that they should upgrade to a newer gdb
>>>>
>>>> Neither of these feel great, but on balance I guess 2 is better?
>>>>
>>>> Luis: I think that rather than doing (2) with a QEMU namespace,
>>>> we should define a gdb namespace for this. That makes it clear
>>>> that this is still a gdb-upstream-sanctioned way of exposing
>>>> the pauth registers.
>>>
>>> That should be fine as well, and would work to side-step the gdb 12 bug so it doesn't crash.
>>>
>>> We could name the feature "org.gnu.gdb.aarch64.pauth_v2" or somesuch, and slowly stop using the original
>>> "org.gnu.gdb.aarch64.pauth" feature. I can document the requirements for a compliant pauth_v2.
>>>
>>> FYI, I've pushed a better documentation for the arm/aarch64 xml descriptions here:
>>>
>>> https://sourceware.org/git?p=binutils-gdb.git;a=commit;h=d7001b29e9f256dfc60acb481d9df8f91f2ee623
>>> https://sourceware.org/git?p=binutils-gdb.git;a=commit;h=e0994165d1b8469dfc27b09b62ac74862d535812
>
>> Just an update on this. I had a chat with Richard Henderson yesterday, and it might actually be easier and more convenient to backport
>> fixes to older gdb versions (at least gdb-12 and gdb-11, but gdb-10 and gdb-9 are also affected). This will ensure those won't crash when
>> they connect to a qemu that advertises the pauth feature.
>>
>> It also means we won't need qemu-side changes. My understanding is that we're close to the 8.0.0 release, and the code is already in place.
>
> Having run into this problem in another couple of situations, one of
> which involved gdb 10, I think I'm increasingly favouring option
> 2 here. The affected gdbs seem to be quite widely deployed, and
> the bug results in crashes even for users who didn't really
> care about pauth. So I'd rather we didn't release a QEMU 8.0
> which crashes these affected deployed gdbs.
>

Are the affected gdb's packaged by distros? If so, a backport the distros can pick up
will solve this in a quick package update.

If we decide qemu should now emit a different xml for pauth, it will fix the crashes, but it also
means older gdb's (9/10/11/12) will not be able to backtrace properly through pauth-signed frames.

Maybe that's a reasonable drawback for qemu users?

If someone decides to implement a debugging stub that reports pauth (fast models, for example), it will
also crash gdb, so I still plan to do the backport anyway.

> So:
>   (a) if on the gdb side you can define (within the next week) a
> suitable new XML name you want QEMU to expose, we can commit a
> change to switch to that before we do the 8.0 release

pauth_v2 sounds about as good as any other for me.

>   (b) if that's too tight a timescale, we can commit a patch which
> just stops QEMU from exposing the pauth.xml, and we can come up
> with a better solution after 8.0 releases
>
> In fact, I think I'm going to submit a patch to do (b) for
> now and we can follow up with a patch for (a) if we want.
>
> thanks
> -- PMM

IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.