[PATCH v2] aarch64: Add MTE mode tunable

[Adhemerval Zanella Netto]

On 11/09/26 04:04, Yury Khrustalev wrote:
> On Thu, Sep 10, 2026 at 11:30:01AM -0300, Adhemerval Zanella Netto wrote:
>>
>> On 07/09/26 11:02, Yury Khrustalev wrote:
>>> Add tunable glibc.cpu.mtemode for selecting tag checking fault mode
>>> for Memory Tagging Extension (MTE): disabled, enabled, sync, or async.
>>> Default is disabled.
>>>
>>> ---
>>> Passes regression in aarch64. OK for trunk?
>>>
>>> Please note that this patch does not include support for MTE binary
>>> marking as there is no intention to implement it at this stage.
>>> However, this patch is designed to support such an extension in the
>>> future.
>>>
>>> Relevant discussion:
>>> https://inbox.sourceware.org/libc-alpha/ansmVLinBE86-SZ4@arm.com/
>>>
>>> Base commit: fe03757f67
>>>
>>> Changes in v2:
>>>  - Don't add the 'mte' property to the cpu_features struct.
>>>  - Define bool variable for MTE for ifunc resolvers based on
>>>    the MTE mode value.
>>>  - v1: https://inbox.sourceware.org/libc-alpha/20260904094800.1810189-1-yury.khrustalev@arm.com/
>>
>> It is not clear to me why add this support without any MTE usage within
>> glibc. All these MTE enablement can done by the application itself, if
>> it requires so (like a MTE aware malloc).
>>
>> I think this feature should added along the MTE malloc you seems to be
>> working on.
> 
> One step at a time. Just avoiding long patch series that are difficult
> to review and maintain. This change is essential ground work and is
> required to unblock work on a couple of other things. If we keep it
> attached to them, it creates unnecessary overhead of rebasing and testing
> overlapping changes.
In understand that, but it is really hard to review and analyze the MTE
proposal without the full picture of what kind of MTE support you aim
to implement.

For instance, there are references for STACK and GLOBAL in this patch
even though you state that it is out of scope.