On Mon, Dec 4, 2023, at 14:37, John Paul Adrian Glaubitz wrote:
> On Mon, 2023-12-04 at 14:06 +0100, Arnd Bergmann wrote:
>
>> I would also suggest raising the minimum SPARC32 level to that of
>> leon3 (SPARCv8e with CAS), which is what glibc requires
>> anyway for futex().
>
> I thought there was user-space emulation for CAS in these cases but I might
> be wrong.
The problem with emulating CAS is that this needs to be serialized
with sys_futex(). Without Andreas's patches, sparc32 kernels don't
even have provide futex(), so it doesn't work at all. On sparc64
kernels or leon3+ with the futex patch, sparc32 userspace can work,
as long as it uses native CAS, but this would in turn break if
userspace ends up using emulated CAS.
Arnd
