On Tue, 2 Mar 2021 10:01:09 +0100, Ard Biesheuvel wrote:
> [ TL;DR for the non-ARM folks on CC: disabling softirq processing when using
> SIMD in kernel mode could reduce complexity and improve performance, but we
> need to decide whether we can do this, and how much softirq processing
> latency we can tolerate. If we can find a satisfactory solution for this,
> we might do the same for x86 and 32-bit ARM as well.
>
> However, based on preliminary off-list discussions with peterz and luto, it
> seems that for x86, there is a preference for using per-CPU buffers to
> preserve/restore the task context's kernel mode SIMD state when the task is
> interrupted to perform kernel mode SIMD in softirq context. On arm64, we
> actually had this arrangement before, and removed it because it made
> reasoning about preserving/restoring userland SVE state (32 SIMD registers
> of up to 2 kbit in size) rather complex. ]
>
> [...]
Applied to arm64 (for-next/neon-softirqs-disabled), thanks!
[1/9] arm64: assembler: remove conditional NEON yield macros
https://git.kernel.org/arm64/c/27248fe1abb2
[2/9] arm64: assembler: introduce wxN aliases for wN registers
https://git.kernel.org/arm64/c/4c4dcd3541f8
[3/9] arm64: fpsimd: run kernel mode NEON with softirqs disabled
https://git.kernel.org/arm64/c/13150149aa6d
--
Catalin
