Re: [PATCH] arm64/fpsimd: Suppress SVE access traps when loading FPSIMD state

From: Mark Brown
Date: Mon Aug 21 2023 - 08:46:28 EST


On Mon, Aug 21, 2023 at 01:10:50PM +0100, Will Deacon wrote:
> On Mon, Aug 07, 2023 at 11:20:38PM +0100, Mark Brown wrote:

> > I did some instrumentation which counted the number of SVE access traps
> > and the number of times we loaded FPSIMD only register state for each task.
> > Testing with Debian Bookworm this showed that during boot the overwhelming
> > majority of tasks triggered another SVE access trap more than 50% of the
> > time after loading FPSIMD only state with a substantial number near 100%,
> > though some programs had a very small number of SVE accesses most likely
> > from startup. There were few tasks in the range 5-45%, most tasks either

> Do you have any performance numbers to motivate this change? It would be
> interesting, for example, to see how changing the timeout value affects
> the results for some real workloads.

I agree, unfortunately I don't yet have a hardware setup so I can't do
those benchmarks myself at the minute - I just have the access trap
counting I mentioned in the commit message. Last time I saw numbers the
access traps were about 70% of a syscall in a microbenchmark and we're
already suppressing them for non-blocking syscalls so I'd not expect
anything *too* revolutionary.

Attachment: signature.asc
Description: PGP signature