On Tue, Aug 25, 2026 at 5:50 PM Richard Henderson
<[email protected]> wrote:
>
> On 8/22/26 12:08, Matt Turner wrote:
> > The cost is memory: the cache grows from 64 KiB to 1 MiB per vCPU. That
> > is easy to justify for a single-vCPU linux-user process and less obvious
> > for system emulation with many vCPUs, so this may want to be sized by
> > target or made tunable rather than raised unconditionally.
>
> Why do you believe linux-user processes are single-threaded?  I suppose
> when emulating /bin a fair few of them are, but...

I don't -- I'd conflated "one process" with "one CPUState". linux-user
creates a CPUState per guest thread, so a threaded guest pays the
megabyte per thread, just as system emulation pays it per vCPU. Commit
message fixed.

The sizing question stays open, which is why the patch says so. I
don't have a threaded workload where 64 KiB is the better trade and
would welcome one.

Reply via email to