[PATCH v2 37/39] efi: Avoid using sandbox virtio devices

Simon Glass sjg at chromium.org
Thu Aug 22 19:11:11 CEST 2024


Hi Tom,

On Thu, 22 Aug 2024 at 09:13, Tom Rini <trini at konsulko.com> wrote:
>
> On Fri, Aug 16, 2024 at 05:53:59PM -0600, Simon Glass wrote:
> [snip]
> > > Some notes from a little bit of digging: There are almost 1000 sandbox
> > > tests (17-20mins), but qemu_arm only runs 62 (2mins). With 'make
> > > qcheck' it runs about 2500 tests in about 4 minutes. I just got 'make
> > > pcheck' going again and that is a little faster (2.5 mins). Binman
> > > tests run in parallel if you use 'pip install concurrencytest'
> >
> > While on this topic I forgot to mention that the sandbox tests (~800
> > of them) actually run in about 10 seconds. Try 'ut all' (although
> > you'll need my test-fix series to stop it crashing).
> >
> > The reason pytest is so slow with sandbox is that it talks to U-Boot
> > through a pty. Stephen Warren did something clever to stop it being
> > horrible, but it is still very slow.
>
> Seeing about addressing this more would probably be very helpful then as
> most of the runs (by user, I mean) happen via pytest and not directly
> running sandbox I strongly suspect.

Yes, there is a long thread about performance[1] which makes for
interesting reading these days. We do indeed have >1000 tests although
it took more than 2 years. The Python approach is indeed quite slow,
although half as slow as using the real 'expect' library.

I do have one idea that might help, so I filed [2].

Regards,
Simon

[1] https://patchwork.ozlabs.org/project/uboot/patch/1447570381-1361-1-git-send-email-swarren@wwwdotorg.org/
[2] https://source.denx.de/u-boot/u-boot/-/issues/39


More information about the U-Boot mailing list