Re: [PATCH] Revert "firmware: qcom: qseecom: convert to using the TZ allocator"

[Johan Hovold <johan@kernel.org>]

On Mon, Jul 29, 2024 at 12:03:55PM +0200, Bartosz Golaszewski wrote:
> On Mon, Jul 29, 2024 at 11:58 AM Johan Hovold <johan+linaro@kernel.org> wrote:
> >
> > This reverts commit 6612103ec35af6058bb85ab24dae28e119b3c055.
> >
> > Using the "TZ allocator" for qcseecom breaks efivars on machines like
> > the Lenovo ThinkPad X13s and x1e80100 CRD:
> >
> >         qcom_scm firmware:scm: qseecom: scm call failed with error -22
> >
> > Reverting to the 6.10 state makes qseecom work again.
> >
> > Fixes: 6612103ec35a ("firmware: qcom: qseecom: convert to using the TZ allocator")
> > Cc: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
> > Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
> > ---
> > Cc: regressions@lists.linux.dev
> >
> > #regzbot introduced: 6612103ec35a
> 
> How about at least giving me the chance to react to the report and fix
> it instead of reverting it right away?

Lots of folks have been running linux-next on Qualcomm machines for a
month without reporting or fixing the issue. And v10 of the offending
patch was apparently never even tested before being merged.

I'm sure you'll have a few days to look at this before we revert.

I'll be on holiday for a few weeks, but you have an X13s so you should
be able to reproduce this yourself.

> Are there any other messages about SHM bridge/SCM calls in the kernel log?

I've also seen this combo:

	[    3.219296] qcom_scm firmware:scm: qseecom: scm call failed with error -22
        [    3.227153] efivars: get_next_variable: status=8000000000000007

But usually the first message is the only hint why efivars is completely
broken.

> Do you have QCOM_TZMEM_MODE_GENERIC=y or QCOM_TZMEM_MODE_SHM_BRIDGE=y
> in your config? If the latter: can you try changing it to the former
> and retest?

I have the former in my config but have tested both, made no difference.

> > It's a little frustrating to find that no-one tested this properly or
> > even noticed the regression for the past month that this has been
> > sitting in linux-next.
> 
> I have tested many platforms and others have done the same but
> unfortunately cannot possibly test every single use-case on every
> platform. This is what next is for after all.

I doubt this is specific to sc8280xp and x1e80100. Which platforms did
you test qseecom and efivars on?

> > Looks like Maximilian may have hit this with v9 too:
> >
> >         https://lore.kernel.org/lkml/CAMRc=Mf_pvrh2VMfTVE-ZTypyO010p=to-cd8Q745DzSDXLGFw@mail.gmail.com/
> >
> > even if there were further issues with that revision.
> 
> This is a different issue that was fixed in a later iteration.

The symptoms appear to be the same once you get past the locking splats:

	[    2.507347] qcom_scm firmware:scm: qseecom: scm call failed with error -22
	[    2.507813] efivars: get_next_variable: status=8000000000000007

So it's possible that this never worked.

Johan