[PATCH 0/8] Migrate PAS clients to qcom_mdt_pas_load() and qcom_pas_prepare_and_auth_reset()

[Dmitry Baryshkov]

On Wed, Oct 21, 2026 at 06:37:35PM +0530, Mukesh Ojha wrote:
> On Wed, Oct 21, 2026 at 01:32:13PM +0200, Konrad Dybcio wrote:
> > On 9/25/26 5:57 PM, Mukesh Ojha wrote:
> > > In preparation to remove qcom_mdt_load() and qcom_pas_auth_and_reset()
> > > public function, migrate every user to the new PAS context-aware API
> > > qcom_mdt_pas_load() and qcom_pas_prepare_and_auth_reset().
> > > 
> > > This series migrates every remaining caller of the deprecated APIs
> > > (qcom_wcnss, adreno zap shader, iris, venus, IPA, ath12k) and then
> > > removes qcom_mdt_load() and qcom_pas_auth_and_reset() entirely.
> > > 
> > > This series is based on top of although no dependency as such.
> > >   firmware: qcom: pas: Make MDT buffer retention explicit
> > 
> > LLM has the following comments:
> > 
> >  1. Metadata buffer (ctx->ptr) is leaked, and overwritten on
> > reload, because qcom_pas_metadata_release() is never called
> > (patches 1-6)
> >  2. devm_qcom_pas_context_alloc() is called from repeatable runtime
> > paths, so contexts accumulate until unbind (patches 2, 3, 4, 6)
> >  3. use_tzmem is never set, so the shmbridge prepare step never
> > runs on SCM and the commit messages are inaccurate; TEE gets
> > a real mem_phys/mem_size change
> >  4. Not bisectable: patch 4 relies on the NULL mem_region
> > handling that only arrives in patch 7
> >  5. Patches 3 and 4 don't apply to this tree; iris
> > iris_detect_firmware() still needs mem_virt after the load;
> > the base of the series is not stated
> >  6. Patch 7: NULL mem_region now means "map internally", which
> > weakens the -EINVAL sanity check
> >  7. Patch 6: redundant second ioremap_wc() on top of ath12k's
> > persistent memremap()
> >  8. Patch 6: ctx is set under one scm_auth_enabled check and
> > used under another, which is fragile
> >  9. Patch 2: the context now carries the full region size instead of
> > the MDT size, which matters if use_tzmem gets enabled
> > 10. Patch 3: stale qcom_mdt_load() comments remain in
> > iris_firmware.c
> > 11. Patch 8: no issues found
> 
> Thanks.  Will address them along with Sasiko's one.

Just to check. Did you verify your commits with the lockdep and atomic
sleep debugging on?

-- 
With best wishes
Dmitry