Re: [REGRESSION] firmware: qcom: scm: smc: switch to using the SCM allocator

[Bartosz Golaszewski]

On Sat, 7 Sept 2024 at 11:57, Dmitry Baryshkov
<dmitry.baryshkov@xxxxxxxxxx> wrote:
>
> On Sat, Sep 07, 2024 at 01:49:02AM GMT, Rudraksha Gupta wrote:
> > > Looks like qcom_scm_get_tzmem_pool() returns NULL. Not sure how this
> > > happens. Can you confirm that the QCom SCM driver probed correctly?
> >
> > Thanks for looking into this! Please let me know how I can help!
> >
> >
> > I've been building the driver into the kernel I believe. Here's the relevant
> > line in the config:
> >
> > $ rg -i "scm"
> > samsung/expressatt/linux.defconfig
> > 1615:# CONFIG_ARM_SCMI_PROTOCOL is not set
> > 1627:CONFIG_QCOM_SCM=y
> > 1628:# CONFIG_QCOM_SCM_DOWNLOAD_MODE_DEFAULT is not set
> >
> >
> > It seems like the scm driver is being probed:
> >
> > ~ # dmesg | grep scm
> > [    0.066438] qcom_scm: convention: smc legacy
> > [    0.362543] bus: 'platform': add driver qcom_scm
>
> The scm driver is added, but it is not probed as there is no SCM node in
> DT. I'll send and RFT patch to add it. However the issue still persists:
> the driver should not crash if there is no SCM device on a system.
> qcom_scm_*_alloc should fall back if there is no SCM.
>

Ugh, so some SCM calls seem to expect that they can get called without
the SCM driver. It's not very intuitive, I would expect that the
driver must be up for SCM to work at all.

I think we should fall back to using kzalloc() in such cases as
there's no struct device to use with the DMA alloc APIs. I'll prepare
a patch.

Bart