Re: [PATCH] firmware: arm_scmi: Fix duplicate workqueue name

From: Sudeep Holla
Date: Wed Oct 14 2020 - 13:35:27 EST


On Wed, Oct 14, 2020 at 06:32:42PM +0100, Sudeep Holla wrote:
> On Wed, Oct 14, 2020 at 10:08:32AM -0700, Florian Fainelli wrote:
> > On 10/14/20 2:18 AM, Sudeep Holla wrote:
> > > Hi Florian,
> > >
> > > Thanks for the patch, it shows someone else is also using this and
> > > testing 😉.
> > >
> > > On Tue, Oct 13, 2020 at 07:17:37PM -0700, Florian Fainelli wrote:
> > >> When more than a single SCMI device are present in the system, the
> > >> creation of the notification workqueue with the WQ_SYSFS flag will lead
> > >> to the following sysfs duplicate node warning:
> > >>
> > >
> > > Please trim the calltrace next time without timestamp and register raw
> > > hex values.
> >
> > Will do, thanks!
> >
>
> Thanks!
>
> > >
> > > You using this on 32-bit platform ? If so, thanks for additional test
> > > coverage.
> >
> > We have a mix of ARMv7/LPAE (Brahma-B15) and ARMv8 (Brahma-B53,
> > Cortex-A72) devices that we regularly test with 32-bit and 64-bit kernels.
> >
>
> Ah OK, good to know.
>
> [...]
>
> > >> Fix this by using dev_name(handle->dev) which guarantees that the name is
> > >> unique and this also helps correlate which notification workqueue corresponds
> > >> to which SCMI device instance.
> > >>
> > >
> > > I am curious as how multiple SCMI instances are used. We know few limitations
> > > in the code to handle that yet, so interested to know if you are carrying
> > > more patches/fixes.
> >
> > We currently have two SCMI device nodes in Device Tree:
> >
> > - the first one is responsible for all of the base, performance, sensors
> > protocols and is present on all of the chips listed above
> >
> > - the second one is responsible for a proprietary protocol through which
> > we encapsulate a variety of operations towards a secure agent in the
> > system, this is only present in a subset of devices.
> >
>
> And any particular reasons it can't exist in the same node. And also are
> they talking to different SCMI firmware implementation meaning different
> location in the system. The reason I ask is we have notion of one platform
> with agent id = 0 as per the specification. It can be split in terms
> of implementation and can have some side band communication amongst
> themselves but can't have agent ID other than 0. It violates specification.
>
> I don't have issues split it into 2 or more SCMI device as long as it
> doesn't provide notion of existence of multiple SCMI platform firmware
> implementations with different agent ID.
>
> Also Cristian has posted patches to support custom protocols[1]. It would
> be good if you can take a look/review/test/comment...
>

Pressed enter too early, link added now.

--
Regards,
Sudeep

[1] https://lore.kernel.org/lkml/20201014150545.44807-1-cristian.marussi@xxxxxxx