Re: [PATCH 2/2] remoteproc: qcom: Check glink->edge in glink_subdev_stop()

[Stephan Gerhold <stephan.gerhold@linaro.org>]

On Tue, Apr 14, 2026 at 11:23:50AM +0800, Jingyi Wang wrote:
> On 4/10/2026 10:15 PM, Stephan Gerhold wrote:
> > On Thu, Apr 09, 2026 at 01:46:22AM -0700, Jingyi Wang wrote:
> > > For rproc that doing attach, glink_subdev_start() is called only when
> > > attach successfully. If rproc_report_crash() is called in the attach
> > > function, rproc_boot_recovery()->rproc_stop()->glink_subdev_stop() could
> > > be called and cause NULL pointer dereference:
> > > 
> > >   Unable to handle kernel NULL pointer dereference at virtual address 0000000000000300
> > >   Mem abort info:
> > >   ...
> > >   pc : qcom_glink_smem_unregister+0x14/0x48 [qcom_glink_smem]
> > >   lr : glink_subdev_stop+0x1c/0x30 [qcom_common]
> > >   ...
> > >   Call trace:
> > >    qcom_glink_smem_unregister+0x14/0x48 [qcom_glink_smem] (P)
> > >    glink_subdev_stop+0x1c/0x30 [qcom_common]
> > >    rproc_stop+0x58/0x17c
> > >    rproc_trigger_recovery+0xb0/0x150
> > >    rproc_crash_handler_work+0xa4/0xc4
> > >    process_scheduled_works+0x18c/0x2d8
> > >    worker_thread+0x144/0x280
> > >    kthread+0x124/0x138
> > >    ret_from_fork+0x10/0x20
> > >   Code: a9be7bfd 910003fd a90153f3 aa0003f3 (b9430000)
> > >   ---[ end trace 0000000000000000 ]---
> > > 
> > > Add NULL pointer check in the glink_subdev_stop() to make sure
> > > qcom_glink_smem_unregister() will not be called if glink_subdev_start()
> > > is not called.
> > > 
> > 
> > You mention the actual root problem here: Why is glink_subdev_stop()
> > called if glink_subdev_start() wasn't called?
> > 
> > The call to rproc_start_subdevices() in __rproc_attach() makes sure that
> > all subdevices are in consistent state when exiting the function (either
> > prepared+started or stopped+unprepared). Only if all subdevices were
> > started successfully, the rproc->state is changed to RPROC_ATTACHED.
> > 
> > In your case, attaching the rproc failed so the rproc->state should be
> > still RPROC_DETACHED. All subdevices should be stopped+unprepared. We
> > shouldn't stop/unprepare any subdevices again in this state, they all
> > might crash like glink does here.
> > 
> > We know that subdevices are already stopped+unprepared in RPROC_DETACHED
> > state, so I think you just need to skip rproc_stop_subdevices() and
> > rproc_unprepare_subdevices() inside rproc_stop() in this case, see diff
> > below.
> > 
> > @@ -1708,8 +1709,9 @@ static int rproc_stop(struct rproc *rproc, bool crashed)
> >   	if (!rproc->ops->stop)
> >   		return -EINVAL;
> > -	/* Stop any subdevices for the remote processor */
> > -	rproc_stop_subdevices(rproc, crashed);
> > +	/* Stop any subdevices for the remote processor if it was attached */
> > +	if (rproc->state != RPROC_DETACHED)
> > +		rproc_stop_subdevices(rproc, crashed);
> >   	/* the installed resource table is no longer accessible */
> >   	ret = rproc_reset_rsc_table_on_stop(rproc);
> > @@ -1726,7 +1728,8 @@ static int rproc_stop(struct rproc *rproc, bool crashed)
> >   		return ret;
> >   	}
> > -	rproc_unprepare_subdevices(rproc);
> > +	if (rproc->state != RPROC_DETACHED)
> > +		rproc_unprepare_subdevices(rproc);
> >   	rproc->state = RPROC_OFFLINE;
> 
> In this case, rproc_crash_handler_work()->rproc_trigger_recovery()->
> rproc_boot_recovery()->rproc_stop()->glink_subdev_stop() is called,
> "rproc->state = RPROC_CRASHED" is set in the rproc_crash_handler_work
> before rproc_boot_recovery is called, so checking RPROC_DETACHED can
> not work for this case.
> 

You're right, I forgot about that. I think we need a more generic
solution for this though. rproc_stop_subdevices() should not be called
without a prior call to rproc_start_subdevices().

I think there are a couple of options for this:

 - Add a bool "subdevs_started" to struct rproc and manage that
   separately from the rproc->state.

 - Track the rproc state before the crash separately (something like
   rproc->state_before_crash) and check that in the stop path.

 - Add a new state RPROC_CRASHED_DETACHED to make sure the states are
   unique.

 - ...

Does the same issue also exist in qcom_pas_stop() of "[PATCH v5 4/5]
remoteproc: qcom: pas: Add late attach support for subsystems" [1]?
There you check for pas->rproc->state != RPROC_ATTACHED, wouldn't this
also fail for the RPROC_CRASHED case?

Thanks,
Stephan

[1]: https://lore.kernel.org/linux-arm-msm/20260409-knp-soccp-v5-4-805a492124da@oss.qualcomm.com/