Re: [PATCH] iommu/arm-smmu: don't unregister on shutdown

[Vladimir Oltean]

On Wed, Dec 14, 2022 at 09:25:45PM +0000, Robin Murphy wrote:
> > > The change itself looks sensible. The point of this shutdown hook is simply
> > > not to leave active translations in place that might confuse future software
> > > after reboot/kexec; any housekeeping in the current kernel state is a waste
> > > of time anyway. Fancy doing the same for SMMUv3 as well?
> > 
> > I can try, but I won't have hardware to test.
> > 
> > Basically the only thing truly relevant for shutdown from arm_smmu_device_remove()
> > is arm_smmu_device_disable(), would you agree to a patch which changes
> > things as below?
> > 
> > diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> > index 6d5df91c5c46..d4d8bfee9feb 100644
> > --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> > +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> > @@ -3854,7 +3854,9 @@ static int arm_smmu_device_remove(struct platform_device *pdev)
> >   static void arm_smmu_device_shutdown(struct platform_device *pdev)
> >   {
> > -	arm_smmu_device_remove(pdev);
> > +	struct arm_smmu_device *smmu = platform_get_drvdata(pdev);
> > +
> > +	arm_smmu_device_disable(smmu);
> >   }
> >   static const struct of_device_id arm_smmu_of_match[] = {
> 
> 
> Looks fine to me! I'll let Will decide if he'd still prefer to do the full
> remove-calls-shutdown reversal here as well for complete consistency, but I
> reckon the minimal diff is no bad thing :)

The reason why I did it this way is that if remove() still called
shutdown(), it would have looked like this here:

static void arm_smmu_device_shutdown(struct platform_device *pdev)
{
	struct arm_smmu_device *smmu = platform_get_drvdata(pdev);

	arm_smmu_device_disable(smmu);
}

static int arm_smmu_device_remove(struct platform_device *pdev)
{
	struct arm_smmu_device *smmu = platform_get_drvdata(pdev);

	iommu_device_unregister(&smmu->iommu);
	iommu_device_sysfs_remove(&smmu->iommu);
	arm_smmu_device_shutdown(pdev);
	iopf_queue_free(smmu->evtq.iopf);

	return 0;
}

Not really that beneficial. I also didn't want to reorder any
operations, they seem to be done in reverse order of what is being done
in arm_smmu_device_probe().