Re: [RESEND PATCH v17 2/5] iommu/arm-smmu: Invoke pm_runtime during probe, add/remove device

[Will Deacon]

On Mon, Nov 26, 2018 at 04:56:42PM +0530, Vivek Gautam wrote:
> On 11/26/2018 11:33 AM, Vivek Gautam wrote:
> >On 11/24/2018 12:06 AM, Will Deacon wrote:
> >>On Thu, Nov 22, 2018 at 05:32:24PM +0530, Vivek Gautam wrote:
> >>>On Wed, Nov 21, 2018 at 11:09 PM Will Deacon <will.deacon@xxxxxxx>
> >>>wrote:
> >>>>On Fri, Nov 16, 2018 at 04:54:27PM +0530, Vivek Gautam wrote:
> >>>>>From: Sricharan R <sricharan@xxxxxxxxxxxxxx>
> >>>>>
> >>>>>The smmu device probe/remove and add/remove master device callbacks
> >>>>>gets called when the smmu is not linked to its master, that is
> >>>>>without
> >>>>>the context of the master device. So calling runtime apis in those
> >>>>>places
> >>>>>separately.
> >>>>>Global locks are also initialized before enabling runtime pm as the
> >>>>>runtime_resume() calls device_reset() which does tlb_sync_global()
> >>>>>that ultimately requires locks to be initialized.
> >>>>>
> >>>>>Signed-off-by: Sricharan R <sricharan@xxxxxxxxxxxxxx>
> >>>>>[vivek: Cleanup pm runtime calls]
> >>>>>Signed-off-by: Vivek Gautam <vivek.gautam@xxxxxxxxxxxxxx>
> >>>>>Reviewed-by: Tomasz Figa <tfiga@xxxxxxxxxxxx>
> >>>>>Tested-by: Srinivas Kandagatla <srinivas.kandagatla@xxxxxxxxxx>
> >>>>>Reviewed-by: Robin Murphy <robin.murphy@xxxxxxx>
> >>>>>---
> >>>>>  drivers/iommu/arm-smmu.c | 101
> >>>>>++++++++++++++++++++++++++++++++++++++++++-----
> >>>>>  1 file changed, 91 insertions(+), 10 deletions(-)
> >>>>Given that you're doing the get/put in the TLBI ops unconditionally:
> >>>>
> >>>>>  static void arm_smmu_flush_iotlb_all(struct iommu_domain *domain)
> >>>>>  {
> >>>>>       struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
> >>>>>+     struct arm_smmu_device *smmu = smmu_domain->smmu;
> >>>>>
> >>>>>-     if (smmu_domain->tlb_ops)
> >>>>>+     if (smmu_domain->tlb_ops) {
> >>>>>+             arm_smmu_rpm_get(smmu);
> >>>>>smmu_domain->tlb_ops->tlb_flush_all(smmu_domain);
> >>>>>+             arm_smmu_rpm_put(smmu);
> >>>>>+     }
> >>>>>  }
> >>>>>
> >>>>>  static void arm_smmu_iotlb_sync(struct iommu_domain *domain)
> >>>>>  {
> >>>>>       struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
> >>>>>+     struct arm_smmu_device *smmu = smmu_domain->smmu;
> >>>>>
> >>>>>-     if (smmu_domain->tlb_ops)
> >>>>>+     if (smmu_domain->tlb_ops) {
> >>>>>+             arm_smmu_rpm_get(smmu);
> >>>>>smmu_domain->tlb_ops->tlb_sync(smmu_domain);
> >>>>>+             arm_smmu_rpm_put(smmu);
> >>>>>+     }
> >>>>Why do you need them around the map/unmap calls as well?
> >>>We still have .tlb_add_flush path?
> >>Ok, so we could add the ops around that as well. Right now, we've got
> >>the runtime pm hooks crossing two parts of the API.
> >
> >Sure, will do that then, and remove the runtime pm hooks from map/unmap.
> 
> I missed this earlier -
> We are adding runtime pm hooks in the 'iommu_ops' callbacks and not really
> to
> 'tlb_ops'. So how the runtime pm hooks crossing the paths?
> '.map/.unmap' iommu_ops don't call '.flush_iotlb_all' or '.iotlb_sync'
> iommu_ops
> anywhere.
> 
> E.g., only callers to domain->ops->flush_iotlb_all() are:
> iommu_dma_flush_iotlb_all(), or iommu_flush_tlb_all() which are not in
> map/unmap paths.

Yes, sorry, I got confused here and completely misled you. In which case,
your original patch is ok because it intercepts the core IOMMU API via
iommu_ops. Apologies.

At that level, should we also annotate arm_smmu_iova_to_phys_hard()
for the iova_to_phys() implementation?

With that detail and clock bits sorted out, we should be able to get this
queued at last.

Will