Re: [PATCH 06/20] drm/msm: Use iommu_paging_domain_alloc()

[Baolu Lu <baolu.lu@linux.intel.com>]

On 5/30/24 3:58 PM, Dmitry Baryshkov wrote:
> On Thu, 30 May 2024 at 04:59, Baolu Lu<baolu.lu@linux.intel.com>  wrote:
>> On 5/29/24 4:21 PM, Dmitry Baryshkov wrote:
>>> On Wed, May 29, 2024 at 01:32:36PM +0800, Lu Baolu wrote:
>>>> The domain allocated in msm_iommu_new() is for the @dev. Replace
>>>> iommu_domain_alloc() with iommu_paging_domain_alloc() to make it explicit.
>>>>
>>>> Update msm_iommu_new() to always return ERR_PTR in failure cases instead
>>>> of NULL.
>>> Please don't mix unrelated changes, because ...
>>>
>>>> Signed-off-by: Lu Baolu<baolu.lu@linux.intel.com>
>>>> ---
>>>>    drivers/gpu/drm/msm/msm_iommu.c | 8 ++++----
>>>>    1 file changed, 4 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/msm/msm_iommu.c b/drivers/gpu/drm/msm/msm_iommu.c
>>>> index d5512037c38b..f7e28d4b5f62 100644
>>>> --- a/drivers/gpu/drm/msm/msm_iommu.c
>>>> +++ b/drivers/gpu/drm/msm/msm_iommu.c
>>>> @@ -407,9 +407,9 @@ struct msm_mmu *msm_iommu_new(struct device *dev, unsigned long quirks)
>>>>       struct msm_iommu *iommu;
>>>>       int ret;
>>>>
>>>> -    domain = iommu_domain_alloc(dev->bus);
>>>> -    if (!domain)
>>>> -            return NULL;
>>>> +    domain = iommu_paging_domain_alloc(dev);
>>>> +    if (IS_ERR(domain))
>>>> +            return ERR_CAST(domain);
>>>>
>>>>       iommu_set_pgtable_quirks(domain, quirks);
>>>>
>>>> @@ -441,7 +441,7 @@ struct msm_mmu *msm_iommu_gpu_new(struct device *dev, struct msm_gpu *gpu, unsig
>>>>       struct msm_mmu *mmu;
>>>>
>>>>       mmu = msm_iommu_new(dev, quirks);
>>>> -    if (IS_ERR_OR_NULL(mmu))
>>>> +    if (IS_ERR(mmu))
>>>>               return mmu;
>>> NAK, not having an IOMMU is a poor but legit usecase for some of devices
>>> which don't have IOMMU support yet (for example because of the buggy
>>> implementation for which we were not able to get all the hooks in).
>>>
>>> Please don't break compatibility for existing platforms.
>> Sure. I will remove this line of change. Though I have no idea in which
>> case msm_iommu_new() could return NULL after this patch.
> So, even without this chunk you are going to break the no-IOMMU case.
> Please don't. This will result in a regression report and a revert.
> 
> Instead please provide a way for the existing drivers to continue
> working. For example, something like:
> 
> if (IS_ERR(mmu) && ERR_PTR(mmu) == -ENODEV))
>      return NULL;

Oh I see. msm_iommu_new() returning NULL indicates a no-IOMMU case,
right? So perhaps we can make it explicit like below?

         if (!device_iommu_mapped(dev))
                 return NULL;

         domain = iommu_paging_domain_alloc(dev);
         if (IS_ERR(domain))
                 return ERR_CAST(domain);

Best regards,
baolu