On Fri, Mar 24, 2023 at 09:59:27PM +0100, Marek Szyprowski wrote:
> Hi Jason,
>
> On 21.03.2023 17:42, Jason Gunthorpe wrote:
> > On Tue, Mar 21, 2023 at 04:43:42PM +0100, Marek Szyprowski wrote:
> >> On 21.03.2023 15:41, Jason Gunthorpe wrote:
> >>> On Thu, Mar 16, 2023 at 12:25:14AM +0100, Marek Szyprowski wrote:
> >>>> There are some subtle differences between release_device() and
> >>>> set_platform_dma_ops() callbacks, so separate those two callbacks. Device
> >>>> links should be removed only in release_device(), because they were
> >>>> created in probe_device() on purpose and they are needed for proper
> >>>> Exynos IOMMU driver operation. While fixing this, remove the conditional
> >>>> code as it is not really needed.
> >>>>
> >>>> Reported-by: Jason Gunthorpe <jgg@xxxxxxxx>
> >>>> Fixes: 189d496b48b1 ("iommu/exynos: Add missing set_platform_dma_ops callback")
> >>>> Signed-off-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> >>>> ---
> >>>> v2:
> >>>> - keep set_platform_dma_ops only on ARM 32bit
> >>>>
> >>>> Some more background why set_platform_dma_ops is needed on ARM 32bit is
> >>>> available here:
> >>>> https://lore.kernel.org/all/9a12fcac-c347-5d81-acef-1124c50d0c37@xxxxxxx/
> >>>> ---
> >>>> drivers/iommu/exynos-iommu.c | 17 ++++++++++-------
> >>>> 1 file changed, 10 insertions(+), 7 deletions(-)
> >>> It seems OK, but do you know what state the device is left in after
> >>> exynos_iommu_detach_device ? Ie is it blocking or identity?
> >> identity
> > Can you do this cleanup like this instead?
>
> Frankly speaking I would like to have a minimal fix (like my $subject
> patch) applied to v6.3-rcX ASAP and then apply this identity domain
> support on top of than for -next (6.4). I've checked and your proposed
> patch works fine in my case, so You can send it as formal patch.
I thought this was a cosmetic fix, do you have an actual bug here? If
so we should split it as you say, but you should describe the actual
bug in the commit message.
Jason
