Re: [PATCH v1] drm/xe: Drop dma mappings for wedged device

[Raag Jadav <raag.jadav@intel.com>]

On Tue, Mar 24, 2026 at 03:37:21PM +0000, Matthew Auld wrote:
> On 24/03/2026 15:14, Raag Jadav wrote:
> > On Tue, Mar 24, 2026 at 12:52:30PM +0000, Matthew Auld wrote:
> > > On 24/03/2026 07:13, Raag Jadav wrote:
> > > > As per uapi documentation[1], the prerequisite for wedged device is to
> > > > drop all dma mappings. Reuse xe_bo_pci_dev_remove_pinned() for this,
> > > > which iterates over external bo list and removes all dma mappings.
> > > > 
> > > > [1] Documentation/gpu/drm-uapi.rst
> > > 
> > > Can you point to where it says that? Do you just mean the: "disabling DMA to
> > > system memory" ?
> > > 
> > > One other thing that maybe jumps out is:
> > > 
> > > "All existing mmaps should be invalidated and
> > > page faults should be redirected to a dummy page"
> > > 
> > > Are we also missing that? We have the dummy page flow, but do we actually
> > > force everything to be refaulted?
> > 
> > I tried this with commit c020fff70d75 but it clearly doesn't cover
> > everything.
> 
> Ah right. Yeah, in theory if you do something like:
> 
> addr = mmap(bo);
> addr[0] = 0xdeadbeaf;
> igt_assert_eq(addr[0], 0xdeadbeaf);
> wedge();
> igt_assert_eq(addr[0], 0);
> 
> It should fail the second assert, I think. With the below fix in theory the
> access after the wedge will now trigger a new fault which will then map your
> dummy page (which is hopefully all zeroes), which I think is what the doc
> wants.

Weirdly, I remember testing something like above with commit c020fff70d75
and it worked even without below fix. Or perhaps I missed some obvious
detail in my testing :/

Let me revisit this.

Raag

> > > Something like:
> > > 
> > >   /* Clear all CPU mappings pointing to this device */
> > >   unmap_mapping_range(dev->anon_inode->i_mapping, 0, 0, 1);
> > 
> > Sure.
> > 
> > > > Signed-off-by: Raag Jadav <raag.jadav@intel.com>
> > > > ---
> > > > PS: This is pretty much uncharted territory for me, so please consider
> > > > this an RFC.
> > > > 
> > > >    drivers/gpu/drm/xe/xe_bo_evict.c | 8 +++++++-
> > > >    drivers/gpu/drm/xe/xe_bo_evict.h | 1 +
> > > >    drivers/gpu/drm/xe/xe_device.c   | 2 ++
> > > >    3 files changed, 10 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/gpu/drm/xe/xe_bo_evict.c b/drivers/gpu/drm/xe/xe_bo_evict.c
> > > > index 7661fca7f278..f741cda50b2d 100644
> > > > --- a/drivers/gpu/drm/xe/xe_bo_evict.c
> > > > +++ b/drivers/gpu/drm/xe/xe_bo_evict.c
> > > > @@ -270,7 +270,13 @@ int xe_bo_restore_late(struct xe_device *xe)
> > > >    	return ret;
> > > >    }
> > > > -static void xe_bo_pci_dev_remove_pinned(struct xe_device *xe)
> > > > +/**
> > > > + * xe_bo_pci_dev_remove_pinned() - Unmap external bos
> > > > + * @xe: xe device
> > > > + *
> > > > + * Drop dma mappings of all external pinned bos.
> > > > + */
> > > > +void xe_bo_pci_dev_remove_pinned(struct xe_device *xe)
> > > >    {
> > > >    	struct xe_tile *tile;
> > > >    	unsigned int id;
> > > > diff --git a/drivers/gpu/drm/xe/xe_bo_evict.h b/drivers/gpu/drm/xe/xe_bo_evict.h
> > > > index e8385cb7f5e9..6ce27e272780 100644
> > > > --- a/drivers/gpu/drm/xe/xe_bo_evict.h
> > > > +++ b/drivers/gpu/drm/xe/xe_bo_evict.h
> > > > @@ -15,6 +15,7 @@ void xe_bo_notifier_unprepare_all_pinned(struct xe_device *xe);
> > > >    int xe_bo_restore_early(struct xe_device *xe);
> > > >    int xe_bo_restore_late(struct xe_device *xe);
> > > > +void xe_bo_pci_dev_remove_pinned(struct xe_device *xe);
> > > >    void xe_bo_pci_dev_remove_all(struct xe_device *xe);
> > > >    int xe_bo_pinned_init(struct xe_device *xe);
> > > > diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> > > > index 207ad2eea412..ac51b04560df 100644
> > > > --- a/drivers/gpu/drm/xe/xe_device.c
> > > > +++ b/drivers/gpu/drm/xe/xe_device.c
> > > > @@ -1351,6 +1351,8 @@ void xe_device_declare_wedged(struct xe_device *xe)
> > > >    	for_each_gt(gt, xe, id)
> > > >    		xe_gt_declare_wedged(gt);
> > > > +	xe_bo_pci_dev_remove_pinned(xe);
> > > 
> > > AFAIK this just removes the iommu mappings for kernel BOs (small subset of
> > > BOs), if there are any. Also if you are not using iommu, then dma between
> > > GPU and system memory is still possible. And for userspace BOs nothing
> > > changes. But I guess this is still better than nothing and will maybe catch
> > > some misuse?
> > 
> > Yeah, I just floated this to start the discussion. Thanks for the pointers,
> > will explore this.
> > 
> > Raag
> > 
> > > >    	if (xe_device_wedged(xe)) {
> > > >    		/*
> > > >    		 * XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET is intended for debugging
> > > 
>