drm/i915: Retry gtt fault when out of fence registers

If we can't find a free fence register to handle a fault in the GMADR
range just return VM_FAULT_NOPAGE without populating the PTE so that
userspace will retry the access and trigger another fault. Eventually
we should find a free fence and the fault will get properly handled.

A further improvement idea might be to reserve a fence (or one per CPU?)
for the express purpose of handling faults without having to retry. But
that would require some additional work.

Looks like this may have gotten broken originally by
commit 39965b3 ("drm/i915: don't trash the gtt when running out of fences")
as that changed the errno to -EDEADLK which wasn't handle by the gtt
fault code either. But later in commit 2feeb52 ("drm/i915/gt: Fix
-EDEADLK handling regression") I changed it again to -ENOBUFS as -EDEADLK
was now getting used for the ww mutex dance. So this fix only makes
sense after that last commit.

Cc: stable@vger.kernel.org
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/9479
Fixes: 2feeb52 ("drm/i915/gt: Fix -EDEADLK handling regression")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231012132801.16292-1-ville.syrjala@linux.intel.com
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
(cherry picked from commit 7f403ca)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drivers/gpu/drm/i915/gem/i915_gem_mman.c

@@ -235,6 +235,7 @@ static vm_fault_t i915_error_to_vmf_fault(int err)
 	case 0:
 	case -EAGAIN:
 	case -ENOSPC: /* transient failure to evict? */
+	case -ENOBUFS: /* temporarily out of fences? */
 	case -ERESTARTSYS:
 	case -EINTR:
 	case -EBUSY: