[PATCH v4 0/8] drm/xe/madvise: Add support for purgeable buffer objects

[PATCH v4 0/8] drm/xe/madvise: Add support for purgeable buffer objects

[Arvind Yadav]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=y, Size: 5549 bytes --]

This patch series introduces comprehensive support for purgeable buffer objects
in the Xe driver, enabling userspace to provide memory usage hints for better
memory management under system pressure.

Overview:

Purgeable memory allows applications to mark buffer objects as "not currently
needed" (DONTNEED), making them eligible for kernel reclamation during memory
pressure. This helps prevent OOM conditions and enables more efficient GPU
memory utilization for workloads with temporary or regeneratable data (caches,
intermediate results, decoded frames, etc.).

Purgeable BO Lifecycle:
1. WILLNEED (default): BO actively needed, kernel preserves backing store
2. DONTNEED (user hint): BO contents discardable, eligible for purging
3. PURGED (kernel action): Backing store reclaimed during memory pressure

Key Design Principles:
  - i915 compatibility: "Once purged, always purged" semantics - purged BOs
    remain permanently invalid and must be destroyed/recreated
  - Per-VMA state tracking: Each VMA tracks its own purgeable state, BO is
    only marked DONTNEED when ALL VMAs across ALL VMs agree (Thomas Hellström)
  - Safety first: Imported/exported dma-bufs blocked from purgeable state -
    no visibility into external device usage (Matt Roper)
  - Multiple protection layers: Validation in madvise, VM bind, mmap, and
    fault handlers
  - Async TLB invalidation: Uses xe_bo_trigger_rebind() for non-blocking
    GPU mapping invalidation
  - Scratch PTE support: Fault-mode VMs use scratch pages for safe zero reads
    on purged BO access.
  - Purgeable state is not applied to imported/exported dma-bufs,
    those BOs always behave as WILLNEED.
  - TTM shrinker integration: Encapsulated helpers manage xe_ttm_tt->purgeable
    flag and shrinker page accounting (shrinkable vs purgeable buckets)

v2 Changes:
  - Reordered patches: Moved shared BO helper before main implementation for
    proper dependency order
  - Fixed reference counting in mmap offset validation (use drm_gem_object_put)
  - Removed incorrect claims about madvise(WILLNEED) restoring purged BOs
  - Fixed error code documentation inconsistencies
  - Initialize purge_state_val fields to prevent kernel memory leaks
  - Use xe_bo_trigger_rebind() for async TLB invalidation (Thomas Hellström)
  - Add NULL rebind with scratch PTEs for fault mode (Thomas Hellström)
  - Implement i915-compatible retained field logic (Thomas Hellström)
  - Skip BO validation for purged BOs in page fault handler (crash fix)
  - Add scratch VM check in page fault path (non-scratch VMs fail fault)

v3 Changes (addressing Matt and Thomas Hellström feedback):
  - Per-VMA purgeable state tracking: Added xe_vma->purgeable_state field
  - Complete VMA check: xe_bo_all_vmas_dontneed() walks all VMAs across all
    VMs to ensure unanimous DONTNEED before marking BO purgeable
  - VMA unbind recheck: Added xe_bo_recheck_purgeable_on_vma_unbind() to
    re-evaluate BO state when VMAs are destroyed
  - Block external dma-bufs: Added xe_bo_is_external_dmabuf() check using
    drm_gem_is_imported() and obj->dma_buf to prevent purging imported/exported BOs
  - Consistent lockdep enforcement: Added xe_bo_assert_held() to all helpers
    that access madv_purgeable state
  - Simplified page table logic: Renamed is_null to is_null_or_purged in
    xe_pt_stage_bind_entry() - purged BOs treated identically to null VMAs
  - Removed unnecessary checks: Dropped redundant "&& bo" check in xe_ttm_bo_purge()
  - Xe-specific warnings: Changed drm_warn() to XE_WARN_ON() in purge path
  - Moved purge checks under locks: Purge state validation now done after
    acquiring dma-resv lock in vma_lock_and_validate() and xe_pagefault_begin()
  - Race-free fault handling: Removed unlocked purge check from
    xe_pagefault_handle_vma(), moved to locked xe_pagefault_begin()
  - Shrinker helper functions: Added xe_bo_set_purgeable_shrinker() and
    xe_bo_clear_purgeable_shrinker() to encapsulate TTM purgeable flag updates
    and shrinker page accounting, improving code clarity and maintainability

v4 Changes (addressing Matt and Thomas Hellström feedback):
  - UAPI: Removed '__u64 reserved' field from purge_state_val union to fit
    16-byte size constraint (Matt)
  - Changed madv_purgeable from atomic_t to u32 across all patches (Matt)
  - CPU fault handling: Added purged check to fastpath (xe_bo_cpu_fault_fastpath)
    to prevent hang when accessing existing mmap of purged BO

Arvind Yadav (7):
  drm/xe/bo: Add purgeable bo state tracking and field madv to xe_bo
  drm/xe/madvise: Implement purgeable buffer object support
  drm/xe/bo: Handle CPU faults on purged buffer objects
  drm/xe/vm: Prevent binding of purged buffer objects
  drm/xe/madvise: Implement per-VMA purgeable state tracking
  drm/xe/madvise: Block imported and exported dma-bufs
  drm/xe/bo: Add purgeable shrinker state helpers

Himal Prasad Ghimiray (1):
  drm/xe/uapi: Add UAPI support for purgeable buffer objects

 drivers/gpu/drm/xe/xe_bo.c         | 127 +++++++++++++++--
 drivers/gpu/drm/xe/xe_bo.h         |  59 ++++++++
 drivers/gpu/drm/xe/xe_bo_types.h   |   3 +
 drivers/gpu/drm/xe/xe_pagefault.c  |  12 ++
 drivers/gpu/drm/xe/xe_pt.c         |  38 +++++-
 drivers/gpu/drm/xe/xe_vm.c         |  46 +++++--
 drivers/gpu/drm/xe/xe_vm_madvise.c | 210 +++++++++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_vm_madvise.h |   3 +
 drivers/gpu/drm/xe/xe_vm_types.h   |  11 ++
 include/uapi/drm/xe_drm.h          |  37 +++++
 10 files changed, 519 insertions(+), 27 deletions(-)

-- 
2.43.0

[PATCH v4 1/8] drm/xe/uapi: Add UAPI support for purgeable buffer objects

[Arvind Yadav]

From: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>

Extend the DRM_XE_MADVISE ioctl to support purgeable buffer object
management by adding DRM_XE_VMA_ATTR_PURGEABLE_STATE attribute type.

This allows userspace applications to provide memory usage hints to
the kernel for better memory management under pressure:

This allows userspace applications to provide memory usage hints to
the kernel for better memory management under pressure:

- WILLNEED: Buffer is needed and should not be purged. If the BO was
  previously purged, retained field returns 0 indicating backing store
  was lost (once purged, always purged semantics matching i915).

- DONTNEED: Buffer is not currently needed and may be purged by the
  kernel under memory pressure to free resources. Only applies to
  non-shared BOs.

The implementation includes a 'retained' output field (matching i915's
drm_i915_gem_madvise.retained) that indicates whether the BO's backing
store still exists (1) or has been purged (0).

v2:
  - Add PURGED state for read-only status, change ioctl to DRM_IOWR,
    add retained field for i915 compatibility

v3:
  - UAPI rule should not be changed (Matthew Brost)
  - Make 'retained' a userptr (Matthew Brost)

v4:
  - You cannot make this part of the union (purge_state_val) larger
    than the existing union (16 bytes). So just drop the '__u64 reserved'
    field. (Matt)

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 include/uapi/drm/xe_drm.h | 37 +++++++++++++++++++++++++++++++++++++
 1 file changed, 37 insertions(+)

diff --git a/include/uapi/drm/xe_drm.h b/include/uapi/drm/xe_drm.h
index 077e66a682e2..7b3901e4b85e 100644
--- a/include/uapi/drm/xe_drm.h
+++ b/include/uapi/drm/xe_drm.h
@@ -2099,6 +2099,7 @@ struct drm_xe_madvise {
 #define DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC	0
 #define DRM_XE_MEM_RANGE_ATTR_ATOMIC		1
 #define DRM_XE_MEM_RANGE_ATTR_PAT		2
+#define DRM_XE_VMA_ATTR_PURGEABLE_STATE		3
 	/** @type: type of attribute */
 	__u32 type;
 
@@ -2189,6 +2190,42 @@ struct drm_xe_madvise {
 			/** @pat_index.reserved: Reserved */
 			__u64 reserved;
 		} pat_index;
+
+		/**
+		 * @purge_state_val: Purgeable state configuration
+		 *
+		 * Used when @type == DRM_XE_VMA_ATTR_PURGEABLE_STATE.
+		 *
+		 * Configures the purgeable state of buffer objects in the specified
+		 * virtual address range. This allows applications to hint to the kernel
+		 * about bo's usage patterns for better memory management.
+		 *
+		 * Supported values for @purge_state_val.val:
+		 *  - DRM_XE_VMA_PURGEABLE_STATE_WILLNEED (0): Marks BO as needed.
+		 *    If BO was purged, returns retained=0 (backing store lost).
+		 *
+		 *  - DRM_XE_VMA_PURGEABLE_STATE_DONTNEED (1): Hints that BO is not
+		 *    currently needed. Kernel may purge it under memory pressure.
+		 *    Only applies to non-shared BOs. Returns retained=1 if not purged.
+		 */
+		struct {
+#define DRM_XE_VMA_PURGEABLE_STATE_WILLNEED	0
+#define DRM_XE_VMA_PURGEABLE_STATE_DONTNEED	1
+			/** @purge_state_val.val: value for DRM_XE_VMA_ATTR_PURGEABLE_STATE */
+			__u32 val;
+
+			/* @purge_state_val.pad */
+			__u32 pad;
+			/**
+			 * @purge_state_val.retained: Pointer to output field for backing
+			 * store status.
+			 *
+			 * Userspace provides a pointer to u32. Kernel writes to it:
+			 * 1 if backing store exists, 0 if purged.
+			 * Similar to i915's drm_i915_gem_madvise.retained field.
+			 */
+			__u64 retained;
+		} purge_state_val;
 	};
 
 	/** @reserved: Reserved */
-- 
2.43.0

Re: [PATCH v4 1/8] drm/xe/uapi: Add UAPI support for purgeable buffer objects

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:47AM +0530, Arvind Yadav wrote:
> From: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> 
> Extend the DRM_XE_MADVISE ioctl to support purgeable buffer object
> management by adding DRM_XE_VMA_ATTR_PURGEABLE_STATE attribute type.
> 
> This allows userspace applications to provide memory usage hints to
> the kernel for better memory management under pressure:
> 
> This allows userspace applications to provide memory usage hints to
> the kernel for better memory management under pressure:
> 
> - WILLNEED: Buffer is needed and should not be purged. If the BO was
>   previously purged, retained field returns 0 indicating backing store
>   was lost (once purged, always purged semantics matching i915).
> 
> - DONTNEED: Buffer is not currently needed and may be purged by the
>   kernel under memory pressure to free resources. Only applies to
>   non-shared BOs.
> 
> The implementation includes a 'retained' output field (matching i915's
> drm_i915_gem_madvise.retained) that indicates whether the BO's backing
> store still exists (1) or has been purged (0).
> 
> v2:
>   - Add PURGED state for read-only status, change ioctl to DRM_IOWR,
>     add retained field for i915 compatibility
> 
> v3:
>   - UAPI rule should not be changed (Matthew Brost)
>   - Make 'retained' a userptr (Matthew Brost)
> 
> v4:
>   - You cannot make this part of the union (purge_state_val) larger
>     than the existing union (16 bytes). So just drop the '__u64 reserved'
>     field. (Matt)
> 
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  include/uapi/drm/xe_drm.h | 37 +++++++++++++++++++++++++++++++++++++
>  1 file changed, 37 insertions(+)
> 
> diff --git a/include/uapi/drm/xe_drm.h b/include/uapi/drm/xe_drm.h
> index 077e66a682e2..7b3901e4b85e 100644
> --- a/include/uapi/drm/xe_drm.h
> +++ b/include/uapi/drm/xe_drm.h
> @@ -2099,6 +2099,7 @@ struct drm_xe_madvise {
>  #define DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC	0
>  #define DRM_XE_MEM_RANGE_ATTR_ATOMIC		1
>  #define DRM_XE_MEM_RANGE_ATTR_PAT		2
> +#define DRM_XE_VMA_ATTR_PURGEABLE_STATE		3
>  	/** @type: type of attribute */
>  	__u32 type;
>  
> @@ -2189,6 +2190,42 @@ struct drm_xe_madvise {
>  			/** @pat_index.reserved: Reserved */
>  			__u64 reserved;
>  		} pat_index;
> +
> +		/**
> +		 * @purge_state_val: Purgeable state configuration
> +		 *
> +		 * Used when @type == DRM_XE_VMA_ATTR_PURGEABLE_STATE.
> +		 *
> +		 * Configures the purgeable state of buffer objects in the specified
> +		 * virtual address range. This allows applications to hint to the kernel
> +		 * about bo's usage patterns for better memory management.
> +		 *
> +		 * Supported values for @purge_state_val.val:
> +		 *  - DRM_XE_VMA_PURGEABLE_STATE_WILLNEED (0): Marks BO as needed.
> +		 *    If BO was purged, returns retained=0 (backing store lost).
> +		 *
> +		 *  - DRM_XE_VMA_PURGEABLE_STATE_DONTNEED (1): Hints that BO is not
> +		 *    currently needed. Kernel may purge it under memory pressure.
> +		 *    Only applies to non-shared BOs. Returns retained=1 if not purged.
> +		 */
> +		struct {
> +#define DRM_XE_VMA_PURGEABLE_STATE_WILLNEED	0
> +#define DRM_XE_VMA_PURGEABLE_STATE_DONTNEED	1
> +			/** @purge_state_val.val: value for DRM_XE_VMA_ATTR_PURGEABLE_STATE */
> +			__u32 val;
> +
> +			/* @purge_state_val.pad */
> +			__u32 pad;
> +			/**
> +			 * @purge_state_val.retained: Pointer to output field for backing
> +			 * store status.
> +			 *
> +			 * Userspace provides a pointer to u32. Kernel writes to it:
> +			 * 1 if backing store exists, 0 if purged.
> +			 * Similar to i915's drm_i915_gem_madvise.retained field.
> +			 */
> +			__u64 retained;

@Rodrigo + @Thomas.

Here we are adding a userptr for copying the result back to userspace
because the existing madvise IOCTL is IOW. If we could change this IOCTL
to IOWR, we could let DRM handle this for us. I'm not sure whether
changing an IOCTL from IOW to IOWR is allowed. My understanding is that
the ioctl encoding is basically the same aside from some control bits,
so switching from IOW → IOWR shouldn’t break anything, but I'm unclear
on the rules. 

What do you two think?

Matt

> +		} purge_state_val;
>  	};
>  
>  	/** @reserved: Reserved */
> -- 
> 2.43.0
> 

Re: [PATCH v4 1/8] drm/xe/uapi: Add UAPI support for purgeable buffer objects

[Vivi, Rodrigo]

On Tue, 2026-01-20 at 09:20 -0800, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:47AM +0530, Arvind Yadav wrote:
> > From: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> > 
> > Extend the DRM_XE_MADVISE ioctl to support purgeable buffer object
> > management by adding DRM_XE_VMA_ATTR_PURGEABLE_STATE attribute
> > type.
> > 
> > This allows userspace applications to provide memory usage hints to
> > the kernel for better memory management under pressure:
> > 
> > This allows userspace applications to provide memory usage hints to
> > the kernel for better memory management under pressure:
> > 
> > - WILLNEED: Buffer is needed and should not be purged. If the BO
> > was
> >   previously purged, retained field returns 0 indicating backing
> > store
> >   was lost (once purged, always purged semantics matching i915).
> > 
> > - DONTNEED: Buffer is not currently needed and may be purged by the
> >   kernel under memory pressure to free resources. Only applies to
> >   non-shared BOs.
> > 
> > The implementation includes a 'retained' output field (matching
> > i915's
> > drm_i915_gem_madvise.retained) that indicates whether the BO's
> > backing
> > store still exists (1) or has been purged (0).
> > 
> > v2:
> >   - Add PURGED state for read-only status, change ioctl to
> > DRM_IOWR,
> >     add retained field for i915 compatibility
> > 
> > v3:
> >   - UAPI rule should not be changed (Matthew Brost)
> >   - Make 'retained' a userptr (Matthew Brost)
> > 
> > v4:
> >   - You cannot make this part of the union (purge_state_val) larger
> >     than the existing union (16 bytes). So just drop the '__u64
> > reserved'
> >     field. (Matt)
> > 
> > Cc: Matthew Brost <matthew.brost@intel.com>
> > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > Signed-off-by: Himal Prasad Ghimiray
> > <himal.prasad.ghimiray@intel.com>
> > Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> > ---
> >  include/uapi/drm/xe_drm.h | 37
> > +++++++++++++++++++++++++++++++++++++
> >  1 file changed, 37 insertions(+)
> > 
> > diff --git a/include/uapi/drm/xe_drm.h b/include/uapi/drm/xe_drm.h
> > index 077e66a682e2..7b3901e4b85e 100644
> > --- a/include/uapi/drm/xe_drm.h
> > +++ b/include/uapi/drm/xe_drm.h
> > @@ -2099,6 +2099,7 @@ struct drm_xe_madvise {
> >  #define DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC	0
> >  #define DRM_XE_MEM_RANGE_ATTR_ATOMIC		1
> >  #define DRM_XE_MEM_RANGE_ATTR_PAT		2
> > +#define DRM_XE_VMA_ATTR_PURGEABLE_STATE		3
> >  	/** @type: type of attribute */
> >  	__u32 type;
> >  
> > @@ -2189,6 +2190,42 @@ struct drm_xe_madvise {
> >  			/** @pat_index.reserved: Reserved */
> >  			__u64 reserved;
> >  		} pat_index;
> > +
> > +		/**
> > +		 * @purge_state_val: Purgeable state configuration
> > +		 *
> > +		 * Used when @type ==
> > DRM_XE_VMA_ATTR_PURGEABLE_STATE.
> > +		 *
> > +		 * Configures the purgeable state of buffer
> > objects in the specified
> > +		 * virtual address range. This allows applications
> > to hint to the kernel
> > +		 * about bo's usage patterns for better memory
> > management.
> > +		 *
> > +		 * Supported values for @purge_state_val.val:
> > +		 *  - DRM_XE_VMA_PURGEABLE_STATE_WILLNEED (0):
> > Marks BO as needed.
> > +		 *    If BO was purged, returns retained=0
> > (backing store lost).
> > +		 *
> > +		 *  - DRM_XE_VMA_PURGEABLE_STATE_DONTNEED (1):
> > Hints that BO is not
> > +		 *    currently needed. Kernel may purge it under
> > memory pressure.
> > +		 *    Only applies to non-shared BOs. Returns
> > retained=1 if not purged.
> > +		 */
> > +		struct {
> > +#define DRM_XE_VMA_PURGEABLE_STATE_WILLNEED	0
> > +#define DRM_XE_VMA_PURGEABLE_STATE_DONTNEED	1
> > +			/** @purge_state_val.val: value for
> > DRM_XE_VMA_ATTR_PURGEABLE_STATE */
> > +			__u32 val;
> > +
> > +			/* @purge_state_val.pad */
> > +			__u32 pad;
> > +			/**
> > +			 * @purge_state_val.retained: Pointer to
> > output field for backing
> > +			 * store status.
> > +			 *
> > +			 * Userspace provides a pointer to u32.
> > Kernel writes to it:
> > +			 * 1 if backing store exists, 0 if purged.
> > +			 * Similar to i915's
> > drm_i915_gem_madvise.retained field.
> > +			 */
> > +			__u64 retained;
> 
> @Rodrigo + @Thomas.
> 
> Here we are adding a userptr for copying the result back to userspace
> because the existing madvise IOCTL is IOW. If we could change this
> IOCTL
> to IOWR, we could let DRM handle this for us. I'm not sure whether
> changing an IOCTL from IOW to IOWR is allowed. My understanding is
> that
> the ioctl encoding is basically the same aside from some control
> bits,
> so switching from IOW → IOWR shouldn’t break anything, but I'm
> unclear
> on the rules. 

I'm not so confident it wouldn't break anything as I'm wondering if
this gets into the final number construction... and old userspace with
new kernel would hit a different slot.

> 
> What do you two think?

Perhaps we would need a new ioctl and deprecate this with time?

> 
> Matt
> 
> > +		} purge_state_val;
> >  	};
> >  
> >  	/** @reserved: Reserved */
> > -- 
> > 2.43.0
> > 

[PATCH v4 2/8] drm/xe/bo: Add purgeable bo state tracking and field madv to xe_bo

[Arvind Yadav]

Add infrastructure for tracking purgeable state of buffer objects.
This includes:

Introduce enum xe_madv_purgeable_state with three states:
   - XE_MADV_PURGEABLE_WILLNEED (0): BO is needed and should not be
     purged. This is the default state for all BOs.

   - XE_MADV_PURGEABLE_DONTNEED (1): BO is not currently needed and
     can be purged by the kernel under memory pressure to reclaim
     resources. Only non-shared BOs can be marked as DONTNEED.

   - XE_MADV_PURGEABLE_PURGED (2): BO has been purged by the kernel.
     Accessing a purged BO results in error. Follows i915 semantics
     where once purged, the BO remains permanently invalid ("once
     purged, always purged").

Add atomic_t madv field to struct xe_bo for state tracking
  of purgeable state across concurrent access paths

v2:
  - Add xe_bo_is_purged() helper, improve state documentation

v3:
  - Add the kernel doc(Matthew Brost)
  - Add the new helpers xe_bo_madv_is_dontneed(Matthew Brost)

v4:
  - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 drivers/gpu/drm/xe/xe_bo.h       | 56 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_bo_types.h |  3 ++
 2 files changed, 59 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
index 8ab4474129c3..00e93b3065c9 100644
--- a/drivers/gpu/drm/xe/xe_bo.h
+++ b/drivers/gpu/drm/xe/xe_bo.h
@@ -86,6 +86,28 @@
 
 #define XE_PCI_BARRIER_MMAP_OFFSET	(0x50 << XE_PTE_SHIFT)
 
+/**
+ * enum xe_madv_purgeable_state - Buffer object purgeable state enumeration
+ *
+ * This enum defines the possible purgeable states for a buffer object,
+ * allowing userspace to provide memory usage hints to the kernel for
+ * better memory management under pressure.
+ *
+ * @XE_MADV_PURGEABLE_WILLNEED: The buffer object is needed and should not be purged.
+ * This is the default state.
+ * @XE_MADV_PURGEABLE_DONTNEED: The buffer object is not currently needed and can be
+ * purged by the kernel under memory pressure.
+ * @XE_MADV_PURGEABLE_PURGED: The buffer object has been purged by the kernel.
+ *
+ * Accessing a purged buffer will result in an error. Per i915 semantics,
+ * once purged, a BO remains permanently invalid and must be destroyed and recreated.
+ */
+enum xe_madv_purgeable_state {
+	XE_MADV_PURGEABLE_WILLNEED,
+	XE_MADV_PURGEABLE_DONTNEED,
+	XE_MADV_PURGEABLE_PURGED,
+};
+
 struct sg_table;
 
 struct xe_bo *xe_bo_alloc(void);
@@ -214,6 +236,40 @@ static inline bool xe_bo_is_protected(const struct xe_bo *bo)
 	return bo->pxp_key_instance;
 }
 
+/**
+ * xe_bo_is_purged() - Check if buffer object has been purged
+ * @bo: The buffer object to check
+ *
+ * Checks if the buffer object's backing store has been discarded by the
+ * kernel due to memory pressure after being marked as purgeable (DONTNEED).
+ * Once purged, the BO cannot be restored and any attempt to use it will fail.
+ *
+ * Context: Caller must hold the BO's dma-resv lock
+ * Return: true if the BO has been purged, false otherwise
+ */
+static inline bool xe_bo_is_purged(struct xe_bo *bo)
+{
+	xe_bo_assert_held(bo);
+	return bo->madv_purgeable == XE_MADV_PURGEABLE_PURGED;
+}
+
+/**
+ * xe_bo_madv_is_dontneed() - Check if BO is marked as DONTNEED
+ * @bo: The buffer object to check
+ *
+ * Checks if userspace has marked this BO as DONTNEED (i.e., its contents
+ * are not currently needed and can be discarded under memory pressure).
+ * This is used internally to decide whether a BO is eligible for purging.
+ *
+ * Context: Caller must hold the BO's dma-resv lock
+ * Return: true if the BO is marked DONTNEED, false otherwise
+ */
+static inline bool xe_bo_madv_is_dontneed(struct xe_bo *bo)
+{
+	xe_bo_assert_held(bo);
+	return bo->madv_purgeable == XE_MADV_PURGEABLE_DONTNEED;
+}
+
 static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo)
 {
 	if (likely(bo)) {
diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h
index d4fe3c8dca5b..6acfed0c0bb4 100644
--- a/drivers/gpu/drm/xe/xe_bo_types.h
+++ b/drivers/gpu/drm/xe/xe_bo_types.h
@@ -108,6 +108,9 @@ struct xe_bo {
 	 * from default
 	 */
 	u64 min_align;
+
+	/** @madv_purgeable: user space advise on BO purgeability */
+	u32 madv_purgeable;
 };
 
 #endif
-- 
2.43.0

Re: [PATCH v4 2/8] drm/xe/bo: Add purgeable bo state tracking and field madv to xe_bo

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:48AM +0530, Arvind Yadav wrote:
> Add infrastructure for tracking purgeable state of buffer objects.
> This includes:
> 
> Introduce enum xe_madv_purgeable_state with three states:
>    - XE_MADV_PURGEABLE_WILLNEED (0): BO is needed and should not be
>      purged. This is the default state for all BOs.
> 
>    - XE_MADV_PURGEABLE_DONTNEED (1): BO is not currently needed and
>      can be purged by the kernel under memory pressure to reclaim
>      resources. Only non-shared BOs can be marked as DONTNEED.
> 
>    - XE_MADV_PURGEABLE_PURGED (2): BO has been purged by the kernel.
>      Accessing a purged BO results in error. Follows i915 semantics
>      where once purged, the BO remains permanently invalid ("once
>      purged, always purged").
> 
> Add atomic_t madv field to struct xe_bo for state tracking
>   of purgeable state across concurrent access paths
> 
> v2:
>   - Add xe_bo_is_purged() helper, improve state documentation
> 
> v3:
>   - Add the kernel doc(Matthew Brost)
>   - Add the new helpers xe_bo_madv_is_dontneed(Matthew Brost)
> 
> v4:
>   - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
> 
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_bo.h       | 56 ++++++++++++++++++++++++++++++++
>  drivers/gpu/drm/xe/xe_bo_types.h |  3 ++
>  2 files changed, 59 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
> index 8ab4474129c3..00e93b3065c9 100644
> --- a/drivers/gpu/drm/xe/xe_bo.h
> +++ b/drivers/gpu/drm/xe/xe_bo.h
> @@ -86,6 +86,28 @@
>  
>  #define XE_PCI_BARRIER_MMAP_OFFSET	(0x50 << XE_PTE_SHIFT)
>  
> +/**
> + * enum xe_madv_purgeable_state - Buffer object purgeable state enumeration
> + *
> + * This enum defines the possible purgeable states for a buffer object,
> + * allowing userspace to provide memory usage hints to the kernel for
> + * better memory management under pressure.
> + *
> + * @XE_MADV_PURGEABLE_WILLNEED: The buffer object is needed and should not be purged.
> + * This is the default state.
> + * @XE_MADV_PURGEABLE_DONTNEED: The buffer object is not currently needed and can be
> + * purged by the kernel under memory pressure.
> + * @XE_MADV_PURGEABLE_PURGED: The buffer object has been purged by the kernel.
> + *
> + * Accessing a purged buffer will result in an error. Per i915 semantics,
> + * once purged, a BO remains permanently invalid and must be destroyed and recreated.
> + */
> +enum xe_madv_purgeable_state {
> +	XE_MADV_PURGEABLE_WILLNEED,
> +	XE_MADV_PURGEABLE_DONTNEED,
> +	XE_MADV_PURGEABLE_PURGED,
> +};
> +
>  struct sg_table;
>  
>  struct xe_bo *xe_bo_alloc(void);
> @@ -214,6 +236,40 @@ static inline bool xe_bo_is_protected(const struct xe_bo *bo)
>  	return bo->pxp_key_instance;
>  }
>  
> +/**
> + * xe_bo_is_purged() - Check if buffer object has been purged
> + * @bo: The buffer object to check
> + *
> + * Checks if the buffer object's backing store has been discarded by the
> + * kernel due to memory pressure after being marked as purgeable (DONTNEED).
> + * Once purged, the BO cannot be restored and any attempt to use it will fail.
> + *
> + * Context: Caller must hold the BO's dma-resv lock
> + * Return: true if the BO has been purged, false otherwise
> + */
> +static inline bool xe_bo_is_purged(struct xe_bo *bo)
> +{
> +	xe_bo_assert_held(bo);
> +	return bo->madv_purgeable == XE_MADV_PURGEABLE_PURGED;
> +}
> +
> +/**
> + * xe_bo_madv_is_dontneed() - Check if BO is marked as DONTNEED
> + * @bo: The buffer object to check
> + *
> + * Checks if userspace has marked this BO as DONTNEED (i.e., its contents
> + * are not currently needed and can be discarded under memory pressure).
> + * This is used internally to decide whether a BO is eligible for purging.
> + *
> + * Context: Caller must hold the BO's dma-resv lock
> + * Return: true if the BO is marked DONTNEED, false otherwise
> + */
> +static inline bool xe_bo_madv_is_dontneed(struct xe_bo *bo)
> +{
> +	xe_bo_assert_held(bo);
> +	return bo->madv_purgeable == XE_MADV_PURGEABLE_DONTNEED;
> +}
> +
>  static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo)
>  {
>  	if (likely(bo)) {
> diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h
> index d4fe3c8dca5b..6acfed0c0bb4 100644
> --- a/drivers/gpu/drm/xe/xe_bo_types.h
> +++ b/drivers/gpu/drm/xe/xe_bo_types.h
> @@ -108,6 +108,9 @@ struct xe_bo {
>  	 * from default
>  	 */
>  	u64 min_align;
> +
> +	/** @madv_purgeable: user space advise on BO purgeability */

, protected by BO's dma-resv lock.

Everything else LGTM.

Matt

> +	u32 madv_purgeable;
>  };
>  
>  #endif
> -- 
> 2.43.0
> 

Re: [PATCH v4 2/8] drm/xe/bo: Add purgeable bo state tracking and field madv to xe_bo

[Yadav, Arvind]

On 20-01-2026 23:15, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:48AM +0530, Arvind Yadav wrote:
>> Add infrastructure for tracking purgeable state of buffer objects.
>> This includes:
>>
>> Introduce enum xe_madv_purgeable_state with three states:
>>     - XE_MADV_PURGEABLE_WILLNEED (0): BO is needed and should not be
>>       purged. This is the default state for all BOs.
>>
>>     - XE_MADV_PURGEABLE_DONTNEED (1): BO is not currently needed and
>>       can be purged by the kernel under memory pressure to reclaim
>>       resources. Only non-shared BOs can be marked as DONTNEED.
>>
>>     - XE_MADV_PURGEABLE_PURGED (2): BO has been purged by the kernel.
>>       Accessing a purged BO results in error. Follows i915 semantics
>>       where once purged, the BO remains permanently invalid ("once
>>       purged, always purged").
>>
>> Add atomic_t madv field to struct xe_bo for state tracking
>>    of purgeable state across concurrent access paths
>>
>> v2:
>>    - Add xe_bo_is_purged() helper, improve state documentation
>>
>> v3:
>>    - Add the kernel doc(Matthew Brost)
>>    - Add the new helpers xe_bo_madv_is_dontneed(Matthew Brost)
>>
>> v4:
>>    - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
>>
>> Cc: Matthew Brost <matthew.brost@intel.com>
>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
>> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
>> ---
>>   drivers/gpu/drm/xe/xe_bo.h       | 56 ++++++++++++++++++++++++++++++++
>>   drivers/gpu/drm/xe/xe_bo_types.h |  3 ++
>>   2 files changed, 59 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
>> index 8ab4474129c3..00e93b3065c9 100644
>> --- a/drivers/gpu/drm/xe/xe_bo.h
>> +++ b/drivers/gpu/drm/xe/xe_bo.h
>> @@ -86,6 +86,28 @@
>>   
>>   #define XE_PCI_BARRIER_MMAP_OFFSET	(0x50 << XE_PTE_SHIFT)
>>   
>> +/**
>> + * enum xe_madv_purgeable_state - Buffer object purgeable state enumeration
>> + *
>> + * This enum defines the possible purgeable states for a buffer object,
>> + * allowing userspace to provide memory usage hints to the kernel for
>> + * better memory management under pressure.
>> + *
>> + * @XE_MADV_PURGEABLE_WILLNEED: The buffer object is needed and should not be purged.
>> + * This is the default state.
>> + * @XE_MADV_PURGEABLE_DONTNEED: The buffer object is not currently needed and can be
>> + * purged by the kernel under memory pressure.
>> + * @XE_MADV_PURGEABLE_PURGED: The buffer object has been purged by the kernel.
>> + *
>> + * Accessing a purged buffer will result in an error. Per i915 semantics,
>> + * once purged, a BO remains permanently invalid and must be destroyed and recreated.
>> + */
>> +enum xe_madv_purgeable_state {
>> +	XE_MADV_PURGEABLE_WILLNEED,
>> +	XE_MADV_PURGEABLE_DONTNEED,
>> +	XE_MADV_PURGEABLE_PURGED,
>> +};
>> +
>>   struct sg_table;
>>   
>>   struct xe_bo *xe_bo_alloc(void);
>> @@ -214,6 +236,40 @@ static inline bool xe_bo_is_protected(const struct xe_bo *bo)
>>   	return bo->pxp_key_instance;
>>   }
>>   
>> +/**
>> + * xe_bo_is_purged() - Check if buffer object has been purged
>> + * @bo: The buffer object to check
>> + *
>> + * Checks if the buffer object's backing store has been discarded by the
>> + * kernel due to memory pressure after being marked as purgeable (DONTNEED).
>> + * Once purged, the BO cannot be restored and any attempt to use it will fail.
>> + *
>> + * Context: Caller must hold the BO's dma-resv lock
>> + * Return: true if the BO has been purged, false otherwise
>> + */
>> +static inline bool xe_bo_is_purged(struct xe_bo *bo)
>> +{
>> +	xe_bo_assert_held(bo);
>> +	return bo->madv_purgeable == XE_MADV_PURGEABLE_PURGED;
>> +}
>> +
>> +/**
>> + * xe_bo_madv_is_dontneed() - Check if BO is marked as DONTNEED
>> + * @bo: The buffer object to check
>> + *
>> + * Checks if userspace has marked this BO as DONTNEED (i.e., its contents
>> + * are not currently needed and can be discarded under memory pressure).
>> + * This is used internally to decide whether a BO is eligible for purging.
>> + *
>> + * Context: Caller must hold the BO's dma-resv lock
>> + * Return: true if the BO is marked DONTNEED, false otherwise
>> + */
>> +static inline bool xe_bo_madv_is_dontneed(struct xe_bo *bo)
>> +{
>> +	xe_bo_assert_held(bo);
>> +	return bo->madv_purgeable == XE_MADV_PURGEABLE_DONTNEED;
>> +}
>> +
>>   static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo)
>>   {
>>   	if (likely(bo)) {
>> diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h
>> index d4fe3c8dca5b..6acfed0c0bb4 100644
>> --- a/drivers/gpu/drm/xe/xe_bo_types.h
>> +++ b/drivers/gpu/drm/xe/xe_bo_types.h
>> @@ -108,6 +108,9 @@ struct xe_bo {
>>   	 * from default
>>   	 */
>>   	u64 min_align;
>> +
>> +	/** @madv_purgeable: user space advise on BO purgeability */
> , protected by BO's dma-resv lock.
>
> Everything else LGTM.


Noted, I will do the changes as per suggestion.


Thanks,
Arvind

>
> Matt
>
>> +	u32 madv_purgeable;
>>   };
>>   
>>   #endif
>> -- 
>> 2.43.0
>>

Re: [PATCH v4 2/8] drm/xe/bo: Add purgeable bo state tracking and field madv to xe_bo

[Thomas Hellström]

On Tue, 2026-01-20 at 09:45 -0800, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:48AM +0530, Arvind Yadav wrote:
> > Add infrastructure for tracking purgeable state of buffer objects.
> > This includes:
> > 
> > Introduce enum xe_madv_purgeable_state with three states:
> >    - XE_MADV_PURGEABLE_WILLNEED (0): BO is needed and should not be
> >      purged. This is the default state for all BOs.
> > 
> >    - XE_MADV_PURGEABLE_DONTNEED (1): BO is not currently needed and
> >      can be purged by the kernel under memory pressure to reclaim
> >      resources. Only non-shared BOs can be marked as DONTNEED.
> > 
> >    - XE_MADV_PURGEABLE_PURGED (2): BO has been purged by the
> > kernel.
> >      Accessing a purged BO results in error. Follows i915 semantics
> >      where once purged, the BO remains permanently invalid ("once
> >      purged, always purged").
> > 
> > Add atomic_t madv field to struct xe_bo for state tracking
> >   of purgeable state across concurrent access paths
> > 
> > v2:
> >   - Add xe_bo_is_purged() helper, improve state documentation
> > 
> > v3:
> >   - Add the kernel doc(Matthew Brost)
> >   - Add the new helpers xe_bo_madv_is_dontneed(Matthew Brost)
> > 
> > v4:
> >   - @madv_purgeable atomic_t → u32 change across all relevant
> > patches. (Matt)
> > 
> > Cc: Matthew Brost <matthew.brost@intel.com>
> > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> > Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_bo.h       | 56
> > ++++++++++++++++++++++++++++++++
> >  drivers/gpu/drm/xe/xe_bo_types.h |  3 ++
> >  2 files changed, 59 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_bo.h
> > b/drivers/gpu/drm/xe/xe_bo.h
> > index 8ab4474129c3..00e93b3065c9 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.h
> > +++ b/drivers/gpu/drm/xe/xe_bo.h
> > @@ -86,6 +86,28 @@
> >  
> >  #define XE_PCI_BARRIER_MMAP_OFFSET	(0x50 << XE_PTE_SHIFT)
> >  
> > +/**
> > + * enum xe_madv_purgeable_state - Buffer object purgeable state
> > enumeration
> > + *
> > + * This enum defines the possible purgeable states for a buffer
> > object,
> > + * allowing userspace to provide memory usage hints to the kernel
> > for
> > + * better memory management under pressure.
> > + *
> > + * @XE_MADV_PURGEABLE_WILLNEED: The buffer object is needed and
> > should not be purged.
> > + * This is the default state.
> > + * @XE_MADV_PURGEABLE_DONTNEED: The buffer object is not currently
> > needed and can be
> > + * purged by the kernel under memory pressure.
> > + * @XE_MADV_PURGEABLE_PURGED: The buffer object has been purged by
> > the kernel.
> > + *
> > + * Accessing a purged buffer will result in an error. Per i915
> > semantics,
> > + * once purged, a BO remains permanently invalid and must be
> > destroyed and recreated.
> > + */
> > +enum xe_madv_purgeable_state {
> > +	XE_MADV_PURGEABLE_WILLNEED,
> > +	XE_MADV_PURGEABLE_DONTNEED,
> > +	XE_MADV_PURGEABLE_PURGED,
> > +};
> > +
> >  struct sg_table;
> >  
> >  struct xe_bo *xe_bo_alloc(void);
> > @@ -214,6 +236,40 @@ static inline bool xe_bo_is_protected(const
> > struct xe_bo *bo)
> >  	return bo->pxp_key_instance;
> >  }
> >  
> > +/**
> > + * xe_bo_is_purged() - Check if buffer object has been purged
> > + * @bo: The buffer object to check
> > + *
> > + * Checks if the buffer object's backing store has been discarded
> > by the
> > + * kernel due to memory pressure after being marked as purgeable
> > (DONTNEED).
> > + * Once purged, the BO cannot be restored and any attempt to use
> > it will fail.
> > + *
> > + * Context: Caller must hold the BO's dma-resv lock
> > + * Return: true if the BO has been purged, false otherwise
> > + */
> > +static inline bool xe_bo_is_purged(struct xe_bo *bo)
> > +{
> > +	xe_bo_assert_held(bo);
> > +	return bo->madv_purgeable == XE_MADV_PURGEABLE_PURGED;
> > +}
> > +
> > +/**
> > + * xe_bo_madv_is_dontneed() - Check if BO is marked as DONTNEED
> > + * @bo: The buffer object to check
> > + *
> > + * Checks if userspace has marked this BO as DONTNEED (i.e., its
> > contents
> > + * are not currently needed and can be discarded under memory
> > pressure).
> > + * This is used internally to decide whether a BO is eligible for
> > purging.
> > + *
> > + * Context: Caller must hold the BO's dma-resv lock
> > + * Return: true if the BO is marked DONTNEED, false otherwise
> > + */
> > +static inline bool xe_bo_madv_is_dontneed(struct xe_bo *bo)
> > +{
> > +	xe_bo_assert_held(bo);
> > +	return bo->madv_purgeable == XE_MADV_PURGEABLE_DONTNEED;
> > +}
> > +
> >  static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo)
> >  {
> >  	if (likely(bo)) {
> > diff --git a/drivers/gpu/drm/xe/xe_bo_types.h
> > b/drivers/gpu/drm/xe/xe_bo_types.h
> > index d4fe3c8dca5b..6acfed0c0bb4 100644
> > --- a/drivers/gpu/drm/xe/xe_bo_types.h
> > +++ b/drivers/gpu/drm/xe/xe_bo_types.h
> > @@ -108,6 +108,9 @@ struct xe_bo {
> >  	 * from default
> >  	 */
> >  	u64 min_align;
> > +
> > +	/** @madv_purgeable: user space advise on BO purgeability
> > */
> 
> , protected by BO's dma-resv lock.
> 
> Everything else LGTM.

Agree, LGTM.

/Thomas


> 
> Matt
> 
> > +	u32 madv_purgeable;
> >  };
> >  
> >  #endif
> > -- 
> > 2.43.0
> > 

[PATCH v4 3/8] drm/xe/madvise: Implement purgeable buffer object support

[Arvind Yadav]

This allows userspace applications to provide memory usage hints to
the kernel for better memory management under pressure:

Add the core implementation for purgeable buffer objects, enabling memory
reclamation of user-designated DONTNEED buffers during eviction.

This patch implements the purge operation and state machine transitions:

Purgeable States (from xe_madv_purgeable_state):
 - WILLNEED (0): BO should be retained, actively used
 - DONTNEED (1): BO eligible for purging, not currently needed
 - PURGED (2): BO backing store reclaimed, permanently invalid

Design Rationale:
  - Async TLB invalidation via trigger_rebind (no blocking xe_vm_invalidate_vma)
  - i915 compatibility: retained field, "once purged always purged" semantics
  - Shared BO protection prevents multi-process memory corruption
  - Scratch PTE reuse avoids new infrastructure, safe for fault mode

v2:
  - Use xe_bo_trigger_rebind() for async TLB invalidation (Thomas Hellström)
  - Add NULL rebind with scratch PTEs for fault mode (Thomas Hellström)
  - Implement i915-compatible retained field logic (Thomas Hellström)
  - Skip BO validation for purged BOs in page fault handler (crash fix)
  - Add scratch VM check in page fault path (non-scratch VMs fail fault)
  - Force clear_pt for non-scratch VMs to avoid phys addr 0 mapping (review fix)
  - Add !is_purged check to resource cursor setup to prevent stale access

v3:
  - Rebase as xe_gt_pagefault.c is gone upstream and replaced
    with xe_pagefault.c (Matthew Brost)
  - Xe specific warn on (Matthew Brost)
  - Call helpers for madv_purgeable access(Matthew Brost)
  - Remove bo NULL check(Matthew Brost)
  - Use xe_bo_assert_held instead of dma assert(Matthew Brost)
  - Move the xe_bo_is_purged check under the dma-resv lock( by Matt)
  - Drop is_purged from xe_pt_stage_bind_entry and just set is_null to true
    for purged BO rename s/is_null/is_null_or_purged (by Matt)
  - UAPI rule should not be changed.(Matthew Brost)
  - Make 'retained' a userptr (Matthew Brost)

v4:
  - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 drivers/gpu/drm/xe/xe_bo.c         | 61 +++++++++++++++++----
 drivers/gpu/drm/xe/xe_pagefault.c  | 12 ++++
 drivers/gpu/drm/xe/xe_pt.c         | 38 +++++++++++--
 drivers/gpu/drm/xe/xe_vm.c         | 11 +++-
 drivers/gpu/drm/xe/xe_vm_madvise.c | 88 ++++++++++++++++++++++++++++++
 5 files changed, 191 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 408c74216fdf..d0a6d340b255 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -836,6 +836,43 @@ static int xe_bo_move_notify(struct xe_bo *bo,
 	return 0;
 }
 
+/**
+ * xe_ttm_bo_purge() - Purge buffer object backing store
+ * @ttm_bo: The TTM buffer object to purge
+ * @ctx: TTM operation context
+ *
+ * This function purges the backing store of a BO marked as DONTNEED and
+ * triggers rebind to invalidate stale GPU mappings. For fault-mode VMs,
+ * this zaps the PTEs. The next GPU access will trigger a page fault and
+ * perform NULL rebind (scratch pages or clear PTEs based on VM config).
+ */
+static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
+{
+	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
+	struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
+
+	if (ttm_bo->ttm) {
+		struct ttm_placement place = {};
+		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
+
+		drm_WARN_ON(&xe->drm, ret);
+		if (!ret) {
+			if (xe_bo_madv_is_dontneed(bo)) {
+				bo->madv_purgeable = XE_MADV_PURGEABLE_PURGED;
+
+				/*
+				 * Trigger rebind to invalidate stale GPU mappings.
+				 * - Non-fault mode: Marks VMAs for rebind
+				 * - Fault mode: Zaps PTEs (sets to 0), next access triggers fault
+				 *   and NULL rebind with scratch/clear PTEs per VM config
+				 */
+				ret = xe_bo_trigger_rebind(xe, bo, ctx);
+				XE_WARN_ON(ret);
+			}
+		}
+	}
+}
+
 static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
 		      struct ttm_operation_ctx *ctx,
 		      struct ttm_resource *new_mem,
@@ -855,6 +892,15 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
 				  ttm && ttm_tt_is_populated(ttm)) ? true : false;
 	int ret = 0;
 
+	/*
+	 * Purge only non-shared BOs explicitly marked DONTNEED by userspace.
+	 * The move_notify callback will handle invalidation asynchronously.
+	 */
+	if (evict && xe_bo_madv_is_dontneed(bo)) {
+		xe_ttm_bo_purge(ttm_bo, ctx);
+		return 0;
+	}
+
 	/* Bo creation path, moving to system or TT. */
 	if ((!old_mem && ttm) && !handle_system_ccs) {
 		if (new_mem->mem_type == XE_PL_TT)
@@ -1604,18 +1650,6 @@ static void xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo)
 	}
 }
 
-static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
-{
-	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
-
-	if (ttm_bo->ttm) {
-		struct ttm_placement place = {};
-		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
-
-		drm_WARN_ON(&xe->drm, ret);
-	}
-}
-
 static void xe_ttm_bo_swap_notify(struct ttm_buffer_object *ttm_bo)
 {
 	struct ttm_operation_ctx ctx = {
@@ -2196,6 +2230,9 @@ struct xe_bo *xe_bo_init_locked(struct xe_device *xe, struct xe_bo *bo,
 #endif
 	INIT_LIST_HEAD(&bo->vram_userfault_link);
 
+	/* Initialize purge advisory state */
+	bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
+
 	drm_gem_private_object_init(&xe->drm, &bo->ttm.base, size);
 
 	if (resv) {
diff --git a/drivers/gpu/drm/xe/xe_pagefault.c b/drivers/gpu/drm/xe/xe_pagefault.c
index 6bee53d6ffc3..e3ace179e9cf 100644
--- a/drivers/gpu/drm/xe/xe_pagefault.c
+++ b/drivers/gpu/drm/xe/xe_pagefault.c
@@ -59,6 +59,18 @@ static int xe_pagefault_begin(struct drm_exec *exec, struct xe_vma *vma,
 	if (!bo)
 		return 0;
 
+	/*
+	 * Check if BO is purged (under dma-resv lock).
+	 * For purged BOs:
+	 * - Scratch VMs: Skip validation, rebind will use scratch PTEs
+	 * - Non-scratch VMs: FAIL the page fault (no scratch page available)
+	 */
+	if (unlikely(xe_bo_is_purged(bo))) {
+		if (!xe_vm_has_scratch(vm))
+			return -EACCES;
+		return 0;
+	}
+
 	return need_vram_move ? xe_bo_migrate(bo, vram->placement, NULL, exec) :
 		xe_bo_validate(bo, vm, true, exec);
 }
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 6703a7049227..c8c66300e25b 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -533,20 +533,26 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
 	/* Is this a leaf entry ?*/
 	if (level == 0 || xe_pt_hugepte_possible(addr, next, level, xe_walk)) {
 		struct xe_res_cursor *curs = xe_walk->curs;
-		bool is_null = xe_vma_is_null(xe_walk->vma);
-		bool is_vram = is_null ? false : xe_res_is_vram(curs);
+		struct xe_bo *bo = xe_vma_bo(xe_walk->vma);
+		bool is_null_or_purged = xe_vma_is_null(xe_walk->vma) ||
+					 (bo && xe_bo_is_purged(bo));
+		bool is_vram = is_null_or_purged ? false : xe_res_is_vram(curs);
 
 		XE_WARN_ON(xe_walk->va_curs_start != addr);
 
 		if (xe_walk->clear_pt) {
 			pte = 0;
 		} else {
-			pte = vm->pt_ops->pte_encode_vma(is_null ? 0 :
+			/*
+			 * For purged BOs, treat like null VMAs - pass address 0.
+			 * The pte_encode_vma will set XE_PTE_NULL flag for scratch mapping.
+			 */
+			pte = vm->pt_ops->pte_encode_vma(is_null_or_purged ? 0 :
 							 xe_res_dma(curs) +
 							 xe_walk->dma_offset,
 							 xe_walk->vma,
 							 pat_index, level);
-			if (!is_null)
+			if (!is_null_or_purged)
 				pte |= is_vram ? xe_walk->default_vram_pte :
 					xe_walk->default_system_pte;
 
@@ -570,7 +576,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
 		if (unlikely(ret))
 			return ret;
 
-		if (!is_null && !xe_walk->clear_pt)
+		if (!is_null_or_purged && !xe_walk->clear_pt)
 			xe_res_next(curs, next - addr);
 		xe_walk->va_curs_start = next;
 		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K << level);
@@ -723,6 +729,26 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
 	};
 	struct xe_pt *pt = vm->pt_root[tile->id];
 	int ret;
+	bool is_purged = false;
+
+	/*
+	 * Check if BO is purged:
+	 * - Scratch VMs: Use scratch PTEs (XE_PTE_NULL) for safe zero reads
+	 * - Non-scratch VMs: Clear PTEs to zero (non-present) to avoid mapping to phys addr 0
+	 *
+	 * For non-scratch VMs, we force clear_pt=true so leaf PTEs become completely
+	 * zero instead of creating a PRESENT mapping to physical address 0.
+	 */
+	if (bo && xe_bo_is_purged(bo)) {
+		is_purged = true;
+
+		/*
+		 * For non-scratch VMs, a NULL rebind should use zero PTEs
+		 * (non-present), not a present PTE to phys 0.
+		 */
+		if (!xe_vm_has_scratch(vm))
+			xe_walk.clear_pt = true;
+	}
 
 	if (range) {
 		/* Move this entire thing to xe_svm.c? */
@@ -762,7 +788,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
 	if (!range)
 		xe_bo_assert_held(bo);
 
-	if (!xe_vma_is_null(vma) && !range) {
+	if (!xe_vma_is_null(vma) && !range && !is_purged) {
 		if (xe_vma_is_userptr(vma))
 			xe_res_first_dma(to_userptr_vma(vma)->userptr.pages.dma_addr, 0,
 					 xe_vma_size(vma), &curs);
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 694f592a0f01..c3a5fe76ff96 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -1359,6 +1359,9 @@ static u64 xelp_pte_encode_bo(struct xe_bo *bo, u64 bo_offset,
 static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
 			       u16 pat_index, u32 pt_level)
 {
+	struct xe_bo *bo = xe_vma_bo(vma);
+	struct xe_vm *vm = xe_vma_vm(vma);
+
 	pte |= XE_PAGE_PRESENT;
 
 	if (likely(!xe_vma_read_only(vma)))
@@ -1367,7 +1370,13 @@ static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
 	pte |= pte_encode_pat_index(pat_index, pt_level);
 	pte |= pte_encode_ps(pt_level);
 
-	if (unlikely(xe_vma_is_null(vma)))
+	/*
+	 * NULL PTEs redirect to scratch page (return zeros on read).
+	 * Set for: 1) explicit null VMAs, 2) purged BOs on scratch VMs.
+	 * Never set NULL flag without scratch page - causes undefined behavior.
+	 */
+	if (unlikely(xe_vma_is_null(vma) ||
+		     (bo && xe_bo_is_purged(bo) && xe_vm_has_scratch(vm))))
 		pte |= XE_PTE_NULL;
 
 	return pte;
diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
index add9a6ca2390..dfeab9e24a09 100644
--- a/drivers/gpu/drm/xe/xe_vm_madvise.c
+++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
@@ -179,6 +179,56 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
 	}
 }
 
+/*
+ * Handle purgeable buffer object advice for DONTNEED/WILLNEED/PURGED.
+ * Returns true if any BO was purged, false otherwise.
+ * Caller must copy retained value to userspace after releasing locks.
+ */
+static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
+				       struct xe_vma **vmas, int num_vmas,
+				       struct drm_xe_madvise *op)
+{
+	bool has_purged_bo = false;
+	int i;
+
+	xe_assert(vm->xe, op->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE);
+
+	for (i = 0; i < num_vmas; i++) {
+		struct xe_bo *bo = xe_vma_bo(vmas[i]);
+
+		if (!bo)
+			continue;
+
+		/* BO must be locked before modifying madv state */
+		xe_bo_assert_held(bo);
+
+		/*
+		 * Once purged, always purged. Cannot transition back to WILLNEED.
+		 * This matches i915 semantics where purged BOs are permanently invalid.
+		 */
+		if (xe_bo_is_purged(bo)) {
+			has_purged_bo = true;
+			continue;
+		}
+
+		switch (op->purge_state_val.val) {
+		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
+			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
+			break;
+		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
+			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
+			break;
+		default:
+			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
+				 op->purge_state_val.val);
+			return false;
+		}
+	}
+
+	/* Return whether any BO was purged; caller will copy to user after unlocking */
+	return has_purged_bo;
+}
+
 typedef void (*madvise_func)(struct xe_device *xe, struct xe_vm *vm,
 			     struct xe_vma **vmas, int num_vmas,
 			     struct drm_xe_madvise *op,
@@ -306,6 +356,16 @@ static bool madvise_args_are_sane(struct xe_device *xe, const struct drm_xe_madv
 			return false;
 		break;
 	}
+	case DRM_XE_VMA_ATTR_PURGEABLE_STATE:
+	{
+		u32 val = args->purge_state_val.val;
+
+		if (XE_IOCTL_DBG(xe, !(val == DRM_XE_VMA_PURGEABLE_STATE_WILLNEED ||
+				       val == DRM_XE_VMA_PURGEABLE_STATE_DONTNEED)))
+			return false;
+
+		break;
+	}
 	default:
 		if (XE_IOCTL_DBG(xe, 1))
 			return false;
@@ -465,6 +525,34 @@ int xe_vm_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *fil
 					goto err_fini;
 			}
 		}
+		if (args->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE) {
+			bool has_purged_bo;
+
+			has_purged_bo = xe_vm_madvise_purgeable_bo(xe, vm, madvise_range.vmas,
+								   madvise_range.num_vmas, args);
+
+			/* Release BO locks */
+			drm_exec_fini(&exec);
+			kfree(madvise_range.vmas);
+			up_write(&vm->lock);
+
+			/*
+			 * Set retained flag to indicate if backing store still exists.
+			 * Matches i915: retained = 1 if not purged, 0 if purged.
+			 * Must copy_to_user AFTER releasing ALL locks to avoid circular dependency.
+			 */
+			if (args->purge_state_val.retained) {
+				u32 retained = !has_purged_bo;
+
+				if (copy_to_user(u64_to_user_ptr(args->purge_state_val.retained),
+						 &retained, sizeof(retained)))
+					drm_warn(&vm->xe->drm, "Failed to copy retained value to user\n");
+			}
+
+			/* Final cleanup for early return */
+			xe_vm_put(vm);
+			return 0;
+		}
 	}
 
 	if (madvise_range.has_svm_userptr_vmas) {
-- 
2.43.0

Re: [PATCH v4 3/8] drm/xe/madvise: Implement purgeable buffer object support

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:49AM +0530, Arvind Yadav wrote:
> This allows userspace applications to provide memory usage hints to
> the kernel for better memory management under pressure:
> 
> Add the core implementation for purgeable buffer objects, enabling memory
> reclamation of user-designated DONTNEED buffers during eviction.
> 
> This patch implements the purge operation and state machine transitions:
> 
> Purgeable States (from xe_madv_purgeable_state):
>  - WILLNEED (0): BO should be retained, actively used
>  - DONTNEED (1): BO eligible for purging, not currently needed
>  - PURGED (2): BO backing store reclaimed, permanently invalid
> 
> Design Rationale:
>   - Async TLB invalidation via trigger_rebind (no blocking xe_vm_invalidate_vma)
>   - i915 compatibility: retained field, "once purged always purged" semantics
>   - Shared BO protection prevents multi-process memory corruption
>   - Scratch PTE reuse avoids new infrastructure, safe for fault mode
> 
> v2:
>   - Use xe_bo_trigger_rebind() for async TLB invalidation (Thomas Hellström)
>   - Add NULL rebind with scratch PTEs for fault mode (Thomas Hellström)
>   - Implement i915-compatible retained field logic (Thomas Hellström)
>   - Skip BO validation for purged BOs in page fault handler (crash fix)
>   - Add scratch VM check in page fault path (non-scratch VMs fail fault)
>   - Force clear_pt for non-scratch VMs to avoid phys addr 0 mapping (review fix)
>   - Add !is_purged check to resource cursor setup to prevent stale access
> 
> v3:
>   - Rebase as xe_gt_pagefault.c is gone upstream and replaced
>     with xe_pagefault.c (Matthew Brost)
>   - Xe specific warn on (Matthew Brost)
>   - Call helpers for madv_purgeable access(Matthew Brost)
>   - Remove bo NULL check(Matthew Brost)
>   - Use xe_bo_assert_held instead of dma assert(Matthew Brost)
>   - Move the xe_bo_is_purged check under the dma-resv lock( by Matt)
>   - Drop is_purged from xe_pt_stage_bind_entry and just set is_null to true
>     for purged BO rename s/is_null/is_null_or_purged (by Matt)
>   - UAPI rule should not be changed.(Matthew Brost)
>   - Make 'retained' a userptr (Matthew Brost)
> 
> v4:
>   - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
> 
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_bo.c         | 61 +++++++++++++++++----
>  drivers/gpu/drm/xe/xe_pagefault.c  | 12 ++++
>  drivers/gpu/drm/xe/xe_pt.c         | 38 +++++++++++--
>  drivers/gpu/drm/xe/xe_vm.c         | 11 +++-
>  drivers/gpu/drm/xe/xe_vm_madvise.c | 88 ++++++++++++++++++++++++++++++
>  5 files changed, 191 insertions(+), 19 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index 408c74216fdf..d0a6d340b255 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -836,6 +836,43 @@ static int xe_bo_move_notify(struct xe_bo *bo,
>  	return 0;
>  }
>  
> +/**
> + * xe_ttm_bo_purge() - Purge buffer object backing store
> + * @ttm_bo: The TTM buffer object to purge
> + * @ctx: TTM operation context
> + *
> + * This function purges the backing store of a BO marked as DONTNEED and
> + * triggers rebind to invalidate stale GPU mappings. For fault-mode VMs,
> + * this zaps the PTEs. The next GPU access will trigger a page fault and
> + * perform NULL rebind (scratch pages or clear PTEs based on VM config).
> + */
> +static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
> +{
> +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> +	struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
> +

xe_bo_assert_held(bo);

> +	if (ttm_bo->ttm) {
> +		struct ttm_placement place = {};
> +		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
> +
> +		drm_WARN_ON(&xe->drm, ret);

I think since 'xe' in available here, you should use xe_assert in place
of drm_WARN_ON.

> +		if (!ret) {
> +			if (xe_bo_madv_is_dontneed(bo)) {
> +				bo->madv_purgeable = XE_MADV_PURGEABLE_PURGED;

Helper to set madv_purgeable state /w lockdep assert?

Also perhaps assert valid state transitions in the helper (e.g., you
cannot tranistion out of XE_MADV_PURGEABLE_PURGED.

> +
> +				/*
> +				 * Trigger rebind to invalidate stale GPU mappings.
> +				 * - Non-fault mode: Marks VMAs for rebind
> +				 * - Fault mode: Zaps PTEs (sets to 0), next access triggers fault
> +				 *   and NULL rebind with scratch/clear PTEs per VM config
> +				 */
> +				ret = xe_bo_trigger_rebind(xe, bo, ctx);
> +				XE_WARN_ON(ret);

I think xe_bo_trigger_rebind is allowed to fail if ctx->no_wait_gpu is
set. In both the faulting fast path and certain parts of the shrinker we
set this. So I think any error returned from xe_bo_trigger_rebind needs
to propagte up the call stack.

> +			}
> +		}
> +	}
> +}
> +
>  static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
>  		      struct ttm_operation_ctx *ctx,
>  		      struct ttm_resource *new_mem,
> @@ -855,6 +892,15 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
>  				  ttm && ttm_tt_is_populated(ttm)) ? true : false;
>  	int ret = 0;
>  
> +	/*
> +	 * Purge only non-shared BOs explicitly marked DONTNEED by userspace.
> +	 * The move_notify callback will handle invalidation asynchronously.
> +	 */
> +	if (evict && xe_bo_madv_is_dontneed(bo)) {
> +		xe_ttm_bo_purge(ttm_bo, ctx);

With above, we need to send errors from xe_ttm_bo_purge up the call
stack.

> +		return 0;
> +	}
> +
>  	/* Bo creation path, moving to system or TT. */
>  	if ((!old_mem && ttm) && !handle_system_ccs) {
>  		if (new_mem->mem_type == XE_PL_TT)
> @@ -1604,18 +1650,6 @@ static void xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo)
>  	}
>  }
>  
> -static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
> -{
> -	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> -
> -	if (ttm_bo->ttm) {
> -		struct ttm_placement place = {};
> -		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
> -
> -		drm_WARN_ON(&xe->drm, ret);
> -	}
> -}
> -
>  static void xe_ttm_bo_swap_notify(struct ttm_buffer_object *ttm_bo)
>  {
>  	struct ttm_operation_ctx ctx = {
> @@ -2196,6 +2230,9 @@ struct xe_bo *xe_bo_init_locked(struct xe_device *xe, struct xe_bo *bo,
>  #endif
>  	INIT_LIST_HEAD(&bo->vram_userfault_link);
>  
> +	/* Initialize purge advisory state */
> +	bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> +
>  	drm_gem_private_object_init(&xe->drm, &bo->ttm.base, size);
>  
>  	if (resv) {
> diff --git a/drivers/gpu/drm/xe/xe_pagefault.c b/drivers/gpu/drm/xe/xe_pagefault.c
> index 6bee53d6ffc3..e3ace179e9cf 100644
> --- a/drivers/gpu/drm/xe/xe_pagefault.c
> +++ b/drivers/gpu/drm/xe/xe_pagefault.c
> @@ -59,6 +59,18 @@ static int xe_pagefault_begin(struct drm_exec *exec, struct xe_vma *vma,
>  	if (!bo)
>  		return 0;
>  
> +	/*
> +	 * Check if BO is purged (under dma-resv lock).
> +	 * For purged BOs:
> +	 * - Scratch VMs: Skip validation, rebind will use scratch PTEs
> +	 * - Non-scratch VMs: FAIL the page fault (no scratch page available)
> +	 */
> +	if (unlikely(xe_bo_is_purged(bo))) {
> +		if (!xe_vm_has_scratch(vm))
> +			return -EACCES;
> +		return 0;
> +	}
> +
>  	return need_vram_move ? xe_bo_migrate(bo, vram->placement, NULL, exec) :
>  		xe_bo_validate(bo, vm, true, exec);
>  }
> diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
> index 6703a7049227..c8c66300e25b 100644
> --- a/drivers/gpu/drm/xe/xe_pt.c
> +++ b/drivers/gpu/drm/xe/xe_pt.c
> @@ -533,20 +533,26 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
>  	/* Is this a leaf entry ?*/
>  	if (level == 0 || xe_pt_hugepte_possible(addr, next, level, xe_walk)) {
>  		struct xe_res_cursor *curs = xe_walk->curs;
> -		bool is_null = xe_vma_is_null(xe_walk->vma);
> -		bool is_vram = is_null ? false : xe_res_is_vram(curs);
> +		struct xe_bo *bo = xe_vma_bo(xe_walk->vma);
> +		bool is_null_or_purged = xe_vma_is_null(xe_walk->vma) ||
> +					 (bo && xe_bo_is_purged(bo));
> +		bool is_vram = is_null_or_purged ? false : xe_res_is_vram(curs);
>  
>  		XE_WARN_ON(xe_walk->va_curs_start != addr);
>  
>  		if (xe_walk->clear_pt) {
>  			pte = 0;
>  		} else {
> -			pte = vm->pt_ops->pte_encode_vma(is_null ? 0 :
> +			/*
> +			 * For purged BOs, treat like null VMAs - pass address 0.
> +			 * The pte_encode_vma will set XE_PTE_NULL flag for scratch mapping.
> +			 */
> +			pte = vm->pt_ops->pte_encode_vma(is_null_or_purged ? 0 :
>  							 xe_res_dma(curs) +
>  							 xe_walk->dma_offset,
>  							 xe_walk->vma,
>  							 pat_index, level);
> -			if (!is_null)
> +			if (!is_null_or_purged)
>  				pte |= is_vram ? xe_walk->default_vram_pte :
>  					xe_walk->default_system_pte;
>  
> @@ -570,7 +576,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
>  		if (unlikely(ret))
>  			return ret;
>  
> -		if (!is_null && !xe_walk->clear_pt)
> +		if (!is_null_or_purged && !xe_walk->clear_pt)
>  			xe_res_next(curs, next - addr);
>  		xe_walk->va_curs_start = next;
>  		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K << level);
> @@ -723,6 +729,26 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
>  	};
>  	struct xe_pt *pt = vm->pt_root[tile->id];
>  	int ret;
> +	bool is_purged = false;
> +
> +	/*
> +	 * Check if BO is purged:
> +	 * - Scratch VMs: Use scratch PTEs (XE_PTE_NULL) for safe zero reads
> +	 * - Non-scratch VMs: Clear PTEs to zero (non-present) to avoid mapping to phys addr 0
> +	 *
> +	 * For non-scratch VMs, we force clear_pt=true so leaf PTEs become completely
> +	 * zero instead of creating a PRESENT mapping to physical address 0.
> +	 */
> +	if (bo && xe_bo_is_purged(bo)) {
> +		is_purged = true;
> +
> +		/*
> +		 * For non-scratch VMs, a NULL rebind should use zero PTEs
> +		 * (non-present), not a present PTE to phys 0.
> +		 */
> +		if (!xe_vm_has_scratch(vm))
> +			xe_walk.clear_pt = true;
> +	}
>  
>  	if (range) {
>  		/* Move this entire thing to xe_svm.c? */
> @@ -762,7 +788,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
>  	if (!range)
>  		xe_bo_assert_held(bo);
>  
> -	if (!xe_vma_is_null(vma) && !range) {
> +	if (!xe_vma_is_null(vma) && !range && !is_purged) {
>  		if (xe_vma_is_userptr(vma))
>  			xe_res_first_dma(to_userptr_vma(vma)->userptr.pages.dma_addr, 0,
>  					 xe_vma_size(vma), &curs);
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index 694f592a0f01..c3a5fe76ff96 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -1359,6 +1359,9 @@ static u64 xelp_pte_encode_bo(struct xe_bo *bo, u64 bo_offset,
>  static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
>  			       u16 pat_index, u32 pt_level)
>  {
> +	struct xe_bo *bo = xe_vma_bo(vma);
> +	struct xe_vm *vm = xe_vma_vm(vma);
> +
>  	pte |= XE_PAGE_PRESENT;
>  
>  	if (likely(!xe_vma_read_only(vma)))
> @@ -1367,7 +1370,13 @@ static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
>  	pte |= pte_encode_pat_index(pat_index, pt_level);
>  	pte |= pte_encode_ps(pt_level);
>  
> -	if (unlikely(xe_vma_is_null(vma)))
> +	/*
> +	 * NULL PTEs redirect to scratch page (return zeros on read).
> +	 * Set for: 1) explicit null VMAs, 2) purged BOs on scratch VMs.
> +	 * Never set NULL flag without scratch page - causes undefined behavior.
> +	 */
> +	if (unlikely(xe_vma_is_null(vma) ||
> +		     (bo && xe_bo_is_purged(bo) && xe_vm_has_scratch(vm))))
>  		pte |= XE_PTE_NULL;
>  
>  	return pte;
> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
> index add9a6ca2390..dfeab9e24a09 100644
> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> @@ -179,6 +179,56 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
>  	}
>  }
>  
> +/*:
> + * Handle purgeable buffer object advice for DONTNEED/WILLNEED/PURGED.
> + * Returns true if any BO was purged, false otherwise.
> + * Caller must copy retained value to userspace after releasing locks.
> + */
> +static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
> +				       struct xe_vma **vmas, int num_vmas,
> +				       struct drm_xe_madvise *op)

Shouldn't this check be a vfunc in madvise_funcs?

Also I think you can hook into xe_madvise_details for the return value /
final copy to user.

> +{
> +	bool has_purged_bo = false;
> +	int i;
> +
> +	xe_assert(vm->xe, op->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE);
> +
> +	for (i = 0; i < num_vmas; i++) {
> +		struct xe_bo *bo = xe_vma_bo(vmas[i]);
> +
> +		if (!bo)
> +			continue;
> +
> +		/* BO must be locked before modifying madv state */
> +		xe_bo_assert_held(bo);
> +
> +		/*
> +		 * Once purged, always purged. Cannot transition back to WILLNEED.
> +		 * This matches i915 semantics where purged BOs are permanently invalid.
> +		 */
> +		if (xe_bo_is_purged(bo)) {
> +			has_purged_bo = true;
> +			continue;
> +		}
> +
> +		switch (op->purge_state_val.val) {
> +		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
> +			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> +			break;
> +		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
> +			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;

Use above suggested helper to set this state?

> +			break;
> +		default:
> +			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
> +				 op->purge_state_val.val);
> +			return false;
> +		}
> +	}
> +
> +	/* Return whether any BO was purged; caller will copy to user after unlocking */
> +	return has_purged_bo;
> +}
> +
>  typedef void (*madvise_func)(struct xe_device *xe, struct xe_vm *vm,
>  			     struct xe_vma **vmas, int num_vmas,
>  			     struct drm_xe_madvise *op,
> @@ -306,6 +356,16 @@ static bool madvise_args_are_sane(struct xe_device *xe, const struct drm_xe_madv
>  			return false;
>  		break;
>  	}
> +	case DRM_XE_VMA_ATTR_PURGEABLE_STATE:
> +	{
> +		u32 val = args->purge_state_val.val;
> +
> +		if (XE_IOCTL_DBG(xe, !(val == DRM_XE_VMA_PURGEABLE_STATE_WILLNEED ||
> +				       val == DRM_XE_VMA_PURGEABLE_STATE_DONTNEED)))
> +			return false;
> +
> +		break;
> +	}
>  	default:
>  		if (XE_IOCTL_DBG(xe, 1))
>  			return false;
> @@ -465,6 +525,34 @@ int xe_vm_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *fil
>  					goto err_fini;
>  			}
>  		}
> +		if (args->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE) {
> +			bool has_purged_bo;
> +
> +			has_purged_bo = xe_vm_madvise_purgeable_bo(xe, vm, madvise_range.vmas,
> +								   madvise_range.num_vmas, args);
> +

Again use the existing vfuncs here.

> +			/* Release BO locks */
> +			drm_exec_fini(&exec);
> +			kfree(madvise_range.vmas);
> +			up_write(&vm->lock);
> +
> +			/*
> +			 * Set retained flag to indicate if backing store still exists.
> +			 * Matches i915: retained = 1 if not purged, 0 if purged.
> +			 * Must copy_to_user AFTER releasing ALL locks to avoid circular dependency.
> +			 */
> +			if (args->purge_state_val.retained) {
> +				u32 retained = !has_purged_bo;
> +
> +				if (copy_to_user(u64_to_user_ptr(args->purge_state_val.retained),
> +						 &retained, sizeof(retained)))

I don't think remained needs to be a u64 - maybe a u16? Will comment on
uAPI too.

> +					drm_warn(&vm->xe->drm, "Failed to copy retained value to user\n");

See above, use xe_madvise_details_fini for the final copy to user.

Matt

> +			}
> +
> +			/* Final cleanup for early return */
> +			xe_vm_put(vm);
> +			return 0;
> +		}
>  	}
>  
>  	if (madvise_range.has_svm_userptr_vmas) {
> -- 
> 2.43.0
> 

Re: [PATCH v4 3/8] drm/xe/madvise: Implement purgeable buffer object support

[Matthew Brost]

On Tue, Jan 20, 2026 at 08:58:05AM -0800, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:49AM +0530, Arvind Yadav wrote:
> > This allows userspace applications to provide memory usage hints to
> > the kernel for better memory management under pressure:
> > 
> > Add the core implementation for purgeable buffer objects, enabling memory
> > reclamation of user-designated DONTNEED buffers during eviction.
> > 
> > This patch implements the purge operation and state machine transitions:
> > 
> > Purgeable States (from xe_madv_purgeable_state):
> >  - WILLNEED (0): BO should be retained, actively used
> >  - DONTNEED (1): BO eligible for purging, not currently needed
> >  - PURGED (2): BO backing store reclaimed, permanently invalid
> > 
> > Design Rationale:
> >   - Async TLB invalidation via trigger_rebind (no blocking xe_vm_invalidate_vma)
> >   - i915 compatibility: retained field, "once purged always purged" semantics
> >   - Shared BO protection prevents multi-process memory corruption
> >   - Scratch PTE reuse avoids new infrastructure, safe for fault mode
> > 
> > v2:
> >   - Use xe_bo_trigger_rebind() for async TLB invalidation (Thomas Hellström)
> >   - Add NULL rebind with scratch PTEs for fault mode (Thomas Hellström)
> >   - Implement i915-compatible retained field logic (Thomas Hellström)
> >   - Skip BO validation for purged BOs in page fault handler (crash fix)
> >   - Add scratch VM check in page fault path (non-scratch VMs fail fault)
> >   - Force clear_pt for non-scratch VMs to avoid phys addr 0 mapping (review fix)
> >   - Add !is_purged check to resource cursor setup to prevent stale access
> > 
> > v3:
> >   - Rebase as xe_gt_pagefault.c is gone upstream and replaced
> >     with xe_pagefault.c (Matthew Brost)
> >   - Xe specific warn on (Matthew Brost)
> >   - Call helpers for madv_purgeable access(Matthew Brost)
> >   - Remove bo NULL check(Matthew Brost)
> >   - Use xe_bo_assert_held instead of dma assert(Matthew Brost)
> >   - Move the xe_bo_is_purged check under the dma-resv lock( by Matt)
> >   - Drop is_purged from xe_pt_stage_bind_entry and just set is_null to true
> >     for purged BO rename s/is_null/is_null_or_purged (by Matt)
> >   - UAPI rule should not be changed.(Matthew Brost)
> >   - Make 'retained' a userptr (Matthew Brost)
> > 
> > v4:
> >   - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
> > 
> > Cc: Matthew Brost <matthew.brost@intel.com>
> > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> > Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_bo.c         | 61 +++++++++++++++++----
> >  drivers/gpu/drm/xe/xe_pagefault.c  | 12 ++++
> >  drivers/gpu/drm/xe/xe_pt.c         | 38 +++++++++++--
> >  drivers/gpu/drm/xe/xe_vm.c         | 11 +++-
> >  drivers/gpu/drm/xe/xe_vm_madvise.c | 88 ++++++++++++++++++++++++++++++
> >  5 files changed, 191 insertions(+), 19 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> > index 408c74216fdf..d0a6d340b255 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > @@ -836,6 +836,43 @@ static int xe_bo_move_notify(struct xe_bo *bo,
> >  	return 0;
> >  }
> >  
> > +/**
> > + * xe_ttm_bo_purge() - Purge buffer object backing store
> > + * @ttm_bo: The TTM buffer object to purge
> > + * @ctx: TTM operation context
> > + *
> > + * This function purges the backing store of a BO marked as DONTNEED and
> > + * triggers rebind to invalidate stale GPU mappings. For fault-mode VMs,
> > + * this zaps the PTEs. The next GPU access will trigger a page fault and
> > + * perform NULL rebind (scratch pages or clear PTEs based on VM config).
> > + */
> > +static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
> > +{
> > +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> > +	struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
> > +
> 
> xe_bo_assert_held(bo);
> 
> > +	if (ttm_bo->ttm) {
> > +		struct ttm_placement place = {};
> > +		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
> > +
> > +		drm_WARN_ON(&xe->drm, ret);
> 
> I think since 'xe' in available here, you should use xe_assert in place
> of drm_WARN_ON.
> 
> > +		if (!ret) {
> > +			if (xe_bo_madv_is_dontneed(bo)) {
> > +				bo->madv_purgeable = XE_MADV_PURGEABLE_PURGED;
> 
> Helper to set madv_purgeable state /w lockdep assert?
> 
> Also perhaps assert valid state transitions in the helper (e.g., you
> cannot tranistion out of XE_MADV_PURGEABLE_PURGED.
> 
> > +
> > +				/*
> > +				 * Trigger rebind to invalidate stale GPU mappings.
> > +				 * - Non-fault mode: Marks VMAs for rebind
> > +				 * - Fault mode: Zaps PTEs (sets to 0), next access triggers fault
> > +				 *   and NULL rebind with scratch/clear PTEs per VM config
> > +				 */
> > +				ret = xe_bo_trigger_rebind(xe, bo, ctx);
> > +				XE_WARN_ON(ret);
> 
> I think xe_bo_trigger_rebind is allowed to fail if ctx->no_wait_gpu is
> set. In both the faulting fast path and certain parts of the shrinker we
> set this. So I think any error returned from xe_bo_trigger_rebind needs
> to propagte up the call stack.
> 
> > +			}
> > +		}
> > +	}
> > +}
> > +
> >  static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
> >  		      struct ttm_operation_ctx *ctx,
> >  		      struct ttm_resource *new_mem,
> > @@ -855,6 +892,15 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
> >  				  ttm && ttm_tt_is_populated(ttm)) ? true : false;
> >  	int ret = 0;
> >  
> > +	/*
> > +	 * Purge only non-shared BOs explicitly marked DONTNEED by userspace.
> > +	 * The move_notify callback will handle invalidation asynchronously.
> > +	 */
> > +	if (evict && xe_bo_madv_is_dontneed(bo)) {
> > +		xe_ttm_bo_purge(ttm_bo, ctx);
> 
> With above, we need to send errors from xe_ttm_bo_purge up the call
> stack.
> 
> > +		return 0;
> > +	}
> > +
> >  	/* Bo creation path, moving to system or TT. */
> >  	if ((!old_mem && ttm) && !handle_system_ccs) {
> >  		if (new_mem->mem_type == XE_PL_TT)
> > @@ -1604,18 +1650,6 @@ static void xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo)
> >  	}
> >  }
> >  
> > -static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
> > -{
> > -	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> > -
> > -	if (ttm_bo->ttm) {
> > -		struct ttm_placement place = {};
> > -		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
> > -
> > -		drm_WARN_ON(&xe->drm, ret);
> > -	}
> > -}
> > -
> >  static void xe_ttm_bo_swap_notify(struct ttm_buffer_object *ttm_bo)
> >  {
> >  	struct ttm_operation_ctx ctx = {
> > @@ -2196,6 +2230,9 @@ struct xe_bo *xe_bo_init_locked(struct xe_device *xe, struct xe_bo *bo,
> >  #endif
> >  	INIT_LIST_HEAD(&bo->vram_userfault_link);
> >  
> > +	/* Initialize purge advisory state */
> > +	bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> > +
> >  	drm_gem_private_object_init(&xe->drm, &bo->ttm.base, size);
> >  
> >  	if (resv) {
> > diff --git a/drivers/gpu/drm/xe/xe_pagefault.c b/drivers/gpu/drm/xe/xe_pagefault.c
> > index 6bee53d6ffc3..e3ace179e9cf 100644
> > --- a/drivers/gpu/drm/xe/xe_pagefault.c
> > +++ b/drivers/gpu/drm/xe/xe_pagefault.c
> > @@ -59,6 +59,18 @@ static int xe_pagefault_begin(struct drm_exec *exec, struct xe_vma *vma,
> >  	if (!bo)
> >  		return 0;
> >  
> > +	/*
> > +	 * Check if BO is purged (under dma-resv lock).
> > +	 * For purged BOs:
> > +	 * - Scratch VMs: Skip validation, rebind will use scratch PTEs
> > +	 * - Non-scratch VMs: FAIL the page fault (no scratch page available)
> > +	 */
> > +	if (unlikely(xe_bo_is_purged(bo))) {
> > +		if (!xe_vm_has_scratch(vm))
> > +			return -EACCES;
> > +		return 0;
> > +	}
> > +
> >  	return need_vram_move ? xe_bo_migrate(bo, vram->placement, NULL, exec) :
> >  		xe_bo_validate(bo, vm, true, exec);
> >  }
> > diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
> > index 6703a7049227..c8c66300e25b 100644
> > --- a/drivers/gpu/drm/xe/xe_pt.c
> > +++ b/drivers/gpu/drm/xe/xe_pt.c
> > @@ -533,20 +533,26 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
> >  	/* Is this a leaf entry ?*/
> >  	if (level == 0 || xe_pt_hugepte_possible(addr, next, level, xe_walk)) {
> >  		struct xe_res_cursor *curs = xe_walk->curs;
> > -		bool is_null = xe_vma_is_null(xe_walk->vma);
> > -		bool is_vram = is_null ? false : xe_res_is_vram(curs);
> > +		struct xe_bo *bo = xe_vma_bo(xe_walk->vma);
> > +		bool is_null_or_purged = xe_vma_is_null(xe_walk->vma) ||
> > +					 (bo && xe_bo_is_purged(bo));
> > +		bool is_vram = is_null_or_purged ? false : xe_res_is_vram(curs);
> >  
> >  		XE_WARN_ON(xe_walk->va_curs_start != addr);
> >  
> >  		if (xe_walk->clear_pt) {
> >  			pte = 0;
> >  		} else {
> > -			pte = vm->pt_ops->pte_encode_vma(is_null ? 0 :
> > +			/*
> > +			 * For purged BOs, treat like null VMAs - pass address 0.
> > +			 * The pte_encode_vma will set XE_PTE_NULL flag for scratch mapping.
> > +			 */
> > +			pte = vm->pt_ops->pte_encode_vma(is_null_or_purged ? 0 :
> >  							 xe_res_dma(curs) +
> >  							 xe_walk->dma_offset,
> >  							 xe_walk->vma,
> >  							 pat_index, level);
> > -			if (!is_null)
> > +			if (!is_null_or_purged)
> >  				pte |= is_vram ? xe_walk->default_vram_pte :
> >  					xe_walk->default_system_pte;
> >  
> > @@ -570,7 +576,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
> >  		if (unlikely(ret))
> >  			return ret;
> >  
> > -		if (!is_null && !xe_walk->clear_pt)
> > +		if (!is_null_or_purged && !xe_walk->clear_pt)
> >  			xe_res_next(curs, next - addr);
> >  		xe_walk->va_curs_start = next;
> >  		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K << level);
> > @@ -723,6 +729,26 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
> >  	};
> >  	struct xe_pt *pt = vm->pt_root[tile->id];
> >  	int ret;
> > +	bool is_purged = false;
> > +
> > +	/*
> > +	 * Check if BO is purged:
> > +	 * - Scratch VMs: Use scratch PTEs (XE_PTE_NULL) for safe zero reads
> > +	 * - Non-scratch VMs: Clear PTEs to zero (non-present) to avoid mapping to phys addr 0
> > +	 *
> > +	 * For non-scratch VMs, we force clear_pt=true so leaf PTEs become completely
> > +	 * zero instead of creating a PRESENT mapping to physical address 0.
> > +	 */
> > +	if (bo && xe_bo_is_purged(bo)) {
> > +		is_purged = true;
> > +
> > +		/*
> > +		 * For non-scratch VMs, a NULL rebind should use zero PTEs
> > +		 * (non-present), not a present PTE to phys 0.
> > +		 */
> > +		if (!xe_vm_has_scratch(vm))
> > +			xe_walk.clear_pt = true;
> > +	}
> >  
> >  	if (range) {
> >  		/* Move this entire thing to xe_svm.c? */
> > @@ -762,7 +788,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
> >  	if (!range)
> >  		xe_bo_assert_held(bo);
> >  
> > -	if (!xe_vma_is_null(vma) && !range) {
> > +	if (!xe_vma_is_null(vma) && !range && !is_purged) {
> >  		if (xe_vma_is_userptr(vma))
> >  			xe_res_first_dma(to_userptr_vma(vma)->userptr.pages.dma_addr, 0,
> >  					 xe_vma_size(vma), &curs);
> > diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> > index 694f592a0f01..c3a5fe76ff96 100644
> > --- a/drivers/gpu/drm/xe/xe_vm.c
> > +++ b/drivers/gpu/drm/xe/xe_vm.c
> > @@ -1359,6 +1359,9 @@ static u64 xelp_pte_encode_bo(struct xe_bo *bo, u64 bo_offset,
> >  static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
> >  			       u16 pat_index, u32 pt_level)
> >  {
> > +	struct xe_bo *bo = xe_vma_bo(vma);
> > +	struct xe_vm *vm = xe_vma_vm(vma);
> > +
> >  	pte |= XE_PAGE_PRESENT;
> >  
> >  	if (likely(!xe_vma_read_only(vma)))
> > @@ -1367,7 +1370,13 @@ static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
> >  	pte |= pte_encode_pat_index(pat_index, pt_level);
> >  	pte |= pte_encode_ps(pt_level);
> >  
> > -	if (unlikely(xe_vma_is_null(vma)))
> > +	/*
> > +	 * NULL PTEs redirect to scratch page (return zeros on read).
> > +	 * Set for: 1) explicit null VMAs, 2) purged BOs on scratch VMs.
> > +	 * Never set NULL flag without scratch page - causes undefined behavior.
> > +	 */
> > +	if (unlikely(xe_vma_is_null(vma) ||
> > +		     (bo && xe_bo_is_purged(bo) && xe_vm_has_scratch(vm))))
> >  		pte |= XE_PTE_NULL;
> >  
> >  	return pte;
> > diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > index add9a6ca2390..dfeab9e24a09 100644
> > --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > @@ -179,6 +179,56 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
> >  	}
> >  }
> >  
> > +/*:
> > + * Handle purgeable buffer object advice for DONTNEED/WILLNEED/PURGED.
> > + * Returns true if any BO was purged, false otherwise.
> > + * Caller must copy retained value to userspace after releasing locks.
> > + */
> > +static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
> > +				       struct xe_vma **vmas, int num_vmas,
> > +				       struct drm_xe_madvise *op)
> 
> Shouldn't this check be a vfunc in madvise_funcs?
> 
> Also I think you can hook into xe_madvise_details for the return value /
> final copy to user.
> 
> > +{
> > +	bool has_purged_bo = false;
> > +	int i;
> > +
> > +	xe_assert(vm->xe, op->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE);
> > +
> > +	for (i = 0; i < num_vmas; i++) {
> > +		struct xe_bo *bo = xe_vma_bo(vmas[i]);
> > +
> > +		if (!bo)
> > +			continue;
> > +
> > +		/* BO must be locked before modifying madv state */
> > +		xe_bo_assert_held(bo);
> > +
> > +		/*
> > +		 * Once purged, always purged. Cannot transition back to WILLNEED.
> > +		 * This matches i915 semantics where purged BOs are permanently invalid.
> > +		 */
> > +		if (xe_bo_is_purged(bo)) {
> > +			has_purged_bo = true;
> > +			continue;
> > +		}
> > +
> > +		switch (op->purge_state_val.val) {
> > +		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
> > +			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> > +			break;
> > +		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
> > +			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
> 
> Use above suggested helper to set this state?
> 
> > +			break;
> > +		default:
> > +			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
> > +				 op->purge_state_val.val);
> > +			return false;
> > +		}
> > +	}
> > +
> > +	/* Return whether any BO was purged; caller will copy to user after unlocking */
> > +	return has_purged_bo;
> > +}
> > +
> >  typedef void (*madvise_func)(struct xe_device *xe, struct xe_vm *vm,
> >  			     struct xe_vma **vmas, int num_vmas,
> >  			     struct drm_xe_madvise *op,
> > @@ -306,6 +356,16 @@ static bool madvise_args_are_sane(struct xe_device *xe, const struct drm_xe_madv
> >  			return false;
> >  		break;
> >  	}
> > +	case DRM_XE_VMA_ATTR_PURGEABLE_STATE:
> > +	{
> > +		u32 val = args->purge_state_val.val;
> > +
> > +		if (XE_IOCTL_DBG(xe, !(val == DRM_XE_VMA_PURGEABLE_STATE_WILLNEED ||
> > +				       val == DRM_XE_VMA_PURGEABLE_STATE_DONTNEED)))
> > +			return false;
> > +
> > +		break;
> > +	}
> >  	default:
> >  		if (XE_IOCTL_DBG(xe, 1))
> >  			return false;
> > @@ -465,6 +525,34 @@ int xe_vm_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *fil
> >  					goto err_fini;
> >  			}
> >  		}
> > +		if (args->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE) {
> > +			bool has_purged_bo;
> > +
> > +			has_purged_bo = xe_vm_madvise_purgeable_bo(xe, vm, madvise_range.vmas,
> > +								   madvise_range.num_vmas, args);
> > +
> 
> Again use the existing vfuncs here.
> 
> > +			/* Release BO locks */
> > +			drm_exec_fini(&exec);
> > +			kfree(madvise_range.vmas);
> > +			up_write(&vm->lock);
> > +
> > +			/*
> > +			 * Set retained flag to indicate if backing store still exists.
> > +			 * Matches i915: retained = 1 if not purged, 0 if purged.
> > +			 * Must copy_to_user AFTER releasing ALL locks to avoid circular dependency.
> > +			 */
> > +			if (args->purge_state_val.retained) {
> > +				u32 retained = !has_purged_bo;
> > +
> > +				if (copy_to_user(u64_to_user_ptr(args->purge_state_val.retained),
> > +						 &retained, sizeof(retained)))
> 
> I don't think remained needs to be a u64 - maybe a u16? Will comment on
> uAPI too.
> 

Ignore this, I forgot purge_state_val.retained is a userptr so u64 is
correct. Let me follow on if we are allowed to change IOCTLs from IOW ->
IOWR. I am really unclear on the rules that part of the uAPI.

> > +					drm_warn(&vm->xe->drm, "Failed to copy retained value to user\n");
> 
> See above, use xe_madvise_details_fini for the final copy to user.
> 
> Matt
> 
> > +			}
> > +
> > +			/* Final cleanup for early return */
> > +			xe_vm_put(vm);
> > +			return 0;
> > +		}
> >  	}
> >  
> >  	if (madvise_range.has_svm_userptr_vmas) {
> > -- 
> > 2.43.0
> > 

Re: [PATCH v4 3/8] drm/xe/madvise: Implement purgeable buffer object support

[Yadav, Arvind]

On 20-01-2026 22:45, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 08:58:05AM -0800, Matthew Brost wrote:
>> On Tue, Jan 20, 2026 at 11:38:49AM +0530, Arvind Yadav wrote:
>>> This allows userspace applications to provide memory usage hints to
>>> the kernel for better memory management under pressure:
>>>
>>> Add the core implementation for purgeable buffer objects, enabling memory
>>> reclamation of user-designated DONTNEED buffers during eviction.
>>>
>>> This patch implements the purge operation and state machine transitions:
>>>
>>> Purgeable States (from xe_madv_purgeable_state):
>>>   - WILLNEED (0): BO should be retained, actively used
>>>   - DONTNEED (1): BO eligible for purging, not currently needed
>>>   - PURGED (2): BO backing store reclaimed, permanently invalid
>>>
>>> Design Rationale:
>>>    - Async TLB invalidation via trigger_rebind (no blocking xe_vm_invalidate_vma)
>>>    - i915 compatibility: retained field, "once purged always purged" semantics
>>>    - Shared BO protection prevents multi-process memory corruption
>>>    - Scratch PTE reuse avoids new infrastructure, safe for fault mode
>>>
>>> v2:
>>>    - Use xe_bo_trigger_rebind() for async TLB invalidation (Thomas Hellström)
>>>    - Add NULL rebind with scratch PTEs for fault mode (Thomas Hellström)
>>>    - Implement i915-compatible retained field logic (Thomas Hellström)
>>>    - Skip BO validation for purged BOs in page fault handler (crash fix)
>>>    - Add scratch VM check in page fault path (non-scratch VMs fail fault)
>>>    - Force clear_pt for non-scratch VMs to avoid phys addr 0 mapping (review fix)
>>>    - Add !is_purged check to resource cursor setup to prevent stale access
>>>
>>> v3:
>>>    - Rebase as xe_gt_pagefault.c is gone upstream and replaced
>>>      with xe_pagefault.c (Matthew Brost)
>>>    - Xe specific warn on (Matthew Brost)
>>>    - Call helpers for madv_purgeable access(Matthew Brost)
>>>    - Remove bo NULL check(Matthew Brost)
>>>    - Use xe_bo_assert_held instead of dma assert(Matthew Brost)
>>>    - Move the xe_bo_is_purged check under the dma-resv lock( by Matt)
>>>    - Drop is_purged from xe_pt_stage_bind_entry and just set is_null to true
>>>      for purged BO rename s/is_null/is_null_or_purged (by Matt)
>>>    - UAPI rule should not be changed.(Matthew Brost)
>>>    - Make 'retained' a userptr (Matthew Brost)
>>>
>>> v4:
>>>    - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
>>>
>>> Cc: Matthew Brost <matthew.brost@intel.com>
>>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>>> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
>>> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
>>> ---
>>>   drivers/gpu/drm/xe/xe_bo.c         | 61 +++++++++++++++++----
>>>   drivers/gpu/drm/xe/xe_pagefault.c  | 12 ++++
>>>   drivers/gpu/drm/xe/xe_pt.c         | 38 +++++++++++--
>>>   drivers/gpu/drm/xe/xe_vm.c         | 11 +++-
>>>   drivers/gpu/drm/xe/xe_vm_madvise.c | 88 ++++++++++++++++++++++++++++++
>>>   5 files changed, 191 insertions(+), 19 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
>>> index 408c74216fdf..d0a6d340b255 100644
>>> --- a/drivers/gpu/drm/xe/xe_bo.c
>>> +++ b/drivers/gpu/drm/xe/xe_bo.c
>>> @@ -836,6 +836,43 @@ static int xe_bo_move_notify(struct xe_bo *bo,
>>>   	return 0;
>>>   }
>>>   
>>> +/**
>>> + * xe_ttm_bo_purge() - Purge buffer object backing store
>>> + * @ttm_bo: The TTM buffer object to purge
>>> + * @ctx: TTM operation context
>>> + *
>>> + * This function purges the backing store of a BO marked as DONTNEED and
>>> + * triggers rebind to invalidate stale GPU mappings. For fault-mode VMs,
>>> + * this zaps the PTEs. The next GPU access will trigger a page fault and
>>> + * perform NULL rebind (scratch pages or clear PTEs based on VM config).
>>> + */
>>> +static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
>>> +{
>>> +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
>>> +	struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
>>> +
>> xe_bo_assert_held(bo);


Noted,  I will add xe_bo_assert_held() at the top of the purge helper

>>
>>> +	if (ttm_bo->ttm) {
>>> +		struct ttm_placement place = {};
>>> +		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
>>> +
>>> +		drm_WARN_ON(&xe->drm, ret);
>> I think since 'xe' in available here, you should use xe_assert in place
>> of drm_WARN_ON.
Agreed. Switched this to xe_assert(xe, !ret) since this is Xe-specific 
and we already have xe.
>>
>>> +		if (!ret) {
>>> +			if (xe_bo_madv_is_dontneed(bo)) {
>>> +				bo->madv_purgeable = XE_MADV_PURGEABLE_PURGED;
>> Helper to set madv_purgeable state /w lockdep assert?
>>
>> Also perhaps assert valid state transitions in the helper (e.g., you
>> cannot tranistion out of XE_MADV_PURGEABLE_PURGED.


Noted,  I will xe_bo_set_purgeable_state() which asserts the BO lock and 
enforces "once PURGED always PURGED" (reject transitions out of PURGED).

>>
>>> +
>>> +				/*
>>> +				 * Trigger rebind to invalidate stale GPU mappings.
>>> +				 * - Non-fault mode: Marks VMAs for rebind
>>> +				 * - Fault mode: Zaps PTEs (sets to 0), next access triggers fault
>>> +				 *   and NULL rebind with scratch/clear PTEs per VM config
>>> +				 */
>>> +				ret = xe_bo_trigger_rebind(xe, bo, ctx);
>>> +				XE_WARN_ON(ret);
>> I think xe_bo_trigger_rebind is allowed to fail if ctx->no_wait_gpu is
>> set. In both the faulting fast path and certain parts of the shrinker we
>> set this. So I think any error returned from xe_bo_trigger_rebind needs
>> to propagte up the call stack.


Agreed. I will changed xe_ttm_bo_purge() to return int and propagate 
errors. xe_bo_move() now returns the error when purge/rebind fails, 
rather than warning and continuing.

>>> +			}
>>> +		}
>>> +	}
>>> +}
>>> +
>>>   static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
>>>   		      struct ttm_operation_ctx *ctx,
>>>   		      struct ttm_resource *new_mem,
>>> @@ -855,6 +892,15 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
>>>   				  ttm && ttm_tt_is_populated(ttm)) ? true : false;
>>>   	int ret = 0;
>>>   
>>> +	/*
>>> +	 * Purge only non-shared BOs explicitly marked DONTNEED by userspace.
>>> +	 * The move_notify callback will handle invalidation asynchronously.
>>> +	 */
>>> +	if (evict && xe_bo_madv_is_dontneed(bo)) {
>>> +		xe_ttm_bo_purge(ttm_bo, ctx);
>> With above, we need to send errors from xe_ttm_bo_purge up the call
>> stack.


Noted.

>>
>>> +		return 0;
>>> +	}
>>> +
>>>   	/* Bo creation path, moving to system or TT. */
>>>   	if ((!old_mem && ttm) && !handle_system_ccs) {
>>>   		if (new_mem->mem_type == XE_PL_TT)
>>> @@ -1604,18 +1650,6 @@ static void xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo)
>>>   	}
>>>   }
>>>   
>>> -static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
>>> -{
>>> -	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
>>> -
>>> -	if (ttm_bo->ttm) {
>>> -		struct ttm_placement place = {};
>>> -		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
>>> -
>>> -		drm_WARN_ON(&xe->drm, ret);
>>> -	}
>>> -}
>>> -
>>>   static void xe_ttm_bo_swap_notify(struct ttm_buffer_object *ttm_bo)
>>>   {
>>>   	struct ttm_operation_ctx ctx = {
>>> @@ -2196,6 +2230,9 @@ struct xe_bo *xe_bo_init_locked(struct xe_device *xe, struct xe_bo *bo,
>>>   #endif
>>>   	INIT_LIST_HEAD(&bo->vram_userfault_link);
>>>   
>>> +	/* Initialize purge advisory state */
>>> +	bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
>>> +
>>>   	drm_gem_private_object_init(&xe->drm, &bo->ttm.base, size);
>>>   
>>>   	if (resv) {
>>> diff --git a/drivers/gpu/drm/xe/xe_pagefault.c b/drivers/gpu/drm/xe/xe_pagefault.c
>>> index 6bee53d6ffc3..e3ace179e9cf 100644
>>> --- a/drivers/gpu/drm/xe/xe_pagefault.c
>>> +++ b/drivers/gpu/drm/xe/xe_pagefault.c
>>> @@ -59,6 +59,18 @@ static int xe_pagefault_begin(struct drm_exec *exec, struct xe_vma *vma,
>>>   	if (!bo)
>>>   		return 0;
>>>   
>>> +	/*
>>> +	 * Check if BO is purged (under dma-resv lock).
>>> +	 * For purged BOs:
>>> +	 * - Scratch VMs: Skip validation, rebind will use scratch PTEs
>>> +	 * - Non-scratch VMs: FAIL the page fault (no scratch page available)
>>> +	 */
>>> +	if (unlikely(xe_bo_is_purged(bo))) {
>>> +		if (!xe_vm_has_scratch(vm))
>>> +			return -EACCES;
>>> +		return 0;
>>> +	}
>>> +
>>>   	return need_vram_move ? xe_bo_migrate(bo, vram->placement, NULL, exec) :
>>>   		xe_bo_validate(bo, vm, true, exec);
>>>   }
>>> diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
>>> index 6703a7049227..c8c66300e25b 100644
>>> --- a/drivers/gpu/drm/xe/xe_pt.c
>>> +++ b/drivers/gpu/drm/xe/xe_pt.c
>>> @@ -533,20 +533,26 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
>>>   	/* Is this a leaf entry ?*/
>>>   	if (level == 0 || xe_pt_hugepte_possible(addr, next, level, xe_walk)) {
>>>   		struct xe_res_cursor *curs = xe_walk->curs;
>>> -		bool is_null = xe_vma_is_null(xe_walk->vma);
>>> -		bool is_vram = is_null ? false : xe_res_is_vram(curs);
>>> +		struct xe_bo *bo = xe_vma_bo(xe_walk->vma);
>>> +		bool is_null_or_purged = xe_vma_is_null(xe_walk->vma) ||
>>> +					 (bo && xe_bo_is_purged(bo));
>>> +		bool is_vram = is_null_or_purged ? false : xe_res_is_vram(curs);
>>>   
>>>   		XE_WARN_ON(xe_walk->va_curs_start != addr);
>>>   
>>>   		if (xe_walk->clear_pt) {
>>>   			pte = 0;
>>>   		} else {
>>> -			pte = vm->pt_ops->pte_encode_vma(is_null ? 0 :
>>> +			/*
>>> +			 * For purged BOs, treat like null VMAs - pass address 0.
>>> +			 * The pte_encode_vma will set XE_PTE_NULL flag for scratch mapping.
>>> +			 */
>>> +			pte = vm->pt_ops->pte_encode_vma(is_null_or_purged ? 0 :
>>>   							 xe_res_dma(curs) +
>>>   							 xe_walk->dma_offset,
>>>   							 xe_walk->vma,
>>>   							 pat_index, level);
>>> -			if (!is_null)
>>> +			if (!is_null_or_purged)
>>>   				pte |= is_vram ? xe_walk->default_vram_pte :
>>>   					xe_walk->default_system_pte;
>>>   
>>> @@ -570,7 +576,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
>>>   		if (unlikely(ret))
>>>   			return ret;
>>>   
>>> -		if (!is_null && !xe_walk->clear_pt)
>>> +		if (!is_null_or_purged && !xe_walk->clear_pt)
>>>   			xe_res_next(curs, next - addr);
>>>   		xe_walk->va_curs_start = next;
>>>   		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K << level);
>>> @@ -723,6 +729,26 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
>>>   	};
>>>   	struct xe_pt *pt = vm->pt_root[tile->id];
>>>   	int ret;
>>> +	bool is_purged = false;
>>> +
>>> +	/*
>>> +	 * Check if BO is purged:
>>> +	 * - Scratch VMs: Use scratch PTEs (XE_PTE_NULL) for safe zero reads
>>> +	 * - Non-scratch VMs: Clear PTEs to zero (non-present) to avoid mapping to phys addr 0
>>> +	 *
>>> +	 * For non-scratch VMs, we force clear_pt=true so leaf PTEs become completely
>>> +	 * zero instead of creating a PRESENT mapping to physical address 0.
>>> +	 */
>>> +	if (bo && xe_bo_is_purged(bo)) {
>>> +		is_purged = true;
>>> +
>>> +		/*
>>> +		 * For non-scratch VMs, a NULL rebind should use zero PTEs
>>> +		 * (non-present), not a present PTE to phys 0.
>>> +		 */
>>> +		if (!xe_vm_has_scratch(vm))
>>> +			xe_walk.clear_pt = true;
>>> +	}
>>>   
>>>   	if (range) {
>>>   		/* Move this entire thing to xe_svm.c? */
>>> @@ -762,7 +788,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
>>>   	if (!range)
>>>   		xe_bo_assert_held(bo);
>>>   
>>> -	if (!xe_vma_is_null(vma) && !range) {
>>> +	if (!xe_vma_is_null(vma) && !range && !is_purged) {
>>>   		if (xe_vma_is_userptr(vma))
>>>   			xe_res_first_dma(to_userptr_vma(vma)->userptr.pages.dma_addr, 0,
>>>   					 xe_vma_size(vma), &curs);
>>> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
>>> index 694f592a0f01..c3a5fe76ff96 100644
>>> --- a/drivers/gpu/drm/xe/xe_vm.c
>>> +++ b/drivers/gpu/drm/xe/xe_vm.c
>>> @@ -1359,6 +1359,9 @@ static u64 xelp_pte_encode_bo(struct xe_bo *bo, u64 bo_offset,
>>>   static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
>>>   			       u16 pat_index, u32 pt_level)
>>>   {
>>> +	struct xe_bo *bo = xe_vma_bo(vma);
>>> +	struct xe_vm *vm = xe_vma_vm(vma);
>>> +
>>>   	pte |= XE_PAGE_PRESENT;
>>>   
>>>   	if (likely(!xe_vma_read_only(vma)))
>>> @@ -1367,7 +1370,13 @@ static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
>>>   	pte |= pte_encode_pat_index(pat_index, pt_level);
>>>   	pte |= pte_encode_ps(pt_level);
>>>   
>>> -	if (unlikely(xe_vma_is_null(vma)))
>>> +	/*
>>> +	 * NULL PTEs redirect to scratch page (return zeros on read).
>>> +	 * Set for: 1) explicit null VMAs, 2) purged BOs on scratch VMs.
>>> +	 * Never set NULL flag without scratch page - causes undefined behavior.
>>> +	 */
>>> +	if (unlikely(xe_vma_is_null(vma) ||
>>> +		     (bo && xe_bo_is_purged(bo) && xe_vm_has_scratch(vm))))
>>>   		pte |= XE_PTE_NULL;
>>>   
>>>   	return pte;
>>> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> index add9a6ca2390..dfeab9e24a09 100644
>>> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> @@ -179,6 +179,56 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
>>>   	}
>>>   }
>>>   
>>> +/*:
>>> + * Handle purgeable buffer object advice for DONTNEED/WILLNEED/PURGED.
>>> + * Returns true if any BO was purged, false otherwise.
>>> + * Caller must copy retained value to userspace after releasing locks.
>>> + */
>>> +static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
>>> +				       struct xe_vma **vmas, int num_vmas,
>>> +				       struct drm_xe_madvise *op)
>> Shouldn't this check be a vfunc in madvise_funcs?
>>
>> Also I think you can hook into xe_madvise_details for the return value /
>> final copy to user.


Yes. I will move purgeable handling into the madvise_funcs[] table as a 
proper madvise vfunc (madvise_purgeable). The retained return is now 
tracked via xe_madvise_details and copied back in 
xe_madvise_details_fini(), so we keep the "copy_to_user after dropping 
locks" rule without a special-case early return.

>>> +{
>>> +	bool has_purged_bo = false;
>>> +	int i;
>>> +
>>> +	xe_assert(vm->xe, op->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE);
>>> +
>>> +	for (i = 0; i < num_vmas; i++) {
>>> +		struct xe_bo *bo = xe_vma_bo(vmas[i]);
>>> +
>>> +		if (!bo)
>>> +			continue;
>>> +
>>> +		/* BO must be locked before modifying madv state */
>>> +		xe_bo_assert_held(bo);
>>> +
>>> +		/*
>>> +		 * Once purged, always purged. Cannot transition back to WILLNEED.
>>> +		 * This matches i915 semantics where purged BOs are permanently invalid.
>>> +		 */
>>> +		if (xe_bo_is_purged(bo)) {
>>> +			has_purged_bo = true;
>>> +			continue;
>>> +		}
>>> +
>>> +		switch (op->purge_state_val.val) {
>>> +		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
>>> +			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
>>> +			break;
>>> +		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
>>> +			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
>> Use above suggested helper to set this state?


Yes, converted to xe_bo_set_purgeable_state().

>>
>>> +			break;
>>> +		default:
>>> +			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
>>> +				 op->purge_state_val.val);
>>> +			return false;
>>> +		}
>>> +	}
>>> +
>>> +	/* Return whether any BO was purged; caller will copy to user after unlocking */
>>> +	return has_purged_bo;
>>> +}
>>> +
>>>   typedef void (*madvise_func)(struct xe_device *xe, struct xe_vm *vm,
>>>   			     struct xe_vma **vmas, int num_vmas,
>>>   			     struct drm_xe_madvise *op,
>>> @@ -306,6 +356,16 @@ static bool madvise_args_are_sane(struct xe_device *xe, const struct drm_xe_madv
>>>   			return false;
>>>   		break;
>>>   	}
>>> +	case DRM_XE_VMA_ATTR_PURGEABLE_STATE:
>>> +	{
>>> +		u32 val = args->purge_state_val.val;
>>> +
>>> +		if (XE_IOCTL_DBG(xe, !(val == DRM_XE_VMA_PURGEABLE_STATE_WILLNEED ||
>>> +				       val == DRM_XE_VMA_PURGEABLE_STATE_DONTNEED)))
>>> +			return false;
>>> +
>>> +		break;
>>> +	}
>>>   	default:
>>>   		if (XE_IOCTL_DBG(xe, 1))
>>>   			return false;
>>> @@ -465,6 +525,34 @@ int xe_vm_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *fil
>>>   					goto err_fini;
>>>   			}
>>>   		}
>>> +		if (args->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE) {
>>> +			bool has_purged_bo;
>>> +
>>> +			has_purged_bo = xe_vm_madvise_purgeable_bo(xe, vm, madvise_range.vmas,
>>> +								   madvise_range.num_vmas, args);
>>> +
>> Again use the existing vfuncs here.


Noted,

>>
>>> +			/* Release BO locks */
>>> +			drm_exec_fini(&exec);
>>> +			kfree(madvise_range.vmas);
>>> +			up_write(&vm->lock);
>>> +
>>> +			/*
>>> +			 * Set retained flag to indicate if backing store still exists.
>>> +			 * Matches i915: retained = 1 if not purged, 0 if purged.
>>> +			 * Must copy_to_user AFTER releasing ALL locks to avoid circular dependency.
>>> +			 */
>>> +			if (args->purge_state_val.retained) {
>>> +				u32 retained = !has_purged_bo;
>>> +
>>> +				if (copy_to_user(u64_to_user_ptr(args->purge_state_val.retained),
>>> +						 &retained, sizeof(retained)))
>> I don't think remained needs to be a u64 - maybe a u16? Will comment on
>> uAPI too.
>>
> Ignore this, I forgot purge_state_val.retained is a userptr so u64 is
> correct. Let me follow on if we are allowed to change IOCTLs from IOW ->
> IOWR. I am really unclear on the rules that part of the uAPI.


sure,

>
>>> +					drm_warn(&vm->xe->drm, "Failed to copy retained value to user\n");
>> See above, use xe_madvise_details_fini for the final copy to user.


Noted -  The retained pointer is stored in details during init, and the 
final copy happens in xe_madvise_details_fini().


Thanks,
Arvind

>>
>> Matt
>>
>>> +			}
>>> +
>>> +			/* Final cleanup for early return */
>>> +			xe_vm_put(vm);
>>> +			return 0;
>>> +		}
>>>   	}
>>>   
>>>   	if (madvise_range.has_svm_userptr_vmas) {
>>> -- 
>>> 2.43.0
>>>

Re: [PATCH v4 3/8] drm/xe/madvise: Implement purgeable buffer object support

[Thomas Hellström]

On Tue, 2026-01-20 at 08:58 -0800, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:49AM +0530, Arvind Yadav wrote:
> > This allows userspace applications to provide memory usage hints to
> > the kernel for better memory management under pressure:
> > 
> > Add the core implementation for purgeable buffer objects, enabling
> > memory
> > reclamation of user-designated DONTNEED buffers during eviction.
> > 
> > This patch implements the purge operation and state machine
> > transitions:
> > 
> > Purgeable States (from xe_madv_purgeable_state):
> >  - WILLNEED (0): BO should be retained, actively used
> >  - DONTNEED (1): BO eligible for purging, not currently needed
> >  - PURGED (2): BO backing store reclaimed, permanently invalid
> > 
> > Design Rationale:
> >   - Async TLB invalidation via trigger_rebind (no blocking
> > xe_vm_invalidate_vma)
> >   - i915 compatibility: retained field, "once purged always purged"
> > semantics
> >   - Shared BO protection prevents multi-process memory corruption
> >   - Scratch PTE reuse avoids new infrastructure, safe for fault
> > mode
> > 
> > v2:
> >   - Use xe_bo_trigger_rebind() for async TLB invalidation (Thomas
> > Hellström)
> >   - Add NULL rebind with scratch PTEs for fault mode (Thomas
> > Hellström)
> >   - Implement i915-compatible retained field logic (Thomas
> > Hellström)
> >   - Skip BO validation for purged BOs in page fault handler (crash
> > fix)
> >   - Add scratch VM check in page fault path (non-scratch VMs fail
> > fault)
> >   - Force clear_pt for non-scratch VMs to avoid phys addr 0 mapping
> > (review fix)
> >   - Add !is_purged check to resource cursor setup to prevent stale
> > access
> > 
> > v3:
> >   - Rebase as xe_gt_pagefault.c is gone upstream and replaced
> >     with xe_pagefault.c (Matthew Brost)
> >   - Xe specific warn on (Matthew Brost)
> >   - Call helpers for madv_purgeable access(Matthew Brost)
> >   - Remove bo NULL check(Matthew Brost)
> >   - Use xe_bo_assert_held instead of dma assert(Matthew Brost)
> >   - Move the xe_bo_is_purged check under the dma-resv lock( by
> > Matt)
> >   - Drop is_purged from xe_pt_stage_bind_entry and just set is_null
> > to true
> >     for purged BO rename s/is_null/is_null_or_purged (by Matt)
> >   - UAPI rule should not be changed.(Matthew Brost)
> >   - Make 'retained' a userptr (Matthew Brost)
> > 
> > v4:
> >   - @madv_purgeable atomic_t → u32 change across all relevant
> > patches. (Matt)
> > 
> > Cc: Matthew Brost <matthew.brost@intel.com>
> > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> > Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_bo.c         | 61 +++++++++++++++++----
> >  drivers/gpu/drm/xe/xe_pagefault.c  | 12 ++++
> >  drivers/gpu/drm/xe/xe_pt.c         | 38 +++++++++++--
> >  drivers/gpu/drm/xe/xe_vm.c         | 11 +++-
> >  drivers/gpu/drm/xe/xe_vm_madvise.c | 88
> > ++++++++++++++++++++++++++++++
> >  5 files changed, 191 insertions(+), 19 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c
> > b/drivers/gpu/drm/xe/xe_bo.c
> > index 408c74216fdf..d0a6d340b255 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > @@ -836,6 +836,43 @@ static int xe_bo_move_notify(struct xe_bo *bo,
> >  	return 0;
> >  }
> >  
> > +/**
> > + * xe_ttm_bo_purge() - Purge buffer object backing store
> > + * @ttm_bo: The TTM buffer object to purge
> > + * @ctx: TTM operation context
> > + *
> > + * This function purges the backing store of a BO marked as
> > DONTNEED and
> > + * triggers rebind to invalidate stale GPU mappings. For fault-
> > mode VMs,
> > + * this zaps the PTEs. The next GPU access will trigger a page
> > fault and
> > + * perform NULL rebind (scratch pages or clear PTEs based on VM
> > config).
> > + */
> > +static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo,
> > struct ttm_operation_ctx *ctx)
> > +{
> > +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> > +	struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
> > +
> 
> xe_bo_assert_held(bo);
> 
> > +	if (ttm_bo->ttm) {
> > +		struct ttm_placement place = {};
> > +		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
> > +
> > +		drm_WARN_ON(&xe->drm, ret);
> 
> I think since 'xe' in available here, you should use xe_assert in
> place
> of drm_WARN_ON.
> 
> > +		if (!ret) {
> > +			if (xe_bo_madv_is_dontneed(bo)) {
> > +				bo->madv_purgeable =
> > XE_MADV_PURGEABLE_PURGED;
> 
> Helper to set madv_purgeable state /w lockdep assert?
> 
> Also perhaps assert valid state transitions in the helper (e.g., you
> cannot tranistion out of XE_MADV_PURGEABLE_PURGED.
> 
> > +
> > +				/*
> > +				 * Trigger rebind to invalidate
> > stale GPU mappings.
> > +				 * - Non-fault mode: Marks VMAs
> > for rebind
> > +				 * - Fault mode: Zaps PTEs (sets
> > to 0), next access triggers fault
> > +				 *   and NULL rebind with
> > scratch/clear PTEs per VM config
> > +				 */
> > +				ret = xe_bo_trigger_rebind(xe, bo,
> > ctx);
> > +				XE_WARN_ON(ret);
> 
> I think xe_bo_trigger_rebind is allowed to fail if ctx->no_wait_gpu
> is
> set. In both the faulting fast path and certain parts of the shrinker
> we
> set this. So I think any error returned from xe_bo_trigger_rebind
> needs
> to propagte up the call stack.

If possible, I think we should call xe_bo_move_notify(), which will in
turn call xe_bo_trigger_rebind() rather than call
xe_bo_trigger_rebind(), since xe_bo_move_notify() is intended to unbind
/ unmap everything needed before a bo move / purge. In this case
xe_bo_trigger_rebind() may be sufficient, but perhaps not in the
future.

> 
> > +			}
> > +		}
> > +	}
> > +}
> > +
> >  static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool
> > evict,
> >  		      struct ttm_operation_ctx *ctx,
> >  		      struct ttm_resource *new_mem,
> > @@ -855,6 +892,15 @@ static int xe_bo_move(struct ttm_buffer_object
> > *ttm_bo, bool evict,
> >  				  ttm && ttm_tt_is_populated(ttm))
> > ? true : false;
> >  	int ret = 0;
> >  
> > +	/*
> > +	 * Purge only non-shared BOs explicitly marked DONTNEED by
> > userspace.
> > +	 * The move_notify callback will handle invalidation
> > asynchronously.
> > +	 */
> > +	if (evict && xe_bo_madv_is_dontneed(bo)) {
> > +		xe_ttm_bo_purge(ttm_bo, ctx);
> 
> With above, we need to send errors from xe_ttm_bo_purge up the call
> stack.
> 
> > +		return 0;
> > +	}
> > +
> >  	/* Bo creation path, moving to system or TT. */
> >  	if ((!old_mem && ttm) && !handle_system_ccs) {
> >  		if (new_mem->mem_type == XE_PL_TT)
> > @@ -1604,18 +1650,6 @@ static void
> > xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo)
> >  	}
> >  }
> >  
> > -static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo,
> > struct ttm_operation_ctx *ctx)
> > -{
> > -	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> > -
> > -	if (ttm_bo->ttm) {
> > -		struct ttm_placement place = {};
> > -		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
> > -
> > -		drm_WARN_ON(&xe->drm, ret);
> > -	}
> > -}
> > -
> >  static void xe_ttm_bo_swap_notify(struct ttm_buffer_object
> > *ttm_bo)
> >  {
> >  	struct ttm_operation_ctx ctx = {
> > @@ -2196,6 +2230,9 @@ struct xe_bo *xe_bo_init_locked(struct
> > xe_device *xe, struct xe_bo *bo,
> >  #endif
> >  	INIT_LIST_HEAD(&bo->vram_userfault_link);
> >  
> > +	/* Initialize purge advisory state */
> > +	bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> > +
> >  	drm_gem_private_object_init(&xe->drm, &bo->ttm.base,
> > size);
> >  
> >  	if (resv) {
> > diff --git a/drivers/gpu/drm/xe/xe_pagefault.c
> > b/drivers/gpu/drm/xe/xe_pagefault.c
> > index 6bee53d6ffc3..e3ace179e9cf 100644
> > --- a/drivers/gpu/drm/xe/xe_pagefault.c
> > +++ b/drivers/gpu/drm/xe/xe_pagefault.c
> > @@ -59,6 +59,18 @@ static int xe_pagefault_begin(struct drm_exec
> > *exec, struct xe_vma *vma,
> >  	if (!bo)
> >  		return 0;
> >  
> > +	/*
> > +	 * Check if BO is purged (under dma-resv lock).
> > +	 * For purged BOs:
> > +	 * - Scratch VMs: Skip validation, rebind will use scratch
> > PTEs
> > +	 * - Non-scratch VMs: FAIL the page fault (no scratch page
> > available)
> > +	 */
> > +	if (unlikely(xe_bo_is_purged(bo))) {
> > +		if (!xe_vm_has_scratch(vm))
> > +			return -EACCES;
> > +		return 0;
> > +	}
> > +
> >  	return need_vram_move ? xe_bo_migrate(bo, vram->placement,
> > NULL, exec) :
> >  		xe_bo_validate(bo, vm, true, exec);
> >  }
> > diff --git a/drivers/gpu/drm/xe/xe_pt.c
> > b/drivers/gpu/drm/xe/xe_pt.c
> > index 6703a7049227..c8c66300e25b 100644
> > --- a/drivers/gpu/drm/xe/xe_pt.c
> > +++ b/drivers/gpu/drm/xe/xe_pt.c
> > @@ -533,20 +533,26 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent,
> > pgoff_t offset,
> >  	/* Is this a leaf entry ?*/
> >  	if (level == 0 || xe_pt_hugepte_possible(addr, next,
> > level, xe_walk)) {
> >  		struct xe_res_cursor *curs = xe_walk->curs;
> > -		bool is_null = xe_vma_is_null(xe_walk->vma);
> > -		bool is_vram = is_null ? false :
> > xe_res_is_vram(curs);
> > +		struct xe_bo *bo = xe_vma_bo(xe_walk->vma);
> > +		bool is_null_or_purged = xe_vma_is_null(xe_walk-
> > >vma) ||
> > +					 (bo &&
> > xe_bo_is_purged(bo));
> > +		bool is_vram = is_null_or_purged ? false :
> > xe_res_is_vram(curs);
> >  
> >  		XE_WARN_ON(xe_walk->va_curs_start != addr);
> >  
> >  		if (xe_walk->clear_pt) {
> >  			pte = 0;
> >  		} else {
> > -			pte = vm->pt_ops->pte_encode_vma(is_null ?
> > 0 :
> > +			/*
> > +			 * For purged BOs, treat like null VMAs -
> > pass address 0.
> > +			 * The pte_encode_vma will set XE_PTE_NULL
> > flag for scratch mapping.
> > +			 */
> > +			pte = vm->pt_ops-
> > >pte_encode_vma(is_null_or_purged ? 0 :
> >  							
> > xe_res_dma(curs) +
> >  							 xe_walk-
> > >dma_offset,
> >  							 xe_walk-
> > >vma,
> >  							
> > pat_index, level);
> > -			if (!is_null)
> > +			if (!is_null_or_purged)
> >  				pte |= is_vram ? xe_walk-
> > >default_vram_pte :
> >  					xe_walk-
> > >default_system_pte;
> >  
> > @@ -570,7 +576,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent,
> > pgoff_t offset,
> >  		if (unlikely(ret))
> >  			return ret;
> >  
> > -		if (!is_null && !xe_walk->clear_pt)
> > +		if (!is_null_or_purged && !xe_walk->clear_pt)
> >  			xe_res_next(curs, next - addr);
> >  		xe_walk->va_curs_start = next;
> >  		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K <<
> > level);
> > @@ -723,6 +729,26 @@ xe_pt_stage_bind(struct xe_tile *tile, struct
> > xe_vma *vma,
> >  	};
> >  	struct xe_pt *pt = vm->pt_root[tile->id];
> >  	int ret;
> > +	bool is_purged = false;
> > +
> > +	/*
> > +	 * Check if BO is purged:
> > +	 * - Scratch VMs: Use scratch PTEs (XE_PTE_NULL) for safe
> > zero reads
> > +	 * - Non-scratch VMs: Clear PTEs to zero (non-present) to
> > avoid mapping to phys addr 0
> > +	 *
> > +	 * For non-scratch VMs, we force clear_pt=true so leaf
> > PTEs become completely
> > +	 * zero instead of creating a PRESENT mapping to physical
> > address 0.
> > +	 */
> > +	if (bo && xe_bo_is_purged(bo)) {
> > +		is_purged = true;
> > +
> > +		/*
> > +		 * For non-scratch VMs, a NULL rebind should use
> > zero PTEs
> > +		 * (non-present), not a present PTE to phys 0.
> > +		 */
> > +		if (!xe_vm_has_scratch(vm))
> > +			xe_walk.clear_pt = true;
> > +	}
> >  
> >  	if (range) {
> >  		/* Move this entire thing to xe_svm.c? */
> > @@ -762,7 +788,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct
> > xe_vma *vma,
> >  	if (!range)
> >  		xe_bo_assert_held(bo);
> >  
> > -	if (!xe_vma_is_null(vma) && !range) {
> > +	if (!xe_vma_is_null(vma) && !range && !is_purged) {
> >  		if (xe_vma_is_userptr(vma))
> >  			xe_res_first_dma(to_userptr_vma(vma)-
> > >userptr.pages.dma_addr, 0,
> >  					 xe_vma_size(vma), &curs);
> > diff --git a/drivers/gpu/drm/xe/xe_vm.c
> > b/drivers/gpu/drm/xe/xe_vm.c
> > index 694f592a0f01..c3a5fe76ff96 100644
> > --- a/drivers/gpu/drm/xe/xe_vm.c
> > +++ b/drivers/gpu/drm/xe/xe_vm.c
> > @@ -1359,6 +1359,9 @@ static u64 xelp_pte_encode_bo(struct xe_bo
> > *bo, u64 bo_offset,
> >  static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
> >  			       u16 pat_index, u32 pt_level)
> >  {
> > +	struct xe_bo *bo = xe_vma_bo(vma);
> > +	struct xe_vm *vm = xe_vma_vm(vma);
> > +
> >  	pte |= XE_PAGE_PRESENT;
> >  
> >  	if (likely(!xe_vma_read_only(vma)))
> > @@ -1367,7 +1370,13 @@ static u64 xelp_pte_encode_vma(u64 pte,
> > struct xe_vma *vma,
> >  	pte |= pte_encode_pat_index(pat_index, pt_level);
> >  	pte |= pte_encode_ps(pt_level);
> >  
> > -	if (unlikely(xe_vma_is_null(vma)))
> > +	/*
> > +	 * NULL PTEs redirect to scratch page (return zeros on
> > read).
> > +	 * Set for: 1) explicit null VMAs, 2) purged BOs on
> > scratch VMs.
> > +	 * Never set NULL flag without scratch page - causes
> > undefined behavior.
> > +	 */
> > +	if (unlikely(xe_vma_is_null(vma) ||
> > +		     (bo && xe_bo_is_purged(bo) &&
> > xe_vm_has_scratch(vm))))
> >  		pte |= XE_PTE_NULL;
> >  
> >  	return pte;
> > diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > index add9a6ca2390..dfeab9e24a09 100644
> > --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > @@ -179,6 +179,56 @@ static void madvise_pat_index(struct xe_device
> > *xe, struct xe_vm *vm,
> >  	}
> >  }
> >  
> > +/*:
> > + * Handle purgeable buffer object advice for
> > DONTNEED/WILLNEED/PURGED.
> > + * Returns true if any BO was purged, false otherwise.
> > + * Caller must copy retained value to userspace after releasing
> > locks.
> > + */
> > +static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe,
> > struct xe_vm *vm,
> > +				       struct xe_vma **vmas, int
> > num_vmas,
> > +				       struct drm_xe_madvise *op)
> 
> Shouldn't this check be a vfunc in madvise_funcs?
> 
> Also I think you can hook into xe_madvise_details for the return
> value /
> final copy to user.
> 
> > +{
> > +	bool has_purged_bo = false;
> > +	int i;
> > +
> > +	xe_assert(vm->xe, op->type ==
> > DRM_XE_VMA_ATTR_PURGEABLE_STATE);
> > +
> > +	for (i = 0; i < num_vmas; i++) {
> > +		struct xe_bo *bo = xe_vma_bo(vmas[i]);
> > +
> > +		if (!bo)
> > +			continue;
> > +
> > +		/* BO must be locked before modifying madv state
> > */
> > +		xe_bo_assert_held(bo);
> > +
> > +		/*
> > +		 * Once purged, always purged. Cannot transition
> > back to WILLNEED.
> > +		 * This matches i915 semantics where purged BOs
> > are permanently invalid.
> > +		 */
> > +		if (xe_bo_is_purged(bo)) {
> > +			has_purged_bo = true;
> > +			continue;
> > +		}
> > +
> > +		switch (op->purge_state_val.val) {
> > +		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
> > +			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_WILLNEED;
> > +			break;
> > +		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
> > +			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_DONTNEED;
> 
> Use above suggested helper to set this state?
> 
> > +			break;
> > +		default:
> > +			drm_warn(&vm->xe->drm, "Invalid madvice
> > value = %d\n",
> > +				 op->purge_state_val.val);
> > +			return false;
> > +		}
> > +	}
> > +
> > +	/* Return whether any BO was purged; caller will copy to
> > user after unlocking */
> > +	return has_purged_bo;
> > +}
> > +
> >  typedef void (*madvise_func)(struct xe_device *xe, struct xe_vm
> > *vm,
> >  			     struct xe_vma **vmas, int num_vmas,
> >  			     struct drm_xe_madvise *op,
> > @@ -306,6 +356,16 @@ static bool madvise_args_are_sane(struct
> > xe_device *xe, const struct drm_xe_madv
> >  			return false;
> >  		break;
> >  	}
> > +	case DRM_XE_VMA_ATTR_PURGEABLE_STATE:
> > +	{
> > +		u32 val = args->purge_state_val.val;
> > +
> > +		if (XE_IOCTL_DBG(xe, !(val ==
> > DRM_XE_VMA_PURGEABLE_STATE_WILLNEED ||
> > +				       val ==
> > DRM_XE_VMA_PURGEABLE_STATE_DONTNEED)))
> > +			return false;
> > +
> > +		break;
> > +	}
> >  	default:
> >  		if (XE_IOCTL_DBG(xe, 1))
> >  			return false;
> > @@ -465,6 +525,34 @@ int xe_vm_madvise_ioctl(struct drm_device
> > *dev, void *data, struct drm_file *fil
> >  					goto err_fini;
> >  			}
> >  		}
> > +		if (args->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE)
> > {
> > +			bool has_purged_bo;
> > +
> > +			has_purged_bo =
> > xe_vm_madvise_purgeable_bo(xe, vm, madvise_range.vmas,
> > +								  
> > madvise_range.num_vmas, args);
> > +
> 
> Again use the existing vfuncs here.
> 
> > +			/* Release BO locks */
> > +			drm_exec_fini(&exec);
> > +			kfree(madvise_range.vmas);
> > +			up_write(&vm->lock);
> > +
> > +			/*
> > +			 * Set retained flag to indicate if
> > backing store still exists.
> > +			 * Matches i915: retained = 1 if not
> > purged, 0 if purged.
> > +			 * Must copy_to_user AFTER releasing ALL
> > locks to avoid circular dependency.
> > +			 */
> > +			if (args->purge_state_val.retained) {
> > +				u32 retained = !has_purged_bo;
> > +
> > +				if
> > (copy_to_user(u64_to_user_ptr(args->purge_state_val.retained),
> > +						 &retained,
> > sizeof(retained)))
> 
> I don't think remained needs to be a u64 - maybe a u16? Will comment
> on
> uAPI too.
> 
> > +					drm_warn(&vm->xe->drm,
> > "Failed to copy retained value to user\n");
> 
> See above, use xe_madvise_details_fini for the final copy to user.

Can we use put_user() rather than copy_to_user() ?

Also, should the IOCTL return a failure in this case?

Another option is ofc to assert that retained is set to false on IOCTL
call, so that if put_user() fails, UMD will not try to reuse a bo whose
retained state is unclear.


Thanks,
Thomas


> 
> Matt
> 
> > +			}
> > +
> > +			/* Final cleanup for early return */
> > +			xe_vm_put(vm);
> > +			return 0;
> > +		}
> >  	}
> >  
> >  	if (madvise_range.has_svm_userptr_vmas) {
> > -- 
> > 2.43.0
> > 

Re: [PATCH v4 3/8] drm/xe/madvise: Implement purgeable buffer object support

[Yadav, Arvind]

On 22-01-2026 21:00, Thomas Hellström wrote:
> On Tue, 2026-01-20 at 08:58 -0800, Matthew Brost wrote:
>> On Tue, Jan 20, 2026 at 11:38:49AM +0530, Arvind Yadav wrote:
>>> This allows userspace applications to provide memory usage hints to
>>> the kernel for better memory management under pressure:
>>>
>>> Add the core implementation for purgeable buffer objects, enabling
>>> memory
>>> reclamation of user-designated DONTNEED buffers during eviction.
>>>
>>> This patch implements the purge operation and state machine
>>> transitions:
>>>
>>> Purgeable States (from xe_madv_purgeable_state):
>>>   - WILLNEED (0): BO should be retained, actively used
>>>   - DONTNEED (1): BO eligible for purging, not currently needed
>>>   - PURGED (2): BO backing store reclaimed, permanently invalid
>>>
>>> Design Rationale:
>>>    - Async TLB invalidation via trigger_rebind (no blocking
>>> xe_vm_invalidate_vma)
>>>    - i915 compatibility: retained field, "once purged always purged"
>>> semantics
>>>    - Shared BO protection prevents multi-process memory corruption
>>>    - Scratch PTE reuse avoids new infrastructure, safe for fault
>>> mode
>>>
>>> v2:
>>>    - Use xe_bo_trigger_rebind() for async TLB invalidation (Thomas
>>> Hellström)
>>>    - Add NULL rebind with scratch PTEs for fault mode (Thomas
>>> Hellström)
>>>    - Implement i915-compatible retained field logic (Thomas
>>> Hellström)
>>>    - Skip BO validation for purged BOs in page fault handler (crash
>>> fix)
>>>    - Add scratch VM check in page fault path (non-scratch VMs fail
>>> fault)
>>>    - Force clear_pt for non-scratch VMs to avoid phys addr 0 mapping
>>> (review fix)
>>>    - Add !is_purged check to resource cursor setup to prevent stale
>>> access
>>>
>>> v3:
>>>    - Rebase as xe_gt_pagefault.c is gone upstream and replaced
>>>      with xe_pagefault.c (Matthew Brost)
>>>    - Xe specific warn on (Matthew Brost)
>>>    - Call helpers for madv_purgeable access(Matthew Brost)
>>>    - Remove bo NULL check(Matthew Brost)
>>>    - Use xe_bo_assert_held instead of dma assert(Matthew Brost)
>>>    - Move the xe_bo_is_purged check under the dma-resv lock( by
>>> Matt)
>>>    - Drop is_purged from xe_pt_stage_bind_entry and just set is_null
>>> to true
>>>      for purged BO rename s/is_null/is_null_or_purged (by Matt)
>>>    - UAPI rule should not be changed.(Matthew Brost)
>>>    - Make 'retained' a userptr (Matthew Brost)
>>>
>>> v4:
>>>    - @madv_purgeable atomic_t → u32 change across all relevant
>>> patches. (Matt)
>>>
>>> Cc: Matthew Brost <matthew.brost@intel.com>
>>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>>> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
>>> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
>>> ---
>>>   drivers/gpu/drm/xe/xe_bo.c         | 61 +++++++++++++++++----
>>>   drivers/gpu/drm/xe/xe_pagefault.c  | 12 ++++
>>>   drivers/gpu/drm/xe/xe_pt.c         | 38 +++++++++++--
>>>   drivers/gpu/drm/xe/xe_vm.c         | 11 +++-
>>>   drivers/gpu/drm/xe/xe_vm_madvise.c | 88
>>> ++++++++++++++++++++++++++++++
>>>   5 files changed, 191 insertions(+), 19 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/xe/xe_bo.c
>>> b/drivers/gpu/drm/xe/xe_bo.c
>>> index 408c74216fdf..d0a6d340b255 100644
>>> --- a/drivers/gpu/drm/xe/xe_bo.c
>>> +++ b/drivers/gpu/drm/xe/xe_bo.c
>>> @@ -836,6 +836,43 @@ static int xe_bo_move_notify(struct xe_bo *bo,
>>>   	return 0;
>>>   }
>>>   
>>> +/**
>>> + * xe_ttm_bo_purge() - Purge buffer object backing store
>>> + * @ttm_bo: The TTM buffer object to purge
>>> + * @ctx: TTM operation context
>>> + *
>>> + * This function purges the backing store of a BO marked as
>>> DONTNEED and
>>> + * triggers rebind to invalidate stale GPU mappings. For fault-
>>> mode VMs,
>>> + * this zaps the PTEs. The next GPU access will trigger a page
>>> fault and
>>> + * perform NULL rebind (scratch pages or clear PTEs based on VM
>>> config).
>>> + */
>>> +static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo,
>>> struct ttm_operation_ctx *ctx)
>>> +{
>>> +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
>>> +	struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
>>> +
>> xe_bo_assert_held(bo);
>>
>>> +	if (ttm_bo->ttm) {
>>> +		struct ttm_placement place = {};
>>> +		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
>>> +
>>> +		drm_WARN_ON(&xe->drm, ret);
>> I think since 'xe' in available here, you should use xe_assert in
>> place
>> of drm_WARN_ON.
>>
>>> +		if (!ret) {
>>> +			if (xe_bo_madv_is_dontneed(bo)) {
>>> +				bo->madv_purgeable =
>>> XE_MADV_PURGEABLE_PURGED;
>> Helper to set madv_purgeable state /w lockdep assert?
>>
>> Also perhaps assert valid state transitions in the helper (e.g., you
>> cannot tranistion out of XE_MADV_PURGEABLE_PURGED.
>>
>>> +
>>> +				/*
>>> +				 * Trigger rebind to invalidate
>>> stale GPU mappings.
>>> +				 * - Non-fault mode: Marks VMAs
>>> for rebind
>>> +				 * - Fault mode: Zaps PTEs (sets
>>> to 0), next access triggers fault
>>> +				 *   and NULL rebind with
>>> scratch/clear PTEs per VM config
>>> +				 */
>>> +				ret = xe_bo_trigger_rebind(xe, bo,
>>> ctx);
>>> +				XE_WARN_ON(ret);
>> I think xe_bo_trigger_rebind is allowed to fail if ctx->no_wait_gpu
>> is
>> set. In both the faulting fast path and certain parts of the shrinker
>> we
>> set this. So I think any error returned from xe_bo_trigger_rebind
>> needs
>> to propagte up the call stack.
> If possible, I think we should call xe_bo_move_notify(), which will in
> turn call xe_bo_trigger_rebind() rather than call
> xe_bo_trigger_rebind(), since xe_bo_move_notify() is intended to unbind
> / unmap everything needed before a bo move / purge. In this case
> xe_bo_trigger_rebind() may be sufficient, but perhaps not in the
> future.


Agree with both points. Will do the change.

>>> +			}
>>> +		}
>>> +	}
>>> +}
>>> +
>>>   static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool
>>> evict,
>>>   		      struct ttm_operation_ctx *ctx,
>>>   		      struct ttm_resource *new_mem,
>>> @@ -855,6 +892,15 @@ static int xe_bo_move(struct ttm_buffer_object
>>> *ttm_bo, bool evict,
>>>   				  ttm && ttm_tt_is_populated(ttm))
>>> ? true : false;
>>>   	int ret = 0;
>>>   
>>> +	/*
>>> +	 * Purge only non-shared BOs explicitly marked DONTNEED by
>>> userspace.
>>> +	 * The move_notify callback will handle invalidation
>>> asynchronously.
>>> +	 */
>>> +	if (evict && xe_bo_madv_is_dontneed(bo)) {
>>> +		xe_ttm_bo_purge(ttm_bo, ctx);
>> With above, we need to send errors from xe_ttm_bo_purge up the call
>> stack.
>>
>>> +		return 0;
>>> +	}
>>> +
>>>   	/* Bo creation path, moving to system or TT. */
>>>   	if ((!old_mem && ttm) && !handle_system_ccs) {
>>>   		if (new_mem->mem_type == XE_PL_TT)
>>> @@ -1604,18 +1650,6 @@ static void
>>> xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo)
>>>   	}
>>>   }
>>>   
>>> -static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo,
>>> struct ttm_operation_ctx *ctx)
>>> -{
>>> -	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
>>> -
>>> -	if (ttm_bo->ttm) {
>>> -		struct ttm_placement place = {};
>>> -		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
>>> -
>>> -		drm_WARN_ON(&xe->drm, ret);
>>> -	}
>>> -}
>>> -
>>>   static void xe_ttm_bo_swap_notify(struct ttm_buffer_object
>>> *ttm_bo)
>>>   {
>>>   	struct ttm_operation_ctx ctx = {
>>> @@ -2196,6 +2230,9 @@ struct xe_bo *xe_bo_init_locked(struct
>>> xe_device *xe, struct xe_bo *bo,
>>>   #endif
>>>   	INIT_LIST_HEAD(&bo->vram_userfault_link);
>>>   
>>> +	/* Initialize purge advisory state */
>>> +	bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
>>> +
>>>   	drm_gem_private_object_init(&xe->drm, &bo->ttm.base,
>>> size);
>>>   
>>>   	if (resv) {
>>> diff --git a/drivers/gpu/drm/xe/xe_pagefault.c
>>> b/drivers/gpu/drm/xe/xe_pagefault.c
>>> index 6bee53d6ffc3..e3ace179e9cf 100644
>>> --- a/drivers/gpu/drm/xe/xe_pagefault.c
>>> +++ b/drivers/gpu/drm/xe/xe_pagefault.c
>>> @@ -59,6 +59,18 @@ static int xe_pagefault_begin(struct drm_exec
>>> *exec, struct xe_vma *vma,
>>>   	if (!bo)
>>>   		return 0;
>>>   
>>> +	/*
>>> +	 * Check if BO is purged (under dma-resv lock).
>>> +	 * For purged BOs:
>>> +	 * - Scratch VMs: Skip validation, rebind will use scratch
>>> PTEs
>>> +	 * - Non-scratch VMs: FAIL the page fault (no scratch page
>>> available)
>>> +	 */
>>> +	if (unlikely(xe_bo_is_purged(bo))) {
>>> +		if (!xe_vm_has_scratch(vm))
>>> +			return -EACCES;
>>> +		return 0;
>>> +	}
>>> +
>>>   	return need_vram_move ? xe_bo_migrate(bo, vram->placement,
>>> NULL, exec) :
>>>   		xe_bo_validate(bo, vm, true, exec);
>>>   }
>>> diff --git a/drivers/gpu/drm/xe/xe_pt.c
>>> b/drivers/gpu/drm/xe/xe_pt.c
>>> index 6703a7049227..c8c66300e25b 100644
>>> --- a/drivers/gpu/drm/xe/xe_pt.c
>>> +++ b/drivers/gpu/drm/xe/xe_pt.c
>>> @@ -533,20 +533,26 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent,
>>> pgoff_t offset,
>>>   	/* Is this a leaf entry ?*/
>>>   	if (level == 0 || xe_pt_hugepte_possible(addr, next,
>>> level, xe_walk)) {
>>>   		struct xe_res_cursor *curs = xe_walk->curs;
>>> -		bool is_null = xe_vma_is_null(xe_walk->vma);
>>> -		bool is_vram = is_null ? false :
>>> xe_res_is_vram(curs);
>>> +		struct xe_bo *bo = xe_vma_bo(xe_walk->vma);
>>> +		bool is_null_or_purged = xe_vma_is_null(xe_walk-
>>>> vma) ||
>>> +					 (bo &&
>>> xe_bo_is_purged(bo));
>>> +		bool is_vram = is_null_or_purged ? false :
>>> xe_res_is_vram(curs);
>>>   
>>>   		XE_WARN_ON(xe_walk->va_curs_start != addr);
>>>   
>>>   		if (xe_walk->clear_pt) {
>>>   			pte = 0;
>>>   		} else {
>>> -			pte = vm->pt_ops->pte_encode_vma(is_null ?
>>> 0 :
>>> +			/*
>>> +			 * For purged BOs, treat like null VMAs -
>>> pass address 0.
>>> +			 * The pte_encode_vma will set XE_PTE_NULL
>>> flag for scratch mapping.
>>> +			 */
>>> +			pte = vm->pt_ops-
>>>> pte_encode_vma(is_null_or_purged ? 0 :
>>>   							
>>> xe_res_dma(curs) +
>>>   							 xe_walk-
>>>> dma_offset,
>>>   							 xe_walk-
>>>> vma,
>>>   							
>>> pat_index, level);
>>> -			if (!is_null)
>>> +			if (!is_null_or_purged)
>>>   				pte |= is_vram ? xe_walk-
>>>> default_vram_pte :
>>>   					xe_walk-
>>>> default_system_pte;
>>>   
>>> @@ -570,7 +576,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent,
>>> pgoff_t offset,
>>>   		if (unlikely(ret))
>>>   			return ret;
>>>   
>>> -		if (!is_null && !xe_walk->clear_pt)
>>> +		if (!is_null_or_purged && !xe_walk->clear_pt)
>>>   			xe_res_next(curs, next - addr);
>>>   		xe_walk->va_curs_start = next;
>>>   		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K <<
>>> level);
>>> @@ -723,6 +729,26 @@ xe_pt_stage_bind(struct xe_tile *tile, struct
>>> xe_vma *vma,
>>>   	};
>>>   	struct xe_pt *pt = vm->pt_root[tile->id];
>>>   	int ret;
>>> +	bool is_purged = false;
>>> +
>>> +	/*
>>> +	 * Check if BO is purged:
>>> +	 * - Scratch VMs: Use scratch PTEs (XE_PTE_NULL) for safe
>>> zero reads
>>> +	 * - Non-scratch VMs: Clear PTEs to zero (non-present) to
>>> avoid mapping to phys addr 0
>>> +	 *
>>> +	 * For non-scratch VMs, we force clear_pt=true so leaf
>>> PTEs become completely
>>> +	 * zero instead of creating a PRESENT mapping to physical
>>> address 0.
>>> +	 */
>>> +	if (bo && xe_bo_is_purged(bo)) {
>>> +		is_purged = true;
>>> +
>>> +		/*
>>> +		 * For non-scratch VMs, a NULL rebind should use
>>> zero PTEs
>>> +		 * (non-present), not a present PTE to phys 0.
>>> +		 */
>>> +		if (!xe_vm_has_scratch(vm))
>>> +			xe_walk.clear_pt = true;
>>> +	}
>>>   
>>>   	if (range) {
>>>   		/* Move this entire thing to xe_svm.c? */
>>> @@ -762,7 +788,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct
>>> xe_vma *vma,
>>>   	if (!range)
>>>   		xe_bo_assert_held(bo);
>>>   
>>> -	if (!xe_vma_is_null(vma) && !range) {
>>> +	if (!xe_vma_is_null(vma) && !range && !is_purged) {
>>>   		if (xe_vma_is_userptr(vma))
>>>   			xe_res_first_dma(to_userptr_vma(vma)-
>>>> userptr.pages.dma_addr, 0,
>>>   					 xe_vma_size(vma), &curs);
>>> diff --git a/drivers/gpu/drm/xe/xe_vm.c
>>> b/drivers/gpu/drm/xe/xe_vm.c
>>> index 694f592a0f01..c3a5fe76ff96 100644
>>> --- a/drivers/gpu/drm/xe/xe_vm.c
>>> +++ b/drivers/gpu/drm/xe/xe_vm.c
>>> @@ -1359,6 +1359,9 @@ static u64 xelp_pte_encode_bo(struct xe_bo
>>> *bo, u64 bo_offset,
>>>   static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
>>>   			       u16 pat_index, u32 pt_level)
>>>   {
>>> +	struct xe_bo *bo = xe_vma_bo(vma);
>>> +	struct xe_vm *vm = xe_vma_vm(vma);
>>> +
>>>   	pte |= XE_PAGE_PRESENT;
>>>   
>>>   	if (likely(!xe_vma_read_only(vma)))
>>> @@ -1367,7 +1370,13 @@ static u64 xelp_pte_encode_vma(u64 pte,
>>> struct xe_vma *vma,
>>>   	pte |= pte_encode_pat_index(pat_index, pt_level);
>>>   	pte |= pte_encode_ps(pt_level);
>>>   
>>> -	if (unlikely(xe_vma_is_null(vma)))
>>> +	/*
>>> +	 * NULL PTEs redirect to scratch page (return zeros on
>>> read).
>>> +	 * Set for: 1) explicit null VMAs, 2) purged BOs on
>>> scratch VMs.
>>> +	 * Never set NULL flag without scratch page - causes
>>> undefined behavior.
>>> +	 */
>>> +	if (unlikely(xe_vma_is_null(vma) ||
>>> +		     (bo && xe_bo_is_purged(bo) &&
>>> xe_vm_has_scratch(vm))))
>>>   		pte |= XE_PTE_NULL;
>>>   
>>>   	return pte;
>>> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> b/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> index add9a6ca2390..dfeab9e24a09 100644
>>> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> @@ -179,6 +179,56 @@ static void madvise_pat_index(struct xe_device
>>> *xe, struct xe_vm *vm,
>>>   	}
>>>   }
>>>   
>>> +/*:
>>> + * Handle purgeable buffer object advice for
>>> DONTNEED/WILLNEED/PURGED.
>>> + * Returns true if any BO was purged, false otherwise.
>>> + * Caller must copy retained value to userspace after releasing
>>> locks.
>>> + */
>>> +static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe,
>>> struct xe_vm *vm,
>>> +				       struct xe_vma **vmas, int
>>> num_vmas,
>>> +				       struct drm_xe_madvise *op)
>> Shouldn't this check be a vfunc in madvise_funcs?
>>
>> Also I think you can hook into xe_madvise_details for the return
>> value /
>> final copy to user.
>>
>>> +{
>>> +	bool has_purged_bo = false;
>>> +	int i;
>>> +
>>> +	xe_assert(vm->xe, op->type ==
>>> DRM_XE_VMA_ATTR_PURGEABLE_STATE);
>>> +
>>> +	for (i = 0; i < num_vmas; i++) {
>>> +		struct xe_bo *bo = xe_vma_bo(vmas[i]);
>>> +
>>> +		if (!bo)
>>> +			continue;
>>> +
>>> +		/* BO must be locked before modifying madv state
>>> */
>>> +		xe_bo_assert_held(bo);
>>> +
>>> +		/*
>>> +		 * Once purged, always purged. Cannot transition
>>> back to WILLNEED.
>>> +		 * This matches i915 semantics where purged BOs
>>> are permanently invalid.
>>> +		 */
>>> +		if (xe_bo_is_purged(bo)) {
>>> +			has_purged_bo = true;
>>> +			continue;
>>> +		}
>>> +
>>> +		switch (op->purge_state_val.val) {
>>> +		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
>>> +			bo->madv_purgeable =
>>> XE_MADV_PURGEABLE_WILLNEED;
>>> +			break;
>>> +		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
>>> +			bo->madv_purgeable =
>>> XE_MADV_PURGEABLE_DONTNEED;
>> Use above suggested helper to set this state?
>>
>>> +			break;
>>> +		default:
>>> +			drm_warn(&vm->xe->drm, "Invalid madvice
>>> value = %d\n",
>>> +				 op->purge_state_val.val);
>>> +			return false;
>>> +		}
>>> +	}
>>> +
>>> +	/* Return whether any BO was purged; caller will copy to
>>> user after unlocking */
>>> +	return has_purged_bo;
>>> +}
>>> +
>>>   typedef void (*madvise_func)(struct xe_device *xe, struct xe_vm
>>> *vm,
>>>   			     struct xe_vma **vmas, int num_vmas,
>>>   			     struct drm_xe_madvise *op,
>>> @@ -306,6 +356,16 @@ static bool madvise_args_are_sane(struct
>>> xe_device *xe, const struct drm_xe_madv
>>>   			return false;
>>>   		break;
>>>   	}
>>> +	case DRM_XE_VMA_ATTR_PURGEABLE_STATE:
>>> +	{
>>> +		u32 val = args->purge_state_val.val;
>>> +
>>> +		if (XE_IOCTL_DBG(xe, !(val ==
>>> DRM_XE_VMA_PURGEABLE_STATE_WILLNEED ||
>>> +				       val ==
>>> DRM_XE_VMA_PURGEABLE_STATE_DONTNEED)))
>>> +			return false;
>>> +
>>> +		break;
>>> +	}
>>>   	default:
>>>   		if (XE_IOCTL_DBG(xe, 1))
>>>   			return false;
>>> @@ -465,6 +525,34 @@ int xe_vm_madvise_ioctl(struct drm_device
>>> *dev, void *data, struct drm_file *fil
>>>   					goto err_fini;
>>>   			}
>>>   		}
>>> +		if (args->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE)
>>> {
>>> +			bool has_purged_bo;
>>> +
>>> +			has_purged_bo =
>>> xe_vm_madvise_purgeable_bo(xe, vm, madvise_range.vmas,
>>> +								
>>> madvise_range.num_vmas, args);
>>> +
>> Again use the existing vfuncs here.
>>
>>> +			/* Release BO locks */
>>> +			drm_exec_fini(&exec);
>>> +			kfree(madvise_range.vmas);
>>> +			up_write(&vm->lock);
>>> +
>>> +			/*
>>> +			 * Set retained flag to indicate if
>>> backing store still exists.
>>> +			 * Matches i915: retained = 1 if not
>>> purged, 0 if purged.
>>> +			 * Must copy_to_user AFTER releasing ALL
>>> locks to avoid circular dependency.
>>> +			 */
>>> +			if (args->purge_state_val.retained) {
>>> +				u32 retained = !has_purged_bo;
>>> +
>>> +				if
>>> (copy_to_user(u64_to_user_ptr(args->purge_state_val.retained),
>>> +						 &retained,
>>> sizeof(retained)))
>> I don't think remained needs to be a u64 - maybe a u16? Will comment
>> on
>> uAPI too.
>>
>>> +					drm_warn(&vm->xe->drm,
>>> "Failed to copy retained value to user\n");
>> See above, use xe_madvise_details_fini for the final copy to user.
> Can we use put_user() rather than copy_to_user() ?


Agreed, Will change to put_user().


>
> Also, should the IOCTL return a failure in this case?

yes. If we can't communicate the retained state to userspace, the IOCTL 
should fail. Will return -EFAULT.

>
> Another option is ofc to assert that retained is set to false on IOCTL
> call, so that if put_user() fails, UMD will not try to reuse a bo whose
> retained state is unclear.


Agreed, I will add a check that retained == 0 on IOCTL entry and reject 
the call otherwise. Combined with returning -EFAULT on put_user() 
failure, this guarantees userspace never observes or relies on an 
uncertain retained state.

Thanks,
~Arvind

>
>
> Thanks,
> Thomas
>
>
>> Matt
>>
>>> +			}
>>> +
>>> +			/* Final cleanup for early return */
>>> +			xe_vm_put(vm);
>>> +			return 0;
>>> +		}
>>>   	}
>>>   
>>>   	if (madvise_range.has_svm_userptr_vmas) {
>>> -- 
>>> 2.43.0
>>>

Re: [PATCH v4 3/8] drm/xe/madvise: Implement purgeable buffer object support

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:49AM +0530, Arvind Yadav wrote:
> This allows userspace applications to provide memory usage hints to
> the kernel for better memory management under pressure:
> 
> Add the core implementation for purgeable buffer objects, enabling memory
> reclamation of user-designated DONTNEED buffers during eviction.
> 
> This patch implements the purge operation and state machine transitions:
> 
> Purgeable States (from xe_madv_purgeable_state):
>  - WILLNEED (0): BO should be retained, actively used
>  - DONTNEED (1): BO eligible for purging, not currently needed
>  - PURGED (2): BO backing store reclaimed, permanently invalid
> 
> Design Rationale:
>   - Async TLB invalidation via trigger_rebind (no blocking xe_vm_invalidate_vma)
>   - i915 compatibility: retained field, "once purged always purged" semantics
>   - Shared BO protection prevents multi-process memory corruption
>   - Scratch PTE reuse avoids new infrastructure, safe for fault mode
> 
> v2:
>   - Use xe_bo_trigger_rebind() for async TLB invalidation (Thomas Hellström)
>   - Add NULL rebind with scratch PTEs for fault mode (Thomas Hellström)
>   - Implement i915-compatible retained field logic (Thomas Hellström)
>   - Skip BO validation for purged BOs in page fault handler (crash fix)
>   - Add scratch VM check in page fault path (non-scratch VMs fail fault)
>   - Force clear_pt for non-scratch VMs to avoid phys addr 0 mapping (review fix)
>   - Add !is_purged check to resource cursor setup to prevent stale access
> 
> v3:
>   - Rebase as xe_gt_pagefault.c is gone upstream and replaced
>     with xe_pagefault.c (Matthew Brost)
>   - Xe specific warn on (Matthew Brost)
>   - Call helpers for madv_purgeable access(Matthew Brost)
>   - Remove bo NULL check(Matthew Brost)
>   - Use xe_bo_assert_held instead of dma assert(Matthew Brost)
>   - Move the xe_bo_is_purged check under the dma-resv lock( by Matt)
>   - Drop is_purged from xe_pt_stage_bind_entry and just set is_null to true
>     for purged BO rename s/is_null/is_null_or_purged (by Matt)
>   - UAPI rule should not be changed.(Matthew Brost)
>   - Make 'retained' a userptr (Matthew Brost)
> 
> v4:
>   - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
> 
> Cc: Matthew Brost <matthew.brost@intel.com>

One last nit here - it is fine you want to implement parts of the IOCTL
eariler in the series to make this eaiser to review but please don't
flip on functionality of the IOCTL until all parts are in place so we
can't biscet the tree and get half of the IOCTLs functionality.

Matt

> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_bo.c         | 61 +++++++++++++++++----
>  drivers/gpu/drm/xe/xe_pagefault.c  | 12 ++++
>  drivers/gpu/drm/xe/xe_pt.c         | 38 +++++++++++--
>  drivers/gpu/drm/xe/xe_vm.c         | 11 +++-
>  drivers/gpu/drm/xe/xe_vm_madvise.c | 88 ++++++++++++++++++++++++++++++
>  5 files changed, 191 insertions(+), 19 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index 408c74216fdf..d0a6d340b255 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -836,6 +836,43 @@ static int xe_bo_move_notify(struct xe_bo *bo,
>  	return 0;
>  }
>  
> +/**
> + * xe_ttm_bo_purge() - Purge buffer object backing store
> + * @ttm_bo: The TTM buffer object to purge
> + * @ctx: TTM operation context
> + *
> + * This function purges the backing store of a BO marked as DONTNEED and
> + * triggers rebind to invalidate stale GPU mappings. For fault-mode VMs,
> + * this zaps the PTEs. The next GPU access will trigger a page fault and
> + * perform NULL rebind (scratch pages or clear PTEs based on VM config).
> + */
> +static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
> +{
> +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> +	struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
> +
> +	if (ttm_bo->ttm) {
> +		struct ttm_placement place = {};
> +		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
> +
> +		drm_WARN_ON(&xe->drm, ret);
> +		if (!ret) {
> +			if (xe_bo_madv_is_dontneed(bo)) {
> +				bo->madv_purgeable = XE_MADV_PURGEABLE_PURGED;
> +
> +				/*
> +				 * Trigger rebind to invalidate stale GPU mappings.
> +				 * - Non-fault mode: Marks VMAs for rebind
> +				 * - Fault mode: Zaps PTEs (sets to 0), next access triggers fault
> +				 *   and NULL rebind with scratch/clear PTEs per VM config
> +				 */
> +				ret = xe_bo_trigger_rebind(xe, bo, ctx);
> +				XE_WARN_ON(ret);
> +			}
> +		}
> +	}
> +}
> +
>  static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
>  		      struct ttm_operation_ctx *ctx,
>  		      struct ttm_resource *new_mem,
> @@ -855,6 +892,15 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
>  				  ttm && ttm_tt_is_populated(ttm)) ? true : false;
>  	int ret = 0;
>  
> +	/*
> +	 * Purge only non-shared BOs explicitly marked DONTNEED by userspace.
> +	 * The move_notify callback will handle invalidation asynchronously.
> +	 */
> +	if (evict && xe_bo_madv_is_dontneed(bo)) {
> +		xe_ttm_bo_purge(ttm_bo, ctx);
> +		return 0;
> +	}
> +
>  	/* Bo creation path, moving to system or TT. */
>  	if ((!old_mem && ttm) && !handle_system_ccs) {
>  		if (new_mem->mem_type == XE_PL_TT)
> @@ -1604,18 +1650,6 @@ static void xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo)
>  	}
>  }
>  
> -static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx)
> -{
> -	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> -
> -	if (ttm_bo->ttm) {
> -		struct ttm_placement place = {};
> -		int ret = ttm_bo_validate(ttm_bo, &place, ctx);
> -
> -		drm_WARN_ON(&xe->drm, ret);
> -	}
> -}
> -
>  static void xe_ttm_bo_swap_notify(struct ttm_buffer_object *ttm_bo)
>  {
>  	struct ttm_operation_ctx ctx = {
> @@ -2196,6 +2230,9 @@ struct xe_bo *xe_bo_init_locked(struct xe_device *xe, struct xe_bo *bo,
>  #endif
>  	INIT_LIST_HEAD(&bo->vram_userfault_link);
>  
> +	/* Initialize purge advisory state */
> +	bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> +
>  	drm_gem_private_object_init(&xe->drm, &bo->ttm.base, size);
>  
>  	if (resv) {
> diff --git a/drivers/gpu/drm/xe/xe_pagefault.c b/drivers/gpu/drm/xe/xe_pagefault.c
> index 6bee53d6ffc3..e3ace179e9cf 100644
> --- a/drivers/gpu/drm/xe/xe_pagefault.c
> +++ b/drivers/gpu/drm/xe/xe_pagefault.c
> @@ -59,6 +59,18 @@ static int xe_pagefault_begin(struct drm_exec *exec, struct xe_vma *vma,
>  	if (!bo)
>  		return 0;
>  
> +	/*
> +	 * Check if BO is purged (under dma-resv lock).
> +	 * For purged BOs:
> +	 * - Scratch VMs: Skip validation, rebind will use scratch PTEs
> +	 * - Non-scratch VMs: FAIL the page fault (no scratch page available)
> +	 */
> +	if (unlikely(xe_bo_is_purged(bo))) {
> +		if (!xe_vm_has_scratch(vm))
> +			return -EACCES;
> +		return 0;
> +	}
> +
>  	return need_vram_move ? xe_bo_migrate(bo, vram->placement, NULL, exec) :
>  		xe_bo_validate(bo, vm, true, exec);
>  }
> diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
> index 6703a7049227..c8c66300e25b 100644
> --- a/drivers/gpu/drm/xe/xe_pt.c
> +++ b/drivers/gpu/drm/xe/xe_pt.c
> @@ -533,20 +533,26 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
>  	/* Is this a leaf entry ?*/
>  	if (level == 0 || xe_pt_hugepte_possible(addr, next, level, xe_walk)) {
>  		struct xe_res_cursor *curs = xe_walk->curs;
> -		bool is_null = xe_vma_is_null(xe_walk->vma);
> -		bool is_vram = is_null ? false : xe_res_is_vram(curs);
> +		struct xe_bo *bo = xe_vma_bo(xe_walk->vma);
> +		bool is_null_or_purged = xe_vma_is_null(xe_walk->vma) ||
> +					 (bo && xe_bo_is_purged(bo));
> +		bool is_vram = is_null_or_purged ? false : xe_res_is_vram(curs);
>  
>  		XE_WARN_ON(xe_walk->va_curs_start != addr);
>  
>  		if (xe_walk->clear_pt) {
>  			pte = 0;
>  		} else {
> -			pte = vm->pt_ops->pte_encode_vma(is_null ? 0 :
> +			/*
> +			 * For purged BOs, treat like null VMAs - pass address 0.
> +			 * The pte_encode_vma will set XE_PTE_NULL flag for scratch mapping.
> +			 */
> +			pte = vm->pt_ops->pte_encode_vma(is_null_or_purged ? 0 :
>  							 xe_res_dma(curs) +
>  							 xe_walk->dma_offset,
>  							 xe_walk->vma,
>  							 pat_index, level);
> -			if (!is_null)
> +			if (!is_null_or_purged)
>  				pte |= is_vram ? xe_walk->default_vram_pte :
>  					xe_walk->default_system_pte;
>  
> @@ -570,7 +576,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
>  		if (unlikely(ret))
>  			return ret;
>  
> -		if (!is_null && !xe_walk->clear_pt)
> +		if (!is_null_or_purged && !xe_walk->clear_pt)
>  			xe_res_next(curs, next - addr);
>  		xe_walk->va_curs_start = next;
>  		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K << level);
> @@ -723,6 +729,26 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
>  	};
>  	struct xe_pt *pt = vm->pt_root[tile->id];
>  	int ret;
> +	bool is_purged = false;
> +
> +	/*
> +	 * Check if BO is purged:
> +	 * - Scratch VMs: Use scratch PTEs (XE_PTE_NULL) for safe zero reads
> +	 * - Non-scratch VMs: Clear PTEs to zero (non-present) to avoid mapping to phys addr 0
> +	 *
> +	 * For non-scratch VMs, we force clear_pt=true so leaf PTEs become completely
> +	 * zero instead of creating a PRESENT mapping to physical address 0.
> +	 */
> +	if (bo && xe_bo_is_purged(bo)) {
> +		is_purged = true;
> +
> +		/*
> +		 * For non-scratch VMs, a NULL rebind should use zero PTEs
> +		 * (non-present), not a present PTE to phys 0.
> +		 */
> +		if (!xe_vm_has_scratch(vm))
> +			xe_walk.clear_pt = true;
> +	}
>  
>  	if (range) {
>  		/* Move this entire thing to xe_svm.c? */
> @@ -762,7 +788,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
>  	if (!range)
>  		xe_bo_assert_held(bo);
>  
> -	if (!xe_vma_is_null(vma) && !range) {
> +	if (!xe_vma_is_null(vma) && !range && !is_purged) {
>  		if (xe_vma_is_userptr(vma))
>  			xe_res_first_dma(to_userptr_vma(vma)->userptr.pages.dma_addr, 0,
>  					 xe_vma_size(vma), &curs);
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index 694f592a0f01..c3a5fe76ff96 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -1359,6 +1359,9 @@ static u64 xelp_pte_encode_bo(struct xe_bo *bo, u64 bo_offset,
>  static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
>  			       u16 pat_index, u32 pt_level)
>  {
> +	struct xe_bo *bo = xe_vma_bo(vma);
> +	struct xe_vm *vm = xe_vma_vm(vma);
> +
>  	pte |= XE_PAGE_PRESENT;
>  
>  	if (likely(!xe_vma_read_only(vma)))
> @@ -1367,7 +1370,13 @@ static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
>  	pte |= pte_encode_pat_index(pat_index, pt_level);
>  	pte |= pte_encode_ps(pt_level);
>  
> -	if (unlikely(xe_vma_is_null(vma)))
> +	/*
> +	 * NULL PTEs redirect to scratch page (return zeros on read).
> +	 * Set for: 1) explicit null VMAs, 2) purged BOs on scratch VMs.
> +	 * Never set NULL flag without scratch page - causes undefined behavior.
> +	 */
> +	if (unlikely(xe_vma_is_null(vma) ||
> +		     (bo && xe_bo_is_purged(bo) && xe_vm_has_scratch(vm))))
>  		pte |= XE_PTE_NULL;
>  
>  	return pte;
> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
> index add9a6ca2390..dfeab9e24a09 100644
> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> @@ -179,6 +179,56 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
>  	}
>  }
>  
> +/*
> + * Handle purgeable buffer object advice for DONTNEED/WILLNEED/PURGED.
> + * Returns true if any BO was purged, false otherwise.
> + * Caller must copy retained value to userspace after releasing locks.
> + */
> +static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
> +				       struct xe_vma **vmas, int num_vmas,
> +				       struct drm_xe_madvise *op)
> +{
> +	bool has_purged_bo = false;
> +	int i;
> +
> +	xe_assert(vm->xe, op->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE);
> +
> +	for (i = 0; i < num_vmas; i++) {
> +		struct xe_bo *bo = xe_vma_bo(vmas[i]);
> +
> +		if (!bo)
> +			continue;
> +
> +		/* BO must be locked before modifying madv state */
> +		xe_bo_assert_held(bo);
> +
> +		/*
> +		 * Once purged, always purged. Cannot transition back to WILLNEED.
> +		 * This matches i915 semantics where purged BOs are permanently invalid.
> +		 */
> +		if (xe_bo_is_purged(bo)) {
> +			has_purged_bo = true;
> +			continue;
> +		}
> +
> +		switch (op->purge_state_val.val) {
> +		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
> +			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> +			break;
> +		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
> +			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
> +			break;
> +		default:
> +			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
> +				 op->purge_state_val.val);
> +			return false;
> +		}
> +	}
> +
> +	/* Return whether any BO was purged; caller will copy to user after unlocking */
> +	return has_purged_bo;
> +}
> +
>  typedef void (*madvise_func)(struct xe_device *xe, struct xe_vm *vm,
>  			     struct xe_vma **vmas, int num_vmas,
>  			     struct drm_xe_madvise *op,
> @@ -306,6 +356,16 @@ static bool madvise_args_are_sane(struct xe_device *xe, const struct drm_xe_madv
>  			return false;
>  		break;
>  	}
> +	case DRM_XE_VMA_ATTR_PURGEABLE_STATE:
> +	{
> +		u32 val = args->purge_state_val.val;
> +
> +		if (XE_IOCTL_DBG(xe, !(val == DRM_XE_VMA_PURGEABLE_STATE_WILLNEED ||
> +				       val == DRM_XE_VMA_PURGEABLE_STATE_DONTNEED)))
> +			return false;
> +
> +		break;
> +	}
>  	default:
>  		if (XE_IOCTL_DBG(xe, 1))
>  			return false;
> @@ -465,6 +525,34 @@ int xe_vm_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *fil
>  					goto err_fini;
>  			}
>  		}
> +		if (args->type == DRM_XE_VMA_ATTR_PURGEABLE_STATE) {
> +			bool has_purged_bo;
> +
> +			has_purged_bo = xe_vm_madvise_purgeable_bo(xe, vm, madvise_range.vmas,
> +								   madvise_range.num_vmas, args);
> +
> +			/* Release BO locks */
> +			drm_exec_fini(&exec);
> +			kfree(madvise_range.vmas);
> +			up_write(&vm->lock);
> +
> +			/*
> +			 * Set retained flag to indicate if backing store still exists.
> +			 * Matches i915: retained = 1 if not purged, 0 if purged.
> +			 * Must copy_to_user AFTER releasing ALL locks to avoid circular dependency.
> +			 */
> +			if (args->purge_state_val.retained) {
> +				u32 retained = !has_purged_bo;
> +
> +				if (copy_to_user(u64_to_user_ptr(args->purge_state_val.retained),
> +						 &retained, sizeof(retained)))
> +					drm_warn(&vm->xe->drm, "Failed to copy retained value to user\n");
> +			}
> +
> +			/* Final cleanup for early return */
> +			xe_vm_put(vm);
> +			return 0;
> +		}
>  	}
>  
>  	if (madvise_range.has_svm_userptr_vmas) {
> -- 
> 2.43.0
> 

[PATCH v4 4/8] drm/xe/bo: Handle CPU faults on purged buffer objects

[Arvind Yadav]

Return error when CPU attempts to access a purged buffer object.
Purged BOs have their backing store reclaimed by the kernel, making
CPU access invalid. The fault handler returns SIGBUS to userspace,
matching i915 semantics.

The purged check is added to both CPU fault paths:
- Fastpath (xe_bo_cpu_fault_fastpath): Returns error immediately
  under dma-resv lock, preventing attempts to migrate/validate freed pages
- Slowpath (xe_bo_cpu_fault): Returns -EFAULT under drm_exec lock,
  converted to VM_FAULT_SIGBUS

Without the fastpath check, accessing existing mmap mappings of purged BOs
would trigger xe_bo_fault_migrate() on freed backing store, causing kernel
hangs or crashes.

v2:
  - Added xe_bo_is_purged(bo) instead of atomic_read.
  - Avoids leaks and keeps drm_dev_exit() while returning.

v3:
  - Move xe_bo_is_purged check under a dma-resv lock (Matthew Brost)

v4:
  - Add purged check to fastpath (xe_bo_cpu_fault_fastpath) to prevent
    hang when accessing existing mmap of purged BO

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 drivers/gpu/drm/xe/xe_bo.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index d0a6d340b255..cc547915161d 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -1934,6 +1934,12 @@ static vm_fault_t xe_bo_cpu_fault_fastpath(struct vm_fault *vmf, struct xe_devic
 	if (!dma_resv_trylock(tbo->base.resv))
 		goto out_validation;
 
+	/* Purged BOs have no backing store - fault to userspace */
+	if (xe_bo_is_purged(bo)) {
+		ret = VM_FAULT_SIGBUS;
+		goto out_unlock;
+	}
+
 	if (xe_ttm_bo_is_imported(tbo)) {
 		ret = VM_FAULT_SIGBUS;
 		drm_dbg(&xe->drm, "CPU trying to access an imported buffer object.\n");
@@ -2024,6 +2030,12 @@ static vm_fault_t xe_bo_cpu_fault(struct vm_fault *vmf)
 		if (err)
 			break;
 
+		/* Purged BOs have no backing store - fault to userspace */
+		if (xe_bo_is_purged(bo)) {
+			err = -EFAULT;
+			break;
+		}
+
 		if (xe_ttm_bo_is_imported(tbo)) {
 			err = -EFAULT;
 			drm_dbg(&xe->drm, "CPU trying to access an imported buffer object.\n");
-- 
2.43.0

Re: [PATCH v4 4/8] drm/xe/bo: Handle CPU faults on purged buffer objects

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:50AM +0530, Arvind Yadav wrote:
> Return error when CPU attempts to access a purged buffer object.
> Purged BOs have their backing store reclaimed by the kernel, making
> CPU access invalid. The fault handler returns SIGBUS to userspace,
> matching i915 semantics.
> 
> The purged check is added to both CPU fault paths:
> - Fastpath (xe_bo_cpu_fault_fastpath): Returns error immediately
>   under dma-resv lock, preventing attempts to migrate/validate freed pages
> - Slowpath (xe_bo_cpu_fault): Returns -EFAULT under drm_exec lock,
>   converted to VM_FAULT_SIGBUS
> 
> Without the fastpath check, accessing existing mmap mappings of purged BOs
> would trigger xe_bo_fault_migrate() on freed backing store, causing kernel
> hangs or crashes.
> 
> v2:
>   - Added xe_bo_is_purged(bo) instead of atomic_read.
>   - Avoids leaks and keeps drm_dev_exit() while returning.
> 
> v3:
>   - Move xe_bo_is_purged check under a dma-resv lock (Matthew Brost)
> 
> v4:
>   - Add purged check to fastpath (xe_bo_cpu_fault_fastpath) to prevent
>     hang when accessing existing mmap of purged BO
> 
> Cc: Matthew Brost <matthew.brost@intel.com>

Reviewed-by: Matthew Brost <matthew.brost@intel.com>

> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_bo.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index d0a6d340b255..cc547915161d 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -1934,6 +1934,12 @@ static vm_fault_t xe_bo_cpu_fault_fastpath(struct vm_fault *vmf, struct xe_devic
>  	if (!dma_resv_trylock(tbo->base.resv))
>  		goto out_validation;
>  
> +	/* Purged BOs have no backing store - fault to userspace */
> +	if (xe_bo_is_purged(bo)) {
> +		ret = VM_FAULT_SIGBUS;
> +		goto out_unlock;
> +	}
> +
>  	if (xe_ttm_bo_is_imported(tbo)) {
>  		ret = VM_FAULT_SIGBUS;
>  		drm_dbg(&xe->drm, "CPU trying to access an imported buffer object.\n");
> @@ -2024,6 +2030,12 @@ static vm_fault_t xe_bo_cpu_fault(struct vm_fault *vmf)
>  		if (err)
>  			break;
>  
> +		/* Purged BOs have no backing store - fault to userspace */
> +		if (xe_bo_is_purged(bo)) {
> +			err = -EFAULT;
> +			break;
> +		}
> +
>  		if (xe_ttm_bo_is_imported(tbo)) {
>  			err = -EFAULT;
>  			drm_dbg(&xe->drm, "CPU trying to access an imported buffer object.\n");
> -- 
> 2.43.0
> 

Re: [PATCH v4 4/8] drm/xe/bo: Handle CPU faults on purged buffer objects

[Thomas Hellström]

On Tue, 2026-01-20 at 11:38 +0530, Arvind Yadav wrote:
> Return error when CPU attempts to access a purged buffer object.
> Purged BOs have their backing store reclaimed by the kernel, making
> CPU access invalid. The fault handler returns SIGBUS to userspace,
> matching i915 semantics.
> 
> The purged check is added to both CPU fault paths:
> - Fastpath (xe_bo_cpu_fault_fastpath): Returns error immediately
>   under dma-resv lock, preventing attempts to migrate/validate freed
> pages
> - Slowpath (xe_bo_cpu_fault): Returns -EFAULT under drm_exec lock,
>   converted to VM_FAULT_SIGBUS
> 
> Without the fastpath check, accessing existing mmap mappings of
> purged BOs
> would trigger xe_bo_fault_migrate() on freed backing store, causing
> kernel
> hangs or crashes.
> 
> v2:
>   - Added xe_bo_is_purged(bo) instead of atomic_read.
>   - Avoids leaks and keeps drm_dev_exit() while returning.
> 
> v3:
>   - Move xe_bo_is_purged check under a dma-resv lock (Matthew Brost)
> 
> v4:
>   - Add purged check to fastpath (xe_bo_cpu_fault_fastpath) to
> prevent
>     hang when accessing existing mmap of purged BO
> 
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>

Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

> ---
>  drivers/gpu/drm/xe/xe_bo.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index d0a6d340b255..cc547915161d 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -1934,6 +1934,12 @@ static vm_fault_t
> xe_bo_cpu_fault_fastpath(struct vm_fault *vmf, struct xe_devic
>  	if (!dma_resv_trylock(tbo->base.resv))
>  		goto out_validation;
>  
> +	/* Purged BOs have no backing store - fault to userspace */
> +	if (xe_bo_is_purged(bo)) {
> +		ret = VM_FAULT_SIGBUS;
> +		goto out_unlock;
> +	}
> +
>  	if (xe_ttm_bo_is_imported(tbo)) {
>  		ret = VM_FAULT_SIGBUS;
>  		drm_dbg(&xe->drm, "CPU trying to access an imported
> buffer object.\n");
> @@ -2024,6 +2030,12 @@ static vm_fault_t xe_bo_cpu_fault(struct
> vm_fault *vmf)
>  		if (err)
>  			break;
>  
> +		/* Purged BOs have no backing store - fault to
> userspace */
> +		if (xe_bo_is_purged(bo)) {
> +			err = -EFAULT;
> +			break;
> +		}
> +
>  		if (xe_ttm_bo_is_imported(tbo)) {
>  			err = -EFAULT;
>  			drm_dbg(&xe->drm, "CPU trying to access an
> imported buffer object.\n");

[PATCH v4 5/8] drm/xe/vm: Prevent binding of purged buffer objects

[Arvind Yadav]

Add check_purged parameter to vma_lock_and_validate() to block
new mapping operations on purged BOs while allowing cleanup
operations to proceed.

Purged BOs have their backing pages freed by the kernel. New
mapping operations (MAP, PREFETCH, REMAP) must be rejected with
-EINVAL to prevent GPU access to invalid memory. Cleanup
operations (UNMAP) must be allowed so applications can release
resources after detecting purge via the retained field.

REMAP operations require mixed handling - reject new prev/next
VMAs if the BO is purged, but allow the unmap portion to proceed
for cleanup.

The check_purged parameter distinguishes between these cases:
true for new mappings (must reject), false for cleanup (allow).

v2:
  - Clarify that purged BOs are permanently invalid (i915 semantics)
  - Remove incorrect claim about madvise(WILLNEED) restoring purged BOs

v3:
  - Move xe_bo_is_purged check under vma_lock_and_validate (Matthew Brost)
  - Add check_purged parameter to distinguish new mappings from cleanup
  - Allow UNMAP operations to prevent resource leaks
  - Handle REMAP operation's dual nature (cleanup + new mappings)

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 drivers/gpu/drm/xe/xe_vm.c | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index c3a5fe76ff96..f250daae3012 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -2883,7 +2883,7 @@ static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm,
 }
 
 static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
-				 bool res_evict, bool validate)
+				 bool res_evict, bool validate, bool check_purged)
 {
 	struct xe_bo *bo = xe_vma_bo(vma);
 	struct xe_vm *vm = xe_vma_vm(vma);
@@ -2892,6 +2892,11 @@ static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
 	if (bo) {
 		if (!bo->vm)
 			err = drm_exec_lock_obj(exec, &bo->ttm.base);
+
+		/* Reject new mappings to purged BOs; allow cleanup operations */
+		if (!err && check_purged && xe_bo_is_purged(bo))
+			err = -EINVAL;
+
 		if (!err && validate)
 			err = xe_bo_validate(bo, vm,
 					     !xe_vm_in_preempt_fence_mode(vm) &&
@@ -2990,7 +2995,8 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
 			err = vma_lock_and_validate(exec, op->map.vma,
 						    res_evict,
 						    !xe_vm_in_fault_mode(vm) ||
-						    op->map.immediate);
+						    op->map.immediate,
+						    true);
 		break;
 	case DRM_GPUVA_OP_REMAP:
 		err = check_ufence(gpuva_to_vma(op->base.remap.unmap->va));
@@ -2999,13 +3005,13 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
 
 		err = vma_lock_and_validate(exec,
 					    gpuva_to_vma(op->base.remap.unmap->va),
-					    res_evict, false);
+					    res_evict, false, false);
 		if (!err && op->remap.prev)
 			err = vma_lock_and_validate(exec, op->remap.prev,
-						    res_evict, true);
+						    res_evict, true, true);
 		if (!err && op->remap.next)
 			err = vma_lock_and_validate(exec, op->remap.next,
-						    res_evict, true);
+						    res_evict, true, true);
 		break;
 	case DRM_GPUVA_OP_UNMAP:
 		err = check_ufence(gpuva_to_vma(op->base.unmap.va));
@@ -3014,7 +3020,7 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
 
 		err = vma_lock_and_validate(exec,
 					    gpuva_to_vma(op->base.unmap.va),
-					    res_evict, false);
+					    res_evict, false, false);
 		break;
 	case DRM_GPUVA_OP_PREFETCH:
 	{
@@ -3029,7 +3035,7 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
 
 		err = vma_lock_and_validate(exec,
 					    gpuva_to_vma(op->base.prefetch.va),
-					    res_evict, false);
+					    res_evict, false, true);
 		if (!err && !xe_vma_has_no_bo(vma))
 			err = xe_bo_migrate(xe_vma_bo(vma),
 					    region_to_mem_type[region],
-- 
2.43.0

Re: [PATCH v4 5/8] drm/xe/vm: Prevent binding of purged buffer objects

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:51AM +0530, Arvind Yadav wrote:
> Add check_purged parameter to vma_lock_and_validate() to block
> new mapping operations on purged BOs while allowing cleanup
> operations to proceed.
> 
> Purged BOs have their backing pages freed by the kernel. New
> mapping operations (MAP, PREFETCH, REMAP) must be rejected with
> -EINVAL to prevent GPU access to invalid memory. Cleanup
> operations (UNMAP) must be allowed so applications can release
> resources after detecting purge via the retained field.
> 
> REMAP operations require mixed handling - reject new prev/next
> VMAs if the BO is purged, but allow the unmap portion to proceed
> for cleanup.
> 
> The check_purged parameter distinguishes between these cases:
> true for new mappings (must reject), false for cleanup (allow).
> 
> v2:
>   - Clarify that purged BOs are permanently invalid (i915 semantics)
>   - Remove incorrect claim about madvise(WILLNEED) restoring purged BOs
> 
> v3:
>   - Move xe_bo_is_purged check under vma_lock_and_validate (Matthew Brost)
>   - Add check_purged parameter to distinguish new mappings from cleanup
>   - Allow UNMAP operations to prevent resource leaks
>   - Handle REMAP operation's dual nature (cleanup + new mappings)
> 
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_vm.c | 20 +++++++++++++-------
>  1 file changed, 13 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index c3a5fe76ff96..f250daae3012 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -2883,7 +2883,7 @@ static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm,
>  }
> 
>  static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
> -				 bool res_evict, bool validate)
> +				 bool res_evict, bool validate, bool check_purged)

It probably time to add something like this to avoid transposing arguments.

struct lock_and_validate_flags {
	bool res_evict;
	bool validate;
	bool check_purged;
};

Logic in the patch looks correct though.

Matt

>  {
>  	struct xe_bo *bo = xe_vma_bo(vma);
>  	struct xe_vm *vm = xe_vma_vm(vma);
> @@ -2892,6 +2892,11 @@ static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
>  	if (bo) {
>  		if (!bo->vm)
>  			err = drm_exec_lock_obj(exec, &bo->ttm.base);
> +
> +		/* Reject new mappings to purged BOs; allow cleanup operations */
> +		if (!err && check_purged && xe_bo_is_purged(bo))
> +			err = -EINVAL;
> +
>  		if (!err && validate)
>  			err = xe_bo_validate(bo, vm,
>  					     !xe_vm_in_preempt_fence_mode(vm) &&
> @@ -2990,7 +2995,8 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
>  			err = vma_lock_and_validate(exec, op->map.vma,
>  						    res_evict,
>  						    !xe_vm_in_fault_mode(vm) ||
> -						    op->map.immediate);
> +						    op->map.immediate,
> +						    true);
>  		break;
>  	case DRM_GPUVA_OP_REMAP:
>  		err = check_ufence(gpuva_to_vma(op->base.remap.unmap->va));
> @@ -2999,13 +3005,13 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
>  
>  		err = vma_lock_and_validate(exec,
>  					    gpuva_to_vma(op->base.remap.unmap->va),
> -					    res_evict, false);
> +					    res_evict, false, false);
>  		if (!err && op->remap.prev)
>  			err = vma_lock_and_validate(exec, op->remap.prev,
> -						    res_evict, true);
> +						    res_evict, true, true);
>  		if (!err && op->remap.next)
>  			err = vma_lock_and_validate(exec, op->remap.next,
> -						    res_evict, true);
> +						    res_evict, true, true);
>  		break;
>  	case DRM_GPUVA_OP_UNMAP:
>  		err = check_ufence(gpuva_to_vma(op->base.unmap.va));
> @@ -3014,7 +3020,7 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
>  
>  		err = vma_lock_and_validate(exec,
>  					    gpuva_to_vma(op->base.unmap.va),
> -					    res_evict, false);
> +					    res_evict, false, false);
>  		break;
>  	case DRM_GPUVA_OP_PREFETCH:
>  	{
> @@ -3029,7 +3035,7 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
>  
>  		err = vma_lock_and_validate(exec,
>  					    gpuva_to_vma(op->base.prefetch.va),
> -					    res_evict, false);
> +					    res_evict, false, true);
>  		if (!err && !xe_vma_has_no_bo(vma))
>  			err = xe_bo_migrate(xe_vma_bo(vma),
>  					    region_to_mem_type[region],
> -- 
> 2.43.0
> 

Re: [PATCH v4 5/8] drm/xe/vm: Prevent binding of purged buffer objects

[Yadav, Arvind]

[-- Attachment #1: Type: text/plain, Size: 4829 bytes --]


On 20-01-2026 22:57, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:51AM +0530, Arvind Yadav wrote:
>> Add check_purged parameter to vma_lock_and_validate() to block
>> new mapping operations on purged BOs while allowing cleanup
>> operations to proceed.
>>
>> Purged BOs have their backing pages freed by the kernel. New
>> mapping operations (MAP, PREFETCH, REMAP) must be rejected with
>> -EINVAL to prevent GPU access to invalid memory. Cleanup
>> operations (UNMAP) must be allowed so applications can release
>> resources after detecting purge via the retained field.
>>
>> REMAP operations require mixed handling - reject new prev/next
>> VMAs if the BO is purged, but allow the unmap portion to proceed
>> for cleanup.
>>
>> The check_purged parameter distinguishes between these cases:
>> true for new mappings (must reject), false for cleanup (allow).
>>
>> v2:
>>    - Clarify that purged BOs are permanently invalid (i915 semantics)
>>    - Remove incorrect claim about madvise(WILLNEED) restoring purged BOs
>>
>> v3:
>>    - Move xe_bo_is_purged check under vma_lock_and_validate (Matthew Brost)
>>    - Add check_purged parameter to distinguish new mappings from cleanup
>>    - Allow UNMAP operations to prevent resource leaks
>>    - Handle REMAP operation's dual nature (cleanup + new mappings)
>>
>> Cc: Matthew Brost<matthew.brost@intel.com>
>> Cc: Thomas Hellström<thomas.hellstrom@linux.intel.com>
>> Cc: Himal Prasad Ghimiray<himal.prasad.ghimiray@intel.com>
>> Signed-off-by: Arvind Yadav<arvind.yadav@intel.com>
>> ---
>>   drivers/gpu/drm/xe/xe_vm.c | 20 +++++++++++++-------
>>   1 file changed, 13 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
>> index c3a5fe76ff96..f250daae3012 100644
>> --- a/drivers/gpu/drm/xe/xe_vm.c
>> +++ b/drivers/gpu/drm/xe/xe_vm.c
>> @@ -2883,7 +2883,7 @@ static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm,
>>   }
>>
>>   static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
>> -				 bool res_evict, bool validate)
>> +				 bool res_evict, bool validate, bool check_purged)
> It probably time to add something like this to avoid transposing arguments.
>
> struct lock_and_validate_flags {
> 	bool res_evict;
> 	bool validate;
> 	bool check_purged;
> };
>
> Logic in the patch looks correct though.


Noted, I will add "struct xe_lock_and_validate_flags" Thanks, Arvind

>
> Matt
>
>>   {
>>   	struct xe_bo *bo = xe_vma_bo(vma);
>>   	struct xe_vm *vm = xe_vma_vm(vma);
>> @@ -2892,6 +2892,11 @@ static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
>>   	if (bo) {
>>   		if (!bo->vm)
>>   			err = drm_exec_lock_obj(exec, &bo->ttm.base);
>> +
>> +		/* Reject new mappings to purged BOs; allow cleanup operations */
>> +		if (!err && check_purged && xe_bo_is_purged(bo))
>> +			err = -EINVAL;
>> +
>>   		if (!err && validate)
>>   			err = xe_bo_validate(bo, vm,
>>   					     !xe_vm_in_preempt_fence_mode(vm) &&
>> @@ -2990,7 +2995,8 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
>>   			err = vma_lock_and_validate(exec, op->map.vma,
>>   						    res_evict,
>>   						    !xe_vm_in_fault_mode(vm) ||
>> -						    op->map.immediate);
>> +						    op->map.immediate,
>> +						    true);
>>   		break;
>>   	case DRM_GPUVA_OP_REMAP:
>>   		err = check_ufence(gpuva_to_vma(op->base.remap.unmap->va));
>> @@ -2999,13 +3005,13 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
>>   
>>   		err = vma_lock_and_validate(exec,
>>   					    gpuva_to_vma(op->base.remap.unmap->va),
>> -					    res_evict, false);
>> +					    res_evict, false, false);
>>   		if (!err && op->remap.prev)
>>   			err = vma_lock_and_validate(exec, op->remap.prev,
>> -						    res_evict, true);
>> +						    res_evict, true, true);
>>   		if (!err && op->remap.next)
>>   			err = vma_lock_and_validate(exec, op->remap.next,
>> -						    res_evict, true);
>> +						    res_evict, true, true);
>>   		break;
>>   	case DRM_GPUVA_OP_UNMAP:
>>   		err = check_ufence(gpuva_to_vma(op->base.unmap.va));
>> @@ -3014,7 +3020,7 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
>>   
>>   		err = vma_lock_and_validate(exec,
>>   					    gpuva_to_vma(op->base.unmap.va),
>> -					    res_evict, false);
>> +					    res_evict, false, false);
>>   		break;
>>   	case DRM_GPUVA_OP_PREFETCH:
>>   	{
>> @@ -3029,7 +3035,7 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm,
>>   
>>   		err = vma_lock_and_validate(exec,
>>   					    gpuva_to_vma(op->base.prefetch.va),
>> -					    res_evict, false);
>> +					    res_evict, false, true);
>>   		if (!err && !xe_vma_has_no_bo(vma))
>>   			err = xe_bo_migrate(xe_vma_bo(vma),
>>   					    region_to_mem_type[region],
>> -- 
>> 2.43.0
>>

[-- Attachment #2: Type: text/html, Size: 5739 bytes --]

Re: [PATCH v4 5/8] drm/xe/vm: Prevent binding of purged buffer objects

[Thomas Hellström]

On Fri, 2026-01-23 at 11:11 +0530, Yadav, Arvind wrote:
> 
> On 20-01-2026 22:57, Matthew Brost wrote:
> > On Tue, Jan 20, 2026 at 11:38:51AM +0530, Arvind Yadav wrote:
> > > Add check_purged parameter to vma_lock_and_validate() to block
> > > new mapping operations on purged BOs while allowing cleanup
> > > operations to proceed.
> > > 
> > > Purged BOs have their backing pages freed by the kernel. New
> > > mapping operations (MAP, PREFETCH, REMAP) must be rejected with
> > > -EINVAL to prevent GPU access to invalid memory. Cleanup
> > > operations (UNMAP) must be allowed so applications can release
> > > resources after detecting purge via the retained field.
> > > 
> > > REMAP operations require mixed handling - reject new prev/next
> > > VMAs if the BO is purged, but allow the unmap portion to proceed
> > > for cleanup.
> > > 
> > > The check_purged parameter distinguishes between these cases:
> > > true for new mappings (must reject), false for cleanup (allow).
> > > 
> > > v2:
> > >    - Clarify that purged BOs are permanently invalid (i915
> > > semantics)
> > >    - Remove incorrect claim about madvise(WILLNEED) restoring
> > > purged BOs
> > > 
> > > v3:
> > >    - Move xe_bo_is_purged check under vma_lock_and_validate
> > > (Matthew Brost)
> > >    - Add check_purged parameter to distinguish new mappings from
> > > cleanup
> > >    - Allow UNMAP operations to prevent resource leaks
> > >    - Handle REMAP operation's dual nature (cleanup + new
> > > mappings)
> > > 
> > > Cc: Matthew Brost<matthew.brost@intel.com>
> > > Cc: Thomas Hellström<thomas.hellstrom@linux.intel.com>
> > > Cc: Himal Prasad Ghimiray<himal.prasad.ghimiray@intel.com>
> > > Signed-off-by: Arvind Yadav<arvind.yadav@intel.com>
> > > ---
> > >   drivers/gpu/drm/xe/xe_vm.c | 20 +++++++++++++-------
> > >   1 file changed, 13 insertions(+), 7 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/xe/xe_vm.c
> > > b/drivers/gpu/drm/xe/xe_vm.c
> > > index c3a5fe76ff96..f250daae3012 100644
> > > --- a/drivers/gpu/drm/xe/xe_vm.c
> > > +++ b/drivers/gpu/drm/xe/xe_vm.c
> > > @@ -2883,7 +2883,7 @@ static void vm_bind_ioctl_ops_unwind(struct
> > > xe_vm *vm,
> > >   }
> > > 
> > >   static int vma_lock_and_validate(struct drm_exec *exec, struct
> > > xe_vma *vma,
> > > -				 bool res_evict, bool validate)
> > > +				 bool res_evict, bool validate,
> > > bool check_purged)
> > It probably time to add something like this to avoid transposing
> > arguments.
> > 
> > struct lock_and_validate_flags {
> > 	bool res_evict;
> > 	bool validate;
> > 	bool check_purged;
> > };
> > 
> > Logic in the patch looks correct though.
> 
> 
> Noted, I will add "struct xe_lock_and_validate_flags" Thanks, Arvind

Note that if you follow the pattern of struct xe_bo_shink_flags,
passing the struct as a const and using

struct lock_and_validate_flags {
	u32 res_evict : 1;
        u32 validate : 1;
        u32 check_purged : 1;
};

This will be more or less equivalent to passing a bit-field with type-
checking.

Some reviewers frown on using "bool" in compound types although we
accept that in the xe driver.

Otherwise patch LGTM as well.

/Thomas


> 
> > 
> > Matt
> > 
> > >   {
> > >   	struct xe_bo *bo = xe_vma_bo(vma);
> > >   	struct xe_vm *vm = xe_vma_vm(vma);
> > > @@ -2892,6 +2892,11 @@ static int vma_lock_and_validate(struct
> > > drm_exec *exec, struct xe_vma *vma,
> > >   	if (bo) {
> > >   		if (!bo->vm)
> > >   			err = drm_exec_lock_obj(exec, &bo-
> > > >ttm.base);
> > > +
> > > +		/* Reject new mappings to purged BOs; allow
> > > cleanup operations */
> > > +		if (!err && check_purged && xe_bo_is_purged(bo))
> > > +			err = -EINVAL;
> > > +
> > >   		if (!err && validate)
> > >   			err = xe_bo_validate(bo, vm,
> > >   					    
> > > !xe_vm_in_preempt_fence_mode(vm) &&
> > > @@ -2990,7 +2995,8 @@ static int op_lock_and_prep(struct drm_exec
> > > *exec, struct xe_vm *vm,
> > >   			err = vma_lock_and_validate(exec, op-
> > > >map.vma,
> > >   						    res_evict,
> > >   						   
> > > !xe_vm_in_fault_mode(vm) ||
> > > -						    op-
> > > >map.immediate);
> > > +						    op-
> > > >map.immediate,
> > > +						    true);
> > >   		break;
> > >   	case DRM_GPUVA_OP_REMAP:
> > >   		err = check_ufence(gpuva_to_vma(op-
> > > >base.remap.unmap->va));
> > > @@ -2999,13 +3005,13 @@ static int op_lock_and_prep(struct
> > > drm_exec *exec, struct xe_vm *vm,
> > >   
> > >   		err = vma_lock_and_validate(exec,
> > >   					    gpuva_to_vma(op-
> > > >base.remap.unmap->va),
> > > -					    res_evict, false);
> > > +					    res_evict, false,
> > > false);
> > >   		if (!err && op->remap.prev)
> > >   			err = vma_lock_and_validate(exec, op-
> > > >remap.prev,
> > > -						    res_evict,
> > > true);
> > > +						    res_evict,
> > > true, true);
> > >   		if (!err && op->remap.next)
> > >   			err = vma_lock_and_validate(exec, op-
> > > >remap.next,
> > > -						    res_evict,
> > > true);
> > > +						    res_evict,
> > > true, true);
> > >   		break;
> > >   	case DRM_GPUVA_OP_UNMAP:
> > >   		err = check_ufence(gpuva_to_vma(op-
> > > >base.unmap.va));
> > > @@ -3014,7 +3020,7 @@ static int op_lock_and_prep(struct drm_exec
> > > *exec, struct xe_vm *vm,
> > >   
> > >   		err = vma_lock_and_validate(exec,
> > >   					    gpuva_to_vma(op-
> > > >base.unmap.va),
> > > -					    res_evict, false);
> > > +					    res_evict, false,
> > > false);
> > >   		break;
> > >   	case DRM_GPUVA_OP_PREFETCH:
> > >   	{
> > > @@ -3029,7 +3035,7 @@ static int op_lock_and_prep(struct drm_exec
> > > *exec, struct xe_vm *vm,
> > >   
> > >   		err = vma_lock_and_validate(exec,
> > >   					    gpuva_to_vma(op-
> > > >base.prefetch.va),
> > > -					    res_evict, false);
> > > +					    res_evict, false,
> > > true);
> > >   		if (!err && !xe_vma_has_no_bo(vma))
> > >   			err = xe_bo_migrate(xe_vma_bo(vma),
> > >   					   
> > > region_to_mem_type[region],
> > > -- 
> > > 2.43.0

Re: [PATCH v4 5/8] drm/xe/vm: Prevent binding of purged buffer objects

[Yadav, Arvind]

On 23-01-2026 18:07, Thomas Hellström wrote:
> On Fri, 2026-01-23 at 11:11 +0530, Yadav, Arvind wrote:
>> On 20-01-2026 22:57, Matthew Brost wrote:
>>> On Tue, Jan 20, 2026 at 11:38:51AM +0530, Arvind Yadav wrote:
>>>> Add check_purged parameter to vma_lock_and_validate() to block
>>>> new mapping operations on purged BOs while allowing cleanup
>>>> operations to proceed.
>>>>
>>>> Purged BOs have their backing pages freed by the kernel. New
>>>> mapping operations (MAP, PREFETCH, REMAP) must be rejected with
>>>> -EINVAL to prevent GPU access to invalid memory. Cleanup
>>>> operations (UNMAP) must be allowed so applications can release
>>>> resources after detecting purge via the retained field.
>>>>
>>>> REMAP operations require mixed handling - reject new prev/next
>>>> VMAs if the BO is purged, but allow the unmap portion to proceed
>>>> for cleanup.
>>>>
>>>> The check_purged parameter distinguishes between these cases:
>>>> true for new mappings (must reject), false for cleanup (allow).
>>>>
>>>> v2:
>>>>     - Clarify that purged BOs are permanently invalid (i915
>>>> semantics)
>>>>     - Remove incorrect claim about madvise(WILLNEED) restoring
>>>> purged BOs
>>>>
>>>> v3:
>>>>     - Move xe_bo_is_purged check under vma_lock_and_validate
>>>> (Matthew Brost)
>>>>     - Add check_purged parameter to distinguish new mappings from
>>>> cleanup
>>>>     - Allow UNMAP operations to prevent resource leaks
>>>>     - Handle REMAP operation's dual nature (cleanup + new
>>>> mappings)
>>>>
>>>> Cc: Matthew Brost<matthew.brost@intel.com>
>>>> Cc: Thomas Hellström<thomas.hellstrom@linux.intel.com>
>>>> Cc: Himal Prasad Ghimiray<himal.prasad.ghimiray@intel.com>
>>>> Signed-off-by: Arvind Yadav<arvind.yadav@intel.com>
>>>> ---
>>>>    drivers/gpu/drm/xe/xe_vm.c | 20 +++++++++++++-------
>>>>    1 file changed, 13 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/xe/xe_vm.c
>>>> b/drivers/gpu/drm/xe/xe_vm.c
>>>> index c3a5fe76ff96..f250daae3012 100644
>>>> --- a/drivers/gpu/drm/xe/xe_vm.c
>>>> +++ b/drivers/gpu/drm/xe/xe_vm.c
>>>> @@ -2883,7 +2883,7 @@ static void vm_bind_ioctl_ops_unwind(struct
>>>> xe_vm *vm,
>>>>    }
>>>>
>>>>    static int vma_lock_and_validate(struct drm_exec *exec, struct
>>>> xe_vma *vma,
>>>> -				 bool res_evict, bool validate)
>>>> +				 bool res_evict, bool validate,
>>>> bool check_purged)
>>> It probably time to add something like this to avoid transposing
>>> arguments.
>>>
>>> struct lock_and_validate_flags {
>>> 	bool res_evict;
>>> 	bool validate;
>>> 	bool check_purged;
>>> };
>>>
>>> Logic in the patch looks correct though.
>>
>> Noted, I will add "struct xe_lock_and_validate_flags" Thanks, Arvind
> Note that if you follow the pattern of struct xe_bo_shink_flags,
> passing the struct as a const and using
>
> struct lock_and_validate_flags {
> 	u32 res_evict : 1;
>          u32 validate : 1;
>          u32 check_purged : 1;
> };
>
> This will be more or less equivalent to passing a bit-field with type-
> checking.
>
> Some reviewers frown on using "bool" in compound types although we
> accept that in the xe driver.


Noted, I will do the changes as per suggestion..


Thanks,
Arvind

> Otherwise patch LGTM as well.
>
> /Thomas
>
>
>>> Matt
>>>
>>>>    {
>>>>    	struct xe_bo *bo = xe_vma_bo(vma);
>>>>    	struct xe_vm *vm = xe_vma_vm(vma);
>>>> @@ -2892,6 +2892,11 @@ static int vma_lock_and_validate(struct
>>>> drm_exec *exec, struct xe_vma *vma,
>>>>    	if (bo) {
>>>>    		if (!bo->vm)
>>>>    			err = drm_exec_lock_obj(exec, &bo-
>>>>> ttm.base);
>>>> +
>>>> +		/* Reject new mappings to purged BOs; allow
>>>> cleanup operations */
>>>> +		if (!err && check_purged && xe_bo_is_purged(bo))
>>>> +			err = -EINVAL;
>>>> +
>>>>    		if (!err && validate)
>>>>    			err = xe_bo_validate(bo, vm,
>>>>    					
>>>> !xe_vm_in_preempt_fence_mode(vm) &&
>>>> @@ -2990,7 +2995,8 @@ static int op_lock_and_prep(struct drm_exec
>>>> *exec, struct xe_vm *vm,
>>>>    			err = vma_lock_and_validate(exec, op-
>>>>> map.vma,
>>>>    						    res_evict,
>>>>    						
>>>> !xe_vm_in_fault_mode(vm) ||
>>>> -						    op-
>>>>> map.immediate);
>>>> +						    op-
>>>>> map.immediate,
>>>> +						    true);
>>>>    		break;
>>>>    	case DRM_GPUVA_OP_REMAP:
>>>>    		err = check_ufence(gpuva_to_vma(op-
>>>>> base.remap.unmap->va));
>>>> @@ -2999,13 +3005,13 @@ static int op_lock_and_prep(struct
>>>> drm_exec *exec, struct xe_vm *vm,
>>>>    
>>>>    		err = vma_lock_and_validate(exec,
>>>>    					    gpuva_to_vma(op-
>>>>> base.remap.unmap->va),
>>>> -					    res_evict, false);
>>>> +					    res_evict, false,
>>>> false);
>>>>    		if (!err && op->remap.prev)
>>>>    			err = vma_lock_and_validate(exec, op-
>>>>> remap.prev,
>>>> -						    res_evict,
>>>> true);
>>>> +						    res_evict,
>>>> true, true);
>>>>    		if (!err && op->remap.next)
>>>>    			err = vma_lock_and_validate(exec, op-
>>>>> remap.next,
>>>> -						    res_evict,
>>>> true);
>>>> +						    res_evict,
>>>> true, true);
>>>>    		break;
>>>>    	case DRM_GPUVA_OP_UNMAP:
>>>>    		err = check_ufence(gpuva_to_vma(op-
>>>>> base.unmap.va));
>>>> @@ -3014,7 +3020,7 @@ static int op_lock_and_prep(struct drm_exec
>>>> *exec, struct xe_vm *vm,
>>>>    
>>>>    		err = vma_lock_and_validate(exec,
>>>>    					    gpuva_to_vma(op-
>>>>> base.unmap.va),
>>>> -					    res_evict, false);
>>>> +					    res_evict, false,
>>>> false);
>>>>    		break;
>>>>    	case DRM_GPUVA_OP_PREFETCH:
>>>>    	{
>>>> @@ -3029,7 +3035,7 @@ static int op_lock_and_prep(struct drm_exec
>>>> *exec, struct xe_vm *vm,
>>>>    
>>>>    		err = vma_lock_and_validate(exec,
>>>>    					    gpuva_to_vma(op-
>>>>> base.prefetch.va),
>>>> -					    res_evict, false);
>>>> +					    res_evict, false,
>>>> true);
>>>>    		if (!err && !xe_vma_has_no_bo(vma))
>>>>    			err = xe_bo_migrate(xe_vma_bo(vma),
>>>>    					
>>>> region_to_mem_type[region],
>>>> -- 
>>>> 2.43.0

[PATCH v4 6/8] drm/xe/madvise: Implement per-VMA purgeable state tracking

[Arvind Yadav]

Track purgeable state per-VMA instead of using a coarse shared
BO check. This prevents purging shared BOs until all VMAs across
all VMs are marked DONTNEED.

Add xe_bo_all_vmas_dontneed() to check all VMAs before marking
a BO purgeable. Add xe_bo_recheck_purgeable_on_vma_unbind() to
handle state transitions when VMAs are destroyed - if all
remaining VMAs are DONTNEED the BO can become purgeable, or if
no VMAs remain it transitions to WILLNEED.

The per-VMA purgeable_state field stores the madvise hint for
each mapping. Shared BOs can only be purged when all VMAs
unanimously indicate DONTNEED.

v3:
  - This addresses Thomas Hellström's feedback: "loop over all vmas
    attached to the bo and check that they all say WONTNEED. This will
    also need a check at VMA unbinding"

v4:
  - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 drivers/gpu/drm/xe/xe_vm.c         | 15 +++++-
 drivers/gpu/drm/xe/xe_vm_madvise.c | 84 +++++++++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_vm_madvise.h |  3 ++
 drivers/gpu/drm/xe/xe_vm_types.h   | 11 ++++
 4 files changed, 111 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index f250daae3012..9543960b5613 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -40,6 +40,7 @@
 #include "xe_tile.h"
 #include "xe_tlb_inval.h"
 #include "xe_trace_bo.h"
+#include "xe_vm_madvise.h"
 #include "xe_wa.h"
 
 static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
@@ -1079,12 +1080,18 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 static void xe_vma_destroy_late(struct xe_vma *vma)
 {
 	struct xe_vm *vm = xe_vma_vm(vma);
+	struct xe_bo *bo = NULL;
 
 	if (vma->ufence) {
 		xe_sync_ufence_put(vma->ufence);
 		vma->ufence = NULL;
 	}
 
+	/* Get BO reference for purgeable state re-check */
+	if (!xe_vma_is_userptr(vma) && !xe_vma_is_null(vma) &&
+	    !xe_vma_is_cpu_addr_mirror(vma))
+		bo = xe_vma_bo(vma);
+
 	if (xe_vma_is_userptr(vma)) {
 		struct xe_userptr_vma *uvma = to_userptr_vma(vma);
 
@@ -1093,7 +1100,13 @@ static void xe_vma_destroy_late(struct xe_vma *vma)
 	} else if (xe_vma_is_null(vma) || xe_vma_is_cpu_addr_mirror(vma)) {
 		xe_vm_put(vm);
 	} else {
-		xe_bo_put(xe_vma_bo(vma));
+		/* Trylock safe for async context; madvise corrects failures */
+		if (bo && dma_resv_trylock(bo->ttm.base.resv)) {
+			xe_bo_recheck_purgeable_on_vma_unbind(bo);
+			dma_resv_unlock(bo->ttm.base.resv);
+		}
+
+		xe_bo_put(bo);
 	}
 
 	xe_vma_free(vma);
diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
index dfeab9e24a09..27b6ad65b314 100644
--- a/drivers/gpu/drm/xe/xe_vm_madvise.c
+++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
@@ -12,6 +12,7 @@
 #include "xe_pat.h"
 #include "xe_pt.h"
 #include "xe_svm.h"
+#include "xe_vm.h"
 
 struct xe_vmas_in_madvise_range {
 	u64 addr;
@@ -179,6 +180,80 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
 	}
 }
 
+/**
+ * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are marked DONTNEED
+ * @bo: Buffer object
+ *
+ * Check all VMAs across all VMs to determine if BO can be purged.
+ * Shared BOs require unanimous DONTNEED state from all mappings.
+ *
+ * Caller must hold BO dma-resv lock.
+ *
+ * Return: true if all VMAs are DONTNEED, false otherwise
+ */
+static bool xe_bo_all_vmas_dontneed(struct xe_bo *bo)
+{
+	struct drm_gpuvm_bo *vm_bo;
+	struct drm_gpuva *gpuva;
+	struct drm_gem_object *obj = &bo->ttm.base;
+	bool has_vmas = false;
+
+	dma_resv_assert_held(bo->ttm.base.resv);
+
+	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
+		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
+			struct xe_vma *vma = gpuva_to_vma(gpuva);
+
+			has_vmas = true;
+
+			/* Any non-DONTNEED VMA prevents purging */
+			if (READ_ONCE(vma->purgeable_state) != XE_MADV_PURGEABLE_DONTNEED)
+				return false;
+		}
+	}
+
+	/* No VMAs means not purgeable */
+	if (!has_vmas)
+		return false;
+
+	return true;
+}
+
+/**
+ * xe_bo_recheck_purgeable_on_vma_unbind() - Re-evaluate BO purgeable state after VMA unbind
+ * @bo: Buffer object
+ *
+ * When a VMA is unbound, re-check if the BO's purgeable state should change.
+ * Destroyed VMAs may allow the BO to become purgeable if all remaining VMAs
+ * are DONTNEED, or require transition to WILLNEED if no VMAs remain.
+ *
+ * Called from VMA destruction path with BO dma-resv lock held.
+ */
+void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo)
+{
+	if (!bo)
+		return;
+
+	dma_resv_assert_held(bo->ttm.base.resv);
+
+	/*
+	 * Once purged, always purged. Cannot transition back to WILLNEED.
+	 * This matches i915 semantics where purged BOs are permanently invalid.
+	 */
+	if (bo->madv_purgeable == XE_MADV_PURGEABLE_PURGED)
+		return;
+
+	if (xe_bo_all_vmas_dontneed(bo)) {
+		/* All VMAs are DONTNEED - mark BO purgeable */
+		if (bo->madv_purgeable != XE_MADV_PURGEABLE_DONTNEED)
+			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
+	} else {
+		/* At least one VMA is WILLNEED - BO must not be purgeable */
+		if (bo->madv_purgeable != XE_MADV_PURGEABLE_WILLNEED)
+			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
+	}
+}
+
 /*
  * Handle purgeable buffer object advice for DONTNEED/WILLNEED/PURGED.
  * Returns true if any BO was purged, false otherwise.
@@ -213,10 +288,17 @@ static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
 
 		switch (op->purge_state_val.val) {
 		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
+			vmas[i]->purgeable_state = XE_MADV_PURGEABLE_WILLNEED;
+
+			/* Mark VMA WILLNEED - BO becomes non-purgeable immediately */
 			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
 			break;
 		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
-			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
+			vmas[i]->purgeable_state = XE_MADV_PURGEABLE_DONTNEED;
+
+			/* Mark BO purgeable only if all VMAs are DONTNEED */
+			if (xe_bo_all_vmas_dontneed(bo))
+				bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
 			break;
 		default:
 			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.h b/drivers/gpu/drm/xe/xe_vm_madvise.h
index b0e1fc445f23..61868f851949 100644
--- a/drivers/gpu/drm/xe/xe_vm_madvise.h
+++ b/drivers/gpu/drm/xe/xe_vm_madvise.h
@@ -8,8 +8,11 @@
 
 struct drm_device;
 struct drm_file;
+struct xe_bo;
 
 int xe_vm_madvise_ioctl(struct drm_device *dev, void *data,
 			struct drm_file *file);
 
+void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo);
+
 #endif
diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
index 437f64202f3b..94ca9d033b06 100644
--- a/drivers/gpu/drm/xe/xe_vm_types.h
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -150,6 +150,17 @@ struct xe_vma {
 	 */
 	bool skip_invalidation;
 
+	/**
+	 * @purgeable_state: Purgeable hint for this VMA mapping
+	 *
+	 * Per-VMA purgeable state from madvise. Valid states are WILLNEED (0)
+	 * or DONTNEED (1). Shared BOs require all VMAs to be DONTNEED before
+	 * the BO can be purged. PURGED state exists only at BO level.
+	 *
+	 * Protected by BO dma-resv lock. Set via DRM_IOCTL_XE_MADVISE.
+	 */
+	u32 purgeable_state;
+
 	/**
 	 * @ufence: The user fence that was provided with MAP.
 	 * Needs to be signalled before UNMAP can be processed.
-- 
2.43.0

Re: [PATCH v4 6/8] drm/xe/madvise: Implement per-VMA purgeable state tracking

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:52AM +0530, Arvind Yadav wrote:
> Track purgeable state per-VMA instead of using a coarse shared
> BO check. This prevents purging shared BOs until all VMAs across
> all VMs are marked DONTNEED.
> 
> Add xe_bo_all_vmas_dontneed() to check all VMAs before marking
> a BO purgeable. Add xe_bo_recheck_purgeable_on_vma_unbind() to
> handle state transitions when VMAs are destroyed - if all
> remaining VMAs are DONTNEED the BO can become purgeable, or if
> no VMAs remain it transitions to WILLNEED.
> 
> The per-VMA purgeable_state field stores the madvise hint for
> each mapping. Shared BOs can only be purged when all VMAs
> unanimously indicate DONTNEED.
> 
> v3:
>   - This addresses Thomas Hellström's feedback: "loop over all vmas
>     attached to the bo and check that they all say WONTNEED. This will
>     also need a check at VMA unbinding"
> 
> v4:
>   - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
> 
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_vm.c         | 15 +++++-
>  drivers/gpu/drm/xe/xe_vm_madvise.c | 84 +++++++++++++++++++++++++++++-
>  drivers/gpu/drm/xe/xe_vm_madvise.h |  3 ++
>  drivers/gpu/drm/xe/xe_vm_types.h   | 11 ++++
>  4 files changed, 111 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index f250daae3012..9543960b5613 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -40,6 +40,7 @@
>  #include "xe_tile.h"
>  #include "xe_tlb_inval.h"
>  #include "xe_trace_bo.h"
> +#include "xe_vm_madvise.h"
>  #include "xe_wa.h"
>  
>  static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
> @@ -1079,12 +1080,18 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
>  static void xe_vma_destroy_late(struct xe_vma *vma)
>  {
>  	struct xe_vm *vm = xe_vma_vm(vma);
> +	struct xe_bo *bo = NULL;
>  
>  	if (vma->ufence) {
>  		xe_sync_ufence_put(vma->ufence);
>  		vma->ufence = NULL;
>  	}
>  
> +	/* Get BO reference for purgeable state re-check */
> +	if (!xe_vma_is_userptr(vma) && !xe_vma_is_null(vma) &&
> +	    !xe_vma_is_cpu_addr_mirror(vma))
> +		bo = xe_vma_bo(vma);

I think xe_vma_bo just returns NULL if any of the above conditions are
met, so I believe is ok to just blindly call xe_vma_bo as you have a
NULL check on the BO below.

> +
>  	if (xe_vma_is_userptr(vma)) {
>  		struct xe_userptr_vma *uvma = to_userptr_vma(vma);
>  
> @@ -1093,7 +1100,13 @@ static void xe_vma_destroy_late(struct xe_vma *vma)
>  	} else if (xe_vma_is_null(vma) || xe_vma_is_cpu_addr_mirror(vma)) {
>  		xe_vm_put(vm);
>  	} else {
> -		xe_bo_put(xe_vma_bo(vma));
> +		/* Trylock safe for async context; madvise corrects failures */
> +		if (bo && dma_resv_trylock(bo->ttm.base.resv)) {
> +			xe_bo_recheck_purgeable_on_vma_unbind(bo);

Also I don't think the correct place to call this.

I believe you can call this function in xe_vma_destroy after
drm_gpuva_unlink. You also have BO lock there too so no need from this
trylock path.

> +			dma_resv_unlock(bo->ttm.base.resv);
> +		}
> +
> +		xe_bo_put(bo);
>  	}
>  
>  	xe_vma_free(vma);
> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
> index dfeab9e24a09..27b6ad65b314 100644
> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> @@ -12,6 +12,7 @@
>  #include "xe_pat.h"
>  #include "xe_pt.h"
>  #include "xe_svm.h"
> +#include "xe_vm.h"
>  
>  struct xe_vmas_in_madvise_range {
>  	u64 addr;
> @@ -179,6 +180,80 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
>  	}
>  }
>  
> +/**
> + * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are marked DONTNEED
> + * @bo: Buffer object
> + *
> + * Check all VMAs across all VMs to determine if BO can be purged.
> + * Shared BOs require unanimous DONTNEED state from all mappings.
> + *
> + * Caller must hold BO dma-resv lock.
> + *
> + * Return: true if all VMAs are DONTNEED, false otherwise
> + */
> +static bool xe_bo_all_vmas_dontneed(struct xe_bo *bo)
> +{
> +	struct drm_gpuvm_bo *vm_bo;
> +	struct drm_gpuva *gpuva;
> +	struct drm_gem_object *obj = &bo->ttm.base;
> +	bool has_vmas = false;
> +
> +	dma_resv_assert_held(bo->ttm.base.resv);
> +
> +	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
> +		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
> +			struct xe_vma *vma = gpuva_to_vma(gpuva);
> +
> +			has_vmas = true;
> +
> +			/* Any non-DONTNEED VMA prevents purging */
> +			if (READ_ONCE(vma->purgeable_state) != XE_MADV_PURGEABLE_DONTNEED)

You don't need the READ_ONCE as purgeable_state is only accessed under
the dma-resv lock, also there isn't a WRITE_ONCE anywhere.

> +				return false;
> +		}
> +	}
> +
> +	/* No VMAs means not purgeable */

No VMAs means it is purgeable, right?

> +	if (!has_vmas)
> +		return false;
> +
> +	return true;
> +}
> +
> +/**
> + * xe_bo_recheck_purgeable_on_vma_unbind() - Re-evaluate BO purgeable state after VMA unbind
> + * @bo: Buffer object
> + *
> + * When a VMA is unbound, re-check if the BO's purgeable state should change.
> + * Destroyed VMAs may allow the BO to become purgeable if all remaining VMAs
> + * are DONTNEED, or require transition to WILLNEED if no VMAs remain.
> + *
> + * Called from VMA destruction path with BO dma-resv lock held.
> + */
> +void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo)
> +{
> +	if (!bo)
> +		return;
> +
> +	dma_resv_assert_held(bo->ttm.base.resv);
> +
> +	/*
> +	 * Once purged, always purged. Cannot transition back to WILLNEED.
> +	 * This matches i915 semantics where purged BOs are permanently invalid.
> +	 */
> +	if (bo->madv_purgeable == XE_MADV_PURGEABLE_PURGED)
> +		return;
> +
> +	if (xe_bo_all_vmas_dontneed(bo)) {
> +		/* All VMAs are DONTNEED - mark BO purgeable */
> +		if (bo->madv_purgeable != XE_MADV_PURGEABLE_DONTNEED)
> +			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
> +	} else {
> +		/* At least one VMA is WILLNEED - BO must not be purgeable */
> +		if (bo->madv_purgeable != XE_MADV_PURGEABLE_WILLNEED)
> +			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> +	}
> +}
> +
>  /*
>   * Handle purgeable buffer object advice for DONTNEED/WILLNEED/PURGED.
>   * Returns true if any BO was purged, false otherwise.
> @@ -213,10 +288,17 @@ static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
>  
>  		switch (op->purge_state_val.val) {
>  		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
> +			vmas[i]->purgeable_state = XE_MADV_PURGEABLE_WILLNEED;
> +
> +			/* Mark VMA WILLNEED - BO becomes non-purgeable immediately */
>  			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
>  			break;
>  		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
> -			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
> +			vmas[i]->purgeable_state = XE_MADV_PURGEABLE_DONTNEED;
> +
> +			/* Mark BO purgeable only if all VMAs are DONTNEED */
> +			if (xe_bo_all_vmas_dontneed(bo))
> +				bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
>  			break;
>  		default:
>  			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.h b/drivers/gpu/drm/xe/xe_vm_madvise.h
> index b0e1fc445f23..61868f851949 100644
> --- a/drivers/gpu/drm/xe/xe_vm_madvise.h
> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.h
> @@ -8,8 +8,11 @@
>  
>  struct drm_device;
>  struct drm_file;
> +struct xe_bo;
>  
>  int xe_vm_madvise_ioctl(struct drm_device *dev, void *data,
>  			struct drm_file *file);
>  
> +void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo);
> +
>  #endif
> diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
> index 437f64202f3b..94ca9d033b06 100644
> --- a/drivers/gpu/drm/xe/xe_vm_types.h
> +++ b/drivers/gpu/drm/xe/xe_vm_types.h
> @@ -150,6 +150,17 @@ struct xe_vma {
>  	 */
>  	bool skip_invalidation;
>  
> +	/**
> +	 * @purgeable_state: Purgeable hint for this VMA mapping
> +	 *
> +	 * Per-VMA purgeable state from madvise. Valid states are WILLNEED (0)
> +	 * or DONTNEED (1). Shared BOs require all VMAs to be DONTNEED before
> +	 * the BO can be purged. PURGED state exists only at BO level.
> +	 *
> +	 * Protected by BO dma-resv lock. Set via DRM_IOCTL_XE_MADVISE.
> +	 */
> +	u32 purgeable_state;
> +

I think xe_vma_mem_attr is a better place for this field.

Matt

>  	/**
>  	 * @ufence: The user fence that was provided with MAP.
>  	 * Needs to be signalled before UNMAP can be processed.
> -- 
> 2.43.0
> 

Re: [PATCH v4 6/8] drm/xe/madvise: Implement per-VMA purgeable state tracking

[Yadav, Arvind]

On 20-01-2026 23:11, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:52AM +0530, Arvind Yadav wrote:
>> Track purgeable state per-VMA instead of using a coarse shared
>> BO check. This prevents purging shared BOs until all VMAs across
>> all VMs are marked DONTNEED.
>>
>> Add xe_bo_all_vmas_dontneed() to check all VMAs before marking
>> a BO purgeable. Add xe_bo_recheck_purgeable_on_vma_unbind() to
>> handle state transitions when VMAs are destroyed - if all
>> remaining VMAs are DONTNEED the BO can become purgeable, or if
>> no VMAs remain it transitions to WILLNEED.
>>
>> The per-VMA purgeable_state field stores the madvise hint for
>> each mapping. Shared BOs can only be purged when all VMAs
>> unanimously indicate DONTNEED.
>>
>> v3:
>>    - This addresses Thomas Hellström's feedback: "loop over all vmas
>>      attached to the bo and check that they all say WONTNEED. This will
>>      also need a check at VMA unbinding"
>>
>> v4:
>>    - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
>>
>> Cc: Matthew Brost <matthew.brost@intel.com>
>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
>> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
>> ---
>>   drivers/gpu/drm/xe/xe_vm.c         | 15 +++++-
>>   drivers/gpu/drm/xe/xe_vm_madvise.c | 84 +++++++++++++++++++++++++++++-
>>   drivers/gpu/drm/xe/xe_vm_madvise.h |  3 ++
>>   drivers/gpu/drm/xe/xe_vm_types.h   | 11 ++++
>>   4 files changed, 111 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
>> index f250daae3012..9543960b5613 100644
>> --- a/drivers/gpu/drm/xe/xe_vm.c
>> +++ b/drivers/gpu/drm/xe/xe_vm.c
>> @@ -40,6 +40,7 @@
>>   #include "xe_tile.h"
>>   #include "xe_tlb_inval.h"
>>   #include "xe_trace_bo.h"
>> +#include "xe_vm_madvise.h"
>>   #include "xe_wa.h"
>>   
>>   static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
>> @@ -1079,12 +1080,18 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
>>   static void xe_vma_destroy_late(struct xe_vma *vma)
>>   {
>>   	struct xe_vm *vm = xe_vma_vm(vma);
>> +	struct xe_bo *bo = NULL;
>>   
>>   	if (vma->ufence) {
>>   		xe_sync_ufence_put(vma->ufence);
>>   		vma->ufence = NULL;
>>   	}
>>   
>> +	/* Get BO reference for purgeable state re-check */
>> +	if (!xe_vma_is_userptr(vma) && !xe_vma_is_null(vma) &&
>> +	    !xe_vma_is_cpu_addr_mirror(vma))
>> +		bo = xe_vma_bo(vma);
> I think xe_vma_bo just returns NULL if any of the above conditions are
> met, so I believe is ok to just blindly call xe_vma_bo as you have a
> NULL check on the BO below.


Right, xe_vma_bo() already returns NULL for those cases. Will simplify 
to just bo = xe_vma_bo(vma).

>
>> +
>>   	if (xe_vma_is_userptr(vma)) {
>>   		struct xe_userptr_vma *uvma = to_userptr_vma(vma);
>>   
>> @@ -1093,7 +1100,13 @@ static void xe_vma_destroy_late(struct xe_vma *vma)
>>   	} else if (xe_vma_is_null(vma) || xe_vma_is_cpu_addr_mirror(vma)) {
>>   		xe_vm_put(vm);
>>   	} else {
>> -		xe_bo_put(xe_vma_bo(vma));
>> +		/* Trylock safe for async context; madvise corrects failures */
>> +		if (bo && dma_resv_trylock(bo->ttm.base.resv)) {
>> +			xe_bo_recheck_purgeable_on_vma_unbind(bo);
> Also I don't think the correct place to call this.
>
> I believe you can call this function in xe_vma_destroy after
> drm_gpuva_unlink. You also have BO lock there too so no need from this
> trylock path.


Good catch! Will move to xe_vma_destroy() after drm_gpuva_unlink() where 
we already have the BO lock. No trylock needed there.

>> +			dma_resv_unlock(bo->ttm.base.resv);
>> +		}
>> +
>> +		xe_bo_put(bo);
>>   	}
>>   
>>   	xe_vma_free(vma);
>> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
>> index dfeab9e24a09..27b6ad65b314 100644
>> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
>> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
>> @@ -12,6 +12,7 @@
>>   #include "xe_pat.h"
>>   #include "xe_pt.h"
>>   #include "xe_svm.h"
>> +#include "xe_vm.h"
>>   
>>   struct xe_vmas_in_madvise_range {
>>   	u64 addr;
>> @@ -179,6 +180,80 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
>>   	}
>>   }
>>   
>> +/**
>> + * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are marked DONTNEED
>> + * @bo: Buffer object
>> + *
>> + * Check all VMAs across all VMs to determine if BO can be purged.
>> + * Shared BOs require unanimous DONTNEED state from all mappings.
>> + *
>> + * Caller must hold BO dma-resv lock.
>> + *
>> + * Return: true if all VMAs are DONTNEED, false otherwise
>> + */
>> +static bool xe_bo_all_vmas_dontneed(struct xe_bo *bo)
>> +{
>> +	struct drm_gpuvm_bo *vm_bo;
>> +	struct drm_gpuva *gpuva;
>> +	struct drm_gem_object *obj = &bo->ttm.base;
>> +	bool has_vmas = false;
>> +
>> +	dma_resv_assert_held(bo->ttm.base.resv);
>> +
>> +	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
>> +		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
>> +			struct xe_vma *vma = gpuva_to_vma(gpuva);
>> +
>> +			has_vmas = true;
>> +
>> +			/* Any non-DONTNEED VMA prevents purging */
>> +			if (READ_ONCE(vma->purgeable_state) != XE_MADV_PURGEABLE_DONTNEED)
> You don't need the READ_ONCE as purgeable_state is only accessed under
> the dma-resv lock, also there isn't a WRITE_ONCE anywhere.


Noted - I will remove READ_ONCE() - the dma-resv lock provides the 
necessary protection.

>
>> +				return false;
>> +		}
>> +	}
>> +
>> +	/* No VMAs means not purgeable */
> No VMAs means it is purgeable, right?


I kept “no VMAs => not purgeable” since otherwise we can mark BOs 
purgeable without any userspace DONTNEED hint (No VMAs means no user 
hint at all).

>> +	if (!has_vmas)
>> +		return false;
>> +
>> +	return true;
>> +}
>> +
>> +/**
>> + * xe_bo_recheck_purgeable_on_vma_unbind() - Re-evaluate BO purgeable state after VMA unbind
>> + * @bo: Buffer object
>> + *
>> + * When a VMA is unbound, re-check if the BO's purgeable state should change.
>> + * Destroyed VMAs may allow the BO to become purgeable if all remaining VMAs
>> + * are DONTNEED, or require transition to WILLNEED if no VMAs remain.
>> + *
>> + * Called from VMA destruction path with BO dma-resv lock held.
>> + */
>> +void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo)
>> +{
>> +	if (!bo)
>> +		return;
>> +
>> +	dma_resv_assert_held(bo->ttm.base.resv);
>> +
>> +	/*
>> +	 * Once purged, always purged. Cannot transition back to WILLNEED.
>> +	 * This matches i915 semantics where purged BOs are permanently invalid.
>> +	 */
>> +	if (bo->madv_purgeable == XE_MADV_PURGEABLE_PURGED)
>> +		return;
>> +
>> +	if (xe_bo_all_vmas_dontneed(bo)) {
>> +		/* All VMAs are DONTNEED - mark BO purgeable */
>> +		if (bo->madv_purgeable != XE_MADV_PURGEABLE_DONTNEED)
>> +			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
>> +	} else {
>> +		/* At least one VMA is WILLNEED - BO must not be purgeable */
>> +		if (bo->madv_purgeable != XE_MADV_PURGEABLE_WILLNEED)
>> +			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
>> +	}
>> +}
>> +
>>   /*
>>    * Handle purgeable buffer object advice for DONTNEED/WILLNEED/PURGED.
>>    * Returns true if any BO was purged, false otherwise.
>> @@ -213,10 +288,17 @@ static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
>>   
>>   		switch (op->purge_state_val.val) {
>>   		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
>> +			vmas[i]->purgeable_state = XE_MADV_PURGEABLE_WILLNEED;
>> +
>> +			/* Mark VMA WILLNEED - BO becomes non-purgeable immediately */
>>   			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
>>   			break;
>>   		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
>> -			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
>> +			vmas[i]->purgeable_state = XE_MADV_PURGEABLE_DONTNEED;
>> +
>> +			/* Mark BO purgeable only if all VMAs are DONTNEED */
>> +			if (xe_bo_all_vmas_dontneed(bo))
>> +				bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
>>   			break;
>>   		default:
>>   			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
>> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.h b/drivers/gpu/drm/xe/xe_vm_madvise.h
>> index b0e1fc445f23..61868f851949 100644
>> --- a/drivers/gpu/drm/xe/xe_vm_madvise.h
>> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.h
>> @@ -8,8 +8,11 @@
>>   
>>   struct drm_device;
>>   struct drm_file;
>> +struct xe_bo;
>>   
>>   int xe_vm_madvise_ioctl(struct drm_device *dev, void *data,
>>   			struct drm_file *file);
>>   
>> +void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo);
>> +
>>   #endif
>> diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
>> index 437f64202f3b..94ca9d033b06 100644
>> --- a/drivers/gpu/drm/xe/xe_vm_types.h
>> +++ b/drivers/gpu/drm/xe/xe_vm_types.h
>> @@ -150,6 +150,17 @@ struct xe_vma {
>>   	 */
>>   	bool skip_invalidation;
>>   
>> +	/**
>> +	 * @purgeable_state: Purgeable hint for this VMA mapping
>> +	 *
>> +	 * Per-VMA purgeable state from madvise. Valid states are WILLNEED (0)
>> +	 * or DONTNEED (1). Shared BOs require all VMAs to be DONTNEED before
>> +	 * the BO can be purged. PURGED state exists only at BO level.
>> +	 *
>> +	 * Protected by BO dma-resv lock. Set via DRM_IOCTL_XE_MADVISE.
>> +	 */
>> +	u32 purgeable_state;
>> +
> I think xe_vma_mem_attr is a better place for this field.


Noted - I will move purgeable_state into xe_vma_mem_attr for better 
organization.


Thanks,
Arvind

>
> Matt
>
>>   	/**
>>   	 * @ufence: The user fence that was provided with MAP.
>>   	 * Needs to be signalled before UNMAP can be processed.
>> -- 
>> 2.43.0
>>

Re: [PATCH v4 6/8] drm/xe/madvise: Implement per-VMA purgeable state tracking

[Thomas Hellström]

On Tue, 2026-01-20 at 09:41 -0800, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:52AM +0530, Arvind Yadav wrote:
> > Track purgeable state per-VMA instead of using a coarse shared
> > BO check. This prevents purging shared BOs until all VMAs across
> > all VMs are marked DONTNEED.
> > 
> > Add xe_bo_all_vmas_dontneed() to check all VMAs before marking
> > a BO purgeable. Add xe_bo_recheck_purgeable_on_vma_unbind() to
> > handle state transitions when VMAs are destroyed - if all
> > remaining VMAs are DONTNEED the BO can become purgeable, or if
> > no VMAs remain it transitions to WILLNEED.
> > 
> > The per-VMA purgeable_state field stores the madvise hint for
> > each mapping. Shared BOs can only be purged when all VMAs
> > unanimously indicate DONTNEED.
> > 
> > v3:
> >   - This addresses Thomas Hellström's feedback: "loop over all vmas
> >     attached to the bo and check that they all say WONTNEED. This
> > will
> >     also need a check at VMA unbinding"
> > 
> > v4:
> >   - @madv_purgeable atomic_t → u32 change across all relevant
> > patches. (Matt)
> > 
> > Cc: Matthew Brost <matthew.brost@intel.com>
> > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> > Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_vm.c         | 15 +++++-
> >  drivers/gpu/drm/xe/xe_vm_madvise.c | 84
> > +++++++++++++++++++++++++++++-
> >  drivers/gpu/drm/xe/xe_vm_madvise.h |  3 ++
> >  drivers/gpu/drm/xe/xe_vm_types.h   | 11 ++++
> >  4 files changed, 111 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_vm.c
> > b/drivers/gpu/drm/xe/xe_vm.c
> > index f250daae3012..9543960b5613 100644
> > --- a/drivers/gpu/drm/xe/xe_vm.c
> > +++ b/drivers/gpu/drm/xe/xe_vm.c
> > @@ -40,6 +40,7 @@
> >  #include "xe_tile.h"
> >  #include "xe_tlb_inval.h"
> >  #include "xe_trace_bo.h"
> > +#include "xe_vm_madvise.h"
> >  #include "xe_wa.h"
> >  
> >  static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
> > @@ -1079,12 +1080,18 @@ static struct xe_vma *xe_vma_create(struct
> > xe_vm *vm,
> >  static void xe_vma_destroy_late(struct xe_vma *vma)
> >  {
> >  	struct xe_vm *vm = xe_vma_vm(vma);
> > +	struct xe_bo *bo = NULL;
> >  
> >  	if (vma->ufence) {
> >  		xe_sync_ufence_put(vma->ufence);
> >  		vma->ufence = NULL;
> >  	}
> >  
> > +	/* Get BO reference for purgeable state re-check */
> > +	if (!xe_vma_is_userptr(vma) && !xe_vma_is_null(vma) &&
> > +	    !xe_vma_is_cpu_addr_mirror(vma))
> > +		bo = xe_vma_bo(vma);
> 
> I think xe_vma_bo just returns NULL if any of the above conditions
> are
> met, so I believe is ok to just blindly call xe_vma_bo as you have a
> NULL check on the BO below.
> 
> > +
> >  	if (xe_vma_is_userptr(vma)) {
> >  		struct xe_userptr_vma *uvma = to_userptr_vma(vma);
> >  
> > @@ -1093,7 +1100,13 @@ static void xe_vma_destroy_late(struct
> > xe_vma *vma)
> >  	} else if (xe_vma_is_null(vma) ||
> > xe_vma_is_cpu_addr_mirror(vma)) {
> >  		xe_vm_put(vm);
> >  	} else {
> > -		xe_bo_put(xe_vma_bo(vma));
> > +		/* Trylock safe for async context; madvise
> > corrects failures */
> > +		if (bo && dma_resv_trylock(bo->ttm.base.resv)) {
> > +			xe_bo_recheck_purgeable_on_vma_unbind(bo);
> 
> Also I don't think the correct place to call this.
> 
> I believe you can call this function in xe_vma_destroy after
> drm_gpuva_unlink. You also have BO lock there too so no need from
> this
> trylock path.
> 
> > +			dma_resv_unlock(bo->ttm.base.resv);
> > +		}
> > +
> > +		xe_bo_put(bo);
> >  	}
> >  
> >  	xe_vma_free(vma);
> > diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > index dfeab9e24a09..27b6ad65b314 100644
> > --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > @@ -12,6 +12,7 @@
> >  #include "xe_pat.h"
> >  #include "xe_pt.h"
> >  #include "xe_svm.h"
> > +#include "xe_vm.h"
> >  
> >  struct xe_vmas_in_madvise_range {
> >  	u64 addr;
> > @@ -179,6 +180,80 @@ static void madvise_pat_index(struct xe_device
> > *xe, struct xe_vm *vm,
> >  	}
> >  }
> >  
> > +/**
> > + * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are
> > marked DONTNEED
> > + * @bo: Buffer object
> > + *
> > + * Check all VMAs across all VMs to determine if BO can be purged.
> > + * Shared BOs require unanimous DONTNEED state from all mappings.
> > + *
> > + * Caller must hold BO dma-resv lock.
> > + *
> > + * Return: true if all VMAs are DONTNEED, false otherwise
> > + */
> > +static bool xe_bo_all_vmas_dontneed(struct xe_bo *bo)
> > +{
> > +	struct drm_gpuvm_bo *vm_bo;
> > +	struct drm_gpuva *gpuva;
> > +	struct drm_gem_object *obj = &bo->ttm.base;
> > +	bool has_vmas = false;
> > +
> > +	dma_resv_assert_held(bo->ttm.base.resv);
> > +
> > +	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
> > +		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
> > +			struct xe_vma *vma = gpuva_to_vma(gpuva);
> > +
> > +			has_vmas = true;
> > +
> > +			/* Any non-DONTNEED VMA prevents purging
> > */
> > +			if (READ_ONCE(vma->purgeable_state) !=
> > XE_MADV_PURGEABLE_DONTNEED)
> 
> You don't need the READ_ONCE as purgeable_state is only accessed
> under
> the dma-resv lock, also there isn't a WRITE_ONCE anywhere.
> 
> > +				return false;
> > +		}
> > +	}
> > +
> > +	/* No VMAs means not purgeable */
> 
> No VMAs means it is purgeable, right?

Then it would be purgeable just after creation, even if CPU is
accessing. However I think we should consider the case where a bo is
purgeable and it's single VMA goes away. Then I'd say it's still
purgeable.

I guess this is our last chance to, after discussing for years,
revert to a bo madvise UAPI before the UAPI is set in stone, given the
above corner-cases. I must admit I'm not 100% sure myself, given CPU
madvises are always based on the virtual address map, and we try to
mimic that as much as possible. 

> 
> > +	if (!has_vmas)
> > +		return false;
> > +
> > +	return true;
> > +}
> > +
> > +/**
> > + * xe_bo_recheck_purgeable_on_vma_unbind() - Re-evaluate BO
> > purgeable state after VMA unbind
> > + * @bo: Buffer object
> > + *
> > + * When a VMA is unbound, re-check if the BO's purgeable state
> > should change.
> > + * Destroyed VMAs may allow the BO to become purgeable if all
> > remaining VMAs
> > + * are DONTNEED, or require transition to WILLNEED if no VMAs
> > remain.
> > + *
> > + * Called from VMA destruction path with BO dma-resv lock held.
> > + */
> > +void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo)
> > +{
> > +	if (!bo)
> > +		return;
> > +
> > +	dma_resv_assert_held(bo->ttm.base.resv);
> > +
> > +	/*
> > +	 * Once purged, always purged. Cannot transition back to
> > WILLNEED.
> > +	 * This matches i915 semantics where purged BOs are
> > permanently invalid.
> > +	 */
> > +	if (bo->madv_purgeable == XE_MADV_PURGEABLE_PURGED)
> > +		return;

	Perhaps replace the below with

	bo->madv_purgeable = xe_bo_all_vmas_dontneed(bo) ?
		XE_MADV_PURGEABLE_DONTNEED : 		XE_MADV_PURGEABLE_WILLNEED; 
> > +
> > +	if (xe_bo_all_vmas_dontneed(bo)) {
> > +		/* All VMAs are DONTNEED - mark BO purgeable */
> > +		if (bo->madv_purgeable !=
> > XE_MADV_PURGEABLE_DONTNEED)
> > +			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_DONTNEED;
> > +	} else {
> > +		/* At least one VMA is WILLNEED - BO must not be
> > purgeable */
> > +		if (bo->madv_purgeable !=
> > XE_MADV_PURGEABLE_WILLNEED)
> > +			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_WILLNEED;
> > +	}
> > +}
> > +
> >  /*
> >   * Handle purgeable buffer object advice for
> > DONTNEED/WILLNEED/PURGED.
> >   * Returns true if any BO was purged, false otherwise.
> > @@ -213,10 +288,17 @@ static bool xe_vm_madvise_purgeable_bo(struct
> > xe_device *xe, struct xe_vm *vm,
> >  
> >  		switch (op->purge_state_val.val) {
> >  		case DRM_XE_VMA_PURGEABLE_STATE_WILLNEED:
> > +			vmas[i]->purgeable_state =
> > XE_MADV_PURGEABLE_WILLNEED;
> > +
> > +			/* Mark VMA WILLNEED - BO becomes non-
> > purgeable immediately */
> >  			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_WILLNEED;
> >  			break;
> >  		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
> > -			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_DONTNEED;
> > +			vmas[i]->purgeable_state =
> > XE_MADV_PURGEABLE_DONTNEED;
> > +
> > +			/* Mark BO purgeable only if all VMAs are
> > DONTNEED */
> > +			if (xe_bo_all_vmas_dontneed(bo))
> > +				bo->madv_purgeable =
> > XE_MADV_PURGEABLE_DONTNEED;
> >  			break;
> >  		default:
> >  			drm_warn(&vm->xe->drm, "Invalid madvice
> > value = %d\n",

This is part of the IOCTL processing, right? Then we should use
XE_IOCTL_DBG() on invalid values, or xe_assert() if input values are
already checked elsewhere.


> > diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.h
> > b/drivers/gpu/drm/xe/xe_vm_madvise.h
> > index b0e1fc445f23..61868f851949 100644
> > --- a/drivers/gpu/drm/xe/xe_vm_madvise.h
> > +++ b/drivers/gpu/drm/xe/xe_vm_madvise.h
> > @@ -8,8 +8,11 @@
> >  
> >  struct drm_device;
> >  struct drm_file;
> > +struct xe_bo;
> >  
> >  int xe_vm_madvise_ioctl(struct drm_device *dev, void *data,
> >  			struct drm_file *file);
> >  
> > +void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo);
> > +
> >  #endif
> > diff --git a/drivers/gpu/drm/xe/xe_vm_types.h
> > b/drivers/gpu/drm/xe/xe_vm_types.h
> > index 437f64202f3b..94ca9d033b06 100644
> > --- a/drivers/gpu/drm/xe/xe_vm_types.h
> > +++ b/drivers/gpu/drm/xe/xe_vm_types.h
> > @@ -150,6 +150,17 @@ struct xe_vma {
> >  	 */
> >  	bool skip_invalidation;
> >  
> > +	/**
> > +	 * @purgeable_state: Purgeable hint for this VMA mapping
> > +	 *
> > +	 * Per-VMA purgeable state from madvise. Valid states are
> > WILLNEED (0)
> > +	 * or DONTNEED (1). Shared BOs require all VMAs to be
> > DONTNEED before
> > +	 * the BO can be purged. PURGED state exists only at BO
> > level.
> > +	 *
> > +	 * Protected by BO dma-resv lock. Set via
> > DRM_IOCTL_XE_MADVISE.
> > +	 */
> > +	u32 purgeable_state;
> > +
> 
> I think xe_vma_mem_attr is a better place for this field.

+1.

/Thomas


> 
> Matt
> 
> >  	/**
> >  	 * @ufence: The user fence that was provided with MAP.
> >  	 * Needs to be signalled before UNMAP can be processed.
> > -- 
> > 2.43.0
> > 

[PATCH v4 7/8] drm/xe/madvise: Block imported and exported dma-bufs

[Arvind Yadav]

Prevent marking imported or exported dma-bufs as purgeable.
External devices may be accessing these buffers without our
knowledge, making purging unsafe.

Check drm_gem_is_imported() for buffers created by other
drivers and obj->dma_buf for buffers exported to other
drivers. Silently skip these BOs during madvise processing.

This follows drm_gem_shmem's purgeable implementation and
prevents data corruption from purging actively-used shared
buffers.

v3:
   - Addresses review feedback from Matt Roper about handling
     imported/exported BOs correctly in the purgeable BO
     implementation.

v4:
   - Check should be add to xe_vm_madvise_purgeable_bo.

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 drivers/gpu/drm/xe/xe_vm_madvise.c | 33 ++++++++++++++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
index 27b6ad65b314..5808fef89777 100644
--- a/drivers/gpu/drm/xe/xe_vm_madvise.c
+++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
@@ -180,6 +180,31 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
 	}
 }
 
+/**
+ * xe_bo_is_external_dmabuf() - Check if BO is imported or exported dma-buf
+ * @bo: Buffer object
+ *
+ * Prevent marking imported or exported dma-bufs as purgeable.
+ * External devices may be accessing these buffers without our
+ * knowledge, making purging unsafe.
+ *
+ * Return: true if BO is imported or exported, false otherwise
+ */
+static bool xe_bo_is_external_dmabuf(struct xe_bo *bo)
+{
+	struct drm_gem_object *obj = &bo->ttm.base;
+
+	/* Imported from another driver */
+	if (drm_gem_is_imported(obj))
+		return true;
+
+	/* Exported to another driver */
+	if (obj->dma_buf)
+		return true;
+
+	return false;
+}
+
 /**
  * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are marked DONTNEED
  * @bo: Buffer object
@@ -200,6 +225,10 @@ static bool xe_bo_all_vmas_dontneed(struct xe_bo *bo)
 
 	dma_resv_assert_held(bo->ttm.base.resv);
 
+	/* External dma-bufs cannot be purgeable */
+	if (xe_bo_is_external_dmabuf(bo))
+		return false;
+
 	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
 		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
 			struct xe_vma *vma = gpuva_to_vma(gpuva);
@@ -277,6 +306,10 @@ static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
 		/* BO must be locked before modifying madv state */
 		xe_bo_assert_held(bo);
 
+		/* Skip external dma-bufs */
+		if (xe_bo_is_external_dmabuf(bo))
+			continue;
+
 		/*
 		 * Once purged, always purged. Cannot transition back to WILLNEED.
 		 * This matches i915 semantics where purged BOs are permanently invalid.
-- 
2.43.0

Re: [PATCH v4 7/8] drm/xe/madvise: Block imported and exported dma-bufs

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:53AM +0530, Arvind Yadav wrote:
> Prevent marking imported or exported dma-bufs as purgeable.
> External devices may be accessing these buffers without our
> knowledge, making purging unsafe.
> 
> Check drm_gem_is_imported() for buffers created by other
> drivers and obj->dma_buf for buffers exported to other
> drivers. Silently skip these BOs during madvise processing.
> 
> This follows drm_gem_shmem's purgeable implementation and
> prevents data corruption from purging actively-used shared
> buffers.
> 
> v3:
>    - Addresses review feedback from Matt Roper about handling
>      imported/exported BOs correctly in the purgeable BO
>      implementation.
> 
> v4:
>    - Check should be add to xe_vm_madvise_purgeable_bo.
> 
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>

@Thomas - couple questions below here I need a 2nd opinion on.

> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_vm_madvise.c | 33 ++++++++++++++++++++++++++++++
>  1 file changed, 33 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
> index 27b6ad65b314..5808fef89777 100644
> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> @@ -180,6 +180,31 @@ static void madvise_pat_index(struct xe_device *xe, struct xe_vm *vm,
>  	}
>  }
>  
> +/**
> + * xe_bo_is_external_dmabuf() - Check if BO is imported or exported dma-buf
> + * @bo: Buffer object
> + *
> + * Prevent marking imported or exported dma-bufs as purgeable.
> + * External devices may be accessing these buffers without our
> + * knowledge, making purging unsafe.
> + *
> + * Return: true if BO is imported or exported, false otherwise
> + */
> +static bool xe_bo_is_external_dmabuf(struct xe_bo *bo)
> +{

@Thomas

Should we have this check more generic? e.g., if a BO is not tied to a
VM, we don't allow purablity to be set?

> +	struct drm_gem_object *obj = &bo->ttm.base;
> +
> +	/* Imported from another driver */
> +	if (drm_gem_is_imported(obj))
> +		return true;
> +
> +	/* Exported to another driver */
> +	if (obj->dma_buf)
> +		return true;
> +
> +	return false;
> +}
> +
>  /**
>   * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are marked DONTNEED
>   * @bo: Buffer object
> @@ -200,6 +225,10 @@ static bool xe_bo_all_vmas_dontneed(struct xe_bo *bo)
>  
>  	dma_resv_assert_held(bo->ttm.base.resv);
>  
> +	/* External dma-bufs cannot be purgeable */
> +	if (xe_bo_is_external_dmabuf(bo))
> +		return false;
> +
>  	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
>  		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
>  			struct xe_vma *vma = gpuva_to_vma(gpuva);
> @@ -277,6 +306,10 @@ static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
>  		/* BO must be locked before modifying madv state */
>  		xe_bo_assert_held(bo);
>  
> +		/* Skip external dma-bufs */
> +		if (xe_bo_is_external_dmabuf(bo))
> +			continue;

@Thomas

I think instead of silently continuing here we fail the IOCTL with
-EINVAL?

What do you think?

Matt

> +
>  		/*
>  		 * Once purged, always purged. Cannot transition back to WILLNEED.
>  		 * This matches i915 semantics where purged BOs are permanently invalid.
> -- 
> 2.43.0
> 

Re: [PATCH v4 7/8] drm/xe/madvise: Block imported and exported dma-bufs

[Thomas Hellström]

On Tue, 2026-01-20 at 09:51 -0800, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:53AM +0530, Arvind Yadav wrote:
> > Prevent marking imported or exported dma-bufs as purgeable.
> > External devices may be accessing these buffers without our
> > knowledge, making purging unsafe.
> > 
> > Check drm_gem_is_imported() for buffers created by other
> > drivers and obj->dma_buf for buffers exported to other
> > drivers. Silently skip these BOs during madvise processing.
> > 
> > This follows drm_gem_shmem's purgeable implementation and
> > prevents data corruption from purging actively-used shared
> > buffers.
> > 
> > v3:
> >    - Addresses review feedback from Matt Roper about handling
> >      imported/exported BOs correctly in the purgeable BO
> >      implementation.
> > 
> > v4:
> >    - Check should be add to xe_vm_madvise_purgeable_bo.
> > 
> > Cc: Matthew Brost <matthew.brost@intel.com>
> > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> 
> @Thomas - couple questions below here I need a 2nd opinion on.
> 
> > Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> > Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_vm_madvise.c | 33
> > ++++++++++++++++++++++++++++++
> >  1 file changed, 33 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > index 27b6ad65b314..5808fef89777 100644
> > --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > @@ -180,6 +180,31 @@ static void madvise_pat_index(struct xe_device
> > *xe, struct xe_vm *vm,
> >  	}
> >  }
> >  
> > +/**
> > + * xe_bo_is_external_dmabuf() - Check if BO is imported or
> > exported dma-buf
> > + * @bo: Buffer object
> > + *
> > + * Prevent marking imported or exported dma-bufs as purgeable.
> > + * External devices may be accessing these buffers without our
> > + * knowledge, making purging unsafe.
> > + *
> > + * Return: true if BO is imported or exported, false otherwise
> > + */
> > +static bool xe_bo_is_external_dmabuf(struct xe_bo *bo)
> > +{
> 
> @Thomas
> 
> Should we have this check more generic? e.g., if a BO is not tied to
> a
> VM, we don't allow purablity to be set?

I think if NEO is going to implement this in one form or another, we
need to support also external bos as long as they're not exported,
since NEO refuses to use local bos.

> 
> > +	struct drm_gem_object *obj = &bo->ttm.base;
> > +
> > +	/* Imported from another driver */
> > +	if (drm_gem_is_imported(obj))
> > +		return true;
> > +
> > +	/* Exported to another driver */
> > +	if (obj->dma_buf)
> > +		return true;
> > +
> > +	return false;
> > +}
> > +
> >  /**
> >   * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are
> > marked DONTNEED
> >   * @bo: Buffer object
> > @@ -200,6 +225,10 @@ static bool xe_bo_all_vmas_dontneed(struct
> > xe_bo *bo)
> >  
> >  	dma_resv_assert_held(bo->ttm.base.resv);
> >  
> > +	/* External dma-bufs cannot be purgeable */
> > +	if (xe_bo_is_external_dmabuf(bo))
> > +		return false;
> > +
> >  	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
> >  		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
> >  			struct xe_vma *vma = gpuva_to_vma(gpuva);
> > @@ -277,6 +306,10 @@ static bool xe_vm_madvise_purgeable_bo(struct
> > xe_device *xe, struct xe_vm *vm,
> >  		/* BO must be locked before modifying madv state
> > */
> >  		xe_bo_assert_held(bo);
> >  
> > +		/* Skip external dma-bufs */
> > +		if (xe_bo_is_external_dmabuf(bo))
> > +			continue;
> 
> @Thomas
> 
> I think instead of silently continuing here we fail the IOCTL with
> -EINVAL?
> 
> What do you think?

Hmm. If we view the export as yet another map, then IMO silently
continuing would probably be the best option and consistent with
sharing across VMs.

Finally I wonder if we should update the purgeable state on final dma-
buf attach.

Floating an idea here:

Could the code be simplified if we maintain a map_count and a
purgeable_count on each bo? Bo is purgeable iff map_count ==
purgeable_count. vmas and dma-buf attachments are included in
map_count.

Thanks,
Thomas


> 
> Matt
> 
> > +
> >  		/*
> >  		 * Once purged, always purged. Cannot transition
> > back to WILLNEED.
> >  		 * This matches i915 semantics where purged BOs
> > are permanently invalid.
> > -- 
> > 2.43.0
> > 

Re: [PATCH v4 7/8] drm/xe/madvise: Block imported and exported dma-bufs

[Yadav, Arvind]

On 23-01-2026 19:01, Thomas Hellström wrote:
> On Tue, 2026-01-20 at 09:51 -0800, Matthew Brost wrote:
>> On Tue, Jan 20, 2026 at 11:38:53AM +0530, Arvind Yadav wrote:
>>> Prevent marking imported or exported dma-bufs as purgeable.
>>> External devices may be accessing these buffers without our
>>> knowledge, making purging unsafe.
>>>
>>> Check drm_gem_is_imported() for buffers created by other
>>> drivers and obj->dma_buf for buffers exported to other
>>> drivers. Silently skip these BOs during madvise processing.
>>>
>>> This follows drm_gem_shmem's purgeable implementation and
>>> prevents data corruption from purging actively-used shared
>>> buffers.
>>>
>>> v3:
>>>     - Addresses review feedback from Matt Roper about handling
>>>       imported/exported BOs correctly in the purgeable BO
>>>       implementation.
>>>
>>> v4:
>>>     - Check should be add to xe_vm_madvise_purgeable_bo.
>>>
>>> Cc: Matthew Brost <matthew.brost@intel.com>
>>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>> @Thomas - couple questions below here I need a 2nd opinion on.
>>
>>> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
>>> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
>>> ---
>>>   drivers/gpu/drm/xe/xe_vm_madvise.c | 33
>>> ++++++++++++++++++++++++++++++
>>>   1 file changed, 33 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> b/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> index 27b6ad65b314..5808fef89777 100644
>>> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
>>> @@ -180,6 +180,31 @@ static void madvise_pat_index(struct xe_device
>>> *xe, struct xe_vm *vm,
>>>   	}
>>>   }
>>>   
>>> +/**
>>> + * xe_bo_is_external_dmabuf() - Check if BO is imported or
>>> exported dma-buf
>>> + * @bo: Buffer object
>>> + *
>>> + * Prevent marking imported or exported dma-bufs as purgeable.
>>> + * External devices may be accessing these buffers without our
>>> + * knowledge, making purging unsafe.
>>> + *
>>> + * Return: true if BO is imported or exported, false otherwise
>>> + */
>>> +static bool xe_bo_is_external_dmabuf(struct xe_bo *bo)
>>> +{
>> @Thomas
>>
>> Should we have this check more generic? e.g., if a BO is not tied to
>> a
>> VM, we don't allow purablity to be set?
> I think if NEO is going to implement this in one form or another, we
> need to support also external bos as long as they're not exported,
> since NEO refuses to use local bos.


You're right, and I believe the current implementation should support that.

To clarify: "external" here refers only to dma-buf sharing (imported or
exported), not to placement. System-memory BOs that NEO typically uses
remain fully eligible for purging.

The xe_bo_is_dmabuf_shared() check specifically targets:
   - Imported dma-bufs (we don't own backing store)
   - Exported dma-bufs (external devices may have active mappings)

Regular Xe-owned system BOs can be marked DONTNEED and purged under memory
pressure. We only skip cases where the BO is shared via dma-buf, following
the same pattern as drm_gem_shmem_is_purgeable().

Please let me know if I've misunderstood the NEO requirements.


>>> +	struct drm_gem_object *obj = &bo->ttm.base;
>>> +
>>> +	/* Imported from another driver */
>>> +	if (drm_gem_is_imported(obj))
>>> +		return true;
>>> +
>>> +	/* Exported to another driver */
>>> +	if (obj->dma_buf)
>>> +		return true;
>>> +
>>> +	return false;
>>> +}
>>> +
>>>   /**
>>>    * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are
>>> marked DONTNEED
>>>    * @bo: Buffer object
>>> @@ -200,6 +225,10 @@ static bool xe_bo_all_vmas_dontneed(struct
>>> xe_bo *bo)
>>>   
>>>   	dma_resv_assert_held(bo->ttm.base.resv);
>>>   
>>> +	/* External dma-bufs cannot be purgeable */
>>> +	if (xe_bo_is_external_dmabuf(bo))
>>> +		return false;
>>> +
>>>   	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
>>>   		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
>>>   			struct xe_vma *vma = gpuva_to_vma(gpuva);
>>> @@ -277,6 +306,10 @@ static bool xe_vm_madvise_purgeable_bo(struct
>>> xe_device *xe, struct xe_vm *vm,
>>>   		/* BO must be locked before modifying madv state
>>> */
>>>   		xe_bo_assert_held(bo);
>>>   
>>> +		/* Skip external dma-bufs */
>>> +		if (xe_bo_is_external_dmabuf(bo))
>>> +			continue;
>> @Thomas
>>
>> I think instead of silently continuing here we fail the IOCTL with
>> -EINVAL?
>>
>> What do you think?
> Hmm. If we view the export as yet another map, then IMO silently
> continuing would probably be the best option and consistent with
> sharing across VMs.
Noted, I'll keep the silent skip behavior for dma-buf shared BOs to 
match the "unanimous DONTNEED" model.
>
> Finally I wonder if we should update the purgeable state on final dma-
> buf attach.


Good point. Currently the check only happens at madvise time, so there's 
a window where a BO marked DONTNEED could later be exported.
Would it make sense to hook into the export path to force transition to 
WILLNEED? That would make the "export as implicit mapping" model complete,
though it adds complexity. Let me know your preference.

>
> Floating an idea here:
>
> Could the code be simplified if we maintain a map_count and a
> purgeable_count on each bo? Bo is purgeable iff map_count ==
> purgeable_count. vmas and dma-buf attachments are included in
> map_count.


That's good idea and would avoid VMA iteration. The tradeoff is 
maintaining correct counts across all VMA and dma-buf lifecycle paths.

For this series, would you prefer I pursue the counter-based refactor 
now, or is the current iteration approach acceptable
with counters as a future optimization? I'm happy to go either direction 
based on your feedback.


Thanks,
Arvind

>
> Thanks,
> Thomas
>
>
>> Matt
>>
>>> +
>>>   		/*
>>>   		 * Once purged, always purged. Cannot transition
>>> back to WILLNEED.
>>>   		 * This matches i915 semantics where purged BOs
>>> are permanently invalid.
>>> -- 
>>> 2.43.0
>>>

Re: [PATCH v4 7/8] drm/xe/madvise: Block imported and exported dma-bufs

[Thomas Hellström]

On Fri, 2026-01-30 at 13:52 +0530, Yadav, Arvind wrote:
> 
> On 23-01-2026 19:01, Thomas Hellström wrote:
> > On Tue, 2026-01-20 at 09:51 -0800, Matthew Brost wrote:
> > > On Tue, Jan 20, 2026 at 11:38:53AM +0530, Arvind Yadav wrote:
> > > > Prevent marking imported or exported dma-bufs as purgeable.
> > > > External devices may be accessing these buffers without our
> > > > knowledge, making purging unsafe.
> > > > 
> > > > Check drm_gem_is_imported() for buffers created by other
> > > > drivers and obj->dma_buf for buffers exported to other
> > > > drivers. Silently skip these BOs during madvise processing.
> > > > 
> > > > This follows drm_gem_shmem's purgeable implementation and
> > > > prevents data corruption from purging actively-used shared
> > > > buffers.
> > > > 
> > > > v3:
> > > >     - Addresses review feedback from Matt Roper about handling
> > > >       imported/exported BOs correctly in the purgeable BO
> > > >       implementation.
> > > > 
> > > > v4:
> > > >     - Check should be add to xe_vm_madvise_purgeable_bo.
> > > > 
> > > > Cc: Matthew Brost <matthew.brost@intel.com>
> > > > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > > @Thomas - couple questions below here I need a 2nd opinion on.
> > > 
> > > > Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> > > > Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> > > > ---
> > > >   drivers/gpu/drm/xe/xe_vm_madvise.c | 33
> > > > ++++++++++++++++++++++++++++++
> > > >   1 file changed, 33 insertions(+)
> > > > 
> > > > diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > > > b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > > > index 27b6ad65b314..5808fef89777 100644
> > > > --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > > > +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > > > @@ -180,6 +180,31 @@ static void madvise_pat_index(struct
> > > > xe_device
> > > > *xe, struct xe_vm *vm,
> > > >   	}
> > > >   }
> > > >   
> > > > +/**
> > > > + * xe_bo_is_external_dmabuf() - Check if BO is imported or
> > > > exported dma-buf
> > > > + * @bo: Buffer object
> > > > + *
> > > > + * Prevent marking imported or exported dma-bufs as purgeable.
> > > > + * External devices may be accessing these buffers without our
> > > > + * knowledge, making purging unsafe.
> > > > + *
> > > > + * Return: true if BO is imported or exported, false otherwise
> > > > + */
> > > > +static bool xe_bo_is_external_dmabuf(struct xe_bo *bo)
> > > > +{
> > > @Thomas
> > > 
> > > Should we have this check more generic? e.g., if a BO is not tied
> > > to
> > > a
> > > VM, we don't allow purablity to be set?
> > I think if NEO is going to implement this in one form or another,
> > we
> > need to support also external bos as long as they're not exported,
> > since NEO refuses to use local bos.
> 
> 
> You're right, and I believe the current implementation should support
> that.
> 
> To clarify: "external" here refers only to dma-buf sharing (imported
> or
> exported), not to placement. System-memory BOs that NEO typically
> uses
> remain fully eligible for purging.
> 
> The xe_bo_is_dmabuf_shared() check specifically targets:
>    - Imported dma-bufs (we don't own backing store)
>    - Exported dma-bufs (external devices may have active mappings)
> 
> Regular Xe-owned system BOs can be marked DONTNEED and purged under
> memory
> pressure. We only skip cases where the BO is shared via dma-buf,
> following
> the same pattern as drm_gem_shmem_is_purgeable().
> 
> Please let me know if I've misunderstood the NEO requirements.

No it sounds right. We have
*local BOs, those that Matt refers to as tied to a VM. They are not
sharable.
*external BOs. Those may or may not be exported as dma-bufs. NEO uses
only these, and they need to be purgeable as well, as long as they are
not exported or imported dma-bufs.

Thanks,
Thomas



> 
> 
> > > > +	struct drm_gem_object *obj = &bo->ttm.base;
> > > > +
> > > > +	/* Imported from another driver */
> > > > +	if (drm_gem_is_imported(obj))
> > > > +		return true;
> > > > +
> > > > +	/* Exported to another driver */
> > > > +	if (obj->dma_buf)
> > > > +		return true;
> > > > +
> > > > +	return false;
> > > > +}
> > > > +
> > > >   /**
> > > >    * xe_bo_all_vmas_dontneed() - Check if all VMAs of a BO are
> > > > marked DONTNEED
> > > >    * @bo: Buffer object
> > > > @@ -200,6 +225,10 @@ static bool xe_bo_all_vmas_dontneed(struct
> > > > xe_bo *bo)
> > > >   
> > > >   	dma_resv_assert_held(bo->ttm.base.resv);
> > > >   
> > > > +	/* External dma-bufs cannot be purgeable */
> > > > +	if (xe_bo_is_external_dmabuf(bo))
> > > > +		return false;
> > > > +
> > > >   	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
> > > >   		drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
> > > >   			struct xe_vma *vma =
> > > > gpuva_to_vma(gpuva);
> > > > @@ -277,6 +306,10 @@ static bool
> > > > xe_vm_madvise_purgeable_bo(struct
> > > > xe_device *xe, struct xe_vm *vm,
> > > >   		/* BO must be locked before modifying madv
> > > > state
> > > > */
> > > >   		xe_bo_assert_held(bo);
> > > >   
> > > > +		/* Skip external dma-bufs */
> > > > +		if (xe_bo_is_external_dmabuf(bo))
> > > > +			continue;
> > > @Thomas
> > > 
> > > I think instead of silently continuing here we fail the IOCTL
> > > with
> > > -EINVAL?
> > > 
> > > What do you think?
> > Hmm. If we view the export as yet another map, then IMO silently
> > continuing would probably be the best option and consistent with
> > sharing across VMs.
> Noted, I'll keep the silent skip behavior for dma-buf shared BOs to 
> match the "unanimous DONTNEED" model.
> > 
> > Finally I wonder if we should update the purgeable state on final
> > dma-
> > buf attach.
> 
> 
> Good point. Currently the check only happens at madvise time, so
> there's 
> a window where a BO marked DONTNEED could later be exported.
> Would it make sense to hook into the export path to force transition
> to 
> WILLNEED? That would make the "export as implicit mapping" model
> complete,
> though it adds complexity. Let me know your preference.
> 
> > 
> > Floating an idea here:
> > 
> > Could the code be simplified if we maintain a map_count and a
> > purgeable_count on each bo? Bo is purgeable iff map_count ==
> > purgeable_count. vmas and dma-buf attachments are included in
> > map_count.
> 
> 
> That's good idea and would avoid VMA iteration. The tradeoff is 
> maintaining correct counts across all VMA and dma-buf lifecycle
> paths.
> 
> For this series, would you prefer I pursue the counter-based refactor
> now, or is the current iteration approach acceptable
> with counters as a future optimization? I'm happy to go either
> direction 
> based on your feedback.
> 
> 
> Thanks,
> Arvind
> 
> > 
> > Thanks,
> > Thomas
> > 
> > 
> > > Matt
> > > 
> > > > +
> > > >   		/*
> > > >   		 * Once purged, always purged. Cannot
> > > > transition
> > > > back to WILLNEED.
> > > >   		 * This matches i915 semantics where purged
> > > > BOs
> > > > are permanently invalid.
> > > > -- 
> > > > 2.43.0
> > > > 

[PATCH v4 8/8] drm/xe/bo: Add purgeable shrinker state helpers

[Arvind Yadav]

Encapsulate TTM purgeable flag updates and shrinker page accounting
into helper functions. This prevents desynchronization between the
TTM tt->purgeable flag and the shrinker's page bucket counters.

Without these helpers, direct manipulation of xe_ttm_tt->purgeable
risks forgetting to update the corresponding shrinker counters,
leading to incorrect memory pressure calculations.

Add xe_bo_set_purgeable_shrinker() and xe_bo_clear_purgeable_shrinker()
which atomically update both the TTM flag and transfer pages between
the shrinkable and purgeable buckets.

v4:
  - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 drivers/gpu/drm/xe/xe_bo.c         | 60 ++++++++++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_bo.h         |  3 ++
 drivers/gpu/drm/xe/xe_vm_madvise.c | 13 +++++--
 3 files changed, 73 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index cc547915161d..2b1448ea3aed 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -836,6 +836,66 @@ static int xe_bo_move_notify(struct xe_bo *bo,
 	return 0;
 }
 
+/**
+ * xe_bo_set_purgeable_shrinker() - Mark BO purgeable and update shrinker
+ * @bo: Buffer object
+ *
+ * Transfers pages from shrinkable to purgeable bucket. Shrinker can now
+ * discard pages immediately without swapping. Caller holds BO lock.
+ */
+void xe_bo_set_purgeable_shrinker(struct xe_bo *bo)
+{
+	struct ttm_buffer_object *ttm_bo = &bo->ttm;
+	struct ttm_tt *tt = ttm_bo->ttm;
+	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
+	struct xe_ttm_tt *xe_tt;
+
+	xe_bo_assert_held(bo);
+
+	if (!tt || !ttm_tt_is_populated(tt))
+		return;
+
+	xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
+
+	if (!xe_tt->purgeable) {
+		xe_tt->purgeable = true;
+		/* Transfer pages from shrinkable to purgeable count */
+		xe_shrinker_mod_pages(xe->mem.shrinker,
+				      -(long)tt->num_pages,
+				      tt->num_pages);
+	}
+}
+
+/**
+ * xe_bo_clear_purgeable_shrinker() - Mark BO non-purgeable and update shrinker
+ * @bo: Buffer object
+ *
+ * Transfers pages from purgeable to shrinkable bucket. Shrinker must now
+ * swap pages instead of discarding. Caller holds BO lock.
+ */
+void xe_bo_clear_purgeable_shrinker(struct xe_bo *bo)
+{
+	struct ttm_buffer_object *ttm_bo = &bo->ttm;
+	struct ttm_tt *tt = ttm_bo->ttm;
+	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
+	struct xe_ttm_tt *xe_tt;
+
+	xe_bo_assert_held(bo);
+
+	if (!tt || !ttm_tt_is_populated(tt))
+		return;
+
+	xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
+
+	if (xe_tt->purgeable) {
+		xe_tt->purgeable = false;
+		/* Transfer pages from purgeable to shrinkable count */
+		xe_shrinker_mod_pages(xe->mem.shrinker,
+				      tt->num_pages,
+				      -(long)tt->num_pages);
+	}
+}
+
 /**
  * xe_ttm_bo_purge() - Purge buffer object backing store
  * @ttm_bo: The TTM buffer object to purge
diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
index 00e93b3065c9..681495e905af 100644
--- a/drivers/gpu/drm/xe/xe_bo.h
+++ b/drivers/gpu/drm/xe/xe_bo.h
@@ -270,6 +270,9 @@ static inline bool xe_bo_madv_is_dontneed(struct xe_bo *bo)
 	return bo->madv_purgeable == XE_MADV_PURGEABLE_DONTNEED;
 }
 
+void xe_bo_set_purgeable_shrinker(struct xe_bo *bo);
+void xe_bo_clear_purgeable_shrinker(struct xe_bo *bo);
+
 static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo)
 {
 	if (likely(bo)) {
diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
index 5808fef89777..0fb07a1ed3ae 100644
--- a/drivers/gpu/drm/xe/xe_vm_madvise.c
+++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
@@ -274,12 +274,16 @@ void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo)
 
 	if (xe_bo_all_vmas_dontneed(bo)) {
 		/* All VMAs are DONTNEED - mark BO purgeable */
-		if (bo->madv_purgeable != XE_MADV_PURGEABLE_DONTNEED)
+		if (bo->madv_purgeable != XE_MADV_PURGEABLE_DONTNEED) {
 			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
+			xe_bo_set_purgeable_shrinker(bo);
+		}
 	} else {
 		/* At least one VMA is WILLNEED - BO must not be purgeable */
-		if (bo->madv_purgeable != XE_MADV_PURGEABLE_WILLNEED)
+		if (bo->madv_purgeable != XE_MADV_PURGEABLE_WILLNEED) {
 			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
+			xe_bo_clear_purgeable_shrinker(bo);
+		}
 	}
 }
 
@@ -325,13 +329,16 @@ static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
 
 			/* Mark VMA WILLNEED - BO becomes non-purgeable immediately */
 			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
+			xe_bo_clear_purgeable_shrinker(bo);
 			break;
 		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
 			vmas[i]->purgeable_state = XE_MADV_PURGEABLE_DONTNEED;
 
 			/* Mark BO purgeable only if all VMAs are DONTNEED */
-			if (xe_bo_all_vmas_dontneed(bo))
+			if (xe_bo_all_vmas_dontneed(bo)) {
 				bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
+				xe_bo_set_purgeable_shrinker(bo);
+			}
 			break;
 		default:
 			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
-- 
2.43.0

Re: [PATCH v4 8/8] drm/xe/bo: Add purgeable shrinker state helpers

[Matthew Brost]

On Tue, Jan 20, 2026 at 11:38:54AM +0530, Arvind Yadav wrote:
> Encapsulate TTM purgeable flag updates and shrinker page accounting
> into helper functions. This prevents desynchronization between the
> TTM tt->purgeable flag and the shrinker's page bucket counters.
> 
> Without these helpers, direct manipulation of xe_ttm_tt->purgeable
> risks forgetting to update the corresponding shrinker counters,
> leading to incorrect memory pressure calculations.
> 
> Add xe_bo_set_purgeable_shrinker() and xe_bo_clear_purgeable_shrinker()
> which atomically update both the TTM flag and transfer pages between
> the shrinkable and purgeable buckets.
> 
> v4:
>   - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
> 
> Cc: Matthew Brost <matthew.brost@intel.com>

I think this patch is right but best to double check with Thomas as he
wrote the shrinker logic and is the expert here.

Matt

> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_bo.c         | 60 ++++++++++++++++++++++++++++++
>  drivers/gpu/drm/xe/xe_bo.h         |  3 ++
>  drivers/gpu/drm/xe/xe_vm_madvise.c | 13 +++++--
>  3 files changed, 73 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index cc547915161d..2b1448ea3aed 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -836,6 +836,66 @@ static int xe_bo_move_notify(struct xe_bo *bo,
>  	return 0;
>  }
>  
> +/**
> + * xe_bo_set_purgeable_shrinker() - Mark BO purgeable and update shrinker
> + * @bo: Buffer object
> + *
> + * Transfers pages from shrinkable to purgeable bucket. Shrinker can now
> + * discard pages immediately without swapping. Caller holds BO lock.
> + */
> +void xe_bo_set_purgeable_shrinker(struct xe_bo *bo)
> +{
> +	struct ttm_buffer_object *ttm_bo = &bo->ttm;
> +	struct ttm_tt *tt = ttm_bo->ttm;
> +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> +	struct xe_ttm_tt *xe_tt;
> +
> +	xe_bo_assert_held(bo);
> +
> +	if (!tt || !ttm_tt_is_populated(tt))
> +		return;
> +
> +	xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
> +
> +	if (!xe_tt->purgeable) {
> +		xe_tt->purgeable = true;
> +		/* Transfer pages from shrinkable to purgeable count */
> +		xe_shrinker_mod_pages(xe->mem.shrinker,
> +				      -(long)tt->num_pages,
> +				      tt->num_pages);
> +	}
> +}
> +
> +/**
> + * xe_bo_clear_purgeable_shrinker() - Mark BO non-purgeable and update shrinker
> + * @bo: Buffer object
> + *
> + * Transfers pages from purgeable to shrinkable bucket. Shrinker must now
> + * swap pages instead of discarding. Caller holds BO lock.
> + */
> +void xe_bo_clear_purgeable_shrinker(struct xe_bo *bo)
> +{
> +	struct ttm_buffer_object *ttm_bo = &bo->ttm;
> +	struct ttm_tt *tt = ttm_bo->ttm;
> +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> +	struct xe_ttm_tt *xe_tt;
> +
> +	xe_bo_assert_held(bo);
> +
> +	if (!tt || !ttm_tt_is_populated(tt))
> +		return;
> +
> +	xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
> +
> +	if (xe_tt->purgeable) {
> +		xe_tt->purgeable = false;
> +		/* Transfer pages from purgeable to shrinkable count */
> +		xe_shrinker_mod_pages(xe->mem.shrinker,
> +				      tt->num_pages,
> +				      -(long)tt->num_pages);
> +	}
> +}
> +
>  /**
>   * xe_ttm_bo_purge() - Purge buffer object backing store
>   * @ttm_bo: The TTM buffer object to purge
> diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
> index 00e93b3065c9..681495e905af 100644
> --- a/drivers/gpu/drm/xe/xe_bo.h
> +++ b/drivers/gpu/drm/xe/xe_bo.h
> @@ -270,6 +270,9 @@ static inline bool xe_bo_madv_is_dontneed(struct xe_bo *bo)
>  	return bo->madv_purgeable == XE_MADV_PURGEABLE_DONTNEED;
>  }
>  
> +void xe_bo_set_purgeable_shrinker(struct xe_bo *bo);
> +void xe_bo_clear_purgeable_shrinker(struct xe_bo *bo);
> +
>  static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo)
>  {
>  	if (likely(bo)) {
> diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
> index 5808fef89777..0fb07a1ed3ae 100644
> --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> @@ -274,12 +274,16 @@ void xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo)
>  
>  	if (xe_bo_all_vmas_dontneed(bo)) {
>  		/* All VMAs are DONTNEED - mark BO purgeable */
> -		if (bo->madv_purgeable != XE_MADV_PURGEABLE_DONTNEED)
> +		if (bo->madv_purgeable != XE_MADV_PURGEABLE_DONTNEED) {
>  			bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
> +			xe_bo_set_purgeable_shrinker(bo);
> +		}
>  	} else {
>  		/* At least one VMA is WILLNEED - BO must not be purgeable */
> -		if (bo->madv_purgeable != XE_MADV_PURGEABLE_WILLNEED)
> +		if (bo->madv_purgeable != XE_MADV_PURGEABLE_WILLNEED) {
>  			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> +			xe_bo_clear_purgeable_shrinker(bo);
> +		}
>  	}
>  }
>  
> @@ -325,13 +329,16 @@ static bool xe_vm_madvise_purgeable_bo(struct xe_device *xe, struct xe_vm *vm,
>  
>  			/* Mark VMA WILLNEED - BO becomes non-purgeable immediately */
>  			bo->madv_purgeable = XE_MADV_PURGEABLE_WILLNEED;
> +			xe_bo_clear_purgeable_shrinker(bo);
>  			break;
>  		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
>  			vmas[i]->purgeable_state = XE_MADV_PURGEABLE_DONTNEED;
>  
>  			/* Mark BO purgeable only if all VMAs are DONTNEED */
> -			if (xe_bo_all_vmas_dontneed(bo))
> +			if (xe_bo_all_vmas_dontneed(bo)) {
>  				bo->madv_purgeable = XE_MADV_PURGEABLE_DONTNEED;
> +				xe_bo_set_purgeable_shrinker(bo);
> +			}
>  			break;
>  		default:
>  			drm_warn(&vm->xe->drm, "Invalid madvice value = %d\n",
> -- 
> 2.43.0
> 

Re: [PATCH v4 8/8] drm/xe/bo: Add purgeable shrinker state helpers

[Thomas Hellström]

On Tue, 2026-01-20 at 09:58 -0800, Matthew Brost wrote:
> On Tue, Jan 20, 2026 at 11:38:54AM +0530, Arvind Yadav wrote:
> > Encapsulate TTM purgeable flag updates and shrinker page accounting
> > into helper functions. This prevents desynchronization between the
> > TTM tt->purgeable flag and the shrinker's page bucket counters.
> > 
> > Without these helpers, direct manipulation of xe_ttm_tt->purgeable
> > risks forgetting to update the corresponding shrinker counters,
> > leading to incorrect memory pressure calculations.
> > 
> > Add xe_bo_set_purgeable_shrinker() and
> > xe_bo_clear_purgeable_shrinker()
> > which atomically update both the TTM flag and transfer pages
> > between
> > the shrinkable and purgeable buckets.
> > 
> > v4:
> >   - @madv_purgeable atomic_t → u32 change across all relevant
> > patches. (Matt)
> > 
> > Cc: Matthew Brost <matthew.brost@intel.com>
> 
> I think this patch is right but best to double check with Thomas as
> he
> wrote the shrinker logic and is the expert here.
> 
> Matt


Looks correct to me.

@Arvind: Did you look at removing the xe_tt purgeable field (which I
always thought of as an intermediate until we had purgeable bos) and
replace it with the bo->madv_purgeable field in the shrinker code. I'm
not sure how complicated it would be, TBH but feel free to leave that
as a follow-up patch.

Also make sure the xe_bo_shrink xe_live_ktest subtest still works since
it relies heavily on shrinkable bos. I don't think CI runs it because
of the execution time and insufficient memory allocation amount
heuristics. We should also file a jira to rewrite that test to a user-
space IGT test now that we support shrinkable bos. I'll take a look at
that.

Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

> 
> > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> > Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_bo.c         | 60
> > ++++++++++++++++++++++++++++++
> >  drivers/gpu/drm/xe/xe_bo.h         |  3 ++
> >  drivers/gpu/drm/xe/xe_vm_madvise.c | 13 +++++--
> >  3 files changed, 73 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c
> > b/drivers/gpu/drm/xe/xe_bo.c
> > index cc547915161d..2b1448ea3aed 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > @@ -836,6 +836,66 @@ static int xe_bo_move_notify(struct xe_bo *bo,
> >  	return 0;
> >  }
> >  
> > +/**
> > + * xe_bo_set_purgeable_shrinker() - Mark BO purgeable and update
> > shrinker
> > + * @bo: Buffer object
> > + *
> > + * Transfers pages from shrinkable to purgeable bucket. Shrinker
> > can now
> > + * discard pages immediately without swapping. Caller holds BO
> > lock.
> > + */
> > +void xe_bo_set_purgeable_shrinker(struct xe_bo *bo)
> > +{
> > +	struct ttm_buffer_object *ttm_bo = &bo->ttm;
> > +	struct ttm_tt *tt = ttm_bo->ttm;
> > +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> > +	struct xe_ttm_tt *xe_tt;
> > +
> > +	xe_bo_assert_held(bo);
> > +
> > +	if (!tt || !ttm_tt_is_populated(tt))
> > +		return;
> > +
> > +	xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
> > +
> > +	if (!xe_tt->purgeable) {
> > +		xe_tt->purgeable = true;
> > +		/* Transfer pages from shrinkable to purgeable
> > count */
> > +		xe_shrinker_mod_pages(xe->mem.shrinker,
> > +				      -(long)tt->num_pages,
> > +				      tt->num_pages);
> > +	}
> > +}
> > +
> > +/**
> > + * xe_bo_clear_purgeable_shrinker() - Mark BO non-purgeable and
> > update shrinker
> > + * @bo: Buffer object
> > + *
> > + * Transfers pages from purgeable to shrinkable bucket. Shrinker
> > must now
> > + * swap pages instead of discarding. Caller holds BO lock.
> > + */
> > +void xe_bo_clear_purgeable_shrinker(struct xe_bo *bo)
> > +{
> > +	struct ttm_buffer_object *ttm_bo = &bo->ttm;
> > +	struct ttm_tt *tt = ttm_bo->ttm;
> > +	struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
> > +	struct xe_ttm_tt *xe_tt;
> > +
> > +	xe_bo_assert_held(bo);
> > +
> > +	if (!tt || !ttm_tt_is_populated(tt))
> > +		return;
> > +
> > +	xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
> > +
> > +	if (xe_tt->purgeable) {
> > +		xe_tt->purgeable = false;
> > +		/* Transfer pages from purgeable to shrinkable
> > count */
> > +		xe_shrinker_mod_pages(xe->mem.shrinker,
> > +				      tt->num_pages,
> > +				      -(long)tt->num_pages);
> > +	}
> > +}
> > +
> >  /**
> >   * xe_ttm_bo_purge() - Purge buffer object backing store
> >   * @ttm_bo: The TTM buffer object to purge
> > diff --git a/drivers/gpu/drm/xe/xe_bo.h
> > b/drivers/gpu/drm/xe/xe_bo.h
> > index 00e93b3065c9..681495e905af 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.h
> > +++ b/drivers/gpu/drm/xe/xe_bo.h
> > @@ -270,6 +270,9 @@ static inline bool
> > xe_bo_madv_is_dontneed(struct xe_bo *bo)
> >  	return bo->madv_purgeable == XE_MADV_PURGEABLE_DONTNEED;
> >  }
> >  
> > +void xe_bo_set_purgeable_shrinker(struct xe_bo *bo);
> > +void xe_bo_clear_purgeable_shrinker(struct xe_bo *bo);
> > +
> >  static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo)
> >  {
> >  	if (likely(bo)) {
> > diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > index 5808fef89777..0fb07a1ed3ae 100644
> > --- a/drivers/gpu/drm/xe/xe_vm_madvise.c
> > +++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
> > @@ -274,12 +274,16 @@ void
> > xe_bo_recheck_purgeable_on_vma_unbind(struct xe_bo *bo)
> >  
> >  	if (xe_bo_all_vmas_dontneed(bo)) {
> >  		/* All VMAs are DONTNEED - mark BO purgeable */
> > -		if (bo->madv_purgeable !=
> > XE_MADV_PURGEABLE_DONTNEED)
> > +		if (bo->madv_purgeable !=
> > XE_MADV_PURGEABLE_DONTNEED) {
> >  			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_DONTNEED;
> > +			xe_bo_set_purgeable_shrinker(bo);
> > +		}
> >  	} else {
> >  		/* At least one VMA is WILLNEED - BO must not be
> > purgeable */
> > -		if (bo->madv_purgeable !=
> > XE_MADV_PURGEABLE_WILLNEED)
> > +		if (bo->madv_purgeable !=
> > XE_MADV_PURGEABLE_WILLNEED) {
> >  			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_WILLNEED;
> > +			xe_bo_clear_purgeable_shrinker(bo);
> > +		}
> >  	}
> >  }
> >  
> > @@ -325,13 +329,16 @@ static bool xe_vm_madvise_purgeable_bo(struct
> > xe_device *xe, struct xe_vm *vm,
> >  
> >  			/* Mark VMA WILLNEED - BO becomes non-
> > purgeable immediately */
> >  			bo->madv_purgeable =
> > XE_MADV_PURGEABLE_WILLNEED;
> > +			xe_bo_clear_purgeable_shrinker(bo);
> >  			break;
> >  		case DRM_XE_VMA_PURGEABLE_STATE_DONTNEED:
> >  			vmas[i]->purgeable_state =
> > XE_MADV_PURGEABLE_DONTNEED;
> >  
> >  			/* Mark BO purgeable only if all VMAs are
> > DONTNEED */
> > -			if (xe_bo_all_vmas_dontneed(bo))
> > +			if (xe_bo_all_vmas_dontneed(bo)) {
> >  				bo->madv_purgeable =
> > XE_MADV_PURGEABLE_DONTNEED;
> > +				xe_bo_set_purgeable_shrinker(bo);
> > +			}
> >  			break;
> >  		default:
> >  			drm_warn(&vm->xe->drm, "Invalid madvice
> > value = %d\n",
> > -- 
> > 2.43.0
> > 

✗ CI.checkpatch: warning for drm/xe/madvise: Add support for purgeable buffer objects (rev5)

[Patchwork]

== Series Details ==

Series: drm/xe/madvise: Add support for purgeable buffer objects (rev5)
URL   : https://patchwork.freedesktop.org/series/156651/
State : warning

== Summary ==

+ KERNEL=/kernel
+ git clone https://gitlab.freedesktop.org/drm/maintainer-tools mt
Cloning into 'mt'...
warning: redirecting to https://gitlab.freedesktop.org/drm/maintainer-tools.git/
+ git -C mt rev-list -n1 origin/master
ee83616c430ce70bd254bd2774d143a5733c8666
+ cd /kernel
+ git config --global --add safe.directory /kernel
+ git log -n1
commit b8e43aa90f01e0196366622cb23da4c031bb10bb
Author: Arvind Yadav <arvind.yadav@intel.com>
Date:   Tue Jan 20 11:38:54 2026 +0530

    drm/xe/bo: Add purgeable shrinker state helpers
    
    Encapsulate TTM purgeable flag updates and shrinker page accounting
    into helper functions. This prevents desynchronization between the
    TTM tt->purgeable flag and the shrinker's page bucket counters.
    
    Without these helpers, direct manipulation of xe_ttm_tt->purgeable
    risks forgetting to update the corresponding shrinker counters,
    leading to incorrect memory pressure calculations.
    
    Add xe_bo_set_purgeable_shrinker() and xe_bo_clear_purgeable_shrinker()
    which atomically update both the TTM flag and transfer pages between
    the shrinkable and purgeable buckets.
    
    v4:
      - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)
    
    Cc: Matthew Brost <matthew.brost@intel.com>
    Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
    Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
    Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
+ /mt/dim checkpatch 3319a649c1a1d4389f48c561537b9915b1789800 drm-intel
6540a04e9e19 drm/xe/uapi: Add UAPI support for purgeable buffer objects
bd39a2b12369 drm/xe/bo: Add purgeable bo state tracking and field madv to xe_bo
-:37: WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?)
#37: 
  - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)

total: 0 errors, 1 warnings, 0 checks, 77 lines checked
aa339e71bdc8 drm/xe/madvise: Implement purgeable buffer object support
-:23: WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?)
#23: 
  - Async TLB invalidation via trigger_rebind (no blocking xe_vm_invalidate_vma)

total: 0 errors, 1 warnings, 0 checks, 304 lines checked
1b71ab7b8171 drm/xe/bo: Handle CPU faults on purged buffer objects
eb146765e124 drm/xe/vm: Prevent binding of purged buffer objects
277c31862829 drm/xe/madvise: Implement per-VMA purgeable state tracking
-:29: WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?)
#29: 
  - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)

total: 0 errors, 1 warnings, 0 checks, 172 lines checked
37e2d276b275 drm/xe/madvise: Block imported and exported dma-bufs
b8e43aa90f01 drm/xe/bo: Add purgeable shrinker state helpers
-:22: WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?)
#22: 
  - @madv_purgeable atomic_t → u32 change across all relevant patches. (Matt)

total: 0 errors, 1 warnings, 0 checks, 110 lines checked

✗ CI.KUnit: failure for drm/xe/madvise: Add support for purgeable buffer objects (rev5)

[Patchwork]

== Series Details ==

Series: drm/xe/madvise: Add support for purgeable buffer objects (rev5)
URL   : https://patchwork.freedesktop.org/series/156651/
State : failure

== Summary ==

+ trap cleanup EXIT
+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/xe/.kunitconfig
[06:14:58] Configuring KUnit Kernel ...
Generating .config ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig
[06:15:02] Building KUnit Kernel ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig
Building with:
$ make all compile_commands.json scripts_gdb ARCH=um O=.kunit --jobs=48
[06:15:34] Starting KUnit Kernel (1/1)...
[06:15:34] ============================================================
Running tests with:
$ .kunit/linux kunit.enable=1 mem=1G console=tty kunit_shutdown=halt
[06:15:34] ================== guc_buf (11 subtests) ===================
[06:15:34] [PASSED] test_smallest
[06:15:34] [PASSED] test_largest
[06:15:34] [PASSED] test_granular
[06:15:34] [PASSED] test_unique
[06:15:34] [PASSED] test_overlap
[06:15:34] [PASSED] test_reusable
[06:15:34] [PASSED] test_too_big
[06:15:34] [PASSED] test_flush
[06:15:34] [PASSED] test_lookup
[06:15:34] [PASSED] test_data
[06:15:34] [PASSED] test_class
[06:15:34] ===================== [PASSED] guc_buf =====================
[06:15:34] =================== guc_dbm (7 subtests) ===================
[06:15:34] [PASSED] test_empty
[06:15:34] [PASSED] test_default
[06:15:34] ======================== test_size  ========================
[06:15:34] [PASSED] 4
[06:15:34] [PASSED] 8
[06:15:34] [PASSED] 32
[06:15:34] [PASSED] 256
[06:15:34] ==================== [PASSED] test_size ====================
[06:15:34] ======================= test_reuse  ========================
[06:15:34] [PASSED] 4
[06:15:34] [PASSED] 8
[06:15:34] [PASSED] 32
[06:15:34] [PASSED] 256
[06:15:34] =================== [PASSED] test_reuse ====================
[06:15:34] =================== test_range_overlap  ====================
[06:15:34] [PASSED] 4
[06:15:34] [PASSED] 8
[06:15:34] [PASSED] 32
[06:15:34] [PASSED] 256
[06:15:34] =============== [PASSED] test_range_overlap ================
[06:15:34] =================== test_range_compact  ====================
[06:15:34] [PASSED] 4
[06:15:34] [PASSED] 8
[06:15:34] [PASSED] 32
[06:15:34] [PASSED] 256
[06:15:34] =============== [PASSED] test_range_compact ================
[06:15:34] ==================== test_range_spare  =====================
[06:15:34] [PASSED] 4
[06:15:34] [PASSED] 8
[06:15:34] [PASSED] 32
[06:15:34] [PASSED] 256
[06:15:34] ================ [PASSED] test_range_spare =================
[06:15:34] ===================== [PASSED] guc_dbm =====================
[06:15:34] =================== guc_idm (6 subtests) ===================
[06:15:34] [PASSED] bad_init
[06:15:34] [PASSED] no_init
[06:15:34] [PASSED] init_fini
[06:15:34] [PASSED] check_used
[06:15:34] [PASSED] check_quota
[06:15:34] [PASSED] check_all
[06:15:34] ===================== [PASSED] guc_idm =====================
[06:15:34] ================== no_relay (3 subtests) ===================
[06:15:34] [PASSED] xe_drops_guc2pf_if_not_ready
[06:15:34] [PASSED] xe_drops_guc2vf_if_not_ready
[06:15:34] [PASSED] xe_rejects_send_if_not_ready
[06:15:34] ==================== [PASSED] no_relay =====================
[06:15:34] ================== pf_relay (14 subtests) ==================
[06:15:34] [PASSED] pf_rejects_guc2pf_too_short
[06:15:34] [PASSED] pf_rejects_guc2pf_too_long
[06:15:34] [PASSED] pf_rejects_guc2pf_no_payload
[06:15:34] [PASSED] pf_fails_no_payload
[06:15:34] [PASSED] pf_fails_bad_origin
[06:15:34] [PASSED] pf_fails_bad_type
[06:15:34] [PASSED] pf_txn_reports_error
[06:15:34] [PASSED] pf_txn_sends_pf2guc
[06:15:34] [PASSED] pf_sends_pf2guc
[06:15:34] [SKIPPED] pf_loopback_nop
[06:15:34] [SKIPPED] pf_loopback_echo
[06:15:34] [SKIPPED] pf_loopback_fail
[06:15:34] [SKIPPED] pf_loopback_busy
[06:15:34] [SKIPPED] pf_loopback_retry
[06:15:34] ==================== [PASSED] pf_relay =====================
[06:15:34] ================== vf_relay (3 subtests) ===================
[06:15:34] [PASSED] vf_rejects_guc2vf_too_short
[06:15:34] [PASSED] vf_rejects_guc2vf_too_long
[06:15:34] [PASSED] vf_rejects_guc2vf_no_payload
[06:15:34] ==================== [PASSED] vf_relay =====================
[06:15:34] ================ pf_gt_config (6 subtests) =================
[06:15:34] [PASSED] fair_contexts_1vf
[06:15:34] [PASSED] fair_doorbells_1vf
[06:15:34] [PASSED] fair_ggtt_1vf
[06:15:34] ====================== fair_contexts  ======================
[06:15:34] [PASSED] 1 VF
[06:15:34] [PASSED] 2 VFs
[06:15:34] [PASSED] 3 VFs
[06:15:34] [PASSED] 4 VFs
[06:15:34] [PASSED] 5 VFs
[06:15:34] [PASSED] 6 VFs
[06:15:34] [PASSED] 7 VFs
[06:15:34] [PASSED] 8 VFs
[06:15:34] [PASSED] 9 VFs
[06:15:34] [PASSED] 10 VFs
[06:15:34] [PASSED] 11 VFs
[06:15:34] [PASSED] 12 VFs
[06:15:34] [PASSED] 13 VFs
[06:15:34] [PASSED] 14 VFs
[06:15:34] [PASSED] 15 VFs
[06:15:34] [PASSED] 16 VFs
[06:15:34] [PASSED] 17 VFs
[06:15:34] [PASSED] 18 VFs
[06:15:34] [PASSED] 19 VFs
[06:15:34] [PASSED] 20 VFs
[06:15:34] [PASSED] 21 VFs
[06:15:34] [PASSED] 22 VFs
[06:15:34] [PASSED] 23 VFs
[06:15:34] [PASSED] 24 VFs
[06:15:34] [PASSED] 25 VFs
[06:15:34] [PASSED] 26 VFs
[06:15:34] [PASSED] 27 VFs
[06:15:34] [PASSED] 28 VFs
[06:15:34] [PASSED] 29 VFs
[06:15:34] [PASSED] 30 VFs
[06:15:34] [PASSED] 31 VFs
[06:15:34] [PASSED] 32 VFs
[06:15:34] [PASSED] 33 VFs
[06:15:34] [PASSED] 34 VFs
[06:15:34] [PASSED] 35 VFs
[06:15:34] [PASSED] 36 VFs
[06:15:34] [PASSED] 37 VFs
[06:15:34] [PASSED] 38 VFs
[06:15:34] [PASSED] 39 VFs
[06:15:34] [PASSED] 40 VFs
[06:15:34] [PASSED] 41 VFs
[06:15:34] [PASSED] 42 VFs
[06:15:34] [PASSED] 43 VFs
[06:15:34] [PASSED] 44 VFs
[06:15:34] [PASSED] 45 VFs
[06:15:34] [PASSED] 46 VFs
[06:15:34] [PASSED] 47 VFs
[06:15:34] [PASSED] 48 VFs
[06:15:34] [PASSED] 49 VFs
[06:15:34] [PASSED] 50 VFs
[06:15:34] [PASSED] 51 VFs
[06:15:34] [PASSED] 52 VFs
[06:15:34] [PASSED] 53 VFs
[06:15:34] [PASSED] 54 VFs
[06:15:34] [PASSED] 55 VFs
[06:15:34] [PASSED] 56 VFs
[06:15:34] [PASSED] 57 VFs
[06:15:34] [PASSED] 58 VFs
[06:15:34] [PASSED] 59 VFs
[06:15:34] [PASSED] 60 VFs
[06:15:34] [PASSED] 61 VFs
[06:15:34] [PASSED] 62 VFs
[06:15:34] [PASSED] 63 VFs
[06:15:34] ================== [PASSED] fair_contexts ==================
[06:15:34] ===================== fair_doorbells  ======================
[06:15:34] [PASSED] 1 VF
[06:15:34] [PASSED] 2 VFs
[06:15:34] [PASSED] 3 VFs
[06:15:34] [PASSED] 4 VFs
[06:15:34] [PASSED] 5 VFs
[06:15:34] [PASSED] 6 VFs
[06:15:34] [PASSED] 7 VFs
[06:15:34] [PASSED] 8 VFs
[06:15:34] [PASSED] 9 VFs
[06:15:34] [PASSED] 10 VFs
[06:15:34] [PASSED] 11 VFs
[06:15:34] [PASSED] 12 VFs
[06:15:34] [PASSED] 13 VFs
[06:15:34] [PASSED] 14 VFs
[06:15:34] [PASSED] 15 VFs
[06:15:34] [PASSED] 16 VFs
[06:15:34] [PASSED] 17 VFs
[06:15:34] [PASSED] 18 VFs
[06:15:34] [PASSED] 19 VFs
[06:15:34] [PASSED] 20 VFs
[06:15:34] [PASSED] 21 VFs
[06:15:34] [PASSED] 22 VFs
[06:15:34] [PASSED] 23 VFs
[06:15:34] [PASSED] 24 VFs
[06:15:34] [PASSED] 25 VFs
[06:15:34] [PASSED] 26 VFs
[06:15:34] [PASSED] 27 VFs
[06:15:34] [PASSED] 28 VFs
[06:15:34] [PASSED] 29 VFs
[06:15:34] [PASSED] 30 VFs
[06:15:34] [PASSED] 31 VFs
[06:15:34] [PASSED] 32 VFs
[06:15:34] [PASSED] 33 VFs
[06:15:34] [PASSED] 34 VFs
[06:15:34] [PASSED] 35 VFs
[06:15:34] [PASSED] 36 VFs
[06:15:34] [PASSED] 37 VFs
[06:15:34] [PASSED] 38 VFs
[06:15:34] [PASSED] 39 VFs
[06:15:34] [PASSED] 40 VFs
[06:15:34] [PASSED] 41 VFs
[06:15:34] [PASSED] 42 VFs
[06:15:34] [PASSED] 43 VFs
[06:15:34] [PASSED] 44 VFs
[06:15:34] [PASSED] 45 VFs
[06:15:34] [PASSED] 46 VFs
[06:15:34] [PASSED] 47 VFs
[06:15:34] [PASSED] 48 VFs
[06:15:34] [PASSED] 49 VFs
[06:15:34] [PASSED] 50 VFs
[06:15:34] [PASSED] 51 VFs
[06:15:34] [PASSED] 52 VFs
[06:15:34] [PASSED] 53 VFs
[06:15:34] [PASSED] 54 VFs
[06:15:34] [PASSED] 55 VFs
[06:15:34] [PASSED] 56 VFs
[06:15:34] [PASSED] 57 VFs
[06:15:34] [PASSED] 58 VFs
[06:15:34] [PASSED] 59 VFs
[06:15:34] [PASSED] 60 VFs
[06:15:34] [PASSED] 61 VFs
[06:15:34] [PASSED] 62 VFs
[06:15:34] [PASSED] 63 VFs
[06:15:34] ================= [PASSED] fair_doorbells ==================
[06:15:34] ======================== fair_ggtt  ========================
[06:15:34] [PASSED] 1 VF
[06:15:34] [PASSED] 2 VFs
[06:15:34] [PASSED] 3 VFs
[06:15:34] [PASSED] 4 VFs
[06:15:34] [PASSED] 5 VFs
[06:15:34] [PASSED] 6 VFs
[06:15:34] [PASSED] 7 VFs
[06:15:34] [PASSED] 8 VFs
[06:15:34] [PASSED] 9 VFs
[06:15:34] [PASSED] 10 VFs
[06:15:34] [PASSED] 11 VFs
[06:15:34] [PASSED] 12 VFs
[06:15:34] [PASSED] 13 VFs
[06:15:34] [PASSED] 14 VFs
[06:15:34] [PASSED] 15 VFs
[06:15:34] [PASSED] 16 VFs
[06:15:34] [PASSED] 17 VFs
[06:15:34] [PASSED] 18 VFs
[06:15:34] [PASSED] 19 VFs
[06:15:34] [PASSED] 20 VFs
[06:15:34] [PASSED] 21 VFs
[06:15:34] [PASSED] 22 VFs
[06:15:34] [PASSED] 23 VFs
[06:15:34] [PASSED] 24 VFs
[06:15:34] [PASSED] 25 VFs
[06:15:34] [PASSED] 26 VFs
[06:15:34] [PASSED] 27 VFs
[06:15:34] [PASSED] 28 VFs
[06:15:34] [PASSED] 29 VFs
[06:15:34] [PASSED] 30 VFs
[06:15:34] [PASSED] 31 VFs
[06:15:34] [PASSED] 32 VFs
[06:15:34] [PASSED] 33 VFs
[06:15:34] [PASSED] 34 VFs
[06:15:34] [PASSED] 35 VFs
[06:15:34] [PASSED] 36 VFs
[06:15:34] [PASSED] 37 VFs
[06:15:34] [PASSED] 38 VFs
[06:15:34] [PASSED] 39 VFs
[06:15:34] [PASSED] 40 VFs
[06:15:34] [PASSED] 41 VFs
[06:15:34] [PASSED] 42 VFs
[06:15:34] [PASSED] 43 VFs
[06:15:34] [PASSED] 44 VFs
[06:15:34] [PASSED] 45 VFs
[06:15:34] [PASSED] 46 VFs
[06:15:34] [PASSED] 47 VFs
[06:15:34] [PASSED] 48 VFs
[06:15:34] [PASSED] 49 VFs
[06:15:34] [PASSED] 50 VFs
[06:15:34] [PASSED] 51 VFs
[06:15:34] [PASSED] 52 VFs
[06:15:34] [PASSED] 53 VFs
[06:15:34] [PASSED] 54 VFs
[06:15:34] [PASSED] 55 VFs
[06:15:34] [PASSED] 56 VFs
[06:15:34] [PASSED] 57 VFs
[06:15:34] [PASSED] 58 VFs
[06:15:34] [PASSED] 59 VFs
[06:15:34] [PASSED] 60 VFs
[06:15:34] [PASSED] 61 VFs
[06:15:34] [PASSED] 62 VFs
[06:15:34] [PASSED] 63 VFs
[06:15:34] ==================== [PASSED] fair_ggtt ====================
[06:15:34] ================== [PASSED] pf_gt_config ===================
[06:15:34] ===================== lmtt (1 subtest) =====================
[06:15:34] ======================== test_ops  =========================
[06:15:34] [PASSED] 2-level
[06:15:34] [PASSED] multi-level
[06:15:34] ==================== [PASSED] test_ops =====================
[06:15:34] ====================== [PASSED] lmtt =======================
[06:15:34] ================= pf_service (11 subtests) =================
[06:15:34] [PASSED] pf_negotiate_any
[06:15:34] [PASSED] pf_negotiate_base_match
[06:15:34] [PASSED] pf_negotiate_base_newer
[06:15:34] [PASSED] pf_negotiate_base_next
[06:15:34] [SKIPPED] pf_negotiate_base_older
[06:15:34] [PASSED] pf_negotiate_base_prev
[06:15:34] [PASSED] pf_negotiate_latest_match
[06:15:34] [PASSED] pf_negotiate_latest_newer
[06:15:34] [PASSED] pf_negotiate_latest_next
[06:15:34] [SKIPPED] pf_negotiate_latest_older
[06:15:34] [SKIPPED] pf_negotiate_latest_prev
[06:15:34] =================== [PASSED] pf_service ====================
[06:15:34] ================= xe_guc_g2g (2 subtests) ==================
[06:15:34] ============== xe_live_guc_g2g_kunit_default  ==============
[06:15:34] ========= [SKIPPED] xe_live_guc_g2g_kunit_default ==========
[06:15:34] ============== xe_live_guc_g2g_kunit_allmem  ===============
[06:15:34] ========== [SKIPPED] xe_live_guc_g2g_kunit_allmem ==========
[06:15:34] =================== [SKIPPED] xe_guc_g2g ===================
[06:15:34] =================== xe_mocs (2 subtests) ===================
[06:15:34] ================ xe_live_mocs_kernel_kunit  ================
[06:15:34] =========== [SKIPPED] xe_live_mocs_kernel_kunit ============
[06:15:34] ================ xe_live_mocs_reset_kunit  =================
[06:15:34] ============ [SKIPPED] xe_live_mocs_reset_kunit ============
[06:15:34] ==================== [SKIPPED] xe_mocs =====================
[06:15:34] ================= xe_migrate (2 subtests) ==================
[06:15:34] ================= xe_migrate_sanity_kunit  =================
[06:15:34] ============ [SKIPPED] xe_migrate_sanity_kunit =============
[06:15:34] ================== xe_validate_ccs_kunit  ==================
[06:15:34] ============= [SKIPPED] xe_validate_ccs_kunit ==============
[06:15:34] =================== [SKIPPED] xe_migrate ===================
[06:15:34] ================== xe_dma_buf (1 subtest) ==================
[06:15:34] ==================== xe_dma_buf_kunit  =====================
[06:15:34] ================ [SKIPPED] xe_dma_buf_kunit ================
[06:15:34] =================== [SKIPPED] xe_dma_buf ===================
[06:15:34] ================= xe_bo_shrink (1 subtest) =================
[06:15:34] =================== xe_bo_shrink_kunit  ====================
[06:15:34] =============== [SKIPPED] xe_bo_shrink_kunit ===============
[06:15:34] ================== [SKIPPED] xe_bo_shrink ==================
[06:15:34] ==================== xe_bo (2 subtests) ====================
[06:15:34] ================== xe_ccs_migrate_kunit  ===================
[06:15:34] ============== [SKIPPED] xe_ccs_migrate_kunit ==============
[06:15:34] ==================== xe_bo_evict_kunit  ====================
[06:15:34] =============== [SKIPPED] xe_bo_evict_kunit ================
[06:15:34] ===================== [SKIPPED] xe_bo ======================
[06:15:34] ==================== args (13 subtests) ====================
[06:15:34] [PASSED] count_args_test
[06:15:34] [PASSED] call_args_example
[06:15:34] [PASSED] call_args_test
[06:15:34] [PASSED] drop_first_arg_example
[06:15:34] [PASSED] drop_first_arg_test
[06:15:34] [PASSED] first_arg_example
[06:15:34] [PASSED] first_arg_test
[06:15:34] [PASSED] last_arg_example
[06:15:34] [PASSED] last_arg_test
[06:15:34] [PASSED] pick_arg_example
[06:15:34] [PASSED] if_args_example
[06:15:34] [PASSED] if_args_test
[06:15:34] [PASSED] sep_comma_example
[06:15:34] ====================== [PASSED] args =======================
[06:15:34] =================== xe_pci (3 subtests) ====================
[06:15:34] ==================== check_graphics_ip  ====================
[06:15:34] [PASSED] 12.00 Xe_LP
[06:15:34] [PASSED] 12.10 Xe_LP+
[06:15:34] [PASSED] 12.55 Xe_HPG
[06:15:34] [PASSED] 12.60 Xe_HPC
[06:15:34] [PASSED] 12.70 Xe_LPG
[06:15:34] [PASSED] 12.71 Xe_LPG
[06:15:34] [PASSED] 12.74 Xe_LPG+
[06:15:34] [PASSED] 20.01 Xe2_HPG
[06:15:34] [PASSED] 20.02 Xe2_HPG
[06:15:34] [PASSED] 20.04 Xe2_LPG
[06:15:34] [PASSED] 30.00 Xe3_LPG
[06:15:34] [PASSED] 30.01 Xe3_LPG
[06:15:34] [PASSED] 30.03 Xe3_LPG
[06:15:34] [PASSED] 30.04 Xe3_LPG
[06:15:34] [PASSED] 30.05 Xe3_LPG
[06:15:34] [PASSED] 35.11 Xe3p_XPC
[06:15:34] ================ [PASSED] check_graphics_ip ================
[06:15:34] ===================== check_media_ip  ======================
[06:15:34] [PASSED] 12.00 Xe_M
[06:15:34] [PASSED] 12.55 Xe_HPM
[06:15:34] [PASSED] 13.00 Xe_LPM+
[06:15:34] [PASSED] 13.01 Xe2_HPM
[06:15:34] [PASSED] 20.00 Xe2_LPM
[06:15:34] [PASSED] 30.00 Xe3_LPM
[06:15:34] [PASSED] 30.02 Xe3_LPM
[06:15:34] [PASSED] 35.00 Xe3p_LPM
[06:15:34] [PASSED] 35.03 Xe3p_HPM
[06:15:34] ================= [PASSED] check_media_ip ==================
[06:15:34] =================== check_platform_desc  ===================
[06:15:34] [PASSED] 0x9A60 (TIGERLAKE)
[06:15:34] [PASSED] 0x9A68 (TIGERLAKE)
[06:15:34] [PASSED] 0x9A70 (TIGERLAKE)
[06:15:34] [PASSED] 0x9A40 (TIGERLAKE)
[06:15:34] [PASSED] 0x9A49 (TIGERLAKE)
[06:15:34] [PASSED] 0x9A59 (TIGERLAKE)
[06:15:34] [PASSED] 0x9A78 (TIGERLAKE)
[06:15:34] [PASSED] 0x9AC0 (TIGERLAKE)
[06:15:34] [PASSED] 0x9AC9 (TIGERLAKE)
[06:15:34] [PASSED] 0x9AD9 (TIGERLAKE)
[06:15:34] [PASSED] 0x9AF8 (TIGERLAKE)
[06:15:34] [PASSED] 0x4C80 (ROCKETLAKE)
[06:15:34] [PASSED] 0x4C8A (ROCKETLAKE)
[06:15:34] [PASSED] 0x4C8B (ROCKETLAKE)
[06:15:34] [PASSED] 0x4C8C (ROCKETLAKE)
[06:15:34] [PASSED] 0x4C90 (ROCKETLAKE)
[06:15:34] [PASSED] 0x4C9A (ROCKETLAKE)
[06:15:34] [PASSED] 0x4680 (ALDERLAKE_S)
[06:15:34] [PASSED] 0x4682 (ALDERLAKE_S)
[06:15:34] [PASSED] 0x4688 (ALDERLAKE_S)
[06:15:34] [PASSED] 0x468A (ALDERLAKE_S)
[06:15:34] [PASSED] 0x468B (ALDERLAKE_S)
[06:15:34] [PASSED] 0x4690 (ALDERLAKE_S)
[06:15:34] [PASSED] 0x4692 (ALDERLAKE_S)
[06:15:34] [PASSED] 0x4693 (ALDERLAKE_S)
[06:15:34] [PASSED] 0x46A0 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46A1 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46A2 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46A3 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46A6 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46A8 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46AA (ALDERLAKE_P)
[06:15:34] [PASSED] 0x462A (ALDERLAKE_P)
[06:15:34] [PASSED] 0x4626 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x4628 (ALDERLAKE_P)
stty: 'standard input': Inappropriate ioctl for device
[06:15:34] [PASSED] 0x46B0 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46B1 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46B2 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46B3 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46C0 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46C1 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46C2 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46C3 (ALDERLAKE_P)
[06:15:34] [PASSED] 0x46D0 (ALDERLAKE_N)
[06:15:34] [PASSED] 0x46D1 (ALDERLAKE_N)
[06:15:34] [PASSED] 0x46D2 (ALDERLAKE_N)
[06:15:34] [PASSED] 0x46D3 (ALDERLAKE_N)
[06:15:34] [PASSED] 0x46D4 (ALDERLAKE_N)
[06:15:34] [PASSED] 0xA721 (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA7A1 (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA7A9 (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA7AC (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA7AD (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA720 (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA7A0 (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA7A8 (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA7AA (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA7AB (ALDERLAKE_P)
[06:15:34] [PASSED] 0xA780 (ALDERLAKE_S)
[06:15:34] [PASSED] 0xA781 (ALDERLAKE_S)
[06:15:34] [PASSED] 0xA782 (ALDERLAKE_S)
[06:15:34] [PASSED] 0xA783 (ALDERLAKE_S)
[06:15:34] [PASSED] 0xA788 (ALDERLAKE_S)
[06:15:34] [PASSED] 0xA789 (ALDERLAKE_S)
[06:15:34] [PASSED] 0xA78A (ALDERLAKE_S)
[06:15:34] [PASSED] 0xA78B (ALDERLAKE_S)
[06:15:34] [PASSED] 0x4905 (DG1)
[06:15:34] [PASSED] 0x4906 (DG1)
[06:15:34] [PASSED] 0x4907 (DG1)
[06:15:34] [PASSED] 0x4908 (DG1)
[06:15:34] [PASSED] 0x4909 (DG1)
[06:15:34] [PASSED] 0x56C0 (DG2)
[06:15:34] [PASSED] 0x56C2 (DG2)
[06:15:34] [PASSED] 0x56C1 (DG2)
[06:15:34] [PASSED] 0x7D51 (METEORLAKE)
[06:15:34] [PASSED] 0x7DD1 (METEORLAKE)
[06:15:34] [PASSED] 0x7D41 (METEORLAKE)
[06:15:34] [PASSED] 0x7D67 (METEORLAKE)
[06:15:34] [PASSED] 0xB640 (METEORLAKE)
[06:15:34] [PASSED] 0x56A0 (DG2)
[06:15:34] [PASSED] 0x56A1 (DG2)
[06:15:34] [PASSED] 0x56A2 (DG2)
[06:15:34] [PASSED] 0x56BE (DG2)
[06:15:34] [PASSED] 0x56BF (DG2)
[06:15:34] [PASSED] 0x5690 (DG2)
[06:15:34] [PASSED] 0x5691 (DG2)
[06:15:34] [PASSED] 0x5692 (DG2)
[06:15:34] [PASSED] 0x56A5 (DG2)
[06:15:34] [PASSED] 0x56A6 (DG2)
[06:15:34] [PASSED] 0x56B0 (DG2)
[06:15:34] [PASSED] 0x56B1 (DG2)
[06:15:34] [PASSED] 0x56BA (DG2)
[06:15:34] [PASSED] 0x56BB (DG2)
[06:15:34] [PASSED] 0x56BC (DG2)
[06:15:34] [PASSED] 0x56BD (DG2)
[06:15:34] [PASSED] 0x5693 (DG2)
[06:15:34] [PASSED] 0x5694 (DG2)
[06:15:34] [PASSED] 0x5695 (DG2)
[06:15:34] [PASSED] 0x56A3 (DG2)
[06:15:34] [PASSED] 0x56A4 (DG2)
[06:15:34] [PASSED] 0x56B2 (DG2)
[06:15:34] [PASSED] 0x56B3 (DG2)
[06:15:34] [PASSED] 0x5696 (DG2)
[06:15:34] [PASSED] 0x5697 (DG2)
[06:15:34] [PASSED] 0xB69 (PVC)
[06:15:34] [PASSED] 0xB6E (PVC)
[06:15:34] [PASSED] 0xBD4 (PVC)
[06:15:34] [PASSED] 0xBD5 (PVC)
[06:15:34] [PASSED] 0xBD6 (PVC)
[06:15:34] [PASSED] 0xBD7 (PVC)
[06:15:34] [PASSED] 0xBD8 (PVC)
[06:15:34] [PASSED] 0xBD9 (PVC)
[06:15:34] [PASSED] 0xBDA (PVC)
[06:15:34] [PASSED] 0xBDB (PVC)
[06:15:34] [PASSED] 0xBE0 (PVC)
[06:15:34] [PASSED] 0xBE1 (PVC)
[06:15:34] [PASSED] 0xBE5 (PVC)
[06:15:34] [PASSED] 0x7D40 (METEORLAKE)
[06:15:34] [PASSED] 0x7D45 (METEORLAKE)
[06:15:34] [PASSED] 0x7D55 (METEORLAKE)
[06:15:34] [PASSED] 0x7D60 (METEORLAKE)
[06:15:34] [PASSED] 0x7DD5 (METEORLAKE)
[06:15:34] [PASSED] 0x6420 (LUNARLAKE)
[06:15:34] [PASSED] 0x64A0 (LUNARLAKE)
[06:15:34] [PASSED] 0x64B0 (LUNARLAKE)
[06:15:34] [PASSED] 0xE202 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE209 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE20B (BATTLEMAGE)
[06:15:34] [PASSED] 0xE20C (BATTLEMAGE)
[06:15:34] [PASSED] 0xE20D (BATTLEMAGE)
[06:15:34] [PASSED] 0xE210 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE211 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE212 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE216 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE220 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE221 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE222 (BATTLEMAGE)
[06:15:34] [PASSED] 0xE223 (BATTLEMAGE)
[06:15:34] [PASSED] 0xB080 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB081 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB082 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB083 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB084 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB085 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB086 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB087 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB08F (PANTHERLAKE)
[06:15:34] [PASSED] 0xB090 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB0A0 (PANTHERLAKE)
[06:15:34] [PASSED] 0xB0B0 (PANTHERLAKE)
[06:15:34] [PASSED] 0xFD80 (PANTHERLAKE)
[06:15:34] [PASSED] 0xFD81 (PANTHERLAKE)
[06:15:34] [PASSED] 0xD740 (NOVALAKE_S)
[06:15:34] [PASSED] 0xD741 (NOVALAKE_S)
[06:15:34] [PASSED] 0xD742 (NOVALAKE_S)
[06:15:34] [PASSED] 0xD743 (NOVALAKE_S)
[06:15:34] [PASSED] 0xD744 (NOVALAKE_S)
[06:15:34] [PASSED] 0xD745 (NOVALAKE_S)
[06:15:34] [PASSED] 0x674C (CRESCENTISLAND)
[06:15:34] =============== [PASSED] check_platform_desc ===============
[06:15:34] ===================== [PASSED] xe_pci ======================
[06:15:34] =================== xe_rtp (2 subtests) ====================
[06:15:34] =============== xe_rtp_process_to_sr_tests  ================
[06:15:34] [PASSED] coalesce-same-reg
[06:15:34] [PASSED] no-match-no-add
[06:15:34] [PASSED] match-or
[06:15:34] [PASSED] match-or-xfail
[06:15:34] [PASSED] no-match-no-add-multiple-rules
[06:15:34] [PASSED] two-regs-two-entries
[06:15:34] [PASSED] clr-one-set-other
[06:15:34] [PASSED] set-field
[06:15:34] [PASSED] conflict-duplicate
[06:15:34] [PASSED] conflict-not-disjoint
[06:15:34] [PASSED] conflict-reg-type
[06:15:34] =========== [PASSED] xe_rtp_process_to_sr_tests ============
[06:15:34] ================== xe_rtp_process_tests  ===================
[06:15:34] [PASSED] active1
[06:15:34] [PASSED] active2
[06:15:34] [PASSED] active-inactive
[06:15:34] [PASSED] inactive-active
[06:15:34] [PASSED] inactive-1st_or_active-inactive
[06:15:34] [PASSED] inactive-2nd_or_active-inactive
[06:15:34] [PASSED] inactive-last_or_active-inactive
[06:15:34] [PASSED] inactive-no_or_active-inactive
[06:15:34] ============== [PASSED] xe_rtp_process_tests ===============
[06:15:34] ===================== [PASSED] xe_rtp ======================
[06:15:34] ==================== xe_wa (1 subtest) =====================
[06:15:34] ======================== xe_wa_gt  =========================
[06:15:34] [PASSED] TIGERLAKE B0
[06:15:34] [PASSED] DG1 A0
[06:15:34] [PASSED] DG1 B0
[06:15:34] [PASSED] ALDERLAKE_S A0
[06:15:34] [PASSED] ALDERLAKE_S B0
[06:15:34] [PASSED] ALDERLAKE_S C0
[06:15:34] [PASSED] ALDERLAKE_S D0
[06:15:34] [PASSED] ALDERLAKE_P A0
[06:15:34] [PASSED] ALDERLAKE_P B0
[06:15:34] [PASSED] ALDERLAKE_P C0
[06:15:34] [PASSED] ALDERLAKE_S RPLS D0
[06:15:34] [PASSED] ALDERLAKE_P RPLU E0
[06:15:34] [PASSED] DG2 G10 C0
[06:15:34] [PASSED] DG2 G11 B1
[06:15:34] [PASSED] DG2 G12 A1
[06:15:34] [PASSED] METEORLAKE 12.70(Xe_LPG) A0 13.00(Xe_LPM+) A0
[06:15:34] [PASSED] METEORLAKE 12.71(Xe_LPG) A0 13.00(Xe_LPM+) A0
[06:15:34] [PASSED] METEORLAKE 12.74(Xe_LPG+) A0 13.00(Xe_LPM+) A0
[06:15:34] [PASSED] LUNARLAKE 20.04(Xe2_LPG) A0 20.00(Xe2_LPM) A0
[06:15:34] [PASSED] LUNARLAKE 20.04(Xe2_LPG) B0 20.00(Xe2_LPM) A0
[06:15:34] [PASSED] BATTLEMAGE 20.01(Xe2_HPG) A0 13.01(Xe2_HPM) A1
[06:15:34] [PASSED] PANTHERLAKE 30.00(Xe3_LPG) A0 30.00(Xe3_LPM) A0
[06:15:34] ==================== [PASSED] xe_wa_gt =====================
[06:15:34] ====================== [PASSED] xe_wa ======================
[06:15:34] ============================================================
[06:15:34] Testing complete. Ran 512 tests: passed: 494, skipped: 18
[06:15:34] Elapsed time: 36.276s total, 4.183s configuring, 31.575s building, 0.466s running

+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/tests/.kunitconfig
[06:15:34] Configuring KUnit Kernel ...
Regenerating .config ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig
[06:15:36] Building KUnit Kernel ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig
Building with:
$ make all compile_commands.json scripts_gdb ARCH=um O=.kunit --jobs=48
ERROR:root:../drivers/gpu/drm/tests/drm_hdmi_state_helper_test.c: In function ‘drm_test_check_reject_hdr_infoframe_bpc_10’:
../drivers/gpu/drm/tests/drm_hdmi_state_helper_test.c:2926:49: warning: passing argument 6 of ‘drm_property_replace_blob_from_id’ makes integer from pointer without a cast [-Wint-conversion]
 2926 |                                                 &replaced);
      |                                                 ^~~~~~~~~
      |                                                 |
      |                                                 bool * {aka _Bool *}
In file included from ../include/drm/drm_connector.h:33,
                 from ../include/drm/drm_modes.h:33,
                 from ../include/drm/drm_crtc.h:32,
                 from ../include/drm/drm_atomic.h:31,
                 from ../drivers/gpu/drm/tests/drm_hdmi_state_helper_test.c:7:
../include/drm/drm_property.h:287:47: note: expected ‘ssize_t’ {aka ‘long int’} but argument is of type ‘bool *’ {aka ‘_Bool *’}
  287 |                                       ssize_t max_size,
      |                                       ~~~~~~~~^~~~~~~~
../drivers/gpu/drm/tests/drm_hdmi_state_helper_test.c:2922:15: error: too few arguments to function ‘drm_property_replace_blob_from_id’
 2922 |         ret = drm_property_replace_blob_from_id(drm,
      |               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../include/drm/drm_property.h:282:5: note: declared here
  282 | int drm_property_replace_blob_from_id(struct drm_device *dev,
      |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
make[7]: *** [../scripts/Makefile.build:287: drivers/gpu/drm/tests/drm_hdmi_state_helper_test.o] Error 1
make[7]: *** Waiting for unfinished jobs....
make[6]: *** [../scripts/Makefile.build:544: drivers/gpu/drm/tests] Error 2
make[6]: *** Waiting for unfinished jobs....
make[5]: *** [../scripts/Makefile.build:544: drivers/gpu/drm] Error 2
make[4]: *** [../scripts/Makefile.build:544: drivers/gpu] Error 2
make[3]: *** [../scripts/Makefile.build:544: drivers] Error 2
make[2]: *** [/kernel/Makefile:2054: .] Error 2
make[1]: *** [/kernel/Makefile:248: __sub-make] Error 2
make: *** [Makefile:248: __sub-make] Error 2

+ cleanup
++ stat -c %u:%g /kernel
+ chown -R 1003:1003 /kernel