Re: [PATCH v5 3/3] drm/xe: implement VM_BIND decompression in vm_bind_ioctl

[Matthew Auld <matthew.auld@intel.com>]

On 07/11/2025 11:48, Nitin Gote wrote:
> Implement handling of VM_BIND(..., DECOMPRESS) in xe_vm_bind_ioctl.
> 
> Key changes:
> - Parse and record per-op intent (op->map.request_decompress) when the
>    DECOMPRESS flag is present.
> - Introduce xe_pat_index_get_comp_en() helper to check if a PAT index
>    has compression enabled via the XE2_COMP_EN bit.
> - Validate DECOMPRESS preconditions in the ioctl path:
>    - Only valid for MAP ops.
>    - The provided pat_index must select the device's "no-compression" PAT.
>    - Only meaningful on devices with flat CCS and the required XE2+
>      otherwise return -EOPNOTSUPP.
>    - Use XE_IOCTL_DBG for uAPI sanity checks.
> - Implement xe_bo_decompress():
>    For VRAM BOs run xe_bo_move_notify(), reserve one fence slot,
>    schedule xe_migrate_resolve(), and attach the returned fence
>    with DMA_RESV_USAGE_KERNEL. Non-VRAM cases are silent no-ops.
> - Wire scheduling into vma_lock_and_validate() so VM_BIND will schedule
>    decompression when request_decompress is set.
> - Handle fault-mode VMs by performing decompression synchronously during
>    the bind process, ensuring that the resolve is completed before the bind
>    finishes.
> 
> This schedules an in-place GPU resolve (xe_migrate_resolve) for
> decompression.
> 
> v5: (Matt)
>     - Correct the condition check of xe_pat_index_get_comp_en
> 
> v4: (Matt)
>     - Introduce xe_pat_index_get_comp_en(), which checks
>       XE2_COMP_EN for the pat_index
>     - .interruptible should be true, everything else false
> 
> v3: (Matt)
>     - s/xe_bo_schedule_decompress/xe_bo_decompress
>     - skip the decrompress step if the BO isn't in VRAM
>     - start/size not required in xe_bo_schedule_decompress
>     - Use xe_bo_move_notify instead of xe_vm_invalidate_vma
>       with respect to invalidation.
>     - Nits
> 
> v2:
>     - Move decompression work out of vm_bind ioctl. (Matt)
>     - Put that work in a small helper at the BO/migrate layer invoke it
>       from vma_lock_and_validate which already runs under drm_exec.
>     - Move lightweight checks to vm_bind_ioctl_check_args (Matthew Auld)
> 
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Matthew Auld <matthew.auld@intel.com>
> Acked-by: Michal Mrozek <michal.mrozek@intel.com>
> Signed-off-by: Nitin Gote <nitin.r.gote@intel.com>
> ---
> Hi Matt,
> 
> You're correct. The polarity of the xe_pat_index_get_comp_en
> check was wrong. I misread it, sorry for the confusion.
> Corrected it in this version.
> 
> Thank you,
> -Nitin
> 
> 
>   drivers/gpu/drm/xe/xe_bo.c       | 52 ++++++++++++++++++++++++++++++++
>   drivers/gpu/drm/xe/xe_bo.h       |  2 ++
>   drivers/gpu/drm/xe/xe_pat.c      |  6 ++++
>   drivers/gpu/drm/xe/xe_pat.h      | 10 ++++++
>   drivers/gpu/drm/xe/xe_vm.c       | 40 +++++++++++++++++-------
>   drivers/gpu/drm/xe/xe_vm_types.h |  2 ++
>   6 files changed, 101 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index b0bd31d14bb9..95331285c418 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -3330,6 +3330,58 @@ int xe_gem_mmap_offset_ioctl(struct drm_device *dev, void *data,
>   	return 0;
>   }
>   
> +/**
> + * xe_bo_decompress - schedule in-place decompress and install fence
> + * @bo:   buffer object (caller should hold drm_exec reservations for VM+BO)
> + * @vm:   VM containing the VMA range
> + *
> + * Schedules an in-place resolve via the migrate layer and installs the
> + * returned dma_fence into the BO kernel reservation slot (DMA_RESV_USAGE_KERNEL).
> + * In preempt fence mode, this operation interrupts hardware execution
> + * which is expensive. Page fault mode is recommended for better performance.

Maybe add a comment somewhere here that this only currently works on 
dgpu? In case someone tries to use this on igpu also, which below will 
silently skip.

> + *
> + * Returns 0 on success, negative errno on error.
> + */
> +int xe_bo_decompress(struct xe_bo *bo, struct xe_vm *vm)
> +{
> +	struct xe_tile *tile = xe_device_get_root_tile(vm->xe);

Look like we could extract this from the bo instead:

xe = xe_bo_device(bo);
xe_device_get_root_tile(xe);

And we can then drop the vm arg?

> +	struct dma_fence *decomp_fence = NULL;
> +	int err = 0;

Nit: Please put err at the end of the block. Reverse Christmas tree.

> +	struct ttm_operation_ctx op_ctx = {
> +		.interruptible = true,
> +		.no_wait_gpu = false,
> +		.gfp_retry_mayfail = false,
> +	};
> +
> +	/* Silently skip decompression for non-VRAM buffers */
> +	if (!bo->ttm.resource || !mem_type_is_vram(bo->ttm.resource->mem_type))
> +		return 0;
> +
> +	/* Notify before scheduling resolve */
> +	err = xe_bo_move_notify(bo, &op_ctx);
> +	if (err)
> +		return err;
> +
> +	/* Reserve fence slot before scheduling */
> +	err = dma_resv_reserve_fences(bo->ttm.base.resv, 1);
> +	if (err)
> +		return err;
> +
> +	/* Schedule the in-place decompression */
> +	decomp_fence = xe_migrate_resolve(tile->migrate,
> +					  bo,
> +					  bo->ttm.resource);
> +
> +	if (IS_ERR(decomp_fence))
> +		return PTR_ERR(decomp_fence);
> +
> +	/* Install kernel-usage fence */
> +	dma_resv_add_fence(bo->ttm.base.resv, decomp_fence, DMA_RESV_USAGE_KERNEL);
> +	dma_fence_put(decomp_fence);
> +
> +	return 0;
> +}
> +
>   /**
>    * xe_bo_lock() - Lock the buffer object's dma_resv object
>    * @bo: The struct xe_bo whose lock is to be taken
> diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
> index 911d5b90461a..346b50a1247c 100644
> --- a/drivers/gpu/drm/xe/xe_bo.h
> +++ b/drivers/gpu/drm/xe/xe_bo.h
> @@ -309,6 +309,8 @@ int xe_bo_dumb_create(struct drm_file *file_priv,
>   
>   bool xe_bo_needs_ccs_pages(struct xe_bo *bo);
>   
> +int xe_bo_decompress(struct xe_bo *bo, struct xe_vm *vm);
> +
>   static inline size_t xe_bo_ccs_pages_start(struct xe_bo *bo)
>   {
>   	return PAGE_ALIGN(xe_bo_size(bo));
> diff --git a/drivers/gpu/drm/xe/xe_pat.c b/drivers/gpu/drm/xe/xe_pat.c
> index 68171cceea18..1b4d5d3def0f 100644
> --- a/drivers/gpu/drm/xe/xe_pat.c
> +++ b/drivers/gpu/drm/xe/xe_pat.c
> @@ -196,6 +196,12 @@ u16 xe_pat_index_get_coh_mode(struct xe_device *xe, u16 pat_index)
>   	return xe->pat.table[pat_index].coh_mode;
>   }
>   
> +bool xe_pat_index_get_comp_en(struct xe_device *xe, u16 pat_index)
> +{
> +	WARN_ON(pat_index >= xe->pat.n_entries);
> +	return !!(xe->pat.table[pat_index].value & XE2_COMP_EN);
> +}

Looks like there no now three in-flight series each implementing a 
helper like this. But all slightly different. Also not sure which is 
going be be merged first. I personally prefer making this the same as 
coh_mode above, and have a comp_en field, but either works.

https://patchwork.freedesktop.org/series/156658/
https://patchwork.freedesktop.org/series/157036/

Wondering if it would make sense to split this helper into a separate 
patch and merge that first, and then have all three series converge on 
that? Also hopefully prevents accidently merging multiple helpers for 
this, depending on the merge order?

+ Sanjay & Xin

> +
>   static void program_pat(struct xe_gt *gt, const struct xe_pat_table_entry table[],
>   			int n_entries)
>   {
> diff --git a/drivers/gpu/drm/xe/xe_pat.h b/drivers/gpu/drm/xe/xe_pat.h
> index 05dae03a5f54..b8559120989e 100644
> --- a/drivers/gpu/drm/xe/xe_pat.h
> +++ b/drivers/gpu/drm/xe/xe_pat.h
> @@ -58,4 +58,14 @@ int xe_pat_dump(struct xe_gt *gt, struct drm_printer *p);
>    */
>   u16 xe_pat_index_get_coh_mode(struct xe_device *xe, u16 pat_index);
>   
> +/**
> + * xe_pat_index_get_comp_en - Extract the compression enable flag for
> + * the given pat_index.
> + * @xe: xe device
> + * @pat_index: The pat_index to query
> + *
> + * Return: true if compression is enabled for this pat_index, false otherwise.
> + */
> +bool xe_pat_index_get_comp_en(struct xe_device *xe, u16 pat_index);
> +
>   #endif
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index 8fb5cc6a69ec..a9927e665ad0 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -2302,6 +2302,7 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_vma_ops *vops,
>   				op->map.vma_flags |= XE_VMA_DUMPABLE;
>   			if (flags & DRM_XE_VM_BIND_FLAG_MADVISE_AUTORESET)
>   				op->map.vma_flags |= XE_VMA_MADV_AUTORESET;
> +			op->map.request_decompress = flags & DRM_XE_VM_BIND_FLAG_DECOMPRESS;
>   			op->map.pat_index = pat_index;
>   			op->map.invalidate_on_bind =
>   				__xe_vm_needs_clear_scratch_pages(vm, flags);
> @@ -2838,7 +2839,7 @@ 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 request_decompress)
>   {
>   	struct xe_bo *bo = xe_vma_bo(vma);
>   	struct xe_vm *vm = xe_vma_vm(vma);
> @@ -2851,6 +2852,12 @@ static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
>   			err = xe_bo_validate(bo, vm,
>   					     !xe_vm_in_preempt_fence_mode(vm) &&
>   					     res_evict, exec);
> +
> +		if (err)
> +			return err;
> +
> +		if (request_decompress)
> +			err = xe_bo_decompress(bo, vm);
>   	}
>   
>   	return err;
> @@ -2938,7 +2945,8 @@ 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,
> +						    op->map.request_decompress);
>   		break;
>   	case DRM_GPUVA_OP_REMAP:
>   		err = check_ufence(gpuva_to_vma(op->base.remap.unmap->va));
> @@ -2947,13 +2955,13 @@ 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, false);
>   		if (!err && op->remap.next)
>   			err = vma_lock_and_validate(exec, op->remap.next,
> -						    res_evict, true);
> +						    res_evict, true, false);
>   		break;
>   	case DRM_GPUVA_OP_UNMAP:
>   		err = check_ufence(gpuva_to_vma(op->base.unmap.va));
> @@ -2962,7 +2970,7 @@ 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:
>   	{
> @@ -2977,7 +2985,7 @@ 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, false);
>   		if (!err && !xe_vma_has_no_bo(vma))
>   			err = xe_bo_migrate(xe_vma_bo(vma),
>   					    region_to_mem_type[region],
> @@ -3299,7 +3307,8 @@ ALLOW_ERROR_INJECTION(vm_bind_ioctl_ops_execute, ERRNO);
>   	 DRM_XE_VM_BIND_FLAG_DUMPABLE | \
>   	 DRM_XE_VM_BIND_FLAG_CHECK_PXP | \
>   	 DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR | \
> -	 DRM_XE_VM_BIND_FLAG_MADVISE_AUTORESET)
> +	 DRM_XE_VM_BIND_FLAG_MADVISE_AUTORESET | \
> +	 DRM_XE_VM_BIND_FLAG_DECOMPRESS)
>   
>   #ifdef TEST_VM_OPS_ERROR
>   #define SUPPORTED_FLAGS	(SUPPORTED_FLAGS_STUB | FORCE_OP_ERROR)
> @@ -3357,6 +3366,7 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, struct xe_vm *vm,
>   		bool is_null = flags & DRM_XE_VM_BIND_FLAG_NULL;
>   		bool is_cpu_addr_mirror = flags &
>   			DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR;
> +		bool is_decompress = flags & DRM_XE_VM_BIND_FLAG_DECOMPRESS;
>   		u16 pat_index = (*bind_ops)[i].pat_index;
>   		u16 coh_mode;
>   
> @@ -3391,7 +3401,9 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, struct xe_vm *vm,
>   		    XE_IOCTL_DBG(xe, obj_offset && (is_null ||
>   						    is_cpu_addr_mirror)) ||
>   		    XE_IOCTL_DBG(xe, op != DRM_XE_VM_BIND_OP_MAP &&
> -				 (is_null || is_cpu_addr_mirror)) ||
> +				 (is_decompress || is_null || is_cpu_addr_mirror)) ||
> +		    XE_IOCTL_DBG(xe, is_decompress &&
> +				 xe_pat_index_get_comp_en(xe, pat_index)) ||
>   		    XE_IOCTL_DBG(xe, !obj &&
>   				 op == DRM_XE_VM_BIND_OP_MAP &&
>   				 !is_null && !is_cpu_addr_mirror) ||
> @@ -3411,8 +3423,8 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, struct xe_vm *vm,
>   				 op == DRM_XE_VM_BIND_OP_PREFETCH) ||
>   		    XE_IOCTL_DBG(xe, prefetch_region &&
>   				 op != DRM_XE_VM_BIND_OP_PREFETCH) ||
> -		    XE_IOCTL_DBG(xe,  (prefetch_region != DRM_XE_CONSULT_MEM_ADVISE_PREF_LOC &&
> -				       !(BIT(prefetch_region) & xe->info.mem_region_mask))) ||
> +		    XE_IOCTL_DBG(xe, (prefetch_region != DRM_XE_CONSULT_MEM_ADVISE_PREF_LOC &&
> +				      !(BIT(prefetch_region) & xe->info.mem_region_mask))) ||

Is this just fixing whitespace? If so, please consider splitting 
unrelated styling/cosmetic fixes into a separate patch.

Otherwise lgtm,
Reviewed-by: Matthew Auld <matthew.auld@intel.com>

>   		    XE_IOCTL_DBG(xe, obj &&
>   				 op == DRM_XE_VM_BIND_OP_UNMAP) ||
>   		    XE_IOCTL_DBG(xe, (flags & DRM_XE_VM_BIND_FLAG_MADVISE_AUTORESET) &&
> @@ -3429,6 +3441,12 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, struct xe_vm *vm,
>   			err = -EINVAL;
>   			goto free_bind_ops;
>   		}
> +
> +		if (is_decompress && (XE_IOCTL_DBG(xe, !xe_device_has_flat_ccs(xe)) ||
> +				      XE_IOCTL_DBG(xe, GRAPHICS_VER(xe) < 20))) {
> +			err = -EOPNOTSUPP;
> +			goto free_bind_ops;
> +		}
>   	}
>   
>   	return 0;
> diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
> index ccd6cc090309..f4962ca0df09 100644
> --- a/drivers/gpu/drm/xe/xe_vm_types.h
> +++ b/drivers/gpu/drm/xe/xe_vm_types.h
> @@ -346,6 +346,8 @@ struct xe_vma_op_map {
>   	bool immediate;
>   	/** @read_only: Read only */
>   	bool invalidate_on_bind;
> +	/** @request_decompress: schedule decompression for GPU map */
> +	bool request_decompress;
>   	/** @pat_index: The pat index to use for this operation. */
>   	u16 pat_index;
>   };