diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 4f077c11e8e2..78d19fd4670a 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -2278,9 +2278,26 @@ int xe_gem_mmap_offset_ioctl(struct drm_device *dev, void *data,
 	    XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
 		return -EINVAL;
 
-	if (XE_IOCTL_DBG(xe, args->flags))
+	if (XE_IOCTL_DBG(xe, args->flags &
+			 ~DRM_XE_MMAP_OFFSET_FLAG_PCI_BARRIER))
 		return -EINVAL;
 
+	if (args->flags & DRM_XE_MMAP_OFFSET_FLAG_PCI_BARRIER) {
+		if (XE_IOCTL_DBG(xe, !IS_DGFX(xe)))
+			return -EINVAL;
+
+		if (XE_IOCTL_DBG(xe, args->handle))
+			return -EINVAL;
+
+		if (XE_IOCTL_DBG(xe, PAGE_SIZE > SZ_4K))
+			return -EINVAL;
+
+		BUILD_BUG_ON(((XE_PCI_BARRIER_MMAP_OFFSET >> XE_PTE_SHIFT) +
+			      SZ_4K) >= DRM_FILE_PAGE_OFFSET_START);
+		args->offset = XE_PCI_BARRIER_MMAP_OFFSET;
+		return 0;
+	}
+
 	gem_obj = drm_gem_object_lookup(file, args->handle);
 	if (XE_IOCTL_DBG(xe, !gem_obj))
 		return -ENOENT;
diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
index d9386ab03140..04995c5ced32 100644
--- a/drivers/gpu/drm/xe/xe_bo.h
+++ b/drivers/gpu/drm/xe/xe_bo.h
@@ -75,6 +75,8 @@
 
 #define XE_BO_PROPS_INVALID	(-1)
 
+#define XE_PCI_BARRIER_MMAP_OFFSET	(0x50 << XE_PTE_SHIFT)
+
 struct sg_table;
 
 struct xe_bo *xe_bo_alloc(void);
diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index 5cbc96b214fe..6ecbf7dd396c 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -232,12 +232,117 @@ static long xe_drm_compat_ioctl(struct file *file, unsigned int cmd, unsigned lo
 #define xe_drm_compat_ioctl NULL
 #endif
 
+static void barrier_open(struct vm_area_struct *vma)
+{
+	drm_dev_get(vma->vm_private_data);
+}
+
+static void barrier_close(struct vm_area_struct *vma)
+{
+	drm_dev_put(vma->vm_private_data);
+}
+
+static void barrier_release_dummy_page(struct drm_device *dev, void *res)
+{
+	struct page *dummy_page = (struct page *)res;
+
+	__free_page(dummy_page);
+}
+
+static vm_fault_t barrier_fault(struct vm_fault *vmf)
+{
+	struct drm_device *dev = vmf->vma->vm_private_data;
+	struct vm_area_struct *vma = vmf->vma;
+	vm_fault_t ret = VM_FAULT_NOPAGE;
+	pgprot_t prot;
+	int idx;
+
+	prot = vm_get_page_prot(vma->vm_flags);
+
+	if (drm_dev_enter(dev, &idx)) {
+		unsigned long pfn;
+
+#define LAST_DB_PAGE_OFFSET 0x7ff001
+		pfn = PHYS_PFN(pci_resource_start(to_pci_dev(dev->dev), 0) +
+				LAST_DB_PAGE_OFFSET);
+		ret = vmf_insert_pfn_prot(vma, vma->vm_start, pfn,
+					  pgprot_noncached(prot));
+		drm_dev_exit(idx);
+	} else {
+		struct page *page;
+
+		/* Allocate new dummy page to map all the VA range in this VMA to it*/
+		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+		if (!page)
+			return VM_FAULT_OOM;
+
+		/* Set the page to be freed using drmm release action */
+		if (drmm_add_action_or_reset(dev, barrier_release_dummy_page, page))
+			return VM_FAULT_OOM;
+
+		ret = vmf_insert_pfn_prot(vma, vma->vm_start, page_to_pfn(page),
+					  prot);
+	}
+
+	return ret;
+}
+
+static const struct vm_operations_struct vm_ops_barrier = {
+	.open = barrier_open,
+	.close = barrier_close,
+	.fault = barrier_fault,
+};
+
+static int xe_pci_barrier_mmap(struct file *filp,
+			       struct vm_area_struct *vma)
+{
+	struct drm_file *priv = filp->private_data;
+	struct drm_device *dev = priv->minor->dev;
+	struct xe_device *xe = to_xe_device(dev);
+
+	if (!IS_DGFX(xe))
+		return -EINVAL;
+
+	if (vma->vm_end - vma->vm_start > SZ_4K)
+		return -EINVAL;
+
+	if (is_cow_mapping(vma->vm_flags))
+		return -EINVAL;
+
+	if (vma->vm_flags & (VM_READ | VM_EXEC))
+		return -EINVAL;
+
+	vm_flags_clear(vma, VM_MAYREAD | VM_MAYEXEC);
+	vm_flags_set(vma, VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP | VM_IO);
+	vma->vm_ops = &vm_ops_barrier;
+	vma->vm_private_data = dev;
+	drm_dev_get(vma->vm_private_data);
+
+	return 0;
+}
+
+static int xe_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+	struct drm_file *priv = filp->private_data;
+	struct drm_device *dev = priv->minor->dev;
+
+	if (drm_dev_is_unplugged(dev))
+		return -ENODEV;
+
+	switch (vma->vm_pgoff) {
+	case XE_PCI_BARRIER_MMAP_OFFSET >> XE_PTE_SHIFT:
+		return xe_pci_barrier_mmap(filp, vma);
+	}
+
+	return drm_gem_mmap(filp, vma);
+}
+
 static const struct file_operations xe_driver_fops = {
 	.owner = THIS_MODULE,
 	.open = drm_open,
 	.release = drm_release_noglobal,
 	.unlocked_ioctl = xe_drm_ioctl,
-	.mmap = drm_gem_mmap,
+	.mmap = xe_mmap,
 	.poll = drm_poll,
 	.read = drm_read,
 	.compat_ioctl = xe_drm_compat_ioctl,
diff --git a/include/uapi/drm/xe_drm.h b/include/uapi/drm/xe_drm.h
index f62689ca861a..cac607a30f6d 100644
--- a/include/uapi/drm/xe_drm.h
+++ b/include/uapi/drm/xe_drm.h
@@ -811,6 +811,32 @@ struct drm_xe_gem_create {
 
 /**
  * struct drm_xe_gem_mmap_offset - Input of &DRM_IOCTL_XE_GEM_MMAP_OFFSET
+ *
+ * The @flags can be:
+ *  - %DRM_XE_MMAP_OFFSET_FLAG_PCI_BARRIER - For user to query special offset
+ *  for use in mmap ioctl. Writing to the returned mmap address will generate a
+ *  PCI memory barrier with low overhead (avoiding IOCTL call as well as writing
+ *  to VRAM which would also add overhead), acting like an MI_MEM_FENCE
+ *  instruction.
+ *
+ *  Note: The mmap size can be at most 4K, due to HW limitations. As a result
+ *  this interface is only supported on CPU architectures that support 4K page
+ *  size. The mmap_offset ioctl will detect this and gracefully return an
+ *  error, where userspace is expected to have a different fallback method for
+ *  triggering a barrier.
+ *
+ *  Roughly the usage would be as follows:
+ *
+ *  .. code-block:: C
+ *
+ *  struct drm_xe_gem_mmap_offset mmo = {
+ *	.handle = 0, // must be set to 0
+ *	.flags = DRM_XE_MMAP_OFFSET_FLAG_PCI_BARRIER,
+ *  };
+ *
+ *  err = ioctl(fd, DRM_IOCTL_XE_GEM_MMAP_OFFSET, &mmo);
+ *  map = mmap(NULL, size, PROT_WRITE, MAP_SHARED, fd, mmo.offset);
+ *  map[i] = 0xdeadbeaf; // issue barrier
  */
 struct drm_xe_gem_mmap_offset {
 	/** @extensions: Pointer to the first extension struct, if any */
@@ -819,7 +845,8 @@ struct drm_xe_gem_mmap_offset {
 	/** @handle: Handle for the object being mapped. */
 	__u32 handle;
 
-	/** @flags: Must be zero */
+#define DRM_XE_MMAP_OFFSET_FLAG_PCI_BARRIER     (1 << 0)
+	/** @flags: Flags */
 	__u32 flags;
 
 	/** @offset: The fake offset to use for subsequent mmap call */