On 10/28/2025 1:06 AM, Karol Wachowski wrote:
From: Jacek Lawrynowicz <[email protected]>
Add support for creating buffer objects from user pointers
in the Intel NPU driver.
This feels redundant. $SUBJECT already mentions adding the support, and
in the Intel NPU driver is a given.
Introduce a new ioctl `drm_ivpu_bo_create_from_userptr` that allows
users to create a GEM buffer object from a user pointer to a memory region.
The user pointer must be page-aligned and the memory region must remain
valid for the lifetime of the buffer object.
This seems good, but perhaps incomplete. Why do this? What benifit would
this new IOCTL give the user? Increases performance when the user has a
specific usecase?
diff --git a/drivers/accel/ivpu/ivpu_gem_userptr.c
b/drivers/accel/ivpu/ivpu_gem_userptr.c
new file mode 100644
index 000000000000..c5b64e0f2b13
--- /dev/null
+++ b/drivers/accel/ivpu/ivpu_gem_userptr.c
@@ -0,0 +1,207 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2020-2025 Intel Corporation
+ */
+
+#include <linux/dma-buf.h>
+#include <linux/err.h>
+#include <linux/highmem.h>
+#include <linux/mm.h>
+#include <linux/mman.h>
+#include <linux/scatterlist.h>
+#include <linux/slab.h>
+#include <linux/capability.h>
+
+#include <drm/drm_device.h>
+#include <drm/drm_file.h>
+#include <drm/drm_gem.h>
+
+#include "ivpu_drv.h"
+#include "ivpu_gem.h"
+
+#define IVPU_BO_USERPTR_MAX_SIZE (4ULL * SZ_1G)
Is this not the same as SZ_4G?
+/**
+ * struct drm_ivpu_bo_create_from_userptr - Create dma-buf from user pointer
+ *
+ * Create a GEM buffer object from a user pointer to a memory region.
+ */
+struct drm_ivpu_bo_create_from_userptr {
+ /** @user_ptr: User pointer to memory region (must be page aligned) */
+ __u64 user_ptr;
+
+ /** @size: Size of the memory region in bytes (must be page aligned) */
Mention max size?
+ __u64 size;
+
+ /**
+ * @flags:
+ *
+ * Supported flags:
+ *
+ * %DRM_IVPU_BO_HIGH_MEM:
+ *
+ * Allocate VPU address from >4GB range.
+ *
+ * %DRM_IVPU_BO_DMA_MEM:
+ *
+ * Allocate from DMA memory range accessible by hardware DMA.
+ *
+ * %DRM_IVPU_BO_READ_ONLY:
+ *
+ * Allocate as a read-only buffer object.
+ */
+ __u32 flags;
+
+ /** @handle: Returned GEM object handle */
+ __u32 handle;
+
+ /** @vpu_addr: Returned VPU virtual address */
+ __u64 vpu_addr;
+};
+
/**
* struct drm_ivpu_bo_info - Query buffer object info
*/
