Merge tag 'drm-accel-2022-11-22' of...

Merge tag 'drm-accel-2022-11-22' of https://git.kernel.org/pub/scm/linux/kernel/git/ogabbay/accel into drm-next

This tag contains the patches that add the new compute acceleration
subsystem, which is part of the DRM subsystem.

The patches:
- Add a new directory at drivers/accel.
- Add a new major (261) for compute accelerators.
- Add a new DRM minor type for compute accelerators.
- Integrate the accel core code with DRM core code.
- Add documentation for the accel subsystem.
Signed-off-by: Dave Airlie <airlied@redhat.com>

some acks from the list (some are in the patch series):
Acked-by: Daniel Stone <daniels@collabora.com>
Acked-by: Sonal Santan <sonal.santan@amd.com>
Acked-by: Maxime Ripard <maxime@cerno.tech>
Acked-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Tested-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>

From: Oded Gabbay <ogabbay@kernel.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20221122112222.GA352082@ogabbay-vm-u20.habana-labs.com

Documentation/accel/index.rst

+.. SPDX-License-Identifier: GPL-2.0
+
+====================
+Compute Accelerators
+====================
+
+.. toctree::
+   :maxdepth: 1
+
+   introduction
+
+.. only::  subproject and html
+
+   Indices
+   =======
+
+   * :ref:`genindex`

Documentation/accel/introduction.rst

+.. SPDX-License-Identifier: GPL-2.0
+
+============
+Introduction
+============
+
+The Linux compute accelerators subsystem is designed to expose compute
+accelerators in a common way to user-space and provide a common set of
+functionality.
+
+These devices can be either stand-alone ASICs or IP blocks inside an SoC/GPU.
+Although these devices are typically designed to accelerate
+Machine-Learning (ML) and/or Deep-Learning (DL) computations, the accel layer
+is not limited to handling these types of accelerators.
+
+Typically, a compute accelerator will belong to one of the following
+categories:
+
+- Edge AI - doing inference at an edge device. It can be an embedded ASIC/FPGA,
+  or an IP inside a SoC (e.g. laptop web camera). These devices
+  are typically configured using registers and can work with or without DMA.
+
+- Inference data-center - single/multi user devices in a large server. This
+  type of device can be stand-alone or an IP inside a SoC or a GPU. It will
+  have on-board DRAM (to hold the DL topology), DMA engines and
+  command submission queues (either kernel or user-space queues).
+  It might also have an MMU to manage multiple users and might also enable
+  virtualization (SR-IOV) to support multiple VMs on the same device. In
+  addition, these devices will usually have some tools, such as profiler and
+  debugger.
+
+- Training data-center - Similar to Inference data-center cards, but typically
+  have more computational power and memory b/w (e.g. HBM) and will likely have
+  a method of scaling-up/out, i.e. connecting to other training cards inside
+  the server or in other servers, respectively.
+
+All these devices typically have different runtime user-space software stacks,
+that are tailored-made to their h/w. In addition, they will also probably
+include a compiler to generate programs to their custom-made computational
+engines. Typically, the common layer in user-space will be the DL frameworks,
+such as PyTorch and TensorFlow.
+
+Sharing code with DRM
+=====================
+
+Because this type of devices can be an IP inside GPUs or have similar
+characteristics as those of GPUs, the accel subsystem will use the
+DRM subsystem's code and functionality. i.e. the accel core code will
+be part of the DRM subsystem and an accel device will be a new type of DRM
+device.
+
+This will allow us to leverage the extensive DRM code-base and
+collaborate with DRM developers that have experience with this type of
+devices. In addition, new features that will be added for the accelerator
+drivers can be of use to GPU drivers as well.
+
+Differentiation from GPUs
+=========================
+
+Because we want to prevent the extensive user-space graphic software stack
+from trying to use an accelerator as a GPU, the compute accelerators will be
+differentiated from GPUs by using a new major number and new device char files.
+
+Furthermore, the drivers will be located in a separate place in the kernel
+tree - drivers/accel/.
+
+The accelerator devices will be exposed to the user space with the dedicated
+261 major number and will have the following convention:
+
+- device char files - /dev/accel/accel*
+- sysfs             - /sys/class/accel/accel*/
+- debugfs           - /sys/kernel/debug/accel/accel*/
+
+Getting Started
+===============
+
+First, read the DRM documentation at Documentation/gpu/index.rst.
+Not only it will explain how to write a new DRM driver but it will also
+contain all the information on how to contribute, the Code Of Conduct and
+what is the coding style/documentation. All of that is the same for the
+accel subsystem.
+
+Second, make sure the kernel is configured with CONFIG_DRM_ACCEL.
+
+To expose your device as an accelerator, two changes are needed to
+be done in your driver (as opposed to a standard DRM driver):
+
+- Add the DRIVER_COMPUTE_ACCEL feature flag in your drm_driver's
+  driver_features field. It is important to note that this driver feature is
+  mutually exclusive with DRIVER_RENDER and DRIVER_MODESET. Devices that want
+  to expose both graphics and compute device char files should be handled by
+  two drivers that are connected using the auxiliary bus framework.
+
+- Change the open callback in your driver fops structure to accel_open().
+  Alternatively, your driver can use DEFINE_DRM_ACCEL_FOPS macro to easily
+  set the correct function operations pointers structure.
+
+External References
+===================
+
+email threads
+-------------
+
+* `Initial discussion on the New subsystem for acceleration devices <https://lkml.org/lkml/2022/7/31/83>`_ - Oded Gabbay (2022)
+* `patch-set to add the new subsystem <https://lkml.org/lkml/2022/10/22/544>`_ - Oded Gabbay (2022)
+
+Conference talks
+----------------
+
+* `LPC 2022 Accelerators BOF outcomes summary <https://airlied.blogspot.com/2022/09/accelerators-bof-outcomes-summary.html>`_ - Dave Airlie (2022)

Documentation/admin-guide/devices.txt

@@ -3080,6 +3080,11 @@
 		  ...
 		  255 = /dev/osd255	256th OSD Device
 
+ 261 char	Compute Acceleration Devices
+		  0 = /dev/accel/accel0	First acceleration device
+		  1 = /dev/accel/accel1	Second acceleration device
+		    ...
+
  384-511 char	RESERVED FOR DYNAMIC ASSIGNMENT
 		Character devices that request a dynamic allocation of major
 		number will take numbers starting from 511 and downward,

Documentation/subsystem-apis.rst

@@ -43,6 +43,7 @@ needed).
    input/index
    hwmon/index
    gpu/index
+   accel/index
    security/index
    sound/index
    crypto/index

MAINTAINERS

@@ -6836,6 +6836,15 @@ F:	include/drm/drm*
 F:	include/linux/vga*
 F:	include/uapi/drm/drm*
 
+DRM COMPUTE ACCELERATORS DRIVERS AND FRAMEWORK
+M:	Oded Gabbay <ogabbay@kernel.org>
+L:	dri-devel@lists.freedesktop.org
+S:	Maintained
+C:	irc://irc.oftc.net/dri-devel
+T:	git https://git.kernel.org/pub/scm/linux/kernel/git/ogabbay/accel.git
+F:	Documentation/accel/
+F:	drivers/accel/
+
 DRM DRIVERS FOR ALLWINNER A10
 M:	Maxime Ripard <mripard@kernel.org>
 M:	Chen-Yu Tsai <wens@csie.org>

drivers/Kconfig

@@ -99,6 +99,8 @@ source "drivers/media/Kconfig"
 
 source "drivers/video/Kconfig"
 
+source "drivers/accel/Kconfig"
+
 source "sound/Kconfig"
 
 source "drivers/hid/Kconfig"

drivers/accel/Kconfig

+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Compute Acceleration device configuration
+#
+# This framework provides support for compute acceleration devices, such
+# as, but not limited to, Machine-Learning and Deep-Learning acceleration
+# devices
+#
+menuconfig DRM_ACCEL
+	bool "Compute Acceleration Framework"
+	depends on DRM
+	help
+	  Framework for device drivers of compute acceleration devices, such
+	  as, but not limited to, Machine-Learning and Deep-Learning
+	  acceleration devices.
+	  If you say Y here, you need to select the module that's right for
+	  your acceleration device from the list below.
+	  This framework is integrated with the DRM subsystem as compute
+	  accelerators and GPUs share a lot in common and can use almost the
+	  same infrastructure code.
+	  Having said that, acceleration devices will have a different
+	  major number than GPUs, and will be exposed to user-space using
+	  different device files, called accel/accel* (in /dev, sysfs
+	  and debugfs).

drivers/accel/drm_accel.c

+// SPDX-License-Identifier: GPL-2.0
+
+/*
+ * Copyright 2022 HabanaLabs, Ltd.
+ * All Rights Reserved.
+ *
+ */
+
+#include <linux/debugfs.h>
+#include <linux/device.h>
+#include <linux/idr.h>
+
+#include <drm/drm_accel.h>
+#include <drm/drm_debugfs.h>
+#include <drm/drm_drv.h>
+#include <drm/drm_file.h>
+#include <drm/drm_ioctl.h>
+#include <drm/drm_print.h>
+
+static DEFINE_SPINLOCK(accel_minor_lock);
+static struct idr accel_minors_idr;
+
+static struct dentry *accel_debugfs_root;
+static struct class *accel_class;
+
+static struct device_type accel_sysfs_device_minor = {
+	.name = "accel_minor"
+};
+
+static char *accel_devnode(struct device *dev, umode_t *mode)
+{
+	return kasprintf(GFP_KERNEL, "accel/%s", dev_name(dev));
+}
+
+static int accel_sysfs_init(void)
+{
+	accel_class = class_create(THIS_MODULE, "accel");
+	if (IS_ERR(accel_class))
+		return PTR_ERR(accel_class);
+
+	accel_class->devnode = accel_devnode;
+
+	return 0;
+}
+
+static void accel_sysfs_destroy(void)
+{
+	if (IS_ERR_OR_NULL(accel_class))
+		return;
+	class_destroy(accel_class);
+	accel_class = NULL;
+}
+
+static int accel_name_info(struct seq_file *m, void *data)
+{
+	struct drm_info_node *node = (struct drm_info_node *) m->private;
+	struct drm_minor *minor = node->minor;
+	struct drm_device *dev = minor->dev;
+	struct drm_master *master;
+
+	mutex_lock(&dev->master_mutex);
+	master = dev->master;
+	seq_printf(m, "%s", dev->driver->name);
+	if (dev->dev)
+		seq_printf(m, " dev=%s", dev_name(dev->dev));
+	if (master && master->unique)
+		seq_printf(m, " master=%s", master->unique);
+	if (dev->unique)
+		seq_printf(m, " unique=%s", dev->unique);
+	seq_puts(m, "\n");
+	mutex_unlock(&dev->master_mutex);
+
+	return 0;
+}
+
+static const struct drm_info_list accel_debugfs_list[] = {
+	{"name", accel_name_info, 0}
+};
+#define ACCEL_DEBUGFS_ENTRIES ARRAY_SIZE(accel_debugfs_list)
+
+/**
+ * accel_debugfs_init() - Initialize debugfs for accel minor
+ * @minor: Pointer to the drm_minor instance.
+ * @minor_id: The minor's id
+ *
+ * This function initializes the drm minor's debugfs members and creates
+ * a root directory for the minor in debugfs. It also creates common files
+ * for accelerators and calls the driver's debugfs init callback.
+ */
+void accel_debugfs_init(struct drm_minor *minor, int minor_id)
+{
+	struct drm_device *dev = minor->dev;
+	char name[64];
+
+	INIT_LIST_HEAD(&minor->debugfs_list);
+	mutex_init(&minor->debugfs_lock);
+	sprintf(name, "%d", minor_id);
+	minor->debugfs_root = debugfs_create_dir(name, accel_debugfs_root);
+
+	drm_debugfs_create_files(accel_debugfs_list, ACCEL_DEBUGFS_ENTRIES,
+				 minor->debugfs_root, minor);
+
+	if (dev->driver->debugfs_init)
+		dev->driver->debugfs_init(minor);
+}
+
+/**
+ * accel_set_device_instance_params() - Set some device parameters for accel device
+ * @kdev: Pointer to the device instance.
+ * @index: The minor's index
+ *
+ * This function creates the dev_t of the device using the accel major and
+ * the device's minor number. In addition, it sets the class and type of the
+ * device instance to the accel sysfs class and device type, respectively.
+ */
+void accel_set_device_instance_params(struct device *kdev, int index)
+{
+	kdev->devt = MKDEV(ACCEL_MAJOR, index);
+	kdev->class = accel_class;
+	kdev->type = &accel_sysfs_device_minor;
+}
+
+/**
+ * accel_minor_alloc() - Allocates a new accel minor
+ *
+ * This function access the accel minors idr and allocates from it
+ * a new id to represent a new accel minor
+ *
+ * Return: A new id on success or error code in case idr_alloc failed
+ */
+int accel_minor_alloc(void)
+{
+	unsigned long flags;
+	int r;
+
+	spin_lock_irqsave(&accel_minor_lock, flags);
+	r = idr_alloc(&accel_minors_idr, NULL, 0, ACCEL_MAX_MINORS, GFP_NOWAIT);
+	spin_unlock_irqrestore(&accel_minor_lock, flags);
+
+	return r;
+}
+
+/**
+ * accel_minor_remove() - Remove an accel minor
+ * @index: The minor id to remove.
+ *
+ * This function access the accel minors idr and removes from
+ * it the member with the id that is passed to this function.
+ */
+void accel_minor_remove(int index)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&accel_minor_lock, flags);
+	idr_remove(&accel_minors_idr, index);
+	spin_unlock_irqrestore(&accel_minor_lock, flags);
+}
+
+/**
+ * accel_minor_replace() - Replace minor pointer in accel minors idr.
+ * @minor: Pointer to the new minor.
+ * @index: The minor id to replace.
+ *
+ * This function access the accel minors idr structure and replaces the pointer
+ * that is associated with an existing id. Because the minor pointer can be
+ * NULL, we need to explicitly pass the index.
+ *
+ * Return: 0 for success, negative value for error
+ */
+void accel_minor_replace(struct drm_minor *minor, int index)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&accel_minor_lock, flags);
+	idr_replace(&accel_minors_idr, minor, index);
+	spin_unlock_irqrestore(&accel_minor_lock, flags);
+}
+
+/*
+ * Looks up the given minor-ID and returns the respective DRM-minor object. The
+ * refence-count of the underlying device is increased so you must release this
+ * object with accel_minor_release().
+ *
+ * The object can be only a drm_minor that represents an accel device.
+ *
+ * As long as you hold this minor, it is guaranteed that the object and the
+ * minor->dev pointer will stay valid! However, the device may get unplugged and
+ * unregistered while you hold the minor.
+ */
+static struct drm_minor *accel_minor_acquire(unsigned int minor_id)
+{
+	struct drm_minor *minor;
+	unsigned long flags;
+
+	spin_lock_irqsave(&accel_minor_lock, flags);
+	minor = idr_find(&accel_minors_idr, minor_id);
+	if (minor)
+		drm_dev_get(minor->dev);
+	spin_unlock_irqrestore(&accel_minor_lock, flags);
+
+	if (!minor) {
+		return ERR_PTR(-ENODEV);
+	} else if (drm_dev_is_unplugged(minor->dev)) {
+		drm_dev_put(minor->dev);
+		return ERR_PTR(-ENODEV);
+	}
+
+	return minor;
+}
+
+static void accel_minor_release(struct drm_minor *minor)
+{
+	drm_dev_put(minor->dev);
+}
+
+/**
+ * accel_open - open method for ACCEL file
+ * @inode: device inode
+ * @filp: file pointer.
+ *
+ * This function must be used by drivers as their &file_operations.open method.
+ * It looks up the correct ACCEL device and instantiates all the per-file
+ * resources for it. It also calls the &drm_driver.open driver callback.
+ *
+ * Return: 0 on success or negative errno value on failure.
+ */
+int accel_open(struct inode *inode, struct file *filp)
+{
+	struct drm_device *dev;
+	struct drm_minor *minor;
+	int retcode;
+
+	minor = accel_minor_acquire(iminor(inode));
+	if (IS_ERR(minor))
+		return PTR_ERR(minor);
+
+	dev = minor->dev;
+
+	atomic_fetch_inc(&dev->open_count);
+
+	/* share address_space across all char-devs of a single device */
+	filp->f_mapping = dev->anon_inode->i_mapping;
+
+	retcode = drm_open_helper(filp, minor);
+	if (retcode)
+		goto err_undo;
+
+	return 0;
+
+err_undo:
+	atomic_dec(&dev->open_count);
+	accel_minor_release(minor);
+	return retcode;
+}
+EXPORT_SYMBOL_GPL(accel_open);
+
+static int accel_stub_open(struct inode *inode, struct file *filp)
+{
+	const struct file_operations *new_fops;
+	struct drm_minor *minor;
+	int err;
+
+	minor = accel_minor_acquire(iminor(inode));
+	if (IS_ERR(minor))
+		return PTR_ERR(minor);
+
+	new_fops = fops_get(minor->dev->driver->fops);
+	if (!new_fops) {
+		err = -ENODEV;
+		goto out;
+	}
+
+	replace_fops(filp, new_fops);
+	if (filp->f_op->open)
+		err = filp->f_op->open(inode, filp);
+	else
+		err = 0;
+
+out:
+	accel_minor_release(minor);
+
+	return err;
+}
+
+static const struct file_operations accel_stub_fops = {
+	.owner = THIS_MODULE,
+	.open = accel_stub_open,
+	.llseek = noop_llseek,
+};
+
+void accel_core_exit(void)
+{
+	unregister_chrdev(ACCEL_MAJOR, "accel");
+	debugfs_remove(accel_debugfs_root);
+	accel_sysfs_destroy();
+	idr_destroy(&accel_minors_idr);
+}
+
+int __init accel_core_init(void)
+{
+	int ret;
+
+	idr_init(&accel_minors_idr);
+
+	ret = accel_sysfs_init();
+	if (ret < 0) {
+		DRM_ERROR("Cannot create ACCEL class: %d\n", ret);
+		goto error;
+	}
+
+	accel_debugfs_root = debugfs_create_dir("accel", NULL);
+
+	ret = register_chrdev(ACCEL_MAJOR, "accel", &accel_stub_fops);
+	if (ret < 0)
+		DRM_ERROR("Cannot register ACCEL major: %d\n", ret);
+
+error:
+	/*
+	 * Any cleanup due to errors will be done in drm_core_exit() that
+	 * will call accel_core_exit()
+	 */
+	return ret;
+}

drivers/gpu/drm/Makefile

@@ -70,6 +70,7 @@ drm-$(CONFIG_DRM_LOAD_EDID_FIRMWARE) += drm_edid_load.o
 drm-$(CONFIG_DRM_PRIVACY_SCREEN) += \
 	drm_privacy_screen.o \
 	drm_privacy_screen_x86.o
+drm-$(CONFIG_DRM_ACCEL) += ../../accel/drm_accel.o
 obj-$(CONFIG_DRM)	+= drm.o
 
 obj-$(CONFIG_DRM_PANEL_ORIENTATION_QUIRKS) += drm_panel_orientation_quirks.o

drivers/gpu/drm/drm_drv.c

@@ -35,6 +35,7 @@
 #include <linux/slab.h>
 #include <linux/srcu.h>
 
+#include <drm/drm_accel.h>
 #include <drm/drm_cache.h>
 #include <drm/drm_client.h>
 #include <drm/drm_color_mgmt.h>
@@ -90,6 +91,8 @@ static struct drm_minor **drm_minor_get_slot(struct drm_device *dev,
 		return &dev->primary;
 	case DRM_MINOR_RENDER:
 		return &dev->render;
+	case DRM_MINOR_ACCEL:
+		return &dev->accel;
 	default:
 		BUG();
 	}
@@ -104,9 +107,13 @@ static void drm_minor_alloc_release(struct drm_device *dev, void *data)
 
 	put_device(minor->kdev);
 
-	spin_lock_irqsave(&drm_minor_lock, flags);
-	idr_remove(&drm_minors_idr, minor->index);
-	spin_unlock_irqrestore(&drm_minor_lock, flags);
+	if (minor->type == DRM_MINOR_ACCEL) {
+		accel_minor_remove(minor->index);
+	} else {
+		spin_lock_irqsave(&drm_minor_lock, flags);
+		idr_remove(&drm_minors_idr, minor->index);
+		spin_unlock_irqrestore(&drm_minor_lock, flags);
+	}
 }
 
 static int drm_minor_alloc(struct drm_device *dev, unsigned int type)
@@ -123,13 +130,17 @@ static int drm_minor_alloc(struct drm_device *dev, unsigned int type)
 	minor->dev = dev;
 
 	idr_preload(GFP_KERNEL);
-	spin_lock_irqsave(&drm_minor_lock, flags);
-	r = idr_alloc(&drm_minors_idr,
-		      NULL,
-		      64 * type,
-		      64 * (type + 1),
-		      GFP_NOWAIT);
-	spin_unlock_irqrestore(&drm_minor_lock, flags);
+	if (type == DRM_MINOR_ACCEL) {
+		r = accel_minor_alloc();
+	} else {
+		spin_lock_irqsave(&drm_minor_lock, flags);
+		r = idr_alloc(&drm_minors_idr,
+			NULL,
+			64 * type,
+			64 * (type + 1),
+			GFP_NOWAIT);
+		spin_unlock_irqrestore(&drm_minor_lock, flags);
+	}
 	idr_preload_end();
 
 	if (r < 0)
@@ -161,10 +172,14 @@ static int drm_minor_register(struct drm_device *dev, unsigned int type)
 	if (!minor)
 		return 0;
 
-	ret = drm_debugfs_init(minor, minor->index, drm_debugfs_root);
-	if (ret) {
-		DRM_ERROR("DRM: Failed to initialize /sys/kernel/debug/dri.\n");
-		goto err_debugfs;
+	if (minor->type == DRM_MINOR_ACCEL) {
+		accel_debugfs_init(minor, minor->index);
+	} else {
+		ret = drm_debugfs_init(minor, minor->index, drm_debugfs_root);
+		if (ret) {
+			DRM_ERROR("DRM: Failed to initialize /sys/kernel/debug/dri.\n");
+			goto err_debugfs;
+		}
 	}
 
 	ret = device_add(minor->kdev);
@@ -172,9 +187,13 @@ static int drm_minor_register(struct drm_device *dev, unsigned int type)
 		goto err_debugfs;
 
 	/* replace NULL with @minor so lookups will succeed from now on */
-	spin_lock_irqsave(&drm_minor_lock, flags);
-	idr_replace(&drm_minors_idr, minor, minor->index);
-	spin_unlock_irqrestore(&drm_minor_lock, flags);
+	if (minor->type == DRM_MINOR_ACCEL) {
+		accel_minor_replace(minor, minor->index);
+	} else {
+		spin_lock_irqsave(&drm_minor_lock, flags);
+		idr_replace(&drm_minors_idr, minor, minor->index);
+		spin_unlock_irqrestore(&drm_minor_lock, flags);
+	}
 
 	DRM_DEBUG("new minor registered %d\n", minor->index);
 	return 0;
@@ -194,9 +213,13 @@ static void drm_minor_unregister(struct drm_device *dev, unsigned int type)
 		return;
 
 	/* replace @minor with NULL so lookups will fail from now on */
-	spin_lock_irqsave(&drm_minor_lock, flags);
-	idr_replace(&drm_minors_idr, NULL, minor->index);
-	spin_unlock_irqrestore(&drm_minor_lock, flags);
+	if (minor->type == DRM_MINOR_ACCEL) {
+		accel_minor_replace(NULL, minor->index);
+	} else {
+		spin_lock_irqsave(&drm_minor_lock, flags);
+		idr_replace(&drm_minors_idr, NULL, minor->index);
+		spin_unlock_irqrestore(&drm_minor_lock, flags);
+	}
 
 	device_del(minor->kdev);
 	dev_set_drvdata(minor->kdev, NULL); /* safety belt */
@@ -603,6 +626,13 @@ static int drm_dev_init(struct drm_device *dev,
 	/* no per-device feature limits by default */
 	dev->driver_features = ~0u;
 
+	if (drm_core_check_feature(dev, DRIVER_COMPUTE_ACCEL) &&
+				(drm_core_check_feature(dev, DRIVER_RENDER) ||
+				drm_core_check_feature(dev, DRIVER_MODESET))) {
+		DRM_ERROR("DRM driver can't be both a compute acceleration and graphics driver\n");
+		return -EINVAL;
+	}
+
 	drm_legacy_init_members(dev);
 	INIT_LIST_HEAD(&dev->filelist);
 	INIT_LIST_HEAD(&dev->filelist_internal);
@@ -628,15 +658,21 @@ static int drm_dev_init(struct drm_device *dev,
 
 	dev->anon_inode = inode;
 
-	if (drm_core_check_feature(dev, DRIVER_RENDER)) {
-		ret = drm_minor_alloc(dev, DRM_MINOR_RENDER);
+	if (drm_core_check_feature(dev, DRIVER_COMPUTE_ACCEL)) {
+		ret = drm_minor_alloc(dev, DRM_MINOR_ACCEL);
 		if (ret)
 			goto err;
-	}
+	} else {
+		if (drm_core_check_feature(dev, DRIVER_RENDER)) {
+			ret = drm_minor_alloc(dev, DRM_MINOR_RENDER);
+			if (ret)
+				goto err;
+		}
 
-	ret = drm_minor_alloc(dev, DRM_MINOR_PRIMARY);
-	if (ret)
-		goto err;
+		ret = drm_minor_alloc(dev, DRM_MINOR_PRIMARY);
+		if (ret)
+			goto err;
+	}
 
 	ret = drm_legacy_create_map_hash(dev);
 	if (ret)
@@ -883,6 +919,10 @@ int drm_dev_register(struct drm_device *dev, unsigned long flags)
 	if (ret)
 		goto err_minors;
 
+	ret = drm_minor_register(dev, DRM_MINOR_ACCEL);
+	if (ret)
+		goto err_minors;
+
 	ret = create_compat_control_link(dev);
 	if (ret)
 		goto err_minors;
@@ -902,12 +942,13 @@ int drm_dev_register(struct drm_device *dev, unsigned long flags)
 		 driver->name, driver->major, driver->minor,
 		 driver->patchlevel, driver->date,
 		 dev->dev ? dev_name(dev->dev) : "virtual device",
-		 dev->primary->index);
+		 dev->primary ? dev->primary->index : dev->accel->index);
 
 	goto out_unlock;
 
 err_minors:
 	remove_compat_control_link(dev);
+	drm_minor_unregister(dev, DRM_MINOR_ACCEL);
 	drm_minor_unregister(dev, DRM_MINOR_PRIMARY);
 	drm_minor_unregister(dev, DRM_MINOR_RENDER);
 out_unlock:
@@ -950,6 +991,7 @@ void drm_dev_unregister(struct drm_device *dev)
 	drm_legacy_rmmaps(dev);
 
 	remove_compat_control_link(dev);
+	drm_minor_unregister(dev, DRM_MINOR_ACCEL);
 	drm_minor_unregister(dev, DRM_MINOR_PRIMARY);
 	drm_minor_unregister(dev, DRM_MINOR_RENDER);
 }
@@ -1034,6 +1076,7 @@ static const struct file_operations drm_stub_fops = {
 static void drm_core_exit(void)
 {
 	drm_privacy_screen_lookup_exit();
+	accel_core_exit();
 	unregister_chrdev(DRM_MAJOR, "drm");
 	debugfs_remove(drm_debugfs_root);
 	drm_sysfs_destroy();
@@ -1061,6 +1104,10 @@ static int __init drm_core_init(void)
 	if (ret < 0)
 		goto error;
 
+	ret = accel_core_init();
+	if (ret < 0)
+		goto error;
+
 	drm_privacy_screen_lookup_init();
 
 	drm_core_init_complete = true;

drivers/gpu/drm/drm_file.c

@@ -326,7 +326,7 @@ static int drm_cpu_valid(void)
  * Creates and initializes a drm_file structure for the file private data in \p
  * filp and add it into the double linked list in \p dev.
  */
-static int drm_open_helper(struct file *filp, struct drm_minor *minor)
+int drm_open_helper(struct file *filp, struct drm_minor *minor)
 {
 	struct drm_device *dev = minor->dev;
 	struct drm_file *priv;

drivers/gpu/drm/drm_sysfs.c

@@ -19,6 +19,7 @@
 #include <linux/kdev_t.h>
 #include <linux/slab.h>
 
+#include <drm/drm_accel.h>
 #include <drm/drm_connector.h>
 #include <drm/drm_device.h>
 #include <drm/drm_file.h>
@@ -471,19 +472,26 @@ struct device *drm_sysfs_minor_alloc(struct drm_minor *minor)
 	struct device *kdev;
 	int r;
 
-	if (minor->type == DRM_MINOR_RENDER)
-		minor_str = "renderD%d";
-	else
-		minor_str = "card%d";
-
 	kdev = kzalloc(sizeof(*kdev), GFP_KERNEL);
 	if (!kdev)
 		return ERR_PTR(-ENOMEM);
 
 	device_initialize(kdev);
-	kdev->devt = MKDEV(DRM_MAJOR, minor->index);
-	kdev->class = drm_class;
-	kdev->type = &drm_sysfs_device_minor;
+
+	if (minor->type == DRM_MINOR_ACCEL) {
+		minor_str = "accel%d";
+		accel_set_device_instance_params(kdev, minor->index);
+	} else {
+		if (minor->type == DRM_MINOR_RENDER)
+			minor_str = "renderD%d";
+		else
+			minor_str = "card%d";
+
+		kdev->devt = MKDEV(DRM_MAJOR, minor->index);
+		kdev->class = drm_class;
+		kdev->type = &drm_sysfs_device_minor;
+	}
+
 	kdev->parent = minor->dev->dev;
 	kdev->release = drm_sysfs_release;
 	dev_set_drvdata(kdev, minor);

include/drm/drm_accel.h

+/* SPDX-License-Identifier: GPL-2.0
+ *
+ * Copyright 2022 HabanaLabs, Ltd.
+ * All Rights Reserved.
+ *
+ */
+
+#ifndef DRM_ACCEL_H_
+#define DRM_ACCEL_H_
+
+#include <drm/drm_file.h>
+
+#define ACCEL_MAJOR		261
+#define ACCEL_MAX_MINORS	256
+
+/**
+ * DRM_ACCEL_FOPS - Default drm accelerators file operations
+ *
+ * This macro provides a shorthand for setting the accelerator file ops in the
+ * &file_operations structure.  If all you need are the default ops, use
+ * DEFINE_DRM_ACCEL_FOPS instead.
+ */
+#define DRM_ACCEL_FOPS \
+	.open		= accel_open,\
+	.release	= drm_release,\
+	.unlocked_ioctl	= drm_ioctl,\
+	.compat_ioctl	= drm_compat_ioctl,\
+	.poll		= drm_poll,\
+	.read		= drm_read,\
+	.llseek		= noop_llseek
+
+/**
+ * DEFINE_DRM_ACCEL_FOPS() - macro to generate file operations for accelerators drivers
+ * @name: name for the generated structure
+ *
+ * This macro autogenerates a suitable &struct file_operations for accelerators based
+ * drivers, which can be assigned to &drm_driver.fops. Note that this structure
+ * cannot be shared between drivers, because it contains a reference to the
+ * current module using THIS_MODULE.
+ *
+ * Note that the declaration is already marked as static - if you need a
+ * non-static version of this you're probably doing it wrong and will break the
+ * THIS_MODULE reference by accident.
+ */
+#define DEFINE_DRM_ACCEL_FOPS(name) \
+	static const struct file_operations name = {\
+		.owner		= THIS_MODULE,\
+		DRM_ACCEL_FOPS,\
+	}
+
+#if IS_ENABLED(CONFIG_DRM_ACCEL)
+
+void accel_core_exit(void);
+int accel_core_init(void);
+void accel_minor_remove(int index);
+int accel_minor_alloc(void);
+void accel_minor_replace(struct drm_minor *minor, int index);
+void accel_set_device_instance_params(struct device *kdev, int index);
+int accel_open(struct inode *inode, struct file *filp);
+void accel_debugfs_init(struct drm_minor *minor, int minor_id);
+
+#else
+
+static inline void accel_core_exit(void)
+{
+}
+
+static inline int __init accel_core_init(void)
+{
+	/* Return 0 to allow drm_core_init to complete successfully */
+	return 0;
+}
+
+static inline void accel_minor_remove(int index)
+{
+}
+
+static inline int accel_minor_alloc(void)
+{
+	return -EOPNOTSUPP;
+}
+
+static inline void accel_minor_replace(struct drm_minor *minor, int index)
+{
+}
+
+static inline void accel_set_device_instance_params(struct device *kdev, int index)
+{
+}
+
+static inline void accel_debugfs_init(struct drm_minor *minor, int minor_id)
+{
+}
+
+#endif /* IS_ENABLED(CONFIG_DRM_ACCEL) */
+
+#endif /* DRM_ACCEL_H_ */

include/drm/drm_device.h

@@ -93,6 +93,9 @@ struct drm_device {
 	/** @render: Render node */
 	struct drm_minor *render;
 
+	/** @accel: Compute Acceleration node */
+	struct drm_minor *accel;
+
 	/**
 	 * @registered:
 	 *

include/drm/drm_drv.h

@@ -96,6 +96,14 @@ enum drm_driver_feature {
 	 * synchronization of command submission.
 	 */
 	DRIVER_SYNCOBJ_TIMELINE         = BIT(6),
+	/**
+	 * @DRIVER_COMPUTE_ACCEL:
+	 *
+	 * Driver supports compute acceleration devices. This flag is mutually exclusive with
+	 * @DRIVER_RENDER and @DRIVER_MODESET. Devices that support both graphics and compute
+	 * acceleration should be handled by two drivers that are connected using auxiliary bus.
+	 */
+	DRIVER_COMPUTE_ACCEL            = BIT(7),
 
 	/* IMPORTANT: Below are all the legacy flags, add new ones above. */
 

include/drm/drm_file.h

@@ -51,11 +51,15 @@ struct file;
 
 /* Note that the order of this enum is ABI (it determines
  * /dev/dri/renderD* numbers).
+ *
+ * Setting DRM_MINOR_ACCEL to 32 gives enough space for more drm minors to
+ * be implemented before we hit any future
  */
 enum drm_minor_type {
 	DRM_MINOR_PRIMARY,
 	DRM_MINOR_CONTROL,
 	DRM_MINOR_RENDER,
+	DRM_MINOR_ACCEL = 32,
 };
 
 /**
@@ -70,7 +74,7 @@ enum drm_minor_type {
 struct drm_minor {
 	/* private: */
 	int index;			/* Minor device number */
-	int type;                       /* Control or render */
+	int type;                       /* Control or render or accel */
 	struct device *kdev;		/* Linux device */
 	struct drm_device *dev;
 
@@ -397,7 +401,22 @@ static inline bool drm_is_render_client(const struct drm_file *file_priv)
 	return file_priv->minor->type == DRM_MINOR_RENDER;
 }
 
+/**
+ * drm_is_accel_client - is this an open file of the compute acceleration node
+ * @file_priv: DRM file
+ *
+ * Returns true if this is an open file of the compute acceleration node, i.e.
+ * &drm_file.minor of @file_priv is a accel minor.
+ *
+ * See also the :ref:`section on accel nodes <drm_accel_node>`.
+ */
+static inline bool drm_is_accel_client(const struct drm_file *file_priv)
+{
+	return file_priv->minor->type == DRM_MINOR_ACCEL;
+}
+
 int drm_open(struct inode *inode, struct file *filp);
+int drm_open_helper(struct file *filp, struct drm_minor *minor);
 ssize_t drm_read(struct file *filp, char __user *buffer,
 		 size_t count, loff_t *offset);
 int drm_release(struct inode *inode, struct file *filp);