Re: [PATCH v9 01/12] vfio: Mediated device Core driver

From: Alex Williamson
Date: Thu Oct 20 2016 - 13:12:44 EST


On Thu, 20 Oct 2016 15:23:53 +0800
Jike Song <jike.song@xxxxxxxxx> wrote:

> On 10/18/2016 05:22 AM, Kirti Wankhede wrote:
> > diff --git a/drivers/vfio/mdev/mdev_core.c b/drivers/vfio/mdev/mdev_core.c
> > new file mode 100644
> > index 000000000000..7db5ec164aeb
> > --- /dev/null
> > +++ b/drivers/vfio/mdev/mdev_core.c
> > @@ -0,0 +1,372 @@
> > +/*
> > + * Mediated device Core Driver
> > + *
> > + * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
> > + * Author: Neo Jia <cjia@xxxxxxxxxx>
> > + * Kirti Wankhede <kwankhede@xxxxxxxxxx>
> > + *
> > + * This program is free software; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License version 2 as
> > + * published by the Free Software Foundation.
> > + */
> > +
> > +#include <linux/module.h>
> > +#include <linux/device.h>
> > +#include <linux/slab.h>
> > +#include <linux/uuid.h>
> > +#include <linux/sysfs.h>
> > +#include <linux/mdev.h>
> > +
> > +#include "mdev_private.h"
> > +
> > +#define DRIVER_VERSION "0.1"
> > +#define DRIVER_AUTHOR "NVIDIA Corporation"
> > +#define DRIVER_DESC "Mediated device Core Driver"
> > +
> > +static LIST_HEAD(parent_list);
> > +static DEFINE_MUTEX(parent_list_lock);
> > +static struct class_compat *mdev_bus_compat_class;
> > +
>
> > +
> > +/*
> > + * mdev_register_device : Register a device
> > + * @dev: device structure representing parent device.
> > + * @ops: Parent device operation structure to be registered.
> > + *
> > + * Add device to list of registered parent devices.
> > + * Returns a negative value on error, otherwise 0.
> > + */
> > +int mdev_register_device(struct device *dev, const struct parent_ops *ops)
> > +{
> > + int ret = 0;
> > + struct parent_device *parent;
> > +
> > + /* check for mandatory ops */
> > + if (!ops || !ops->create || !ops->remove || !ops->supported_type_groups)
> > + return -EINVAL;
> > +
> > + dev = get_device(dev);
> > + if (!dev)
> > + return -EINVAL;
> > +
> > + mutex_lock(&parent_list_lock);
> > +
> > + /* Check for duplicate */
> > + parent = __find_parent_device(dev);
> > + if (parent) {
> > + ret = -EEXIST;
> > + goto add_dev_err;
> > + }
> > +
> > + parent = kzalloc(sizeof(*parent), GFP_KERNEL);
> > + if (!parent) {
> > + ret = -ENOMEM;
> > + goto add_dev_err;
> > + }
> > +
> > + kref_init(&parent->ref);
> > +
> > + parent->dev = dev;
> > + parent->ops = ops;
> > +
> > + ret = parent_create_sysfs_files(parent);
> > + if (ret) {
> > + mutex_unlock(&parent_list_lock);
> > + mdev_put_parent(parent);
> > + return ret;
> > + }
> > +
> > + ret = class_compat_create_link(mdev_bus_compat_class, dev, NULL);
> > + if (ret)
> > + dev_warn(dev, "Failed to create compatibility class link\n");
> > +
> > + list_add(&parent->next, &parent_list);
> > + mutex_unlock(&parent_list_lock);
> > +
> > + dev_info(dev, "MDEV: Registered\n");
> > + return 0;
> > +
> > +add_dev_err:
> > + mutex_unlock(&parent_list_lock);
> > + put_device(dev);
> > + return ret;
> > +}
> > +EXPORT_SYMBOL(mdev_register_device);
>
> > +static int __init mdev_init(void)
> > +{
> > + int ret;
> > +
> > + ret = mdev_bus_register();
> > + if (ret) {
> > + pr_err("Failed to register mdev bus\n");
> > + return ret;
> > + }
> > +
> > + mdev_bus_compat_class = class_compat_register("mdev_bus");
> > + if (!mdev_bus_compat_class) {
> > + mdev_bus_unregister();
> > + return -ENOMEM;
> > + }
> > +
> > + /*
> > + * Attempt to load known vfio_mdev. This gives us a working environment
> > + * without the user needing to explicitly load vfio_mdev driver.
> > + */
> > + request_module_nowait("vfio_mdev");
> > +
> > + return ret;
> > +}
> > +
> > +static void __exit mdev_exit(void)
> > +{
> > + class_compat_unregister(mdev_bus_compat_class);
> > + mdev_bus_unregister();
> > +}
> > +
> > +module_init(mdev_init)
> > +module_exit(mdev_exit)
>
> Hi Kirti,
>
> There is a possible issue: mdev_bus_register is called from mdev_init,
> a module_init, equal to device_initcall if builtin to vmlinux; however,
> the vendor driver, say i915.ko for intel case, have to call
> mdev_register_device from its module_init: at that time, mdev_init
> is still not called.
>
> Not sure if this issue exists with nvidia.ko. Though in most cases we
> are expecting users select mdev as a standalone module, we still won't
> break builtin case.
>
>
> Hi Alex, do you have any suggestion here?

To fully solve the problem of built-in drivers making use of the mdev
infrastructure we'd need to make mdev itself builtin and possibly a
subsystem that is initialized prior to device drivers. Is that really
necessary? Even though i915.ko is often loaded as part of an
initramfs, most systems still build it as a module. I would expect
that standard module dependencies will pull in the necessary mdev and
vfio modules to make this work correctly. I can't say that I'm
prepared to make mdev be a subsystem as would be necessary for builtin
drivers to make use of. Perhaps if such a driver exists it could
somehow do late binding with mdev. i915 should certainly be tested as
a builtin driver to make sure it doesn't fail with mdev support added.
The kvm-vfio device (virt/kvm/vfio.c) makes use of symbol tricks to
avoid hard dependencies between kvm and vfio, perhaps when builtin to
the kernel, i915 could use something like that. Thanks,

Alex