Re: [RFC PATCH v2 0/5] Media Device Allocator API

[Shuah Khan]

On 04/05/2016 12:10 AM, Takashi Iwai wrote:
> On Tue, 05 Apr 2016 05:35:55 +0200,
> Shuah Khan wrote:
>>
>> There are known problems with media device life time management. When media
>> device is released while an media ioctl is in progress, ioctls fail with
>> use-after-free errors and kernel hangs in some cases.
>>
>> Media Device can be in any the following states:
>>
>> - Allocated
>> - Registered (could be tied to more than one driver)
>> - Unregistered, not in use (media device file is not open)
>> - Unregistered, in use (media device file is not open)
>> - Released
>>
>> When media device belongs to  more than one driver, registrations should be
>> tracked to avoid unregistering when one of the drivers does unregister. A new
>> num_drivers field in the struct media_device covers this case. The media device
>> should be unregistered only when the last unregister occurs with num_drivers
>> count zero.
>>
>> When a media device is in use when it is unregistered, it should not be
>> released until the application exits when it detects the unregistered
>> status. Media device that is in use when it is unregistered is moved to
>> to_delete_list. When the last unregister occurs, media device is unregistered
>> and becomes an unregistered, still allocated device. Unregister marks the
>> device to be deleted.
>>
>> When media device belongs to more than one driver, as both drivers could be
>> unbound/bound, driver should not end up getting stale media device that is
>> on its way out. Moving the unregistered media device to to_delete_list helps
>> this case as well.
>>
>> I ran bind/unbind loop tests on uvcvideo, au0828, and snd-usb-audio while
>> running application that does ioctls. Didn't see any use-after-free errors
>> on media device. A couple of known issues seen:
>>
>> 1. When application exits, cdev_put() gets called after media device is
>>    released. This is a known issue to resolve and Media Device Allocator
>>    can't solve this one.
>> 2. When au0828 module is removed and then ioctls fail when cdev_get() looks
>>    for the owning module as au0828 is very often the module that owns the
>>    media devnode. This is a cdev related issue that needs to be resolved and
>>    Media Device Allocator can't solve this one.
>>
>> Shuah Khan (5):
>>   media: Add Media Device Allocator API
>>   media: Add driver count to keep track of media device registrations
>>   media: uvcvideo change to use Media Device Allocator API
>>   media: au0828 change to use Media Device Allocator API
>>   sound/usb: Use Media Controller API to share media resources
> 
> I don't think we need to include usb-audio patch at this stage yet.
> The most important thing for now is to improve / stabilize the API
> itself so that other drivers can use it as is.  Once when the API is
> really stabilized, we create a solid git branch that may be based for
> multiple subsystems, and I'll merge usb-audio stuff through sound git
> tree.

Agreed. I included snd-usb-audio as it provides a good test case for
multiple driver use-case. Yes it is a good idea to have a git branch
for wider testing.

> 
> Also, the previous usb-audio MC implementation had a few serious bugs,
> including quirk NULL dereference.  See the bugzilla below for some fix
> patches to 4.6-rc1:
>   https://bugzilla.kernel.org/show_bug.cgi?id=115561
> Feel free to fold them in, if they are still valid.

I folded them in. It is unfortunate that these bugs were introduced towards
the end when I was making changes to address review comments and I didn't
catch them, especially the quirk NULL dereference.

thanks,
-- Shuah