linux-kernel - Re: [RFC PATCH 2/2] drm/vgem: Enable dmabuf import interfaces

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 7 Apr 2017 09:18:30 -0700
From:   Laura Abbott <labbott@...hat.com>
To:     Chris Wilson <chris@...is-wilson.co.uk>,
        Daniel Vetter <daniel.vetter@...ll.ch>,
        Sean Paul <seanpaul@...omium.org>,
        dri-devel@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
        Sumit Semwal <sumit.semwal@...aro.org>
Subject: Re: [RFC PATCH 2/2] drm/vgem: Enable dmabuf import interfaces

On 04/07/2017 12:39 AM, Chris Wilson wrote:
> On Thu, Apr 06, 2017 at 04:18:33PM -0700, Laura Abbott wrote:
>>
>> Enable the GEM dma-buf import interfaces in addition to the export
>> interfaces. This lets vgem be used as a test source for other allocators
>> (e.g. Ion).
>>
>> +int vgem_gem_get_pages(struct drm_vgem_gem_object *obj)
>> +{
>> +	struct page **pages;
>> +
>> +	if (obj->pages)
>> +		return 0;
>> +
>> +	pages = drm_gem_get_pages(&obj->base);
>> +	if (IS_ERR(pages)) {
>> +		return PTR_ERR(pages);
>> +	}
>> +
>> +	obj->pages = pages;
> 
> That's a significant loss in functionality (it requires the entire
> object to fit into physical memory at the same time and requires a large
> vmalloc for 32b systems), for what? vgem only has the ability to mmap
> and export a fd -- both of which you already have if you have the dmabuf
> fd. The only extra interface is the signaling, which does not yet have a
> direct correspondence on dmabuf.
> 
> (An obvious way to keep both would be to move the get_pages to importing
> and then differentiate in the faulthandler where to get the page from.)
> 

Thanks for pointing this out. I'll look to keep the existing behavior.

> Can you provide more details on how you are using vgem to justify the
> changes? I'm also not particularly happy in losing testing of a virtual
> platform device from our CI.
> -Chris
> 

There exists a test module in the Ion directory
(drivers/staging/android/ion/ion_test.c) today but it's not actually
Ion specific. It registers a platform device and then provides an
ioctl interface to perform a dma_buf attach and map. I proposed
moving this to a generic dma-buf test module
(https://marc.info/?l=dri-devel&m=148952187004611&w=2) and Daniel
suggested that this was roughly what vgem was supposed to do.

Adding the import methods for vgem covers all of what the
Ion test was doing and we don't have to invent yet another ioctl
interface and framework for attaching and mapping. 

Can you clarify about what you mean about 'virtual platform device'?


Thanks,
Laura