[Intel-gfx] [PATCH] drm/i915: Introduce mapping of user pages into video memory (userptr) ioctl

[Daniel Vetter]

On Mon, Apr 8, 2013 at 7:40 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> On Mon, Apr 08, 2013 at 07:18:11PM +0200, Daniel Vetter wrote:
>> On Tue, Feb 12, 2013 at 02:17:22PM +0000, Chris Wilson wrote:
>> > By exporting the ability to map user address and inserting PTEs
>> > representing their backing pages into the GTT, we can exploit UMA in order
>> > to utilize normal application data as a texture source or even as a
>> > render target (depending upon the capabilities of the chipset). This has
>> > a number of uses, with zero-copy downloads to the GPU and efficient
>> > readback making the intermixed streaming of CPU and GPU operations
>> > fairly efficient. This ability has many widespread implications from
>> > faster rendering of client-side software rasterisers (chromium),
>> > mitigation of stalls due to read back (firefox) and to faster pipelining
>> > of texture data (such as pixel buffer objects in GL or data blobs in CL).
>> >
>> > v2: Compile with CONFIG_MMU_NOTIFIER
>> > v3: We can sleep while performing invalidate-range, which we can utilise
>> > to drop our page references prior to the kernel manipulating the vma
>> > (for either discard or cloning) and so protect normal users.
>> > v4: Only run the invalidate notifier if the range intercepts the bo.
>> > v5: Prevent userspace from attempting to GTT mmap non-page aligned buffers
>> >
>> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>>
>> Quick bikeshed:
>> - Still not really in favour of the in-page gtt offset handling ... I
>>   still think that this is just a fancy mmap api, and it better reject
>>   attempts to not map anything aligned to a full page outright.
>
> Strongly disagree.

Ok, let's dig out the beaten arguments here ;-)
- Imo the gtt_offset frobbery is a bit fragile, and getting this right
in the face of ppgtt won't make it better. And yes the only reason we
still have that field is that you've shot down any patch to remove it
citing userptr here. So "it's here already" doesn't count ;-)
- Userptr for i915 is an mmap interface, and that works on pages,
lying to userspace isn't great.
- I don't see why userspace can't do this themselves. I've seen that
it makes things easier in SNA/X, but for a general purpose interface
that argument doesn't cut too much.
- I'm also a bit afraid that our code implicitly assumes that
size/offset are always page-aligned and I kinda want to avoid that we
have to audit for such issues from here on. We've blown up in the past
assuming that size > 0 already, I think we're set to blow up on this
one here.

In any case, if you really want to stick to this I want this to be
explictly track in an obj->reloc_gtt_offset_adjustment or something
which is very loudly yelling at people to make sure no one trips over
it. Tracking the adjustment in a separate field, which would only ever
be used in the reloc code would address all my concerns (safe for the
api ugliness one).

>> - I915_USERPTR_UNSYNCHRONIZED ... eeek. That means that despite everyone
>>   having mmu notifiers enabled in their distro config, you make sure sna
>>   doesn't hit it. Imo not enough testing coverage ;-) Or this there
>>   another reason behind this than "mmu notifiers are too slow"?
>>
>>   Generally I'm a bit sloppy with going root-only for legacy X stuff (like
>>   scanline waits), but this here looks very much generally useful. So not
>>   exemption-material imo.
>
> Strongly disagree. Most of my machines do not have mmu-notifiers and
> would still like to benefit from userptr and I see no reason why we need
> to force mmu-notifiers.

Note that I didn't shout against the mmu_notifier-less support
(although I'm honestly not too happy about it), what I don't like is
the override bit disabling the mmu_notifiers even if we have them.
Since that will mean that the code won't be tested through SNA, and so
has a good chance of being buggy. Once mesa comes around and uses it,
it'll nicely blow up. And one of the reason Jesse is breathing down my
neck to merge this is "other guys are interested in this at intel".

>> - On a quick read I've seen some gtt mmap support remnants. This scares
>>   me, a roundabout njet! would appease. Though I think that should already
>>   happen with the checks we have to reject snoopable buffers?
>
> That's because there are platforms where it is theorectically possible
> and whilst I have no use case for it, I wanted to make it work
> nevertheless. I still think it is possible, but I could not see a way to
> do so without completely replacing the drm vm code.

If I understand things correctly we should be able to block this
simply by refusing to create an mmap offset for a userptr backed
object.

>> - I think we should reject set_cacheing requests on usersptr objects.
>
> I don't think that is strictly required, just like we should not limit
> the user from using set_tiling. (Though the user is never actually going
> to tell the kernel about such tiling...).

Yeah, I guess we could allow people to shot their foot off. Otoh it
adds another dimension to the userptr interface, which we need to make
sure it works. Similar besides set_tiling is probably also
prime_export.

The reason I'd prefer to lock things down is that we have fairly nice
coverage for normal gem objects wrt exercising corner-cases with igt.
If we don't disallow the same corner-cases with userptr, I'd prefer
the tests to also cover those ... Disallowing is cheaper ;-)

>> - union drm_i915_gem_objects freaked me out shortly, until I've noticed
>>   that it's only used for our private slab. Maybe and explicit max in
>>   there? Also, this somewhat defeats that you've moved the userptr stuff
>>   out of the base class - this way we won't save any memory ...
>
> The alternative is to use a union inside the object. Long ago, I had a
> few more objects in there.

Yeah, the entire sg rework nicely unified things \o/

Still, the union seems to be killable with a few lines of patch-diff
on top of this ...

>> - Basic igt to check the above api corner-cases return -EINVAL would be
>>   nice.
>
> Been sitting around for ages, just waiting for the interface to be
> agreed upon.
>
>> - I need to check for deadlocks around the mmu notifier handling. Iirc
>>   that thing takes all mm locks, and our own bo unref code can be called
>>   from all kinds of interesting places. Since each vma object also holds a
>>   ref onto a gem bo I suspect that we do have some fun all here ...
>
> The notifier itself takes no locks, so the locking is whatever state the
> caller sets up. The current locking order is (mm, struct mutex) i.e. the
> same ordering as used the notifier callbacks.

Oops, too scared from the fbdev_notifier - the mmu notifier indeed
uses srcu, so we should be fine here. I need to read the callchains
in-depth though and maybe ask lockdep what it thinks about this first
before I can put my concerns at ease ...
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch