Re: [PATCH] radeon: use vmalloc instead of kmalloc

[Thomas Hellström]

Dave Airlie skrev:
> On Tue, Jun 23, 2009 at 9:35 PM, Peter Zijlstra<peterz@xxxxxxxxxxxxx> wrote:
>   
> > On Mon, 2009-06-22 at 19:26 +0200, Jerome Glisse wrote:
> >     
> > > We don't need to allocated contiguous pages in cs codepath
> > > so use vmalloc instead.
> > >       
> > Best would be to not require >PAGE_SIZE allocations at all of course.
> >     
>
> It gets messy when you have copy from user and spinlocks, it would be nice
> to just parse the userspace PAGE_SIZE at a time, but it would mean dropping
> a lock we probably need to hold.
>
>   
> > But barring that, it would be great to have something like:
> >
> >  ptr = kmalloc(size, GFP_KERNEL | __GFP_NOWARN);
> >  if (!ptr)
> >    ptr = vmalloc(size);
> >     
>
> we have a drm_calloc_large workaround already for the Intel driver which also
> need this.
>   
One problem with multiple vmallocs per command submission is 
performance. Judging from previous work, drivers that are doing this 
tend to get very cpu-hungry. Since Radeon is only allowing a single 
process into the command submission path at once, what about 
pre-allocating a single larger vmalloced buffer at first command 
submission and take care to flush user-space before the submitted 
command stream gets too big.

> > Also, how long do these allocations live? vmalloc space can be quite
> > limited (i386) and it can fragment too.
> >     
>
> Only an ioctl lifetime so they aren't that bad. We however do have some vmaps
> that might be quite large, mainly the fbcon framebuffer (up to 8MB in
> some cases)
>   
That one would be ioremapped, not vmapped right? Not that it matters 
because it's using vmalloc space anyway, but it wouldn't be worse than a 
traditionally ioremapped framebuffer.

Thomas

> Dave.
>
> ------------------------------------------------------------------------------
> Are you an open source citizen? Join us for the Open Source Bridge conference!
> Portland, OR, June 17-19. Two days of sessions, one day of unconference: $250.
> Need another reason to go? 24-hour hacker lounge. Register today!
> http://ad.doubleclick.net/clk;215844324;13503038;v?http://opensourcebridge.org
> --
> _______________________________________________
> Dri-devel mailing list
> Dri-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.sourceforge.net/lists/listinfo/dri-devel
>   

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/