Re: [PATCH 1/6] drm/i915: Fix lock order reversal in GTT pwrite path.

[Peter Zijlstra <peterz@infradead.org>]

On Thu, 2009-03-26 at 17:43 -0700, Jesse Barnes wrote:
> On Wed, 25 Mar 2009 14:45:05 -0700
> Eric Anholt <eric@anholt.net> wrote:
> 
> > Since the pagefault path determines that the lock order we use has to
> > be mmap_sem -> struct_mutex, we can't allow page faults to occur
> > while the struct_mutex is held.  To fix this in pwrite, we first try
> > optimistically to see if we can copy from user without faulting.  If
> > it fails, fall back to using get_user_pages to pin the user's memory,
> > and map those pages atomically when copying it to the GPU.
> > 
> > Signed-off-by: Eric Anholt <eric@anholt.net>
> > ---
> > +	/* Pin the user pages containing the data.  We can't fault
> > while
> > +	 * holding the struct mutex, and all of the pwrite
> > implementations
> > +	 * want to hold it while dereferencing the user data.
> > +	 */
> > +	first_data_page = data_ptr / PAGE_SIZE;
> > +	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
> > +	num_pages = last_data_page - first_data_page + 1;
> > +
> > +	user_pages = kcalloc(num_pages, sizeof(struct page *),
> > GFP_KERNEL);
> > +	if (user_pages == NULL)
> > +		return -ENOMEM;
> 
> If kmalloc limits us to a 128k allocation (and maybe less under
> pressure), then we'll be limited to 128k/8 page pointers on 64 bit, or
> 64M per pwrite...  Is that ok?  Or do we need to handle multiple passes
> here?

While officially supported, a 128k kmalloc is _very_ likely to fail, it
would require an order 5 page allocation to back that, and that is well
outside of comfortable.