Re: chasing the four level page table

[Jon Smirl <jonsmirl@gmail.com>]

On Thu, 06 Jan 2005 19:12:15 +0100, Andi Kleen <ak@muc.de> wrote:
> Yes, you should use get_user_pages() instead if you access real memory.
> If you try to find hardware mappings using that there is no ready
> function for you right now, although I guess it could be added.

drm_follow_page is used like this:

offset = drm_follow_page(pt) | ((unsigned long) pt & ~PAGE_MASK);
map = drm_lookup_map(offset, size, dev);
if (map && map->type == _DRM_AGP) {
	vunmap(pt);
	return;
}

I think pt is a user space address. In DRM AGP memory is mapped into
kernel and user space so the user space address is being converted
into a kernel space one. The kernel space one is used to verify that
the address is a valid mapping to AGP space, if so the page is
unmapped.  I didn't write this code so I'm not 100% sure of what is
going on.

On Thu, 06 Jan 2005 19:12:15 +0100, Andi Kleen <ak@muc.de> wrote:
> Jon Smirl <jonsmirl@gmail.com> writes:
> 
> > The DRM driver contains this routine:
> >
> > drivers/char/drm/drm_memory.h
> >
> > static inline unsigned long
> > drm_follow_page (void *vaddr)
> > {
> >       pgd_t *pgd = pgd_offset_k((unsigned long) vaddr);
> >       pud_t *pud = pud_offset(pgd, (unsigned long) vaddr);
> >       pmd_t *pmd = pmd_offset(pud, (unsigned long) vaddr);
> >       pte_t *ptep = pte_offset_kernel(pmd, (unsigned long) vaddr);
> >       return pte_pfn(*ptep) << PAGE_SHIFT;
> > }
> >
> > No other driver needs to chase the page table like this so there is
> > probably some other way to achieve this. Can someone who knows more
> > about the VM system tell me if there is a way to eliminate this code?
> 
> Yes, you should use get_user_pages() instead if you access real memory.
> If you try to find hardware mappings using that there is no ready
> function for you right now, although I guess it could be added.
> 
> The function is also not quite correct, it should already least take
> the page_table_lock (depending on where you call it from) and check
> p*_none() on each level.
> 
> -Andi
> 

-- 
Jon Smirl
jonsmirl@gmail.com