Re: remap_pfn_range addr page alignment

[Alex Zhang <zhangalex@google.com>]

Ah yea, I was trying to do some device memory mapping where there's an
offset into the page.

Ok, I'll submit a patch that adds a VM_BUG_ON check that "addr" is
page aligned and modify the header to reflect that.

-Alex

On Tue, Nov 5, 2019 at 8:04 AM Vlastimil Babka <vbabka@suse.cz> wrote:
>
> On 10/16/19 9:57 PM, Alex Zhang wrote:
> > Hey there!
> >
> > I was seeing some weird behavior where I was hitting
> > random BUG_ONs in memory.c . After a bit of debugging,
> > I traced it back to remap_pfn_range being my entry point
> > into memory.c . Should the "addr" passed it be page aligned?
>
> Most likely yes. How would you map if not on page boundary? I suspect it
> was a bug in the caller?
>
> > I see no documentation stating that nor do I see any checks
> > in the code for it.
>
> I suspect it was implicitly assumed. Most callers seem to pass
> vma->vm_start as addr, which is page aligned. But 'size' parameter
> alignment is fixed up, and it should be a similar case as addr.
> Maybe a VM_BUG_ON check?
>
> > Alex
> >
>
>