Re: [PATCH v6 03/11] mm, x86: Add support for eXclusive Page Frame Ownership (XPFO)

[Dave Hansen]

On 11/14/2017 07:44 PM, Matthew Wilcox wrote:
> On Mon, Nov 13, 2017 at 02:46:25PM -0800, Dave Hansen wrote:
>> On 11/13/2017 02:20 PM, Dave Hansen wrote:
>>> On 11/09/2017 05:09 PM, Tycho Andersen wrote:
>>>> which I guess is from the additional flags in grow_dev_page() somewhere down
>>>> the stack. Anyway... it seems this is a kernel allocation that's using
>>>> MIGRATE_MOVABLE, so perhaps we need some more fine tuned heuristic than just
>>>> all MOVABLE allocations are un-mapped via xpfo, and all the others are mapped.
>>>>
>>>> Do you have any ideas?
>>>
>>> It still has to do a kmap() or kmap_atomic() to be able to access it.  I
>>> thought you hooked into that.  Why isn't that path getting hit for these?
>>
>> Oh, this looks to be accessing data mapped by a buffer_head.  It
>> (rudely) accesses data via:
>>
>> void set_bh_page(struct buffer_head *bh,
>> ...
>> 	bh->b_data = page_address(page) + offset;
> 
> We don't need to kmap in order to access MOVABLE allocations.  kmap is
> only needed for HIGHMEM allocations.  So there's nothing wrong with ext4
> or set_bh_page().

Yeah, it's definitely not _buggy_.

Although, I do wonder what we should do about these for XPFO.  Should we
just stick a kmap() in there and comment it?  What we really need is a
mechanism to say "use this as a kernel page" and "stop using this as a
kernel page".  kmap() does that... kinda.  It's not a perfect fit, but
it's pretty close.