Re: [PATCH v5 1/2] mm: Allow the page cache to allocate large pages

From: Matthew Wilcox <willy@infradead.org>
To: Michal Hocko <mhocko@kernel.org>
Cc: William Kucharski <william.kucharski@oracle.com>,
	linux-kernel@vger.kernel.org, linux-mm@kvack.org,
	linux-fsdevel@vger.kernel.org,
	Dave Hansen <dave.hansen@linux.intel.com>,
	Song Liu <songliubraving@fb.com>,
	Bob Kasten <robert.a.kasten@intel.com>,
	Mike Kravetz <mike.kravetz@oracle.com>,
	Chad Mynhier <chad.mynhier@oracle.com>,
	"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
	Johannes Weiner <jweiner@fb.com>
Subject: Re: [PATCH v5 1/2] mm: Allow the page cache to allocate large pages
Date: Tue, 3 Sep 2019 09:28:31 -0700	[thread overview]
Message-ID: <20190903162831.GI29434@bombadil.infradead.org> (raw)
In-Reply-To: <20190903121952.GU14028@dhcp22.suse.cz>

On Tue, Sep 03, 2019 at 02:19:52PM +0200, Michal Hocko wrote:
> On Tue 03-09-19 05:11:55, Matthew Wilcox wrote:
> > On Tue, Sep 03, 2019 at 01:57:48PM +0200, Michal Hocko wrote:
> > > On Mon 02-09-19 03:23:40, William Kucharski wrote:
> > > > Add an 'order' argument to __page_cache_alloc() and
> > > > do_read_cache_page(). Ensure the allocated pages are compound pages.
> > > 
> > > Why do we need to touch all the existing callers and change them to use
> > > order 0 when none is actually converted to a different order? This just
> > > seem to add a lot of code churn without a good reason. If anything I
> > > would simply add __page_cache_alloc_order and make __page_cache_alloc
> > > call it with order 0 argument.
> > 
> > Patch 2/2 uses a non-zero order.
> 
> It is a new caller and it can use a new function right?
> 
> > I agree it's a lot of churn without
> > good reason; that's why I tried to add GFP_ORDER flags a few months ago.
> > Unfortunately, you didn't like that approach either.
> 
> Is there any future plan that all/most __page_cache_alloc will get a
> non-zero order argument?

I'm not sure about "most".  It will certainly become more common, as
far as I can tell.

> > > Also is it so much to ask callers to provide __GFP_COMP explicitly?
> > 
> > Yes, it's an unreasonable burden on the callers.
> 
> Care to exaplain why? __GFP_COMP tends to be used in the kernel quite
> extensively.

Most of the places which call this function get their gfp_t from
mapping->gfp_mask.  If we only want to allocate a single page, we
must not set __GFP_COMP.  If we want to allocate a large page, we must
set __GFP_COMP.  Rather than require individual filesystems to concern
themselves with this wart of the GFP interface, we can solve it in the
page cache.