Re: [PATCH] mm,hugetlb: compute page_size_log properly

From: Matthew Wilcox <willy@infradead.org>
To: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: akpm@linux-foundation.org, mhocko@suse.com, ak@linux.intel.com,
	mtk.manpages@gmail.com, linux-mm@kvack.org,
	linux-kernel@vger.kernel.org, Davidlohr Bueso <dbueso@suse.de>,
	khandual@linux.vnet.ibm.com
Subject: Re: [PATCH] mm,hugetlb: compute page_size_log properly
Date: Thu, 13 Apr 2017 05:46:33 -0700	[thread overview]
Message-ID: <20170413124633.GG784@bombadil.infradead.org> (raw)
In-Reply-To: <87wpaoq1zy.fsf@skywalker.in.ibm.com>

On Thu, Apr 13, 2017 at 11:32:09AM +0530, Aneesh Kumar K.V wrote:
> > +#define SHM_HUGE_SHIFT	26
> > +#define SHM_HUGE_MASK	0x3f
> > +#define SHM_HUGE_2MB	(21 << SHM_HUGE_SHIFT)
> > +#define SHM_HUGE_8MB	(23 << SHM_HUGE_SHIFT)
> > +#define SHM_HUGE_1GB	(30 << SHM_HUGE_SHIFT)
> > +#define SHM_HUGE_16GB	(34 << SHM_HUGE_SHIFT)
> 
> This should be in arch/uapi like MAP_HUGE_2M ? That will let arch add
> #defines based on the hugepae size supported by them.

Well, what do we want to happen if source code contains SHM_HUGE_2MB?
Do we want it to fail to compile on ppc, or do we want it to request 2MB
pages and get ... hmm, looks like it fails at runtime (size_to_hstate
ends up returning NULL).  So, yeah, looks like a compile-time failure
would be better.

But speaking of MAP_HUGE_, the only definitions so far are in arch/x86.
Are you going to add the ppc versions?

Also, which header file?  I'm reluctant to add a new header, but
asm/shmbuf.h doesn't seem like a great place to put it.