Message-ID: <539A1CDA.5000709@hp.com>
Date: Thu, 12 Jun 2014 17:34:18 -0400
From: Waiman Long <waiman.long@hp.com>
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:10.0.12) Gecko/20130109 Thunderbird/10.0.12
MIME-Version: 1.0
To: Andrew Morton <akpm@linux-foundation.org>
CC: Mel Gorman <mgorman@suse.de>, Rik van Riel <riel@redhat.com>,
        Ingo Molnar <mingo@kernel.org>, Peter Zijlstra <peterz@infradead.org>,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        Scott J Norton <scott.norton@hp.com>
Subject: Re: [PATCH] mm: Move __vma_address() to internal.h to be inlined
 in huge_memory.c
References: <1402600540-52031-1-git-send-email-Waiman.Long@hp.com> <20140612122546.cfdebdb22bb22c0f767e30b5@linux-foundation.org>
In-Reply-To: <20140612122546.cfdebdb22bb22c0f767e30b5@linux-foundation.org>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org

On 06/12/2014 03:25 PM, Andrew Morton wrote:
> On Thu, 12 Jun 2014 15:15:40 -0400 Waiman Long<Waiman.Long@hp.com>  wrote:
>
>> The vma_address() function which is used to compute the virtual address
>> within a VMA is used only by 2 files in the mm subsystem - rmap.c and
>> huge_memory.c. This function is defined in rmap.c and is inlined by
>> its callers there, but it is also declared as an external function.
>>
>> However, the __split_huge_page() function which calls vma_address()
>> in huge_memory.c is calling it as a real function call. This is not
>> as efficient as an inlined function. This patch moves the underlying
>> inlined __vma_address() function to internal.h to be shared by both
>> the rmap.c and huge_memory.c file.
> This increases huge_memory.o's text+data_bss by 311 bytes, which makes
> me suspect that it is a bad change due to its increase of kernel cache
> footprint.
>
> Perhaps we should be noinlining __vma_address()?

On my test machine, I saw an increase of 144 bytes in the text segment
of huge_memory.o. The size in size is caused by an increase in the size
of the __split_huge_page function. When I remove the

         if (unlikely(is_vm_hugetlb_page(vma)))
                 pgoff = page->index << huge_page_order(page_hstate(page));

check, the increase in size drops down to 24 bytes. As a THP cannot be
a hugetlb page, there is no point in doing this check for a THP. I will
update the patch to pass in an additional argument to disable this
check for __split_huge_page.

-Longman

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/