Date: Mon, 23 Jun 2014 14:00:37 +0400 From: Dmitry Sivachenko <trtrmitya@gmail.com> To: hackers@freebsd.org Subject: Re: mlock() and vm_max_wired Message-ID: <C187B4D5-7D29-426F-BF90-06CFB023B702@gmail.com> In-Reply-To: <D85BD1F5-414C-45FA-B509-C55DC9BE125A@gmail.com> References: <D85BD1F5-414C-45FA-B509-C55DC9BE125A@gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 23 июня 2014 г., at 13:52, Dmitry Sivachenko <trtrmitya@gmail.com> wrote:
> Hello!
>
> I am using FreeBSD-10/stable and have a problem with mlock().
> I have 256GB of memory and a program which does mmap+mlock on a number of data files. It can process one file several times (so it does mmap() + mlock() on the same file more than once).
> I set vm.max_wired=67108864 (67108864 * 4k = 256GB), so it is allowed to mlock() the whole RAM.
>
> The total size of all files is about 180GB.
>
> The program fails with mlock: Resource temporarily unavailable error.
>
> If I increase vm.max_wired even more, the program works fine and after it starts top(1) shows about 186GB of Wired memory.
>
> Why does it fail with vm.max_wired=67108864? Is it a bug or am I missing something?
>
> Thanks.
Forgot to add the I suspect the following fragment in vm_mmap.c, in vm_mlock() function:
if (npages + cnt.v_wire_count > vm_page_max_wired)
return (EAGAIN);
If we mlock() the same region second time, this condition can be true though after mlock() these pages will not increase total locked pages counter.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?C187B4D5-7D29-426F-BF90-06CFB023B702>