On Fri, Nov 19, 2010 at 04:33:54PM -0700, Eric B Munson wrote:
> hugetlbfs_prefault relies on having a file descriptor associated with
> the huge page backed region to prefault the huge pages. This patch
> gives an alternative for prefaulting which will be used when the
> kernel supports MAP_HUGETLB.
>
> Signed-off-by: Eric B Munson <emun...@mgebm.net>
> ---
> hugeutils.c | 59
> ++++++++++++++++++++++++++++++++++++++---------------------
> 1 files changed, 38 insertions(+), 21 deletions(-)
>
> diff --git a/hugeutils.c b/hugeutils.c
> index bf9fb85..ecbfc29 100644
> --- a/hugeutils.c
> +++ b/hugeutils.c
> @@ -956,6 +956,15 @@ int hugetlbfs_unlinked_fd(void)
> #define IOV_LEN 64
> int hugetlbfs_prefault(int fd, void *addr, size_t length)
> {
> + size_t offset;
> + struct iovec iov[IOV_LEN];
> + int ret;
> + int i;
> + int close_fd = -1;
> +
> + if (!__hugetlbfs_prefault)
> + return 0;
> +
> /*
> * The NUMA users of libhugetlbfs' malloc feature are
> * expected to use the numactl program to specify an
> @@ -972,30 +981,38 @@ int hugetlbfs_prefault(int fd, void *addr, size_t
> length)
> * -ENOMEM is returned. The caller is expected to release the entire
> * mapping and optionally it may recover by mapping base pages instead.
> */
> - if (__hugetlbfs_prefault) {
> - int i;
> - size_t offset;
> - struct iovec iov[IOV_LEN];
> - int ret;
> -
> - for (offset = 0; offset < length; ) {
> - for (i = 0; i < IOV_LEN && offset < length; i++) {
> - iov[i].iov_base = addr + offset;
> - iov[i].iov_len = 1;
> - offset += gethugepagesize();
> - }
> - ret = readv(fd, iov, i);
> - if (ret != i) {
> - DEBUG("Got %d of %d requested; err=%d\n", ret,
> - i, ret < 0 ? errno : 0);
> - WARNING("Failed to reserve %ld huge pages "
> - "for new region\n",
> - length / gethugepagesize());
> - return -ENOMEM;
> - }
> +
> + if (fd < 0 && __hugetlb_opts.map_hugetlb) {
> + fd = open("/dev/zero", O_RDONLY);
> + if (fd < 0) {
> + ERROR("Failed to open /dev/zero for reading\n");
> + return -ENOMEM;
> + }
> + close_fd = fd;
> + }
> +
> + for (offset = 0; offset < length; ) {
> + for (i = 0; i < IOV_LEN && offset < length; i++) {
> + iov[i].iov_base = addr + offset;
> + iov[i].iov_len = 1;
> + offset += gethugepagesize();
> + }
> + ret = readv(fd, iov, i);
> + if (ret != i) {
> + DEBUG("Got %d of %d requested; err=%d\n", ret,
> + i, ret < 0 ? errno : 0);
> + WARNING("Failed to reserve %ld huge pages "
> + "for new region\n",
> + length / gethugepagesize());
> + if (close_fd >= 0)
> + close(close_fd);
> + return -ENOMEM;
> }
I'm missing something. How does opening /dev/zero guarantee that we are
checking for huge page availability? This looks like we are just
prefaulting some base pages. Should we consider MAP_HUGETLB and
prefaulting incompatible? That said, the availability of MAP_HUGETLB in
the kernel guarantees that proper reservations are also available due
to the timing the features were introduced.
> }
>
> + if (close_fd >= 0)
> + close(close_fd);
> +
> return 0;
> }
>
> --
> 1.7.1
>
>
> ------------------------------------------------------------------------------
> Beautiful is writing same markup. Internet Explorer 9 supports
> standards for HTML5, CSS3, SVG 1.1, ECMAScript5, and DOM L2 & L3.
> Spend less time writing and rewriting code and more time creating great
> experiences on the web. Be a part of the beta today
> http://p.sf.net/sfu/msIE9-sfdev2dev
> _______________________________________________
> Libhugetlbfs-devel mailing list
> Libhugetlbfs-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/libhugetlbfs-devel
>
--
Mel Gorman
Part-time Phd Student Linux Technology Center
University of Limerick IBM Dublin Software Lab
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Libhugetlbfs-devel mailing list
Libhugetlbfs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libhugetlbfs-devel
