Messages in this thread | | | Date | Mon, 27 Feb 2012 21:55:17 -0500 | From | Steven Truelove <> | Subject | Re: [PATCH] HUGETLBFS: Align memory request to multiple of huge page size to avoid underallocating. |
| |
On 27/02/2012 6:42 PM, Andrew Morton wrote: > On Mon, 27 Feb 2012 09:09:28 -0500 > Steven Truelove<steven.truelove@utoronto.ca> wrote: > >> When calling shmget with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient. Modified hugetlb_file_setup to align requests to the huge page size. >> >> Signed-off-by: Steven Truelove<steven.truelove@utoronto.ca> >> --- >> fs/hugetlbfs/inode.c | 9 ++++++--- >> 1 files changed, 6 insertions(+), 3 deletions(-) >> >> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c >> index 1e85a7a..6c23f09 100644 >> --- a/fs/hugetlbfs/inode.c >> +++ b/fs/hugetlbfs/inode.c >> @@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size, >> struct path path; >> struct dentry *root; >> struct qstr quick_string; >> + struct hstate *hstate; >> + int num_pages; >> >> *user = NULL; >> if (!hugetlbfs_vfsmount) >> @@ -967,10 +969,11 @@ struct file *hugetlb_file_setup(const char *name, size_t size, >> if (!inode) >> goto out_dentry; >> >> + hstate = hstate_inode(inode); >> + num_pages = (size + huge_page_size(hstate) - 1)>> >> + huge_page_shift(hstate); >> error = -ENOMEM; >> - if (hugetlb_reserve_pages(inode, 0, >> - size>> huge_page_shift(hstate_inode(inode)), NULL, >> - acctflag)) >> + if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag)) >> goto out_inode; >> >> d_instantiate(path.dentry, inode); > A few things... > > - sys_mmap_pgoff() does the rounding up prior to calling > hugetlb_file_setup(). ipc/shm.c:newseg() does not. > > We should be consistent here: do it in the caller or the callee, > not both (or neither!). I guess doing it in the callee would be > best. > > - The above code could/should have used ALIGN(). Or round_up(): the > difference presently escapes me, even though it was so obvious that > we left all these things undocumented. > > - What's the point in aligning the length if we don't also look at > the start address? If that isn't a multiple of huge_page_size(), we > will need an additional page. >
Since mmap has an address to check and shmget does not, if the address is going to be checked it will need to be in the caller. If you like, I will leave the size check in hugetlb_file_setup() and remove the size check from mmap_pgoff, but replace it with a check of the address. That will centralize the common check (size of buffer), and let mmap_pgoff check the part that is unique to it. Patch shortly.
Steven Truelove
| |