On Wed, 22 Jul 2015, Kirill A. Shutemov wrote:
> On Tue, Jul 21, 2015 at 03:59:38PM -0400, Eric B Munson wrote:
> > The upcoming mlock(MLOCK_ONFAULT) implementation will need a way to
> > request that all present pages in a range are locked without faulting in
> > pages that are not present. This logic is very close to what the
> > __mm_populate() call handles without faulting pages so the patch pulls
> > out the pieces that can be shared and adds mm_lock_present() to gup.c.
> > The following patch will call it from do_mlock() when MLOCK_ONFAULT is
> > specified.
> >
> > Signed-off-by: Eric B Munson <emunson@xxxxxxxxxx>
> > Cc: Jonathan Corbet <corbet@xxxxxxx>
> > Cc: Vlastimil Babka <vbabka@xxxxxxx>
> > Cc: linux-mm@xxxxxxxxx
> > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > ---
> > mm/gup.c | 172 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++------
> > 1 file changed, 157 insertions(+), 15 deletions(-)
>
> I don't like that you've copy-pasted a lot of code. I think it can be
> solved with new foll flags.
>
> Totally untested patch below split out mlock part of FOLL_POPULATE into
> new FOLL_MLOCK flag. FOLL_POPULATE | FOLL_MLOCK will do what currently
> FOLL_POPULATE does. The new MLOCK_ONFAULT can use just FOLL_MLOCK. It will
> not trigger fault in.
I originally tried to do this by adding a check for VM_LOCKONFAULT in
__get_user_pages() before the call to faultin_page() which would goto
next_page if LOCKONFAULT was specified. With the early out in
__get_user_pages(), all of the tests using lock on fault failed to lock
pages. I will try with a new FOLL flag and see if that can work out.
>
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index c3a2b37365f6..c3834cddfcc7 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -2002,6 +2002,7 @@ static inline struct page *follow_page(struct vm_area_struct *vma,
> #define FOLL_NUMA 0x200 /* force NUMA hinting page fault */
> #define FOLL_MIGRATION 0x400 /* wait for page to replace migration entry */
> #define FOLL_TRIED 0x800 /* a retry, previous pass started an IO */
> +#define FOLL_MLOCK 0x1000 /* mlock the page if the VMA is VM_LOCKED */
>
> typedef int (*pte_fn_t)(pte_t *pte, pgtable_t token, unsigned long addr,
> void *data);
> diff --git a/mm/gup.c b/mm/gup.c
> index a798293fc648..4c7ff23947b9 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -129,7 +129,7 @@ retry:
> */
> mark_page_accessed(page);
> }
> - if ((flags & FOLL_POPULATE) && (vma->vm_flags & VM_LOCKED)) {
> + if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
> /*
> * The preliminary mapping check is mainly to avoid the
> * pointless overhead of lock_page on the ZERO_PAGE
> @@ -299,6 +299,9 @@ static int faultin_page(struct task_struct *tsk, struct vm_area_struct *vma,
> unsigned int fault_flags = 0;
> int ret;
>
> + /* mlock present pages, but not fault in new one */
> + if ((*flags & (FOLL_POPULATE | FOLL_MLOCK)) == FOLL_MLOCK)
> + return -ENOENT;
> /* For mm_populate(), just skip the stack guard page. */
> if ((*flags & FOLL_POPULATE) &&
> (stack_guard_page_start(vma, address) ||
> @@ -890,7 +893,7 @@ long populate_vma_page_range(struct vm_area_struct *vma,
> VM_BUG_ON_VMA(end > vma->vm_end, vma);
> VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
>
> - gup_flags = FOLL_TOUCH | FOLL_POPULATE;
> + gup_flags = FOLL_TOUCH | FOLL_POPULATE | FOLL_MLOCK;
> /*
> * We want to touch writable mappings with a write fault in order
> * to break COW, except for shared mappings because these don't COW
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 8f9a334a6c66..9eeb3bd304fc 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -1306,7 +1306,7 @@ struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
> pmd, _pmd, 1))
> update_mmu_cache_pmd(vma, addr, pmd);
> }
> - if ((flags & FOLL_POPULATE) && (vma->vm_flags & VM_LOCKED)) {
> + if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
> if (page->mapping && trylock_page(page)) {
> lru_add_drain();
> if (page->mapping)
>
> --
> Kirill A. Shutemov
Attachment:
signature.asc
Description: Digital signature