shmem/tmpfs uses pseudo vma to allocate page with correct NUMA policy.
The pseudo vma doesn't have vm_page_prot set. We are going to encode
encryption KeyID in vm_page_prot. Having garbage there causes problems.
Zero out all unused fields in the pseudo vma.
Signed-off-by: Kirill A. Shutemov <[email protected]>
---
mm/shmem.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/mm/shmem.c b/mm/shmem.c
index 9d6c7e595415..693fb82b4b42 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1404,10 +1404,9 @@ static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
struct shmem_inode_info *info, pgoff_t index)
{
/* Create a pseudo vma that just contains the policy */
- vma->vm_start = 0;
+ memset(vma, 0, sizeof(*vma));
/* Bias interleave by inode number to distribute better across nodes */
vma->vm_pgoff = index + info->vfs_inode.i_ino;
- vma->vm_ops = NULL;
vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
}
--
2.17.0
On Thu, 31 May 2018, Kirill A. Shutemov wrote:
> shmem/tmpfs uses pseudo vma to allocate page with correct NUMA policy.
>
> The pseudo vma doesn't have vm_page_prot set. We are going to encode
> encryption KeyID in vm_page_prot. Having garbage there causes problems.
>
> Zero out all unused fields in the pseudo vma.
>
> Signed-off-by: Kirill A. Shutemov <[email protected]>
I won't go so far as to say NAK, but personally I much prefer that we
document what fields actually get used, by initializing only those,
rather than having such a blanket memset.
And you say "We are going to ...": so this should really be part of
some future patchset, shouldn't it?
My opinion might be in the minority: you remind me of a similar
request from Josef some while ago, Cc'ing him.
(I'm very ashamed, by the way, of shmem's pseudo-vma, I think it's
horrid, and just reflects that shmem was an afterthought when NUMA
mempolicies were designed. Internally, we replaced alloc_pages_vma()
throughout by alloc_pages_mpol(), which has no need for pseudo-vmas,
and the advantage of dropping mmap_sem across the bulk of NUMA page
migration. I shall be updating that work in coming months, and hope
to upstream, but no promise from me on the timing - your need for
vm_page_prot likely much sooner.)
Hugh
> ---
> mm/shmem.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 9d6c7e595415..693fb82b4b42 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1404,10 +1404,9 @@ static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
> struct shmem_inode_info *info, pgoff_t index)
> {
> /* Create a pseudo vma that just contains the policy */
> - vma->vm_start = 0;
> + memset(vma, 0, sizeof(*vma));
> /* Bias interleave by inode number to distribute better across nodes */
> vma->vm_pgoff = index + info->vfs_inode.i_ino;
> - vma->vm_ops = NULL;
> vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
> }
>
> --
> 2.17.0
On Thu, 31 May 2018 16:56:02 +0300 "Kirill A. Shutemov" <[email protected]> wrote:
> shmem/tmpfs uses pseudo vma to allocate page with correct NUMA policy.
>
> The pseudo vma doesn't have vm_page_prot set. We are going to encode
> encryption KeyID in vm_page_prot. Having garbage there causes problems.
>
> Zero out all unused fields in the pseudo vma.
>
So there are no known problems in the current mainline kernel?
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1404,10 +1404,9 @@ static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
> struct shmem_inode_info *info, pgoff_t index)
> {
> /* Create a pseudo vma that just contains the policy */
> - vma->vm_start = 0;
> + memset(vma, 0, sizeof(*vma));
> /* Bias interleave by inode number to distribute better across nodes */
> vma->vm_pgoff = index + info->vfs_inode.i_ino;
> - vma->vm_ops = NULL;
> vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
> }
On Thu, 31 May 2018, Andrew Morton wrote:
> On Thu, 31 May 2018 16:56:02 +0300 "Kirill A. Shutemov" <[email protected]> wrote:
>
> > shmem/tmpfs uses pseudo vma to allocate page with correct NUMA policy.
> >
> > The pseudo vma doesn't have vm_page_prot set. We are going to encode
> > encryption KeyID in vm_page_prot. Having garbage there causes problems.
> >
> > Zero out all unused fields in the pseudo vma.
> >
>
> So there are no known problems in the current mainline kernel?
Correct - if we limit ourselves to the area of the shmem pseudo-vma :)
Hugh
On Thu, May 31, 2018 at 10:50:36PM +0000, Hugh Dickins wrote:
> On Thu, 31 May 2018, Kirill A. Shutemov wrote:
>
> > shmem/tmpfs uses pseudo vma to allocate page with correct NUMA policy.
> >
> > The pseudo vma doesn't have vm_page_prot set. We are going to encode
> > encryption KeyID in vm_page_prot. Having garbage there causes problems.
> >
> > Zero out all unused fields in the pseudo vma.
> >
> > Signed-off-by: Kirill A. Shutemov <[email protected]>
>
> I won't go so far as to say NAK, but personally I much prefer that we
> document what fields actually get used, by initializing only those,
> rather than having such a blanket memset.
I recognize value of documentation here. But I still think leaving garbage
in the fields is not a great idea.
>
> And you say "We are going to ...": so this should really be part of
> some future patchset, shouldn't it?
Yeah. It's for MKTME. I just try to push easy patches first.
> My opinion might be in the minority: you remind me of a similar
> request from Josef some while ago, Cc'ing him.
>
> (I'm very ashamed, by the way, of shmem's pseudo-vma, I think it's
> horrid, and just reflects that shmem was an afterthought when NUMA
> mempolicies were designed. Internally, we replaced alloc_pages_vma()
> throughout by alloc_pages_mpol(), which has no need for pseudo-vmas,
> and the advantage of dropping mmap_sem across the bulk of NUMA page
> migration. I shall be updating that work in coming months, and hope
> to upstream, but no promise from me on the timing - your need for
> vm_page_prot likely much sooner.)
I will try to look at how we can get alloc_pages_mpol() implemented.
(Although interleave bias is kinda confusing. I'll need to wrap my head
around the thing.)
--
Kirill A. Shutemov