lkml.org 
[lkml]   [2007]   [Jun]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [shm][hugetlb] Fix get_policy for stacked shared memory files
Date
"Adam Litke" <aglitke@gmail.com> writes:

> On 6/12/07, Eric W. Biederman <ebiederm@xmission.com> wrote:
>> Adam Litke <agl@us.ibm.com> writes:
>>
>> > Here's another breakage as a result of shared memory stacked files :(
>> >
>> > The NUMA policy for a VMA is determined by checking the following (in the
> order
>> > given):

Where that doesn't appear to be what the current code does.
If there is a VMA it doesn't appear that we check task->mempolicy.

>> > 1) vma->vm_ops->get_policy() (if defined)
>> > 2) vma->vm_policy (if defined)
>> > 3) task->mempolicy (if defined)
>> > 4) Fall back to default_policy
>> >
>> > By switching to stacked files for shared memory, get_policy() is now always
> set
>> > to shm_get_policy which is a wrapper function. This causes us to stop at
> step
>> > 1, which yields NULL for hugetlb instead of task->mempolicy which was the
>> > previous (and correct) result.
>> >
>> > This patch modifies the shm_get_policy() wrapper to maintain steps 1-3 for
> the
>> > wrapped vm_ops. Andi and Christoph, does this look right to you?
>>
>> I'm confused.
>>
>> I agree that the behavior you describe is correct.
>> However I only see two code paths were get_policy is called and
>> both of them take a NULL result and change it to task->mempolicy:
>
> The coffee hasn't quite absorbed yet, but don't those two code paths
> take a NULL result from get_policy() and turn it into default_policy,
> not task->mempolicy?

True, a NULL is turned into default_policy. However the basic
confusion remains. I don't see how we ever return task->mempolicy
on those paths if we have a vma, and those appear to be the only
call sites of get_policy.

>> From mm/mempolicy.c
>>
>> > long do_get_mempolicy(int *policy, nodemask_t *nmask,
>> > unsigned long addr, unsigned long flags)
>> > {
>> > int err;
>> > struct mm_struct *mm = current->mm;
>> > struct vm_area_struct *vma = NULL;
>> > struct mempolicy *pol = current->mempolicy;
>> >
>> > cpuset_update_task_memory_state();
>> > if (flags & ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR))
>> > return -EINVAL;
>> > if (flags & MPOL_F_ADDR) {
>> > down_read(&mm->mmap_sem);
>> > vma = find_vma_intersection(mm, addr, addr+1);
>> > if (!vma) {
>> > up_read(&mm->mmap_sem);
>> > return -EFAULT;
>> > }
>> > if (vma->vm_ops && vma->vm_ops->get_policy)
>> > pol = vma->vm_ops->get_policy(vma, addr);
>> > else
>> > pol = vma->vm_policy;
>> > } else if (addr)
>> > return -EINVAL;
>> >
>> > if (!pol)
>> > pol = &default_policy;
>>
>>
>>
>> > /* Return effective policy for a VMA */
>> > static struct mempolicy * get_vma_policy(struct task_struct *task,
>> > struct vm_area_struct *vma, unsigned long addr)
>> > {
>> > struct mempolicy *pol = task->mempolicy;
>> >
>> > if (vma) {
>> > if (vma->vm_ops && vma->vm_ops->get_policy)
>> > pol = vma->vm_ops->get_policy(vma, addr);
>> > else if (vma->vm_policy &&
>> > vma->vm_policy->policy != MPOL_DEFAULT)
>> > pol = vma->vm_policy;
>> > }
>> > if (!pol)
>> > pol = &default_policy;
>> > return pol;
>> > }
>>
>>
>> Does this perhaps need to be:
>> > Signed-off-by: Adam Litke <agl@us.ibm.com>
>> >
>> > diff --git a/ipc/shm.c b/ipc/shm.c
>> > index 4fefbad..8d2672d 100644
>> > --- a/ipc/shm.c
>> > +++ b/ipc/shm.c
>> > @@ -254,8 +254,10 @@ struct mempolicy *shm_get_policy(struct vm_area_struct
>> > *vma, unsigned long addr)
>>
>> + pol = NULL;
>> >
>> > if (sfd->vm_ops->get_policy)
>> > pol = sfd->vm_ops->get_policy(vma, addr);
>> > - else
>> > + else if (vma->vm_policy && vma->vm_policy->policy != MPOL_DEFAULT)
>> > pol = vma->vm_policy;
>> > return pol;
>> > }
>> > #endif
>
> afaict this would provide no way for pol to be set to task->mempolicy
> for hugetlb per my comment above.

Being dense I don't see a way for us to have ever returned task->mempolicy.

So perhaps something else changed. Or something was made more
consistent and another bug was revealed.

I don't truly see the bug in shm_get_policy. I can see why it would
not have the expected affect but that is different.

Eric
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2007-06-12 20:25    [W:0.104 / U:1.956 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site