Hi Muchun,
On 07/28/20 at 11:49am, Muchun Song wrote:
> In the reservation routine, we only check whether the cpuset meets
> the memory allocation requirements. But we ignore the mempolicy of
> MPOL_BIND case. If someone mmap hugetlb succeeds, but the subsequent
> memory allocation may fail due to mempolicy restrictions and receives
> the SIGBUS signal. This can be reproduced by the follow steps.
>
> 1) Compile the test case.
> cd tools/testing/selftests/vm/
> gcc map_hugetlb.c -o map_hugetlb
>
> 2) Pre-allocate huge pages. Suppose there are 2 numa nodes in the
> system. Each node will pre-allocate one huge page.
> echo 2 > /proc/sys/vm/nr_hugepages
>
> 3) Run test case(mmap 4MB). We receive the SIGBUS signal.
> numactl --membind=0 ./map_hugetlb 4
I think supporting the mempolicy of MPOL_BIND case is a good idea.
I am wondering what about the other mempolicy cases, e.g MPOL_INTERLEAVE,
MPOL_PREFERRED. Asking these because we already have similar handling in
sysfs, proc nr_hugepages_mempolicy writting. Please see
__nr_hugepages_store_common() for detail.
Thanks
Baoquan
>
> With this patch applied, the mmap will fail in the step 3) and throw
> "mmap: Cannot allocate memory".
>
> Signed-off-by: Muchun Song <[email protected]>
> Reported-by: Jianchao Guo <[email protected]>
> Suggested-by: Michal Hocko <[email protected]>
> Reviewed-by: Mike Kravetz <[email protected]>
> ---
> changelog in v4:
> 1) Fix compilation errors with !CONFIG_NUMA.
>
> changelog in v3:
> 1) Do not allocate nodemask on the stack.
> 2) Update comment.
>
> changelog in v2:
> 1) Reuse policy_nodemask().
>
> include/linux/mempolicy.h | 14 ++++++++++++++
> mm/hugetlb.c | 22 ++++++++++++++++++----
> mm/mempolicy.c | 2 +-
> 3 files changed, 33 insertions(+), 5 deletions(-)
>
> diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
> index ea9c15b60a96..0656ece1ccf1 100644
> --- a/include/linux/mempolicy.h
> +++ b/include/linux/mempolicy.h
> @@ -152,6 +152,15 @@ extern int huge_node(struct vm_area_struct *vma,
> extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
> extern bool mempolicy_nodemask_intersects(struct task_struct *tsk,
> const nodemask_t *mask);
> +extern nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy);
> +
> +static inline nodemask_t *policy_nodemask_current(gfp_t gfp)
> +{
> + struct mempolicy *mpol = get_task_policy(current);
> +
> + return policy_nodemask(gfp, mpol);
> +}
> +
> extern unsigned int mempolicy_slab_node(void);
>
> extern enum zone_type policy_zone;
> @@ -281,5 +290,10 @@ static inline int mpol_misplaced(struct page *page, struct vm_area_struct *vma,
> static inline void mpol_put_task_policy(struct task_struct *task)
> {
> }
> +
> +static inline nodemask_t *policy_nodemask_current(gfp_t gfp)
> +{
> + return NULL;
> +}
> #endif /* CONFIG_NUMA */
> #endif
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 589c330df4db..a34458f6a475 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -3463,13 +3463,21 @@ static int __init default_hugepagesz_setup(char *s)
> }
> __setup("default_hugepagesz=", default_hugepagesz_setup);
>
> -static unsigned int cpuset_mems_nr(unsigned int *array)
> +static unsigned int allowed_mems_nr(struct hstate *h)
> {
> int node;
> unsigned int nr = 0;
> + nodemask_t *mpol_allowed;
> + unsigned int *array = h->free_huge_pages_node;
> + gfp_t gfp_mask = htlb_alloc_mask(h);
> +
> + mpol_allowed = policy_nodemask_current(gfp_mask);
>
> - for_each_node_mask(node, cpuset_current_mems_allowed)
> - nr += array[node];
> + for_each_node_mask(node, cpuset_current_mems_allowed) {
> + if (!mpol_allowed ||
> + (mpol_allowed && node_isset(node, *mpol_allowed)))
> + nr += array[node];
> + }
>
> return nr;
> }
> @@ -3648,12 +3656,18 @@ static int hugetlb_acct_memory(struct hstate *h, long delta)
> * we fall back to check against current free page availability as
> * a best attempt and hopefully to minimize the impact of changing
> * semantics that cpuset has.
> + *
> + * Apart from cpuset, we also have memory policy mechanism that
> + * also determines from which node the kernel will allocate memory
> + * in a NUMA system. So similar to cpuset, we also should consider
> + * the memory policy of the current task. Similar to the description
> + * above.
> */
> if (delta > 0) {
> if (gather_surplus_pages(h, delta) < 0)
> goto out;
>
> - if (delta > cpuset_mems_nr(h->free_huge_pages_node)) {
> + if (delta > allowed_mems_nr(h)) {
> return_unused_surplus_pages(h, delta);
> goto out;
> }
> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> index 93fcfc1f2fa2..fce14c3f4f38 100644
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -1873,7 +1873,7 @@ static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
> * Return a nodemask representing a mempolicy for filtering nodes for
> * page allocation
> */
> -static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
> +nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
> {
> /* Lower zones don't get a nodemask applied for MPOL_BIND */
> if (unlikely(policy->mode == MPOL_BIND) &&
> --
> 2.11.0
>
>
On Tue, Jul 28, 2020 at 9:25 PM Baoquan He <[email protected]> wrote:
>
> Hi Muchun,
>
> On 07/28/20 at 11:49am, Muchun Song wrote:
> > In the reservation routine, we only check whether the cpuset meets
> > the memory allocation requirements. But we ignore the mempolicy of
> > MPOL_BIND case. If someone mmap hugetlb succeeds, but the subsequent
> > memory allocation may fail due to mempolicy restrictions and receives
> > the SIGBUS signal. This can be reproduced by the follow steps.
> >
> > 1) Compile the test case.
> > cd tools/testing/selftests/vm/
> > gcc map_hugetlb.c -o map_hugetlb
> >
> > 2) Pre-allocate huge pages. Suppose there are 2 numa nodes in the
> > system. Each node will pre-allocate one huge page.
> > echo 2 > /proc/sys/vm/nr_hugepages
> >
> > 3) Run test case(mmap 4MB). We receive the SIGBUS signal.
> > numactl --membind=0 ./map_hugetlb 4
>
> I think supporting the mempolicy of MPOL_BIND case is a good idea.
> I am wondering what about the other mempolicy cases, e.g MPOL_INTERLEAVE,
> MPOL_PREFERRED. Asking these because we already have similar handling in
> sysfs, proc nr_hugepages_mempolicy writting. Please see
> __nr_hugepages_store_common() for detail.
Yeah, I know the nr_hugepages_mempolicy. But this new code will
help produce a quick failure as described in the commit message
instead of waiting until the page fault routine(and receive a SIGBUG
signal).
>
> Thanks
> Baoquan
>
> >
> > With this patch applied, the mmap will fail in the step 3) and throw
> > "mmap: Cannot allocate memory".
> >
> > Signed-off-by: Muchun Song <[email protected]>
> > Reported-by: Jianchao Guo <[email protected]>
> > Suggested-by: Michal Hocko <[email protected]>
> > Reviewed-by: Mike Kravetz <[email protected]>
> > ---
> > changelog in v4:
> > 1) Fix compilation errors with !CONFIG_NUMA.
> >
> > changelog in v3:
> > 1) Do not allocate nodemask on the stack.
> > 2) Update comment.
> >
> > changelog in v2:
> > 1) Reuse policy_nodemask().
> >
> > include/linux/mempolicy.h | 14 ++++++++++++++
> > mm/hugetlb.c | 22 ++++++++++++++++++----
> > mm/mempolicy.c | 2 +-
> > 3 files changed, 33 insertions(+), 5 deletions(-)
> >
> > diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
> > index ea9c15b60a96..0656ece1ccf1 100644
> > --- a/include/linux/mempolicy.h
> > +++ b/include/linux/mempolicy.h
> > @@ -152,6 +152,15 @@ extern int huge_node(struct vm_area_struct *vma,
> > extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
> > extern bool mempolicy_nodemask_intersects(struct task_struct *tsk,
> > const nodemask_t *mask);
> > +extern nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy);
> > +
> > +static inline nodemask_t *policy_nodemask_current(gfp_t gfp)
> > +{
> > + struct mempolicy *mpol = get_task_policy(current);
> > +
> > + return policy_nodemask(gfp, mpol);
> > +}
> > +
> > extern unsigned int mempolicy_slab_node(void);
> >
> > extern enum zone_type policy_zone;
> > @@ -281,5 +290,10 @@ static inline int mpol_misplaced(struct page *page, struct vm_area_struct *vma,
> > static inline void mpol_put_task_policy(struct task_struct *task)
> > {
> > }
> > +
> > +static inline nodemask_t *policy_nodemask_current(gfp_t gfp)
> > +{
> > + return NULL;
> > +}
> > #endif /* CONFIG_NUMA */
> > #endif
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index 589c330df4db..a34458f6a475 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -3463,13 +3463,21 @@ static int __init default_hugepagesz_setup(char *s)
> > }
> > __setup("default_hugepagesz=", default_hugepagesz_setup);
> >
> > -static unsigned int cpuset_mems_nr(unsigned int *array)
> > +static unsigned int allowed_mems_nr(struct hstate *h)
> > {
> > int node;
> > unsigned int nr = 0;
> > + nodemask_t *mpol_allowed;
> > + unsigned int *array = h->free_huge_pages_node;
> > + gfp_t gfp_mask = htlb_alloc_mask(h);
> > +
> > + mpol_allowed = policy_nodemask_current(gfp_mask);
> >
> > - for_each_node_mask(node, cpuset_current_mems_allowed)
> > - nr += array[node];
> > + for_each_node_mask(node, cpuset_current_mems_allowed) {
> > + if (!mpol_allowed ||
> > + (mpol_allowed && node_isset(node, *mpol_allowed)))
> > + nr += array[node];
> > + }
> >
> > return nr;
> > }
> > @@ -3648,12 +3656,18 @@ static int hugetlb_acct_memory(struct hstate *h, long delta)
> > * we fall back to check against current free page availability as
> > * a best attempt and hopefully to minimize the impact of changing
> > * semantics that cpuset has.
> > + *
> > + * Apart from cpuset, we also have memory policy mechanism that
> > + * also determines from which node the kernel will allocate memory
> > + * in a NUMA system. So similar to cpuset, we also should consider
> > + * the memory policy of the current task. Similar to the description
> > + * above.
> > */
> > if (delta > 0) {
> > if (gather_surplus_pages(h, delta) < 0)
> > goto out;
> >
> > - if (delta > cpuset_mems_nr(h->free_huge_pages_node)) {
> > + if (delta > allowed_mems_nr(h)) {
> > return_unused_surplus_pages(h, delta);
> > goto out;
> > }
> > diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> > index 93fcfc1f2fa2..fce14c3f4f38 100644
> > --- a/mm/mempolicy.c
> > +++ b/mm/mempolicy.c
> > @@ -1873,7 +1873,7 @@ static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
> > * Return a nodemask representing a mempolicy for filtering nodes for
> > * page allocation
> > */
> > -static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
> > +nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
> > {
> > /* Lower zones don't get a nodemask applied for MPOL_BIND */
> > if (unlikely(policy->mode == MPOL_BIND) &&
> > --
> > 2.11.0
> >
> >
>
--
Yours,
Muchun
On 7/28/20 6:24 AM, Baoquan He wrote:
> Hi Muchun,
>
> On 07/28/20 at 11:49am, Muchun Song wrote:
>> In the reservation routine, we only check whether the cpuset meets
>> the memory allocation requirements. But we ignore the mempolicy of
>> MPOL_BIND case. If someone mmap hugetlb succeeds, but the subsequent
>> memory allocation may fail due to mempolicy restrictions and receives
>> the SIGBUS signal. This can be reproduced by the follow steps.
>>
>> 1) Compile the test case.
>> cd tools/testing/selftests/vm/
>> gcc map_hugetlb.c -o map_hugetlb
>>
>> 2) Pre-allocate huge pages. Suppose there are 2 numa nodes in the
>> system. Each node will pre-allocate one huge page.
>> echo 2 > /proc/sys/vm/nr_hugepages
>>
>> 3) Run test case(mmap 4MB). We receive the SIGBUS signal.
>> numactl --membind=0 ./map_hugetlb 4
>
> I think supporting the mempolicy of MPOL_BIND case is a good idea.
> I am wondering what about the other mempolicy cases, e.g MPOL_INTERLEAVE,
> MPOL_PREFERRED. Asking these because we already have similar handling in
> sysfs, proc nr_hugepages_mempolicy writting. Please see
> __nr_hugepages_store_common() for detail.
There is a high level difference in the function of this code and the code
called by the sysfs and proc interfaces. This patch is dealing with reserving
huge pages in the pool for later use. The sysfs and proc interfaces are
allocating huge pages to be added to the pool.
Using mempolicy to decide how to allocate huge pages is pretty straight
forward. Using mempolicy to reserve pages is almost impossible to get
correct. The comment at the beginning of hugetlb_acct_memory() and modified
by this patch summarizes the issues.
IMO, at this time it makes little sense to perform checks for more than
MPOL_BIND at reservation time. If we ever take on the monumental task of
supporting mempolicy directed per-node reservations throughout the life of
a process, support for other policies will need to be taken into account.
--
Mike Kravetz
On 07/28/20 at 09:46am, Mike Kravetz wrote:
> On 7/28/20 6:24 AM, Baoquan He wrote:
> > Hi Muchun,
> >
> > On 07/28/20 at 11:49am, Muchun Song wrote:
> >> In the reservation routine, we only check whether the cpuset meets
> >> the memory allocation requirements. But we ignore the mempolicy of
> >> MPOL_BIND case. If someone mmap hugetlb succeeds, but the subsequent
> >> memory allocation may fail due to mempolicy restrictions and receives
> >> the SIGBUS signal. This can be reproduced by the follow steps.
> >>
> >> 1) Compile the test case.
> >> cd tools/testing/selftests/vm/
> >> gcc map_hugetlb.c -o map_hugetlb
> >>
> >> 2) Pre-allocate huge pages. Suppose there are 2 numa nodes in the
> >> system. Each node will pre-allocate one huge page.
> >> echo 2 > /proc/sys/vm/nr_hugepages
> >>
> >> 3) Run test case(mmap 4MB). We receive the SIGBUS signal.
> >> numactl --membind=0 ./map_hugetlb 4
> >
> > I think supporting the mempolicy of MPOL_BIND case is a good idea.
> > I am wondering what about the other mempolicy cases, e.g MPOL_INTERLEAVE,
> > MPOL_PREFERRED. Asking these because we already have similar handling in
> > sysfs, proc nr_hugepages_mempolicy writting. Please see
> > __nr_hugepages_store_common() for detail.
>
> There is a high level difference in the function of this code and the code
> called by the sysfs and proc interfaces. This patch is dealing with reserving
> huge pages in the pool for later use. The sysfs and proc interfaces are
> allocating huge pages to be added to the pool.
>
> Using mempolicy to decide how to allocate huge pages is pretty straight
> forward. Using mempolicy to reserve pages is almost impossible to get
> correct. The comment at the beginning of hugetlb_acct_memory() and modified
> by this patch summarizes the issues.
>
> IMO, at this time it makes little sense to perform checks for more than
> MPOL_BIND at reservation time. If we ever take on the monumental task of
> supporting mempolicy directed per-node reservations throughout the life of
> a process, support for other policies will need to be taken into account.
I haven't figured out the difficulty of using mempolicy very clearly, will
read more codes and digest and understand your words. Thanks a lot for
these details.
Thanks
Baoquan