When tmpfs has the memory policy interleaved it always starts allocating at each
file at node 0. When there are many small files the lower nodes fill up
disproportionately.
This patch attempts to spread out node usage by starting files at nodes other
then 0. I disturbed the addr parameter since alloc_pages_vma will only use it
when the policy is MPOL_INTERLEAVE. A files preferred node is selected by
the cpu_mem_spread_node rotor.
v2: passed preferred node via addr
v3: using current->cpuset_mem_spread_rotor instead of random_node
Cc: Christoph Lameter <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Lee Schermerhorn <[email protected]>
Cc: KOSAKI Motohiro <[email protected]>
Acked-by: Rik van Riel <[email protected]>
Signed-off-by: Nathan T Zimmer <[email protected]>
---
include/linux/shmem_fs.h | 1 +
mm/shmem.c | 9 +++++++--
2 files changed, 8 insertions(+), 2 deletions(-)
diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
index bef2cf0..cfe8a34 100644
--- a/include/linux/shmem_fs.h
+++ b/include/linux/shmem_fs.h
@@ -17,6 +17,7 @@ struct shmem_inode_info {
char *symlink; /* unswappable short symlink */
};
struct shared_policy policy; /* NUMA memory alloc policy */
+ unsigned long node_offset; /* bias for interleaved nodes */
struct list_head swaplist; /* chain of maybes on swap */
struct list_head xattr_list; /* list of shmem_xattr */
struct inode vfs_inode;
diff --git a/mm/shmem.c b/mm/shmem.c
index a15a466..93801b3 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -64,6 +64,7 @@ static struct vfsmount *shm_mnt;
#include <linux/highmem.h>
#include <linux/seq_file.h>
#include <linux/magic.h>
+#include <linux/cpuset.h>
#include <asm/uaccess.h>
#include <asm/pgtable.h>
@@ -938,9 +939,12 @@ static struct page *shmem_alloc_page(gfp_t gfp,
pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
/*
- * alloc_page_vma() will drop the shared policy reference
+ * alloc_page_vma() will drop the shared policy reference.
+ *
+ * To avoid allocating all tmpfs pages on node 0, we fake up a virtual
+ * address based on this file's predetermined preferred node.
*/
- return alloc_page_vma(gfp, &pvma, 0);
+ return alloc_page_vma(gfp, &pvma, info->node_offset << PAGE_SHIFT);
}
#else /* !CONFIG_NUMA */
#ifdef CONFIG_TMPFS
@@ -1374,6 +1378,7 @@ static struct inode *shmem_get_inode(struct super_block *sb, const struct inode
inode->i_fop = &shmem_file_operations;
mpol_shared_policy_init(&info->policy,
shmem_get_sbmpol(sbinfo));
+ info->node_offset = cpuset_mem_spread_node();
break;
case S_IFDIR:
inc_nlink(inode);
(6/22/12 10:35 AM), Nathan Zimmer wrote:
> When tmpfs has the memory policy interleaved it always starts allocating at each
> file at node 0. When there are many small files the lower nodes fill up
> disproportionately.
> This patch attempts to spread out node usage by starting files at nodes other
> then 0. I disturbed the addr parameter since alloc_pages_vma will only use it
> when the policy is MPOL_INTERLEAVE. A files preferred node is selected by
> the cpu_mem_spread_node rotor.
>
> v2: passed preferred node via addr
> v3: using current->cpuset_mem_spread_rotor instead of random_node
>
> Cc: Christoph Lameter <[email protected]>
> Cc: Nick Piggin <[email protected]>
> Cc: Hugh Dickins <[email protected]>
> Cc: Lee Schermerhorn <[email protected]>
> Cc: KOSAKI Motohiro <[email protected]>
> Acked-by: Rik van Riel <[email protected]>
> Signed-off-by: Nathan T Zimmer <[email protected]>
> ---
>
> include/linux/shmem_fs.h | 1 +
> mm/shmem.c | 9 +++++++--
> 2 files changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
> index bef2cf0..cfe8a34 100644
> --- a/include/linux/shmem_fs.h
> +++ b/include/linux/shmem_fs.h
> @@ -17,6 +17,7 @@ struct shmem_inode_info {
> char *symlink; /* unswappable short symlink */
> };
> struct shared_policy policy; /* NUMA memory alloc policy */
> + unsigned long node_offset; /* bias for interleaved nodes */
> struct list_head swaplist; /* chain of maybes on swap */
> struct list_head xattr_list; /* list of shmem_xattr */
> struct inode vfs_inode;
> diff --git a/mm/shmem.c b/mm/shmem.c
> index a15a466..93801b3 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -64,6 +64,7 @@ static struct vfsmount *shm_mnt;
> #include <linux/highmem.h>
> #include <linux/seq_file.h>
> #include <linux/magic.h>
> +#include <linux/cpuset.h>
>
> #include <asm/uaccess.h>
> #include <asm/pgtable.h>
> @@ -938,9 +939,12 @@ static struct page *shmem_alloc_page(gfp_t gfp,
> pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
>
> /*
> - * alloc_page_vma() will drop the shared policy reference
> + * alloc_page_vma() will drop the shared policy reference.
> + *
> + * To avoid allocating all tmpfs pages on node 0, we fake up a virtual
> + * address based on this file's predetermined preferred node.
> */
> - return alloc_page_vma(gfp, &pvma, 0);
> + return alloc_page_vma(gfp, &pvma, info->node_offset << PAGE_SHIFT);
Still unacceptable. Please stop to pass hacky fake address.
> }
> #else /* !CONFIG_NUMA */
> #ifdef CONFIG_TMPFS
> @@ -1374,6 +1378,7 @@ static struct inode *shmem_get_inode(struct super_block *sb, const struct inode
> inode->i_fop = &shmem_file_operations;
> mpol_shared_policy_init(&info->policy,
> shmem_get_sbmpol(sbinfo));
> + info->node_offset = cpuset_mem_spread_node();
cpuset_mem_spread_node is designed for using allocation time.
On Fri, Jun 22, 2012 at 08:35:14PM -0400, KOSAKI Motohiro wrote:
> (6/22/12 10:35 AM), Nathan Zimmer wrote:
> > When tmpfs has the memory policy interleaved it always starts allocating at each
> > file at node 0. When there are many small files the lower nodes fill up
> > disproportionately.
> > This patch attempts to spread out node usage by starting files at nodes other
> > then 0. I disturbed the addr parameter since alloc_pages_vma will only use it
> > when the policy is MPOL_INTERLEAVE. A files preferred node is selected by
> > the cpu_mem_spread_node rotor.
> >
> > v2: passed preferred node via addr
> > v3: using current->cpuset_mem_spread_rotor instead of random_node
> >
> > Cc: Christoph Lameter <[email protected]>
> > Cc: Nick Piggin <[email protected]>
> > Cc: Hugh Dickins <[email protected]>
> > Cc: Lee Schermerhorn <[email protected]>
> > Cc: KOSAKI Motohiro <[email protected]>
> > Acked-by: Rik van Riel <[email protected]>
> > Signed-off-by: Nathan T Zimmer <[email protected]>
> > ---
> >
> > include/linux/shmem_fs.h | 1 +
> > mm/shmem.c | 9 +++++++--
> > 2 files changed, 8 insertions(+), 2 deletions(-)
> >
> > diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
> > index bef2cf0..cfe8a34 100644
> > --- a/include/linux/shmem_fs.h
> > +++ b/include/linux/shmem_fs.h
> > @@ -17,6 +17,7 @@ struct shmem_inode_info {
> > char *symlink; /* unswappable short symlink */
> > };
> > struct shared_policy policy; /* NUMA memory alloc policy */
> > + unsigned long node_offset; /* bias for interleaved nodes */
> > struct list_head swaplist; /* chain of maybes on swap */
> > struct list_head xattr_list; /* list of shmem_xattr */
> > struct inode vfs_inode;
> > diff --git a/mm/shmem.c b/mm/shmem.c
> > index a15a466..93801b3 100644
> > --- a/mm/shmem.c
> > +++ b/mm/shmem.c
> > @@ -64,6 +64,7 @@ static struct vfsmount *shm_mnt;
> > #include <linux/highmem.h>
> > #include <linux/seq_file.h>
> > #include <linux/magic.h>
> > +#include <linux/cpuset.h>
> >
> > #include <asm/uaccess.h>
> > #include <asm/pgtable.h>
> > @@ -938,9 +939,12 @@ static struct page *shmem_alloc_page(gfp_t gfp,
> > pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
> >
> > /*
> > - * alloc_page_vma() will drop the shared policy reference
> > + * alloc_page_vma() will drop the shared policy reference.
> > + *
> > + * To avoid allocating all tmpfs pages on node 0, we fake up a virtual
> > + * address based on this file's predetermined preferred node.
> > */
> > - return alloc_page_vma(gfp, &pvma, 0);
> > + return alloc_page_vma(gfp, &pvma, info->node_offset << PAGE_SHIFT);
>
> Still unacceptable. Please stop to pass hacky fake address.
I am open to suggestion here. The reason it was placed there was because the
effect would only occur when the policy was interleaved and it would be clear
the call to mpol_shared_policy_lookup should still used the unaltered index.
>
> > }
> > #else /* !CONFIG_NUMA */
> > #ifdef CONFIG_TMPFS
> > @@ -1374,6 +1378,7 @@ static struct inode *shmem_get_inode(struct super_block *sb, const struct inode
> > inode->i_fop = &shmem_file_operations;
> > mpol_shared_policy_init(&info->policy,
> > shmem_get_sbmpol(sbinfo));
> > + info->node_offset = cpuset_mem_spread_node();
>
> cpuset_mem_spread_node is designed for using allocation time.
>
>
>
>