[PATCH] Correct alignment of huge page requests.

[PATCH] Correct alignment of huge page requests.

[Steven Truelove]

When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient.  Modified hugetlb_file_setup() to align requests to the huge page size.  Also modified mmap_pgoff() to avoid duplicating this check and to align against the start address.

Signed-off-by: Steven Truelove <steven.truelove@utoronto.ca>
---
 fs/hugetlbfs/inode.c |    9 ++++++---
 mm/mmap.c            |    6 +++++-
 2 files changed, 11 insertions(+), 4 deletions(-)

diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index 1e85a7a..b4bed46 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
 	struct path path;
 	struct dentry *root;
 	struct qstr quick_string;
+	struct hstate *hstate;
+	int num_pages;
 
 	*user = NULL;
 	if (!hugetlbfs_vfsmount)
@@ -967,10 +969,11 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
 	if (!inode)
 		goto out_dentry;
 
+	hstate = hstate_inode(inode);
+	num_pages = ALIGN(size, huge_page_size(hstate)) >>
+			huge_page_shift(hstate);
 	error = -ENOMEM;
-	if (hugetlb_reserve_pages(inode, 0,
-			size >> huge_page_shift(hstate_inode(inode)), NULL,
-			acctflag))
+	if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
 		goto out_inode;
 
 	d_instantiate(path.dentry, inode);
diff --git a/mm/mmap.c b/mm/mmap.c
index 3f758c7..1f44ccf 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -1098,8 +1098,12 @@ SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
 		 * taken when vm_ops->mmap() is called
 		 * A dummy user value is used because we are not locking
 		 * memory so no accounting is necessary
+		 * Length is increased by the amount necessary to align
+		 * the base address to the huge page size.
+		 * hugetlb_file_setup() aligns the end of the buffer to
+		 * the huge page size.
 		 */
-		len = ALIGN(len, huge_page_size(&default_hstate));
+		len += ALIGN(addr, huge_page_size(&default_hstate)) - addr;
 		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
 						&user, HUGETLB_ANONHUGE_INODE);
 		if (IS_ERR(file))
-- 
1.7.3.4

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] Correct alignment of huge page requests.

[Andrew Morton]

On Mon, 27 Feb 2012 23:00:28 -0500
Steven Truelove <steven.truelove@utoronto.ca> wrote:

> When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient.  Modified hugetlb_file_setup() to align requests to the huge page size.  Also modified mmap_pgoff() to avoid duplicating this check and to align against the start address.
> 

I don't think this is quite right.

Suppose huge_page_size is 4096, addr=4095, len=4098.  So we're mapping
three pages: the last byte of the first page, all of the second page
and the first byte of the third page. 

> --- a/fs/hugetlbfs/inode.c
> +++ b/fs/hugetlbfs/inode.c
> @@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
>  	struct path path;
>  	struct dentry *root;
>  	struct qstr quick_string;
> +	struct hstate *hstate;
> +	int num_pages;
>  
>  	*user = NULL;
>  	if (!hugetlbfs_vfsmount)
> @@ -967,10 +969,11 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
>  	if (!inode)
>  		goto out_dentry;
>  
> +	hstate = hstate_inode(inode);
> +	num_pages = ALIGN(size, huge_page_size(hstate)) >>
> +			huge_page_shift(hstate);
>  	error = -ENOMEM;
> -	if (hugetlb_reserve_pages(inode, 0,
> -			size >> huge_page_shift(hstate_inode(inode)), NULL,
> -			acctflag))
> +	if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
>  		goto out_inode;
>  
>  	d_instantiate(path.dentry, inode);
> diff --git a/mm/mmap.c b/mm/mmap.c
> index 3f758c7..1f44ccf 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -1098,8 +1098,12 @@ SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
>  		 * taken when vm_ops->mmap() is called
>  		 * A dummy user value is used because we are not locking
>  		 * memory so no accounting is necessary
> +		 * Length is increased by the amount necessary to align
> +		 * the base address to the huge page size.
> +		 * hugetlb_file_setup() aligns the end of the buffer to
> +		 * the huge page size.
>  		 */
> -		len = ALIGN(len, huge_page_size(&default_hstate));
> +		len += ALIGN(addr, huge_page_size(&default_hstate)) - addr;
>  		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
>  						&user, HUGETLB_ANONHUGE_INODE);

mmap_pgoff() will change `len' from 4098 to 4099.  hugetlb_file_setup()
will round that up to 8192 and will decide to reserve two pages, not
three.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH] Correct alignment of huge page requests.

[Steven Truelove]

When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient.  Modified hugetlb_file_setup() to align requests to the huge page size, and to accept an address argument so that all alignment checks can be performed in hugetlb_file_setup(), rather than in its callers.  Changed newseg and mmap_pgoff to match new prototype and eliminated a now redundant alignment check.
---
 fs/hugetlbfs/inode.c    |   12 ++++++++----
 include/linux/hugetlb.h |    3 ++-
 ipc/shm.c               |    2 +-
 mm/mmap.c               |    6 +++---
 4 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index 1e85a7a..a97b7cc 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -928,7 +928,7 @@ static int can_do_hugetlb_shm(void)
 	return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 }
 
-struct file *hugetlb_file_setup(const char *name, size_t size,
+struct file *hugetlb_file_setup(const char *name, unsigned long addr, size_t size,
 				vm_flags_t acctflag,
 				struct user_struct **user, int creat_flags)
 {
@@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
 	struct path path;
 	struct dentry *root;
 	struct qstr quick_string;
+	struct hstate *hstate;
+	int num_pages;
 
 	*user = NULL;
 	if (!hugetlbfs_vfsmount)
@@ -967,10 +969,12 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
 	if (!inode)
 		goto out_dentry;
 
+	hstate = hstate_inode(inode);
+	size += addr & ~huge_page_mask(hstate);
+	num_pages = ALIGN(size, huge_page_size(hstate)) >>
+			huge_page_shift(hstate);
 	error = -ENOMEM;
-	if (hugetlb_reserve_pages(inode, 0,
-			size >> huge_page_shift(hstate_inode(inode)), NULL,
-			acctflag))
+	if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
 		goto out_inode;
 
 	d_instantiate(path.dentry, inode);
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index d9d6c86..4b9e59d 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -164,7 +164,8 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 
 extern const struct file_operations hugetlbfs_file_operations;
 extern const struct vm_operations_struct hugetlb_vm_ops;
-struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
+struct file *hugetlb_file_setup(const char *name, unsigned long addr,
+				size_t size, vm_flags_t acct,
 				struct user_struct **user, int creat_flags);
 int hugetlb_get_quota(struct address_space *mapping, long delta);
 void hugetlb_put_quota(struct address_space *mapping, long delta);
diff --git a/ipc/shm.c b/ipc/shm.c
index b76be5b..406c5b2 100644
--- a/ipc/shm.c
+++ b/ipc/shm.c
@@ -482,7 +482,7 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 		/* hugetlb_file_setup applies strict accounting */
 		if (shmflg & SHM_NORESERVE)
 			acctflag = VM_NORESERVE;
-		file = hugetlb_file_setup(name, size, acctflag,
+		file = hugetlb_file_setup(name, 0, size, acctflag,
 					&shp->mlock_user, HUGETLB_SHMFS_INODE);
 	} else {
 		/*
diff --git a/mm/mmap.c b/mm/mmap.c
index 3f758c7..4bf211a 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -1099,9 +1099,9 @@ SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
 		 * A dummy user value is used because we are not locking
 		 * memory so no accounting is necessary
 		 */
-		len = ALIGN(len, huge_page_size(&default_hstate));
-		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
-						&user, HUGETLB_ANONHUGE_INODE);
+		file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
+						VM_NORESERVE, &user,
+						HUGETLB_ANONHUGE_INODE);
 		if (IS_ERR(file))
 			return PTR_ERR(file);
 	}
-- 
1.7.3.4

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] Correct alignment of huge page requests.

[Steven Truelove]

Sorry, I just realized I forgot to sign off the patch.  Will resend now.


On 01/03/2012 9:41 PM, Steven Truelove wrote:
> When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient.  Modified hugetlb_file_setup() to align requests to the huge page size, and to accept an address argument so that all alignment checks can be performed in hugetlb_file_setup(), rather than in its callers.  Changed newseg and mmap_pgoff to match new prototype and eliminated a now redundant alignment check.
> ---
>   fs/hugetlbfs/inode.c    |   12 ++++++++----
>   include/linux/hugetlb.h |    3 ++-
>   ipc/shm.c               |    2 +-
>   mm/mmap.c               |    6 +++---
>   4 files changed, 14 insertions(+), 9 deletions(-)
>
> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
> index 1e85a7a..a97b7cc 100644
> --- a/fs/hugetlbfs/inode.c
> +++ b/fs/hugetlbfs/inode.c
> @@ -928,7 +928,7 @@ static int can_do_hugetlb_shm(void)
>   	return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
>   }
>
> -struct file *hugetlb_file_setup(const char *name, size_t size,
> +struct file *hugetlb_file_setup(const char *name, unsigned long addr, size_t size,
>   				vm_flags_t acctflag,
>   				struct user_struct **user, int creat_flags)
>   {
> @@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
>   	struct path path;
>   	struct dentry *root;
>   	struct qstr quick_string;
> +	struct hstate *hstate;
> +	int num_pages;
>
>   	*user = NULL;
>   	if (!hugetlbfs_vfsmount)
> @@ -967,10 +969,12 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
>   	if (!inode)
>   		goto out_dentry;
>
> +	hstate = hstate_inode(inode);
> +	size += addr&  ~huge_page_mask(hstate);
> +	num_pages = ALIGN(size, huge_page_size(hstate))>>
> +			huge_page_shift(hstate);
>   	error = -ENOMEM;
> -	if (hugetlb_reserve_pages(inode, 0,
> -			size>>  huge_page_shift(hstate_inode(inode)), NULL,
> -			acctflag))
> +	if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
>   		goto out_inode;
>
>   	d_instantiate(path.dentry, inode);
> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> index d9d6c86..4b9e59d 100644
> --- a/include/linux/hugetlb.h
> +++ b/include/linux/hugetlb.h
> @@ -164,7 +164,8 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
>
>   extern const struct file_operations hugetlbfs_file_operations;
>   extern const struct vm_operations_struct hugetlb_vm_ops;
> -struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
> +struct file *hugetlb_file_setup(const char *name, unsigned long addr,
> +				size_t size, vm_flags_t acct,
>   				struct user_struct **user, int creat_flags);
>   int hugetlb_get_quota(struct address_space *mapping, long delta);
>   void hugetlb_put_quota(struct address_space *mapping, long delta);
> diff --git a/ipc/shm.c b/ipc/shm.c
> index b76be5b..406c5b2 100644
> --- a/ipc/shm.c
> +++ b/ipc/shm.c
> @@ -482,7 +482,7 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
>   		/* hugetlb_file_setup applies strict accounting */
>   		if (shmflg&  SHM_NORESERVE)
>   			acctflag = VM_NORESERVE;
> -		file = hugetlb_file_setup(name, size, acctflag,
> +		file = hugetlb_file_setup(name, 0, size, acctflag,
>   					&shp->mlock_user, HUGETLB_SHMFS_INODE);
>   	} else {
>   		/*
> diff --git a/mm/mmap.c b/mm/mmap.c
> index 3f758c7..4bf211a 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -1099,9 +1099,9 @@ SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
>   		 * A dummy user value is used because we are not locking
>   		 * memory so no accounting is necessary
>   		 */
> -		len = ALIGN(len, huge_page_size(&default_hstate));
> -		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
> -						&user, HUGETLB_ANONHUGE_INODE);
> +		file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
> +						VM_NORESERVE,&user,
> +						HUGETLB_ANONHUGE_INODE);
>   		if (IS_ERR(file))
>   			return PTR_ERR(file);
>   	}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH] Correct alignment of huge page requests.

[Steven Truelove]

When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient.  Modified hugetlb_file_setup() to align requests to the huge page size, and to accept an address argument so that all alignment checks can be performed in hugetlb_file_setup(), rather than in its callers.  Changed newseg and mmap_pgoff to match new prototype and eliminated a now redundant alignment check.

Signed-off-by: Steven Truelove <steven.truelove@utoronto.ca>
---
 fs/hugetlbfs/inode.c    |   12 ++++++++----
 include/linux/hugetlb.h |    3 ++-
 ipc/shm.c               |    2 +-
 mm/mmap.c               |    6 +++---
 4 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index 1e85a7a..a97b7cc 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -928,7 +928,7 @@ static int can_do_hugetlb_shm(void)
 	return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 }
 
-struct file *hugetlb_file_setup(const char *name, size_t size,
+struct file *hugetlb_file_setup(const char *name, unsigned long addr, size_t size,
 				vm_flags_t acctflag,
 				struct user_struct **user, int creat_flags)
 {
@@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
 	struct path path;
 	struct dentry *root;
 	struct qstr quick_string;
+	struct hstate *hstate;
+	int num_pages;
 
 	*user = NULL;
 	if (!hugetlbfs_vfsmount)
@@ -967,10 +969,12 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
 	if (!inode)
 		goto out_dentry;
 
+	hstate = hstate_inode(inode);
+	size += addr & ~huge_page_mask(hstate);
+	num_pages = ALIGN(size, huge_page_size(hstate)) >>
+			huge_page_shift(hstate);
 	error = -ENOMEM;
-	if (hugetlb_reserve_pages(inode, 0,
-			size >> huge_page_shift(hstate_inode(inode)), NULL,
-			acctflag))
+	if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
 		goto out_inode;
 
 	d_instantiate(path.dentry, inode);
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index d9d6c86..4b9e59d 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -164,7 +164,8 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 
 extern const struct file_operations hugetlbfs_file_operations;
 extern const struct vm_operations_struct hugetlb_vm_ops;
-struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
+struct file *hugetlb_file_setup(const char *name, unsigned long addr,
+				size_t size, vm_flags_t acct,
 				struct user_struct **user, int creat_flags);
 int hugetlb_get_quota(struct address_space *mapping, long delta);
 void hugetlb_put_quota(struct address_space *mapping, long delta);
diff --git a/ipc/shm.c b/ipc/shm.c
index b76be5b..406c5b2 100644
--- a/ipc/shm.c
+++ b/ipc/shm.c
@@ -482,7 +482,7 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 		/* hugetlb_file_setup applies strict accounting */
 		if (shmflg & SHM_NORESERVE)
 			acctflag = VM_NORESERVE;
-		file = hugetlb_file_setup(name, size, acctflag,
+		file = hugetlb_file_setup(name, 0, size, acctflag,
 					&shp->mlock_user, HUGETLB_SHMFS_INODE);
 	} else {
 		/*
diff --git a/mm/mmap.c b/mm/mmap.c
index 3f758c7..4bf211a 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -1099,9 +1099,9 @@ SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
 		 * A dummy user value is used because we are not locking
 		 * memory so no accounting is necessary
 		 */
-		len = ALIGN(len, huge_page_size(&default_hstate));
-		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
-						&user, HUGETLB_ANONHUGE_INODE);
+		file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
+						VM_NORESERVE, &user,
+						HUGETLB_ANONHUGE_INODE);
 		if (IS_ERR(file))
 			return PTR_ERR(file);
 	}
-- 
1.7.3.4

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] Correct alignment of huge page requests.

[Naoya Horiguchi]

On Thu, Mar 01, 2012 at 09:58:41PM -0500, Steven Truelove wrote:
> When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient.  Modified hugetlb_file_setup() to align requests to the huge page size, and to accept an address argument so that all alignment checks can be performed in hugetlb_file_setup(), rather than in its callers.  Changed newseg and mmap_pgoff to match new prototype and eliminated a now redundant alignment check.

I think only rounding up request size in shmget() is not sufficient,
because later shmat() also have alignment check and fails to mmap()
to unaligned address.
Maybe file->f_op->get_unmapped_area() (or hugetlb_get_unmapped_area())
should have round up code, I think.
Could you try it?

And a few comments below,

> Signed-off-by: Steven Truelove <steven.truelove@utoronto.ca>
> ---
>  fs/hugetlbfs/inode.c    |   12 ++++++++----
>  include/linux/hugetlb.h |    3 ++-
>  ipc/shm.c               |    2 +-
>  mm/mmap.c               |    6 +++---
>  4 files changed, 14 insertions(+), 9 deletions(-)
> 
> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
> index 1e85a7a..a97b7cc 100644
> --- a/fs/hugetlbfs/inode.c
> +++ b/fs/hugetlbfs/inode.c
> @@ -928,7 +928,7 @@ static int can_do_hugetlb_shm(void)
>  	return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
>  }
>  
> -struct file *hugetlb_file_setup(const char *name, size_t size,
> +struct file *hugetlb_file_setup(const char *name, unsigned long addr, size_t size,

Just a nitpick, this line is over 80 characters.
checkpatch.pl should warn.

>  				vm_flags_t acctflag,
>  				struct user_struct **user, int creat_flags)
>  {
> @@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
>  	struct path path;
>  	struct dentry *root;
>  	struct qstr quick_string;
> +	struct hstate *hstate;
> +	int num_pages;

Is unsigned long better?

Thanks,
Naoya

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] Correct alignment of huge page requests.

[Hillf Danton]

On Fri, Mar 2, 2012 at 10:58 AM, Steven Truelove
<steven.truelove@utoronto.ca> wrote:
> When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient.  Modified hugetlb_file_setup() to align requests to the huge page size, and to accept an address argument so that all alignment checks can be performed in hugetlb_file_setup(), rather than in its callers.  Changed newseg and mmap_pgoff to match new prototype and eliminated a now redundant alignment check.
>
> Signed-off-by: Steven Truelove <steven.truelove@utoronto.ca>

Acked-by: Hillf Danton <dhillf@gmail.com>

> ---
>  fs/hugetlbfs/inode.c    |   12 ++++++++----
>  include/linux/hugetlb.h |    3 ++-
>  ipc/shm.c               |    2 +-
>  mm/mmap.c               |    6 +++---
>  4 files changed, 14 insertions(+), 9 deletions(-)
>
> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
> index 1e85a7a..a97b7cc 100644
> --- a/fs/hugetlbfs/inode.c
> +++ b/fs/hugetlbfs/inode.c
> @@ -928,7 +928,7 @@ static int can_do_hugetlb_shm(void)
>        return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
>  }
>
> -struct file *hugetlb_file_setup(const char *name, size_t size,
> +struct file *hugetlb_file_setup(const char *name, unsigned long addr, size_t size,
>                                vm_flags_t acctflag,
>                                struct user_struct **user, int creat_flags)
>  {
> @@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
>        struct path path;
>        struct dentry *root;
>        struct qstr quick_string;
> +       struct hstate *hstate;
> +       int num_pages;
>
>        *user = NULL;
>        if (!hugetlbfs_vfsmount)
> @@ -967,10 +969,12 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
>        if (!inode)
>                goto out_dentry;
>
> +       hstate = hstate_inode(inode);
> +       size += addr & ~huge_page_mask(hstate);
> +       num_pages = ALIGN(size, huge_page_size(hstate)) >>
> +                       huge_page_shift(hstate);
>        error = -ENOMEM;
> -       if (hugetlb_reserve_pages(inode, 0,
> -                       size >> huge_page_shift(hstate_inode(inode)), NULL,
> -                       acctflag))
> +       if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
>                goto out_inode;
>
>        d_instantiate(path.dentry, inode);
> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> index d9d6c86..4b9e59d 100644
> --- a/include/linux/hugetlb.h
> +++ b/include/linux/hugetlb.h
> @@ -164,7 +164,8 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
>
>  extern const struct file_operations hugetlbfs_file_operations;
>  extern const struct vm_operations_struct hugetlb_vm_ops;
> -struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
> +struct file *hugetlb_file_setup(const char *name, unsigned long addr,
> +                               size_t size, vm_flags_t acct,
>                                struct user_struct **user, int creat_flags);
>  int hugetlb_get_quota(struct address_space *mapping, long delta);
>  void hugetlb_put_quota(struct address_space *mapping, long delta);
> diff --git a/ipc/shm.c b/ipc/shm.c
> index b76be5b..406c5b2 100644
> --- a/ipc/shm.c
> +++ b/ipc/shm.c
> @@ -482,7 +482,7 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
>                /* hugetlb_file_setup applies strict accounting */
>                if (shmflg & SHM_NORESERVE)
>                        acctflag = VM_NORESERVE;
> -               file = hugetlb_file_setup(name, size, acctflag,
> +               file = hugetlb_file_setup(name, 0, size, acctflag,
>                                        &shp->mlock_user, HUGETLB_SHMFS_INODE);
>        } else {
>                /*
> diff --git a/mm/mmap.c b/mm/mmap.c
> index 3f758c7..4bf211a 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -1099,9 +1099,9 @@ SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
>                 * A dummy user value is used because we are not locking
>                 * memory so no accounting is necessary
>                 */
> -               len = ALIGN(len, huge_page_size(&default_hstate));
> -               file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
> -                                               &user, HUGETLB_ANONHUGE_INODE);
> +               file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
> +                                               VM_NORESERVE, &user,
> +                                               HUGETLB_ANONHUGE_INODE);
>                if (IS_ERR(file))
>                        return PTR_ERR(file);
>        }
> --
> 1.7.3.4
>
> --
> 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/
>
>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] Correct alignment of huge page requests.

[Steven Truelove]

On 03/03/2012 10:02 PM, Naoya Horiguchi wrote:
> On Thu, Mar 01, 2012 at 09:58:41PM -0500, Steven Truelove wrote:
>> When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient.  Modified hugetlb_file_setup() to align requests to the huge page size, and to accept an address argument so that all alignment checks can be performed in hugetlb_file_setup(), rather than in its callers.  Changed newseg and mmap_pgoff to match new prototype and eliminated a now redundant alignment check.
> I think only rounding up request size in shmget() is not sufficient,
> because later shmat() also have alignment check and fails to mmap()
> to unaligned address.
> Maybe file->f_op->get_unmapped_area() (or hugetlb_get_unmapped_area())
> should have round up code, I think.
> Could you try it?

Because the allocation is done in shmget() and the the address is not 
provided until shmat(), I don't see a way to make this work reasonably.  
I would argue that only allowing aligned addresses, or allowing the 
kernel to choose the address, is a reasonable restriction on SHM_HUGETLB 
usage.

Regarding your other comments, I will submit a revised patch.

Thanks,

Steven Truelove


> And a few comments below,
>
>> Signed-off-by: Steven Truelove<steven.truelove@utoronto.ca>
>> ---
>>   fs/hugetlbfs/inode.c    |   12 ++++++++----
>>   include/linux/hugetlb.h |    3 ++-
>>   ipc/shm.c               |    2 +-
>>   mm/mmap.c               |    6 +++---
>>   4 files changed, 14 insertions(+), 9 deletions(-)
>>
>> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
>> index 1e85a7a..a97b7cc 100644
>> --- a/fs/hugetlbfs/inode.c
>> +++ b/fs/hugetlbfs/inode.c
>> @@ -928,7 +928,7 @@ static int can_do_hugetlb_shm(void)
>>   	return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
>>   }
>>
>> -struct file *hugetlb_file_setup(const char *name, size_t size,
>> +struct file *hugetlb_file_setup(const char *name, unsigned long addr, size_t size,
> Just a nitpick, this line is over 80 characters.
> checkpatch.pl should warn.
>
>>   				vm_flags_t acctflag,
>>   				struct user_struct **user, int creat_flags)
>>   {
>> @@ -938,6 +938,8 @@ struct file *hugetlb_file_setup(const char *name, size_t size,
>>   	struct path path;
>>   	struct dentry *root;
>>   	struct qstr quick_string;
>> +	struct hstate *hstate;
>> +	int num_pages;
> Is unsigned long better?
>
> Thanks,
> Naoya
>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>