[PATCH v6.13-rc1] arm64: mte: set VM_MTE_ALLOWED for hugetlbfs at correct place

[PATCH v6.13-rc1] arm64: mte: set VM_MTE_ALLOWED for hugetlbfs at correct place

[Yang Shi]

The commit 5de195060b2e ("mm: resolve faulty mmap_region() error path
behaviour") moved vm flags validation before fop->mmap for file
mappings.  But when commit 25c17c4b55de ("hugetlb: arm64: add mte support")
was rebased on top of it, the hugetlbfs part was missed.  Mmapping
hugetlbfs file may not have MAP_HUGETLB set.

Fixes: 25c17c4b55de ("hugetlb: arm64: add mte support")
Signed-off-by: Yang Shi <yang@os.amperecomputing.com>
---
 arch/arm64/include/asm/mman.h | 3 ++-
 fs/hugetlbfs/inode.c          | 2 +-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/include/asm/mman.h b/arch/arm64/include/asm/mman.h
index 1d53022fc7e1..21df8bbd2668 100644
--- a/arch/arm64/include/asm/mman.h
+++ b/arch/arm64/include/asm/mman.h
@@ -7,6 +7,7 @@
 #ifndef BUILD_VDSO
 #include <linux/compiler.h>
 #include <linux/fs.h>
+#include <linux/hugetlb.h>
 #include <linux/shmem_fs.h>
 #include <linux/types.h>
 
@@ -44,7 +45,7 @@ static inline unsigned long arch_calc_vm_flag_bits(struct file *file,
 	if (system_supports_mte()) {
 		if (flags & (MAP_ANONYMOUS | MAP_HUGETLB))
 			return VM_MTE_ALLOWED;
-		if (shmem_file(file))
+		if (shmem_file(file) || is_file_hugepages(file))
 			return VM_MTE_ALLOWED;
 	}
 
diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index 1bbf783b244a..2dea122e5b93 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -113,7 +113,7 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 	 * way when do_mmap unwinds (may be important on powerpc
 	 * and ia64).
 	 */
-	vm_flags_set(vma, VM_HUGETLB | VM_DONTEXPAND | VM_MTE_ALLOWED);
+	vm_flags_set(vma, VM_HUGETLB | VM_DONTEXPAND);
 	vma->vm_ops = &hugetlb_vm_ops;
 
 	ret = seal_check_write(info->seals, vma);
-- 
2.41.0

Re: [PATCH v6.13-rc1] arm64: mte: set VM_MTE_ALLOWED for hugetlbfs at correct place

[Catalin Marinas]

On Tue, Nov 19, 2024 at 12:09:14PM -0800, Yang Shi wrote:
> The commit 5de195060b2e ("mm: resolve faulty mmap_region() error path
> behaviour") moved vm flags validation before fop->mmap for file
> mappings.  But when commit 25c17c4b55de ("hugetlb: arm64: add mte support")
> was rebased on top of it, the hugetlbfs part was missed.  Mmapping
> hugetlbfs file may not have MAP_HUGETLB set.
> 
> Fixes: 25c17c4b55de ("hugetlb: arm64: add mte support")
> Signed-off-by: Yang Shi <yang@os.amperecomputing.com>

Since 5de195060b2e went into -rc7 and my arm64 tree was based on -rc1, I
did not test this combination. The patch looks fine, so:

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>

So I either merge it via the arm64 tree at -rc1 or Linus picks it up and
applies it directly during the merging window. Both options are fine,
it's not essential that MTE + hugetlbfs works at -rc1. Adding Linus and
Andrew in case they have any preference (and leaving the patch in place
below).

Thanks,

Catalin

> ---
>  arch/arm64/include/asm/mman.h | 3 ++-
>  fs/hugetlbfs/inode.c          | 2 +-
>  2 files changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/mman.h b/arch/arm64/include/asm/mman.h
> index 1d53022fc7e1..21df8bbd2668 100644
> --- a/arch/arm64/include/asm/mman.h
> +++ b/arch/arm64/include/asm/mman.h
> @@ -7,6 +7,7 @@
>  #ifndef BUILD_VDSO
>  #include <linux/compiler.h>
>  #include <linux/fs.h>
> +#include <linux/hugetlb.h>
>  #include <linux/shmem_fs.h>
>  #include <linux/types.h>
>  
> @@ -44,7 +45,7 @@ static inline unsigned long arch_calc_vm_flag_bits(struct file *file,
>  	if (system_supports_mte()) {
>  		if (flags & (MAP_ANONYMOUS | MAP_HUGETLB))
>  			return VM_MTE_ALLOWED;
> -		if (shmem_file(file))
> +		if (shmem_file(file) || is_file_hugepages(file))
>  			return VM_MTE_ALLOWED;
>  	}
>  
> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
> index 1bbf783b244a..2dea122e5b93 100644
> --- a/fs/hugetlbfs/inode.c
> +++ b/fs/hugetlbfs/inode.c
> @@ -113,7 +113,7 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
>  	 * way when do_mmap unwinds (may be important on powerpc
>  	 * and ia64).
>  	 */
> -	vm_flags_set(vma, VM_HUGETLB | VM_DONTEXPAND | VM_MTE_ALLOWED);
> +	vm_flags_set(vma, VM_HUGETLB | VM_DONTEXPAND);
>  	vma->vm_ops = &hugetlb_vm_ops;
>  
>  	ret = seal_check_write(info->seals, vma);
> -- 

Re: [PATCH v6.13-rc1] arm64: mte: set VM_MTE_ALLOWED for hugetlbfs at correct place

[Yang Shi]

On 11/21/24 10:22 AM, Catalin Marinas wrote:
> On Tue, Nov 19, 2024 at 12:09:14PM -0800, Yang Shi wrote:
>> The commit 5de195060b2e ("mm: resolve faulty mmap_region() error path
>> behaviour") moved vm flags validation before fop->mmap for file
>> mappings.  But when commit 25c17c4b55de ("hugetlb: arm64: add mte support")
>> was rebased on top of it, the hugetlbfs part was missed.  Mmapping
>> hugetlbfs file may not have MAP_HUGETLB set.
>>
>> Fixes: 25c17c4b55de ("hugetlb: arm64: add mte support")
>> Signed-off-by: Yang Shi <yang@os.amperecomputing.com>
> Since 5de195060b2e went into -rc7 and my arm64 tree was based on -rc1, I
> did not test this combination. The patch looks fine, so:
>
> Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
>
> So I either merge it via the arm64 tree at -rc1 or Linus picks it up and
> applies it directly during the merging window. Both options are fine,
> it's not essential that MTE + hugetlbfs works at -rc1. Adding Linus and
> Andrew in case they have any preference (and leaving the patch in place
> below).

Thanks, Catalin. Either way is fine to me too.

>
> Thanks,
>
> Catalin
>
>> ---
>>   arch/arm64/include/asm/mman.h | 3 ++-
>>   fs/hugetlbfs/inode.c          | 2 +-
>>   2 files changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/arm64/include/asm/mman.h b/arch/arm64/include/asm/mman.h
>> index 1d53022fc7e1..21df8bbd2668 100644
>> --- a/arch/arm64/include/asm/mman.h
>> +++ b/arch/arm64/include/asm/mman.h
>> @@ -7,6 +7,7 @@
>>   #ifndef BUILD_VDSO
>>   #include <linux/compiler.h>
>>   #include <linux/fs.h>
>> +#include <linux/hugetlb.h>
>>   #include <linux/shmem_fs.h>
>>   #include <linux/types.h>
>>   
>> @@ -44,7 +45,7 @@ static inline unsigned long arch_calc_vm_flag_bits(struct file *file,
>>   	if (system_supports_mte()) {
>>   		if (flags & (MAP_ANONYMOUS | MAP_HUGETLB))
>>   			return VM_MTE_ALLOWED;
>> -		if (shmem_file(file))
>> +		if (shmem_file(file) || is_file_hugepages(file))
>>   			return VM_MTE_ALLOWED;
>>   	}
>>   
>> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
>> index 1bbf783b244a..2dea122e5b93 100644
>> --- a/fs/hugetlbfs/inode.c
>> +++ b/fs/hugetlbfs/inode.c
>> @@ -113,7 +113,7 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
>>   	 * way when do_mmap unwinds (may be important on powerpc
>>   	 * and ia64).
>>   	 */
>> -	vm_flags_set(vma, VM_HUGETLB | VM_DONTEXPAND | VM_MTE_ALLOWED);
>> +	vm_flags_set(vma, VM_HUGETLB | VM_DONTEXPAND);
>>   	vma->vm_ops = &hugetlb_vm_ops;
>>   
>>   	ret = seal_check_write(info->seals, vma);
>> -- 

Re: [PATCH v6.13-rc1] arm64: mte: set VM_MTE_ALLOWED for hugetlbfs at correct place

[Catalin Marinas]

On Tue, 19 Nov 2024 12:09:14 -0800, Yang Shi wrote:
> The commit 5de195060b2e ("mm: resolve faulty mmap_region() error path
> behaviour") moved vm flags validation before fop->mmap for file
> mappings.  But when commit 25c17c4b55de ("hugetlb: arm64: add mte support")
> was rebased on top of it, the hugetlbfs part was missed.  Mmapping
> hugetlbfs file may not have MAP_HUGETLB set.
> 
> 
> [...]

Applied to arm64 (for-next/fixes), thanks!

[1/1] arm64: mte: set VM_MTE_ALLOWED for hugetlbfs at correct place
      https://git.kernel.org/arm64/c/49ccf2c3cafb

-- 
Catalin