Skip to contentSkip to conversation

Does shmem_getpage==>shmem_alloc_page==>alloc_page_vma hold mmap_sem?

DISCUSSION Paul Jackson 2 participants 3 messages Nov 6, 2005 Memory management

lore
Share & tools
lore
#0 lore raw

Andi,

The comment in mm/mempolicy.c for alloc_page_vma() states:

Should be called with the mm_sem of the vma hold.

However it seems that the call chain (#ifdef CONFIG_NUMA):

shmem_getpage ==> shmem_alloc_page ==> alloc_page_vma

where shmem_getpage() is called from many of the mm/shmem.c file operations, is called without holding mmap_sem. There is no mention of mmap_sem in the entire mm/shmem.c file.

This doesn't seem right.

Signature · 9 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
#1 lore raw
Quoted 5 lines · Paul Jackson

The comment in mm/mempolicy.c for alloc_page_vma() states:

Should be called with the mm_sem of the vma hold.

Quoted 7 lines

However it seems that the call chain (#ifdef CONFIG_NUMA):

shmem_getpage ==> shmem_alloc_page ==> alloc_page_vma

where shmem_getpage() is called from many of the mm/shmem.c file operations, is called without holding mmap_sem. There is no mention of mmap_sem in the entire mm/shmem.c file.

It's safe but horrid. Look closer and you'll find there isn't even an mm to hold the mmap_sem of. The struct vm_area_struct is on the stack of shmem_alloc_page, and exists solely to apply mempolicy to a shmem file via an interface designed for mempolicy on vmas.

So far as I know, it works fine; but that interface really ought to be redesigned some time - it looks like a quick hack that stuck.

Hugh

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#2 lore raw
Quoted 1 line · Hugh

It's safe but horrid.

Ok - thanks for the explanation.

Now I don't feel so bad about some of my cpuset locking hacks.

(Yes, Andrew, I'm still looking at my latest hack, the down_write_trylock() call in cpuset.c refresh_mems.)

Signature · 9 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>