On Thu, Jun 03, 2010 at 09:02:09AM +1000, Dave Chinner wrote: > Did you skip it unconditionally, or only when a transaction was > required? xfs_vm_releasepage is mostly a no-op if no transaction is required. If we have neither delalloc nor unwritten buffer we do not actually enter xfs_page_state_convert, and ->releasepage also doesn't touch unampped buffers at all. > The scary part is that I've seen stack traces (i.e. most stack used) > through this reclaim path for delalloc conversion even for > allocations that are GFP_NOFS and the only thing saving us from > deadlocks is th PF_FSTRANS check. Even worse is that > shrinker_page_list() will call try_to_release_pages() without > checking whether it's allowed to enter the filesystem or not, so we > can be doing block allocation in places we've specifically told the > memory allocation subsystem not to.... s/shrinker_page_list/shrink_page_list/ and s/try_to_release_pages/try_to_release_page/ above. shrink_page_list takes the gfp_mask for try_to_release_page from the scan_control structure passed to it from all the top of the long callchain. I can't find anobvious bug, but this could cause a lot more harm. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>