There is currently a deadlock in the DAX PMD fault path because it ends up taking the same mapping->i_mmap_rwsem twice. The first take is a i_mmap_lock_read() near the beginning of __dax_pmd_fault(), and the second is __dax_pmd_fault() -> unmap_mapping_range() -> i_mmap_lock_write().
We are re-working the way that DAX does its locking for v4.4, so for now just disable DAX PMD faults and fall back to PAGE_SIZE faults to make sure we don't hit this deadlock. Signed-off-by: Ross Zwisler <[email protected]> --- fs/dax.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/fs/dax.c b/fs/dax.c index f364c90..d42217d 100644 --- a/fs/dax.c +++ b/fs/dax.c @@ -534,6 +534,13 @@ int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address, unsigned long pfn; int result = 0; + /* + * There is currently a deadlock in the DAX PMD fault path. Disable + * this path and fall back to the normal dax_fault() path for now + * while we rework locking. + */ + return VM_FAULT_FALLBACK; + /* Fall back to PTEs if we're going to COW */ if (write && !(vma->vm_flags & VM_SHARED)) return VM_FAULT_FALLBACK; -- 2.1.0 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
