On 04/04/2018 03:03, David Rientjes wrote: > On Tue, 3 Apr 2018, David Rientjes wrote: > >>>>> I found the root cause of this lockdep warning. >>>>> >>>>> In mmap_region(), unmap_region() may be called while vma_link() has not >>>>> been >>>>> called. This happens during the error path if call_mmap() failed. >>>>> >>>>> The only to fix that particular case is to call >>>>> seqcount_init(&vma->vm_sequence) when initializing the vma in >>>>> mmap_region(). >>>>> >>>> >>>> Ack, although that would require a fixup to dup_mmap() as well. >>> >>> You're right, I'll fix that too. >>> >> >> I also think the following is needed: >> >> diff --git a/fs/exec.c b/fs/exec.c >> --- a/fs/exec.c >> +++ b/fs/exec.c >> @@ -312,6 +312,10 @@ static int __bprm_mm_init(struct linux_binprm *bprm) >> vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | >> VM_STACK_INCOMPLETE_SETUP; >> vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); >> INIT_LIST_HEAD(&vma->anon_vma_chain); >> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT >> + seqcount_init(&vma->vm_sequence); >> + atomic_set(&vma->vm_ref_count, 0); >> +#endif >> >> err = insert_vm_struct(mm, vma); >> if (err) >> > > Ugh, I think there are a number of other places where this is needed as > well in mm/mmap.c. I think it would be better to just create a new > alloc_vma(unsigned long flags) that all vma allocators can use and for > CONFIG_SPECULATIVE_PAGE_FAULT will initialize the seqcount_t and atomic_t. >
I don't think this is really needed, most of the time the vma structure is allocated cleared and is then link to rb tree via vma_link. The only case generating a locked warning is when the vma is referenced in the error path before being linked in the mm tree and that is not usual.
