Re: [PATCH] Initialise mmu_notifier_range correctly

[Jerome Glisse]

On Wed, Jan 02, 2019 at 08:18:33PM -0800, Matthew Wilcox wrote:
> On Wed, Jan 02, 2019 at 07:32:08PM -0800, John Hubbard wrote:
> > Having the range struct declared in separate places from the mmu_notifier_range_init()
> > calls is not great. But I'm not sure I see a way to make it significantly cleaner, given
> > that __follow_pte_pmd uses the range pointer as a way to decide to issue the mmn calls.
> 
> Yeah, I don't think there's anything we can do.  But I started reviewing
> the comments, and they don't make sense together:
> 
>                 /*
>                  * Note because we provide range to follow_pte_pmd it will
>                  * call mmu_notifier_invalidate_range_start() on our behalf
>                  * before taking any lock.
>                  */
>                 if (follow_pte_pmd(vma->vm_mm, address, &range,
>                                    &ptep, &pmdp, &ptl))
>                         continue;
> 
>                 /*
>                  * No need to call mmu_notifier_invalidate_range() as we are
>                  * downgrading page table protection not changing it to point
>                  * to a new page.
>                  *
>                  * See Documentation/vm/mmu_notifier.rst
>                  */
> 
> So if we don't call mmu_notifier_invalidate_range, why are we calling
> mmu_notifier_invalidate_range_start and mmu_notifier_invalidate_range_end?
> ie, why not this ...

Thus comments looks wrong to me ... we need to call
mmu_notifier_invalidate_range() those are use by
IOMMU. I might be to blame for those comments thought.


> 
> diff --git a/fs/dax.c b/fs/dax.c
> index 6959837cc465..905340149924 100644
> --- a/fs/dax.c
> +++ b/fs/dax.c
> @@ -777,7 +777,6 @@ static void dax_entry_mkclean(struct address_space *mapping, pgoff_t index,
>  
>  	i_mmap_lock_read(mapping);
>  	vma_interval_tree_foreach(vma, &mapping->i_mmap, index, index) {
> -		struct mmu_notifier_range range;
>  		unsigned long address;
>  
>  		cond_resched();
> @@ -787,12 +786,7 @@ static void dax_entry_mkclean(struct address_space *mapping, pgoff_t index,
>  
>  		address = pgoff_address(index, vma);
>  
> -		/*
> -		 * Note because we provide start/end to follow_pte_pmd it will
> -		 * call mmu_notifier_invalidate_range_start() on our behalf
> -		 * before taking any lock.
> -		 */
> -		if (follow_pte_pmd(vma->vm_mm, address, &range,
> +		if (follow_pte_pmd(vma->vm_mm, address, NULL,
>  				   &ptep, &pmdp, &ptl))
>  			continue;
>  
> @@ -834,8 +828,6 @@ static void dax_entry_mkclean(struct address_space *mapping, pgoff_t index,
>  unlock_pte:
>  			pte_unmap_unlock(ptep, ptl);
>  		}
> -
> -		mmu_notifier_invalidate_range_end(&range);
>  	}
>  	i_mmap_unlock_read(mapping);
>  }