Re: [PATCH 2/2] dax: Giant hack

[Dave Chinner <david@fromorbit.com>]

On Tue, Jan 26, 2016 at 11:08:28PM -0500, Matthew Wilcox wrote:
> From: Matthew Wilcox <willy@linux.intel.com>
> 
> This glop of impossible-to-review code implements a number of ideas that
> need to be separated out.
> 
>  - Eliminate vm_ops->huge_fault.  The core calls ->fault instead and callers
>    who set VM_HUGEPAGE should be prepared to deal with FAULT_FLAG_SIZE_PMD
>    (and larger)
>  - Switch back to calling ->page_mkwrite instead of ->pfn_mkwrite.  DAX now
>    always has a page to lock, and no other imlementations of ->pfn_mkwrite
>    exist.
>  - dax_mkwrite splits out from dax_fault.  dax_fault will now never call
>    get_block() to allocate a block; only to see if a block has been allocated.
>    dax_mkwrite will always attempt to allocate a block.
>  - Filesystems now take their DAX allocation mutex in exclusive/write mode
>    when calling dax_mkwrite.
>  - Split out dax_insert_pmd_mapping() from dax_pmd_fault and share it with
>    the new dax_pmd_mkwrite
>  - Change dax_pmd_write to take a vm_fault argument like the rest of the
>    family of functions.
> 
> Signed-off-by: Matthew Wilcox <willy@linux.intel.com>

I haven't looked really closely at the patch itself, but I like the
direction you're taking here.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com