Re: [PATCH v5 00/12] Lockless update of reference count protected by spinlock

[Waiman Long <waiman.long@hp.com>]

On 07/05/2013 01:54 PM, Al Viro wrote:
> On Fri, Jul 05, 2013 at 11:29:10AM -0400, Waiman Long wrote:
>>> I've just pushed such commit into vfs.git#for-linus; please, do the rest
>>> on top of it.  And keep it bisectable, i.e. so that at any intermediate
>>> point the tree would build and work.
>> I am sorry. I didn't change anything in the dentry structure in
>> patch 3. So putting patches 4-11 on top of it won't break the build.
> *gyah*...  I'd missed the !@#!# macro you've added there.  Could you
> explain the reasons for using it at all?  Not to mention anything
> else, you've missed
> #  define d_refcount(d)                 ((d)->d_count)
> in Lustre.  What's the point of your macro (d_refcount ->  d_count), anyway?
> All references outside of fs/namei.c, fs/dcache.c, include/linux/dcache.h
> should be via d_count(dentry) anyway...

I am sorry that I am still using the latest 3.10 bits that I pull in 
last week as the basis for my patchset. I want to get my changes 
stabilized before looking at the latest bits. So I did missed the latest 
3.11 bits that are merged this week including the Lustre change that you 
mentioned. I will pull in the latest bit and rebase my patch on top of 
it. I guess I also need to make changes to that d_refcount() macro as 
well as codes that reference it. Are you aware of other upcoming patches 
that may conflict with my patch?

The d_refcount macro in patch 3 is to make the name change first so that 
I don't need to change them in the last patch. This is to make the last 
patch easier to review by moving those irrelevant name changes away from it.

Regards,
Longman