Re: [PATCH 1/3] idr: make idr_get_next() good for rcu_read_lock()

[Randy Dunlap <rdunlap-/UHa2rfvQTnk1uMJSBkQmQ@public.gmane.org>]

On 01/20/2012 07:45 PM, Hugh Dickins wrote:
> On Fri, 20 Jan 2012, Andrew Morton wrote:
>> On Thu, 19 Jan 2012 12:48:48 -0800 (PST)
>> Hugh Dickins <hughd-hpIqsD4AKlfQT0dZR+AlfA@public.gmane.org> wrote:
>>> Copied comment on RCU locking from idr_find().
>>>
>>> + *
>>> + * This function can be called under rcu_read_lock(), given that the leaf
>>> + * pointers lifetimes are correctly managed.
>>
>> Awkward comment.  It translates to "..., because the leaf pointers
>> lifetimes are correctly managed".
>>
>> Is that what we really meant?  Or did we mean "..., provided the leaf
>> pointers lifetimes are correctly managed"?
> 
> You are right, and part of me realized that even as I copied in the
> comment.  I wanted to express the same optimism for idr_get_next() 
> as was already expressed for idr_find() - whatever it meant ;)
> 
> I thought it was meaning a bit of both: idr.c is managing its end well
> enough that rcu_read_lock() can now be used, but the caller has to
> manage their locking and lifetimes appropriately too.
> 
>>
>> Also, "pointers" should have been "pointer" or "pointer's"!
> 
> You're afraid of Linus turning his "its/it's" wrath from Al to yourself.
> 
> Since "lifetimes" is in the plural, I think it would have to be
> "pointers'" - I _think_ that's correct, rather than "pointers's".

That seems correct to me also.

> But then, it's not the lifetimes of the pointers, but the lifetimes
> of the objects that they point to, that's in question.  So what it
> ought to say is...
> 
> ... falls asleep.

ack.

and thanks for doing all of that radix tree test harness work, Hugh.

-- 
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***