Re: [PATCH] remove hugetlb_instantiation_mutex

[Adam Litke <agl@us.ibm.com>]

Hey... I am amazed at how quickly you came back with a patch for this :)
Thanks for looking at it.  Unfortunately there is one show-stopper and I
have some reservations (pun definitely intended) with your approach:

First, your patch does not pass the libhugetlbfs test
'alloc-instantiate-race' which was written to tickle the the race which
the mutex was introduced to solve.  Your patch works for shared
mappings, but not for the private case.

Second, the introduction of another pair of global counters triggers my
internal warning system... These global counters are known to cause
problems with NUMA and cpusets.  Have you considered these interactions?
Additionally, the commit/rollback logic you are using closely parallels
what we already do with the huge page reservation mechanism.  Is there
any way you could integrate your stuff into the reservation system to
avoid all the duplicated logic?

-- 
Adam Litke - (agl at us.ibm.com)
IBM Linux Technology Center