Dynamic huge page pool resizing.

In most real-world scenarios, configuring the size of the hugetlb pool
correctly is a difficult task.  If too few pages are allocated to the pool,
applications using MAP_SHARED may fail to mmap() a hugepage region and
applications using MAP_PRIVATE may receive SIGBUS.  Isolating too much memory
in the hugetlb pool means it is not available for other uses, especially those
programs not using huge pages.

The obvious answer is to let the hugetlb pool grow and shrink in response to
the runtime demand for huge pages.  The work Mel Gorman has been doing to
establish a memory zone for movable memory allocations makes dynamically
resizing the hugetlb pool reliable within the limits of that zone.  This patch
series implements dynamic pool resizing for private and shared mappings while
being careful to maintain existing semantics.  Please reply with your comments
and feedback; even just to say whether it would be a useful feature to you.
Thanks.

How it works
============

Upon depletion of the hugetlb pool, rather than reporting an error immediately,
first try and allocate the needed huge pages directly from the buddy allocator.
Care must be taken to avoid unbounded growth of the hugetlb pool, so the
hugetlb filesystem quota is used to limit overall pool size.

The real work begins when we decide there is a shortage of huge pages.  What
happens next depends on whether the pages are for a private or shared mapping.
Private mappings are straightforward.  At fault time, if alloc_huge_page()
fails, we allocate a page from the buddy allocator and increment the source
node's surplus_huge_pages counter.  When free_huge_page() is called for a page
on a node with a surplus, the page is freed directly to the buddy allocator
instead of the hugetlb pool.

Because shared mappings require all of the pages to be reserved up front, some
additional work must be done at mmap() to support them.  We determine the
reservation shortage and allocate the required number of pages all at once.
These pages are then added to the hugetlb pool and marked reserved.  Where that
is not possible the mmap() will fail.  As with private mappings, the
appropriate surplus counters are updated.  Since reserved huge pages won't
necessarily be used by the process, we can't be sure that free_huge_page() will
always be called to return surplus pages to the buddy allocator.  To prevent
the huge page pool from bloating, we must free unused surplus pages when their
reservation has ended.

Controlling it
==============

With the entire patch series applied, pool resizing is off by default so unless
specific action is taken, the semantics are unchanged.

To take advantage of the flexibility afforded by this patch series one must
tolerate a change in semantics.  To control hugetlb pool growth, the following
techniques can be employed:

 * A sysctl tunable to enable/disable the feature entirely
 * The size= mount option for hugetlbfs filesystems to limit pool size

Performance
===========

When contiguous memory is readily available, it is expected that the cost of
dynamicly resizing the pool will be small.  This series has been performance
tested with 'stream' to measure this cost.

Stream (http://www.cs.virginia.edu/stream/) was linked with libhugetlbfs to
enable remapping of the text and data/bss segments into huge pages.

Stream with small array
-----------------------
Baseline: 	nr_hugepages = 0, No libhugetlbfs segment remapping
Preallocated:	nr_hugepages = 5, Text and data/bss remapping
Dynamic:	nr_hugepages = 0, Text and data/bss remapping

				Rate (MB/s)
Function	Baseline	Preallocated	Dynamic
Copy:		4695.6266	5942.8371	5982.2287
Scale:		4451.5776	5017.1419	5658.7843
Add:		5815.8849	7927.7827	8119.3552
Triad:		5949.4144	8527.6492	8110.6903

Stream with large array
-----------------------
Baseline: 	nr_hugepages =  0, No libhugetlbfs segment remapping
Preallocated:	nr_hugepages = 67, Text and data/bss remapping
Dynamic:	nr_hugepages =  0, Text and data/bss remapping

				Rate (MB/s)
Function	Baseline	Preallocated	Dynamic
Copy:		2227.8281	2544.2732	2546.4947
Scale:		2136.3208	2430.7294	2421.2074
Add:		2773.1449	4004.0021	3999.4331
Triad:		2748.4502	3777.0109	3773.4970

* All numbers are averages taken from 10 consecutive runs with a maximum
  standard deviation of 1.3 percent noted.

This patch:

Simply move update_and_free_page() so that it can be reused later in this
patch series.  The implementation is not changed.
Signed-off-by: NAdam Litke <agl@us.ibm.com>
Acked-by: NAndy Whitcroft <apw@shadowen.org>
Acked-by: NDave McCracken <dave.mccracken@oracle.com>
Acked-by: NWilliam Irwin <bill.irwin@oracle.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Ken Chen <kenchen@google.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Current ia64 kernel flushes icache by lazy_mmu_prot_update() *after*
set_pte().  This is too late.  This patch removes lazy_mmu_prot_update and
add modfied set_pte() for flushing if necessary.

This patch flush icache of a page when
	new pte has exec bit.
	&& new pte has present bit
	&& new pte is user's page.
	&& (old *ptep is not present
            || new pte's pfn is not same to old *ptep's ptn)
	&& new pte's page has no Pg_arch_1 bit.
	   Pg_arch_1 is set when a page is cache consistent.

I think this condition checks are much easier to understand than considering
"Where sync_icache_dcache() should be inserted ?".

pte_user() for ia64 was removed by http://lkml.org/lkml/2007/6/12/67 as
clean-up. So, I added it again.
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The virtual address space argument of clear_user_highpage is supposed to be
the virtual address where the page being cleared will eventually be mapped.
 This allows architectures with virtually indexed caches a few clever
tricks.  That sort of trick falls over in painful ways if the virtual
address argument is wrong.
Signed-off-by: NRalf Baechle <ralf@linux-mips.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This patch proposes fixes to the reference counting of memory policy in the
page allocation paths and in show_numa_map().  Extracted from my "Memory
Policy Cleanups and Enhancements" series as stand-alone.

Shared policy lookup [shmem] has always added a reference to the policy,
but this was never unrefed after page allocation or after formatting the
numa map data.

Default system policy should not require additional ref counting, nor
should the current task's task policy.  However, show_numa_map() calls
get_vma_policy() to examine what may be [likely is] another task's policy.
The latter case needs protection against freeing of the policy.

This patch adds a reference count to a mempolicy returned by
get_vma_policy() when the policy is a vma policy or another task's
mempolicy.  Again, shared policy is already reference counted on lookup.  A
matching "unref" [__mpol_free()] is performed in alloc_page_vma() for
shared and vma policies, and in show_numa_map() for shared and another
task's mempolicy.  We can call __mpol_free() directly, saving an admittedly
inexpensive inline NULL test, because we know we have a non-NULL policy.

Handling policy ref counts for hugepages is a bit trickier.
huge_zonelist() returns a zone list that might come from a shared or vma
'BIND policy.  In this case, we should hold the reference until after the
huge page allocation in dequeue_hugepage().  The patch modifies
huge_zonelist() to return a pointer to the mempolicy if it needs to be
unref'd after allocation.

Kernel Build [16cpu, 32GB, ia64] - average of 10 runs:

		w/o patch	w/ refcount patch
	    Avg	  Std Devn	   Avg	  Std Devn
Real:	 100.59	    0.38	 100.63	    0.43
User:	1209.60	    0.37	1209.91	    0.31
System:   81.52	    0.42	  81.64	    0.34
Signed-off-by: NLee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: NAndi Kleen <ak@suse.de>
Cc: Christoph Lameter <clameter@sgi.com>
Acked-by: NMel Gorman <mel@csn.ul.ie>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It seems a simple mistake was made when converting follow_hugetlb_page()
over to the VM_FAULT flags bitmasks (in "mm: fault feedback #2", commit
83c54070).

By using the wrong bitmask, hugetlb_fault() failures are not being
recognized.  This results in an infinite loop whenever follow_hugetlb_page
is involved in a failed fault.
Signed-off-by: NAdam Litke <agl@us.ibm.com>
Acked-by: NNick Piggin <npiggin@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

dequeue_huge_page() has a serious memory leak upon hugetlb page
allocation.  The for loop continues on allocating hugetlb pages out of
all allowable zone, where this function is supposedly only dequeue one
and only one pages.

Fixed it by breaking out of the for loop once a hugetlb page is found.
Signed-off-by: NKen Chen <kenchen@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Use appropriate accessor function to set compound page destructor
function.

Cc:  William Irwin <wli@holomorphy.com>
Signed-off-by: NAkinobu Mita <akinobu.mita@gmail.com>
Acked-by: NAdam Litke <agl@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The fix to that race in alloc_fresh_huge_page() which could give an illegal
node ID did not need nid_lock at all: the fix was to replace static int nid
by static int prev_nid and do the work on local int nid.  nid_lock did make
sure that racers strictly roundrobin the nodes, but that's not something we
need to enforce strictly.  Kill nid_lock.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb.c: In function `dequeue_huge_page':
mm/hugetlb.c:72: warning: 'nid' might be used uninitialized in this function

Cc: Christoph Lameter <clameter@sgi.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <hermes@gibson.dropbear.id.au>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This patch completes Linus's wish that the fault return codes be made into
bit flags, which I agree makes everything nicer.  This requires requires
all handle_mm_fault callers to be modified (possibly the modifications
should go further and do things like fault accounting in handle_mm_fault --
however that would be for another patch).

[akpm@linux-foundation.org: fix alpha build]
[akpm@linux-foundation.org: fix s390 build]
[akpm@linux-foundation.org: fix sparc build]
[akpm@linux-foundation.org: fix sparc64 build]
[akpm@linux-foundation.org: fix ia64 build]
Signed-off-by: NNick Piggin <npiggin@suse.de>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ian Molton <spyro@f2s.com>
Cc: Bryan Wu <bryan.wu@analog.com>
Cc: Mikael Starvik <starvik@axis.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Roman Zippel <zippel@linux-m68k.org>
Cc: Greg Ungerer <gerg@uclinux.org>
Cc: Matthew Wilcox <willy@debian.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp>
Cc: Richard Curnow <rc@rc0.org.uk>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp>
Cc: Chris Zankel <chris@zankel.net>
Acked-by: NKyle McMartin <kyle@mcmartin.ca>
Acked-by: NHaavard Skinnemoen <hskinnemoen@atmel.com>
Acked-by: NRalf Baechle <ralf@linux-mips.org>
Acked-by: NAndi Kleen <ak@muc.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
[ Still apparently needs some ARM and PPC loving - Linus ]
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Change ->fault prototype.  We now return an int, which contains
VM_FAULT_xxx code in the low byte, and FAULT_RET_xxx code in the next byte.
 FAULT_RET_ code tells the VM whether a page was found, whether it has been
locked, and potentially other things.  This is not quite the way he wanted
it yet, but that's changed in the next patch (which requires changes to
arch code).

This means we no longer set VM_CAN_INVALIDATE in the vma in order to say
that a page is locked which requires filemap_nopage to go away (because we
can no longer remain backward compatible without that flag), but we were
going to do that anyway.

struct fault_data is renamed to struct vm_fault as Linus asked. address
is now a void __user * that we should firmly encourage drivers not to use
without really good reason.

The page is now returned via a page pointer in the vm_fault struct.
Signed-off-by: NNick Piggin <npiggin@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NRobert P. J. Day <rpjday@mindspring.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Huge pages are not movable so are not allocated from ZONE_MOVABLE.  However,
as ZONE_MOVABLE will always have pages that can be migrated or reclaimed, it
can be used to satisfy hugepage allocations even when the system has been
running a long time.  This allows an administrator to resize the hugepage pool
at runtime depending on the size of ZONE_MOVABLE.

This patch adds a new sysctl called hugepages_treat_as_movable.  When a
non-zero value is written to it, future allocations for the huge page pool
will use ZONE_MOVABLE.  Despite huge pages being non-movable, we do not
introduce additional external fragmentation of note as huge pages are always
the largest contiguous block we care about.

[akpm@linux-foundation.org: various fixes]
Signed-off-by: NMel Gorman <mel@csn.ul.ie>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

That static `nid' index needs locking.  Without it we can end up calling
alloc_pages_node() with an illegal node ID and the kernel crashes.
Acked-by: Ngurudas pai <gurudas.pai@oracle.com>
Cc: <stable@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

nid is initialized to numa_node_id() but will either be overwritten in
the loop or not used in the conditional. So remove the initialization.
Signed-off-by: NNishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Some changes done a while ago to avoid pounding on ptep_set_access_flags and
update_mmu_cache in some race situations break sun4c which requires
update_mmu_cache() to always be called on minor faults.

This patch reworks ptep_set_access_flags() semantics, implementations and
callers so that it's now responsible for returning whether an update is
necessary or not (basically whether the PTE actually changed).  This allow
fixing the sparc implementation to always return 1 on sun4c.

[akpm@linux-foundation.org: fixes, cleanups]
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: David Miller <davem@davemloft.net>
Cc: Mark Fortescue <mark@mtfhpc.demon.co.uk>
Acked-by: NWilliam Lee Irwin III <wli@holomorphy.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When cpuset is configured, it breaks the strict hugetlb page reservation as
the accounting is done on a global variable.  Such reservation is
completely rubbish in the presence of cpuset because the reservation is not
checked against page availability for the current cpuset.  Application can
still potentially OOM'ed by kernel with lack of free htlb page in cpuset
that the task is in.  Attempt to enforce strict accounting with cpuset is
almost impossible (or too ugly) because cpuset is too fluid that task or
memory node can be dynamically moved between cpusets.

The change of semantics for shared hugetlb mapping with cpuset is
undesirable.  However, in order to preserve some of the semantics, we fall
back to check against current free page availability as a best attempt and
hopefully to minimize the impact of changing semantics that cpuset has on
hugetlb.
Signed-off-by: NKen Chen <kenchen@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The internal hugetlb resv_huge_pages variable can permanently leak nonzero
value in the error path of hugetlb page fault handler when hugetlb page is
used in combination of cpuset.  The leaked count can permanently trap N
number of hugetlb pages in unusable "reserved" state.

Steps to reproduce the bug:

  (1) create two cpuset, user1 and user2
  (2) reserve 50 htlb pages in cpuset user1
  (3) attempt to shmget/shmat 50 htlb page inside cpuset user2
  (4) kernel oom the user process in step 3
  (5) ipcrm the shm segment

At this point resv_huge_pages will have a count of 49, even though
there are no active hugetlbfs file nor hugetlb shared memory segment
in the system.  The leak is permanent and there is no recovery method
other than system reboot. The leaked count will hold up all future use
of that many htlb pages in all cpusets.

The culprit is that the error path of alloc_huge_page() did not
properly undo the change it made to resv_huge_page, causing
inconsistent state.
Signed-off-by: NKen Chen <kenchen@google.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Adam Litke <agl@us.ibm.com>
Cc: Martin Bligh <mbligh@google.com>
Acked-by: NDavid Gibson <dwg@au1.ibm.com>
Cc: <stable@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

__unmap_hugepage_range() is buggy that it does not preserve dirty state of
huge_pte when unmapping hugepage range.  It causes data corruption in the
event of dop_caches being used by sys admin.  For example, an application
creates a hugetlb file, modify pages, then unmap it.  While leaving the
hugetlb file alive, comes along sys admin doing a "echo 3 >
/proc/sys/vm/drop_caches".

drop_pagecache_sb() will happily free all pages that aren't marked dirty if
there are no active mapping.  Later when application remaps the hugetlb
file back and all data are gone, triggering catastrophic flip over on
application.

Not only that, the internal resv_huge_pages count will also get all messed
up.  Fix it up by marking page dirty appropriately.
Signed-off-by: NKen Chen <kenchen@google.com>
Cc: "Nish Aravamudan" <nish.aravamudan@gmail.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: <stable@kernel.org>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

To allow a more effective copy_user_highpage() on certain architectures,
a vma argument is added to the function and cow_user_page() allowing
the implementation of these functions to check for the VM_EXEC bit.

The main part of this patch was originally written by Ralf Baechle;
Atushi Nemoto did the the debugging.
Signed-off-by: NAtsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: NRalf Baechle <ralf@linux-mips.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Elaborate the API for calling cpuset_zone_allowed(), so that users have to
explicitly choose between the two variants:

  cpuset_zone_allowed_hardwall()
  cpuset_zone_allowed_softwall()

Until now, whether or not you got the hardwall flavor depended solely on
whether or not you or'd in the __GFP_HARDWALL gfp flag to the gfp_mask
argument.

If you didn't specify __GFP_HARDWALL, you implicitly got the softwall
version.

Unfortunately, this meant that users would end up with the softwall version
without thinking about it.  Since only the softwall version might sleep,
this led to bugs with possible sleeping in interrupt context on more than
one occassion.

The hardwall version requires that the current tasks mems_allowed allows
the node of the specified zone (or that you're in interrupt or that
__GFP_THISNODE is set or that you're on a one cpuset system.)

The softwall version, depending on the gfp_mask, might allow a node if it
was allowed in the nearest enclusing cpuset marked mem_exclusive (which
requires taking the cpuset lock 'callback_mutex' to evaluate.)

This patch removes the cpuset_zone_allowed() call, and forces the caller to
explicitly choose between the hardwall and the softwall case.

If the caller wants the gfp_mask to determine this choice, they should (1)
be sure they can sleep or that __GFP_HARDWALL is set, and (2) invoke the
cpuset_zone_allowed_softwall() routine.

This adds another 100 or 200 bytes to the kernel text space, due to the few
lines of nearly duplicate code at the top of both cpuset_zone_allowed_*
routines.  It should save a few instructions executed for the calls that
turned into calls of cpuset_zone_allowed_hardwall, thanks to not having to
set (before the call) then check (within the call) the __GFP_HARDWALL flag.

For the most critical call, from get_page_from_freelist(), the same
instructions are executed as before -- the old cpuset_zone_allowed()
routine it used to call is the same code as the
cpuset_zone_allowed_softwall() routine that it calls now.

Not a perfect win, but seems worth it, to reduce this chance of hitting a
sleeping with irq off complaint again.
Signed-off-by: NPaul Jackson <pj@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Currently we we use the lru head link of the second page of a compound page
to hold its destructor.  This was ok when it was purely an internal
implmentation detail.  However, hugetlbfs overrides this destructor
violating the layering.  Abstract this out as explicit calls, also
introduce a type for the callback function allowing them to be type
checked.  For each callback we pre-declare the function, causing a type
error on definition rather than on use elsewhere.

[akpm@osdl.org: cleanups]
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Imprecise RSS accounting is an irritating ill effect with pt sharing.  After
consulted with several VM experts, I have tried various methods to solve that
problem: (1) iterate through all mm_structs that share the PT and increment
count; (2) keep RSS count in page table structure and then sum them up at
reporting time.  None of the above methods yield any satisfactory
implementation.

Since process RSS accounting is pure information only, I propose we don't
count them at all for hugetlb page.  rlimit has such field, though there is
absolutely no enforcement on limiting that resource.  One other method is to
account all RSS at hugetlb mmap time regardless they are faulted or not.  I
opt for the simplicity of no accounting at all.

Hugetlb page are special, they are reserved up front in global reservation
pool and is not reclaimable.  From physical memory resource point of view, it
is already consumed regardless whether there are users using them.

If the concern is that RSS can be used to control resource allocation, we
already can specify hugetlb fs size limit and sysadmin can enforce that at
mount time.  Combined with the two points mentioned above, I fail to see if
there is anything got affected because of this patch.
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Acked-by: NHugh Dickins <hugh@veritas.com>
Cc: Dave McCracken <dmccr@us.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Adam Litke <agl@us.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Following up with the work on shared page table done by Dave McCracken.  This
set of patch target shared page table for hugetlb memory only.

The shared page table is particular useful in the situation of large number of
independent processes sharing large shared memory segments.  In the normal
page case, the amount of memory saved from process' page table is quite
significant.  For hugetlb, the saving on page table memory is not the primary
objective (as hugetlb itself already cuts down page table overhead
significantly), instead, the purpose of using shared page table on hugetlb is
to allow faster TLB refill and smaller cache pollution upon TLB miss.

With PT sharing, pte entries are shared among hundreds of processes, the cache
consumption used by all the page table is smaller and in return, application
gets much higher cache hit ratio.  One other effect is that cache hit ratio
with hardware page walker hitting on pte in cache will be higher and this
helps to reduce tlb miss latency.  These two effects contribute to higher
application performance.
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Acked-by: NHugh Dickins <hugh@veritas.com>
Cc: Dave McCracken <dmccr@us.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Adam Litke <agl@us.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

If you truncated an mmap'ed hugetlbfs file, then faulted on the truncated
area, /proc/meminfo's HugePages_Rsvd wrapped hugely "negative".  Reinstate my
preliminary i_size check before attempting to allocate the page (though this
only fixes the most obvious case: more work will be needed here).
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

commit fe1668ae causes kernel to oops with
libhugetlbfs test suite.  The problem is that hugetlb pages can be shared
by multiple mappings.  Multiple threads can fight over page->lru in the
unmap path and bad things happen.  We now serialize __unmap_hugepage_range
to void concurrent linked list manipulation.  Such serialization is also
needed for shared page table page on hugetlb area.  This patch will fixed
the bug and also serve as a prepatch for shared page table.
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Spotted by Hugh that hugetlb page is free'ed back to global pool before
performing any TLB flush in unmap_hugepage_range().  This potentially allow
threads to abuse free-alloc race condition.

The generic tlb gather code is unsuitable to use by hugetlb, I just open
coded a page gathering list and delayed put_page until tlb flush is
performed.

Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Acked-by: NWilliam Irwin <wli@holomorphy.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

There are many places where we need to determine the node of a zone.
Currently we use a difficult to read sequence of pointer dereferencing.
Put that into an inline function and use throughout VM.  Maybe we can find
a way to optimize the lookup in the future.
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

I found two location in hugetlb.c where we chase pointer instead of using
page_to_nid().  Page_to_nid is more effective and can get the node directly
from page flags.
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Current hugetlb strict accounting for shared mapping always assume mapping
starts at zero file offset and reserves pages between zero and size of the
file.  This assumption often reserves (or lock down) a lot more pages then
necessary if application maps at none zero file offset.  libhugetlbfs is
one example that requires proper reservation on shared mapping starts at
none zero offset.

This patch extends the reservation and hugetlb strict accounting to support
any arbitrary pair of (offset, len), resulting a much more robust and
accurate scheme.  More importantly, it won't lock down any hugetlb pages
outside file mapping.
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Acked-by: NAdam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

With strict page reservation, I think kernel should enforce number of free
hugetlb page don't fall below reserved count.  Currently it is possible in
the sysctl path.  Add proper check in sysctl to disallow that.
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

git-commit: d5d4b0aa
"[PATCH] optimize follow_hugetlb_page" breaks mlock on hugepage areas.

I mis-interpret pages argument and made get_page() unconditional.  It
should only get a ref count when "pages" argument is non-null.

Credit goes to Adam Litke who spotted the bug.
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Acked-by: NAdam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Fix bogus node loop in hugetlb.c alloc_fresh_huge_page(), which was
assuming that nodes are numbered contiguously from 0 to num_online_nodes().
Once the hotplug folks get this far, that will be false.
Signed-off-by: NPaul Jackson <pj@sgi.com>
Acked-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

follow_hugetlb_page() walks a range of user virtual address and then fills
in list of struct page * into an array that is passed from the argument
list.  It also gets a reference count via get_page().  For compound page,
get_page() actually traverse back to head page via page_private() macro and
then adds a reference count to the head page.  Since we are doing a virt to
pte look up, kernel already has a struct page pointer into the head page.
So instead of traverse into the small unit page struct and then follow a
link back to the head page, optimize that with incrementing the reference
count directly on the head page.

The benefit is that we don't take a cache miss on accessing page struct for
the corresponding user address and more importantly, not to pollute the
cache with a "not very useful" round trip of pointer chasing.  This adds a
moderate performance gain on an I/O intensive database transaction
workload.
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Originally, mm/hugetlb.c just handled the hugepage physical allocation path
and its {alloc,free}_huge_page() functions were used from the arch specific
hugepage code.  These days those functions are only used with mm/hugetlb.c
itself.  Therefore, this patch makes them static and removes their
prototypes from hugetlb.h.  This requires a small rearrangement of code in
mm/hugetlb.c to avoid a forward declaration.

This patch causes no regressions on the libhugetlbfs testsuite (ppc64,
POWER5).
Signed-off-by: NDavid Gibson <dwg@au1.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

These days, hugepages are demand-allocated at first fault time.  There's a
somewhat dubious (and racy) heuristic when making a new mmap() to check if
there are enough available hugepages to fully satisfy that mapping.

A particularly obvious case where the heuristic breaks down is where a
process maps its hugepages not as a single chunk, but as a bunch of
individually mmap()ed (or shmat()ed) blocks without touching and
instantiating the pages in between allocations.  In this case the size of
each block is compared against the total number of available hugepages.
It's thus easy for the process to become overcommitted, because each block
mapping will succeed, although the total number of hugepages required by
all blocks exceeds the number available.  In particular, this defeats such
a program which will detect a mapping failure and adjust its hugepage usage
downward accordingly.

The patch below addresses this problem, by strictly reserving a number of
physical hugepages for hugepage inodes which have been mapped, but not
instatiated.  MAP_SHARED mappings are thus "safe" - they will fail on
mmap(), not later with an OOM SIGKILL.  MAP_PRIVATE mappings can still
trigger an OOM.  (Actually SHARED mappings can technically still OOM, but
only if the sysadmin explicitly reduces the hugepage pool between mapping
and instantiation)

This patch appears to address the problem at hand - it allows DB2 to start
correctly, for instance, which previously suffered the failure described
above.

This patch causes no regressions on the libhugetblfs testsuite, and makes a
test (designed to catch this problem) pass which previously failed (ppc64,
POWER5).
Signed-off-by: NDavid Gibson <dwg@au1.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Currently, no lock or mutex is held between allocating a hugepage and
inserting it into the pagetables / page cache.  When we do go to insert the
page into pagetables or page cache, we recheck and may free the newly
allocated hugepage.  However, since the number of hugepages in the system
is strictly limited, and it's usualy to want to use all of them, this can
still lead to spurious allocation failures.

For example, suppose two processes are both mapping (MAP_SHARED) the same
hugepage file, large enough to consume the entire available hugepage pool.
If they race instantiating the last page in the mapping, they will both
attempt to allocate the last available hugepage.  One will fail, of course,
returning OOM from the fault and thus causing the process to be killed,
despite the fact that the entire mapping can, in fact, be instantiated.

The patch fixes this race by the simple method of adding a (sleeping) mutex
to serialize the hugepage fault path between allocation and insertion into
pagetables and/or page cache.  It would be possible to avoid the
serialization by catching the allocation failures, waiting on some
condition, then rechecking to see if someone else has instantiated the page
for us.  Given the likely frequency of hugepage instantiations, it seems
very doubtful it's worth the extra complexity.

This patch causes no regression on the libhugetlbfs testsuite, and one
test, which can trigger this race now passes where it previously failed.

Actually, the test still sometimes fails, though less often and only as a
shmat() failure, rather processes getting OOM killed by the VM.  The dodgy
heuristic tests in fs/hugetlbfs/inode.c for whether there's enough hugepage
space aren't protected by the new mutex, and would be ugly to do so, so
there's still a race there.  Another patch to replace those tests with
something saner for this reason as well as others coming...
Signed-off-by: NDavid Gibson <dwg@au1.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Move the loops used in mm/hugetlb.c to clear and copy hugepages to their
own functions for clarity.  As we do so, we add some checks of need_resched
- we are, after all copying megabytes of memory here.  We also add
might_sleep() accordingly.  We generally dropped locks around the clear and
copy, already but not everyone has PREEMPT enabled, so we should still be
checking explicitly.

For this to work, we need to remove the clear_huge_page() from
alloc_huge_page(), which is called with the page_table_lock held in the COW
path.  We move the clear_huge_page() to just after the alloc_huge_page() in
the hugepage no-page path.  In the COW path, the new page is about to be
copied over, so clearing it was just a waste of time anyway.  So as a side
effect we also fix the fact that we held the page_table_lock for far too
long in this path by calling alloc_huge_page() under it.

It causes no regressions on the libhugetlbfs testsuite (ppc64, POWER5).
Signed-off-by: NDavid Gibson <dwg@au1.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

2.6.16-rc3 uses hugetlb on-demand paging, but it doesn_t support hugetlb
mprotect.

From: David Gibson <david@gibson.dropbear.id.au>

  Remove a test from the mprotect() path which checks that the mprotect()ed
  range on a hugepage VMA is hugepage aligned (yes, really, the sense of
  is_aligned_hugepage_range() is the opposite of what you'd guess :-/).

  In fact, we don't need this test.  If the given addresses match the
  beginning/end of a hugepage VMA they must already be suitably aligned.  If
  they don't, then mprotect_fixup() will attempt to split the VMA.  The very
  first test in split_vma() will check for a badly aligned address on a
  hugepage VMA and return -EINVAL if necessary.

From: "Chen, Kenneth W" <kenneth.w.chen@intel.com>

  On i386 and x86-64, pte flag _PAGE_PSE collides with _PAGE_PROTNONE.  The
  identify of hugetlb pte is lost when changing page protection via mprotect.
  A page fault occurs later will trigger a bug check in huge_pte_alloc().

  The fix is to always make new pte a hugetlb pte and also to clean up
  legacy code where _PAGE_PRESENT is forced on in the pre-faulting day.
Signed-off-by: NZhang Yanmin <yanmin.zhang@intel.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Signed-off-by: NNishanth Aravamudan <nacc@us.ibm.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>