There is a name clash with ia64 arch code in Andrew's tree. Rename
is_avialable_memory() to is_memory_available() to avoid the clash.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Tony Luck <[email protected]>
Currently we simply attempt to allocate from all allowed nodes using
GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at
all if the recent GFP_THISNODE patch is accepted). If we truly run out of
memory in the whole system then fallback_alloc may return NULL although
memory may still be available if we would perform more thorough reclaim.
This patch changes fallback_alloc() so that we first only inspect all the
per node queues for available slabs. If we find any then we allocate from
those. This avoids slab fragmentation by first getting rid of all partial
allocated slabs on every node before allocating new memory.
If we cannot satisfy the allocation from any per node queue then we extend
a slab. We now call into the page allocator without specifying
GFP_THISNODE. The page allocator will then implement its own fallback (in
the given cpuset context), perform necessary reclaim (again considering not
a single node but the whole set of allowed nodes) and then return pages for
a new slab.
We identify from which node the pages were allocated and then insert the
pages into the corresponding per node structure. In order to do so we need
to modify cache_grow() to take a parameter that specifies the new slab.
kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be
able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE
needs to be specified when calling cache_grow().
One key advantage is that the decision from which node to allocate new
memory is removed from slab fallback processing. The patch allows to go
back to use of the page allocators fallback/reclaim logic.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The intent of GFP_THISNODE is to make sure that an allocation occurs on a
particular node. If this is not possible then NULL needs to be returned so
that the caller can choose what to do next on its own (the slab allocator
depends on that).
However, GFP_THISNODE currently triggers reclaim before returning a failure
(GFP_THISNODE means GFP_NORETRY is set). If we have over allocated a node
then we will currently do some reclaim before returning NULL. The caller
may want memory from other nodes before reclaim should be triggered. (If
the caller wants reclaim then he can directly use __GFP_THISNODE instead).
There is no flag to avoid reclaim in the page allocator and adding yet
another GFP_xx flag would be difficult given that we are out of available
flags.
So just compare and see if all bits for GFP_THISNODE (__GFP_THISNODE,
__GFP_NORETRY and __GFP_NOWARN) are set. If so then we return NULL before
waking up kswapd.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This addresses two issues:
1. Kmalloc_node() may intermittently return NULL if we are allocating
from the current node and are unable to obtain memory for the current
node from the page allocator. This is because we call ___cache_alloc()
if nodeid == numa_node_id() and ____cache_alloc is not able to fallback
to other nodes.
This was introduced in the 2.6.19 development cycle. <= 2.6.18 in
that case does not do a restricted allocation and blindly trusts the
page allocator to have given us memory from the indicated node. It
inserts the page regardless of the node it came from into the queues for
the current node.
2. If kmalloc_node() is used on a node that has not been bootstrapped
yet then we may try to pass an invalid node number to
____cache_alloc_node() triggering a BUG().
Change the function to call fallback_alloc() instead. Only call
fallback_alloc() if we are allowed to fallback at all. The need to
handle a node not bootstrapped yet also first surfaced in the 2.6.19
cycle.
Update the comments since they were still describing the old kmalloc_node
from 2.6.12.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Replace all uses of kmem_cache_t with struct kmem_cache.
The patch was generated using the following script:
#!/bin/sh
#
# Replace one string by another in all the kernel sources.
#
set -e
for file in `find * -name "*.c" -o -name "*.h"|xargs grep -l $1`; do
quilt add $file
sed -e "1,\$s/$1/$2/g" $file >/tmp/$$
mv /tmp/$$ $file
quilt refresh
done
The script was run like this
sh replace kmem_cache_t "struct kmem_cache"
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
SLAB_DMA is an alias of GFP_DMA. This is the last one so we
remove the leftover comment too.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
SLAB_NOIO is an alias of GFP_NOIO with a single instance of use.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
SLAB_LEVEL_MASK is only used internally to the slab and is
and alias of GFP_LEVEL_MASK.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
drain_node_pages() currently drains the complete pageset of all pages. If
there are a large number of pages in the queues then we may hold off
interrupts for too long.
Duplicate the method used in free_hot_cold_page. Only drain pcp->batch
pages at one time.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Remove all uses of kmem_cache_t (the most were left in slab.h). The
typedef for kmem_cache_t is then only necessary for other kernel
subsystems. Add a comment to that effect.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The names_cachep is used for getname() and putname(). So lets put it into
fs.h near those two definitions.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
fs_cachep is only used in kernel/exit.c and in kernel/fork.c.
It is used to store fs_struct items so it should be placed in linux/fs_struct.h
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
filp_cachep is only used in fs/file_table.c and in fs/dcache.c where
it is defined.
Move it to related definitions in linux/file.h.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Proper place is in file.h since files_cachep uses are rated to file I/O.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
vm_area_cachep is used to store vm_area_structs. So move to mm.h.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Move sighand_cachep definitioni to linux/signal.h
The sighand cache is only used in fs/exec.c and kernel/fork.c. It is defined
in kernel/fork.c but only used in fs/exec.c.
The sighand_cachep is related to signal processing. So add the definition to
signal.h.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Remove bio_cachep from slab.h - it no longer exists.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The zone table is mostly not needed. If we have a node in the page flags
then we can get to the zone via NODE_DATA() which is much more likely to be
already in the cpu cache.
In case of SMP and UP NODE_DATA() is a constant pointer which allows us to
access an exact replica of zonetable in the node_zones field. In all of
the above cases there will be no need at all for the zone table.
The only remaining case is if in a NUMA system the node numbers do not fit
into the page flags. In that case we make sparse generate a table that
maps sections to nodes and use that table to to figure out the node number.
This table is sized to fit in a single cache line for the known 32 bit
NUMA platform which makes it very likely that the information can be
obtained without a cache miss.
For sparsemem the zone table seems to be have been fairly large based on
the maximum possible number of sections and the number of zones per node.
There is some memory saving by removing zone_table. The main benefit is to
reduce the cache foootprint of the VM from the frequent lookups of zones.
Plus it simplifies the page allocator.
[[email protected]: build fix]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Andy Whitcroft <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The zonelist may contain zones of nodes that have not been bootstrapped and
we will oops if we try to allocate from those zones. So check if the node
information for the slab and the node have been setup before attempting an
allocation. If it has not been setup then skip that zone.
Usually we will not encounter this situation since the slab bootstrap code
avoids falling back before we have setup the respective nodes but we seem
to have a special needs for pppc.
Signed-off-by: Christoph Lameter <[email protected]>
Acked-by: Andy Whitcroft <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Mike Kravetz <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Acked-by: Mel Gorman <[email protected]>
Acked-by: Will Schmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Init list is called with a list parameter that is not equal to the
cachep->nodelists entry under NUMA if more than one node exists. This is
fully legitimatei. One may want to populate the list fields before
switching nodelist pointers.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
When the per cpu sched domains are build then they also need to be placed
on the node where the cpu resides otherwise we will have frequent off node
accesses which will slow down the system.
Signed-off-by: Christoph Lameter <[email protected]>
Acked-by: Ingo Molnar <[email protected]>
Acked-by: Nick Piggin <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Now that we have the node in the hot zone of struct zone we can avoid
accessing zone_pgdat in zone_statistics.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
We do not need to allocate pagesets for unpopulated zones.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch insures that the slab node lists in the NUMA case only contain
slabs that belong to that specific node. All slab allocations use
GFP_THISNODE when calling into the page allocator. If an allocation fails
then we fall back in the slab allocator according to the zonelists appropriate
for a certain context.
This allows a replication of the behavior of alloc_pages and alloc_pages node
in the slab layer.
Currently allocations requested from the page allocator may be redirected via
cpusets to other nodes. This results in remote pages on nodelists and that in
turn results in interrupt latency issues during cache draining. Plus the slab
is handing out memory as local when it is really remote.
Fallback for slab memory allocations will occur within the slab allocator and
not in the page allocator. This is necessary in order to be able to use the
existing pools of objects on the nodes that we fall back to before adding more
pages to a slab.
The fallback function insures that the nodes we fall back to obey cpuset
restrictions of the current context. We do not allocate objects from outside
of the current cpuset context like before.
Note that the implementation of locality constraints within the slab allocator
requires importing logic from the page allocator. This is a mischmash that is
not that great. Other allocators (uncached allocator, vmalloc, huge pages)
face similar problems and have similar minimal reimplementations of the basic
fallback logic of the page allocator. There is another way of implementing a
slab by avoiding per node lists (see modular slab) but this wont work within
the existing slab.
V1->V2:
- Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA
- Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another
#ifdef
[[email protected]: build fix]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
GFP_THISNODE must be set to 0 in the non numa case otherwise we disable retry
and warnings for failing allocations in the SMP and UP case.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The NUMA_BUILD constant is always available and will be set to 1 on
NUMA_BUILDs. That way checks valid only under CONFIG_NUMA can easily be done
without #ifdef CONFIG_NUMA
F.e.
if (NUMA_BUILD && <numa_condition>) {
...
}
[akpm: not a thing we'd normally do, but CONFIG_NUMA is special: it is
causing ifdef explosion in core kernel, so let's see if this is a comfortable
way in whcih to control that]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
kmalloc_node() falls back to ___cache_alloc() under certain conditions and
at that point memory policies may be applied redirecting the allocation
away from the current node. Therefore kmalloc_node(...,numa_node_id()) or
kmalloc_node(...,-1) may not return memory from the local node.
Fix this by doing the policy check in __cache_alloc() instead of
____cache_alloc().
This version here is a cleanup of Kiran's patch.
- Tested on ia64.
- Extra material removed.
- Consolidate the exit path if alternate_node_alloc() returned an object.
[[email protected]: warning fix]
Signed-off-by: Alok N Kataria <[email protected]>
Signed-off-by: Ravikiran Thirumalai <[email protected]>
Signed-off-by: Shai Fultheim <[email protected]>
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The GFP_DMA option usually does nothing on SN2 since all of memory is in thei
DMA zone and the BTE has always been capable of addressing all of memory.
So there is no need to get memory from a restricted range of memory (which
is what GFP_DMA is for).
Remove useless __GFP_DMA option.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Tony Luck <[email protected]>
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: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
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: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Minor performance fix.
If we reclaimed enough slab pages from a zone then we can avoid going off
node with the current allocation. Take care of updating nr_reclaimed when
reclaiming from the slab.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently one can enable slab reclaim by setting an explicit option in
/proc/sys/vm/zone_reclaim_mode. Slab reclaim is then used as a final
option if the freeing of unmapped file backed pages is not enough to free
enough pages to allow a local allocation.
However, that means that the slab can grow excessively and that most memory
of a node may be used by slabs. We have had a case where a machine with
46GB of memory was using 40-42GB for slab. Zone reclaim was effective in
dealing with pagecache pages. However, slab reclaim was only done during
global reclaim (which is a bit rare on NUMA systems).
This patch implements slab reclaim during zone reclaim. Zone reclaim
occurs if there is a danger of an off node allocation. At that point we
1. Shrink the per node page cache if the number of pagecache
pages is more than min_unmapped_ratio percent of pages in a zone.
2. Shrink the slab cache if the number of the nodes reclaimable slab pages
(patch depends on earlier one that implements that counter)
are more than min_slab_ratio (a new /proc/sys/vm tunable).
The shrinking of the slab cache is a bit problematic since it is not node
specific. So we simply calculate what point in the slab we want to reach
(current per node slab use minus the number of pages that neeed to be
allocated) and then repeately run the global reclaim until that is
unsuccessful or we have reached the limit. I hope we will have zone based
slab reclaim at some point which will make that easier.
The default for the min_slab_ratio is 5%
Also remove the slab option from /proc/sys/vm/zone_reclaim_mode.
[[email protected]: cleanups]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Remove the atomic counter for slab_reclaim_pages and replace the counter
and NR_SLAB with two ZVC counter that account for unreclaimable and
reclaimable slab pages: NR_SLAB_RECLAIMABLE and NR_SLAB_UNRECLAIMABLE.
Change the check in vmscan.c to refer to to NR_SLAB_RECLAIMABLE. The
intend seems to be to check for slab pages that could be freed.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
*_pages is a better description of the role of the variable.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The allocpercpu functions __alloc_percpu and __free_percpu() are heavily
using the slab allocator. However, they are conceptually slab. This also
simplifies SLOB (at this point slob may be broken in mm. This should fix
it).
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Matt Mackall <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
If a zone is unpopulated then we do not need to check for pages that are to
be drained and also not for vm counters that may need to be updated.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Free one_page currently adds the page to a fake list and calls
free_page_bulk. Fee_page_bulk takes it off again and then calles
__free_one_page.
Make free_one_page go directly to __free_one_page. Saves list on / off and
a temporary list in free_one_page for higher ordered pages.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
In many places we will need to use the same combination of flags. Specify
a single GFP_THISNODE definition for ease of use in gfp.h.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Profiling really suffers with off node buffers. Fail if no memory is
available on the nodes. The profiling code can deal with these failures
should they occur.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
There are frequent references to *z in get_page_from_freelist.
Add an explicit zone variable that can be used in all these places.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The uncached allocator manages per node pools. Specify __GFP_THISNODE in
order to force allocation on the indicated node or fail. The uncached
allocator has already logic to deal with failing allocations.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Andy Whitcroft <[email protected]>
Cc: Mel Gorman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Add a new gfp flag __GFP_THISNODE to avoid fallback to other nodes. This
flag is essential if a kernel component requires memory to be located on a
certain node. It will be needed for alloc_pages_node() to force allocation
on the indicated node and for alloc_pages() to force allocation on the
current node.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Andy Whitcroft <[email protected]>
Cc: Mel Gorman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
It is fairly easy to get a system to oops by simply sizing a cache via
/proc in such a way that one of the chaches (shared is easiest) becomes
bigger than the maximum allowed slab allocation size. This occurs because
enable_cpucache() fails if it cannot reallocate some caches.
However, enable_cpucache() is used for multiple purposes: resizing caches,
cache creation and bootstrap.
If the slab is already up then we already have working caches. The resize
can fail without a problem. We just need to return the proper error code.
F.e. after this patch:
# echo "size-64 10000 50 1000" >/proc/slabinfo
-bash: echo: write error: Cannot allocate memory
notice no OOPS.
If we are doing a kmem_cache_create() then we also should not panic but
return -ENOMEM.
If on the other hand we do not have a fully bootstrapped slab allocator yet
then we should indeed panic since we are unable to bring up the slab to its
full functionality.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Pekka Enberg <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The ability to free memory allocated to a slab cache is also useful if an
error occurs during setup of a slab. So extract the function.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Pekka Enberg <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
I wonder why we need this bitmask indexing into zone->node_zonelists[]?
We always start with the highest zone and then include all lower zones
if we build zonelists.
Are there really cases where we need allocation from ZONE_DMA or
ZONE_HIGHMEM but not ZONE_NORMAL? It seems that the current implementation
of highest_zone() makes that already impossible.
If we go linear on the index then gfp_zone() == highest_zone() and a lot
of definitions fall by the wayside.
We can now revert back to the use of gfp_zone() in mempolicy.c ;-)
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
After we have done this we can now do some typing cleanup.
The memory policy layer keeps a policy_zone that specifies
the zone that gets memory policies applied. This variable
can now be of type enum zone_type.
The check_highest_zone function and the build_zonelists funnctionm must
then also take a enum zone_type parameter.
Plus there are a number of loops over zones that also should use
zone_type.
We run into some troubles at some points with functions that need a
zone_type variable to become -1. Fix that up.
[[email protected]: fix set_mempolicy() crash]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Paul Jackson <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
There is a check in zonelist_policy that compares pieces of the bitmap
obtained from a gfp mask via GFP_ZONETYPES with a zone number in function
zonelist_policy().
The bitmap is an ORed mask of __GFP_DMA, __GFP_DMA32 and __GFP_HIGHMEM.
The policy_zone is a zone number with the possible values of ZONE_DMA,
ZONE_DMA32, ZONE_HIGHMEM and ZONE_NORMAL. These are two different domains
of values.
For some reason seemed to work before the zone reduction patchset (It
definitely works on SGI boxes since we just have one zone and the check
cannot fail).
With the zone reduction patchset this check definitely fails on systems
with two zones if the system actually has memory in both zones.
This is because ZONE_NORMAL is selected using no __GFP flag at
all and thus gfp_zone(gfpmask) == 0. ZONE_DMA is selected when __GFP_DMA
is set. __GFP_DMA is 0x01. So gfp_zone(gfpmask) == 1.
policy_zone is set to ZONE_NORMAL (==1) if ZONE_NORMAL and ZONE_DMA are
populated.
For ZONE_NORMAL gfp_zone(<no _GFP_DMA>) yields 0 which is <
policy_zone(ZONE_NORMAL) and so policy is not applied to regular memory
allocations!
Instead gfp_zone(__GFP_DMA) == 1 which results in policy being applied
to DMA allocations!
What we realy want in that place is to establish the highest allowable
zone for a given gfp_mask. If the highest zone is higher or equal to the
policy_zone then memory policies need to be applied. We have such
a highest_zone() function in page_alloc.c.
So move the highest_zone() function from mm/page_alloc.c into
include/linux/gfp.h. On the way we simplify the function and use the new
zone_type that was also introduced with the zone reduction patchset plus we
also specify the right type for the gfp flags parameter.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Lee Schermerhorn <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
We cannot check MAX_NR_ZONES since it not defined in the preprocessor
anymore.
So remove the check.
The maximum number of zones per node for i386 is 3 since i386 does not
support ZONE_DMA32.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
eventcounters: Do not display counters for zones that are not available on an
arch
Do not define or display counters for the DMA32 and the HIGHMEM zone if such
zones were not configured.
[[email protected]: s390 fix]
[[email protected]: s390 fix]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Make ZONE_HIGHMEM optional
- ifdef out code and definitions related to CONFIG_HIGHMEM
- __GFP_HIGHMEM falls back to normal allocations if there is no
ZONE_HIGHMEM
- GFP_ZONEMASK becomes 0x01 if there is no DMA32 and no HIGHMEM
zone.
[[email protected]: build fix]
Signed-off-by: Jeff Dike <[email protected]>
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Make ZONE_DMA32 optional
- Add #ifdefs around ZONE_DMA32 specific code and definitions.
- Add CONFIG_ZONE_DMA32 config option and use that for x86_64
that alone needs this zone.
- Remove the use of CONFIG_DMA_IS_DMA32 and CONFIG_DMA_IS_NORMAL
for ia64 and fix up the way per node ZVCs are calculated.
- Fall back to prior GFP_ZONEMASK of 0x03 if there is no
DMA32 zone.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Use enum for zones and reformat zones dependent information
Add comments explaning the use of zones and add a zones_t type for zone
numbers.
Line up information that will be #ifdefd by the following patches.
[[email protected]: comment cleanups]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
page allocator ZONE_HIGHMEM fixups
1. We do not need to do an #ifdef in si_meminfo since both counters
in use are zero if !CONFIG_HIGHMEM.
2. Add #ifdef in si_meminfo_node instead to avoid referencing zone
information for ZONE_HIGHMEM if we do not have HIGHMEM
(may not be there after the following patches).
3. Replace the use of ZONE_HIGHMEM with MAX_NR_ZONES in build_zonelists_node
4. build_zonelists_node: Remove BUG_ON for ZONE_HIGHMEM. Zone will
be optional soon and thus BUG_ON cannot be triggered anymore.
5. init_free_area_core: Replace a use of ZONE_HIGHMEM with NR_MAX_ZONES.
[[email protected]: cleanups]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Move totalhigh_pages and nr_free_highpages() into highmem.c/.h
Move the totalhigh_pages definition into highmem.c/.h. Move the
nr_free_highpages function into highmem.c
[[email protected]: build fix]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Yoichi Yuasa <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Do not display HIGHMEM memory sizes if CONFIG_HIGHMEM is not set.
Make HIGHMEM dependent texts and make display of highmem counters optional
Some texts are depending on CONFIG_HIGHMEM.
Remove those strings and remove the display of highmem counter values if
CONFIG_HIGHMEM is not set.
[[email protected]: remove some ifdefs]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Fix array initialization in lots of arches
The number of zones may now be reduced from 4 to 2 for many arches. Fix the
array initialization for the zones array for all architectures so that it is
not initializing a fixed number of elements.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
I keep seeing zones on various platforms that are never used and wonder why we
compile support for them into the kernel. Counters show up for HIGHMEM and
DMA32 that are alway zero.
This patch allows the removal of ZONE_DMA32 for non x86_64 architectures and
it will get rid of ZONE_HIGHMEM for arches not using highmem (like 64 bit
architectures). If an arch does not define CONFIG_HIGHMEM then ZONE_HIGHMEM
will not be defined. Similarly if an arch does not define CONFIG_ZONE_DMA32
then ZONE_DMA32 will not be defined.
No current architecture uses all the 4 zones (DMA,DMA32,NORMAL,HIGH) that we
have now. The patchset will reduce the number of zones for all platforms.
On many platforms that do not have DMA32 or HIGHMEM this will reduce the
number of zones by 50%. F.e. ia64 only uses DMA and NORMAL.
Large amounts of memory can be saved for larger systemss that may have a few
hundred NUMA nodes.
With ZONE_DMA32 and ZONE_HIGHMEM support optional MAX_NR_ZONES will be 2 for
many non i386 platforms and even for i386 without CONFIG_HIGHMEM set.
Tested on ia64, x86_64 and on i386 with and without highmem.
The patchset consists of 11 patches that are following this message.
One could go even further than this patchset and also make ZONE_DMA optional
because some platforms do not need a separate DMA zone and can do DMA to all
of memory. This could reduce MAX_NR_ZONES to 1. Such a patchset will
hopefully follow soon.
This patch:
Fix strange uses of MAX_NR_ZONES
Sometimes we use MAX_NR_ZONES - x to refer to a zone. Make that explicit.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The ZVC counter update threshold is currently set to a fixed value of 32.
This patch sets up the threshold depending on the number of processors and
the sizes of the zones in the system.
With the current threshold of 32, I was able to observe slight contention
when more than 130-140 processors concurrently updated the counters. The
contention vanished when I either increased the threshold to 64 or used
Andrew's idea of overstepping the interval (see ZVC overstep patch).
However, we saw contention again at 220-230 processors. So we need higher
values for larger systems.
But the current default is already a bit of an overkill for smaller
systems. Some systems have tiny zones where precision matters. For
example i386 and x86_64 have 16M DMA zones and either 900M ZONE_NORMAL or
ZONE_DMA32. These are even present on SMP and NUMA systems.
The patch here sets up a threshold based on the number of processors in the
system and the size of the zone that these counters are used for. The
threshold should grow logarithmically, so we use fls() as an easy
approximation.
Results of tests on a system with 1024 processors (4TB RAM)
The following output is from a test allocating 1GB of memory concurrently
on each processor (Forking the process. So contention on mmap_sem and the
pte locks is not a factor):
X MIN
TYPE: CPUS WALL WALL SYS USER TOTCPU
fork 1 0.552 0.552 0.540 0.012 0.552
fork 4 0.552 0.548 2.164 0.036 2.200
fork 16 0.564 0.548 8.812 0.164 8.976
fork 128 0.580 0.572 72.204 1.208 73.412
fork 256 1.300 0.660 310.400 2.160 312.560
fork 512 3.512 0.696 1526.836 4.816 1531.652
fork 1020 20.024 0.700 17243.176 6.688 17249.863
So a threshold of 32 is fine up to 128 processors. At 256 processors contention
becomes a factor.
Overstepping the counter (earlier patch) improves the numbers a bit:
fork 4 0.552 0.548 2.164 0.040 2.204
fork 16 0.552 0.548 8.640 0.148 8.788
fork 128 0.556 0.548 69.676 0.956 70.632
fork 256 0.876 0.636 212.468 2.108 214.576
fork 512 2.276 0.672 997.324 4.260 1001.584
fork 1020 13.564 0.680 11586.436 6.088 11592.523
Still contention at 512 and 1020. Contention at 1020 is down by a third.
256 still has a slight bit of contention.
After this patch the counter threshold will be set to 125 which reduces
contention significantly:
fork 128 0.560 0.548 69.776 0.932 70.708
fork 256 0.636 0.556 143.460 2.036 145.496
fork 512 0.640 0.548 284.244 4.236 288.480
fork 1020 1.500 0.588 1326.152 8.892 1335.044
[[email protected]: !SMP build fix]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Increments and decrements are usually grouped rather than mixed. We can
optimize the inc and dec functions for that case.
Increment and decrement the counters by 50% more than the threshold in
those cases and set the differential accordingly. This decreases the need
to update the atomic counters.
The idea came originally from Andrew Morton. The overstepping alone was
sufficient to address the contention issue found when updating the global
and the per zone counters from 160 processors.
Also remove some code in dec_zone_page_state.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Change both the NODES_SHIFT and the NR_CPUS so that even big machines
can boot all nodes and processors with a generic kernel.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Tony Luck <[email protected]>
It turns out that there is a way to build a kernel with NUMA and no SMP.
In that case we are missing one definition __inc_zone_state.
Provide that missing __inc_zone_state.
(akpm: NUMA && !SMP sounds odd, but I am told "But there is the concept of
cpuless nodes. A NUMA system without SMP has a single processor but multiple
memory nodes. This used to work before on IA64 (wasn't aware of it, never seen
anyone with this kind of thing).")
Acked-by: Tony Luck <[email protected]>
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
It turns out that it is advantageous to leave a small portion of unmapped file
backed pages if all of a zone's pages (or almost all pages) are allocated and
so the page allocator has to go off-node.
This allows recently used file I/O buffers to stay on the node and
reduces the times that zone reclaim is invoked if file I/O occurs
when we run out of memory in a zone.
The problem is that zone reclaim runs too frequently when the page cache is
used for file I/O (read write and therefore unmapped pages!) alone and we have
almost all pages of the zone allocated. Zone reclaim may remove 32 unmapped
pages. File I/O will use these pages for the next read/write requests and the
unmapped pages increase. After the zone has filled up again zone reclaim will
remove it again after only 32 pages. This cycle is too inefficient and there
are potentially too many zone reclaim cycles.
With the 1% boundary we may still remove all unmapped pages for file I/O in
zone reclaim pass. However. it will take a large number of read and writes
to get back to 1% again where we trigger zone reclaim again.
The zone reclaim 2.6.16/17 does not show this behavior because we have a 30
second timeout.
[[email protected]: rename the /proc file and the variable]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Post and discussion:
http://marc.theaimsgroup.com/?t=115074342800003&r=1&w=2
Code in __shrink_node() duplicates code in cache_reap()
Add a new function drain_freelist that removes slabs with objects that are
already free and use that in various places.
This eliminates the __node_shrink() function and provides the interrupt
holdoff reduction from slab_free to code that used to call __node_shrink.
[[email protected]: build fixes]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Pekka Enberg <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The remaining counters in page_state after the zoned VM counter patches
have been applied are all just for show in /proc/vmstat. They have no
essential function for the VM.
We use a simple increment of per cpu variables. In order to avoid the most
severe races we disable preempt. Preempt does not prevent the race between
an increment and an interrupt handler incrementing the same statistics
counter. However, that race is exceedingly rare, we may only loose one
increment or so and there is no requirement (at least not in kernel) that
the vm event counters have to be accurate.
In the non preempt case this results in a simple increment for each
counter. For many architectures this will be reduced by the compiler to a
single instruction. This single instruction is atomic for i386 and x86_64.
And therefore even the rare race condition in an interrupt is avoided for
both architectures in most cases.
The patchset also adds an off switch for embedded systems that allows a
building of linux kernels without these counters.
The implementation of these counters is through inline code that hopefully
results in only a single instruction increment instruction being emitted
(i386, x86_64) or in the increment being hidden though instruction
concurrency (EPIC architectures such as ia64 can get that done).
Benefits:
- VM event counter operations usually reduce to a single inline instruction
on i386 and x86_64.
- No interrupt disable, only preempt disable for the preempt case.
Preempt disable can also be avoided by moving the counter into a spinlock.
- Handling is similar to zoned VM counters.
- Simple and easily extendable.
- Can be omitted to reduce memory use for embedded use.
References:
RFC http://marc.theaimsgroup.com/?l=linux-kernel&m=113512330605497&w=2
RFC http://marc.theaimsgroup.com/?l=linux-kernel&m=114988082814934&w=2
local_t http://marc.theaimsgroup.com/?l=linux-kernel&m=114991748606690&w=2
V2 http://marc.theaimsgroup.com/?t=115014808400007&r=1&w=2
V3 http://marc.theaimsgroup.com/?l=linux-kernel&m=115024767022346&w=2
V4 http://marc.theaimsgroup.com/?l=linux-kernel&m=115047968808926&w=2
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The numa statistics are really event counters. But they are per node and
so we have had special treatment for these counters through additional
fields on the pcp structure. We can now use the per zone nature of the
zoned VM counters to realize these.
This will shrink the size of the pcp structure on NUMA systems. We will
have some room to add additional per zone counters that will all still fit
in the same cacheline.
Bits Prior pcp size Size after patch We can add
------------------------------------------------------------------
64 128 bytes (16 words) 80 bytes (10 words) 48
32 76 bytes (19 words) 56 bytes (14 words) 8 (64 byte cacheline)
72 (128 byte)
Remove the special statistics for numa and replace them with zoned vm
counters. This has the side effect that global sums of these events now
show up in /proc/vmstat.
Also take the opportunity to move the zone_statistics() function from
page_alloc.c into vmstat.c.
Discussions:
V2 http://marc.theaimsgroup.com/?t=115048227000002&r=1&w=2
Signed-off-by: Christoph Lameter <[email protected]>
Acked-by: Andi Kleen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Remove writeback state
We can remove some functions now that were needed to calculate the page state
for writeback control since these statistics are now directly available.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Conversion of nr_bounce to a per zone counter
nr_bounce is only used for proc output. So it could be left as an event
counter. However, the event counters may not be accurate and nr_bounce is
categorizing types of pages in a zone. So we really need this to also be a
per zone counter.
[[email protected]: bugfix]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Conversion of nr_unstable to a per zone counter
We need to do some special modifications to the nfs code since there are
multiple cases of disposition and we need to have a page ref for proper
accounting.
This converts the last critical page state of the VM and therefore we need to
remove several functions that were depending on GET_PAGE_STATE_LAST in order
to make the kernel compile again. We are only left with event type counters
in page state.
[[email protected]: bugfixes]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Conversion of nr_writeback to per zone counter.
This removes the last page_state counter from arch/i386/mm/pgtable.c so we
drop the page_state from there.
[[email protected]: bugfix]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This makes nr_dirty a per zone counter. Looping over all processors is
avoided during writeback state determination.
The counter aggregation for nr_dirty had to be undone in the NFS layer since
we summed up the page counts from multiple zones. Someone more familiar with
NFS should probably review what I have done.
[[email protected]: bugfix]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
- Allows reclaim to access counter without looping over processor counts.
- Allows accurate statistics on how many pages are used in a zone by
the slab. This may become useful to balance slab allocations over
various zones.
[[email protected]: bugfix]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The zone_reclaim_interval was necessary because we were not able to determine
how many unmapped pages exist in a zone. Therefore we had to scan in
intervals to figure out if any pages were unmapped.
With the zoned counters and NR_ANON_PAGES we now know the number of pagecache
pages and the number of mapped pages in a zone. So we can simply skip the
reclaim if there is an insufficient number of unmapped pages. We use
SWAP_CLUSTER_MAX as the boundary.
Drop all support for /proc/sys/vm/zone_reclaim_interval.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The current NR_FILE_MAPPED is used by zone reclaim and the dirty load
calculation as the number of mapped pagecache pages. However, that is not
true. NR_FILE_MAPPED includes the mapped anonymous pages. This patch
separates those and therefore allows an accurate tracking of the anonymous
pages per zone.
It then becomes possible to determine the number of unmapped pages per zone
and we can avoid scanning for unmapped pages if there are none.
Also it may now be possible to determine the mapped/unmapped ratio in
get_dirty_limit. Isnt the number of anonymous pages irrelevant in that
calculation?
Note that this will change the meaning of the number of mapped pages reported
in /proc/vmstat /proc/meminfo and in the per node statistics. This may affect
user space tools that monitor these counters! NR_FILE_MAPPED works like
NR_FILE_DIRTY. It is only valid for pagecache pages.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
We can now access the number of pages in a mapped state in an inexpensive way
in shrink_active_list. So drop the nr_mapped field from scan_control.
[[email protected]: bugfix]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently a single atomic variable is used to establish the size of the page
cache in the whole machine. The zoned VM counters have the same method of
implementation as the nr_pagecache code but also allow the determination of
the pagecache size per zone.
Remove the special implementation for nr_pagecache and make it a zoned counter
named NR_FILE_PAGES.
Updates of the page cache counters are always performed with interrupts off.
We can therefore use the __ variant here.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
nr_mapped is important because it allows a determination of how many pages of
a zone are not mapped, which would allow a more efficient means of determining
when we need to reclaim memory in a zone.
We take the nr_mapped field out of the page state structure and define a new
per zone counter named NR_FILE_MAPPED (the anonymous pages will be split off
from NR_MAPPED in the next patch).
We replace the use of nr_mapped in various kernel locations. This avoids the
looping over all processors in try_to_free_pages(), writeback, reclaim (swap +
zone reclaim).
[[email protected]: bugfix]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Per zone counter infrastructure
The counters that we currently have for the VM are split per processor. The
processor however has not much to do with the zone these pages belong to. We
cannot tell f.e. how many ZONE_DMA pages are dirty.
So we are blind to potentially inbalances in the usage of memory in various
zones. F.e. in a NUMA system we cannot tell how many pages are dirty on a
particular node. If we knew then we could put measures into the VM to balance
the use of memory between different zones and different nodes in a NUMA
system. For example it would be possible to limit the dirty pages per node so
that fast local memory is kept available even if a process is dirtying huge
amounts of pages.
Another example is zone reclaim. We do not know how many unmapped pages exist
per zone. So we just have to try to reclaim. If it is not working then we
pause and try again later. It would be better if we knew when it makes sense
to reclaim unmapped pages from a zone. This patchset allows the determination
of the number of unmapped pages per zone. We can remove the zone reclaim
interval with the counters introduced here.
Futhermore the ability to have various usage statistics available will allow
the development of new NUMA balancing algorithms that may be able to improve
the decision making in the scheduler of when to move a process to another node
and hopefully will also enable automatic page migration through a user space
program that can analyse the memory load distribution and then rebalance
memory use in order to increase performance.
The counter framework here implements differential counters for each processor
in struct zone. The differential counters are consolidated when a threshold
is exceeded (like done in the current implementation for nr_pageache), when
slab reaping occurs or when a consolidation function is called.
Consolidation uses atomic operations and accumulates counters per zone in the
zone structure and also globally in the vm_stat array. VM functions can
access the counts by simply indexing a global or zone specific array.
The arrangement of counters in an array also simplifies processing when output
has to be generated for /proc/*.
Counters can be updated by calling inc/dec_zone_page_state or
_inc/dec_zone_page_state analogous to *_page_state. The second group of
functions can be called if it is known that interrupts are disabled.
Special optimized increment and decrement functions are provided. These can
avoid certain checks and use increment or decrement instructions that an
architecture may provide.
We also add a new CONFIG_DMA_IS_NORMAL that signifies that an architecture can
do DMA to all memory and therefore ZONE_NORMAL will not be populated. This is
only currently set for IA64 SGI SN2 and currently only affects
node_page_state(). In the best case node_page_state can be reduced to
retrieving a single counter for the one zone on the node.
[[email protected]: cleanups]
[[email protected]: export vm_stat[] for filesystems]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
NOTE: ZVC are *not* the lightweight event counters. ZVCs are reliable whereas
event counters do not need to be.
Zone based VM statistics are necessary to be able to determine what the state
of memory in one zone is. In a NUMA system this can be helpful for local
reclaim and other memory optimizations that may be able to shift VM load in
order to get more balanced memory use.
It is also useful to know how the computing load affects the memory
allocations on various zones. This patchset allows the retrieval of that data
from userspace.
The patchset introduces a framework for counters that is a cross between the
existing page_stats --which are simply global counters split per cpu-- and the
approach of deferred incremental updates implemented for nr_pagecache.
Small per cpu 8 bit counters are added to struct zone. If the counter exceeds
certain thresholds then the counters are accumulated in an array of
atomic_long in the zone and in a global array that sums up all zone values.
The small 8 bit counters are next to the per cpu page pointers and so they
will be in high in the cpu cache when pages are allocated and freed.
Access to VM counter information for a zone and for the whole machine is then
possible by simply indexing an array (Thanks to Nick Piggin for pointing out
that approach). The access to the total number of pages of various types does
no longer require the summing up of all per cpu counters.
Benefits of this patchset right now:
- Ability for UP and SMP configuration to determine how memory
is balanced between the DMA, NORMAL and HIGHMEM zones.
- loops over all processors are avoided in writeback and
reclaim paths. We can avoid caching the writeback information
because the needed information is directly accessible.
- Special handling for nr_pagecache removed.
- zone_reclaim_interval vanishes since VM stats can now determine
when it is worth to do local reclaim.
- Fast inline per node page state determination.
- Accurate counters in /sys/devices/system/node/node*/meminfo. Current
counters are counting simply which processor allocated a page somewhere
and guestimate based on that. So the counters were not useful to show
the actual distribution of page use on a specific zone.
- The swap_prefetch patch requires per node statistics in order to
figure out when processors of a node can prefetch. This patch provides
some of the needed numbers.
- Detailed VM counters available in more /proc and /sys status files.
References to earlier discussions:
V1 http://marc.theaimsgroup.com/?l=linux-kernel&m=113511649910826&w=2
V2 http://marc.theaimsgroup.com/?l=linux-kernel&m=114980851924230&w=2
V3 http://marc.theaimsgroup.com/?l=linux-kernel&m=115014697910351&w=2
V4 http://marc.theaimsgroup.com/?l=linux-kernel&m=115024767318740&w=2
Performance tests with AIM7 did not show any regressions. Seems to be a tad
faster even. Tested on ia64/NUMA. Builds fine on i386, SMP / UP. Includes
fixes for s390/arm/uml arch code.
This patch:
Move counter code from page_alloc.c/page-flags.h to vmstat.c/h.
Create vmstat.c/vmstat.h by separating the counter code and the proc
functions.
Move the vm_stat_text array before zoneinfo_show.
[[email protected]: s390 build fix]
[[email protected]: HOTPLUG_CPU build fix]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Heiko Carstens <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Cc: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Hugh clarified the role of VM_LOCKED. So we can now implement page
migration for mlocked pages.
Allow the migration of mlocked pages. This means that try_to_unmap must
unmap mlocked pages in the migration case.
Signed-off-by: Christoph Lameter <[email protected]>
Acked-by: Hugh Dickins <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Hooks for calling vma specific migration functions
With this patch a vma may define a vma->vm_ops->migrate function. That
function may perform page migration on its own (some vmas may not contain page
structs and therefore cannot be handled by regular page migration. Pages in a
vma may require special preparatory treatment before migration is possible
etc) . Only mmap_sem is held when the migration function is called. The
migrate() function gets passed two sets of nodemasks describing the source and
the target of the migration. The flags parameter either contains
MPOL_MF_MOVE which means that only pages used exclusively by
the specified mm should be moved
or
MPOL_MF_MOVE_ALL which means that pages shared with other processes
should also be moved.
The migration function returns 0 on success or an error condition. An error
condition will prevent regular page migration from occurring.
On its own this patch cannot be included since there are no users for this
functionality. But it seems that the uncached allocator will need this
functionality at some point.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Andi Kleen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Remove VM_LOCKED before remap_pfn range from device drivers and get rid of
VM_SHM.
remap_pfn_range() already sets VM_IO. There is no need to set VM_SHM since
it does nothing. VM_LOCKED is of no use since the remap_pfn_range does not
place pages on the LRU. The pages are therefore never subject to swap
anyways. Remove all the vm_flags settings before calling remap_pfn_range.
After removing all the vm_flag settings no use of VM_SHM is left. Drop it.
Signed-off-by: Christoph Lameter <[email protected]>
Acked-by: Hugh Dickins <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The definition of the third parameter is a pointer to an array of virtual
addresses which give us some trouble. The existing code calculated the
wrong address in the array since I used void to avoid having to specify a
type.
I now use the correct type "compat_uptr_t __user *" in the definition of
the function in kernel/compat.c.
However, I used __u32 in syscalls.h. Would have to include compat.h there
in order to provide the same definition which would generate an ugly
include situation.
On both ia64 and x86_64 compat_uptr_t is u32. So this works although
parameter declarations differ.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
sys_move_pages() support for 32bit (i386 plus x86_64 compat layer)
Add support for move_pages() on i386 and also add the compat functions
necessary to run 32 bit binaries on x86_64.
Add compat_sys_move_pages to the x86_64 32bit binary layer. Note that it is
not up to date so I added the missing pieces. Not sure if this is done the
right way.
[[email protected]: compile fix]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Andi Kleen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
move_pages() is used to move individual pages of a process. The function can
be used to determine the location of pages and to move them onto the desired
node. move_pages() returns status information for each page.
long move_pages(pid, number_of_pages_to_move,
addresses_of_pages[],
nodes[] or NULL,
status[],
flags);
The addresses of pages is an array of void * pointing to the
pages to be moved.
The nodes array contains the node numbers that the pages should be moved
to. If a NULL is passed instead of an array then no pages are moved but
the status array is updated. The status request may be used to determine
the page state before issuing another move_pages() to move pages.
The status array will contain the state of all individual page migration
attempts when the function terminates. The status array is only valid if
move_pages() completed successfullly.
Possible page states in status[]:
0..MAX_NUMNODES The page is now on the indicated node.
-ENOENT Page is not present
-EACCES Page is mapped by multiple processes and can only
be moved if MPOL_MF_MOVE_ALL is specified.
-EPERM The page has been mlocked by a process/driver and
cannot be moved.
-EBUSY Page is busy and cannot be moved. Try again later.
-EFAULT Invalid address (no VMA or zero page).
-ENOMEM Unable to allocate memory on target node.
-EIO Unable to write back page. The page must be written
back in order to move it since the page is dirty and the
filesystem does not provide a migration function that
would allow the moving of dirty pages.
-EINVAL A dirty page cannot be moved. The filesystem does not provide
a migration function and has no ability to write back pages.
The flags parameter indicates what types of pages to move:
MPOL_MF_MOVE Move pages that are only mapped by the process.
MPOL_MF_MOVE_ALL Also move pages that are mapped by multiple processes.
Requires sufficient capabilities.
Possible return codes from move_pages()
-ENOENT No pages found that would require moving. All pages
are either already on the target node, not present, had an
invalid address or could not be moved because they were
mapped by multiple processes.
-EINVAL Flags other than MPOL_MF_MOVE(_ALL) specified or an attempt
to migrate pages in a kernel thread.
-EPERM MPOL_MF_MOVE_ALL specified without sufficient priviledges.
or an attempt to move a process belonging to another user.
-EACCES One of the target nodes is not allowed by the current cpuset.
-ENODEV One of the target nodes is not online.
-ESRCH Process does not exist.
-E2BIG Too many pages to move.
-ENOMEM Not enough memory to allocate control array.
-EFAULT Parameters could not be accessed.
A test program for move_pages() may be found with the patches
on ftp.kernel.org:/pub/linux/kernel/people/christoph/pmig/patches-2.6.17-rc4-mm3
From: Christoph Lameter <[email protected]>
Detailed results for sys_move_pages()
Pass a pointer to an integer to get_new_page() that may be used to
indicate where the completion status of a migration operation should be
placed. This allows sys_move_pags() to report back exactly what happened to
each page.
Wish there would be a better way to do this. Looks a bit hacky.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Jes Sorensen <[email protected]>
Cc: KAMEZAWA Hiroyuki <[email protected]>
Cc: Lee Schermerhorn <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Michael Kerrisk <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Instead of passing a list of new pages, pass a function to allocate a new
page. This allows the correct placement of MPOL_INTERLEAVE pages during page
migration. It also further simplifies the callers of migrate pages.
migrate_pages() becomes similar to migrate_pages_to() so drop
migrate_pages_to(). The batching of new page allocations becomes unnecessary.
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Jes Sorensen <[email protected]>
Cc: KAMEZAWA Hiroyuki <[email protected]>
Cc: Lee Schermerhorn <[email protected]>
Cc: Andi Kleen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently migrate_pages() is mess with lots of goto. Extract two functions
from migrate_pages() and get rid of the gotos.
Plus we can just unconditionally set the locked bit on the new page since we
are the only one holding a reference. Locking is to stop others from
accessing the page once we establish references to the new page.
Remove the list_del from move_to_lru in order to have finer control over list
processing.
[[email protected]: add debug check]
Signed-off-by: Christoph Lameter <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Jes Sorensen <[email protected]>
Cc: KAMEZAWA Hiroyuki <[email protected]>
Cc: Lee Schermerhorn <[email protected]>
Cc: Andi Kleen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This implements the use of migration entries to preserve ptes of file backed
pages during migration. Processes can therefore be migrated back and forth
without loosing their connection to pagecache pages.
Note that we implement the migration entries only for linear mappings.
Nonlinear mappings still require the unmapping of the ptes for migration.
And another writepage() ugliness shows up. writepage() can drop the page
lock. Therefore we have to remove migration ptes before calling writepages()
in order to avoid having migration entries point to unlocked pages.
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Christoph Lameter