There is a NUMA memory configuration issue in 2.6.24:
A 2-node machine of ours has got the following memory layout:
Node 0: 0 - 2 Gbytes
Node 0: 4 - 8 Gbytes
Node 1: 8 - 16 Gbytes
Node 0: 16 - 18 Gbytes
"efi_memmap_init()" merges the three last ranges into one.
"register_active_ranges()" is called as follows:
efi_memmap_walk(register_active_ranges, NULL);
i.e. once for the 4 - 18 Gbytes range. It picks up the node
number from the start address, and registers all the memory for
the node #0.
"register_active_ranges()" should be called as follows to
make sure there is no merged address range at its entry:
efi_memmap_walk(filter_memory, register_active_ranges);
"filter_memory()" is similar to "filter_rsvd_memory()",
but the reserved memory ranges are not filtered out.
Signed-off-by: Zoltan Menyhart <Zoltan.Menyhart@bull.net>
Signed-off-by: Tony Luck <tony.luck@intel.com>
show_mem() has no need to print the amount of free swap space manually because
show_free_areas() does this already and is called by the former.
The two outputs only differ in text formatting:
printk("Free swap = %lukB\n", ...);
printk("Free swap: %6ldkB\n", ...);
Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This attached patch significantly shrinks boot memory allocation on ia64.
It does this by not allocating per_cpu areas for cpus that can never
exist.
In the case where acpi does not have any numa node description of the
cpus, I defaulted to assigning the first 32 round-robin on the known
nodes.. For the !CONFIG_ACPI I used for_each_possible_cpu().
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
A simple fix. The existing pernodesize reservation is not taking into
account a second array of pg_data_t structures. This is normally not
important because the PAGE_ALIGN macro reserves adequate space.
I made the compute_pernodesize steps in the same order as the fill_pernode
steps to make the correlation more clear.
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This patchset adds a flags variable to reserve_bootmem() and uses the
BOOTMEM_EXCLUSIVE flag in crashkernel reservation code to detect collisions
between crashkernel area and already used memory.
This patch:
Change the reserve_bootmem() function to accept a new flag BOOTMEM_EXCLUSIVE.
If that flag is set, the function returns with -EBUSY if the memory already
has been reserved in the past. This is to avoid conflicts.
Because that code runs before SMP initialisation, there's no race condition
inside reserve_bootmem_core().
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: fix powerpc build]
Signed-off-by: Bernhard Walle <bwalle@suse.de>
Cc: <linux-arch@vger.kernel.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch set frees the restriction that makedumpfile users should install a
vmlinux file (including the debugging information) into each system.
makedumpfile command is the dump filtering feature for kdump. It creates a
small dumpfile by filtering unnecessary pages for the analysis. To
distinguish unnecessary pages, it needs a vmlinux file including the debugging
information. These days, the debugging package becomes a huge file, and it is
hard to install it into each system.
To solve the problem, kdump developers discussed it at lkml and kexec-ml. As
the result, we reached the conclusion that necessary information for dump
filtering (called "vmcoreinfo") should be embedded into the first kernel file
and it should be accessed through /proc/vmcore during the second kernel.
(http://www.uwsg.iu.edu/hypermail/linux/kernel/0707.0/1806.html)
Dan Aloni created the patch set for the above implementation.
(http://www.uwsg.iu.edu/hypermail/linux/kernel/0707.1/1053.html)
And I updated it for multi architectures and memory models.
(http://lists.infradead.org/pipermail/kexec/2007-August/000479.html)
Signed-off-by: Dan Aloni <da-x@monatomic.org>
Signed-off-by: Ken'ichi Ohmichi <oomichi@mxs.nes.nec.co.jp>
Signed-off-by: Bernhard Walle <bwalle@suse.de>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Equip IA64 sparsemem with a virtual memmap. This is similar to the existing
CONFIG_VIRTUAL_MEM_MAP functionality for DISCONTIGMEM. It uses a PAGE_SIZE
mapping.
This is provided as a minimally intrusive solution. We split the 128TB
VMALLOC area into two 64TB areas and use one for the virtual memmap.
This should replace CONFIG_VIRTUAL_MEM_MAP long term.
[apw@shadowen.org: convert to new helper based initialisation]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When dumping memory via sysrq-m it is possible to take a bogus NMI watchdog
or softlockup watchdog because the dump can take a long time on big memory
systems.
Occasionally tickle the watchdog when doing the dump.
Signed-off-by: Prarit Bhargava <prarit@redhat.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
IA64 is the origin of the quicklist implementation. So cut out the pieces
that are now in core code and modify the functions called.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This is to fix many section mismatches of code related to memory hotplug.
I checked compile with memory hotplug on/off on ia64 and x86-64 box.
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have seen bad_pte_print when testing crashdump on an SN machine in
recent 2.6.20 kernel. There are tons of bad pte print (pfn < max_low_pfn)
reports when the crash kernel boots up, all those reported bad pages
are inside initmem range; That is because if the crash kernel code and
data happens to be at the beginning of the 1st node. build_node_maps in
discontig.c will bypass reserved regions with filter_rsvd_memory. Since
min_low_pfn is calculated in build_node_map, so in this case, min_low_pfn
will be greater than kernel code and data.
Because pages inside initmem are freed and reused later, we saw
pfn_valid check fail on those pages.
I think this theoretically happen on a normal kernel. When I check
min_low_pfn and max_low_pfn calculation in contig.c and discontig.c.
I found more issues than this.
1. min_low_pfn and max_low_pfn calculation is inconsistent between
contig.c and discontig.c,
min_low_pfn is calculated as the first page number of boot memmap in
contig.c (Why? Though this may work at the most of the time, I don't
think it is the right logic). It is calculated as the lowest physical
memory page number bypass reserved regions in discontig.c.
max_low_pfn is calculated include reserved regions in contig.c. It is
calculated exclude reserved regions in discontig.c.
2. If kernel code and data region is happen to be at the begin or the
end of physical memory, when min_low_pfn and max_low_pfn calculation is
bypassed kernel code and data, pages in initmem will report bad.
3. initrd is also in reserved regions, if it is at the begin or at the
end of physical memory, kernel will refuse to reuse the memory. Because
the virt_addr_valid check in free_initrd_mem.
So it is better to fix and clean up those issues.
Calculate min_low_pfn and max_low_pfn in a consistent way.
Signed-off-by: Zou Nan hai <nanhai.zou@intel.com>
Acked-by: Jay Lan <jlan@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Make saved_max_pfn point to max_pfn of entire system.
Without this patch is so that vmcore is zero length on ia64. This is
because saved_max_pfn was wrongly being set to the max_pfn of the crash
kernel's address space, rather than the max_pfg on the physical memory of
the machine - the whole purpose of vmcore is to access physical memory that
is not part of the crash kernel's addresss space.
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Zou Nan hai <nanhai.zou@intel.com>
Sort-Of-Acked-By: Jay Lan <jlan@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
ZONE_DMA less operation for IA64 SGI platform
Disable ZONE_DMA for SGI SN2. All memory is addressable by all devices and we
do not need any special memory pool.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Eliminate arch specific memory_present call ia64 NUMA by utilizing
sparse_memory_present_with_active_regions.
Acked-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Bob Picco <bob.picco@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
While pursuing and unrelated issue with 64Mb granules I noticed a problem
related to inconsistent use of add_active_range. There doesn't appear any
reason to me why FLATMEM versus DISCONTIG_MEM should register memory to
add_active_range with different code. So I've changed the code into a
common implementation.
The other subtle issue fixed by this patch was calling add_active_range in
count_node_pages before granule aligning is performed. We were lucky with
16MB granules but not so with 64MB granules. count_node_pages has reserved
regions filtered out and as a consequence linked kernel text and data
aren't covered by calls to count_node_pages. So linked kernel regions
wasn't reported to add_active_regions. This resulted in free_initmem
causing numerous bad_page reports. This won't occur with this patch
because now all known memory regions are reported by
register_active_ranges.
Acked-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Bob Picco <bob.picco@hp.com>
Acked-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Set saved_max_pfn when discontig memory is in use.
This sets up saved_max_pfn when disctontig memory is in use.
This mirrors the code for contig memory.
This patch does not entirely solve the problem of making vmcore work,
however it does appear to be neccessary. Please consider applying.
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Arch-independent zone-sizing is using indices instead of symbolic names to
offset within an array related to zones (max_zone_pfns). The unintended
impact is that ZONE_DMA and ZONE_NORMAL is initialised on powerpc instead
of ZONE_DMA and ZONE_HIGHMEM when CONFIG_HIGHMEM is set. As a result, the
the machine fails to boot but will boot with CONFIG_HIGHMEM turned off.
The following patch properly initialises the max_zone_pfns[] array and uses
symbolic names instead of indices in each architecture using
arch-independent zone-sizing. Two users have successfully booted their
powerpcs with it (one an ibook G4). It has also been boot tested on x86,
x86_64, ppc64 and ia64. Please merge for 2.6.19-rc2.
Credit to Benjamin Herrenschmidt for identifying the bug and rolling the
first fix. Additional credit to Johannes Berg and Andreas Schwab for
reporting the problem and testing on powerpc.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
[IA64] minor reformatting to vmlinux.lds.S
[IA64] CMC/CPE: Reverse the order of fetching log and checking poll threshold
[IA64] PAL calls need physical mode, stacked
[IA64] ar.fpsr not set on MCA/INIT kernel entry
[IA64] printing support for MCA/INIT
[IA64] trim output of show_mem()
[IA64] show_mem() printk levels
[IA64] Make gp value point to Region 5 in mca handler
Revert "[IA64] Unwire set/get_robust_list"
[IA64] Implement futex primitives
[IA64-SGI] Do not request DMA memory for BTE
[IA64] Move perfmon tables from thread_struct to pfm_context
[IA64] Add interface so modules can discover whether multithreading is on.
[IA64] kprobes: fixup the pagefault exception caused by probehandlers
[IA64] kprobe opcode 16 bytes alignment on IA64
[IA64] esi-support
[IA64] Add "model name" to /proc/cpuinfo
Cut the number of lines of memory info output per node from five
to one line.
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Use the default sysrq printk level for printing show_mem() output both
for disconfig and contig versions. This is consistent with the printk
level used on other architectures (well ia32 at least).
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
contig.c (FLATMEM) requires the same optimization as in discontig.c for show_mem
when VIRTUAL_MEM_MAP is in use. Otherwise FLATMEM has softlockup timeouts.
This was boot tested for memory configuration: SPARSEMEM,
DISCONTIG+VIRTUAL_MEM_MAP, FLATMEM, FLATMEM+VIRTUAL_MEM_MAP and
FLATMEM+VIRTUAL_MEM_MAP with largest memory gap less than LARGE_GAP by
using boot parameter "mem=".
This was boot tested and "echo m >/proc/sysrq-trigger" output evaluated for
: FLATMEM, FLATMEM+VIRTUAL_MEM_MAP, DISCONTIGMEM+VIRTUAL_MEM_MAP and
SPARSEMEM.
Signed-off-by: Bob Picco <bob.picco@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Assure that vmem_map's high endpoint is MAX_ORDER aligned. Not doing so violates
the buddy allocator algorithm. Also anyone using mem=XXX on boot line and
not aligned to MAX_ORDER requires this patch in order to satisfy buddy
allocator. vmem_map always starts at pfn 0. The potentially large MAX_ORDER
on ia64 (due to hugetlbfs) requires that the end of vmem_map be aligned
to MAX_ORDER_NR_PAGES.
This was boot tested for: FLATMEM, FLATMEM+VIRTUAL_MEM_MAP,
DISCONTIGMEM+VIRTUAL_MEM_MAP and SPARSEMEM.
Signed-off-by: Bob Picco <bob.picco@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
I found a bug in memory hot-add code for ia64.
IA64's code has copies of pgdat's array on each node to reduce memory
access over crossing node. This array is used by NODE_DATA() macro. When
new node is hot-added, this pgdat's array should be updated and copied on
new node too.
However, I used for_each_online_node() in scatter_node_data() to copy
it. This meant its array is not copied on new node.
Because initialization of structures for new node was halfway,
so online_node_map couldn't be set at this time.
To copy arrays on new node, I changed it to check value of pgdat_list[]
which is source array of copies. I tested this patch with my Memory Hotadd
emulation on Tiger4. This patch is for 2.6.17-git20.
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This is a patch to allocate pgdat and per node data area for ia64. The size
for them can be calculated by compute_pernodesize().
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This is to refresh node_data[] array for ia64. As I mentioned previous
patches, ia64 has copies of information of pgdat address array on each node as
per node data.
At v2 of node_add, this function used stop_machine_run() to update them. (I
wished that they were copied safety as much as possible.) But, in this patch,
this arrays are just copied simply, and set node_online_map bit after
completion of pgdat initialization.
So, kernel must touch NODE_DATA() macro after checking node_online_map().
(Current code has already done it.) This is more simple way for just
hot-add.....
Note : It will be problem when hot-remove will occur,
because, even if online_map bit is set, kernel may
touch NODE_DATA() due to race condition. :-(
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This is a preparatory patch to make common code for updating of NODE_DATA() of
ia64 between boottime and hotplug.
Current code remembers pgdat address in mem_data which is used at just boot
time. But its information can be used at hotplug time by moving to global
value. The next patch uses this array.
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch modifies ia64's show_mem() to walk the vmem_map page tables and
rapidly skip forward across regions where the page tables are missing.
This prevents the pfn_valid() check from causing numerous unnecessary
page faults.
Without this patch on a 512 node 512 cpu system where every node has four
memory holes, the show_mem() call takes 1 hour 18 minutes. With this
patch, it takes less than 3 seconds.
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Because pgdat_list was linked to pgdat_list in *reverse* order, (By default)
some of arch has to sort it by themselves.
for_each_pgdat has gone..for_each_online_pgdat() uses node_online_map, which
doesn't need to be sorted.
This patch removes codes for sorting pgdat.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add init declaration to cpu initialization functions.
Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
here is the BSP removal support for IA64. Its pretty much the same thing that
was released a while back, but has your feedback incorporated.
- Removed CONFIG_BSP_REMOVE_WORKAROUND and associated cmdline param
- Fixed compile issue with sn2/zx1 due to a undefined fix_b0_for_bsp
- some formatting nits (whitespace etc)
This has been tested on tiger and long back by alex on hp systems as well.
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
The per-node data structures are allocated with strided offsets that are a
function of the node number. This prevents excessive cache-aliasing from
occurring.
On systems with a large number of nodes, the strided offset becomes
too large. This patch restricts the maximum offset to 32MB. This is far larger
than the size of any current L3 cache.
Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
The original memory less node allocation attempted to use NODEDATA_ALIGN for
alignment. The bootmem allocator only allows a power of two alignments. This
causes a BUG_ON for some nodes. For cpu only nodes just allocate with a
PERCPU_PAGE_SIZE alignment.
Some older firmware reports SLIT distances of 0xff and results in bestnode
not being computed. This is now treated correctly.
The failed allocation check was removed because it's redundant. The
bootmem allocator already makes this check.
This fix has been boot tested on 4 node machine which has 4 cpu only nodes
and 1 memory node. Thanks to Pete Keilty for reporting this and helping me
test it.
Signed-off-by: Bob Picco <bob.picco@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
pgdat->node_size_lock is basically only neeeded in one place in the normal
code: show_mem(), which is the arch-specific sysrq-m printing function.
Strictly speaking, the architectures not doing memory hotplug do no need this
locking in show_mem(). However, they are all included for completeness. This
should also make any future consolidation of all of the implementations a
little more straightforward.
This lock is also held in the sparsemem code during a memory removal, as
sections are invalidated. This is the place there pfn_valid() is made false
for a memory area that's being removed. The lock is only required when doing
pfn_valid() operations on memory which the user does not already have a
reference on the page, such as in show_mem().
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch is the minimal set of changes required by ia64 to use SPARSEMEM.
Signed-off-by: Bob Picco <bob.picco@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Jesse Barnes provided the original version of this patch months ago, but
other changes kept conflicting with it, so it got deferred. Greg Edwards
dug it out of obscurity just over a week ago, and almost immediately
another conflicting patch appeared (Bob Picco's memory-less nodes).
I've resolved the conflicts and got it running again. CONFIG_SGI_TIOCX
is set to "y" in defconfig, which causes a Tiger to not boot (oops in
tiocx_init). But that can be resolved later ... get this in now before it
gets stale again.
Signed-off-by: Tony Luck <tony.luck@intel.com>
I reworked how nodes with only CPUs are treated. The patch below seems
simpler to me and has eliminated the complicated routine
reassign_cpu_only_nodes. There isn't any longer the requirement
to modify ACPI NUMA information which was in large part the
complexity introduced in reassign_cpu_only_nodes.
This patch will produce a different number of nodes. For example,
reassign_cpu_only_nodes would reduce two CPUonly nodes and one memory node
configuration to one memory+CPUs node configuration. This patch
doesn't change the number of nodes which means the user will see three. Two
nodes without memory and one node with all the memory.
While doing this patch, I noticed that early_nr_phys_cpus_node isn't serving
any useful purpose. It is called once in find_pernode_space but the value
isn't used to computer pernode space.
Signed-off-by: bob.picco <bob.picco@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This patch effectively eliminates direct use of pgdat->node_mem_map outside
of the DISCONTIG code. On a flat memory system, these fields aren't
currently used, neither are they on a sparsemem system.
There was also a node_mem_map(nid) macro on many architectures. Its use
along with the use of ->node_mem_map itself was not consistent. It has
been removed in favor of two new, more explicit, arch-independent macros:
pgdat_page_nr(pgdat, pagenr)
nid_page_nr(nid, pagenr)
I called them "pgdat" and "nid" because we overload the term "node" to mean
"NUMA node", "DISCONTIG node" or "pg_data_t" in very confusing ways. I
believe the newer names are much clearer.
These macros can be overridden in the sparsemem case with a theoretically
slower operation using node_start_pfn and pfn_to_page(), instead. We could
make this the only behavior if people want, but I don't want to change too
much at once. One thing at a time.
This patch removes more code than it adds.
Compile tested on alpha, alpha discontig, arm, arm-discontig, i386, i386
generic, NUMAQ, Summit, ppc64, ppc64 discontig, and x86_64. Full list
here: http://sr71.net/patches/2.6.12/2.6.12-rc1-mhp2/configs/
Boot tested on NUMAQ, x86 SMP and ppc64 power4/5 LPARs.
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Martin J. Bligh <mbligh@aracnet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch introduces using the quicklists for pgd, pmd, and pte levels
by combining the alloc and free functions into a common set of routines.
This greatly simplifies the reading of this header file.
This patch is simple but necessary for large numa configurations.
It simply ensures that only pages from the local node are added to a
cpus quicklist. This prevents the trapping of pages on a remote nodes
quicklist by starting a process, touching a large number of pages to
fill pmd and pte entries, migrating to another node, and then unmapping
or exiting. With those conditions, the pages get trapped and if the
machine has more than 100 nodes of the same size, the calculation of
the pgtable high water mark will be larger than any single node so page
table cache flushing will never occur.
I ran lmbench lat_proc fork and lat_proc exec on a zx1 with and without
this patch and did not notice any change.
On an sn2 machine, there was a slight improvement which is possibly
due to pages from other nodes trapped on the test node before starting
the run. I did not investigate further.
This patch shrinks the quicklist based upon free memory on the node
instead of the high/low water marks. I have written it to enable
preemption periodically and recalculate the amount to shrink every time
we have freed enough pages that the quicklist size should have grown.
I rescan the nodes zones each pass because other processess may be
draining node memory at the same time as we are adding.
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Tony Luck