CONFIG_DEBUG_PER_CPU_MAPS may return NUMA_NO_NODE when an
early_cpu_to_node() mapping hasn't been initialized. In such a
case, it emits a warning and continues without an issue but
callers may try to use the return value to index into an array.
We can catch those errors and fail silently since a warning has
already been emitted. No current user of numa_add_cpu()
requires this error checking to avoid a crash, but it's better
to be proactive in case a future user happens to have a bug and
a user tries to diagnose it with CONFIG_DEBUG_PER_CPU_MAPS.
Reported-by: Jesper Juhl <jj@chaosbits.net>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Tejun Heo <tj@kernel.org>
LKML-Reference: <alpine.DEB.2.00.1102071407250.7812@chino.kir.corp.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Conflicts:
arch/x86/mm/numa_64.c
Merge reason: fix the conflict, update to latest -rc and pick up this
dependent fix from Yinghai:
e6d2e2b2b1e1: memblock: don't adjust size in memblock_find_base()
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Now that everything else is unified, NUMA initialization can be
unified too.
* numa_init_array() and init_cpu_to_node() are moved from
numa_64 to numa.
* numa_32::initmem_init() is updated to call numa_init_array()
and setup_arch() to call init_cpu_to_node() on 32bit too.
* x86_cpu_to_node_map is now initialized to NUMA_NO_NODE on
32bit too. This is safe now as numa_init_array() will initialize
it early during boot.
This makes NUMA mapping fully initialized before
setup_per_cpu_areas() on 32bit too and thus makes the first
percpu chunk which contains all the static variables and some of
dynamic area allocated with NUMA affinity correctly considered.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: yinghai@kernel.org
Cc: brgerst@gmail.com
Cc: gorcunov@gmail.com
Cc: shaohui.zheng@intel.com
Cc: rientjes@google.com
LKML-Reference: <1295789862-25482-17-git-send-email-tj@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
x86_32 has been managing node_to_cpumask_map explicitly from
map_cpu_to_node() and friends in a rather ugly way. With
previous changes, it's now possible to share the code with
64bit.
* When CONFIG_NUMA_EMU is disabled, numa_add/remove_cpu() are
implemented in numa.c and shared by 32 and 64bit. CONFIG_NUMA_EMU
versions still live in numa_64.c.
NUMA_EMU's dependency on 64bit is planned to be removed and the
above should go away together.
* identify_cpu() now calls numa_add_cpu() for 32bit too. This
makes the explicit mask management from map_cpu_to_node() unnecessary.
* The whole x86_32 specific map_cpu_to_node() chunk is no longer
necessary. Dropped.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: eric.dumazet@gmail.com
Cc: yinghai@kernel.org
Cc: brgerst@gmail.com
Cc: gorcunov@gmail.com
Cc: shaohui.zheng@intel.com
Cc: rientjes@google.com
LKML-Reference: <1295789862-25482-16-git-send-email-tj@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: David Rientjes <rientjes@google.com>
Cc: Shaohui Zheng <shaohui.zheng@intel.com>
Unlike 64bit, 32bit has been using its own cpu_to_node_map[] for
CPU -> NUMA node mapping. Replace it with early_percpu variable
x86_cpu_to_node_map and share the mapping code with 64bit.
* USE_PERCPU_NUMA_NODE_ID is now enabled for 32bit too.
* x86_cpu_to_node_map and numa_set/clear_node() are moved from
numa_64 to numa. For now, on 32bit, x86_cpu_to_node_map is initialized
with 0 instead of NUMA_NO_NODE. This is to avoid introducing unexpected
behavior change and will be updated once init path is unified.
* srat_detect_node() is now enabled for x86_32 too. It calls
numa_set_node() and initializes the mapping making explicit
cpu_to_node_map[] updates from map/unmap_cpu_to_node() unnecessary.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: eric.dumazet@gmail.com
Cc: yinghai@kernel.org
Cc: brgerst@gmail.com
Cc: gorcunov@gmail.com
Cc: penberg@kernel.org
Cc: shaohui.zheng@intel.com
Cc: rientjes@google.com
LKML-Reference: <1295789862-25482-15-git-send-email-tj@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: David Rientjes <rientjes@google.com>
The mapping between cpu/apicid and node is done via
apicid_to_node[] on 64bit and apicid_2_node[] +
apic->x86_32_numa_cpu_node() on 32bit. This difference makes it
difficult to further unify 32 and 64bit NUMA handling.
This patch unifies it by replacing both apicid_to_node[] and
apicid_2_node[] with __apicid_to_node[] array, which is accessed
by two accessors - set_apicid_to_node() and numa_cpu_node(). On
64bit, numa_cpu_node() always consults __apicid_to_node[]
directly while 32bit goes through apic->numa_cpu_node() method
to allow apic implementations to override it.
srat_detect_node() for amd cpus contains workaround for broken
NUMA configuration which assumes relationship between APIC ID,
HT node ID and NUMA topology. Leave it to access
__apicid_to_node[] directly as mapping through CPU might result
in undesirable behavior change. The comment is reformatted and
updated to note the ugliness.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: eric.dumazet@gmail.com
Cc: yinghai@kernel.org
Cc: brgerst@gmail.com
Cc: gorcunov@gmail.com
Cc: shaohui.zheng@intel.com
Cc: rientjes@google.com
LKML-Reference: <1295789862-25482-14-git-send-email-tj@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: David Rientjes <rientjes@google.com>
In order to be able to suppress the use of SRAT tables that
32-bit Linux can't deal with (in one case known to lead to a
non-bootable system, unless disabling ACPI altogether), move the
"numa=" option handling to common code.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Reviewed-by: Thomas Renninger <trenn@suse.de>
Cc: Tejun Heo <tj@kernel.org>
Cc: Thomas Renninger <trenn@suse.de>
LKML-Reference: <4D36B581020000780002D0FF@vpn.id2.novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
"x86, numa: Fake node-to-cpumask for NUMA emulation" broke the
build when CONFIG_DEBUG_PER_CPU_MAPS is set and CONFIG_NUMA_EMU
is not. This is because it is possible to map a cpu to multiple
nodes when NUMA emulation is used; the patch required a physical
node address table to find those nodes that was only available
when CONFIG_NUMA_EMU was enabled.
This extracts the common debug functionality to its own function
for CONFIG_DEBUG_PER_CPU_MAPS and uses it regardless of whether
CONFIG_NUMA_EMU is set or not.
NUMA emulation will now iterate over the set of possible nodes
for each cpu and call the new debug function whereas only the
cpu's node will be used without NUMA emulation enabled.
Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <alpine.DEB.2.00.1012301053590.12995@chino.kir.corp.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We need to access it right way, so make sure that it is mapped already.
Prepare to put page table on local node, and nodemap is used before that.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <4D1933C8.7060105@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
NUMA boot code assumes that physical node ids start at 0, but the DIMMs
that the apic id represents may not be reachable. If this is the case,
node 0 is never online and cpus never end up getting appropriately
assigned to a node. This causes the cpumask of all online nodes to be
empty and machines crash with kernel code assuming online nodes have
valid cpus.
The fix is to appropriately map all the address ranges for physical nodes
and ensure the cpu to node mapping function checks all possible nodes (up
to MAX_NUMNODES) instead of simply checking nodes 0-N, where N is the
number of physical nodes, for valid address ranges.
This requires no longer "compressing" the address ranges of nodes in the
physical node map from 0-N, but rather leave indices in physnodes[] to
represent the actual node id of the physical node. Accordingly, the
topology exported by both amd_get_nodes() and acpi_get_nodes() no longer
must return the number of nodes to iterate through; all such iterations
will now be to MAX_NUMNODES.
This change also passes the end address of system RAM (which may be
different from normal operation if mem= is specified on the command line)
before the physnodes[] array is populated. ACPI parsed nodes are
truncated to fit within the address range that respect the mem=
boundaries and even some physical nodes may become unreachable in such
cases.
When NUMA emulation does succeed, any apicid to node mapping that exists
for unreachable nodes are given default values so that proximity domains
can still be assigned. This is important for node_distance() to
function as desired.
Signed-off-by: David Rientjes <rientjes@google.com>
LKML-Reference: <alpine.DEB.2.00.1012221702090.3701@chino.kir.corp.google.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
It's necessary to fake the node-to-cpumask mapping so that an emulated
node ID returns a cpumask that includes all cpus that have affinity to
the memory it represents.
This is a little intrusive because it requires knowledge of the physical
topology of the system. setup_physnodes() gives us that information, but
since NUMA emulation ends up altering the physnodes array, it's necessary
to reset it before cpus are brought online.
Accordingly, the physnodes array is moved out of init.data and into
cpuinit.data since it will be needed on cpuup callbacks.
This works regardless of whether numa=fake is used on the command line,
or the setup of the fake node succeeds or fails. The physnodes array
always contains the physical topology of the machine if CONFIG_NUMA_EMU
is enabled and can be used to setup the correct node-to-cpumask mappings
in all cases since setup_physnodes() is called whenever the array needs
to be repopulated with the correct data.
To fake the actual mappings, numa_add_cpu() and numa_remove_cpu() are
rewritten for CONFIG_NUMA_EMU so that we first find the physical node to
which each cpu has local affinity, then iterate through all online nodes
to find the emulated nodes that have local affinity to that physical
node, and then finally map the cpu to each of those emulated nodes.
Signed-off-by: David Rientjes <rientjes@google.com>
LKML-Reference: <alpine.DEB.2.00.1012221701520.3701@chino.kir.corp.google.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
This patch adds the equivalent of acpi_fake_nodes() for AMD Northbridge
platforms. The goal is to fake the apicid-to-node mappings for NUMA
emulation so the physical topology of the machine is correctly maintained
within the kernel.
This change also fakes proximity domains for both ACPI and k8 code so the
physical distance between emulated nodes is maintained via
node_distance(). This exports the correct distances via
/sys/devices/system/node/.../distance based on the underlying topology.
A new helper function, fake_physnodes(), is introduced to correctly
invoke the correct NUMA code to fake these two mappings based on the
system type. If there is no underlying NUMA configuration, all cpus are
mapped to node 0 for local distance.
Since acpi_fake_nodes() is no longer called with CONFIG_ACPI_NUMA, it's
prototype can be removed from the header file for such a configuration.
Signed-off-by: David Rientjes <rientjes@google.com>
LKML-Reference: <alpine.DEB.2.00.1012221701360.3701@chino.kir.corp.google.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Not only the naming of the files was confusing, it was even more so for
the function and variable names.
Renamed the K8 NB and NUMA stuff that is also used on other AMD
platforms. This also renames the CONFIG_K8_NUMA option to
CONFIG_AMD_NUMA and the related file k8topology_64.c to
amdtopology_64.c. No functional changes intended.
Signed-off-by: Hans Rosenfeld <hans.rosenfeld@amd.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Xen can reserve huge amounts of memory for pre-ballooning, but that
still shows as RAM in the e820 memory map. early_node_mem could not
find range because of start/end adjusting, and will go through the
fallback path. However, the fallback patch is still using
memblock_x86_find_range_node(), and it is partially top-down because
it go through active_range entries from low to high.
Let's use memblock_find_in_range instead memblock_x86_find_range_node.
So get real top down in fallback path.
We may still need to make memblock_x86_find_range_node to do overall
top_down work.
Reported-by: Jeremy Fitzhardinge <jeremy@goop.org>
Tested-by: Jeremy Fitzhardinge <jeremy@goop.org>
Tested-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <4CC9A9C9.8020700@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
* 'core-memblock-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (74 commits)
x86-64: Only set max_pfn_mapped to 512 MiB if we enter via head_64.S
xen: Cope with unmapped pages when initializing kernel pagetable
memblock, bootmem: Round pfn properly for memory and reserved regions
memblock: Annotate memblock functions with __init_memblock
memblock: Allow memblock_init to be called early
memblock/arm: Fix memblock_region_is_memory() typo
x86, memblock: Remove __memblock_x86_find_in_range_size()
memblock: Fix wraparound in find_region()
x86-32, memblock: Make add_highpages honor early reserved ranges
x86, memblock: Fix crashkernel allocation
arm, memblock: Fix the sparsemem build
memblock: Fix section mismatch warnings
powerpc, memblock: Fix memblock API change fallout
memblock, microblaze: Fix memblock API change fallout
x86: Remove old bootmem code
x86, memblock: Use memblock_memory_size()/memblock_free_memory_size() to get correct dma_reserve
x86: Remove not used early_res code
x86, memblock: Replace e820_/_early string with memblock_
x86: Use memblock to replace early_res
x86, memblock: Use memblock_debug to control debug message print out
...
Fix up trivial conflicts in arch/x86/kernel/setup.c and kernel/Makefile
The file names are somehow misleading as the code is not specific to
AMD K8 CPUs anymore. The files accomodate code for other AMD CPU
northbridges as well.
Same is true for the config option which is valid for AMD CPU
northbridges in general and not specific to K8.
Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com>
LKML-Reference: <20100917160343.GD4958@loge.amd.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Requested by Ingo, Thomas and HPA.
The old bootmem code is no longer necessary, and the transition is
complete. Remove it.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
1.include linux/memblock.h directly. so later could reduce e820.h reference.
2 this patch is done by sed scripts mainly
-v2: use MEMBLOCK_ERROR instead of -1ULL or -1UL
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
1. replace find_e820_area with memblock_find_in_range
2. replace reserve_early with memblock_x86_reserve_range
3. replace free_early with memblock_x86_free_range.
4. NO_BOOTMEM will switch to use memblock too.
5. use _e820, _early wrap in the patch, in following patch, will
replace them all
6. because memblock_x86_free_range support partial free, we can remove some special care
7. Need to make sure that memblock_find_in_range() is called after memblock_x86_fill()
so adjust some calling later in setup.c::setup_arch()
-- corruption_check and mptable_update
-v2: Move reserve_brk() early
Before fill_memblock_area, to avoid overlap between brk and memblock_find_in_range()
that could happen We have more then 128 RAM entry in E820 tables, and
memblock_x86_fill() could use memblock_find_in_range() to find a new place for
memblock.memory.region array.
and We don't need to use extend_brk() after fill_memblock_area()
So move reserve_brk() early before fill_memblock_area().
-v3: Move find_smp_config early
To make sure memblock_find_in_range not find wrong place, if BIOS doesn't put mptable
in right place.
-v4: Treat RESERVED_KERN as RAM in memblock.memory. and they are already in
memblock.reserved already..
use __NOT_KEEP_MEMBLOCK to make sure memblock related code could be freed later.
-v5: Generic version __memblock_find_in_range() is going from high to low, and for 32bit
active_region for 32bit does include high pages
need to replace the limit with memblock.default_alloc_limit, aka get_max_mapped()
-v6: Use current_limit instead
-v7: check with MEMBLOCK_ERROR instead of -1ULL or -1L
-v8: Set memblock_can_resize early to handle EFI with more RAM entries
-v9: update after kmemleak changes in mainline
Suggested-by: David S. Miller <davem@davemloft.net>
Suggested-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
x86 arch specific changes to use generic numa_node_id() based on generic
percpu variable infrastructure. Back out x86's custom version of
numa_node_id()
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Nick Piggin <npiggin@suse.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Eric Whitney <eric.whitney@hp.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'x86-bootmem-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (30 commits)
early_res: Need to save the allocation name in drop_range_partial()
sparsemem: Fix compilation on PowerPC
early_res: Add free_early_partial()
x86: Fix non-bootmem compilation on PowerPC
core: Move early_res from arch/x86 to kernel/
x86: Add find_fw_memmap_area
Move round_up/down to kernel.h
x86: Make 32bit support NO_BOOTMEM
early_res: Enhance check_and_double_early_res
x86: Move back find_e820_area to e820.c
x86: Add find_early_area_size
x86: Separate early_res related code from e820.c
x86: Move bios page reserve early to head32/64.c
sparsemem: Put mem map for one node together.
sparsemem: Put usemap for one node together
x86: Make 64 bit use early_res instead of bootmem before slab
x86: Only call dma32_reserve_bootmem 64bit !CONFIG_NUMA
x86: Make early_node_mem get mem > 4 GB if possible
x86: Dynamically increase early_res array size
x86: Introduce max_early_res and early_res_count
...
Now that numa=fake=<size>[MG] is implemented, it is possible to remove
configurable node size support. The command-line parsing was already
broken (numa=fake=*128, for example, would not work) and since fake nodes
are now interleaved over physical nodes, this support is no longer
required.
Signed-off-by: David Rientjes <rientjes@google.com>
LKML-Reference: <alpine.DEB.2.00.1002151343080.26927@chino.kir.corp.google.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
numa=fake=N specifies the number of fake nodes, N, to partition the
system into and then allocates them by interleaving over physical nodes.
This requires knowledge of the system capacity when attempting to
allocate nodes of a certain size: either very large nodes to benchmark
scalability of code that operates on individual nodes, or very small
nodes to find bugs in the VM.
This patch introduces numa=fake=<size>[MG] so it is possible to specify
the size of each node to allocate. When used, nodes of the size
specified will be allocated and interleaved over the set of physical
nodes.
FAKE_NODE_MIN_SIZE was also moved to the more-appropriate
include/asm/numa_64.h.
Signed-off-by: David Rientjes <rientjes@google.com>
LKML-Reference: <alpine.DEB.2.00.1002151342510.26927@chino.kir.corp.google.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
numa=fake=N uses split_nodes_interleave() to partition the system into N
fake nodes. Each node size must have be a multiple of
FAKE_NODE_MIN_SIZE, otherwise it is possible to get strange alignments.
Because of this, the remaining memory from each node when rounded to
FAKE_NODE_MIN_SIZE is consolidated into a number of "big nodes" that are
bigger than the rest.
The calculation of the number of big nodes is incorrect since it is using
a logical AND operator when it should be multiplying the rounded-off
portion of each node with N.
Signed-off-by: David Rientjes <rientjes@google.com>
LKML-Reference: <alpine.DEB.2.00.1002151342230.26927@chino.kir.corp.google.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Finally we can use early_res to replace bootmem for x86_64 now.
Still can use CONFIG_NO_BOOTMEM to enable it or not.
-v2: fix 32bit compiling about MAX_DMA32_PFN
-v3: folded bug fix from LKML message below
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <4B747239.4070907@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Simplify setup_node_mem: don't use bootmem from other node, instead
just find_e820_area in early_node_mem.
This keeps the boundary between early_res and boot mem more clear, and
lets us only call early_res_to_bootmem() one time instead of for all
nodes.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <1265793639-15071-12-git-send-email-yinghai@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
CPU to node mapping is set via the following sequence:
1. numa_init_array(): Set up roundrobin from cpu to online node
2. init_cpu_to_node(): Set that according to apicid_to_node[]
according to srat only handle the node that
is online, and leave other cpu on node
without ram (aka not online) to still
roundrobin.
3. later call srat_detect_node for Intel/AMD, will use first_online
node or nearby node.
Problem is that setup_per_cpu_areas() is not called between 2 and 3,
the per_cpu for cpu on node with ram is on different node, and could
put that on node with two hops away.
So try to optimize this and add find_near_online_node() and call
init_cpu_to_node().
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <4B07A739.3030104@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In the NUMA bootmem setup failure path we freed nodedata_phys
incorrectly.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <4B07A739.3030104@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Add interleaved NUMA emulation support
This patch interleaves emulated nodes over the system's physical
nodes. This is required for interleave optimizations since
mempolicies, for example, operate by iterating over a nodemask and
act without knowledge of node distances. It can also be used for
testing memory latencies and NUMA bugs in the kernel.
There're a couple of ways to do this:
- divide the number of emulated nodes by the number of physical
nodes and allocate the result on each physical node, or
- allocate each successive emulated node on a different physical
node until all memory is exhausted.
The disadvantage of the first option is, depending on the asymmetry
in node capacities of each physical node, emulated nodes may
substantially differ in size on a particular physical node compared
to another.
The disadvantage of the second option is, also depending on the
asymmetry in node capacities of each physical node, there may be
more emulated nodes allocated on a single physical node as another.
This patch implements the second option; we sacrifice the
possibility that we may have slightly more emulated nodes on a
particular physical node compared to another in lieu of node size
asymmetry.
[ Note that "node capacity" of a physical node is not only a
function of its addressable range, but also is affected by
subtracting out the amount of reserved memory over that range.
NUMA emulation only deals with available, non-reserved memory
quantities. ]
We ensure there is at least a minimal amount of available memory
allocated to each node. We also make sure that at least this
amount of available memory is available in ZONE_DMA32 for any node
that includes both ZONE_DMA32 and ZONE_NORMAL.
This patch also cleans the emulation code up by no longer passing
the statically allocated struct bootnode array among the various
functions. This init.data array is not allocated on the stack since
it may be very large and thus it may be accessed at file scope.
The WARN_ON() for nodes_cover_memory() when faking proximity
domains is removed since it relies on successive nodes always
having greater start addresses than previous nodes; with
interleaving this is no longer always true.
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andreas Herrmann <andreas.herrmann3@amd.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Ankita Garg <ankita@in.ibm.com>
Cc: Len Brown <len.brown@intel.com>
LKML-Reference: <alpine.DEB.1.00.0909251519150.14754@chino.kir.corp.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This is the counterpart to "x86: export k8 physical topology" for
SRAT. It is not as invasive because the acpi code already seperates
node setup into detection and registration steps, with the
exception of registering e820 active regions in
acpi_numa_memory_affinity_init(). This is now moved to
acpi_scan_nodes() if NUMA emulation is disabled or deferred.
acpi_numa_init() now returns a value which specifies whether an
underlying SRAT was located. If so, that topology can be used by
the emulation code to interleave emulated nodes over physical nodes
or to register the nodes for ACPI.
acpi_get_nodes() may now be used to export the srat physical
topology of the machine for NUMA emulation.
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Andreas Herrmann <andreas.herrmann3@amd.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Ankita Garg <ankita@in.ibm.com>
Cc: Len Brown <len.brown@intel.com>
LKML-Reference: <alpine.DEB.1.00.0909251518580.14754@chino.kir.corp.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
To eventually interleave emulated nodes over physical nodes, we
need to know the physical topology of the machine without actually
registering it. This does the k8 node setup in two parts:
detection and registration. NUMA emulation can then used the
physical topology detected to setup the address ranges of emulated
nodes accordingly. If emulation isn't used, the k8 nodes are
registered as normal.
Two formals are added to the x86 NUMA setup functions: `acpi' and
`k8'. These represent whether ACPI or K8 NUMA has been detected;
both cannot be true at the same time. This specifies to the NUMA
emulation code whether an underlying physical NUMA topology exists
and which interface to use.
This patch deals solely with separating the k8 setup path into
Northbridge detection and registration steps and leaves the ACPI
changes for a subsequent patch. The `acpi' formal is added here,
however, to avoid touching all the header files again in the next
patch.
This approach also ensures emulated nodes will not span physical
nodes so the true memory latency is not misrepresented.
k8_get_nodes() may now be used to export the k8 physical topology
of the machine for NUMA emulation.
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Andreas Herrmann <andreas.herrmann3@amd.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Ankita Garg <ankita@in.ibm.com>
Cc: Len Brown <len.brown@intel.com>
LKML-Reference: <alpine.DEB.1.00.0909251518400.14754@chino.kir.corp.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Recently there were some changes to the meaning of node_possible_map,
and it is quite strange:
- the node without memory would be set in node_possible_map
- but some node with less NODE_MIN_SIZE will be kicked out of node_possible_map.
fix it by adding strict_setup_node_bootmem().
Also, remove unparse_node().
so result will be:
1. cpu_to_node() will return online node only (nearest one)
2. apicid_to_node() still returns the node that could be not online but is set
in node_possible_map.
3. node_possible_map will include nodes that mem on it are less NODE_MIN_SIZE
v2: after move_cpus_to_node change.
[ Impact: get node_possible_map right ]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Tested-by: Jack Steiner <steiner@sgi.com>
LKML-Reference: <4A0C49BE.6080800@kernel.org>
[ v3: various small cleanups and comment clarifications ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
after:
| commit b263295dbf
| Author: Christoph Lameter <clameter@sgi.com>
| Date: Wed Jan 30 13:30:47 2008 +0100
|
| x86: 64-bit, make sparsemem vmemmap the only memory model
we don't have MEMORY_HOTPLUG_RESERVE anymore.
Historically, x86-64 had an architecture-specific method for memory hotplug
whereby it scanned the SRAT for physical memory ranges that could be
potentially used for memory hot-add later. By reserving those ranges
without physical memory, the memmap would be allocated and left dormant
until needed. This depended on the DISCONTIG memory model which has been
removed so the code implementing HOTPLUG_RESERVE is now dead.
This patch removes the dead code used by MEMORY_HOTPLUG_RESERVE.
(Changelog authored by Mel.)
v2: updated changelog, and remove hotadd= in doc
[ Impact: remove dead code ]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Reviewed-by: Mel Gorman <mel@csn.ul.ie>
Workflow-found-OK-by: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <4A0C4910.7090508@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
64-bit UMA and NUMA versions of paging_init() are almost identical.
Therefore, merge the copy in mm/numa_64.c to mm/init_64.c to remove
duplicate code.
[ Impact: cleanup ]
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
LKML-Reference: <1241699741.17846.30.camel@penberg-laptop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Commit dc09855 ("x86/uv: fix init of memory-less nodes") causes a
two sockets system (where node-1 doesn't have RAM installed) to crash.
That commit makes node_possible include cpu nodes that do not have memory.
So check boundary in setup_node_bootmem().
[ Impact: fix boot crash on RAM-less NUMA node system ]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Jack Steiner <steiner@sgi.com>
LKML-Reference: <49EF89DF.9090404@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
We are removing cpumask_t in favour of struct cpumask: mainly as a
marker of what code is now CONFIG_CPUMASK_OFFSTACK-safe.
The only non-trivial change here is vector_allocation_domain():
explicitly clear the mask and set the first word, rather than using
assignment.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Impact: reduce kernel memory usage when CONFIG_CPUMASK_OFFSTACK=y
Straightforward conversion: done for 32 and 64 bit kernels.
node_to_cpumask_map is now a cpumask_var_t array.
64-bit used to be a dynamic cpumask_t array, and 32-bit used to be a
static cpumask_t array.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Impact: cleanup
We take the 64-bit code and use it on 32-bit as well. The new file
is called mm/numa.c.
In a minor cleanup, we use cpu_none_mask instead of declaring a local
cpu_mask_none.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
What's happening is that the assertion in mm/page_alloc.c:move_freepages()
is triggering:
BUG_ON(page_zone(start_page) != page_zone(end_page));
Once I knew this is what was happening, I added some annotations:
if (unlikely(page_zone(start_page) != page_zone(end_page))) {
printk(KERN_ERR "move_freepages: Bogus zones: "
"start_page[%p] end_page[%p] zone[%p]\n",
start_page, end_page, zone);
printk(KERN_ERR "move_freepages: "
"start_zone[%p] end_zone[%p]\n",
page_zone(start_page), page_zone(end_page));
printk(KERN_ERR "move_freepages: "
"start_pfn[0x%lx] end_pfn[0x%lx]\n",
page_to_pfn(start_page), page_to_pfn(end_page));
printk(KERN_ERR "move_freepages: "
"start_nid[%d] end_nid[%d]\n",
page_to_nid(start_page), page_to_nid(end_page));
...
And here's what I got:
move_freepages: Bogus zones: start_page[2207d0000] end_page[2207dffc0] zone[fffff8103effcb00]
move_freepages: start_zone[fffff8103effcb00] end_zone[fffff8003fffeb00]
move_freepages: start_pfn[0x81f600] end_pfn[0x81f7ff]
move_freepages: start_nid[1] end_nid[0]
My memory layout on this box is:
[ 0.000000] Zone PFN ranges:
[ 0.000000] Normal 0x00000000 -> 0x0081ff5d
[ 0.000000] Movable zone start PFN for each node
[ 0.000000] early_node_map[8] active PFN ranges
[ 0.000000] 0: 0x00000000 -> 0x00020000
[ 0.000000] 1: 0x00800000 -> 0x0081f7ff
[ 0.000000] 1: 0x0081f800 -> 0x0081fe50
[ 0.000000] 1: 0x0081fed1 -> 0x0081fed8
[ 0.000000] 1: 0x0081feda -> 0x0081fedb
[ 0.000000] 1: 0x0081fedd -> 0x0081fee5
[ 0.000000] 1: 0x0081fee7 -> 0x0081ff51
[ 0.000000] 1: 0x0081ff59 -> 0x0081ff5d
So it's a block move in that 0x81f600-->0x81f7ff region which triggers
the problem.
This patch:
Declaration of early_pfn_to_nid() is scattered over per-arch include
files, and it seems it's complicated to know when the declaration is used.
I think it makes fix-for-memmap-init not easy.
This patch moves all declaration to include/linux/mm.h
After this,
if !CONFIG_NODES_POPULATES_NODE_MAP && !CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
-> Use static definition in include/linux/mm.h
else if !CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
-> Use generic definition in mm/page_alloc.c
else
-> per-arch back end function will be called.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Tested-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reported-by: David Miller <davem@davemlloft.net>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: <stable@kernel.org> [2.6.25.x, 2.6.26.x, 2.6.27.x, 2.6.28.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Impact: bug fix
Don't use per_cpu_offset() to determine if it valid to access a
per-cpu variable for a given cpu number. It is not a valid assumption
on x86-64 anymore. Use cpu_possible() instead.
Signed-off-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: Code movement, no functional change.
Move the 64-bit NUMA code from setup_percpu.c to numa_64.c
Signed-off-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Impact: cleanup, futureproof
In fact, all cpumask ops will only be valid (in general) for bit
numbers < nr_cpu_ids. So use that instead of NR_CPUS in various
places.
This is always safe: no cpu number can be >= nr_cpu_ids, and
nr_cpu_ids is initialized to NR_CPUS at boot.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
There are a lot of places that define either a single bootmem descriptor or an
array of them. Use only one central array with MAX_NUMNODES items instead.
Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Kyle McMartin <kyle@parisc-linux.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are a couple of places where (P)Dprintk is used which is an old
compile time enabled printk wrapper. Convert it to the generic
pr_debug().
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Ingo Molnar