Raghu noticed an issue with excessive memory allocation on power with a
simple cgroup test, specifically, in mem_cgroup_css_alloc ->
for_each_node -> alloc_mem_cgroup_per_zone_info(), which ends up blowing
up the kmalloc-2048 slab (to the order of 200MB for 400 cgroup
directories).
The underlying issue is that NODES_SHIFT on power is 8 (256 NUMA nodes
possible), which defines node_possible_map, which in turn defines the
value of nr_node_ids in setup_nr_node_ids and the iteration of
for_each_node.
In practice, we never see a system with 256 NUMA nodes, and in fact, we
do not support node hotplug on power in the first place, so the nodes
that are online when we come up are the nodes that will be present for
the lifetime of this kernel. So let's, at least, drop the NUMA possible
map down to the online map at runtime. This is similar to what x86 does
in its initialization routines.
mem_cgroup_css_alloc should also be fixed to only iterate over
memory-populated nodes and handle hotplug, but that is a separate
change.
Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The goal behind this patch is to be able to write userland tests for the
VPHN parsing code.
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The first argument to vphn_unpack_associativity() is a const long *, but the
parsing code expects __be64 values actually. Let's move the endian fixing
down for consistency.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The number of values returned by the H_HOME_NODE_ASSOCIATIVITY h_call deserves
to be explicitly defined, for a better understanding of the code.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Some nice cleanups like removing bootmem, and removal of __get_cpu_var().
There is one patch to mm/gup.c. This is the generic GUP implementation, but is
only used by us and arm(64). We have an ack from Steve Capper, and although we
didn't get an ack from Andrew he told us to take the patch through the powerpc
tree.
There's one cxl patch. This is in drivers/misc, but Greg said he was happy for
us to manage fixes for it.
There is an infrastructure patch to support an IPMI driver for OPAL. That patch
also appears in Corey Minyard's IPMI tree, you may see a conflict there.
There is also an RTC driver for OPAL. We weren't able to get any response from
the RTC maintainer, Alessandro Zummo, so in the end we just merged the driver.
The usual batch of Freescale updates from Scott.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=jvyT
-----END PGP SIGNATURE-----
Merge tag 'powerpc-3.19-1' of git://git.kernel.org/pub/scm/linux/kernel/git/mpe/linux
Pull powerpc updates from Michael Ellerman:
"Some nice cleanups like removing bootmem, and removal of
__get_cpu_var().
There is one patch to mm/gup.c. This is the generic GUP
implementation, but is only used by us and arm(64). We have an ack
from Steve Capper, and although we didn't get an ack from Andrew he
told us to take the patch through the powerpc tree.
There's one cxl patch. This is in drivers/misc, but Greg said he was
happy for us to manage fixes for it.
There is an infrastructure patch to support an IPMI driver for OPAL.
There is also an RTC driver for OPAL. We weren't able to get any
response from the RTC maintainer, Alessandro Zummo, so in the end we
just merged the driver.
The usual batch of Freescale updates from Scott"
* tag 'powerpc-3.19-1' of git://git.kernel.org/pub/scm/linux/kernel/git/mpe/linux: (101 commits)
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
powerpc/book3s: Fix partial invalidation of TLBs in MCE code.
powerpc/mm: don't do tlbie for updatepp request with NO HPTE fault
powerpc/xmon: Cleanup the breakpoint flags
powerpc/xmon: Enable HW instruction breakpoint on POWER8
powerpc/mm/thp: Use tlbiel if possible
powerpc/mm/thp: Remove code duplication
powerpc/mm/hugetlb: Sanity check gigantic hugepage count
powerpc/oprofile: Disable pagefaults during user stack read
powerpc/mm: Check for matching hpte without taking hpte lock
powerpc: Drop useless warning in eeh_init()
powerpc/powernv: Cleanup unused MCE definitions/declarations.
powerpc/eeh: Dump PHB diag-data early
powerpc/eeh: Recover EEH error on ownership change for BCM5719
powerpc/eeh: Set EEH_PE_RESET on PE reset
powerpc/eeh: Refactor eeh_reset_pe()
powerpc: Remove more traces of bootmem
powerpc/pseries: Initialise nvram_pstore_info's buf_lock
cxl: Name interrupts in /proc/interrupt
cxl: Return error to PSL if IRQ demultiplexing fails & print clearer warning
...
The OF_RECONFIG notifier callback uses a different structure depending
on whether it is a node change or a property change. This is silly, and
not very safe. Rework the code to use the same data structure regardless
of the type of notifier.
Signed-off-by: Grant Likely <grant.likely@linaro.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Pantelis Antoniou <pantelis.antoniou@konsulko.com>
Cc: <linuxppc-dev@lists.ozlabs.org>
We did part of sparse initialisation in setup_arch and part in
initmem_init. Put them together.
Signed-off-by: Anton Blanchard <anton@samba.org>
Tested-by: Emil Medve <Emilian.Medve@Freescale.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
At the moment we transition from the memblock alloctor to the bootmem
allocator. Gitting rid of the bootmem allocator removes a bunch of
complicated code (most of which I owe the dubious honour of being
responsible for writing).
Signed-off-by: Anton Blanchard <anton@samba.org>
Tested-by: Emil Medve <Emilian.Medve@Freescale.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
We received a report of warning in kernel/sched/core.c where the sched
group was NULL on an LPAR after a topology update. This seems to occur
because after the topology update has moved the CPUs, cpu_to_node is
returning the old value still, which ends up breaking the consistency of
the NUMA topology in the per-cpu maps. Ensure that we update the per-cpu
fields when we re-map CPUs.
Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
There isn't any need to keep referring to update->cpu, as we've already
checked cpu == update->cpu at this point.
Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The associativity domain numbers are obtained from the hypervisor through
registers and written into memory by the guest: the packed array passed to
vphn_unpack_associativity() is then native-endian, unlike what was assumed
in the following commit:
commit b08a2a12e4
Author: Alistair Popple <alistair@popple.id.au>
Date: Wed Aug 7 02:01:44 2013 +1000
powerpc: Make NUMA device node code endian safe
This issue fills the topology with bogus data and makes it unusable. It may
lead to severe performance breakdowns.
We should ideally patch the vphn_unpack_associativity() function to do the
64-bit loads, but this requires some more brain storming.
In the meantime, let's go for a suboptimal and temporary bug fix: this patch
converts each 64-bit value of the packed array to big endian, as expected by
the current parsing code in vphn_unpack_associativity().
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
We have hit a few customer issues with the topology update code (VPHN
and PRRN). It would be nice to be able to debug the notifications coming
from the hypervisor in both cases to the LPAR, as well as to disable
responding to the notifications at boot-time, to narrow down the source
of the problems. Add a basic level of such functionality, similar to the
numa= command-line parameter. We already have a toggle in
/proc/powerpc/topology_updates that allows run-time enabling/disabling,
so the updates can be started at run-time if desired. But the bugs we've
run into have occured during boot or very shortly after coming to login,
and have resulted in a broken NUMA topology.
Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
proc_create can fail, we should check the return value and pass up the
failure.
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Freescale updates from Scott (27 commits):
"Highlights include DMA32 zone support (SATA, USB, etc now works on 64-bit
FSL kernels), MSI changes, 8xx optimizations and cleanup, t104x board
support, and PrPMC PCI enumeration."
this patches changes some error handling logics in numa_setup_cpu(),
when cpu node is not found, so:
if the cpu is possible, but not present, -1 is kept in numa_cpu_lookup_table,
so later, if the cpu is added, we could set correct numa information for it.
if the cpu is present, then we set the first online node to
numa_cpu_lookup_table instead of 0 ( in case 0 might not be an online node? )
Cc: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Cc: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
Acked-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
As Nish suggested, it makes more sense to init the numa node informatiion
for present cpus at boottime, which could also avoid WARN_ON(1) in
numa_setup_cpu().
With this change, we also need to change the smp_prepare_cpus() to set up
numa information only on present cpus.
For those possible, but not present cpus, their numa information
will be set up after they are started, as the original code did before commit
2fabf084b6.
Cc: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Cc: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
Acked-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Tested-by: Cyril Bur <cyril.bur@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
With commit 2fabf084b6 ("powerpc: reorder per-cpu NUMA information's
initialization"), during boottime, cpu_numa_callback() is called
earlier(before their online) for each cpu, and verify_cpu_node_mapping()
uses cpu_to_node() to check whether siblings are in the same node.
It skips the checking for siblings that are not online yet. So the only
check done here is for the bootcpu, which is online at that time. But
the per-cpu numa_node cpu_to_node() uses hasn't been set up yet (which
will be set up in smp_prepare_cpus()).
So I saw something like following reported:
[ 0.000000] CPU thread siblings 1/2/3 and 0 don't belong to the same
node!
As we don't actually do the checking during this early stage, so maybe
we could directly call numa_setup_cpu() in do_init_bootmem().
Cc: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Cc: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
Acked-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Commit 1c98025c6c "powerpc: Dynamic DMA
zone limits" updated how zones are created in paging_init(), but missed
the NUMA version of paging_init(). This was noticed via a linker
error, since dma_pfn_limit_to_zone() was, like the non-NUMA
paging_init(), limited by #ifndef CONFIG_NEED_MULTIPLE_NODES.
It turns out that the NUMA paging_init() was not actually doing
anything different from the standard paging_init(), other than a couple
debug prints, a couple 32-bit-only ifdef sections, and a call to
mark_nonram_nosave(). It's not clear whether mark_nonram_nosave() is
inherently wrong to do for NUMA, or just not useful on targets that
have NUMA, but for now I'm preserving the existing behavior.
Fixes: 1c98025c6c "powerpc: Dynamic DMA zone limits"
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Scott Wood <scottwood@freescale.com>
There is an issue currently where NUMA information is used on powerpc
(and possibly ia64) before it has been read from the device-tree, which
leads to large slab consumption with CONFIG_SLUB and memoryless nodes.
NUMA powerpc non-boot CPU's cpu_to_node/cpu_to_mem is only accurate
after start_secondary(), similar to ia64, which is invoked via
smp_init().
Commit 6ee0578b4d ("workqueue: mark init_workqueues() as
early_initcall()") made init_workqueues() be invoked via
do_pre_smp_initcalls(), which is obviously before the secondary
processors are online.
Additionally, the following commits changed init_workqueues() to use
cpu_to_node to determine the node to use for kthread_create_on_node:
bce903809a ("workqueue: add wq_numa_tbl_len and
wq_numa_possible_cpumask[]")
f3f90ad469 ("workqueue: determine NUMA node of workers accourding to
the allowed cpumask")
Therefore, when init_workqueues() runs, it sees all CPUs as being on
Node 0. On LPARs or KVM guests where Node 0 is memoryless, this leads to
a high number of slab deactivations
(http://www.spinics.net/lists/linux-mm/msg67489.html).
Fix this by initializing the powerpc-specific CPU<->node/local memory
node mapping as early as possible, which on powerpc is
do_init_bootmem(). Currently that function initializes the mapping for
the boot CPU, but we extend it to setup the mapping for all possible
CPUs. Then, in smp_prepare_cpus(), we can correspondingly set the
per-cpu values for all possible CPUs. That ensures that before the
early_initcalls run (and really as early as possible), the per-cpu NUMA
mapping is accurate.
While testing memoryless nodes on PowerKVM guests with a fix to the
workqueue logic to use cpu_to_mem() instead of cpu_to_node(), with a
guest topology of:
available: 2 nodes (0-1)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
node 0 size: 0 MB
node 0 free: 0 MB
node 1 cpus: 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
node 1 size: 16336 MB
node 1 free: 15329 MB
node distances:
node 0 1
0: 10 40
1: 40 10
the slab consumption decreases from
Slab: 932416 kB
SUnreclaim: 902336 kB
to
Slab: 395264 kB
SUnreclaim: 359424 kB
And we a corresponding increase in the slab efficiency from
slab mem objs slabs
used active active
------------------------------------------------------------
kmalloc-16384 337 MB 11.28% 100.00%
task_struct 288 MB 9.93% 100.00%
to
slab mem objs slabs
used active active
------------------------------------------------------------
kmalloc-16384 37 MB 100.00% 100.00%
task_struct 31 MB 100.00% 100.00%
Powerpc didn't support memoryless nodes until recently (64bb80d87f
"powerpc/numa: Enable CONFIG_HAVE_MEMORYLESS_NODES" and 8c27226119
"powerpc/numa: Enable USE_PERCPU_NUMA_NODE_ID"). Those commits also
helped improve memory consumption with these kind of environments.
Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Since v1:
Edited the comment according to Srivatsa's suggestion.
During the testing, we encounter below WARN followed by Oops:
WARNING: at kernel/sched/core.c:6218
...
NIP [c000000000101660] .build_sched_domains+0x11d0/0x1200
LR [c000000000101358] .build_sched_domains+0xec8/0x1200
PACATMSCRATCH [800000000000f032]
Call Trace:
[c00000001b103850] [c000000000101358] .build_sched_domains+0xec8/0x1200
[c00000001b1039a0] [c00000000010aad4] .partition_sched_domains+0x484/0x510
[c00000001b103aa0] [c00000000016d0a8] .rebuild_sched_domains+0x68/0xa0
[c00000001b103b30] [c00000000005cbf0] .topology_work_fn+0x10/0x30
...
Oops: Kernel access of bad area, sig: 11 [#1]
...
NIP [c00000000045c000] .__bitmap_weight+0x60/0xf0
LR [c00000000010132c] .build_sched_domains+0xe9c/0x1200
PACATMSCRATCH [8000000000029032]
Call Trace:
[c00000001b1037a0] [c000000000288ff4] .kmem_cache_alloc_node_trace+0x184/0x3a0
[c00000001b103850] [c00000000010132c] .build_sched_domains+0xe9c/0x1200
[c00000001b1039a0] [c00000000010aad4] .partition_sched_domains+0x484/0x510
[c00000001b103aa0] [c00000000016d0a8] .rebuild_sched_domains+0x68/0xa0
[c00000001b103b30] [c00000000005cbf0] .topology_work_fn+0x10/0x30
...
This was caused by that 'sd->groups == NULL' after building groups, which
was caused by the empty 'sd->span'.
The cpu's domain contained nothing because the cpu was assigned to a wrong
node, due to the following unfortunate sequence of events:
1. The hypervisor sent a topology update to the guest OS, to notify changes
to the cpu-node mapping. However, the update was actually redundant - i.e.,
the "new" mapping was exactly the same as the old one.
2. Due to this, the 'updated_cpus' mask turned out to be empty after exiting
the 'for-loop' in arch_update_cpu_topology().
3. So we ended up calling stop-machine() with an empty cpumask list, which made
stop-machine internally elect cpumask_first(cpu_online_mask), i.e., CPU0 as
the cpu to run the payload (the update_cpu_topology() function).
4. This causes update_cpu_topology() to be run by CPU0. And since 'updates'
is kzalloc()'ed inside arch_update_cpu_topology(), update_cpu_topology()
finds update->cpu as well as update->new_nid to be 0. In other words, we
end up assigning CPU0 (and eventually its siblings) to node 0, incorrectly.
Along with the following wrong updating, it causes the sched-domain rebuild
code to break and crash the system.
Fix this by skipping the topology update in cases where we find that
the topology has not actually changed in reality (ie., spurious updates).
CC: Benjamin Herrenschmidt <benh@kernel.crashing.org>
CC: Paul Mackerras <paulus@samba.org>
CC: Nathan Fontenot <nfont@linux.vnet.ibm.com>
CC: Stephen Rothwell <sfr@canb.auug.org.au>
CC: Andrew Morton <akpm@linux-foundation.org>
CC: Robert Jennings <rcj@linux.vnet.ibm.com>
CC: Jesse Larrew <jlarrew@linux.vnet.ibm.com>
CC: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>
CC: Alistair Popple <alistair@popple.id.au>
Suggested-by: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Pull powerpc updates from Ben Herrenschmidt:
"So here's my next branch for powerpc. A bit late as I was on vacation
last week. It's mostly the same stuff that was in next already, I
just added two patches today which are the wiring up of lockref for
powerpc, which for some reason fell through the cracks last time and
is trivial.
The highlights are, in addition to a bunch of bug fixes:
- Reworked Machine Check handling on kernels running without a
hypervisor (or acting as a hypervisor). Provides hooks to handle
some errors in real mode such as TLB errors, handle SLB errors,
etc...
- Support for retrieving memory error information from the service
processor on IBM servers running without a hypervisor and routing
them to the memory poison infrastructure.
- _PAGE_NUMA support on server processors
- 32-bit BookE relocatable kernel support
- FSL e6500 hardware tablewalk support
- A bunch of new/revived board support
- FSL e6500 deeper idle states and altivec powerdown support
You'll notice a generic mm change here, it has been acked by the
relevant authorities and is a pre-req for our _PAGE_NUMA support"
* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc: (121 commits)
powerpc: Implement arch_spin_is_locked() using arch_spin_value_unlocked()
powerpc: Add support for the optimised lockref implementation
powerpc/powernv: Call OPAL sync before kexec'ing
powerpc/eeh: Escalate error on non-existing PE
powerpc/eeh: Handle multiple EEH errors
powerpc: Fix transactional FP/VMX/VSX unavailable handlers
powerpc: Don't corrupt transactional state when using FP/VMX in kernel
powerpc: Reclaim two unused thread_info flag bits
powerpc: Fix races with irq_work
Move precessing of MCE queued event out from syscall exit path.
pseries/cpuidle: Remove redundant call to ppc64_runlatch_off() in cpu idle routines
powerpc: Make add_system_ram_resources() __init
powerpc: add SATA_MV to ppc64_defconfig
powerpc/powernv: Increase candidate fw image size
powerpc: Add debug checks to catch invalid cpu-to-node mappings
powerpc: Fix the setup of CPU-to-Node mappings during CPU online
powerpc/iommu: Don't detach device without IOMMU group
powerpc/eeh: Hotplug improvement
powerpc/eeh: Call opal_pci_reinit() on powernv for restoring config space
powerpc/eeh: Add restore_config operation
...
There have been some weird bugs in the past where the kernel tried to associate
threads of the same core to different NUMA nodes, and things went haywire after
that point (as expected).
But unfortunately, root-causing such issues have been quite challenging, due to
the lack of appropriate debug checks in the kernel. These bugs usually lead to
some odd soft-lockups in the scheduler's build-sched-domain code in the CPU
hotplug path, which makes it very hard to trace it back to the incorrect
cpu-to-node mappings.
So add appropriate debug checks to catch such invalid cpu-to-node mappings
as early as possible.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
On POWER platforms, the hypervisor can notify the guest kernel about dynamic
changes in the cpu-numa associativity (VPHN topology update). Hence the
cpu-to-node mappings that we got from the firmware during boot, may no longer
be valid after such updates. This is handled using the arch_update_cpu_topology()
hook in the scheduler, and the sched-domains are rebuilt according to the new
mappings.
But unfortunately, at the moment, CPU hotplug ignores these updated mappings
and instead queries the firmware for the cpu-to-numa relationships and uses
them during CPU online. So the kernel can end up assigning wrong NUMA nodes
to CPUs during subsequent CPU hotplug online operations (after booting).
Further, a particularly problematic scenario can result from this bug:
On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8)
threads per core. The switch to Single-Threaded (ST) mode is performed by
offlining all except the first CPU thread in each core. Switching back to
SMT mode involves onlining those other threads back, in each core.
Now consider this scenario:
1. During boot, the kernel gets the cpu-to-node mappings from the firmware
and assigns the CPUs to NUMA nodes appropriately, during CPU online.
2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and
communicates this update to the kernel. The kernel in turn updates its
cpu-to-node associations and rebuilds its sched domains. Everything is
fine so far.
3. Now, the user switches the machine from SMT to ST mode (say, by running
ppc64_cpu --smt=1). This involves offlining all except 1 thread in each
core.
4. The user then tries to switch back from ST to SMT mode (say, by running
ppc64_cpu --smt=4), and this involves onlining those threads back. Since
CPU hotplug ignores the new mappings, it queries the firmware and tries to
associate the newly onlined sibling threads to the old NUMA nodes. This
results in sibling threads within the same core getting associated with
different NUMA nodes, which is incorrect.
The scheduler's build-sched-domains code gets thoroughly confused with this
and enters an infinite loop and causes soft-lockups, as explained in detail
in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings).
So to fix this, use the numa_cpu_lookup_table to remember the updated
cpu-to-node mappings, and use them during CPU hotplug online operations.
Further, we also need to ensure that all threads in a core are assigned to a
common NUMA node, irrespective of whether all those threads were online during
the topology update. To achieve this, we take care not to use cpu_sibling_mask()
since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread()
and set up the mappings manually using the 'threads_per_core' value for that
particular platform. This helps us ensure that we don't hit this bug with any
combination of CPU hotplug and SMT mode switching.
Cc: stable@vger.kernel.org
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Use "pgdat_end_pfn()" instead of "pgdat->node_start_pfn +
pgdat->node_spanned_pages". Simplify the code, no functional change.
Signed-off-by: Xishi Qiu <qiuxishi@huawei.com>
Cc: James Hogan <james.hogan@imgtec.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Simple fixes for sparse warnings in this file.
Resolves:
arch/powerpc/mm/numa.c:198:24:
warning: Using plain integer as NULL pointer
arch/powerpc/mm/numa.c:1157:5:
warning: symbol 'hot_add_node_scn_to_nid' was not declared.
Should it be static?
arch/powerpc/mm/numa.c:1238:28:
warning: Using plain integer as NULL pointer
arch/powerpc/mm/numa.c:1538:6:
warning: symbol 'topology_schedule_update' was not declared.
Should it be static?
Signed-off-by: Robert C Jennings <rcj@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The device tree is big endian so make sure we byteswap on little
endian. We assume any pHyp calls also return big endian results in
memory.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Although the shared_proc field in the lppaca works today, it is
not architected. A shared processor partition will always have a non
zero yield_count so use that instead. Create a wrapper so users
don't have to know about the details.
In order for older kernels to continue to work on KVM we need
to set the shared_proc bit. While here, remove the ugly bitfield.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
When an associativity level change is found for one thread, the
siblings threads need to be updated as well. This is done today
for PRRN in stage_topology_update() but is missing for VPHN in
update_cpu_associativity_changes_mask(). This patch will correctly
update all thread siblings during a topology change.
Without this patch a topology update can result in a CPU in
init_sched_groups_power() getting stuck indefinitely in a loop.
This loop is built in build_sched_groups(). As a result of the thread
moving to a node separate from its siblings the struct sched_group will
have its next pointer set to point to itself rather than the sched_group
struct of the next thread. This happens because we have a domain without
the SD_OVERLAP flag, which is correct, and a topology that doesn't conform
with reality (threads on the same core assigned to different numa nodes).
When this list is traversed by init_sched_groups_power() it will reach
the thread's sched_group structure and loop indefinitely; the cpu will
be stuck at this point.
The bug was exposed when VPHN was enabled in commit b7abef0 (v3.9).
Cc: <stable@vger.kernel.org> [v3.9+]
Reported-by: Jan Stancek <jstancek@redhat.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The topology update code that updates the cpu node registration in sysfs
should not be called while in stop_machine(). The register/unregister
calls take a lock and may sleep.
This patch moves these calls outside of the call to stop_machine().
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
CC: <stable@vger.kernel.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The __cpuinit type of throwaway sections might have made sense
some time ago when RAM was more constrained, but now the savings
do not offset the cost and complications. For example, the fix in
commit 5e427ec2d0 ("x86: Fix bit corruption at CPU resume time")
is a good example of the nasty type of bugs that can be created
with improper use of the various __init prefixes.
After a discussion on LKML[1] it was decided that cpuinit should go
the way of devinit and be phased out. Once all the users are gone,
we can then finally remove the macros themselves from linux/init.h.
This removes all the powerpc uses of the __cpuinit macros. There
are no __CPUINIT users in assembly files in powerpc.
[1] https://lkml.org/lkml/2013/5/20/589
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Josh Boyer <jwboyer@gmail.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Pull powerpc update from Benjamin Herrenschmidt:
"The main highlights this time around are:
- A pile of addition POWER8 bits and nits, such as updated
performance counter support (Michael Ellerman), new branch history
buffer support (Anshuman Khandual), base support for the new PCI
host bridge when not using the hypervisor (Gavin Shan) and other
random related bits and fixes from various contributors.
- Some rework of our page table format by Aneesh Kumar which fixes a
thing or two and paves the way for THP support. THP itself will
not make it this time around however.
- More Freescale updates, including Altivec support on the new e6500
cores, new PCI controller support, and a pile of new boards support
and updates.
- The usual batch of trivial cleanups & fixes"
* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc: (156 commits)
powerpc: Fix build error for book3e
powerpc: Context switch the new EBB SPRs
powerpc: Turn on the EBB H/FSCR bits
powerpc: Replace CPU_FTR_BCTAR with CPU_FTR_ARCH_207S
powerpc: Setup BHRB instructions facility in HFSCR for POWER8
powerpc: Fix interrupt range check on debug exception
powerpc: Update tlbie/tlbiel as per ISA doc
powerpc: Print page size info during boot
powerpc: print both base and actual page size on hash failure
powerpc: Fix hpte_decode to use the correct decoding for page sizes
powerpc: Decode the pte-lp-encoding bits correctly.
powerpc: Use encode avpn where we need only avpn values
powerpc: Reduce PTE table memory wastage
powerpc: Move the pte free routines from common header
powerpc: Reduce the PTE_INDEX_SIZE
powerpc: Switch 16GB and 16MB explicit hugepages to a different page table format
powerpc: New hugepage directory format
powerpc: Don't truncate pgd_index wrongly
powerpc: Don't hard code the size of pte page
powerpc: Save DAR and DSISR in pt_regs on MCE
...
Correct build failure for powerpc/pseries builds with CONFIG_SMP not defined.
The function cpu_sibling_mask has no meaning (or definition) when CONFIG_SMP
is not defined. Additionally, the updating of NUMA affinity for a CPU in a UP
system doesn't really make sense.
This patch ifdef's out the code making the affinity updates for PRRN events to
fix the following build break.
arch/powerpc/mm/numa.c: In function ‘stage_topology_update’:
arch/powerpc/mm/numa.c:1535: error: implicit declaration of function ‘cpu_sibling_mask’
arch/powerpc/mm/numa.c:1535: warning: passing argument 3 of ‘cpumask_or’ makes pointer from integer without a cast
make[1]: *** [arch/powerpc/mm/numa.o] Error 1
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
After merging the cgroup tree, today's linux-next build (powerpc
ppc64_defconfig) failed like this:
arch/powerpc/mm/numa.c: In function 'arch_update_cpu_topology':
arch/powerpc/mm/numa.c:1465:2: error: implicit declaration of function 'kzalloc' [-Werror=implicit-function-declaration]
arch/powerpc/mm/numa.c:1465:10: error: assignment makes pointer from integer without a cast [-Werror]
arch/powerpc/mm/numa.c:1497:2: error: implicit declaration of function 'kfree' [-Werror=implicit-function-declaration]
Caused by commit 30c05350c3 ("powerpc/pseries: Use stop machine to
update cpu maps") from the powerpc tree interacting with (probably)
commit ff794dea52 ("cpuset: remove include of cgroup.h from cpuset.h")
from the cgroup tree. Removing includes from header files is fraught
with danger ...
The former should have added an include of linux/slab.h to
arch/powerpc/mm/numa.c.
I have added the following merge fix patch for today (but it should be
applied to the powerpc tree ASAP).
From: Stephen Rothwell <sfr@canb.auug.org.au>
Date: Mon, 29 Apr 2013 14:01:44 +1000
Subject: [PATCH] powerpc: numa.c: using kzalloc/kfree requires including
slab.h
fixes these build errors:
arch/powerpc/mm/numa.c: In function 'arch_update_cpu_topology':
arch/powerpc/mm/numa.c:1465:2: error: implicit declaration of function 'kzalloc' [-Werror=implicit-function-declaration]
arch/powerpc/mm/numa.c:1465:10: error: assignment makes pointer from integer without a cast [-Werror]
arch/powerpc/mm/numa.c:1497:2: error: implicit declaration of function 'kfree' [-Werror=implicit-function-declaration]
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Pull cgroup updates from Tejun Heo:
- Fixes and a lot of cleanups. Locking cleanup is finally complete.
cgroup_mutex is no longer exposed to individual controlelrs which
used to cause nasty deadlock issues. Li fixed and cleaned up quite a
bit including long standing ones like racy cgroup_path().
- device cgroup now supports proper hierarchy thanks to Aristeu.
- perf_event cgroup now supports proper hierarchy.
- A new mount option "__DEVEL__sane_behavior" is added. As indicated
by the name, this option is to be used for development only at this
point and generates a warning message when used. Unfortunately,
cgroup interface currently has too many brekages and inconsistencies
to implement a consistent and unified hierarchy on top. The new flag
is used to collect the behavior changes which are necessary to
implement consistent unified hierarchy. It's likely that this flag
won't be used verbatim when it becomes ready but will be enabled
implicitly along with unified hierarchy.
The option currently disables some of broken behaviors in cgroup core
and also .use_hierarchy switch in memcg (will be routed through -mm),
which can be used to make very unusual hierarchy where nesting is
partially honored. It will also be used to implement hierarchy
support for blk-throttle which would be impossible otherwise without
introducing a full separate set of control knobs.
This is essentially versioning of interface which isn't very nice but
at this point I can't see any other options which would allow keeping
the interface the same while moving towards hierarchy behavior which
is at least somewhat sane. The planned unified hierarchy is likely
to require some level of adaptation from userland anyway, so I think
it'd be best to take the chance and update the interface such that
it's supportable in the long term.
Maintaining the existing interface does complicate cgroup core but
shouldn't put too much strain on individual controllers and I think
it'd be manageable for the foreseeable future. Maybe we'll be able
to drop it in a decade.
Fix up conflicts (including a semantic one adding a new #include to ppc
that was uncovered by header the file changes) as per Tejun.
* 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup: (45 commits)
cpuset: fix compile warning when CONFIG_SMP=n
cpuset: fix cpu hotplug vs rebuild_sched_domains() race
cpuset: use rebuild_sched_domains() in cpuset_hotplug_workfn()
cgroup: restore the call to eventfd->poll()
cgroup: fix use-after-free when umounting cgroupfs
cgroup: fix broken file xattrs
devcg: remove parent_cgroup.
memcg: force use_hierarchy if sane_behavior
cgroup: remove cgrp->top_cgroup
cgroup: introduce sane_behavior mount option
move cgroupfs_root to include/linux/cgroup.h
cgroup: convert cgroupfs_root flag bits to masks and add CGRP_ prefix
cgroup: make cgroup_path() not print double slashes
Revert "cgroup: remove bind() method from cgroup_subsys."
perf: make perf_event cgroup hierarchical
cgroup: implement cgroup_is_descendant()
cgroup: make sure parent won't be destroyed before its children
cgroup: remove bind() method from cgroup_subsys.
devcg: remove broken_hierarchy tag
cgroup: remove cgroup_lock_is_held()
...
There are instances in which we do not want topology updates to occur.
In order to allow this a /proc interface (/proc/powerpc/topology_updates)
is introduced so that topology updates can be enabled and disabled.
This patch also adds a prrn_is_enabled() call so that PRRN events are
handled in the kernel only if topology updating is enabled.
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The new PRRN firmware feature provides a more convenient and event-driven
interface than VPHN for notifying Linux of changes to the NUMA affinity of
platform resources. However, for practical reasons, it may not be feasible
for some customers to update to the latest firmware. For these customers,
the VPHN feature supported on previous firmware versions may still be the
best option.
The VPHN feature was previously disabled due to races with the load
balancing code when accessing the NUMA cpu maps, but the new stop_machine()
approach protects the NUMA cpu maps from these concurrent accesses. It
should be safe to re-enable this feature now.
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The following patch adds vdso_getcpu_init(), which stores the NUMA node for
a cpu in SPRG3:
Commit 18ad51dd34 ("powerpc: Add VDSO version of getcpu") adds
vdso_getcpu_init(), which stores the NUMA node for a cpu in SPRG3.
This patch ensures that this information is also updated when the NUMA
affinity of a cpu changes.
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The new PRRN firmware feature allows CPU and memory resources to be
transparently reassigned across NUMA boundaries. When this happens, the
kernel must update the node maps to reflect the new affinity information.
Although the NUMA maps can be protected by locking primitives during the
update itself, this is insufficient to prevent concurrent accesses to these
structures. Since cpumask_of_node() hands out a pointer to these
structures, they can still be modified outside of the lock. Furthermore,
tracking down each usage of these pointers and adding locks would be quite
invasive and difficult to maintain.
The approach used is to make a list of affected cpus and call stop_machine
to have the update routine run on each of the affected cpus allowing them
to update themselves. Each cpu finds itself in the list of cpus and makes
the appropriate updates. We need to have each cpu do this for themselves to
handle calls to vdso_getcpu_init() added in a subsequent patch.
Situations like these are best handled using stop_machine(). Since the NUMA
affinity updates are exceptionally rare events, this approach has the
benefit of not adding any overhead while accessing the NUMA maps during
normal operation.
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Platform events such as partition migration or the new PRRN firmware
feature can cause the NUMA characteristics of a CPU to change, and these
changes will be reflected in the device tree nodes for the affected
CPUs.
This patch registers a handler for Open Firmware device tree updates
and reconfigures the CPU and node maps whenever the associativity
changes. Currently, this is accomplished by marking the affected CPUs in
the cpu_associativity_changes_mask and allowing
arch_update_cpu_topology() to retrieve the new associativity information
using hcall_vphn().
Protecting the NUMA cpu maps from concurrent access during an update
operation will be addressed in a subsequent patch in this series.
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Update the numa code to use the updated firmware_has_feature() when checking
for type 1 affinity.
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This function has always been marked as __cpuinit, but is only called
from functions marked as __init and references an __initdata variable.
So change its annotation to __init.
Fixes this build warning:
WARNING: arch/powerpc/mm/built-in.o(.cpuinit.text+0x86): Section mismatch in reference from the function .fake_numa_create_new_node() to the variable .init.data:cmdline
The function __cpuinit .fake_numa_create_new_node() references
a variable __initdata cmdline.
If cmdline is only used by .fake_numa_create_new_node then
annotate cmdline with a matching annotation.
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
The following commit breaks numa distance setup for old powerpc
systems that use form0 encoding in device tree.
commit 41eab6f88f
powerpc/numa: Use form 1 affinity to setup node distance
Device tree node /rtas/ibm,associativity-reference-points would
index into /cpus/PowerPCxxxx/ibm,associativity based on form0 or
form1 encoding detected by ibm,architecture-vec-5 property.
All modern systems use form1 and current kernel code is correct.
However, on older systems with form0 encoding, the numa distance
will get hard coded as LOCAL_DISTANCE for all nodes. This causes
task scheduling anomaly since scheduler will skip building numa
level domain (topmost domain with all cpus) if all numa distances
are same. (value of 'level' in sched_init_numa() will remain 0)
Prior to the above commit:
((from) == (to) ? LOCAL_DISTANCE : REMOTE_DISTANCE)
Restoring compatible behavior with this patch for old powerpc systems
with device tree where numa distance are encoded as form0.
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
This patch moves the definition of the of_drconf_cell struct to asm/prom.h
to make it available for all powerpc/pseries code.
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Acked-by: Rob Herring <rob.herring@calxeda.com>
Acked-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch_update_cpu_topology() should only return 1 when the topology has
actually changed, and should return 0 otherwise.
This patch fixes a potential bug where rebuild_sched_domains() would
reinitialize the sched domains even when the topology hasn't changed.
Signed-off-by: Jesse Larrew <jlarrew@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Nishanth Aravamudan
Greg Kurz
Linus Torvalds
Grant Likely
Anton Blanchard
Michael Ellerman
Li Zhong
Scott Wood
Andrey Utkin
Mike Qiu
Michael Wang
Joe Perches
Tang Chen
Srivatsa S. Bhat
Xishi Qiu
Robert Jennings
Alistair Popple
Nathan Fontenot
Paul Gortmaker
Stephen Rothwell
Cody P Schafer
Jesse Larrew
Vaidyanathan Srinivasan
Benjamin Herrenschmidt