Currently, all newly added memory blocks remain in 'offline' state
unless someone onlines them, some linux distributions carry special udev
rules like:
SUBSYSTEM=="memory", ACTION=="add", ATTR{state}=="offline", ATTR{state}="online"
to make this happen automatically. This is not a great solution for
virtual machines where memory hotplug is being used to address high
memory pressure situations as such onlining is slow and a userspace
process doing this (udev) has a chance of being killed by the OOM killer
as it will probably require to allocate some memory.
Introduce default policy for the newly added memory blocks in
/sys/devices/system/memory/auto_online_blocks file with two possible
values: "offline" which preserves the current behavior and "online"
which causes all newly added memory blocks to go online as soon as
they're added. The default is "offline".
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Kay Sievers <kay@vrfy.org>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Prevent userspace from trying and failing to online ZONE_DEVICE pages
which are meant to never be onlined.
For example on platforms with a udev rule like the following:
SUBSYSTEM=="memory", ACTION=="add", ATTR{state}=="offline", ATTR{state}="online"
...will generate futile attempts to online the ZONE_DEVICE sections.
Example kernel messages:
Built 1 zonelists in Node order, mobility grouping on. Total pages: 1004747
Policy zone: Normal
online_pages [mem 0x248000000-0x24fffffff] failed
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix a bug where a kernel warning is triggered when performing a memory
hotplug on ppc64. This warning may also occur on any architecture that
uses the memory_probe_store interface.
WARNING: at drivers/base/memory.c:200
CPU: 9 PID: 13042 Comm: systemd-udevd Not tainted 4.4.0-rc4-00113-g0bd0f1e-dirty #7
NIP [c00000000055e034] pages_correctly_reserved+0x134/0x1b0
LR [c00000000055e7f8] memory_subsys_online+0x68/0x140
Call Trace:
memory_subsys_online+0x68/0x140
device_online+0xb4/0x120
store_mem_state+0xb0/0x180
dev_attr_store+0x34/0x60
sysfs_kf_write+0x64/0xa0
kernfs_fop_write+0x17c/0x1e0
__vfs_write+0x40/0x160
vfs_write+0xb8/0x200
SyS_write+0x60/0x110
system_call+0x38/0xd0
The warning is triggered because there is a udev rule that automatically
tries to online memory after it has been added. The udev rule varies
from distro to distro, but will generally look something like:
SUBSYSTEM=="memory", ACTION=="add", ATTR{state}=="offline", ATTR{state}="online"
On any architecture that uses memory_probe_store to reserve memory, the
udev rule will be triggered after the first section of the block is
reserved and will subsequently attempt to online the entire block,
interrupting the memory reservation process and causing the warning.
This patch modifies memory_probe_store to add a block of memory with a
single call to add_memory as opposed to looping through and adding each
section individually. A single call to add_memory is protected by the
mem_hotplug mutex which will prevent the udev rule from onlining memory
until the reservation of the entire block is complete.
Signed-off-by: John Allen <jallen@linux.vnet.ibm.com>
Acked-by: Dave Hansen <dave.hansen@intel.com>
Cc: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Right now, section_count is calculated in add_memory_block(). However,
init_memory_block() increments section_count as well, which, at first,
seems like it would lead to an off-by-one error. There is no harm done
because add_memory_block() immediately overwrites the
mem->section_count, but it is messy.
This commit moves the increment out of the common init_memory_block()
(called by both add_memory_block() and register_new_memory()) and adds
it to register_new_memory().
Signed-off-by: Seth Jennings <sjennings@variantweb.net>
Cc: Andrew Banman <abanman@sgi.com>
Cc: Daniel J Blueman <daniel@numascale.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Greg KH <greg@kroah.com>
Cc: Russ Anderson <rja@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit bdee237c03 ("x86: mm: Use 2GB memory block size on large-memory
x86-64 systems") and 982792c782 ("x86, mm: probe memory block size for
generic x86 64bit") introduced large block sizes for x86. This made it
possible to have multiple sections per memory block where previously,
there was a only every one section per block.
Since blocks consist of contiguous ranges of section, there can be holes
in the blocks where sections are not present. If one attempts to
offline such a block, a crash occurs since the code is not designed to
deal with this.
This patch is a quick fix to gaurd against the crash by not allowing
blocks with non-present sections to be offlined.
Addresses https://bugzilla.kernel.org/show_bug.cgi?id=107781
Signed-off-by: Seth Jennings <sjennings@variantweb.net>
Reported-by: Andrew Banman <abanman@sgi.com>
Cc: Daniel J Blueman <daniel@numascale.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Greg KH <greg@kroah.com>
Cc: Russ Anderson <rja@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There's a deadlock when concurrently hot-adding memory through the probe
interface and switching a memory block from offline to online.
When hot-adding memory via the probe interface, add_memory() first takes
mem_hotplug_begin() and then device_lock() is later taken when registering
the newly initialized memory block. This creates a lock dependency of (1)
mem_hotplug.lock (2) dev->mutex.
When switching a memory block from offline to online, dev->mutex is first
grabbed in device_online() when the write(2) transitions an existing
memory block from offline to online, and then online_pages() will take
mem_hotplug_begin().
This creates a lock inversion between mem_hotplug.lock and dev->mutex.
Vitaly reports that this deadlock can happen when kworker handling a probe
event races with systemd-udevd switching a memory block's state.
This patch requires the state transition to take mem_hotplug_begin()
before dev->mutex. Hot-adding memory via the probe interface creates a
memory block while holding mem_hotplug_begin(), there is no way to take
dev->mutex first in this case.
online_pages() and offline_pages() are only called when transitioning
memory block state. We now require that mem_hotplug_begin() is taken
before calling them -- this requires exporting the mem_hotplug_begin() and
mem_hotplug_done() to generic code. In all hot-add and hot-remove cases,
mem_hotplug_begin() is done prior to device_online(). This is all that is
needed to avoid the deadlock.
Signed-off-by: David Rientjes <rientjes@google.com>
Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zhang Zhen <zhenzhang.zhang@huawei.com>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Cc: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently memory-hotplug has two limits:
1. If the memory block is in ZONE_NORMAL, you can change it to
ZONE_MOVABLE, but this memory block must be adjacent to ZONE_MOVABLE.
2. If the memory block is in ZONE_MOVABLE, you can change it to
ZONE_NORMAL, but this memory block must be adjacent to ZONE_NORMAL.
With this patch, we can easy to know a memory block can be onlined to
which zone, and don't need to know the above two limits.
Updated the related Documentation.
[akpm@linux-foundation.org: use conventional comment layout]
[akpm@linux-foundation.org: fix build with CONFIG_MEMORY_HOTREMOVE=n]
[akpm@linux-foundation.org: remove unused local zone_prev]
Signed-off-by: Zhang Zhen <zhenzhang.zhang@huawei.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In store_mem_state(), we have:
...
334 else if (!strncmp(buf, "offline", min_t(int, count, 7)))
335 online_type = -1;
...
355 case -1:
356 ret = device_offline(&mem->dev);
357 break;
...
Here, "offline" is hard coded as -1.
This patch does the following renaming:
ONLINE_KEEP -> MMOP_ONLINE_KEEP
ONLINE_KERNEL -> MMOP_ONLINE_KERNEL
ONLINE_MOVABLE -> MMOP_ONLINE_MOVABLE
and introduces MMOP_OFFLINE = -1 to avoid hard coding.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Hu Tao <hutao@cn.fujitsu.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We use the following command to online a memory_block:
echo online|online_kernel|online_movable > /sys/devices/system/memory/memoryXXX/state
But, if we do the following:
echo online_fhsjkghfkd > /sys/devices/system/memory/memoryXXX/state
the block will also be onlined.
This is because the following code in store_mem_state() does not compare
the whole string, but only the prefix of the string.
store_mem_state()
{
......
328 if (!strncmp(buf, "online_kernel", min_t(int, count, 13)))
Here, only compare the first 13 letters of the string. If we give "online_kernelXXXXXX",
it will be recognized as online_kernel, which is incorrect.
329 online_type = ONLINE_KERNEL;
330 else if (!strncmp(buf, "online_movable", min_t(int, count, 14)))
We have the same problem here,
331 online_type = ONLINE_MOVABLE;
332 else if (!strncmp(buf, "online", min_t(int, count, 6)))
here,
(Here is more problematic. If we give online_movalbe, which is a typo
of online_movable, it will be recognized as online without noticing the
author.)
333 online_type = ONLINE_KEEP;
334 else if (!strncmp(buf, "offline", min_t(int, count, 7)))
and here.
335 online_type = -1;
336 else {
337 ret = -EINVAL;
338 goto err;
339 }
......
}
This patch fixes this problem by using sysfs_streq() to compare the
whole string.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Reported-by: Hu Tao <hutao@cn.fujitsu.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Gu Zheng <guz.fnst@cn.fujitsu.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Seems we all agree that information about SECTION, e.g. section size,
sections per memory block should be kept as kernel internals, and not
exposed to userspace.
This patch updates Documentation/memory-hotplug.txt to refer to memory
blocks instead of memory sections where appropriate and added a
paragraph to explain that memory blocks are made of memory sections.
The documentation update is mostly provided by Nathan.
Also, as end_phys_index in code is actually not the end section id, but
the end memory block id, which should always be the same as phys_index.
So it is removed here.
Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
Reviewed-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When inserting a wrong value to /sys/devices/system/memory/memoryX/state file,
following messages are shown. And device_hotplug_lock is never released.
================================================
[ BUG: lock held when returning to user space! ]
3.12.0-rc4-debug+ #3 Tainted: G W
------------------------------------------------
bash/6442 is leaving the kernel with locks still held!
1 lock held by bash/6442:
#0: (device_hotplug_lock){+.+.+.}, at: [<ffffffff8146cbb5>] lock_device_hotplug_sysfs+0x15/0x50
This issue was introdued by commit fa2be40 (drivers: base: use standard
device online/offline for state change).
This patch releases device_hotplug_lcok when store_mem_state returns EINVAL.
Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Reviewed-by: Toshi Kani <toshi.kani@hp.com>
CC: Seth Jennings <sjenning@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Pull ACPI and power management updates from Rafael Wysocki:
1) ACPI-based PCI hotplug (ACPIPHP) subsystem rework and introduction
of Intel Thunderbolt support on systems that use ACPI for signalling
Thunderbolt hotplug events. This also should make ACPIPHP work in
some cases in which it was known to have problems. From
Rafael J Wysocki, Mika Westerberg and Kirill A Shutemov.
2) ACPI core code cleanups and dock station support cleanups from
Jiang Liu and Rafael J Wysocki.
3) Fixes for locking problems related to ACPI device hotplug from
Rafael J Wysocki.
4) ACPICA update to version 20130725 includig fixes, cleanups, support
for more than 256 GPEs per GPE block and a change to make the ACPI
PM Timer optional (we've seen systems without the PM Timer in the
field already). One of the fixes, related to the DeRefOf operator,
is necessary to prevent some Windows 8 oriented AML from causing
problems to happen. From Bob Moore, Lv Zheng, and Jung-uk Kim.
5) Removal of the old and long deprecated /proc/acpi/event interface
and related driver changes from Thomas Renninger.
6) ACPI and Xen changes to make the reduced hardware sleep work with
the latter from Ben Guthro.
7) ACPI video driver cleanups and a blacklist of systems that should
not tell the BIOS that they are compatible with Windows 8 (or ACPI
backlight and possibly other things will not work on them). From
Felipe Contreras.
8) Assorted ACPI fixes and cleanups from Aaron Lu, Hanjun Guo,
Kuppuswamy Sathyanarayanan, Lan Tianyu, Sachin Kamat, Tang Chen,
Toshi Kani, and Wei Yongjun.
9) cpufreq ondemand governor target frequency selection change to
reduce oscillations between min and max frequencies (essentially,
it causes the governor to choose target frequencies proportional
to load) from Stratos Karafotis.
10) cpufreq fixes allowing sysfs attributes file permissions to be
preserved over suspend/resume cycles Srivatsa S Bhat.
11) Removal of Device Tree parsing for CPU device nodes from multiple
cpufreq drivers that required some changes related to
of_get_cpu_node() to be made in a few architectures and in the
driver core. From Sudeep KarkadaNagesha.
12) cpufreq core fixes and cleanups related to mutual exclusion and
driver module references from Viresh Kumar, Lukasz Majewski and
Rafael J Wysocki.
13) Assorted cpufreq fixes and cleanups from Amit Daniel Kachhap,
Bartlomiej Zolnierkiewicz, Hanjun Guo, Jingoo Han, Joseph Lo,
Julia Lawall, Li Zhong, Mark Brown, Sascha Hauer, Stephen Boyd,
Stratos Karafotis, and Viresh Kumar.
14) Fixes to prevent race conditions in coupled cpuidle from happening
from Colin Cross.
15) cpuidle core fixes and cleanups from Daniel Lezcano and
Tuukka Tikkanen.
16) Assorted cpuidle fixes and cleanups from Daniel Lezcano,
Geert Uytterhoeven, Jingoo Han, Julia Lawall, Linus Walleij,
and Sahara.
17) System sleep tracing changes from Todd E Brandt and Shuah Khan.
18) PNP subsystem conversion to using struct dev_pm_ops for power
management from Shuah Khan.
* tag 'pm+acpi-3.12-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (217 commits)
cpufreq: Don't use smp_processor_id() in preemptible context
cpuidle: coupled: fix race condition between pokes and safe state
cpuidle: coupled: abort idle if pokes are pending
cpuidle: coupled: disable interrupts after entering safe state
ACPI / hotplug: Remove containers synchronously
driver core / ACPI: Avoid device hot remove locking issues
cpufreq: governor: Fix typos in comments
cpufreq: governors: Remove duplicate check of target freq in supported range
cpufreq: Fix timer/workqueue corruption due to double queueing
ACPI / EC: Add ASUSTEK L4R to quirk list in order to validate ECDT
ACPI / thermal: Add check of "_TZD" availability and evaluating result
cpufreq: imx6q: Fix clock enable balance
ACPI: blacklist win8 OSI for buggy laptops
cpufreq: tegra: fix the wrong clock name
cpuidle: Change struct menu_device field types
cpuidle: Add a comment warning about possible overflow
cpuidle: Fix variable domains in get_typical_interval()
cpuidle: Fix menu_device->intervals type
cpuidle: CodingStyle: Break up multiple assignments on single line
cpuidle: Check called function parameter in get_typical_interval()
...
Pull driver core patches from Greg KH:
"Here's the big driver core pull request for 3.12-rc1.
Lots of tiny changes here fixing up the way sysfs attributes are
created, to try to make drivers simpler, and fix a whole class race
conditions with creations of device attributes after the device was
announced to userspace.
All the various pieces are acked by the different subsystem
maintainers"
* tag 'driver-core-3.12-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core: (119 commits)
firmware loader: fix pending_fw_head list corruption
drivers/base/memory.c: introduce help macro to_memory_block
dynamic debug: line queries failing due to uninitialized local variable
sysfs: sysfs_create_groups returns a value.
debugfs: provide debugfs_create_x64() when disabled
rbd: convert bus code to use bus_groups
firmware: dcdbas: use binary attribute groups
sysfs: add sysfs_create/remove_groups for when SYSFS is not enabled
driver core: add #include <linux/sysfs.h> to core files.
HID: convert bus code to use dev_groups
Input: serio: convert bus code to use drv_groups
Input: gameport: convert bus code to use drv_groups
driver core: firmware: use __ATTR_RW()
driver core: core: use DEVICE_ATTR_RO
driver core: bus: use DRIVER_ATTR_WO()
driver core: create write-only attribute macros for devices and drivers
sysfs: create __ATTR_WO()
driver-core: platform: convert bus code to use dev_groups
workqueue: convert bus code to use dev_groups
MEI: convert bus code to use dev_groups
...
device_hotplug_lock is held around the acpi_bus_trim() call in
acpi_scan_hot_remove() which generally removes devices (it removes
ACPI device objects at least, but it may also remove "physical"
device objects through .detach() callbacks of ACPI scan handlers).
Thus, potentially, device sysfs attributes are removed under that
lock and to remove those attributes it is necessary to hold the
s_active references of their directory entries for writing.
On the other hand, the execution of a .show() or .store() callback
from a sysfs attribute is carried out with that attribute's s_active
reference held for reading. Consequently, if any device sysfs
attribute that may be removed from within acpi_scan_hot_remove()
through acpi_bus_trim() has a .store() or .show() callback which
acquires device_hotplug_lock, the execution of that callback may
deadlock with the removal of the attribute. [Unfortunately, the
"online" device attribute of CPUs and memory blocks is one of them.]
To avoid such deadlocks, make all of the sysfs attribute callbacks
that need to lock device hotplug, for example store_online(), use
a special function, lock_device_hotplug_sysfs(), to lock device
hotplug and return the result of that function immediately if it is
not zero. This will cause the s_active reference of the directory
entry in question to be released and the syscall to be restarted
if device_hotplug_lock cannot be acquired.
[show_online() actually doesn't need to lock device hotplug, but
it is useful to serialize it with respect to device_offline() and
device_online() for the same device (in case user space attempts to
run them concurrently) which can be done with the help of
device_lock().]
Reported-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Reported-and-tested-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
There are two ways to set the online/offline state for a memory block:
echo 0|1 > online and echo online|online_kernel|online_movable|offline >
state.
The state attribute can online a memory block with extra data, the
"online type", where the online attribute uses a default online type of
ONLINE_KEEP, same as echo online > state.
Currently there is a state_mutex that provides consistency between the
memory block state and the underlying memory.
The problem is that this code does a lot of things that the common
device layer can do for us, such as the serialization of the
online/offline handlers using the device lock, setting the dev->offline
field, and calling kobject_uevent().
This patch refactors the online/offline code to allow the common
device_[online|offline] functions to be used. The result is a simpler
and more common code path for the two state setting mechanisms. It also
removes the state_mutex from the struct memory_block as the memory block
device lock provides the state consistency.
No functional change is intended by this patch.
Signed-off-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Right now memory_dev_init() maintains the memory block pointer
between iterations of add_memory_section(). This is nasty.
This patch refactors add_memory_section() to become add_memory_block().
The refactoring pulls the section scanning out of memory_dev_init()
and simplifies the signature.
Signed-off-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
