Make NFS use the new d_automount() dentry operation rather than abusing
follow_link() on directories.
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Add an AT_NO_AUTOMOUNT flag to suppress terminal automounting of automount
point directories. This can be used by fstatat() users to permit the
gathering of attributes on an automount point and also prevent
mass-automounting of a directory of automount points by ls.
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
* 'linux-next' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes/pci-2.6:
PCI/PM: Report wakeup events before resuming devices
PCI/PM: Use pm_wakeup_event() directly for reporting wakeup events
PCI: sysfs: Update ROM to include default owner write access
x86/PCI: make Broadcom CNB20LE driver EMBEDDED and EXPERIMENTAL
x86/PCI: don't use native Broadcom CNB20LE driver when ACPI is available
PCI/ACPI: Request _OSC control once for each root bridge (v3)
PCI: enable pci=bfsort by default on future Dell systems
PCI/PCIe: Clear Root PME Status bits early during system resume
PCI: pci-stub: ignore zero-length id parameters
x86/PCI: irq and pci_ids patch for Intel Patsburg
PCI: Skip id checking if no id is passed
PCI: fix __pci_device_probe kernel-doc warning
PCI: make pci_restore_state return void
PCI: Disable ASPM if BIOS asks us to
PCI: Add mask bit definition for MSI-X table
PCI: MSI: Move MSI-X entry definition to pci_regs.h
Fix up trivial conflicts in drivers/net/{skge.c,sky2.c} that had in the
meantime been converted to not use legacy PCI power management, and thus
no longer use pci_restore_state() at all (and that caused trivial
conflicts with the "make pci_restore_state return void" patch)
* git://git.infradead.org/battery-2.6: (21 commits)
power_supply: Add MAX17042 Fuel Gauge Driver
olpc_battery: Fix up XO-1.5 properties list
olpc_battery: Add support for CURRENT_NOW and VOLTAGE_NOW
olpc_battery: Add support for CHARGE_NOW
olpc_battery: Add support for CHARGE_FULL_DESIGN
olpc_battery: Ambient temperature is not available on XO-1.5
jz4740-battery: Should include linux/io.h
s3c_adc_battery: Add gpio_inverted field to pdata
power_supply: Don't use flush_scheduled_work()
power_supply: Fix use after free and memory leak
gpio-charger: Fix potential race between irq handler and probe/remove
gpio-charger: Provide default name for the power_supply
gpio-charger: Check result of kzalloc
jz4740-battery: Check if platform_data is supplied
isp1704_charger: Detect charger after probe
isp1704_charger: Set isp->dev before anything needs it
isp1704_charger: Detect HUB/Host chargers
isp1704_charger: Correct length for storing model
power_supply: Add gpio charger driver
jz4740-battery: Protect against concurrent battery readings
...
* 'vfs-scale-working' of git://git.kernel.org/pub/scm/linux/kernel/git/npiggin/linux-npiggin:
kernel: fix hlist_bl again
cgroups: Fix a lockdep warning at cgroup removal
fs: namei fix ->put_link on wrong inode in do_filp_open
* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/sameo/mfd-2.6: (59 commits)
mfd: ab8500-core chip version cut 2.0 support
mfd: Flag WM831x /IRQ as a wake source
mfd: Convert WM831x away from legacy I2C PM operations
regulator: Support MAX8998/LP3974 DVS-GPIO
mfd: Support LP3974 RTC
i2c: Convert SCx200 driver from using raw PCI to platform device
x86: OLPC: convert olpc-xo1 driver from pci device to platform device
mfd: MAX8998/LP3974 hibernation support
mfd/ab8500: remove spi support
mfd: Remove ARCH_U8500 dependency from AB8500
misc: Make AB8500_PWM driver depend on U8500 due to PWM breakage
mfd: Add __devexit annotation for vx855_remove
mfd: twl6030 irq_data conversion.
gpio: Fix cs5535 printk warnings
misc: Fix cs5535 printk warnings
mfd: Convert Wolfson MFD drivers to use irq_data accessor function
mfd: Convert TWL4030 to new irq_ APIs
mfd: Convert tps6586x driver to new irq_ API
mfd: Convert tc6393xb driver to new irq_ APIs
mfd: Convert t7166xb driver to new irq_ API
...
After recent changes related to wakeup events pm_wakeup_event()
automatically checks if the given device is configured to signal wakeup,
so pci_wakeup_event() may be a static inline function calling
pm_wakeup_event() directly.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Move the evaluation of acpi_pci_osc_control_set() (to request control of
PCI Express native features) into acpi_pci_root_add() to avoid calling
it many times for the same root complex with the same arguments.
Additionally, check if all of the requisite _OSC support bits are set
before calling acpi_pci_osc_control_set() for a given root complex.
References: https://bugzilla.kernel.org/show_bug.cgi?id=20232
Reported-by: Ozan Caglayan <ozan@pardus.org.tr>
Tested-by: Ozan Caglayan <ozan@pardus.org.tr>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
The flags added by commit db16d5ec1f
has no user now. We believe we'll use it soon but considering
patch reviewing, the change itself should be folded into incoming
set of "dirty ratio for memcg" patches.
So, it's better to drop this change from current mainline tree.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Greg Thelen <gthelen@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The MAX17042 is a fuel gauge with an I2C interface for lithium-ion
betteries. Unlike its predecessor MAX17040, MAX17042 uses 16bit
registers. Besides, MAX17042 has much more features than MAX17040; e.g.,
a thermistor, current and current accumulation measurement, battery
internal resistance estimate, average values of measurement, and others.
This patch implements a driver for MAX17042.
In this initial release, we have implemented the most basic features of
a fuel gauge: measure the battery capacity and voltage.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: Anton Vorontsov <cbouatmailru@gmail.com>
__d_rehash is dereferencing an almost-NULL pointer on my ARM926.
CONFIG_SMP=n and CONFIG_DEBUG_SPINLOCK=y.
The faulting instruction is: strne r3, [r2, #4]
and as can be seen from the register dump below, r2 is 0x00000001, hence
the faulting 0x00000005 address.
__d_rehash is essentially:
spin_lock_bucket(b);
entry->d_flags &= ~DCACHE_UNHASHED;
hlist_bl_add_head_rcu(&entry->d_hash, &b->head);
spin_unlock_bucket(b);
which is:
bit_spin_lock(0, (unsigned long *)&b->head.first);
entry->d_flags &= ~DCACHE_UNHASHED;
hlist_bl_add_head_rcu(&entry->d_hash, &b->head);
__bit_spin_unlock(0, (unsigned long *)&b->head.first);
bit_spin_lock(0, ptr) sets bit 0 of *ptr, in this case b->head.first if
CONFIG_SMP or CONFIG_DEBUG_SPINLOCK is set:
#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
while (unlikely(test_and_set_bit_lock(bitnum, addr))) {
while (test_bit(bitnum, addr)) {
preempt_enable();
cpu_relax();
preempt_disable();
}
}
#endif
So, b->head.first starts off NULL, and becomes a non-NULL (address 1).
hlist_bl_add_head_rcu() does this:
static inline void hlist_bl_add_head_rcu(struct hlist_bl_node *n,
struct hlist_bl_head *h)
{
first = hlist_bl_first(h);
n->next = first;
if (first)
first->pprev = &n->next;
It is the store to first->pprev which is faulting.
hlist_bl_first():
static inline struct hlist_bl_node *hlist_bl_first(struct hlist_bl_head *h)
{
return (struct hlist_bl_node *)
((unsigned long)h->first & ~LIST_BL_LOCKMASK);
}
but:
#if defined(CONFIG_SMP)
#define LIST_BL_LOCKMASK 1UL
#else
#define LIST_BL_LOCKMASK 0UL
#endif
So, we have one piece of code which sets bit 0 of addresses, and another
bit of code which doesn't clear it before dereferencing the pointer if
!CONFIG_SMP && CONFIG_DEBUG_SPINLOCK. With the patch below, I can again
sucessfully boot the kernel on my Versatile PB/926 platform.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This patch adds support for chip version 2.0 or cut 2.0.
One new interrupt latch register - latch 12 - is introduced.
Signed-off-by: Mattias Wallin <mattias.wallin@stericsson.com>
Acked-by: Linus Walleij <linus.walleij@stericsson.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
The previous driver did not support BUCK1-DVS3, BUCK1-DVS4, and
BUCK2-DVS2 modes. This patch adds such modes and an option to block
setting buck1/2 voltages out of the preset values.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Acked-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
The first releases of LP3974 have a large delay in RTC registers,
which requires 2 seconds of delay after writing to a rtc register
(recommended by National Semiconductor's engineers)
before reading it.
If "rtc_delay" field of the platform data is true, the rtc driver
assumes that such delays are required. Although we have not seen
LP3974s without requiring such delays, we assume that such LP3974s
will be released soon (or they have done so already) and they are
supported by "lp3974" without setting "rtc_delay" at the platform
data.
This patch adds delays with msleep when writing values to RTC registers
if the platform data has rtc_delay set.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Reviewed-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
This patch makes the driver to save and restore register values
for hibernation.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
This patch adds the ioresources used by subdrivers to
retrieve their interrupt.
Signed-off-by: Mattias Wallin <mattias.wallin@stericsson.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
Allow MFD cells to have pm_runtime_no_callbacks() called on them during
registration. This causes the runtime PM framework to ignore them,
allowing use of runtime PM to suspend the device as a whole even if
not all drivers for the MFD can usefully implement runtime PM. For
example, RTCs are likely to run continuously regardless of the power
state of the system.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
The WM8326 is a high performance variant of the WM832x series with
no software visible differences.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6: (59 commits)
ACPI / PM: Fix build problems for !CONFIG_ACPI related to NVS rework
ACPI: fix resource check message
ACPI / Battery: Update information on info notification and resume
ACPI: Drop device flag wake_capable
ACPI: Always check if _PRW is present before trying to evaluate it
ACPI / PM: Check status of power resources under mutexes
ACPI / PM: Rename acpi_power_off_device()
ACPI / PM: Drop acpi_power_nocheck
ACPI / PM: Drop acpi_bus_get_power()
Platform / x86: Make fujitsu_laptop use acpi_bus_update_power()
ACPI / Fan: Rework the handling of power resources
ACPI / PM: Register power resource devices as soon as they are needed
ACPI / PM: Register acpi_power_driver early
ACPI / PM: Add function for updating device power state consistently
ACPI / PM: Add function for device power state initialization
ACPI / PM: Introduce __acpi_bus_get_power()
ACPI / PM: Introduce function for refcounting device power resources
ACPI / PM: Add functions for manipulating lists of power resources
ACPI / PM: Prevent acpi_power_get_inferred_state() from making changes
ACPICA: Update version to 20101209
...
* 'idle-release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-idle-2.6:
cpuidle/x86/perf: fix power:cpu_idle double end events and throw cpu_idle events from the cpuidle layer
intel_idle: open broadcast clock event
cpuidle: CPUIDLE_FLAG_CHECK_BM is omap3_idle specific
cpuidle: CPUIDLE_FLAG_TLB_FLUSHED is specific to intel_idle
cpuidle: delete unused CPUIDLE_FLAG_SHALLOW, BALANCED, DEEP definitions
SH, cpuidle: delete use of NOP CPUIDLE_FLAGS_SHALLOW
cpuidle: delete NOP CPUIDLE_FLAG_POLL
ACPI: processor_idle: delete use of NOP CPUIDLE_FLAGs
cpuidle: Rename X86 specific idle poll state[0] from C0 to POLL
ACPI, intel_idle: Cleanup idle= internal variables
cpuidle: Make cpuidle_enable_device() call poll_idle_init()
intel_idle: update Sandy Bridge core C-state residency targets
* 'vfs-scale-working' of git://git.kernel.org/pub/scm/linux/kernel/git/npiggin/linux-npiggin:
fs: fix do_last error case when need_reval_dot
nfs: add missing rcu-walk check
fs: hlist UP debug fixup
fs: fix dropping of rcu-walk from force_reval_path
fs: force_reval_path drop rcu-walk before d_invalidate
fs: small rcu-walk documentation fixes
Fixed up trivial conflicts in Documentation/filesystems/porting
* 'stable/gntdev' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/p2m: Fix module linking error.
xen p2m: clear the old pte when adding a page to m2p_override
xen gntdev: use gnttab_map_refs and gnttab_unmap_refs
xen: introduce gnttab_map_refs and gnttab_unmap_refs
xen p2m: transparently change the p2m mappings in the m2p override
xen/gntdev: Fix circular locking dependency
xen/gntdev: stop using "token" argument
xen: gntdev: move use of GNTMAP_contains_pte next to the map_op
xen: add m2p override mechanism
xen: move p2m handling to separate file
xen/gntdev: add VM_PFNMAP to vma
xen/gntdev: allow usermode to map granted pages
xen: define gnttab_set_map_op/unmap_op
Fix up trivial conflict in drivers/xen/Kconfig
Po-Yu Chuang <ratbert.chuang@gmail.com> noticed that hlist_bl_set_first could
crash on a UP system when LIST_BL_LOCKMASK is 0, because
LIST_BL_BUG_ON(!((unsigned long)h->first & LIST_BL_LOCKMASK));
always evaulates to true.
Fix the expression, and also avoid a dependency between bit spinlock
implementation and list bl code (list code shouldn't know anything
except that bit 0 is set when adding and removing elements). Eventually
if a good use case comes up, we might use this list to store 1 or more
arbitrary bits of data, so it really shouldn't be tied to locking either,
but for now they are helpful for debugging.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
In the current implementation mem_cgroup_end_migration() decides whether
the page migration has succeeded or not by checking "oldpage->mapping".
But if we are tring to migrate a shmem swapcache, the page->mapping of it
is NULL from the begining, so the check would be invalid. As a result,
mem_cgroup_end_migration() assumes the migration has succeeded even if
it's not, so "newpage" would be freed while it's not uncharged.
This patch fixes it by passing mem_cgroup_end_migration() the result of
the page migration.
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Introduce a new bit spin lock, PCG_MOVE_LOCK, to synchronize the page
accounting and migration code. This reworks the locking scheme of
_update_stat() and _move_account() by adding new lock bit PCG_MOVE_LOCK,
which is always taken under IRQ disable.
1. If pages are being migrated from a memcg, then updates to that
memcg page statistics are protected by grabbing PCG_MOVE_LOCK using
move_lock_page_cgroup(). In an upcoming commit, memcg dirty page
accounting will be updating memcg page accounting (specifically: num
writeback pages) from IRQ context (softirq). Avoid a deadlocking
nested spin lock attempt by disabling irq on the local processor when
grabbing the PCG_MOVE_LOCK.
2. lock for update_page_stat is used only for avoiding race with
move_account(). So, IRQ awareness of lock_page_cgroup() itself is not
a problem. The problem is between mem_cgroup_update_page_stat() and
mem_cgroup_move_account_page().
Trade-off:
* Changing lock_page_cgroup() to always disable IRQ (or
local_bh) has some impacts on performance and I think
it's bad to disable IRQ when it's not necessary.
* adding a new lock makes move_account() slower. Score is
here.
Performance Impact: moving a 8G anon process.
Before:
real 0m0.792s
user 0m0.000s
sys 0m0.780s
After:
real 0m0.854s
user 0m0.000s
sys 0m0.842s
This score is bad but planned patches for optimization can reduce
this impact.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Greg Thelen <gthelen@google.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Andrea Righi <arighi@develer.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Replace usage of the mem_cgroup_update_file_mapped() memcg
statistic update routine with two new routines:
* mem_cgroup_inc_page_stat()
* mem_cgroup_dec_page_stat()
As before, only the file_mapped statistic is managed. However, these more
general interfaces allow for new statistics to be more easily added. New
statistics are added with memcg dirty page accounting.
Signed-off-by: Greg Thelen <gthelen@google.com>
Signed-off-by: Andrea Righi <arighi@develer.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patchset provides the ability for each cgroup to have independent
dirty page limits.
Limiting dirty memory is like fixing the max amount of dirty (hard to
reclaim) page cache used by a cgroup. So, in case of multiple cgroup
writers, they will not be able to consume more than their designated share
of dirty pages and will be forced to perform write-out if they cross that
limit.
The patches are based on a series proposed by Andrea Righi in Mar 2010.
Overview:
- Add page_cgroup flags to record when pages are dirty, in writeback, or nfs
unstable.
- Extend mem_cgroup to record the total number of pages in each of the
interesting dirty states (dirty, writeback, unstable_nfs).
- Add dirty parameters similar to the system-wide /proc/sys/vm/dirty_*
limits to mem_cgroup. The mem_cgroup dirty parameters are accessible
via cgroupfs control files.
- Consider both system and per-memcg dirty limits in page writeback when
deciding to queue background writeback or block for foreground writeback.
Known shortcomings:
- When a cgroup dirty limit is exceeded, then bdi writeback is employed to
writeback dirty inodes. Bdi writeback considers inodes from any cgroup, not
just inodes contributing dirty pages to the cgroup exceeding its limit.
- When memory.use_hierarchy is set, then dirty limits are disabled. This is a
implementation detail. An enhanced implementation is needed to check the
chain of parents to ensure that no dirty limit is exceeded.
Performance data:
- A page fault microbenchmark workload was used to measure performance, which
can be called in read or write mode:
f = open(foo. $cpu)
truncate(f, 4096)
alarm(60)
while (1) {
p = mmap(f, 4096)
if (write)
*p = 1
else
x = *p
munmap(p)
}
- The workload was called for several points in the patch series in different
modes:
- s_read is a single threaded reader
- s_write is a single threaded writer
- p_read is a 16 thread reader, each operating on a different file
- p_write is a 16 thread writer, each operating on a different file
- Measurements were collected on a 16 core non-numa system using "perf stat
--repeat 3". The -a option was used for parallel (p_*) runs.
- All numbers are page fault rate (M/sec). Higher is better.
- To compare the performance of a kernel without non-memcg compare the first and
last rows, neither has memcg configured. The first row does not include any
of these memcg patches.
- To compare the performance of using memcg dirty limits, compare the baseline
(2nd row titled "w/ memcg") with the the code and memcg enabled (2nd to last
row titled "all patches").
root_cgroup child_cgroup
s_read s_write p_read p_write s_read s_write p_read p_write
mmotm w/o memcg 0.428 0.390 0.429 0.388
mmotm w/ memcg 0.411 0.378 0.391 0.362 0.412 0.377 0.385 0.363
all patches 0.384 0.360 0.370 0.348 0.381 0.363 0.368 0.347
all patches 0.431 0.402 0.427 0.395
w/o memcg
This patch:
Add additional flags to page_cgroup to track dirty pages within a
mem_cgroup.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrea Righi <arighi@develer.com>
Signed-off-by: Greg Thelen <gthelen@google.com>
Acked-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
migrate_pages() -> unmap_and_move() only calls rcu_read_lock() for
anonymous pages, as introduced by git commit
989f89c57e ("fix rcu_read_lock() in page
migraton"). The point of the RCU protection there is part of getting a
stable reference to anon_vma and is only held for anon pages as file pages
are locked which is sufficient protection against freeing.
However, while a file page's mapping is being migrated, the radix tree is
double checked to ensure it is the expected page. This uses
radix_tree_deref_slot() -> rcu_dereference() without the RCU lock held
triggering the following warning.
[ 173.674290] ===================================================
[ 173.676016] [ INFO: suspicious rcu_dereference_check() usage. ]
[ 173.676016] ---------------------------------------------------
[ 173.676016] include/linux/radix-tree.h:145 invoked rcu_dereference_check() without protection!
[ 173.676016]
[ 173.676016] other info that might help us debug this:
[ 173.676016]
[ 173.676016]
[ 173.676016] rcu_scheduler_active = 1, debug_locks = 0
[ 173.676016] 1 lock held by hugeadm/2899:
[ 173.676016] #0: (&(&inode->i_data.tree_lock)->rlock){..-.-.}, at: [<c10e3d2b>] migrate_page_move_mapping+0x40/0x1ab
[ 173.676016]
[ 173.676016] stack backtrace:
[ 173.676016] Pid: 2899, comm: hugeadm Not tainted 2.6.37-rc5-autobuild
[ 173.676016] Call Trace:
[ 173.676016] [<c128cc01>] ? printk+0x14/0x1b
[ 173.676016] [<c1063502>] lockdep_rcu_dereference+0x7d/0x86
[ 173.676016] [<c10e3db5>] migrate_page_move_mapping+0xca/0x1ab
[ 173.676016] [<c10e41ad>] migrate_page+0x23/0x39
[ 173.676016] [<c10e491b>] buffer_migrate_page+0x22/0x107
[ 173.676016] [<c10e48f9>] ? buffer_migrate_page+0x0/0x107
[ 173.676016] [<c10e425d>] move_to_new_page+0x9a/0x1ae
[ 173.676016] [<c10e47e6>] migrate_pages+0x1e7/0x2fa
This patch introduces radix_tree_deref_slot_protected() which calls
rcu_dereference_protected(). Users of it must pass in the
mapping->tree_lock that is protecting this dereference. Holding the tree
lock protects against parallel updaters of the radix tree meaning that
rcu_dereference_protected is allowable.
[akpm@linux-foundation.org: remove unneeded casts]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Milton Miller <miltonm@bga.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: <stable@kernel.org> [2.6.37.early]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Cleanup some code with common compound_trans_head helper.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Avi Kivity <avi@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
MADV_HUGEPAGE and MADV_NOHUGEPAGE were fully effective only if run after
mmap and before touching the memory. While this is enough for most
usages, it's little effort to make madvise more dynamic at runtime on an
existing mapping by making khugepaged aware about madvise.
MADV_HUGEPAGE: register in khugepaged immediately without waiting a page
fault (that may not ever happen if all pages are already mapped and the
"enabled" knob was set to madvise during the initial page faults).
MADV_NOHUGEPAGE: skip vmas marked VM_NOHUGEPAGE in khugepaged to stop
collapsing pages where not needed.
[akpm@linux-foundation.org: tweak comment]
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add madvise MADV_NOHUGEPAGE to mark regions that are not important to be
hugepage backed. Return -EINVAL if the vma is not of an anonymous type,
or the feature isn't built into the kernel. Never silently return
success.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Count each transparent hugepage as HPAGE_PMD_NR pages in the LRU
statistics, so the Active(anon) and Inactive(anon) statistics in
/proc/meminfo are correct.
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This takes advantage of memory compaction to properly generate pages of
order > 0 if regular page reclaim fails and priority level becomes more
severe and we don't reach the proper watermarks.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
For GRU and EPT, we need gup-fast to set referenced bit too (this is why
it's correct to return 0 when shadow_access_mask is zero, it requires
gup-fast to set the referenced bit). qemu-kvm access already sets the
young bit in the pte if it isn't zero-copy, if it's zero copy or a shadow
paging EPT minor fault we relay on gup-fast to signal the page is in
use...
We also need to check the young bits on the secondary pagetables for NPT
and not nested shadow mmu as the data may never get accessed again by the
primary pte.
Without this closer accuracy, we'd have to remove the heuristic that
avoids collapsing hugepages in hugepage virtual regions that have not even
a single subpage in use.
->test_young is full backwards compatible with GRU and other usages that
don't have young bits in pagetables set by the hardware and that should
nuke the secondary mmu mappings when ->clear_flush_young runs just like
EPT does.
Removing the heuristic that checks the young bit in
khugepaged/collapse_huge_page completely isn't so bad either probably but
I thought it was worth it and this makes it reliable.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
An huge pmd can only be mapped if the corresponding 2M virtual range is
fully contained in the vma. At times the VM calls split_vma twice, if the
first split_vma succeeds and the second fail, the first split_vma remains
in effect and it's not rolled back. For split_vma or vma_adjust to fail
an allocation failure is needed so it's a very unlikely event (the out of
memory killer would normally fire before any allocation failure is visible
to kernel and userland and if an out of memory condition happens it's
unlikely to happen exactly here). Nevertheless it's safer to ensure that
no huge pmd can be left around if the vma is adjusted in a way that can't
fit hugepages anymore at the new vm_start/vm_end address.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It's mostly a matter of replacing alloc_pages with alloc_pages_vma after
introducing alloc_pages_vma. khugepaged needs special handling as the
allocation has to happen inside collapse_huge_page where the vma is known
and an error has to be returned to the outer loop to sleep
alloc_sleep_millisecs in case of failure. But it retains the more
efficient logic of handling allocation failures in khugepaged in case of
CONFIG_NUMA=n.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Natively handle huge pmds when changing page tables on behalf of
mprotect().
I left out update_mmu_cache() because we do not need it on x86 anyway but
more importantly the interface works on ptes, not pmds.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Handle transparent huge page pmd entries natively instead of splitting
them into subpages.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add support for transparent hugepages to x86 32bit.
Share the same VM_ bitflag for VM_MAPPED_COPY. mm/nommu.c will never
support transparent hugepages.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
PG_buddy can be converted to _mapcount == -2. So the PG_compound_lock can
be added to page->flags without overflowing (because of the sparse section
bits increasing) with CONFIG_X86_PAE=y and CONFIG_X86_PAT=y. This also
has to move the memory hotplug code from _mapcount to lru.next to avoid
any risk of clashes. We can't use lru.next for PG_buddy removal, but
memory hotplug can use lru.next even more easily than the mapcount
instead.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add khugepaged to relocate fragmented pages into hugepages if new
hugepages become available. (this is indipendent of the defrag logic that
will have to make new hugepages available)
The fundamental reason why khugepaged is unavoidable, is that some memory
can be fragmented and not everything can be relocated. So when a virtual
machine quits and releases gigabytes of hugepages, we want to use those
freely available hugepages to create huge-pmd in the other virtual
machines that may be running on fragmented memory, to maximize the CPU
efficiency at all times. The scan is slow, it takes nearly zero cpu time,
except when it copies data (in which case it means we definitely want to
pay for that cpu time) so it seems a good tradeoff.
In addition to the hugepages being released by other process releasing
memory, we have the strong suspicion that the performance impact of
potentially defragmenting hugepages during or before each page fault could
lead to more performance inconsistency than allocating small pages at
first and having them collapsed into large pages later... if they prove
themselfs to be long lived mappings (khugepaged scan is slow so short
lived mappings have low probability to run into khugepaged if compared to
long lived mappings).
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add hugepage stat information to /proc/vmstat and /proc/meminfo.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
No pmd_trans_huge should ever materialize in migration ptes areas, because
we split the hugepage before migration ptes are instantiated.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add madvise MADV_HUGEPAGE to mark regions that are important to be
hugepage backed. Return -EINVAL if the vma is not of an anonymous type,
or the feature isn't built into the kernel. Never silently return
success.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Lately I've been working to make KVM use hugepages transparently without
the usual restrictions of hugetlbfs. Some of the restrictions I'd like to
see removed:
1) hugepages have to be swappable or the guest physical memory remains
locked in RAM and can't be paged out to swap
2) if a hugepage allocation fails, regular pages should be allocated
instead and mixed in the same vma without any failure and without
userland noticing
3) if some task quits and more hugepages become available in the
buddy, guest physical memory backed by regular pages should be
relocated on hugepages automatically in regions under
madvise(MADV_HUGEPAGE) (ideally event driven by waking up the
kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes
not null)
4) avoidance of reservation and maximization of use of hugepages whenever
possible. Reservation (needed to avoid runtime fatal faliures) may be ok for
1 machine with 1 database with 1 database cache with 1 database cache size
known at boot time. It's definitely not feasible with a virtualization
hypervisor usage like RHEV-H that runs an unknown number of virtual machines
with an unknown size of each virtual machine with an unknown amount of
pagecache that could be potentially useful in the host for guest not using
O_DIRECT (aka cache=off).
hugepages in the virtualization hypervisor (and also in the guest!) are
much more important than in a regular host not using virtualization,
becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24
to 19 in case only the hypervisor uses transparent hugepages, and they
decrease the tlb-miss cacheline accesses from 19 to 15 in case both the
linux hypervisor and the linux guest both uses this patch (though the
guest will limit the addition speedup to anonymous regions only for
now...). Even more important is that the tlb miss handler is much slower
on a NPT/EPT guest than for a regular shadow paging or no-virtualization
scenario. So maximizing the amount of virtual memory cached by the TLB
pays off significantly more with NPT/EPT than without (even if there would
be no significant speedup in the tlb-miss runtime).
The first (and more tedious) part of this work requires allowing the VM to
handle anonymous hugepages mixed with regular pages transparently on
regular anonymous vmas. This is what this patch tries to achieve in the
least intrusive possible way. We want hugepages and hugetlb to be used in
a way so that all applications can benefit without changes (as usual we
leverage the KVM virtualization design: by improving the Linux VM at
large, KVM gets the performance boost too).
The most important design choice is: always fallback to 4k allocation if
the hugepage allocation fails! This is the _very_ opposite of some large
pagecache patches that failed with -EIO back then if a 64k (or similar)
allocation failed...
Second important decision (to reduce the impact of the feature on the
existing pagetable handling code) is that at any time we can split an
hugepage into 512 regular pages and it has to be done with an operation
that can't fail. This way the reliability of the swapping isn't decreased
(no need to allocate memory when we are short on memory to swap) and it's
trivial to plug a split_huge_page* one-liner where needed without
polluting the VM. Over time we can teach mprotect, mremap and friends to
handle pmd_trans_huge natively without calling split_huge_page*. The fact
it can't fail isn't just for swap: if split_huge_page would return -ENOMEM
(instead of the current void) we'd need to rollback the mprotect from the
middle of it (ideally including undoing the split_vma) which would be a
big change and in the very wrong direction (it'd likely be simpler not to
call split_huge_page at all and to teach mprotect and friends to handle
hugepages instead of rolling them back from the middle). In short the
very value of split_huge_page is that it can't fail.
The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and
incremental and it'll just be an "harmless" addition later if this initial
part is agreed upon. It also should be noted that locking-wise replacing
regular pages with hugepages is going to be very easy if compared to what
I'm doing below in split_huge_page, as it will only happen when
page_count(page) matches page_mapcount(page) if we can take the PG_lock
and mmap_sem in write mode. collapse_huge_page will be a "best effort"
that (unlike split_huge_page) can fail at the minimal sign of trouble and
we can try again later. collapse_huge_page will be similar to how KSM
works and the madvise(MADV_HUGEPAGE) will work similar to
madvise(MADV_MERGEABLE).
The default I like is that transparent hugepages are used at page fault
time. This can be changed with
/sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set
to three values "always", "madvise", "never" which mean respectively that
hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions,
or never used. /sys/kernel/mm/transparent_hugepage/defrag instead
controls if the hugepage allocation should defrag memory aggressively
"always", only inside "madvise" regions, or "never".
The pmd_trans_splitting/pmd_trans_huge locking is very solid. The
put_page (from get_user_page users that can't use mmu notifier like
O_DIRECT) that runs against a __split_huge_page_refcount instead was a
pain to serialize in a way that would result always in a coherent page
count for both tail and head. I think my locking solution with a
compound_lock taken only after the page_first is valid and is still a
PageHead should be safe but it surely needs review from SMP race point of
view. In short there is no current existing way to serialize the O_DIRECT
final put_page against split_huge_page_refcount so I had to invent a new
one (O_DIRECT loses knowledge on the mapping status by the time gup_fast
returns so...). And I didn't want to impact all gup/gup_fast users for
now, maybe if we change the gup interface substantially we can avoid this
locking, I admit I didn't think too much about it because changing the gup
unpinning interface would be invasive.
If we ignored O_DIRECT we could stick to the existing compound refcounting
code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM
(and any other mmu notifier user) would call it without FOLL_GET (and if
FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the
current task mmu notifier list yet). But O_DIRECT is fundamental for
decent performance of virtualized I/O on fast storage so we can't avoid it
to solve the race of put_page against split_huge_page_refcount to achieve
a complete hugepage feature for KVM.
Swap and oom works fine (well just like with regular pages ;). MMU
notifier is handled transparently too, with the exception of the young bit
on the pmd, that didn't have a range check but I think KVM will be fine
because the whole point of hugepages is that EPT/NPT will also use a huge
pmd when they notice gup returns pages with PageCompound set, so they
won't care of a range and there's just the pmd young bit to check in that
case.
NOTE: in some cases if the L2 cache is small, this may slowdown and waste
memory during COWs because 4M of memory are accessed in a single fault
instead of 8k (the payoff is that after COW the program can run faster).
So we might want to switch the copy_huge_page (and clear_huge_page too) to
not temporal stores. I also extensively researched ways to avoid this
cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k
up to 1M (I can send those patches that fully implemented prefault) but I
concluded they're not worth it and they add an huge additional complexity
and they remove all tlb benefits until the full hugepage has been faulted
in, to save a little bit of memory and some cache during app startup, but
they still don't improve substantially the cache-trashing during startup
if the prefault happens in >4k chunks. One reason is that those 4k pte
entries copied are still mapped on a perfectly cache-colored hugepage, so
the trashing is the worst one can generate in those copies (cow of 4k page
copies aren't so well colored so they trashes less, but again this results
in software running faster after the page fault). Those prefault patches
allowed things like a pte where post-cow pages were local 4k regular anon
pages and the not-yet-cowed pte entries were pointing in the middle of
some hugepage mapped read-only. If it doesn't payoff substantially with
todays hardware it will payoff even less in the future with larger l2
caches, and the prefault logic would blot the VM a lot. If one is
emebdded transparent_hugepage can be disabled during boot with sysfs or
with the boot commandline parameter transparent_hugepage=0 (or
transparent_hugepage=2 to restrict hugepages inside madvise regions) that
will ensure not a single hugepage is allocated at boot time. It is simple
enough to just disable transparent hugepage globally and let transparent
hugepages be allocated selectively by applications in the MADV_HUGEPAGE
region (both at page fault time, and if enabled with the
collapse_huge_page too through the kernel daemon).
This patch supports only hugepages mapped in the pmd, archs that have
smaller hugepages will not fit in this patch alone. Also some archs like
power have certain tlb limits that prevents mixing different page size in
the same regions so they will not fit in this framework that requires
"graceful fallback" to basic PAGE_SIZE in case of physical memory
fragmentation. hugetlbfs remains a perfect fit for those because its
software limits happen to match the hardware limits. hugetlbfs also
remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped
to be found not fragmented after a certain system uptime and that would be
very expensive to defragment with relocation, so requiring reservation.
hugetlbfs is the "reservation way", the point of transparent hugepages is
not to have any reservation at all and maximizing the use of cache and
hugepages at all times automatically.
Some performance result:
vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep
ages3
memset page fault 1566023
memset tlb miss 453854
memset second tlb miss 453321
random access tlb miss 41635
random access second tlb miss 41658
vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3
memset page fault 1566471
memset tlb miss 453375
memset second tlb miss 453320
random access tlb miss 41636
random access second tlb miss 41637
vmx andrea # ./largepages3
memset page fault 1566642
memset tlb miss 453417
memset second tlb miss 453313
random access tlb miss 41630
random access second tlb miss 41647
vmx andrea # ./largepages3
memset page fault 1566872
memset tlb miss 453418
memset second tlb miss 453315
random access tlb miss 41618
random access second tlb miss 41659
vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage
vmx andrea # ./largepages3
memset page fault 2182476
memset tlb miss 460305
memset second tlb miss 460179
random access tlb miss 44483
random access second tlb miss 44186
vmx andrea # ./largepages3
memset page fault 2182791
memset tlb miss 460742
memset second tlb miss 459962
random access tlb miss 43981
random access second tlb miss 43988
============
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/time.h>
#define SIZE (3UL*1024*1024*1024)
int main()
{
char *p = malloc(SIZE), *p2;
struct timeval before, after;
gettimeofday(&before, NULL);
memset(p, 0, SIZE);
gettimeofday(&after, NULL);
printf("memset page fault %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);
gettimeofday(&before, NULL);
memset(p, 0, SIZE);
gettimeofday(&after, NULL);
printf("memset tlb miss %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);
gettimeofday(&before, NULL);
memset(p, 0, SIZE);
gettimeofday(&after, NULL);
printf("memset second tlb miss %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);
gettimeofday(&before, NULL);
for (p2 = p; p2 < p+SIZE; p2 += 4096)
*p2 = 0;
gettimeofday(&after, NULL);
printf("random access tlb miss %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);
gettimeofday(&before, NULL);
for (p2 = p; p2 < p+SIZE; p2 += 4096)
*p2 = 0;
gettimeofday(&after, NULL);
printf("random access second tlb miss %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);
return 0;
}
============
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
