rt_mutex_trylock() must be called from thread context. It can be
called from atomic regions (preemption or interrupts disabled), but
not from hard/softirq/nmi context. Add a warning to alert abusers.
The reasons for this are:
1) There is a potential deadlock in the slowpath
2) Another cpu which blocks on the rtmutex will boost the task
which allegedly locked the rtmutex, but that cannot work
because the hard/softirq context borrows the task context.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
The rtmutex code is the only user of __HAVE_ARCH_CMPXCHG and we have a few
other user of cmpxchg() which do not care about __HAVE_ARCH_CMPXCHG. This
define was first introduced in 23f78d4a0 ("[PATCH] pi-futex: rt mutex core")
which is v2.6.18. The generic cmpxchg was introduced later in 068fbad288
("Add cmpxchg_local to asm-generic for per cpu atomic operations") which is
v2.6.25.
Back then something was required to get rtmutex working with the fast
path on architectures without cmpxchg and this seems to be the result.
It popped up recently on rt-users because ARM (v6+) does not define
__HAVE_ARCH_CMPXCHG (even that it implements it) which results in slower
locking performance in the fast path.
To put some numbers on it: preempt -RT, am335x, 10 loops of
100000 invocations of rt_spin_lock() + rt_spin_unlock() (time "total" is
the average of the 10 loops for the 100000 invocations, "loop" is
"total / 100000 * 1000"):
cmpxchg | slowpath used || cmpxchg used
| total | loop || total | loop
--------|-----------|-------||------------|-------
ARMv6 | 9129.4 us | 91 ns || 3311.9 us | 33 ns
generic | 9360.2 us | 94 ns || 10834.6 us | 108 ns
----------------------------||--------------------
Forcing it to generic cmpxchg() made things worse for the slowpath and
even worse in cmpxchg() path. It boils down to 14ns more per lock+unlock
in a cache hot loop so it might not be that much in real world.
The last test was a substitute for pre ARMv6 machine but then I was able
to perform the comparison on imx28 which is ARMv5 and therefore is
always is using the generic cmpxchg implementation. And the numbers:
| total | loop
-------- |----------- |--------
slowpath | 263937.2 us | 2639 ns
cmpxchg | 16934.2 us | 169 ns
--------------------------------
The numbers are larger since the machine is slower in general. However,
letting rtmutex use cmpxchg() instead the slowpath seem to improve things.
Since from the ARM (tested on am335x + imx28) point of view always
using cmpxchg() in rt_mutex_lock() + rt_mutex_unlock() makes sense I
would drop the define.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: will.deacon@arm.com
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/20150225175613.GE6823@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Currently, atomic_cmpxchg() is used to get the lock. However, this
is not really necessary if there is more than one task in the queue
and the queue head don't need to reset the tail code. For that case,
a simple write to set the lock bit is enough as the queue head will
be the only one eligible to get the lock as long as it checks that
both the lock and pending bits are not set. The current pending bit
waiting code will ensure that the bit will not be set as soon as the
tail code in the lock is set.
With that change, the are some slight improvement in the performance
of the queued spinlock in the 5M loop micro-benchmark run on a 4-socket
Westere-EX machine as shown in the tables below.
[Standalone/Embedded - same node]
# of tasks Before patch After patch %Change
---------- ----------- ---------- -------
3 2324/2321 2248/2265 -3%/-2%
4 2890/2896 2819/2831 -2%/-2%
5 3611/3595 3522/3512 -2%/-2%
6 4281/4276 4173/4160 -3%/-3%
7 5018/5001 4875/4861 -3%/-3%
8 5759/5750 5563/5568 -3%/-3%
[Standalone/Embedded - different nodes]
# of tasks Before patch After patch %Change
---------- ----------- ---------- -------
3 12242/12237 12087/12093 -1%/-1%
4 10688/10696 10507/10521 -2%/-2%
It was also found that this change produced a much bigger performance
improvement in the newer IvyBridge-EX chip and was essentially to close
the performance gap between the ticket spinlock and queued spinlock.
The disk workload of the AIM7 benchmark was run on a 4-socket
Westmere-EX machine with both ext4 and xfs RAM disks at 3000 users
on a 3.14 based kernel. The results of the test runs were:
AIM7 XFS Disk Test
kernel JPM Real Time Sys Time Usr Time
----- --- --------- -------- --------
ticketlock 5678233 3.17 96.61 5.81
qspinlock 5750799 3.13 94.83 5.97
AIM7 EXT4 Disk Test
kernel JPM Real Time Sys Time Usr Time
----- --- --------- -------- --------
ticketlock 1114551 16.15 509.72 7.11
qspinlock 2184466 8.24 232.99 6.01
The ext4 filesystem run had a much higher spinlock contention than
the xfs filesystem run.
The "ebizzy -m" test was also run with the following results:
kernel records/s Real Time Sys Time Usr Time
----- --------- --------- -------- --------
ticketlock 2075 10.00 216.35 3.49
qspinlock 3023 10.00 198.20 4.80
Signed-off-by: Waiman Long <Waiman.Long@hp.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Daniel J Blueman <daniel@numascale.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Douglas Hatch <doug.hatch@hp.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Paolo Bonzini <paolo.bonzini@gmail.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Scott J Norton <scott.norton@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: virtualization@lists.linux-foundation.org
Cc: xen-devel@lists.xenproject.org
Link: http://lkml.kernel.org/r/1429901803-29771-7-git-send-email-Waiman.Long@hp.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch introduces a new generic queued spinlock implementation that
can serve as an alternative to the default ticket spinlock. Compared
with the ticket spinlock, this queued spinlock should be almost as fair
as the ticket spinlock. It has about the same speed in single-thread
and it can be much faster in high contention situations especially when
the spinlock is embedded within the data structure to be protected.
Only in light to moderate contention where the average queue depth
is around 1-3 will this queued spinlock be potentially a bit slower
due to the higher slowpath overhead.
This queued spinlock is especially suit to NUMA machines with a large
number of cores as the chance of spinlock contention is much higher
in those machines. The cost of contention is also higher because of
slower inter-node memory traffic.
Due to the fact that spinlocks are acquired with preemption disabled,
the process will not be migrated to another CPU while it is trying
to get a spinlock. Ignoring interrupt handling, a CPU can only be
contending in one spinlock at any one time. Counting soft IRQ, hard
IRQ and NMI, a CPU can only have a maximum of 4 concurrent lock waiting
activities. By allocating a set of per-cpu queue nodes and used them
to form a waiting queue, we can encode the queue node address into a
much smaller 24-bit size (including CPU number and queue node index)
leaving one byte for the lock.
Please note that the queue node is only needed when waiting for the
lock. Once the lock is acquired, the queue node can be released to
be used later.
Signed-off-by: Waiman Long <Waiman.Long@hp.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Daniel J Blueman <daniel@numascale.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Douglas Hatch <doug.hatch@hp.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Paolo Bonzini <paolo.bonzini@gmail.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Scott J Norton <scott.norton@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: virtualization@lists.linux-foundation.org
Cc: xen-devel@lists.xenproject.org
Link: http://lkml.kernel.org/r/1429901803-29771-2-git-send-email-Waiman.Long@hp.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
In up_write()/up_read(), rwsem_wake() will be called whenever it
detects that some writers/readers are waiting. The rwsem_wake()
function will take the wait_lock and call __rwsem_do_wake() to do the
real wakeup. For a heavily contended rwsem, doing a spin_lock() on
wait_lock will cause further contention on the heavily contended rwsem
cacheline resulting in delay in the completion of the up_read/up_write
operations.
This patch makes the wait_lock taking and the call to __rwsem_do_wake()
optional if at least one spinning writer is present. The spinning
writer will be able to take the rwsem and call rwsem_wake() later
when it calls up_write(). With the presence of a spinning writer,
rwsem_wake() will now try to acquire the lock using trylock. If that
fails, it will just quit.
Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Waiman Long <Waiman.Long@hp.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Davidlohr Bueso <dave@stgolabs.net>
Acked-by: Jason Low <jason.low2@hp.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Douglas Hatch <doug.hatch@hp.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Scott J Norton <scott.norton@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1430428337-16802-2-git-send-email-Waiman.Long@hp.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull RCU fix from Ingo Molnar:
"An RCU Kconfig fix that eliminates an annoying interactive kconfig
question for CONFIG_RCU_TORTURE_TEST_SLOW_INIT"
* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
rcu: Control grace-period delays directly from value
Pull networking fixes from David Miller:
1) Receive packet length needs to be adjust by 2 on RX to accomodate
the two padding bytes in altera_tse driver. From Vlastimil Setka.
2) If rx frame is dropped due to out of memory in macb driver, we leave
the receive ring descriptors in an undefined state. From Punnaiah
Choudary Kalluri
3) Some netlink subsystems erroneously signal NLM_F_MULTI. That is
only for dumps. Fix from Nicolas Dichtel.
4) Fix mis-use of raw rt->rt_pmtu value in ipv4, one must always go via
the ipv4_mtu() helper. From Herbert Xu.
5) Fix null deref in bridge netfilter, and miscalculated lengths in
jump/goto nf_tables verdicts. From Florian Westphal.
6) Unhash ping sockets properly.
7) Software implementation of BPF divide did 64/32 rather than 64/64
bit divide. The JITs got it right. Fix from Alexei Starovoitov.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (30 commits)
ipv4: Missing sk_nulls_node_init() in ping_unhash().
net: fec: Fix RGMII-ID mode
net/mlx4_en: Schedule napi when RX buffers allocation fails
netxen_nic: use spin_[un]lock_bh around tx_clean_lock
net/mlx4_core: Fix unaligned accesses
mlx4_en: Use correct loop cursor in error path.
cxgb4: Fix MC1 memory offset calculation
bnx2x: Delay during kdump load
net: Fix Kernel Panic in bonding driver debugfs file: rlb_hash_table
net: dsa: Fix scope of eeprom-length property
net: macb: Fix race condition in driver when Rx frame is dropped
hv_netvsc: Fix a bug in netvsc_start_xmit()
altera_tse: Correct rx packet length
mlx4: Fix tx ring affinity_mask creation
tipc: fix problem with parallel link synchronization mechanism
tipc: remove wrong use of NLM_F_MULTI
bridge/nl: remove wrong use of NLM_F_MULTI
bridge/mdb: remove wrong use of NLM_F_MULTI
net: sched: act_connmark: don't zap skb->nfct
trivial: net: systemport: bcmsysport.h: fix 0x0x prefix
...
Pull power management and ACPI fixes from Rafael Wysocki:
"Three regression fixes this time, one for a recent regression in the
cpuidle core affecting multiple systems, one for an inadvertently
added duplicate typedef in ACPICA that breaks compilation with GCC 4.5
and one for an ACPI Smart Battery Subsystem driver regression
introduced during the 3.18 cycle (stable-candidate).
Specifics:
- Fix for a regression in the cpuidle core introduced by one of the
recent commits in the clockevents_notify() removal series that put
a call to a function which had to be executed with disabled
interrupts into a code path running with enabled interrupts (Rafael
J Wysocki)
- Fix for a build problem in ACPICA (with GCC 4.5) introduced by one
of the recent ACPICA tools commits that added a duplicate typedef
to one of the ACPICA's header files by mistake (Olaf Hering)
- Fix for a regression in the ACPI SBS (Smart Battery Subsystem)
driver introduced during the 3.18 development cycle causing the
smart battery manager to be marked as not present when it should be
marked as present (Chris Bainbridge)"
* tag 'pm+acpi-4.1-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
cpuidle: Run tick_broadcast_exit() with disabled interrupts
ACPI / SBS: Enable battery manager when present
ACPICA: remove duplicate u8 typedef
Pull kvm changes from Paolo Bonzini:
"Remove from guest code the handling of task migration during a pvclock
read; instead use the correct protocol in KVM.
This removes the need for task migration notifiers in core scheduler
code"
[ The scheduler people really hated the migration notifiers, so this was
kind of required - Linus ]
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
x86: pvclock: Really remove the sched notifier for cross-cpu migrations
kvm: x86: fix kvmclock update protocol
Commit 335f49196f (sched/idle: Use explicit broadcast oneshot
control function) replaced clockevents_notify() invocations in
cpuidle_idle_call() with direct calls to tick_broadcast_enter()
and tick_broadcast_exit(), but it overlooked the fact that
interrupts were already enabled before calling the latter which
led to functional breakage on systems using idle states with the
CPUIDLE_FLAG_TIMER_STOP flag set.
Fix that by moving the invocations of tick_broadcast_enter()
and tick_broadcast_exit() down into cpuidle_enter_state() where
interrupts are still disabled when tick_broadcast_exit() is
called. Also ensure that interrupts will be disabled before
running tick_broadcast_exit() even if they have been enabled by
the idle state's ->enter callback. Trigger a WARN_ON_ONCE() in
that case, as we generally don't want that to happen for states
with CPUIDLE_FLAG_TIMER_STOP set.
Fixes: 335f49196f (sched/idle: Use explicit broadcast oneshot control function)
Reported-and-tested-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Reported-and-tested-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Pull s390 updates from Martin Schwidefsky:
"One additional new feature for 4.1, a new PRNG based on SHA-512 for
the zcrypt driver.
Two memory management related changes, the page table reallocation for
KVM is removed, and with file ptes gone the encoding of page table
entries is improved.
And three bug fixes"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
s390/zcrypt: Introduce new SHA-512 based Pseudo Random Generator.
s390/mm: change swap pte encoding and pgtable cleanup
s390/mm: correct transfer of dirty & young bits in __pmd_to_pte
s390/bpf: add dependency to z196 features
s390/3215: free memory in error path
s390/kvm: remove delayed reallocation of page tables for KVM
kexec: allocate the kexec control page with KEXEC_CONTROL_MEMORY_GFP
ALU64_DIV instruction should be dividing 64-bit by 64-bit,
whereas do_div() does 64-bit by 32-bit divide.
x64 and arm64 JITs correctly implement 64 by 64 unsigned divide.
llvm BPF backend emits code assuming that ALU64_DIV does 64 by 64.
Fixes: 89aa075832 ("net: sock: allow eBPF programs to be attached to sockets")
Reported-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commits 0a4e6be9ca
and 80f7fdb1c7.
The task migration notifier was originally introduced in order to support
the pvclock vsyscall with non-synchronized TSC, but KVM only supports it
with synchronized TSC. Hence, on KVM the race condition is only needed
due to a bad implementation on the host side, and even then it's so rare
that it's mostly theoretical.
As far as KVM is concerned it's possible to fix the host, avoiding the
additional complexity in the vDSO and the (re)introduction of the task
migration notifier.
Xen, on the other hand, hasn't yet implemented vsyscall support at
all, so we do not care about its plans for non-synchronized TSC.
Reported-by: Peter Zijlstra <peterz@infradead.org>
Suggested-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Peter Zijlstra
Thomas Gleixner
Sebastian Andrzej Siewior
Waiman Long
Ingo Molnar
David Vrabel
Linus Torvalds
David Howells
Rafael J. Wysocki
Alexei Starovoitov
Paolo Bonzini