As the generic rwsem-xadd code is using the appropriate acquire and
release versions of the atomic operations, the arch specific rwsem.h
files will not be that much faster than the generic code as long as the
atomic functions are properly implemented. So we can remove those arch
specific rwsem.h and stop building asm/rwsem.h to reduce maintenance
effort.
Currently, only x86, alpha and ia64 have implemented architecture
specific fast paths. I don't have access to alpha and ia64 systems for
testing, but they are legacy systems that are not likely to be updated
to the latest kernel anyway.
By using a rwsem microbenchmark, the total locking rates on a 4-socket
56-core 112-thread x86-64 system before and after the patch were as
follows (mixed means equal # of read and write locks):
Before Patch After Patch
# of Threads wlock rlock mixed wlock rlock mixed
------------ ----- ----- ----- ----- ----- -----
1 29,201 30,143 29,458 28,615 30,172 29,201
2 6,807 13,299 1,171 7,725 15,025 1,804
4 6,504 12,755 1,520 7,127 14,286 1,345
8 6,762 13,412 764 6,826 13,652 726
16 6,693 15,408 662 6,599 15,938 626
32 6,145 15,286 496 5,549 15,487 511
64 5,812 15,495 60 5,858 15,572 60
There were some run-to-run variations for the multi-thread tests. For
x86-64, using the generic C code fast path seems to be a little bit
faster than the assembly version with low lock contention. Looking at
the assembly version of the fast paths, there are assembly to/from C
code wrappers that save and restore all the callee-clobbered registers
(7 registers on x86-64). The assembly generated from the generic C
code doesn't need to do that. That may explain the slight performance
gain here.
The generic asm rwsem.h can also be merged into kernel/locking/rwsem.h
with no code change as no other code other than those under
kernel/locking needs to access the internal rwsem macros and functions.
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: linux-c6x-dev@linux-c6x.org
Cc: linux-m68k@lists.linux-m68k.org
Cc: linux-riscv@lists.infradead.org
Cc: linux-um@lists.infradead.org
Cc: linux-xtensa@linux-xtensa.org
Cc: linuxppc-dev@lists.ozlabs.org
Cc: nios2-dev@lists.rocketboards.org
Cc: openrisc@lists.librecores.org
Cc: uclinux-h8-devel@lists.sourceforge.jp
Link: https://lkml.kernel.org/r/20190322143008.21313-2-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
[kdrag0n: Dropped conflicting architecture changes]
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Currently, when a reader acquires a lock, it only sets the
RWSEM_READER_OWNED bit in the owner field. The other bits are simply
not used. When debugging hanging cases involving rwsems and readers,
the owner value does not provide much useful information at all.
This patch modifies the current behavior to always store the task_struct
pointer of the last rwsem-acquiring reader in a reader-owned rwsem. This
may be useful in debugging rwsem hanging cases especially if only one
reader is involved. However, the task in the owner field may not the
real owner or one of the real owners at all when the owner value is
examined, for example, in a crash dump. So it is just an additional
hint about the past history.
If CONFIG_DEBUG_RWSEMS=y is enabled, the owner field will be checked at
unlock time too to make sure the task pointer value is valid. That does
have a slight performance cost and so is only enabled as part of that
debug option.
From the performance point of view, it is expected that the changes
shouldn't have any noticeable performance impact. A rwsem microbenchmark
(with 48 worker threads and 1:1 reader/writer ratio) was ran on a
2-socket 24-core 48-thread Haswell system. The locking rates on a
4.19-rc1 based kernel were as follows:
1) Unpatched kernel: 543.3 kops/s
2) Patched kernel: 549.2 kops/s
3) Patched kernel (CONFIG_DEBUG_RWSEMS on): 546.6 kops/s
There was actually a slight increase in performance (1.1%) in this
particular case. Maybe it was caused by the elimination of a branch or
just a testing noise. Turning on the CONFIG_DEBUG_RWSEMS option also
had less than the expected impact on performance.
The least significant 2 bits of the owner value are now used to designate
the rwsem is readers owned and the owners are anonymous.
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/1536265114-10842-1-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
It was found that the use of up_read_non_owner() in NFS was causing
the following warning when DEBUG_RWSEMS was configured.
DEBUG_LOCKS_WARN_ON(sem->owner != ((struct task_struct *)(1UL << 0)))
Looking into the rwsem.c file, it was discovered that the corresponding
down_read_non_owner() function was not setting the owner field properly.
This is fixed now, and the warning should be gone.
Fixes: 5149cbac4235 ("locking/rwsem: Add DEBUG_RWSEMS to look for lock/unlock mismatches")
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Gavin Schenk <g.schenk@eckelmann.de>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: linux-nfs@vger.kernel.org
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/1527168398-4291-1-git-send-email-longman@redhat.com
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
For a rwsem, locking can either be exclusive or shared. The corresponding
exclusive or shared unlock must be used. Otherwise, the protected data
structures may get corrupted or the lock may be in an inconsistent state.
In order to detect such anomaly, a new configuration option DEBUG_RWSEMS
is added which can be enabled to look for such mismatches and print
warnings that that happens.
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Davidlohr Bueso <dave@stgolabs.net>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1522445280-7767-2-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
With the > 4 nesting levels case handled by the commit:
d682b596d993 ("locking/qspinlock: Handle > 4 slowpath nesting levels")
the BUG_ON() call in encode_tail() will never actually be triggered.
Remove it.
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/1551057253-3231-1-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Track the number of slowpath locking operations that are being done
without any MCS node available as well renaming lock_index[123] to make
them more descriptive.
Using these stat counters is one way to find out if a code path is
being exercised.
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: James Morse <james.morse@arm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: SRINIVAS <srinivas.eeda@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Zhenzhong Duan <zhenzhong.duan@oracle.com>
Link: https://lkml.kernel.org/r/1548798828-16156-3-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Four queue nodes per CPU are allocated to enable up to 4 nesting levels
using the per-CPU nodes. Nested NMIs are possible in some architectures.
Still it is very unlikely that we will ever hit more than 4 nested
levels with contention in the slowpath.
When that rare condition happens, however, it is likely that the system
will hang or crash shortly after that. It is not good and we need to
handle this exception case.
This is done by spinning directly on the lock using repeated trylock.
This alternative code path should only be used when there is nested
NMIs. Assuming that the locks used by those NMI handlers will not be
heavily contended, a simple TAS locking should work out.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: James Morse <james.morse@arm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: SRINIVAS <srinivas.eeda@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Zhenzhong Duan <zhenzhong.duan@oracle.com>
Link: https://lkml.kernel.org/r/1548798828-16156-2-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
The qspinlock code supports up to 4 levels of slowpath nesting using
four per-CPU mcs_spinlock structures. For 64-bit architectures, they
fit nicely in one 64-byte cacheline.
For para-virtualized (PV) qspinlocks it needs to store more information
in the per-CPU node structure than there is space for. It uses a trick
to use a second cacheline to hold the extra information that it needs.
So PV qspinlock needs to access two extra cachelines for its information
whereas the native qspinlock code only needs one extra cacheline.
Freshly added counter profiling of the qspinlock code, however, revealed
that it was very rare to use more than two levels of slowpath nesting.
So it doesn't make sense to penalize PV qspinlock code in order to have
four mcs_spinlock structures in the same cacheline to optimize for a case
in the native qspinlock code that rarely happens.
Extend the per-CPU node structure to have two more long words when PV
qspinlock locks are configured to hold the extra data that it needs.
As a result, the PV qspinlock code will enjoy the same benefit of using
just one extra cacheline like the native counterpart, for most cases.
[ mingo: Minor changelog edits. ]
Signed-off-by: Waiman Long <longman@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/1539697507-28084-2-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Both spin locks and write locks currently do:
f0 0f b1 17 lock cmpxchg %edx,(%rdi)
85 c0 test %eax,%eax
75 05 jne [slowpath]
This 'test' insn is superfluous; the cmpxchg insn sets the Z flag
appropriately. Peter pointed out that using atomic_try_cmpxchg_acquire()
will let the compiler know this is true. Comparing before/after
disassemblies show the only effect is to remove this insn.
Take this opportunity to make the spin & write lock code resemble each
other more closely and have similar likely() hints.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Link: http://lkml.kernel.org/r/20180820162639.GC25153@bombadil.infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Removes "#ifndef queued_spin_is_locked" from the generic code: this is
unused and it's reasonable to conclude that it will continue to be unused.
Also removes the comment about spin_is_locked() from mutex_is_locked():
the comment remains valid but not particularly useful.
Suggested-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrea Parri <andrea.parri@amarulasolutions.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: akiyks@gmail.com
Cc: boqun.feng@gmail.com
Cc: dhowells@redhat.com
Cc: j.alglave@ucl.ac.uk
Cc: linux-arch@vger.kernel.org
Cc: luc.maranget@inria.fr
Cc: npiggin@gmail.com
Cc: parri.andrea@gmail.com
Cc: stern@rowland.harvard.edu
Link: http://lkml.kernel.org/r/1526338889-7003-3-git-send-email-paulmck@linux.vnet.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
A qspinlock can be unlocked simply by writing zero to the locked byte.
This can be implemented in the generic code, so do that and remove the
arch-specific override for x86 in the !PV case.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-11-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
[kdrag0n: Stripped x86 part due to conflicts]
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Queued spinlock supports up to 4 levels of lock slowpath nesting -
user context, soft IRQ, hard IRQ and NMI. However, we are not sure how
often the nesting happens.
So add 3 more per-CPU stat counters to track the number of instances where
nesting index goes to 1, 2 and 3 respectively.
On a dual-socket 64-core 128-thread Zen server, the following were the
new stat counter values under different circumstances:
State slowpath index1 index2 index3
----- -------- ------ ------ -------
After bootup 1,012,150 82 0 0
After parallel build + perf-top 125,195,009 82 0 0
So the chance of having more than 2 levels of nesting is extremely low.
[ mingo: Minor changelog edits. ]
Signed-off-by: Waiman Long <longman@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/1539697507-28084-1-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
commit 7aa54be2976550f17c11a1c3e3630002dea39303 upstream.
On x86 we cannot do fetch_or() with a single instruction and thus end up
using a cmpxchg loop, this reduces determinism. Replace the fetch_or()
with a composite operation: tas-pending + load.
Using two instructions of course opens a window we previously did not
have. Consider the scenario:
CPU0 CPU1 CPU2
1) lock
trylock -> (0,0,1)
2) lock
trylock /* fail */
3) unlock -> (0,0,0)
4) lock
trylock -> (0,0,1)
5) tas-pending -> (0,1,1)
load-val <- (0,1,0) from 3
6) clear-pending-set-locked -> (0,0,1)
FAIL: _2_ owners
where 5) is our new composite operation. When we consider each part of
the qspinlock state as a separate variable (as we can when
_Q_PENDING_BITS == 8) then the above is entirely possible, because
tas-pending will only RmW the pending byte, so the later load is able
to observe prior tail and lock state (but not earlier than its own
trylock, which operates on the whole word, due to coherence).
To avoid this we need 2 things:
- the load must come after the tas-pending (obviously, otherwise it
can trivially observe prior state).
- the tas-pending must be a full word RmW instruction, it cannot be an XCHGB for
example, such that we cannot observe other state prior to setting
pending.
On x86 we can realize this by using "LOCK BTS m32, r32" for
tas-pending followed by a regular load.
Note that observing later state is not a problem:
- if we fail to observe a later unlock, we'll simply spin-wait for
that store to become visible.
- if we observe a later xchg_tail(), there is no difference from that
xchg_tail() having taken place before the tas-pending.
Suggested-by: Will Deacon <will.deacon@arm.com>
Reported-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: andrea.parri@amarulasolutions.com
Cc: longman@redhat.com
Fixes: 59fb586b4a07 ("locking/qspinlock: Remove unbounded cmpxchg() loop from locking slowpath")
Link: https://lkml.kernel.org/r/20181003130957.183726335@infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
[bigeasy: GEN_BINARY_RMWcc macro redo]
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
While working my way through the code again; I felt the comments could
use help.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: andrea.parri@amarulasolutions.com
Cc: longman@redhat.com
Link: https://lkml.kernel.org/r/20181003130257.156322446@infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Flip the branch condition after atomic_fetch_or_acquire(_Q_PENDING_VAL)
such that we loose the indent. This also result in a more natural code
flow IMO.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: andrea.parri@amarulasolutions.com
Cc: longman@redhat.com
Link: https://lkml.kernel.org/r/20181003130257.156322446@infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Currently, the qspinlock_stat code tracks only statistical counts in the
PV qspinlock code. However, it may also be useful to track the number
of locking operations done via the pending code vs. the MCS lock queue
slowpath for the non-PV case.
The qspinlock stat code is modified to do that. The stat counter
pv_lock_slowpath is renamed to lock_slowpath so that it can be used by
both the PV and non-PV cases.
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Cc: will.deacon@arm.com
Link: http://lkml.kernel.org/r/1524738868-31318-14-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
When reaching the head of an uncontended queue on the qspinlock slow-path,
using a try_cmpxchg() instead of a cmpxchg() operation to transition the
lock work to _Q_LOCKED_VAL generates slightly better code for x86 and
pretty much identical code for arm64.
Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-13-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
The qspinlock slowpath must ensure that the MCS node is fully initialised
before it can be reached by another other CPU. This is currently enforced
by using a RELEASE operation when updating the tail and also when linking
the node into the waitqueue, since the control dependency off xchg_tail
is insufficient to enforce sufficient ordering, see:
95bcade33a8a ("locking/qspinlock: Ensure node is initialised before updating prev->next")
Back-to-back RELEASE operations may be expensive on some architectures,
particularly those that implement them using fences under the hood. We
can replace the two RELEASE operations with a single smp_wmb() fence and
use RELAXED operations for the subsequent publishing of the node.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-12-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
When a locker reaches the head of the queue and takes the lock, a
concurrent locker may enqueue and force the lock holder to spin
whilst its node->next field is initialised. Rather than open-code
a READ_ONCE/cpu_relax() loop, this can be implemented using
smp_cond_load_relaxed() instead.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-10-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Rather than dig into the counter field of the atomic_t inside the
qspinlock structure so that we can call smp_cond_load_acquire(), use
atomic_cond_read_acquire() instead, which operates on the atomic_t
directly.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-8-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
It's fair to say that our ticket lock has served us well over time, but
it's time to bite the bullet and start using the generic qspinlock code
so we can make use of explicit MCS queuing and potentially better PV
performance in future.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
We can provide an implementation of smp_cond_load_relaxed using READ_ONCE
and __cmpwait_relaxed.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: kdrag0n <dragon@khronodragon.com>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
The following commit:
38b850a73034f ("arm64: spinlock: order spin_{is_locked,unlock_wait} against local locks")
... added an smp_mb() to arch_spin_is_locked(), in order
"to ensure that the lock value is always loaded after any other locks have
been taken by the current CPU", and reported one example (the "insane case"
in ipc/sem.c) relying on such guarantee.
It is however understood that spin_is_locked() is not required to provide
such an ordering guarantee (a guarantee that is currently not provided by
all the implementations/archs), and that callers relying on such ordering
should instead insert suitable memory barriers before acting on the result
of spin_is_locked().
Following a recent auditing [1] of the callers of {,raw_}spin_is_locked(),
revealing that none of them are relying on the ordering guarantee anymore,
this commit removes the leading smp_mb() from the primitive thus reverting
38b850a73034f.
[1] https://marc.info/?l=linux-kernel&m=151981440005264&w=2https://marc.info/?l=linux-kernel&m=152042843808540&w=2https://marc.info/?l=linux-kernel&m=152043346110262&w=2
Signed-off-by: Andrea Parri <andrea.parri@amarulasolutions.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: akiyks@gmail.com
Cc: boqun.feng@gmail.com
Cc: dhowells@redhat.com
Cc: j.alglave@ucl.ac.uk
Cc: linux-arch@vger.kernel.org
Cc: luc.maranget@inria.fr
Cc: npiggin@gmail.com
Cc: parri.andrea@gmail.com
Cc: stern@rowland.harvard.edu
Link: http://lkml.kernel.org/r/1526338889-7003-2-git-send-email-paulmck@linux.vnet.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: kdrag0n <dragon@khronodragon.com>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
The arch_{read,spin,write}_lock_flags() macros are simply mapped to the
non-flags versions by the majority of architectures, so do this in core
code and remove the dummy implementations. Also remove the implementation
in spinlock_up.h, since all callers of do_raw_spin_lock_flags() call
local_irq_save(flags) anyway.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1507055129-12300-4-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
arch_{read,spin,write}_relax() are defined as cpu_relax() by the core
code, so architectures that can't do better (i.e. most of them) don't
need to bother with the dummy definitions.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1507055129-12300-3-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
[kdrag0n: dropped changes to arch/sparc to due to conflicts]
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
The IRQ subsystem allocates a domain name dynamically on its own. By
overwriting it, there is not only a memory leak created, but also a
kfree() on a non-allocated pointer when the IRQ is freed. We can remedy
all of this by simply not overwriting the IRQ domain name.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Signed-off-by: UtsavBalar1231 <utsavbalar1231@gmail.com>
Write back caching is a popular technique which copies data to higher
level caches, backing store or memory.
As per an article in Techopedia [1], enabling Write-Back Caching improves
or reduces write operations to and from a cache and it's data source.
[1] - https://www.techopedia.com/definition/1138/write-back-cache
Change-Id: I0140736d3f921d5e0275f6154219009f84b6cb8b
Signed-off-by: Panchajanya1999 <panchajanya@azure-dev.live>
Signed-off-by: Adithya R <gh0strider.2k18.reborn@gmail.com>
move messages emitting out of sch_tree_lock to avoid holding
this lock too long.
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Yaroslav Furman <yaro330@gmail.com>
This tames quite a bit of the log spam and makes dmesg readable.
Uses work from Danny Lin <danny@kdrag0n.dev>.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Free up some space for tracking inflight and losses for each
bw sample, in upcoming commits.
These timestamps are in microseconds, and are now stored in 32
bits. So they can only hold time intervals up to roughly 2^12 = 4096
seconds. But Linux TCP RTT and RTO tracking has the same 32-bit
microsecond implementation approach and resulting deployment
limitations. So this is not introducing a new limit. And these should
not be a limitation for the foreseeable future.
Effort: net-tcp_bbr
Origin-9xx-SHA1: 238a7e6b5d51625fef1ce7769826a7b21b02ae55
Change-Id: I3b779603797263b52a61ad57c565eb91fe42680c
(cherry-picked from 6cabc5bdb5)
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
In order to track CE marks per rate sample (one round trip), we'll
need to snap the starting tcp delivered_ce acount in the packet
meta header (tcp_skb_cb). But there's not enough space.
Good news is that the "last_in_flight" in the header, used by
NV congestion control, is almost equivalent as "delivered". In
fact "delivered" is better by accounting out-of-order packets
additionally. Therefore we can remove it to make room for the
CE tracking.
This would make delayed ACK detection slightly less accurate but the
impact is negligible since it's not used for any critical control.
Effort: net-tcp_rate
Origin-9xx-SHA1: ddcd46ec85d5f1c4454258af0c54b3254c0d64a7
Change-Id: I1a184aad6d101c981ac7f2f275aa9417ff856910
(cherry-picked from 3de61ba5b8)
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
This commit is a bug fix for the Linux TCP app-limited logic
used for collecting rate (bandwidth) samples.
Previously the app-limited logic only looked for "bubbles" of
silence in between application writes, by checking at the start
of each sendmsg. But "bubbles" of silence can also happen before
retransmits: e.g. bubbles can happen between an application write
and a retransmit, or between two retransmits.
Retransmits are triggered by ACKs or timers. So this commit checks
for bubbles of app-limited silence upon ACKs or timers.
Why does this commit check for app-limited state at the start of
ACKs and timer handling? Because at that point we know whether
inflight was fully using the cwnd. During processing the ACK or
timer event we often change the cwnd; after changing the cwnd we
can't know whether inflight was fully using the old cwnd.
Origin-9xx-SHA1: 3fe9b53291e018407780fb8c356adb5666722cbc
Change-Id: I37221506f5166877c2b110753d39bb0757985e68
(cherry-picked from 80d039a0a0)
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Aggregation effects are extremely common with wifi, cellular, and cable
modem link technologies, ACK decimation in middleboxes, and LRO and GRO
in receiving hosts. The aggregation can happen in either direction,
data or ACKs, but in either case the aggregation effect is visible
to the sender in the ACK stream.
Previously BBR's sending was often limited by cwnd under severe ACK
aggregation/decimation because BBR sized the cwnd at 2*BDP. If packets
were acked in bursts after long delays (e.g. one ACK acking 5*BDP after
5*RTT), BBR's sending was halted after sending 2*BDP over 2*RTT, leaving
the bottleneck idle for potentially long periods. Note that loss-based
congestion control does not have this issue because when facing
aggregation it continues increasing cwnd after bursts of ACKs, growing
cwnd until the buffer is full.
To achieve good throughput in the presence of aggregation effects, this
algorithm allows the BBR sender to put extra data in flight to keep the
bottleneck utilized during silences in the ACK stream that it has evidence
to suggest were caused by aggregation.
A summary of the algorithm: when a burst of packets are acked by a
stretched ACK or a burst of ACKs or both, BBR first estimates the expected
amount of data that should have been acked, based on its estimated
bandwidth. Then the surplus ("extra_acked") is recorded in a windowed-max
filter to estimate the recent level of observed ACK aggregation. Then cwnd
is increased by the ACK aggregation estimate. The larger cwnd avoids BBR
being cwnd-limited in the face of ACK silences that recent history suggests
were caused by aggregation. As a sanity check, the ACK aggregation degree
is upper-bounded by the cwnd (at the time of measurement) and a global max
of BW * 100ms. The algorithm is further described by the following
presentation:
https://datatracker.ietf.org/meeting/101/materials/slides-101-iccrg-an-update-on-bbr-work-at-google-00
In our internal testing, we observed a significant increase in BBR
throughput (measured using netperf), in a basic wifi setup.
- Host1 (sender on ethernet) -> AP -> Host2 (receiver on wifi)
- 2.4 GHz -> BBR before: ~73 Mbps; BBR after: ~102 Mbps; CUBIC: ~100 Mbps
- 5.0 GHz -> BBR before: ~362 Mbps; BBR after: ~593 Mbps; CUBIC: ~601 Mbps
Also, this code is running globally on YouTube TCP connections and produced
significant bandwidth increases for YouTube traffic.
This is based on Ian Swett's max_ack_height_ algorithm from the
QUIC BBR implementation.
Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Because bbr_target_cwnd() is really a general-purpose BBR helper for
computing some volume of inflight data as a function of the estimated
BDP, refactor it into following helper functions:
- bbr_bdp()
- bbr_quantization_budget()
- bbr_inflight()
Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Centralize the code that sets gains used for computing cwnd and pacing
rate. This simplifies the code and makes it easier to change the state
machine or (in the future) dynamically change the gain values and
ensure that the correct gain values are always used.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
This commit fixes a corner case where TCP BBR would enter PROBE_RTT
mode but not reduce its cwnd. If a TCP receiver ACKed less than one
full segment, the number of delivered/acked packets was 0, so that
bbr_set_cwnd() would short-circuit and exit early, without cutting
cwnd to the value we want for PROBE_RTT.
The fix is to instead make sure that even when 0 full packets are
ACKed, we do apply all the appropriate caps, including the cap that
applies in PROBE_RTT mode.
Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Kevin Yang <yyd@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
This patch fix the case where BBR does not exit PROBE_RTT mode when
it restarts from idle. When BBR restarts from idle and if BBR is in
PROBE_RTT mode, BBR should check if it's time to exit PROBE_RTT. If
yes, then BBR should exit PROBE_RTT mode and restore the cwnd to its
full value.
Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Kevin Yang <yyd@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
This patch add a helper function bbr_check_probe_rtt_done() to
1. check the condition to see if bbr should exit probe_rtt mode;
2. process the logic of exiting probe_rtt mode.
Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Kevin Yang <yyd@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Danny Lin <danny@kdrag0n.dev>
