diff options
Diffstat (limited to 'debian/patches-rt/0173-locking-rtmutex-add-rwsem-implementation-based-on-rt.patch')
| -rw-r--r-- | debian/patches-rt/0173-locking-rtmutex-add-rwsem-implementation-based-on-rt.patch | 455 |
1 files changed, 455 insertions, 0 deletions
diff --git a/debian/patches-rt/0173-locking-rtmutex-add-rwsem-implementation-based-on-rt.patch b/debian/patches-rt/0173-locking-rtmutex-add-rwsem-implementation-based-on-rt.patch new file mode 100644 index 000000000..d1595ab45 --- /dev/null +++ b/debian/patches-rt/0173-locking-rtmutex-add-rwsem-implementation-based-on-rt.patch @@ -0,0 +1,455 @@ +From 9095a37abced402fb5197fef8d126069f63ebfb4 Mon Sep 17 00:00:00 2001 +From: Thomas Gleixner <tglx@linutronix.de> +Date: Thu, 12 Oct 2017 17:28:34 +0200 +Subject: [PATCH 173/323] locking/rtmutex: add rwsem implementation based on + rtmutex +Origin: https://www.kernel.org/pub/linux/kernel/projects/rt/5.10/older/patches-5.10.204-rt100.tar.xz + +The RT specific R/W semaphore implementation restricts the number of readers +to one because a writer cannot block on multiple readers and inherit its +priority or budget. + +The single reader restricting is painful in various ways: + + - Performance bottleneck for multi-threaded applications in the page fault + path (mmap sem) + + - Progress blocker for drivers which are carefully crafted to avoid the + potential reader/writer deadlock in mainline. + +The analysis of the writer code paths shows, that properly written RT tasks +should not take them. Syscalls like mmap(), file access which take mmap sem +write locked have unbound latencies which are completely unrelated to mmap +sem. Other R/W sem users like graphics drivers are not suitable for RT tasks +either. + +So there is little risk to hurt RT tasks when the RT rwsem implementation is +changed in the following way: + + - Allow concurrent readers + + - Make writers block until the last reader left the critical section. This + blocking is not subject to priority/budget inheritance. + + - Readers blocked on a writer inherit their priority/budget in the normal + way. + +There is a drawback with this scheme. R/W semaphores become writer unfair +though the applications which have triggered writer starvation (mostly on +mmap_sem) in the past are not really the typical workloads running on a RT +system. So while it's unlikely to hit writer starvation, it's possible. If +there are unexpected workloads on RT systems triggering it, we need to rethink +the approach. + +Signed-off-by: Thomas Gleixner <tglx@linutronix.de> +Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> +--- + include/linux/rwsem-rt.h | 70 +++++++++ + kernel/locking/rwsem-rt.c | 318 ++++++++++++++++++++++++++++++++++++++ + 2 files changed, 388 insertions(+) + create mode 100644 include/linux/rwsem-rt.h + create mode 100644 kernel/locking/rwsem-rt.c + +diff --git a/include/linux/rwsem-rt.h b/include/linux/rwsem-rt.h +new file mode 100644 +index 000000000000..0ba8aae9a198 +--- /dev/null ++++ b/include/linux/rwsem-rt.h +@@ -0,0 +1,70 @@ ++// SPDX-License-Identifier: GPL-2.0-only ++#ifndef _LINUX_RWSEM_RT_H ++#define _LINUX_RWSEM_RT_H ++ ++#ifndef _LINUX_RWSEM_H ++#error "Include rwsem.h" ++#endif ++ ++#include <linux/rtmutex.h> ++#include <linux/swait.h> ++ ++#define READER_BIAS (1U << 31) ++#define WRITER_BIAS (1U << 30) ++ ++struct rw_semaphore { ++ atomic_t readers; ++ struct rt_mutex rtmutex; ++#ifdef CONFIG_DEBUG_LOCK_ALLOC ++ struct lockdep_map dep_map; ++#endif ++}; ++ ++#define __RWSEM_INITIALIZER(name) \ ++{ \ ++ .readers = ATOMIC_INIT(READER_BIAS), \ ++ .rtmutex = __RT_MUTEX_INITIALIZER(name.rtmutex), \ ++ RW_DEP_MAP_INIT(name) \ ++} ++ ++#define DECLARE_RWSEM(lockname) \ ++ struct rw_semaphore lockname = __RWSEM_INITIALIZER(lockname) ++ ++extern void __rwsem_init(struct rw_semaphore *rwsem, const char *name, ++ struct lock_class_key *key); ++ ++#define __init_rwsem(sem, name, key) \ ++do { \ ++ rt_mutex_init(&(sem)->rtmutex); \ ++ __rwsem_init((sem), (name), (key)); \ ++} while (0) ++ ++#define init_rwsem(sem) \ ++do { \ ++ static struct lock_class_key __key; \ ++ \ ++ __init_rwsem((sem), #sem, &__key); \ ++} while (0) ++ ++static inline int rwsem_is_locked(struct rw_semaphore *sem) ++{ ++ return atomic_read(&sem->readers) != READER_BIAS; ++} ++ ++static inline int rwsem_is_contended(struct rw_semaphore *sem) ++{ ++ return atomic_read(&sem->readers) > 0; ++} ++ ++extern void __down_read(struct rw_semaphore *sem); ++extern int __down_read_interruptible(struct rw_semaphore *sem); ++extern int __down_read_killable(struct rw_semaphore *sem); ++extern int __down_read_trylock(struct rw_semaphore *sem); ++extern void __down_write(struct rw_semaphore *sem); ++extern int __must_check __down_write_killable(struct rw_semaphore *sem); ++extern int __down_write_trylock(struct rw_semaphore *sem); ++extern void __up_read(struct rw_semaphore *sem); ++extern void __up_write(struct rw_semaphore *sem); ++extern void __downgrade_write(struct rw_semaphore *sem); ++ ++#endif +diff --git a/kernel/locking/rwsem-rt.c b/kernel/locking/rwsem-rt.c +new file mode 100644 +index 000000000000..a0771c150041 +--- /dev/null ++++ b/kernel/locking/rwsem-rt.c +@@ -0,0 +1,318 @@ ++// SPDX-License-Identifier: GPL-2.0-only ++#include <linux/rwsem.h> ++#include <linux/sched/debug.h> ++#include <linux/sched/signal.h> ++#include <linux/export.h> ++#include <linux/blkdev.h> ++ ++#include "rtmutex_common.h" ++ ++/* ++ * RT-specific reader/writer semaphores ++ * ++ * down_write() ++ * 1) Lock sem->rtmutex ++ * 2) Remove the reader BIAS to force readers into the slow path ++ * 3) Wait until all readers have left the critical region ++ * 4) Mark it write locked ++ * ++ * up_write() ++ * 1) Remove the write locked marker ++ * 2) Set the reader BIAS so readers can use the fast path again ++ * 3) Unlock sem->rtmutex to release blocked readers ++ * ++ * down_read() ++ * 1) Try fast path acquisition (reader BIAS is set) ++ * 2) Take sem->rtmutex.wait_lock which protects the writelocked flag ++ * 3) If !writelocked, acquire it for read ++ * 4) If writelocked, block on sem->rtmutex ++ * 5) unlock sem->rtmutex, goto 1) ++ * ++ * up_read() ++ * 1) Try fast path release (reader count != 1) ++ * 2) Wake the writer waiting in down_write()#3 ++ * ++ * down_read()#3 has the consequence, that rw semaphores on RT are not writer ++ * fair, but writers, which should be avoided in RT tasks (think mmap_sem), ++ * are subject to the rtmutex priority/DL inheritance mechanism. ++ * ++ * It's possible to make the rw semaphores writer fair by keeping a list of ++ * active readers. A blocked writer would force all newly incoming readers to ++ * block on the rtmutex, but the rtmutex would have to be proxy locked for one ++ * reader after the other. We can't use multi-reader inheritance because there ++ * is no way to support that with SCHED_DEADLINE. Implementing the one by one ++ * reader boosting/handover mechanism is a major surgery for a very dubious ++ * value. ++ * ++ * The risk of writer starvation is there, but the pathological use cases ++ * which trigger it are not necessarily the typical RT workloads. ++ */ ++ ++void __rwsem_init(struct rw_semaphore *sem, const char *name, ++ struct lock_class_key *key) ++{ ++#ifdef CONFIG_DEBUG_LOCK_ALLOC ++ /* ++ * Make sure we are not reinitializing a held semaphore: ++ */ ++ debug_check_no_locks_freed((void *)sem, sizeof(*sem)); ++ lockdep_init_map(&sem->dep_map, name, key, 0); ++#endif ++ atomic_set(&sem->readers, READER_BIAS); ++} ++EXPORT_SYMBOL(__rwsem_init); ++ ++int __down_read_trylock(struct rw_semaphore *sem) ++{ ++ int r, old; ++ ++ /* ++ * Increment reader count, if sem->readers < 0, i.e. READER_BIAS is ++ * set. ++ */ ++ for (r = atomic_read(&sem->readers); r < 0;) { ++ old = atomic_cmpxchg(&sem->readers, r, r + 1); ++ if (likely(old == r)) ++ return 1; ++ r = old; ++ } ++ return 0; ++} ++ ++static int __sched __down_read_common(struct rw_semaphore *sem, int state) ++{ ++ struct rt_mutex *m = &sem->rtmutex; ++ struct rt_mutex_waiter waiter; ++ int ret; ++ ++ if (__down_read_trylock(sem)) ++ return 0; ++ ++ /* ++ * Flush blk before ->pi_blocked_on is set. At schedule() time it is too ++ * late if one of the callbacks needs to acquire a sleeping lock. ++ */ ++ if (blk_needs_flush_plug(current)) ++ blk_schedule_flush_plug(current); ++ ++ might_sleep(); ++ raw_spin_lock_irq(&m->wait_lock); ++ /* ++ * Allow readers as long as the writer has not completely ++ * acquired the semaphore for write. ++ */ ++ if (atomic_read(&sem->readers) != WRITER_BIAS) { ++ atomic_inc(&sem->readers); ++ raw_spin_unlock_irq(&m->wait_lock); ++ return 0; ++ } ++ ++ /* ++ * Call into the slow lock path with the rtmutex->wait_lock ++ * held, so this can't result in the following race: ++ * ++ * Reader1 Reader2 Writer ++ * down_read() ++ * down_write() ++ * rtmutex_lock(m) ++ * swait() ++ * down_read() ++ * unlock(m->wait_lock) ++ * up_read() ++ * swake() ++ * lock(m->wait_lock) ++ * sem->writelocked=true ++ * unlock(m->wait_lock) ++ * ++ * up_write() ++ * sem->writelocked=false ++ * rtmutex_unlock(m) ++ * down_read() ++ * down_write() ++ * rtmutex_lock(m) ++ * swait() ++ * rtmutex_lock(m) ++ * ++ * That would put Reader1 behind the writer waiting on ++ * Reader2 to call up_read() which might be unbound. ++ */ ++ rt_mutex_init_waiter(&waiter, false); ++ ret = rt_mutex_slowlock_locked(m, state, NULL, RT_MUTEX_MIN_CHAINWALK, ++ &waiter); ++ /* ++ * The slowlock() above is guaranteed to return with the rtmutex (for ++ * ret = 0) is now held, so there can't be a writer active. Increment ++ * the reader count and immediately drop the rtmutex again. ++ * For ret != 0 we don't hold the rtmutex and need unlock the wait_lock. ++ * We don't own the lock then. ++ */ ++ if (!ret) ++ atomic_inc(&sem->readers); ++ raw_spin_unlock_irq(&m->wait_lock); ++ if (!ret) ++ __rt_mutex_unlock(m); ++ ++ debug_rt_mutex_free_waiter(&waiter); ++ return ret; ++} ++ ++void __down_read(struct rw_semaphore *sem) ++{ ++ int ret; ++ ++ ret = __down_read_common(sem, TASK_UNINTERRUPTIBLE); ++ WARN_ON_ONCE(ret); ++} ++ ++int __down_read_interruptible(struct rw_semaphore *sem) ++{ ++ int ret; ++ ++ ret = __down_read_common(sem, TASK_INTERRUPTIBLE); ++ if (likely(!ret)) ++ return ret; ++ WARN_ONCE(ret != -EINTR, "Unexpected state: %d\n", ret); ++ return -EINTR; ++} ++ ++int __down_read_killable(struct rw_semaphore *sem) ++{ ++ int ret; ++ ++ ret = __down_read_common(sem, TASK_KILLABLE); ++ if (likely(!ret)) ++ return ret; ++ WARN_ONCE(ret != -EINTR, "Unexpected state: %d\n", ret); ++ return -EINTR; ++} ++ ++void __up_read(struct rw_semaphore *sem) ++{ ++ struct rt_mutex *m = &sem->rtmutex; ++ struct task_struct *tsk; ++ ++ /* ++ * sem->readers can only hit 0 when a writer is waiting for the ++ * active readers to leave the critical region. ++ */ ++ if (!atomic_dec_and_test(&sem->readers)) ++ return; ++ ++ might_sleep(); ++ raw_spin_lock_irq(&m->wait_lock); ++ /* ++ * Wake the writer, i.e. the rtmutex owner. It might release the ++ * rtmutex concurrently in the fast path (due to a signal), but to ++ * clean up the rwsem it needs to acquire m->wait_lock. The worst ++ * case which can happen is a spurious wakeup. ++ */ ++ tsk = rt_mutex_owner(m); ++ if (tsk) ++ wake_up_process(tsk); ++ ++ raw_spin_unlock_irq(&m->wait_lock); ++} ++ ++static void __up_write_unlock(struct rw_semaphore *sem, int bias, ++ unsigned long flags) ++{ ++ struct rt_mutex *m = &sem->rtmutex; ++ ++ atomic_add(READER_BIAS - bias, &sem->readers); ++ raw_spin_unlock_irqrestore(&m->wait_lock, flags); ++ __rt_mutex_unlock(m); ++} ++ ++static int __sched __down_write_common(struct rw_semaphore *sem, int state) ++{ ++ struct rt_mutex *m = &sem->rtmutex; ++ unsigned long flags; ++ ++ /* ++ * Flush blk before ->pi_blocked_on is set. At schedule() time it is too ++ * late if one of the callbacks needs to acquire a sleeping lock. ++ */ ++ if (blk_needs_flush_plug(current)) ++ blk_schedule_flush_plug(current); ++ ++ /* Take the rtmutex as a first step */ ++ if (__rt_mutex_lock_state(m, state)) ++ return -EINTR; ++ ++ /* Force readers into slow path */ ++ atomic_sub(READER_BIAS, &sem->readers); ++ might_sleep(); ++ ++ set_current_state(state); ++ for (;;) { ++ raw_spin_lock_irqsave(&m->wait_lock, flags); ++ /* Have all readers left the critical region? */ ++ if (!atomic_read(&sem->readers)) { ++ atomic_set(&sem->readers, WRITER_BIAS); ++ __set_current_state(TASK_RUNNING); ++ raw_spin_unlock_irqrestore(&m->wait_lock, flags); ++ return 0; ++ } ++ ++ if (signal_pending_state(state, current)) { ++ __set_current_state(TASK_RUNNING); ++ __up_write_unlock(sem, 0, flags); ++ return -EINTR; ++ } ++ raw_spin_unlock_irqrestore(&m->wait_lock, flags); ++ ++ if (atomic_read(&sem->readers) != 0) { ++ schedule(); ++ set_current_state(state); ++ } ++ } ++} ++ ++void __sched __down_write(struct rw_semaphore *sem) ++{ ++ __down_write_common(sem, TASK_UNINTERRUPTIBLE); ++} ++ ++int __sched __down_write_killable(struct rw_semaphore *sem) ++{ ++ return __down_write_common(sem, TASK_KILLABLE); ++} ++ ++int __down_write_trylock(struct rw_semaphore *sem) ++{ ++ struct rt_mutex *m = &sem->rtmutex; ++ unsigned long flags; ++ ++ if (!__rt_mutex_trylock(m)) ++ return 0; ++ ++ atomic_sub(READER_BIAS, &sem->readers); ++ ++ raw_spin_lock_irqsave(&m->wait_lock, flags); ++ if (!atomic_read(&sem->readers)) { ++ atomic_set(&sem->readers, WRITER_BIAS); ++ raw_spin_unlock_irqrestore(&m->wait_lock, flags); ++ return 1; ++ } ++ __up_write_unlock(sem, 0, flags); ++ return 0; ++} ++ ++void __up_write(struct rw_semaphore *sem) ++{ ++ struct rt_mutex *m = &sem->rtmutex; ++ unsigned long flags; ++ ++ raw_spin_lock_irqsave(&m->wait_lock, flags); ++ __up_write_unlock(sem, WRITER_BIAS, flags); ++} ++ ++void __downgrade_write(struct rw_semaphore *sem) ++{ ++ struct rt_mutex *m = &sem->rtmutex; ++ unsigned long flags; ++ ++ raw_spin_lock_irqsave(&m->wait_lock, flags); ++ /* Release it and account current as reader */ ++ __up_write_unlock(sem, WRITER_BIAS - 1, flags); ++} +-- +2.43.0 + |