Concurrency and Race
Conditions
Linux Kernel Programming
CIS 4930/COP 5641
MOTIVATION:
EXAMPLE PITFALL IN SCULL
Pitfalls in scull
 Race condition: result of uncontrolled access
to shared data
if (!dptr->data[s_pos]) {
dptr->data[s_pos] = kzalloc(quantum, GFP_KERNEL);
if (!dptr->data[s_pos]) {
goto out;
}
}
Pitfalls in scull
 Race condition: result of uncontrolled access
to shared data
if (!dptr->data[s_pos]) {
dptr->data[s_pos] = kzalloc(quantum, GFP_KERNEL);
if (!dptr->data[s_pos]) {
goto out;
}
}
Pitfalls in scull
 Race condition: result of uncontrolled access
to shared data
if (!dptr->data[s_pos]) {
dptr->data[s_pos] = kmalloc(quantum, GFP_KERNEL);
if (!dptr->data[s_pos]) {
goto out;
}
}
Pitfalls in scull
 Race condition: result of uncontrolled access
to shared data
if (!dptr->data[s_pos]) {
dptr->data[s_pos] = kmalloc(quantum, GFP_KERNEL);
if (!dptr->data[s_pos]) {
goto out;
Memory leak
}
}
MANAGING CONCURRENCY
Concurrency and Its Management
 Sources of concurrency




Multiple user-space processes
Multiple CPUs
Device interrupts
Timers
Some guiding principles
 Try to avoid concurrent access entirely
 Global variables
 Apply locking and mutual exclusion principles
 Implications to device drivers



Use sufficient concurrency mechanisms (depending
on context)
No object can be made available to the kernel until it
can function properly
References to such objects must be tracked for proper
removal
 Avoid “roll your own” solutions
Managing Concurrency
 Atomic operation: all or nothing from the
perspective of other threads
 Critical section: code executed by only one
thread at a time

Not all critical sections are the same


Access from interrupt handlers
Latency constraints
Lock Design Considerations
 Context
 Can another thread be scheduled on the current
processor?
 Assumptions of kernel operation

Breaking assumptions will break code that relies on them
 Time expected to wait for lock
 Considerations



Long


Amount of time lock is expected to be held
Amount of expected contention
Other threads can make better use of the processor
Short

Time to switch to another thread will be longer than just
waiting a short amount of time
Kernel Locking Implementations
 mutex


Sleep if lock cannot be acquired immediately
Allow other threads to use the processor
 spinlock


Continuously try to grab the lock
Generally do not allow sleeping

Why?
MUTEX
Mutex Implementation
 Architecture-dependent code

Optimizations
 Initialization


DEFINE_MUTEX(name)
void mutex_init(struct mutex *lock);
 Various routines




void mutex_lock(struct mutex *lock);
int mutex_lock_interruptible(struct mutex *lock);
int mutex_lock_killable(struct mutex *lock);
void mutex_unlock(struct mutex *lock);
Using mutexes in scull
 scull_dev structure revisited
struct scull_dev {
struct scull_qset *data; /* Pointer to first quantum set
int quantum;
/* the current quantum size
int qset;
/* the current array size
unsigned long size;
/* amount of data stored here
unsigned int access_key; /* used by sculluid & scullpriv
struct mutex mutex;
/* mutual exclusion */
struct cdev cdev;
/* Char device structure
};
*/
*/
*/
*/
*/
*/
Using mutexes in scull
 scull_dev initialization
for (i = 0; i < scull_nr_devs; i++) {
scull_devices[i].quantum = scull_quantum;
scull_devices[i].qset = scull_qset;
mutex_init(&scull_devices[i].mutex); /* before cdev_add */
scull_setup_cdev(&scull_devices[i], i);
}
Using mutexes in scull
 scull_write()
if (mutex_lock_interruptible(&dev->mutex))
return -ERESTARTSYS;
 scull_write ends with
out:
mutex_unlock(&dev->mutex);
return retval;
mutex_lock_interruptible()
returns nonzero
 If can be resubmitted
 Undo
visible changes if any and
restart
 Otherwise return -EINTR
 E.g.,
could not undo changes
mutex_lock_interruptible()
(returns non-zero)
 If can be resubmitted
 Undo
visible changes if any and
restart
 Otherwise return -EINTR
 E.g.,
could not undo changes
Restartable system call
 Automatic restarting of certain interrupted system calls
 Retry with same arguments (values)
 Simplifies user-space programming for dealing with
"interrupted system call“
 POSIX permits an implementation to restart system calls,
but it is not required.
 SUS defines the SA_RESTART flag to provide a means
by which an application can request that an interrupted
system calls be restarted.
 http://pubs.opengroup.org/onlinepubs/009604499/function
s/sigaction.html
 return -ERESTARTSYS
Restartable system call
 Arguments may need to be modified
 return
-ERESTARTSYS_RESTARTBLOCK
 Specify
callback function to modify
arguments

http://lwn.net/Articles/17744/
Userspace write() and
kernelspace *_interruptible()
 From POSIX man page



If write() is interrupted by a signal before it
writes any data, it shall return -1 with errno set
to [EINTR].
If write() is interrupted by a signal after it
successfully writes some data, it shall return
the number of bytes written.
http://pubs.opengroup.org/onlinepubs/009604
499/functions/sigaction.html
mutex_lock_killable()
 mutex_lock()
 Process assumes that it cannot be interrupted by a signal
 Breaking assumption breaks user-kernel space interface
 If process receives fatal signal and mutex_lock() never
returns

Results in an immortal process
 Assumptions/expectations do not apply if process receives
fatal signal


Process that called system call will never return
Does not break assumption since process does not continue
 http://lwn.net/Articles/288056/
MUTEX USAGE AS
COMPLETION (ERROR)
HTTPS://LKML.ORG/LKML/2013/12/2/997
General Pattern
 refcount variable for deciding which thread to
perform cleanup
 Usage




Initialize shared object
Set refcount to number of concurrent threads
Start multiple threads
Last thread cleans up
<do stuff>
mutex_lock(obj->lock);
dead = !--obj->refcount;
mutex_unlock(obj->lock);
if (dead)
free(obj);
fs/pipe.c
__pipe_lock(pipe);
…
spin_lock(&inode->i_lock);
if (!--pipe->files) {
inode->i_pipe = NULL;
kill = 1;
}
spin_unlock(&inode->i_lock);
__pipe_unlock(pipe);
if (kill)
free_pipe_info(pipe);
CPU 1
mutex_lock(obj->lock);
dead = !--obj->refcount;
// refcount was 2, is now 1, dead = 0.
mutex_unlock(obj->lock);
__mutex_fastpath_unlock()
fastpath fails (because mutex is nonpositive
__mutex_unlock_slowpath:
if (__mutex_slowpath_needs_to_unlock())
atomic_set(&lock->count, 1);
but in the meantime, CPU1 is busy still
unlocking:
if (!list_empty(&lock->wait_list)) {
CPU 2
mutex_lock(obj->lock);
// blocks on obj->lock, goes to slowpath
// mutex is negative, CPU2 is in
optimistic
// spinning mode in
__mutex_lock_common
if ((atomic_read(&lock->count) == 1) &&
(atomic_cmpxchg(&lock->count, 1, 0)
== 1)) {
.. and now CPU2 owns the mutex, and goes
on
dead = !--obj->refcount;
// refcount was 1, is now 0, dead = 1.
mutex_unlock(obj->lock);
if (dead)
free(obj);
Conclusion
 Mutex serializes what is inside the mutex, but
not necessarily the lock ITSELF
 Use spinlocks and/or atomic ref counts
 "don't use mutexes to implement
completions"
COMPLETIONS
Completions
 Start and wait for operation to complete (outside current
thread)


Common pattern in kernel programming
E.g., wait for initialization to complete
 Reasons to use instead of mutexes
 Wake up multiple threads
 More efficient
 More meaningful syntax
 Subtle races with mutex implementation code

Cleanup of mutex itself

http://lkml.iu.edu//hypermail/linux/kernel/0107.3/0674.html
https://lkml.org/lkml/2008/4/11/323

 completions
 #include <linux/completion.h>
Completions
 To create a completion
DECLARE_COMPLETION(my_completion);

Or
struct completion my_completion;
init_completion(&my_completion);
 To wait for the completion, call
void wait_for_completion(struct completion *c);
void wait_for_completion_interruptible(struct
completion *c);
void wait_for_completion_timeout(struct completion
*c, unsigned long timeout);
Completions
 To signal a completion event, call one of the
following
/* wake up one waiting thread */
void complete(struct completion *c);
/* wake up multiple waiting threads */
/* need to call INIT_COMPLETION(struct completion c)
to reuse the completion structure */
void complete_all(struct completion *c);
Completions
 Example: misc-modules/complete.c
DECLARE_COMPLETION(comp);
ssize_t complete_read(struct file *filp, char __user *buf,
size_t count, loff_t *pos) {
printk(KERN_DEBUG "process %i (%s) going to sleep\n",
current->pid, current->comm);
wait_for_completion(&comp);
printk(KERN_DEBUG "awoken %i (%s)\n", current->pid,
current->comm);
return 0; /* EOF */
}
Completions
 Example
ssize_t complete_write(struct file *filp,
const char __user *buf, size_t count,
loff_t *pos) {
printk(KERN_DEBUG
"process %i (%s) awakening the readers...\n",
current->pid, current->comm);
complete(&comp);
return count; /* succeed, to avoid retrial */
}
SPINLOCKS
Spinlocks
 Generally used in code that should not sleep

(e.g., interrupt handlers)
 Usually implemented as a single bit


If the lock is available, the bit is set and the
code continues
If the lock is taken, the code enters a tight loop

Repeatedly checks the lock until it become
available
Spinlocks
 Actual implementation varies for different
architectures
 Protect a process from other CPUs and
interrupts

Usually does nothing on uniprocessor
machines

Exception: changing the IRQ masking status
Introduction to Spinlock API
 #include <linux/spinlock.h>
 To initialize, declare
spinlock_t my_lock = SPIN_LOCK_UNLOCKED;
 Or call
void spin_lock_init(spinlock_t *lock);
 To acquire a lock, call
void spin_lock(spinlock_t *lock);

Spinlock waits are uninterruptible
 To release a lock, call
void spin_unlock(spinlock_t *lock);
Spinlocks and Atomic Context
 While holding a spinlock, be atomic

Do not sleep or relinquish the processor

Examples of calls that can sleep
 Copying data to or from user space
 User-space page may need to be on disk…
 Memory allocation
 Memory might not be available

Disable interrupts (on the local CPU) as
needed
 Hold spinlocks for the minimum time possible
The Spinlock Functions
 Four functions to acquire a spinlock
void spin_lock(spinlock_t *lock);
/* disables interrupts on the local CPU */
void spin_lock_irqsave(spinlock_t *lock,
unsigned long flags);
/* only if no other code disabled interrupts */
void spin_lock_irq(spinlock_t *lock);
/* disables software interrupts; leaves hardware
interrupts enabled (e.g. tasklets)*/
void spin_lock_bh(spinlock_t *lock);
The Spinlock Functions
 Four functions to release a spinlock
void spin_unlock(spinlock_t *lock);
/* need to use the same flags variable for locking */
/* need to call spin_lock_irqsave and
spin_unlock_irqrestore in the same function, or your
code may break on some architectures */
void spin_unlock_irqrestore(spinlock_t *lock,
unsigned long flags);
void spin_unlock_irq(spinlock_t *lock);
void spin_unlock_bh(spinlock_t *lock);
Locking Traps
 It is very hard to manage concurrency
 What can possibly go wrong?
Ambiguous Rules
 Shared data structure D, protected by lock L
function A() {
lock(&L);
/* call function B() that accesses D */
unlock(&L);
}
 If function B() calls lock(&L), we have
a deadlock
Ambiguous Rules
 Solution


Have clear entry points to access data
structures
Document assumptions about locking
Lock Ordering Rules
function A() {
function B() {
lock(&L1);
lock(&L2);
/* access D */
unlock(&L2);
unlock(&L1)
}
lock(&L2);
lock(&L1);
/* access D */
unlock(&L1);
unlock(&L2)
}
- Multiple locks should always be acquired in the same order
- Easier said than done
Lock Ordering Rules
function A() {
function B() {
lock(&L1);
X();
unlock(&L1)
lock(&L2);
Y();
unlock(&L2)
}
}
function X() {
function Y() {
lock(&L2);
/* access D */
unlock(&L2);
}
lock(&L1);
/* access D */
unlock(&L1);
}
Lock Ordering Rules of Thumb
 Choose a lock ordering that is local to your
code before taking a lock belonging to a more
central part of the kernel

Lock of central kernel code likely has more
users (more contention)
 Obtain the mutex first before taking the
spinlock

Grabbing a mutex (which can sleep) inside a
spinlock can lead to deadlocks
Fine- Versus Coarse-Grained Locking
 Coarse-grained locking

Poor concurrency
 Fine-grained locking

Need to know which one to acquire

And which order to acquire
 At the device driver level



Start with coarse-grained locking
Refine the granularity as contention arises
Can enable lockstat to check lock holding time
BKL
 Kernel used to have “big kernel lock”


Giant spinlock introduced in Linux 2.0
Only one CPU could be executing locked
kernel code at any time
 BKL has been removed


https://lwn.net/Articles/384855/
https://www.linux.com/learn/tutorials/447301:w
hats-new-in-linux-2639-ding-dong-the-bigkernel-lock-is-dead
Alternatives to Locking
 Lock-free algorithms
 Atomic variables
 Bit operations
 seqlocks
 Read-copy-update (RCU)
Lock-Free Algorithms
 Circular buffer

Producer places data into one end of an array


When the end of the array is reached, the
producer wraps back
Consumer removes data from the other end
Lock-Free Algorithms
 Producer and consumer can access buffer
concurrently without race conditions


Always store the value before updating the
index into the array
Need to make sure that producer/consumer
indices do not overrun each other
 A generic circular buffer is available
 See <linux/kfifo.h>
ATOMIC VARIABLES
Atomic Variables
 If the shared resource is an integer value

Locking is overkill (if supported by processor)
 The kernel provides atomic types
 atomic_t - integer
 atomic64_t – long integer
 Both types must be accessed through special
functions (See <asm/atomic.h>)

SMP safe
Atomic Variables
 Atomic operations
 atomic_sub(amount, &account1);
 atomic_add(amount, &account2);
 A higher level locking must be used
Bit Operations
 Atomic bit operations
 See <asm/bitops.h>

SMP safe
OTHER SYNCHRONIZATION
MECHANISMS
Read-Copy-Update (RCU)
 Assumptions



Reads are common
Writes are rare
Resources accessed via pointers

All references to those resources held by atomic
code
Read-Copy-Update
 Basic idea




The writing thread makes a copy
Make changes to the copy
Switch a few pointers to commit changes
Deallocate the old version when all references
to the old version are gone
EVEN MORE…
seqlocks
 sequential lock
 Designed to protect small, simple, and frequently
accessed resource
 Write access is rare

Must obtain an exclusive lock (spinlock)
 Allow readers free access to the resource
 Lockless
 Operation



Check for collisions with writers
Retry as needed
Not for protecting pointers
seqlocks
Expected non-blocking reader usage:
do {
seq = read_seqbegin(&foo);
...
} while (read_seqretry(&foo, seq));
lglock (local/global locks)
 Fast per-cpu access
 Allows access to other cpu data (slow)
 Implementation

per-CPU array of spinlocks
 Can only be declared as global variables to
avoid overhead and keep things simple
 http://lwn.net/Articles/401738/
brlocks
 Sat Oct 5 14:19:39 2013 -0400

no need to keep brlock macros anymore...
 0f6ed63b170778b9c93fb0ae4017f110c9ee64
16
Reader/Writer Semaphores
 Allow multiple concurrent readers


Single writer (for infrequent writes)
Too many writers can lead to reader
starvation (unbounded waiting)
 #include <linux/rwsem.h>
 Do not follow the return value convention

E.g., returns 1 if successful
 Not interruptible
Reader/Writer Spinlocks
 Analogous to the reader/writer semaphores


Allow multiple readers to enter a critical
section
Provide exclusive access for writers
 #include <linux/spinlock.h>
Reader/Writer Spinlocks
 To declare and initialize, there are two ways
/* static way */
rwlock_t my_rwlock = RW_LOCK_UNLOCKED;
/* dynamic way */
rwlock_t my_rwlock;
rwlock_init(&my_rwlock);
Reader/Writer Spinlocks
 Similar functions are available
void
void
void
void
read_lock(rwlock_t *lock);
read_lock_irqsave(rwlock_t *lock, unsigned long flags);
read_lock_irq(rwlock_t *lock);
read_lock_bh(rwlock_t *lock);
void read_unlock(rwlock_t *lock);
void read_unlock_irqrestore(rwlock_t *lock,
unsigned long flags);
void read_unlock_irq(rwlock_t *lock);
void read_unlock_bh(rwlock_t *lock);
Reader/Writer Spinlocks
 Similar functions are available
void
void
void
void
write_lock(rwlock_t *lock);
write_lock_irqsave(rwlock_t *lock, unsigned long flags);
write_lock_irq(rwlock_t *lock);
write_lock_bh(rwlock_t *lock);
void write_unlock(rwlock_t *lock);
void write_unlock_irqrestore(rwlock_t *lock,
unsigned long flags);
void write_unlock_irq(rwlock_t *lock);
void write_unlock_bh(rwlock_t *lock);