1 Semantics and Behavior of Local Atomic Operations
6 This document explains the purpose of the local atomic operations, how
7 to implement them for any given architecture and shows how they can be used
8 properly. It also stresses on the precautions that must be taken when reading
9 those local variables across CPUs when the order of memory writes matters.
13 * Purpose of local atomic operations
15 Local atomic operations are meant to provide fast and highly reentrant per CPU
16 counters. They minimize the performance cost of standard atomic operations by
17 removing the LOCK prefix and memory barriers normally required to synchronize
20 Having fast per CPU atomic counters is interesting in many cases : it does not
21 require disabling interrupts to protect from interrupt handlers and it permits
22 coherent counters in NMI handlers. It is especially useful for tracing purposes
23 and for various performance monitoring counters.
25 Local atomic operations only guarantee variable modification atomicity wrt the
26 CPU which owns the data. Therefore, care must taken to make sure that only one
27 CPU writes to the local_t data. This is done by using per cpu data and making
28 sure that we modify it from within a preemption safe context. It is however
29 permitted to read local_t data from any CPU : it will then appear to be written
30 out of order wrt other memory writes by the owner CPU.
33 * Implementation for a given architecture
35 It can be done by slightly modifying the standard atomic operations : only
36 their UP variant must be kept. It typically means removing LOCK prefix (on
37 i386 and x86_64) and any SMP sychronization barrier. If the architecture does
38 not have a different behavior between SMP and UP, including asm-generic/local.h
39 in your archtecture's local.h is sufficient.
41 The local_t type is defined as an opaque signed long by embedding an
42 atomic_long_t inside a structure. This is made so a cast from this type to a
43 long fails. The definition looks like :
45 typedef struct { atomic_long_t a; } local_t;
48 * Rules to follow when using local atomic operations
50 - Variables touched by local ops must be per cpu variables.
51 - _Only_ the CPU owner of these variables must write to them.
52 - This CPU can use local ops from any context (process, irq, softirq, nmi, ...)
53 to update its local_t variables.
54 - Preemption (or interrupts) must be disabled when using local ops in
55 process context to make sure the process won't be migrated to a
56 different CPU between getting the per-cpu variable and doing the
58 - When using local ops in interrupt context, no special care must be
59 taken on a mainline kernel, since they will run on the local CPU with
60 preemption already disabled. I suggest, however, to explicitly
61 disable preemption anyway to make sure it will still work correctly on
63 - Reading the local cpu variable will provide the current copy of the
65 - Reads of these variables can be done from any CPU, because updates to
66 "long", aligned, variables are always atomic. Since no memory
67 synchronization is done by the writer CPU, an outdated copy of the
68 variable can be read when reading some _other_ cpu's variables.
71 * Rules to follow when using local atomic operations
73 - Variables touched by local ops must be per cpu variables.
74 - _Only_ the CPU owner of these variables must write to them.
75 - This CPU can use local ops from any context (process, irq, softirq, nmi, ...)
76 to update its local_t variables.
77 - Preemption (or interrupts) must be disabled when using local ops in
78 process context to make sure the process won't be migrated to a
79 different CPU between getting the per-cpu variable and doing the
81 - When using local ops in interrupt context, no special care must be
82 taken on a mainline kernel, since they will run on the local CPU with
83 preemption already disabled. I suggest, however, to explicitly
84 disable preemption anyway to make sure it will still work correctly on
86 - Reading the local cpu variable will provide the current copy of the
88 - Reads of these variables can be done from any CPU, because updates to
89 "long", aligned, variables are always atomic. Since no memory
90 synchronization is done by the writer CPU, an outdated copy of the
91 variable can be read when reading some _other_ cpu's variables.
94 * How to use local atomic operations
96 #include <linux/percpu.h>
97 #include <asm/local.h>
99 static DEFINE_PER_CPU(local_t, counters) = LOCAL_INIT(0);
104 Counting is done on all the bits of a signed long.
106 In preemptible context, use get_cpu_var() and put_cpu_var() around local atomic
107 operations : it makes sure that preemption is disabled around write access to
108 the per cpu variable. For instance :
110 local_inc(&get_cpu_var(counters));
111 put_cpu_var(counters);
113 If you are already in a preemption-safe context, you can directly use
114 __get_cpu_var() instead.
116 local_inc(&__get_cpu_var(counters));
120 * Reading the counters
122 Those local counters can be read from foreign CPUs to sum the count. Note that
123 the data seen by local_read across CPUs must be considered to be out of order
124 relatively to other memory writes happening on the CPU that owns the data.
127 for_each_online_cpu(cpu)
128 sum += local_read(&per_cpu(counters, cpu));
130 If you want to use a remote local_read to synchronize access to a resource
131 between CPUs, explicit smp_wmb() and smp_rmb() memory barriers must be used
132 respectively on the writer and the reader CPUs. It would be the case if you use
133 the local_t variable as a counter of bytes written in a buffer : there should
134 be a smp_wmb() between the buffer write and the counter increment and also a
135 smp_rmb() between the counter read and the buffer read.
138 Here is a sample module which implements a basic per cpu counter using local.h.
143 * Sample module for local.h usage.
147 #include <asm/local.h>
148 #include <linux/module.h>
149 #include <linux/timer.h>
151 static DEFINE_PER_CPU(local_t, counters) = LOCAL_INIT(0);
153 static struct timer_list test_timer;
155 /* IPI called on each CPU. */
156 static void test_each(void *info)
158 /* Increment the counter from a non preemptible context */
159 printk("Increment on cpu %d\n", smp_processor_id());
160 local_inc(&__get_cpu_var(counters));
162 /* This is what incrementing the variable would look like within a
163 * preemptible context (it disables preemption) :
165 * local_inc(&get_cpu_var(counters));
166 * put_cpu_var(counters);
170 static void do_test_timer(unsigned long data)
174 /* Increment the counters */
175 on_each_cpu(test_each, NULL, 0, 1);
176 /* Read all the counters */
177 printk("Counters read from CPU %d\n", smp_processor_id());
178 for_each_online_cpu(cpu) {
179 printk("Read : CPU %d, count %ld\n", cpu,
180 local_read(&per_cpu(counters, cpu)));
182 del_timer(&test_timer);
183 test_timer.expires = jiffies + 1000;
184 add_timer(&test_timer);
187 static int __init test_init(void)
189 /* initialize the timer that will increment the counter */
190 init_timer(&test_timer);
191 test_timer.function = do_test_timer;
192 test_timer.expires = jiffies + 1;
193 add_timer(&test_timer);
198 static void __exit test_exit(void)
200 del_timer_sync(&test_timer);
203 module_init(test_init);
204 module_exit(test_exit);
206 MODULE_LICENSE("GPL");
207 MODULE_AUTHOR("Mathieu Desnoyers");
208 MODULE_DESCRIPTION("Local Atomic Ops");