6 The use of 'struct static_key' directly, is now DEPRECATED. In addition
7 static_key_{true,false}() is also DEPRECATED. IE DO NOT use the following:
9 struct static_key false = STATIC_KEY_INIT_FALSE;
10 struct static_key true = STATIC_KEY_INIT_TRUE;
14 The updated API replacements are:
16 DEFINE_STATIC_KEY_TRUE(key);
17 DEFINE_STATIC_KEY_FALSE(key);
18 DEFINE_STATIC_KEY_ARRAY_TRUE(keys, count);
19 DEFINE_STATIC_KEY_ARRAY_FALSE(keys, count);
20 static_branch_likely()
21 static_branch_unlikely()
25 Static keys allows the inclusion of seldom used features in
26 performance-sensitive fast-path kernel code, via a GCC feature and a code
27 patching technique. A quick example:
29 DEFINE_STATIC_KEY_FALSE(key);
33 if (static_branch_unlikely(&key))
39 static_branch_enable(&key);
41 static_branch_disable(&key);
44 The static_branch_unlikely() branch will be generated into the code with as little
45 impact to the likely code path as possible.
51 Currently, tracepoints are implemented using a conditional branch. The
52 conditional check requires checking a global variable for each tracepoint.
53 Although the overhead of this check is small, it increases when the memory
54 cache comes under pressure (memory cache lines for these global variables may
55 be shared with other memory accesses). As we increase the number of tracepoints
56 in the kernel this overhead may become more of an issue. In addition,
57 tracepoints are often dormant (disabled) and provide no direct kernel
58 functionality. Thus, it is highly desirable to reduce their impact as much as
59 possible. Although tracepoints are the original motivation for this work, other
60 kernel code paths should be able to make use of the static keys facility.
66 gcc (v4.5) adds a new 'asm goto' statement that allows branching to a label:
68 http://gcc.gnu.org/ml/gcc-patches/2009-07/msg01556.html
70 Using the 'asm goto', we can create branches that are either taken or not taken
71 by default, without the need to check memory. Then, at run-time, we can patch
72 the branch site to change the branch direction.
74 For example, if we have a simple branch that is disabled by default:
76 if (static_branch_unlikely(&key))
77 printk("I am the true branch\n");
79 Thus, by default the 'printk' will not be emitted. And the code generated will
80 consist of a single atomic 'no-op' instruction (5 bytes on x86), in the
81 straight-line code path. When the branch is 'flipped', we will patch the
82 'no-op' in the straight-line codepath with a 'jump' instruction to the
83 out-of-line true branch. Thus, changing branch direction is expensive but
84 branch selection is basically 'free'. That is the basic tradeoff of this
87 This lowlevel patching mechanism is called 'jump label patching', and it gives
88 the basis for the static keys facility.
90 3) Static key label API, usage and examples:
93 In order to make use of this optimization you must first define a key:
95 DEFINE_STATIC_KEY_TRUE(key);
99 DEFINE_STATIC_KEY_FALSE(key);
102 The key must be global, that is, it can't be allocated on the stack or dynamically
103 allocated at run-time.
105 The key is then used in code as:
107 if (static_branch_unlikely(&key))
114 if (static_branch_likely(&key))
119 Keys defined via DEFINE_STATIC_KEY_TRUE(), or DEFINE_STATIC_KEY_FALSE, may
120 be used in either static_branch_likely() or static_branch_unlikely()
123 Branch(es) can be set true via:
125 static_branch_enable(&key);
129 static_branch_disable(&key);
131 The branch(es) can then be switched via reference counts:
133 static_branch_inc(&key);
135 static_branch_dec(&key);
137 Thus, 'static_branch_inc()' means 'make the branch true', and
138 'static_branch_dec()' means 'make the branch false' with appropriate
139 reference counting. For example, if the key is initialized true, a
140 static_branch_dec(), will switch the branch to false. And a subsequent
141 static_branch_inc(), will change the branch back to true. Likewise, if the
142 key is initialized false, a 'static_branch_inc()', will change the branch to
143 true. And then a 'static_branch_dec()', will again make the branch false.
145 Where an array of keys is required, it can be defined as:
147 DEFINE_STATIC_KEY_ARRAY_TRUE(keys, count);
151 DEFINE_STATIC_KEY_ARRAY_FALSE(keys, count);
153 4) Architecture level code patching interface, 'jump labels'
156 There are a few functions and macros that architectures must implement in order
157 to take advantage of this optimization. If there is no architecture support, we
158 simply fall back to a traditional, load, test, and jump sequence. Also, the
159 struct jump_entry table must be at least 4-byte aligned because the
160 static_key->entry field makes use of the two least significant bits.
162 * select HAVE_ARCH_JUMP_LABEL, see: arch/x86/Kconfig
164 * #define JUMP_LABEL_NOP_SIZE, see: arch/x86/include/asm/jump_label.h
166 * __always_inline bool arch_static_branch(struct static_key *key, bool branch), see:
167 arch/x86/include/asm/jump_label.h
169 * __always_inline bool arch_static_branch_jump(struct static_key *key, bool branch),
170 see: arch/x86/include/asm/jump_label.h
172 * void arch_jump_label_transform(struct jump_entry *entry, enum jump_label_type type),
173 see: arch/x86/kernel/jump_label.c
175 * __init_or_module void arch_jump_label_transform_static(struct jump_entry *entry, enum jump_label_type type),
176 see: arch/x86/kernel/jump_label.c
179 * struct jump_entry, see: arch/x86/include/asm/jump_label.h
182 5) Static keys / jump label analysis, results (x86_64):
185 As an example, let's add the following branch to 'getppid()', such that the
186 system call now looks like:
188 SYSCALL_DEFINE0(getppid)
192 + if (static_branch_unlikely(&key))
193 + printk("I am the true branch\n");
196 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
202 The resulting instructions with jump labels generated by GCC is:
204 ffffffff81044290 <sys_getppid>:
205 ffffffff81044290: 55 push %rbp
206 ffffffff81044291: 48 89 e5 mov %rsp,%rbp
207 ffffffff81044294: e9 00 00 00 00 jmpq ffffffff81044299 <sys_getppid+0x9>
208 ffffffff81044299: 65 48 8b 04 25 c0 b6 mov %gs:0xb6c0,%rax
209 ffffffff810442a0: 00 00
210 ffffffff810442a2: 48 8b 80 80 02 00 00 mov 0x280(%rax),%rax
211 ffffffff810442a9: 48 8b 80 b0 02 00 00 mov 0x2b0(%rax),%rax
212 ffffffff810442b0: 48 8b b8 e8 02 00 00 mov 0x2e8(%rax),%rdi
213 ffffffff810442b7: e8 f4 d9 00 00 callq ffffffff81051cb0 <pid_vnr>
214 ffffffff810442bc: 5d pop %rbp
215 ffffffff810442bd: 48 98 cltq
216 ffffffff810442bf: c3 retq
217 ffffffff810442c0: 48 c7 c7 e3 54 98 81 mov $0xffffffff819854e3,%rdi
218 ffffffff810442c7: 31 c0 xor %eax,%eax
219 ffffffff810442c9: e8 71 13 6d 00 callq ffffffff8171563f <printk>
220 ffffffff810442ce: eb c9 jmp ffffffff81044299 <sys_getppid+0x9>
222 Without the jump label optimization it looks like:
224 ffffffff810441f0 <sys_getppid>:
225 ffffffff810441f0: 8b 05 8a 52 d8 00 mov 0xd8528a(%rip),%eax # ffffffff81dc9480 <key>
226 ffffffff810441f6: 55 push %rbp
227 ffffffff810441f7: 48 89 e5 mov %rsp,%rbp
228 ffffffff810441fa: 85 c0 test %eax,%eax
229 ffffffff810441fc: 75 27 jne ffffffff81044225 <sys_getppid+0x35>
230 ffffffff810441fe: 65 48 8b 04 25 c0 b6 mov %gs:0xb6c0,%rax
231 ffffffff81044205: 00 00
232 ffffffff81044207: 48 8b 80 80 02 00 00 mov 0x280(%rax),%rax
233 ffffffff8104420e: 48 8b 80 b0 02 00 00 mov 0x2b0(%rax),%rax
234 ffffffff81044215: 48 8b b8 e8 02 00 00 mov 0x2e8(%rax),%rdi
235 ffffffff8104421c: e8 2f da 00 00 callq ffffffff81051c50 <pid_vnr>
236 ffffffff81044221: 5d pop %rbp
237 ffffffff81044222: 48 98 cltq
238 ffffffff81044224: c3 retq
239 ffffffff81044225: 48 c7 c7 13 53 98 81 mov $0xffffffff81985313,%rdi
240 ffffffff8104422c: 31 c0 xor %eax,%eax
241 ffffffff8104422e: e8 60 0f 6d 00 callq ffffffff81715193 <printk>
242 ffffffff81044233: eb c9 jmp ffffffff810441fe <sys_getppid+0xe>
243 ffffffff81044235: 66 66 2e 0f 1f 84 00 data32 nopw %cs:0x0(%rax,%rax,1)
244 ffffffff8104423c: 00 00 00 00
246 Thus, the disable jump label case adds a 'mov', 'test' and 'jne' instruction
247 vs. the jump label case just has a 'no-op' or 'jmp 0'. (The jmp 0, is patched
248 to a 5 byte atomic no-op instruction at boot-time.) Thus, the disabled jump
251 6 (mov) + 2 (test) + 2 (jne) = 10 - 5 (5 byte jump 0) = 5 addition bytes.
253 If we then include the padding bytes, the jump label code saves, 16 total bytes
254 of instruction memory for this small function. In this case the non-jump label
255 function is 80 bytes long. Thus, we have saved 20% of the instruction
256 footprint. We can in fact improve this even further, since the 5-byte no-op
257 really can be a 2-byte no-op since we can reach the branch with a 2-byte jmp.
258 However, we have not yet implemented optimal no-op sizes (they are currently
261 Since there are a number of static key API uses in the scheduler paths,
262 'pipe-test' (also known as 'perf bench sched pipe') can be used to show the
263 performance improvement. Testing done on 3.3.0-rc2:
267 Performance counter stats for 'bash -c /tmp/pipe-test' (50 runs):
269 855.700314 task-clock # 0.534 CPUs utilized ( +- 0.11% )
270 200,003 context-switches # 0.234 M/sec ( +- 0.00% )
271 0 CPU-migrations # 0.000 M/sec ( +- 39.58% )
272 487 page-faults # 0.001 M/sec ( +- 0.02% )
273 1,474,374,262 cycles # 1.723 GHz ( +- 0.17% )
274 <not supported> stalled-cycles-frontend
275 <not supported> stalled-cycles-backend
276 1,178,049,567 instructions # 0.80 insns per cycle ( +- 0.06% )
277 208,368,926 branches # 243.507 M/sec ( +- 0.06% )
278 5,569,188 branch-misses # 2.67% of all branches ( +- 0.54% )
280 1.601607384 seconds time elapsed ( +- 0.07% )
284 Performance counter stats for 'bash -c /tmp/pipe-test' (50 runs):
286 841.043185 task-clock # 0.533 CPUs utilized ( +- 0.12% )
287 200,004 context-switches # 0.238 M/sec ( +- 0.00% )
288 0 CPU-migrations # 0.000 M/sec ( +- 40.87% )
289 487 page-faults # 0.001 M/sec ( +- 0.05% )
290 1,432,559,428 cycles # 1.703 GHz ( +- 0.18% )
291 <not supported> stalled-cycles-frontend
292 <not supported> stalled-cycles-backend
293 1,175,363,994 instructions # 0.82 insns per cycle ( +- 0.04% )
294 206,859,359 branches # 245.956 M/sec ( +- 0.04% )
295 4,884,119 branch-misses # 2.36% of all branches ( +- 0.85% )
297 1.579384366 seconds time elapsed
299 The percentage of saved branches is .7%, and we've saved 12% on
300 'branch-misses'. This is where we would expect to get the most savings, since
301 this optimization is about reducing the number of branches. In addition, we've
302 saved .2% on instructions, and 2.8% on cycles and 1.4% on elapsed time.