Commit | Line | Data |
---|---|---|
3d083395 SR |
1 | /* |
2 | * Code for replacing ftrace calls with jumps. | |
3 | * | |
4 | * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com> | |
5 | * | |
6 | * Thanks goes to Ingo Molnar, for suggesting the idea. | |
7 | * Mathieu Desnoyers, for suggesting postponing the modifications. | |
8 | * Arjan van de Ven, for keeping me straight, and explaining to me | |
9 | * the dangers of modifying code on the run. | |
10 | */ | |
11 | ||
12 | #include <linux/spinlock.h> | |
13 | #include <linux/hardirq.h> | |
14 | #include <linux/ftrace.h> | |
15 | #include <linux/percpu.h> | |
16 | #include <linux/init.h> | |
17 | #include <linux/list.h> | |
18 | ||
395a59d0 | 19 | #include <asm/ftrace.h> |
732f3ca7 | 20 | #include <asm/nops.h> |
3d083395 | 21 | |
3d083395 | 22 | |
dfa60aba SR |
23 | /* Long is fine, even if it is only 4 bytes ;-) */ |
24 | static long *ftrace_nop; | |
3d083395 | 25 | |
3d083395 | 26 | union ftrace_code_union { |
395a59d0 | 27 | char code[MCOUNT_INSN_SIZE]; |
3d083395 SR |
28 | struct { |
29 | char e8; | |
30 | int offset; | |
31 | } __attribute__((packed)); | |
32 | }; | |
33 | ||
395a59d0 | 34 | |
3c1720f0 SR |
35 | static int notrace ftrace_calc_offset(long ip, long addr) |
36 | { | |
37 | return (int)(addr - ip); | |
38 | } | |
3d083395 | 39 | |
3c1720f0 SR |
40 | notrace unsigned char *ftrace_nop_replace(void) |
41 | { | |
42 | return (char *)ftrace_nop; | |
43 | } | |
44 | ||
45 | notrace unsigned char *ftrace_call_replace(unsigned long ip, unsigned long addr) | |
46 | { | |
47 | static union ftrace_code_union calc; | |
3d083395 | 48 | |
3c1720f0 | 49 | calc.e8 = 0xe8; |
395a59d0 | 50 | calc.offset = ftrace_calc_offset(ip + MCOUNT_INSN_SIZE, addr); |
3c1720f0 SR |
51 | |
52 | /* | |
53 | * No locking needed, this must be called via kstop_machine | |
54 | * which in essence is like running on a uniprocessor machine. | |
55 | */ | |
56 | return calc.code; | |
3d083395 SR |
57 | } |
58 | ||
3c1720f0 | 59 | notrace int |
3d083395 SR |
60 | ftrace_modify_code(unsigned long ip, unsigned char *old_code, |
61 | unsigned char *new_code) | |
62 | { | |
dfa60aba SR |
63 | unsigned replaced; |
64 | unsigned old = *(unsigned *)old_code; /* 4 bytes */ | |
65 | unsigned new = *(unsigned *)new_code; /* 4 bytes */ | |
66 | unsigned char newch = new_code[4]; | |
3d083395 SR |
67 | int faulted = 0; |
68 | ||
69 | /* | |
70 | * Note: Due to modules and __init, code can | |
71 | * disappear and change, we need to protect against faulting | |
72 | * as well as code changing. | |
73 | * | |
74 | * No real locking needed, this code is run through | |
75 | * kstop_machine. | |
76 | */ | |
77 | asm volatile ( | |
78 | "1: lock\n" | |
dfa60aba SR |
79 | " cmpxchg %3, (%2)\n" |
80 | " jnz 2f\n" | |
81 | " movb %b4, 4(%2)\n" | |
3d083395 SR |
82 | "2:\n" |
83 | ".section .fixup, \"ax\"\n" | |
a56be3fe SR |
84 | "3: movl $1, %0\n" |
85 | " jmp 2b\n" | |
3d083395 SR |
86 | ".previous\n" |
87 | _ASM_EXTABLE(1b, 3b) | |
88 | : "=r"(faulted), "=a"(replaced) | |
ee4311ad | 89 | : "r"(ip), "r"(new), "c"(newch), |
dfa60aba | 90 | "0"(faulted), "a"(old) |
3d083395 SR |
91 | : "memory"); |
92 | sync_core(); | |
93 | ||
dfa60aba | 94 | if (replaced != old && replaced != new) |
3d083395 SR |
95 | faulted = 2; |
96 | ||
97 | return faulted; | |
98 | } | |
99 | ||
d61f82d0 SR |
100 | notrace int ftrace_update_ftrace_func(ftrace_func_t func) |
101 | { | |
102 | unsigned long ip = (unsigned long)(&ftrace_call); | |
395a59d0 | 103 | unsigned char old[MCOUNT_INSN_SIZE], *new; |
d61f82d0 SR |
104 | int ret; |
105 | ||
395a59d0 | 106 | memcpy(old, &ftrace_call, MCOUNT_INSN_SIZE); |
d61f82d0 SR |
107 | new = ftrace_call_replace(ip, (unsigned long)func); |
108 | ret = ftrace_modify_code(ip, old, new); | |
109 | ||
110 | return ret; | |
111 | } | |
112 | ||
113 | notrace int ftrace_mcount_set(unsigned long *data) | |
114 | { | |
0a37605c SR |
115 | /* mcount is initialized as a nop */ |
116 | *data = 0; | |
d61f82d0 SR |
117 | return 0; |
118 | } | |
119 | ||
120 | int __init ftrace_dyn_arch_init(void *data) | |
3d083395 | 121 | { |
732f3ca7 SR |
122 | extern const unsigned char ftrace_test_p6nop[]; |
123 | extern const unsigned char ftrace_test_nop5[]; | |
124 | extern const unsigned char ftrace_test_jmp[]; | |
125 | int faulted = 0; | |
d61f82d0 | 126 | |
732f3ca7 SR |
127 | /* |
128 | * There is no good nop for all x86 archs. | |
129 | * We will default to using the P6_NOP5, but first we | |
130 | * will test to make sure that the nop will actually | |
131 | * work on this CPU. If it faults, we will then | |
132 | * go to a lesser efficient 5 byte nop. If that fails | |
133 | * we then just use a jmp as our nop. This isn't the most | |
134 | * efficient nop, but we can not use a multi part nop | |
135 | * since we would then risk being preempted in the middle | |
136 | * of that nop, and if we enabled tracing then, it might | |
137 | * cause a system crash. | |
138 | * | |
139 | * TODO: check the cpuid to determine the best nop. | |
140 | */ | |
141 | asm volatile ( | |
142 | "jmp ftrace_test_jmp\n" | |
143 | /* This code needs to stay around */ | |
144 | ".section .text, \"ax\"\n" | |
145 | "ftrace_test_jmp:" | |
146 | "jmp ftrace_test_p6nop\n" | |
147 | ".byte 0x00,0x00,0x00\n" /* 2 byte jmp + 3 bytes */ | |
148 | "ftrace_test_p6nop:" | |
149 | P6_NOP5 | |
150 | "jmp 1f\n" | |
151 | "ftrace_test_nop5:" | |
152 | ".byte 0x66,0x66,0x66,0x66,0x90\n" | |
153 | "jmp 1f\n" | |
154 | ".previous\n" | |
155 | "1:" | |
156 | ".section .fixup, \"ax\"\n" | |
157 | "2: movl $1, %0\n" | |
158 | " jmp ftrace_test_nop5\n" | |
159 | "3: movl $2, %0\n" | |
160 | " jmp 1b\n" | |
161 | ".previous\n" | |
162 | _ASM_EXTABLE(ftrace_test_p6nop, 2b) | |
163 | _ASM_EXTABLE(ftrace_test_nop5, 3b) | |
164 | : "=r"(faulted) : "0" (faulted)); | |
165 | ||
166 | switch (faulted) { | |
167 | case 0: | |
168 | pr_info("ftrace: converting mcount calls to 0f 1f 44 00 00\n"); | |
169 | ftrace_nop = (unsigned long *)ftrace_test_p6nop; | |
170 | break; | |
171 | case 1: | |
172 | pr_info("ftrace: converting mcount calls to 66 66 66 66 90\n"); | |
173 | ftrace_nop = (unsigned long *)ftrace_test_nop5; | |
174 | break; | |
175 | case 2: | |
176 | pr_info("ftrace: converting mcount calls to jmp 1f\n"); | |
177 | ftrace_nop = (unsigned long *)ftrace_test_jmp; | |
178 | break; | |
179 | } | |
180 | ||
181 | /* The return code is retured via data */ | |
182 | *(unsigned long *)data = 0; | |
dfa60aba | 183 | |
3d083395 SR |
184 | return 0; |
185 | } |