1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
|
/*
* Dynamic function tracer architecture backend.
*
* Copyright IBM Corp. 2009,2014
*
* Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>,
* Martin Schwidefsky <schwidefsky@de.ibm.com>
*/
#include <linux/moduleloader.h>
#include <linux/hardirq.h>
#include <linux/uaccess.h>
#include <linux/ftrace.h>
#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/kprobes.h>
#include <trace/syscall.h>
#include <asm/asm-offsets.h>
#include <asm/cacheflush.h>
#include "entry.h"
/*
* The mcount code looks like this:
* stg %r14,8(%r15) # offset 0
* larl %r1,<&counter> # offset 6
* brasl %r14,_mcount # offset 12
* lg %r14,8(%r15) # offset 18
* Total length is 24 bytes. Only the first instruction will be patched
* by ftrace_make_call / ftrace_make_nop.
* The enabled ftrace code block looks like this:
* > brasl %r0,ftrace_caller # offset 0
* larl %r1,<&counter> # offset 6
* brasl %r14,_mcount # offset 12
* lg %r14,8(%r15) # offset 18
* The ftrace function gets called with a non-standard C function call ABI
* where r0 contains the return address. It is also expected that the called
* function only clobbers r0 and r1, but restores r2-r15.
* For module code we can't directly jump to ftrace caller, but need a
* trampoline (ftrace_plt), which clobbers also r1.
* The return point of the ftrace function has offset 24, so execution
* continues behind the mcount block.
* The disabled ftrace code block looks like this:
* > jg .+24 # offset 0
* larl %r1,<&counter> # offset 6
* brasl %r14,_mcount # offset 12
* lg %r14,8(%r15) # offset 18
* The jg instruction branches to offset 24 to skip as many instructions
* as possible.
*/
unsigned long ftrace_plt;
int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
unsigned long addr)
{
return 0;
}
int ftrace_make_nop(struct module *mod, struct dyn_ftrace *rec,
unsigned long addr)
{
struct ftrace_insn orig, new, old;
if (probe_kernel_read(&old, (void *) rec->ip, sizeof(old)))
return -EFAULT;
if (addr == MCOUNT_ADDR) {
/* Initial code replacement; we expect to see stg r14,8(r15) */
orig.opc = 0xe3e0;
orig.disp = 0xf0080024;
ftrace_generate_nop_insn(&new);
} else if (old.opc == BREAKPOINT_INSTRUCTION) {
/*
* If we find a breakpoint instruction, a kprobe has been
* placed at the beginning of the function. We write the
* constant KPROBE_ON_FTRACE_NOP into the remaining four
* bytes of the original instruction so that the kprobes
* handler can execute a nop, if it reaches this breakpoint.
*/
new.opc = orig.opc = BREAKPOINT_INSTRUCTION;
orig.disp = KPROBE_ON_FTRACE_CALL;
new.disp = KPROBE_ON_FTRACE_NOP;
} else {
/* Replace ftrace call with a nop. */
ftrace_generate_call_insn(&orig, rec->ip);
ftrace_generate_nop_insn(&new);
}
/* Verify that the to be replaced code matches what we expect. */
if (memcmp(&orig, &old, sizeof(old)))
return -EINVAL;
if (probe_kernel_write((void *) rec->ip, &new, sizeof(new)))
return -EPERM;
return 0;
}
int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
{
struct ftrace_insn orig, new, old;
if (probe_kernel_read(&old, (void *) rec->ip, sizeof(old)))
return -EFAULT;
if (old.opc == BREAKPOINT_INSTRUCTION) {
/*
* If we find a breakpoint instruction, a kprobe has been
* placed at the beginning of the function. We write the
* constant KPROBE_ON_FTRACE_CALL into the remaining four
* bytes of the original instruction so that the kprobes
* handler can execute a brasl if it reaches this breakpoint.
*/
new.opc = orig.opc = BREAKPOINT_INSTRUCTION;
orig.disp = KPROBE_ON_FTRACE_NOP;
new.disp = KPROBE_ON_FTRACE_CALL;
} else {
/* Replace nop with an ftrace call. */
ftrace_generate_nop_insn(&orig);
ftrace_generate_call_insn(&new, rec->ip);
}
/* Verify that the to be replaced code matches what we expect. */
if (memcmp(&orig, &old, sizeof(old)))
return -EINVAL;
if (probe_kernel_write((void *) rec->ip, &new, sizeof(new)))
return -EPERM;
return 0;
}
int ftrace_update_ftrace_func(ftrace_func_t func)
{
return 0;
}
int __init ftrace_dyn_arch_init(void)
{
return 0;
}
static int __init ftrace_plt_init(void)
{
unsigned int *ip;
ftrace_plt = (unsigned long) module_alloc(PAGE_SIZE);
if (!ftrace_plt)
panic("cannot allocate ftrace plt\n");
ip = (unsigned int *) ftrace_plt;
ip[0] = 0x0d10e310; /* basr 1,0; lg 1,10(1); br 1 */
ip[1] = 0x100a0004;
ip[2] = 0x07f10000;
ip[3] = FTRACE_ADDR >> 32;
ip[4] = FTRACE_ADDR & 0xffffffff;
set_memory_ro(ftrace_plt, 1);
return 0;
}
device_initcall(ftrace_plt_init);
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
/*
* Hook the return address and push it in the stack of return addresses
* in current thread info.
*/
unsigned long prepare_ftrace_return(unsigned long parent, unsigned long ip)
{
struct ftrace_graph_ent trace;
if (unlikely(ftrace_graph_is_dead()))
goto out;
if (unlikely(atomic_read(¤t->tracing_graph_pause)))
goto out;
ip = (ip & PSW_ADDR_INSN) - MCOUNT_INSN_SIZE;
trace.func = ip;
trace.depth = current->curr_ret_stack + 1;
/* Only trace if the calling function expects to. */
if (!ftrace_graph_entry(&trace))
goto out;
if (ftrace_push_return_trace(parent, ip, &trace.depth, 0) == -EBUSY)
goto out;
parent = (unsigned long) return_to_handler;
out:
return parent;
}
NOKPROBE_SYMBOL(prepare_ftrace_return);
/*
* Patch the kernel code at ftrace_graph_caller location. The instruction
* there is branch relative on condition. To enable the ftrace graph code
* block, we simply patch the mask field of the instruction to zero and
* turn the instruction into a nop.
* To disable the ftrace graph code the mask field will be patched to
* all ones, which turns the instruction into an unconditional branch.
*/
int ftrace_enable_ftrace_graph_caller(void)
{
u8 op = 0x04; /* set mask field to zero */
return probe_kernel_write(__va(ftrace_graph_caller)+1, &op, sizeof(op));
}
int ftrace_disable_ftrace_graph_caller(void)
{
u8 op = 0xf4; /* set mask field to all ones */
return probe_kernel_write(__va(ftrace_graph_caller)+1, &op, sizeof(op));
}
#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
|
