tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
powernv: Add OPAL tracepoints
Knowing how long we spend in firmware calls is an important part of
minimising OS jitter.
This patch adds tracepoints to each OPAL call. If tracepoints are
enabled we branch out to a common routine that calls an entry and exit
tracepoint.
This allows us to write tools that monitor the frequency and duration
of OPAL calls, eg:
name count total(ms) min(ms) max(ms) avg(ms) period(ms)
OPAL_HANDLE_INTERRUPT 5 0.199 0.037 0.042 0.040 12547.545
OPAL_POLL_EVENTS 204 2.590 0.012 0.036 0.013 2264.899
OPAL_PCI_MSI_EOI 2830 3.066 0.001 0.005 0.001 81.166
We use jump labels if configured, which means we only add a single
nop instruction to every OPAL call when the tracepoints are disabled.
Signed-off-by: Anton Blanchard <anton@samba.org>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
authored by
Anton Blanchard
and committed by
Benjamin Herrenschmidt
c49f6353
aaad4224
+236
-9
+45
arch/powerpc/include/asm/trace.h
+45
arch/powerpc/include/asm/trace.h
···
99
99
);
100
100
#endif
101
101
102
+
#ifdef CONFIG_PPC_POWERNV
103
+
extern void opal_tracepoint_regfunc(void);
104
+
extern void opal_tracepoint_unregfunc(void);
105
+
106
+
TRACE_EVENT_FN(opal_entry,
107
+
108
+
TP_PROTO(unsigned long opcode, unsigned long *args),
109
+
110
+
TP_ARGS(opcode, args),
111
+
112
+
TP_STRUCT__entry(
113
+
__field(unsigned long, opcode)
114
+
),
115
+
116
+
TP_fast_assign(
117
+
__entry->opcode = opcode;
118
+
),
119
+
120
+
TP_printk("opcode=%lu", __entry->opcode),
121
+
122
+
opal_tracepoint_regfunc, opal_tracepoint_unregfunc
123
+
);
124
+
125
+
TRACE_EVENT_FN(opal_exit,
126
+
127
+
TP_PROTO(unsigned long opcode, unsigned long retval),
128
+
129
+
TP_ARGS(opcode, retval),
130
+
131
+
TP_STRUCT__entry(
132
+
__field(unsigned long, opcode)
133
+
__field(unsigned long, retval)
134
+
),
135
+
136
+
TP_fast_assign(
137
+
__entry->opcode = opcode;
138
+
__entry->retval = retval;
139
+
),
140
+
141
+
TP_printk("opcode=%lu retval=%lu", __entry->opcode, __entry->retval),
142
+
143
+
opal_tracepoint_regfunc, opal_tracepoint_unregfunc
144
+
);
145
+
#endif
146
+
102
147
#endif /* _TRACE_POWERPC_H */
103
148
104
149
#undef TRACE_INCLUDE_PATH
+1
arch/powerpc/platforms/powernv/Makefile
+1
arch/powerpc/platforms/powernv/Makefile
+84
arch/powerpc/platforms/powernv/opal-tracepoints.c
+84
arch/powerpc/platforms/powernv/opal-tracepoints.c
···
1
+
#include <linux/percpu.h>
2
+
#include <linux/jump_label.h>
3
+
#include <asm/trace.h>
4
+
5
+
#ifdef CONFIG_JUMP_LABEL
6
+
struct static_key opal_tracepoint_key = STATIC_KEY_INIT;
7
+
8
+
void opal_tracepoint_regfunc(void)
9
+
{
10
+
static_key_slow_inc(&opal_tracepoint_key);
11
+
}
12
+
13
+
void opal_tracepoint_unregfunc(void)
14
+
{
15
+
static_key_slow_dec(&opal_tracepoint_key);
16
+
}
17
+
#else
18
+
/*
19
+
* We optimise OPAL calls by placing opal_tracepoint_refcount
20
+
* directly in the TOC so we can check if the opal tracepoints are
21
+
* enabled via a single load.
22
+
*/
23
+
24
+
/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
25
+
extern long opal_tracepoint_refcount;
26
+
27
+
void opal_tracepoint_regfunc(void)
28
+
{
29
+
opal_tracepoint_refcount++;
30
+
}
31
+
32
+
void opal_tracepoint_unregfunc(void)
33
+
{
34
+
opal_tracepoint_refcount--;
35
+
}
36
+
#endif
37
+
38
+
/*
39
+
* Since the tracing code might execute OPAL calls we need to guard against
40
+
* recursion.
41
+
*/
42
+
static DEFINE_PER_CPU(unsigned int, opal_trace_depth);
43
+
44
+
void __trace_opal_entry(unsigned long opcode, unsigned long *args)
45
+
{
46
+
unsigned long flags;
47
+
unsigned int *depth;
48
+
49
+
local_irq_save(flags);
50
+
51
+
depth = &__get_cpu_var(opal_trace_depth);
52
+
53
+
if (*depth)
54
+
goto out;
55
+
56
+
(*depth)++;
57
+
preempt_disable();
58
+
trace_opal_entry(opcode, args);
59
+
(*depth)--;
60
+
61
+
out:
62
+
local_irq_restore(flags);
63
+
}
64
+
65
+
void __trace_opal_exit(long opcode, unsigned long retval)
66
+
{
67
+
unsigned long flags;
68
+
unsigned int *depth;
69
+
70
+
local_irq_save(flags);
71
+
72
+
depth = &__get_cpu_var(opal_trace_depth);
73
+
74
+
if (*depth)
75
+
goto out;
76
+
77
+
(*depth)++;
78
+
trace_opal_exit(opcode, retval);
79
+
preempt_enable();
80
+
(*depth)--;
81
+
82
+
out:
83
+
local_irq_restore(flags);
84
+
}
+106
-9
arch/powerpc/platforms/powernv/opal-wrappers.S
+106
-9
arch/powerpc/platforms/powernv/opal-wrappers.S
···
13
13
#include <asm/hvcall.h>
14
14
#include <asm/asm-offsets.h>
15
15
#include <asm/opal.h>
16
+
#include <asm/jump_label.h>
17
+
18
+
.section ".text"
19
+
20
+
#ifdef CONFIG_TRACEPOINTS
21
+
#ifdef CONFIG_JUMP_LABEL
22
+
#define OPAL_BRANCH(LABEL) \
23
+
ARCH_STATIC_BRANCH(LABEL, opal_tracepoint_key)
24
+
#else
25
+
26
+
.section ".toc","aw"
27
+
28
+
.globl opal_tracepoint_refcount
29
+
opal_tracepoint_refcount:
30
+
.llong 0
31
+
32
+
.section ".text"
33
+
34
+
/*
35
+
* We branch around this in early init by using an unconditional cpu
36
+
* feature.
37
+
*/
38
+
#define OPAL_BRANCH(LABEL) \
39
+
BEGIN_FTR_SECTION; \
40
+
b 1f; \
41
+
END_FTR_SECTION(0, 1); \
42
+
ld r12,opal_tracepoint_refcount@toc(r2); \
43
+
std r12,32(r1); \
44
+
cmpdi r12,0; \
45
+
bne- LABEL; \
46
+
1:
47
+
48
+
#endif
49
+
50
+
#else
51
+
#define OPAL_BRANCH(LABEL)
52
+
#endif
16
53
17
54
/* TODO:
18
55
*
19
56
* - Trace irqs in/off (needs saving/restoring all args, argh...)
20
57
* - Get r11 feed up by Dave so I can have better register usage
21
58
*/
59
+
22
60
#define OPAL_CALL(name, token) \
23
61
_GLOBAL(name); \
24
62
mflr r0; \
25
-
mfcr r12; \
26
63
std r0,16(r1); \
64
+
li r0,token; \
65
+
OPAL_BRANCH(opal_tracepoint_entry) \
66
+
mfcr r12; \
27
67
stw r12,8(r1); \
28
68
std r1,PACAR1(r13); \
29
-
li r0,0; \
69
+
li r11,0; \
30
70
mfmsr r12; \
31
-
ori r0,r0,MSR_EE; \
71
+
ori r11,r11,MSR_EE; \
32
72
std r12,PACASAVEDMSR(r13); \
33
-
andc r12,r12,r0; \
73
+
andc r12,r12,r11; \
34
74
mtmsrd r12,1; \
35
-
LOAD_REG_ADDR(r0,opal_return); \
36
-
mtlr r0; \
37
-
li r0,MSR_DR|MSR_IR|MSR_LE;\
38
-
andc r12,r12,r0; \
39
-
li r0,token; \
75
+
LOAD_REG_ADDR(r11,opal_return); \
76
+
mtlr r11; \
77
+
li r11,MSR_DR|MSR_IR|MSR_LE;\
78
+
andc r12,r12,r11; \
40
79
mtspr SPRN_HSRR1,r12; \
41
80
LOAD_REG_ADDR(r11,opal); \
42
81
ld r12,8(r11); \
···
99
60
mtspr SPRN_SRR1,r6;
100
61
mtcr r4;
101
62
rfid
63
+
64
+
#ifdef CONFIG_TRACEPOINTS
65
+
opal_tracepoint_entry:
66
+
stdu r1,-STACKFRAMESIZE(r1)
67
+
std r0,STK_REG(R23)(r1)
68
+
std r3,STK_REG(R24)(r1)
69
+
std r4,STK_REG(R25)(r1)
70
+
std r5,STK_REG(R26)(r1)
71
+
std r6,STK_REG(R27)(r1)
72
+
std r7,STK_REG(R28)(r1)
73
+
std r8,STK_REG(R29)(r1)
74
+
std r9,STK_REG(R30)(r1)
75
+
std r10,STK_REG(R31)(r1)
76
+
mr r3,r0
77
+
addi r4,r1,STK_REG(R24)
78
+
bl __trace_opal_entry
79
+
ld r0,STK_REG(R23)(r1)
80
+
ld r3,STK_REG(R24)(r1)
81
+
ld r4,STK_REG(R25)(r1)
82
+
ld r5,STK_REG(R26)(r1)
83
+
ld r6,STK_REG(R27)(r1)
84
+
ld r7,STK_REG(R28)(r1)
85
+
ld r8,STK_REG(R29)(r1)
86
+
ld r9,STK_REG(R30)(r1)
87
+
ld r10,STK_REG(R31)(r1)
88
+
LOAD_REG_ADDR(r11,opal_tracepoint_return)
89
+
mfcr r12
90
+
std r11,16(r1)
91
+
stw r12,8(r1)
92
+
std r1,PACAR1(r13)
93
+
li r11,0
94
+
mfmsr r12
95
+
ori r11,r11,MSR_EE
96
+
std r12,PACASAVEDMSR(r13)
97
+
andc r12,r12,r11
98
+
mtmsrd r12,1
99
+
LOAD_REG_ADDR(r11,opal_return)
100
+
mtlr r11
101
+
li r11,MSR_DR|MSR_IR|MSR_LE
102
+
andc r12,r12,r11
103
+
mtspr SPRN_HSRR1,r12
104
+
LOAD_REG_ADDR(r11,opal)
105
+
ld r12,8(r11)
106
+
ld r2,0(r11)
107
+
mtspr SPRN_HSRR0,r12
108
+
hrfid
109
+
110
+
opal_tracepoint_return:
111
+
std r3,STK_REG(R31)(r1)
112
+
mr r4,r3
113
+
ld r0,STK_REG(R23)(r1)
114
+
bl __trace_opal_exit
115
+
ld r3,STK_REG(R31)(r1)
116
+
addi r1,r1,STACKFRAMESIZE
117
+
ld r0,16(r1)
118
+
mtlr r0
119
+
blr
120
+
#endif
102
121
103
122
OPAL_CALL(opal_invalid_call, OPAL_INVALID_CALL);
104
123
OPAL_CALL(opal_console_write, OPAL_CONSOLE_WRITE);