tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
powerpc/perf/{hv-gpci, hv-common}: generate requests with counters annotated
This adds (in req-gen/) a framework for defining gpci counter requests.
It uses macro magic similar to ftrace.
Also convert the existing hv-gpci request structures and enum values to
use the new framework (and adjust old users of the structs and enum
values to cope with changes in naming).
In exchange for this macro disaster, we get autogenerated event listing
for GPCI in sysfs, build time field offset checking, and zero
duplication of information about GPCI requests.
Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
authored by
Cody P Schafer
and committed by
Michael Ellerman
9e9f6010
5c5cd7b5
+316
-30
+5
-5
arch/powerpc/perf/hv-common.c
+5
-5
arch/powerpc/perf/hv-common.c
···
9
9
unsigned long r;
10
10
struct p {
11
11
struct hv_get_perf_counter_info_params params;
12
-
struct cv_system_performance_capabilities caps;
12
+
struct hv_gpci_system_performance_capabilities caps;
13
13
} __packed __aligned(sizeof(uint64_t));
14
14
15
15
struct p arg = {
16
16
.params = {
17
17
.counter_request = cpu_to_be32(
18
-
CIR_SYSTEM_PERFORMANCE_CAPABILITIES),
18
+
HV_GPCI_system_performance_capabilities),
19
19
.starting_index = cpu_to_be32(-1),
20
20
.counter_info_version_in = 0,
21
21
}
···
31
31
32
32
caps->version = arg.params.counter_info_version_out;
33
33
caps->collect_privileged = !!arg.caps.perf_collect_privileged;
34
-
caps->ga = !!(arg.caps.capability_mask & CV_CM_GA);
35
-
caps->expanded = !!(arg.caps.capability_mask & CV_CM_EXPANDED);
36
-
caps->lab = !!(arg.caps.capability_mask & CV_CM_LAB);
34
+
caps->ga = !!(arg.caps.capability_mask & HV_GPCI_CM_GA);
35
+
caps->expanded = !!(arg.caps.capability_mask & HV_GPCI_CM_EXPANDED);
36
+
caps->lab = !!(arg.caps.capability_mask & HV_GPCI_CM_LAB);
37
37
38
38
return r;
39
39
}
+76
arch/powerpc/perf/hv-gpci-requests.h
+76
arch/powerpc/perf/hv-gpci-requests.h
···
1
+
2
+
#include "req-gen/_begin.h"
3
+
4
+
/*
5
+
* Based on the document "getPerfCountInfo v1.07"
6
+
*/
7
+
8
+
/*
9
+
* #define REQUEST_NAME counter_request_name
10
+
* #define REQUEST_NUM r_num
11
+
* #define REQUEST_IDX_KIND starting_index_kind
12
+
* #include I(REQUEST_BEGIN)
13
+
* REQUEST(
14
+
* __field(...)
15
+
* __field(...)
16
+
* __array(...)
17
+
* __count(...)
18
+
* )
19
+
* #include I(REQUEST_END)
20
+
*
21
+
* - starting_index_kind is one of the following, depending on the event:
22
+
*
23
+
* chip_id: hardware chip id or -1 for current hw chip
24
+
* phys_processor_idx:
25
+
* 0xffffffffffffffff: or -1, which means it is irrelavant for the event
26
+
*
27
+
* __count(offset, bytes, name):
28
+
* a counter that should be exposed via perf
29
+
* __field(offset, bytes, name)
30
+
* a normal field
31
+
* __array(offset, bytes, name)
32
+
* an array of bytes
33
+
*
34
+
*
35
+
* @bytes for __count, and __field _must_ be a numeral token
36
+
* in decimal, not an expression and not in hex.
37
+
*
38
+
*
39
+
* TODO:
40
+
* - expose secondary index (if any counter ever uses it, only 0xA0
41
+
* appears to use it right now, and it doesn't have any counters)
42
+
* - embed versioning info
43
+
* - include counter descriptions
44
+
*/
45
+
#define REQUEST_NAME dispatch_timebase_by_processor
46
+
#define REQUEST_NUM 0x10
47
+
#define REQUEST_IDX_KIND "phys_processor_idx=?"
48
+
#include I(REQUEST_BEGIN)
49
+
REQUEST(__count(0, 8, processor_time_in_timebase_cycles)
50
+
__field(0x8, 4, hw_processor_id)
51
+
__field(0xC, 2, owning_part_id)
52
+
__field(0xE, 1, processor_state)
53
+
__field(0xF, 1, version)
54
+
__field(0x10, 4, hw_chip_id)
55
+
__field(0x14, 4, phys_module_id)
56
+
__field(0x18, 4, primary_affinity_domain_idx)
57
+
__field(0x1C, 4, secondary_affinity_domain_idx)
58
+
__field(0x20, 4, processor_version)
59
+
__field(0x24, 2, logical_processor_idx)
60
+
__field(0x26, 2, reserved)
61
+
__field(0x28, 4, processor_id_register)
62
+
__field(0x2C, 4, phys_processor_idx)
63
+
)
64
+
#include I(REQUEST_END)
65
+
66
+
#define REQUEST_NAME system_performance_capabilities
67
+
#define REQUEST_NUM 0x40
68
+
#define REQUEST_IDX_KIND "starting_index=0xffffffffffffffff"
69
+
#include I(REQUEST_BEGIN)
70
+
REQUEST(__field(0, 1, perf_collect_privileged)
71
+
__field(0x1, 1, capability_mask)
72
+
__array(0x2, 0xE, reserved)
73
+
)
74
+
#include I(REQUEST_END)
75
+
76
+
#include "req-gen/_end.h"
+23
arch/powerpc/perf/hv-gpci.c
+23
arch/powerpc/perf/hv-gpci.c
···
31
31
/* u32 */
32
32
EVENT_DEFINE_RANGE_FORMAT(request, config, 0, 31);
33
33
/* u32 */
34
+
/*
35
+
* Note that starting_index, phys_processor_idx, sibling_part_id,
36
+
* hw_chip_id, partition_id all refer to the same bit range. They
37
+
* are basically aliases for the starting_index. The specific alias
38
+
* used depends on the event. See REQUEST_IDX_KIND in hv-gpci-requests.h
39
+
*/
34
40
EVENT_DEFINE_RANGE_FORMAT(starting_index, config, 32, 63);
41
+
EVENT_DEFINE_RANGE_FORMAT_LITE(phys_processor_idx, config, 32, 63);
42
+
EVENT_DEFINE_RANGE_FORMAT_LITE(sibling_part_id, config, 32, 63);
43
+
EVENT_DEFINE_RANGE_FORMAT_LITE(hw_chip_id, config, 32, 63);
44
+
EVENT_DEFINE_RANGE_FORMAT_LITE(partition_id, config, 32, 63);
45
+
35
46
/* u16 */
36
47
EVENT_DEFINE_RANGE_FORMAT(secondary_index, config1, 0, 15);
37
48
/* u8 */
···
55
44
static struct attribute *format_attrs[] = {
56
45
&format_attr_request.attr,
57
46
&format_attr_starting_index.attr,
47
+
&format_attr_phys_processor_idx.attr,
48
+
&format_attr_sibling_part_id.attr,
49
+
&format_attr_hw_chip_id.attr,
50
+
&format_attr_partition_id.attr,
58
51
&format_attr_secondary_index.attr,
59
52
&format_attr_counter_info_version.attr,
60
53
···
70
55
static struct attribute_group format_group = {
71
56
.name = "format",
72
57
.attrs = format_attrs,
58
+
};
59
+
60
+
static struct attribute_group event_group = {
61
+
.name = "events",
62
+
.attrs = hv_gpci_event_attrs,
73
63
};
74
64
75
65
#define HV_CAPS_ATTR(_name, _format) \
···
122
102
123
103
static const struct attribute_group *attr_groups[] = {
124
104
&format_group,
105
+
&event_group,
125
106
&interface_group,
126
107
NULL,
127
108
};
···
285
264
int r;
286
265
unsigned long hret;
287
266
struct hv_perf_caps caps;
267
+
268
+
hv_gpci_assert_offsets_correct();
288
269
289
270
if (!firmware_has_feature(FW_FEATURE_LPAR)) {
290
271
pr_debug("not a virtualized system, not enabling\n");
+12
-25
arch/powerpc/perf/hv-gpci.h
+12
-25
arch/powerpc/perf/hv-gpci.h
···
42
42
*/
43
43
#define COUNTER_INFO_VERSION_CURRENT 0x8
44
44
45
-
/*
46
-
* These determine the counter_value[] layout and the meaning of starting_index
47
-
* and secondary_index.
48
-
*
49
-
* Unless otherwise noted, @secondary_index is unused and ignored.
50
-
*/
51
-
enum counter_info_requests {
52
-
53
-
/* GENERAL */
54
-
55
-
/* @starting_index: must be -1 (to refer to the current partition)
56
-
*/
57
-
CIR_SYSTEM_PERFORMANCE_CAPABILITIES = 0X40,
45
+
/* capability mask masks. */
46
+
enum {
47
+
HV_GPCI_CM_GA = (1 << 7),
48
+
HV_GPCI_CM_EXPANDED = (1 << 6),
49
+
HV_GPCI_CM_LAB = (1 << 5)
58
50
};
59
51
60
-
struct cv_system_performance_capabilities {
61
-
/* If != 0, allowed to collect data from other partitions */
62
-
__u8 perf_collect_privileged;
63
-
64
-
/* These following are only valid if counter_info_version >= 0x3 */
65
-
#define CV_CM_GA (1 << 7)
66
-
#define CV_CM_EXPANDED (1 << 6)
67
-
#define CV_CM_LAB (1 << 5)
68
-
/* remaining bits are reserved */
69
-
__u8 capability_mask;
70
-
__u8 reserved[0xE];
71
-
} __packed;
52
+
#define REQUEST_FILE "../hv-gpci-requests.h"
53
+
#define NAME_LOWER hv_gpci
54
+
#define NAME_UPPER HV_GPCI
55
+
#include "req-gen/perf.h"
56
+
#undef REQUEST_FILE
57
+
#undef NAME_LOWER
58
+
#undef NAME_UPPER
72
59
73
60
#endif
+13
arch/powerpc/perf/req-gen/_begin.h
+13
arch/powerpc/perf/req-gen/_begin.h
···
1
+
/* Include paths to be used in interface defining headers */
2
+
#ifndef POWERPC_PERF_REQ_GEN_H_
3
+
#define POWERPC_PERF_REQ_GEN_H_
4
+
5
+
#define CAT2_STR_(t, s) __stringify(t/s)
6
+
#define CAT2_STR(t, s) CAT2_STR_(t, s)
7
+
#define I(...) __VA_ARGS__
8
+
9
+
#endif
10
+
11
+
#define REQ_GEN_PREFIX req-gen
12
+
#define REQUEST_BEGIN CAT2_STR(REQ_GEN_PREFIX, _request-begin.h)
13
+
#define REQUEST_END CAT2_STR(REQ_GEN_PREFIX, _request-end.h)
+5
arch/powerpc/perf/req-gen/_clear.h
+5
arch/powerpc/perf/req-gen/_clear.h
+4
arch/powerpc/perf/req-gen/_end.h
+4
arch/powerpc/perf/req-gen/_end.h
+15
arch/powerpc/perf/req-gen/_request-begin.h
+15
arch/powerpc/perf/req-gen/_request-begin.h
···
1
+
2
+
#define REQUEST(r_contents) \
3
+
REQUEST_(REQUEST_NAME, REQUEST_NUM, REQUEST_IDX_KIND, I(r_contents))
4
+
5
+
#define __field(f_offset, f_bytes, f_name) \
6
+
__field_(REQUEST_NAME, REQUEST_NUM, REQUEST_IDX_KIND, \
7
+
f_offset, f_bytes, f_name)
8
+
9
+
#define __array(f_offset, f_bytes, f_name) \
10
+
__array_(REQUEST_NAME, REQUEST_NUM, REQUEST_IDX_KIND, \
11
+
f_offset, f_bytes, f_name)
12
+
13
+
#define __count(f_offset, f_bytes, f_name) \
14
+
__count_(REQUEST_NAME, REQUEST_NUM, REQUEST_IDX_KIND, \
15
+
f_offset, f_bytes, f_name)
+8
arch/powerpc/perf/req-gen/_request-end.h
+8
arch/powerpc/perf/req-gen/_request-end.h
+155
arch/powerpc/perf/req-gen/perf.h
+155
arch/powerpc/perf/req-gen/perf.h
···
1
+
#ifndef LINUX_POWERPC_PERF_REQ_GEN_PERF_H_
2
+
#define LINUX_POWERPC_PERF_REQ_GEN_PERF_H_
3
+
4
+
#include <linux/perf_event.h>
5
+
6
+
#ifndef REQUEST_FILE
7
+
#error "REQUEST_FILE must be defined before including"
8
+
#endif
9
+
10
+
#ifndef NAME_LOWER
11
+
#error "NAME_LOWER must be defined before including"
12
+
#endif
13
+
14
+
#ifndef NAME_UPPER
15
+
#error "NAME_UPPER must be defined before including"
16
+
#endif
17
+
18
+
#define BE_TYPE_b1 __u8
19
+
#define BE_TYPE_b2 __be16
20
+
#define BE_TYPE_b4 __be32
21
+
#define BE_TYPE_b8 __be64
22
+
23
+
#define BYTES_TO_BE_TYPE(bytes) \
24
+
BE_TYPE_b##bytes
25
+
26
+
#define CAT2_(a, b) a ## b
27
+
#define CAT2(a, b) CAT2_(a, b)
28
+
#define CAT3_(a, b, c) a ## b ## c
29
+
#define CAT3(a, b, c) CAT3_(a, b, c)
30
+
31
+
/*
32
+
* enumerate the request values as
33
+
* <NAME_UPPER>_<request name> = <request value>
34
+
*/
35
+
#define REQUEST_VALUE__(name_upper, r_name) name_upper ## _ ## r_name
36
+
#define REQUEST_VALUE_(name_upper, r_name) REQUEST_VALUE__(name_upper, r_name)
37
+
#define REQUEST_VALUE(r_name) REQUEST_VALUE_(NAME_UPPER, r_name)
38
+
39
+
#include "_clear.h"
40
+
#define REQUEST_(r_name, r_value, r_idx_1, r_fields) \
41
+
REQUEST_VALUE(r_name) = r_value,
42
+
enum CAT2(NAME_LOWER, _requests) {
43
+
#include REQUEST_FILE
44
+
};
45
+
46
+
/*
47
+
* For each request:
48
+
* struct <NAME_LOWER>_<request name> {
49
+
* r_fields
50
+
* };
51
+
*/
52
+
#include "_clear.h"
53
+
#define STRUCT_NAME__(name_lower, r_name) name_lower ## _ ## r_name
54
+
#define STRUCT_NAME_(name_lower, r_name) STRUCT_NAME__(name_lower, r_name)
55
+
#define STRUCT_NAME(r_name) STRUCT_NAME_(NAME_LOWER, r_name)
56
+
#define REQUEST_(r_name, r_value, r_idx_1, r_fields) \
57
+
struct STRUCT_NAME(r_name) { \
58
+
r_fields \
59
+
};
60
+
#define __field_(r_name, r_value, r_idx_1, f_offset, f_bytes, f_name) \
61
+
BYTES_TO_BE_TYPE(f_bytes) f_name;
62
+
#define __count_(r_name, r_value, r_idx_1, f_offset, f_bytes, f_name) \
63
+
__field_(r_name, r_value, r_idx_1, f_offset, f_bytes, f_name)
64
+
#define __array_(r_name, r_value, r_idx_1, a_offset, a_bytes, a_name) \
65
+
__u8 a_name[a_bytes];
66
+
67
+
#include REQUEST_FILE
68
+
69
+
/*
70
+
* Generate a check of the field offsets
71
+
* <NAME_LOWER>_assert_offsets_correct()
72
+
*/
73
+
#include "_clear.h"
74
+
#define REQUEST_(r_name, r_value, index, r_fields) \
75
+
r_fields
76
+
#define __field_(r_name, r_value, r_idx_1, f_offset, f_size, f_name) \
77
+
BUILD_BUG_ON(offsetof(struct STRUCT_NAME(r_name), f_name) != f_offset);
78
+
#define __count_(r_name, r_value, r_idx_1, c_offset, c_size, c_name) \
79
+
__field_(r_name, r_value, r_idx_1, c_offset, c_size, c_name)
80
+
#define __array_(r_name, r_value, r_idx_1, a_offset, a_size, a_name) \
81
+
__field_(r_name, r_value, r_idx_1, a_offset, a_size, a_name)
82
+
83
+
static inline void CAT2(NAME_LOWER, _assert_offsets_correct)(void)
84
+
{
85
+
#include REQUEST_FILE
86
+
}
87
+
88
+
/*
89
+
* Generate event attributes:
90
+
* PMU_EVENT_ATTR_STRING(<request name>_<field name>,
91
+
* <NAME_LOWER>_event_attr_<request name>_<field name>,
92
+
* "request=<request value>"
93
+
* "starting_index=<starting index type>"
94
+
* "counter_info_version=CURRENT_COUNTER_INFO_VERSION"
95
+
* "length=<f_size>"
96
+
* "offset=<f_offset>")
97
+
*
98
+
* TODO: counter_info_version may need to vary, we should interperate the
99
+
* value to some extent
100
+
*/
101
+
#define EVENT_ATTR_NAME__(name, r_name, c_name) \
102
+
name ## _event_attr_ ## r_name ## _ ## c_name
103
+
#define EVENT_ATTR_NAME_(name, r_name, c_name) \
104
+
EVENT_ATTR_NAME__(name, r_name, c_name)
105
+
#define EVENT_ATTR_NAME(r_name, c_name) \
106
+
EVENT_ATTR_NAME_(NAME_LOWER, r_name, c_name)
107
+
108
+
#include "_clear.h"
109
+
#define __field_(r_name, r_value, r_idx_1, f_offset, f_size, f_name)
110
+
#define __array_(r_name, r_value, r_idx_1, a_offset, a_size, a_name)
111
+
#define __count_(r_name, r_value, r_idx_1, c_offset, c_size, c_name) \
112
+
PMU_EVENT_ATTR_STRING( \
113
+
CAT3(r_name, _, c_name), \
114
+
EVENT_ATTR_NAME(r_name, c_name), \
115
+
"request=" __stringify(r_value) "," \
116
+
r_idx_1 "," \
117
+
"counter_info_version=" \
118
+
__stringify(COUNTER_INFO_VERSION_CURRENT) "," \
119
+
"length=" #c_size "," \
120
+
"offset=" #c_offset)
121
+
#define REQUEST_(r_name, r_value, r_idx_1, r_fields) \
122
+
r_fields
123
+
124
+
#include REQUEST_FILE
125
+
126
+
/*
127
+
* Define event attribute array
128
+
* static struct attribute *hv_gpci_event_attrs[] = {
129
+
* &<NAME_LOWER>_event_attr_<request name>_<field name>.attr,
130
+
* };
131
+
*/
132
+
#include "_clear.h"
133
+
#define __field_(r_name, r_value, r_idx_1, f_offset, f_size, f_name)
134
+
#define __count_(r_name, r_value, r_idx_1, c_offset, c_size, c_name) \
135
+
&EVENT_ATTR_NAME(r_name, c_name).attr.attr,
136
+
#define __array_(r_name, r_value, r_idx_1, a_offset, a_size, a_name)
137
+
#define REQUEST_(r_name, r_value, r_idx_1, r_fields) \
138
+
r_fields
139
+
140
+
static __maybe_unused struct attribute *hv_gpci_event_attrs[] = {
141
+
#include REQUEST_FILE
142
+
NULL
143
+
};
144
+
145
+
/* cleanup */
146
+
#include "_clear.h"
147
+
#undef EVENT_ATTR_NAME
148
+
#undef EVENT_ATTR_NAME_
149
+
#undef BIT_NAME
150
+
#undef BIT_NAME_
151
+
#undef STRUCT_NAME
152
+
#undef REQUEST_VALUE
153
+
#undef REQUEST_VALUE_
154
+
155
+
#endif