1 /*
2 * Data gathering module for Linux-VM Monitor Stream, Stage 1.
3 * Collects misc. OS related data (CPU utilization, running processes).
4 *
5 * Copyright IBM Corp. 2003, 2006
6 *
7 * Author: Gerald Schaefer <gerald.schaefer@de.ibm.com>
8 */
9
10 #define KMSG_COMPONENT "appldata"
11 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/slab.h>
16 #include <linux/errno.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/netdevice.h>
19 #include <linux/sched.h>
20 #include <linux/sched/loadavg.h>
21 #include <linux/sched/stat.h>
22 #include <asm/appldata.h>
23 #include <asm/smp.h>
24
25 #include "appldata.h"
26
27 /*
28 * OS data
29 *
30 * This is accessed as binary data by z/VM. If changes to it can't be avoided,
31 * the structure version (product ID, see appldata_base.c) needs to be changed
32 * as well and all documentation and z/VM applications using it must be
33 * updated.
34 *
35 * The record layout is documented in the Linux for zSeries Device Drivers
36 * book:
37 * http://oss.software.ibm.com/developerworks/opensource/linux390/index.shtml
38 */
39 struct appldata_os_per_cpu {
40 u32 per_cpu_user; /* timer ticks spent in user mode */
41 u32 per_cpu_nice; /* ... spent with modified priority */
42 u32 per_cpu_system; /* ... spent in kernel mode */
43 u32 per_cpu_idle; /* ... spent in idle mode */
44
45 /* New in 2.6 */
46 u32 per_cpu_irq; /* ... spent in interrupts */
47 u32 per_cpu_softirq; /* ... spent in softirqs */
48 u32 per_cpu_iowait; /* ... spent while waiting for I/O */
49
50 /* New in modification level 01 */
51 u32 per_cpu_steal; /* ... stolen by hypervisor */
52 u32 cpu_id; /* number of this CPU */
53 } __attribute__((packed));
54
55 struct appldata_os_data {
56 u64 timestamp;
57 u32 sync_count_1; /* after VM collected the record data, */
58 u32 sync_count_2; /* sync_count_1 and sync_count_2 should be the
59 same. If not, the record has been updated on
60 the Linux side while VM was collecting the
61 (possibly corrupt) data */
62
63 u32 nr_cpus; /* number of (virtual) CPUs */
64 u32 per_cpu_size; /* size of the per-cpu data struct */
65 u32 cpu_offset; /* offset of the first per-cpu data struct */
66
67 u32 nr_running; /* number of runnable threads */
68 u32 nr_threads; /* number of threads */
69 u32 avenrun[3]; /* average nr. of running processes during */
70 /* the last 1, 5 and 15 minutes */
71
72 /* New in 2.6 */
73 u32 nr_iowait; /* number of blocked threads
74 (waiting for I/O) */
75
76 /* per cpu data */
77 struct appldata_os_per_cpu os_cpu[0];
78 } __attribute__((packed));
79
80 static struct appldata_os_data *appldata_os_data;
81
82 static struct appldata_ops ops = {
83 .name = "os",
84 .record_nr = APPLDATA_RECORD_OS_ID,
85 .owner = THIS_MODULE,
86 .mod_lvl = {0xF0, 0xF1}, /* EBCDIC "01" */
87 };
88
89
90 /*
91 * appldata_get_os_data()
92 *
93 * gather OS data
94 */
appldata_get_os_data(void * data)95 static void appldata_get_os_data(void *data)
96 {
97 int i, j, rc;
98 struct appldata_os_data *os_data;
99 unsigned int new_size;
100
101 os_data = data;
102 os_data->sync_count_1++;
103
104 os_data->nr_threads = nr_threads;
105 os_data->nr_running = nr_running();
106 os_data->nr_iowait = nr_iowait();
107 os_data->avenrun[0] = avenrun[0] + (FIXED_1/200);
108 os_data->avenrun[1] = avenrun[1] + (FIXED_1/200);
109 os_data->avenrun[2] = avenrun[2] + (FIXED_1/200);
110
111 j = 0;
112 for_each_online_cpu(i) {
113 os_data->os_cpu[j].per_cpu_user =
114 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_USER]);
115 os_data->os_cpu[j].per_cpu_nice =
116 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_NICE]);
117 os_data->os_cpu[j].per_cpu_system =
118 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_SYSTEM]);
119 os_data->os_cpu[j].per_cpu_idle =
120 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IDLE]);
121 os_data->os_cpu[j].per_cpu_irq =
122 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IRQ]);
123 os_data->os_cpu[j].per_cpu_softirq =
124 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_SOFTIRQ]);
125 os_data->os_cpu[j].per_cpu_iowait =
126 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IOWAIT]);
127 os_data->os_cpu[j].per_cpu_steal =
128 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_STEAL]);
129 os_data->os_cpu[j].cpu_id = i;
130 j++;
131 }
132
133 os_data->nr_cpus = j;
134
135 new_size = sizeof(struct appldata_os_data) +
136 (os_data->nr_cpus * sizeof(struct appldata_os_per_cpu));
137 if (ops.size != new_size) {
138 if (ops.active) {
139 rc = appldata_diag(APPLDATA_RECORD_OS_ID,
140 APPLDATA_START_INTERVAL_REC,
141 (unsigned long) ops.data, new_size,
142 ops.mod_lvl);
143 if (rc != 0)
144 pr_err("Starting a new OS data collection "
145 "failed with rc=%d\n", rc);
146
147 rc = appldata_diag(APPLDATA_RECORD_OS_ID,
148 APPLDATA_STOP_REC,
149 (unsigned long) ops.data, ops.size,
150 ops.mod_lvl);
151 if (rc != 0)
152 pr_err("Stopping a faulty OS data "
153 "collection failed with rc=%d\n", rc);
154 }
155 ops.size = new_size;
156 }
157 os_data->timestamp = get_tod_clock();
158 os_data->sync_count_2++;
159 }
160
161
162 /*
163 * appldata_os_init()
164 *
165 * init data, register ops
166 */
appldata_os_init(void)167 static int __init appldata_os_init(void)
168 {
169 int rc, max_size;
170
171 max_size = sizeof(struct appldata_os_data) +
172 (num_possible_cpus() * sizeof(struct appldata_os_per_cpu));
173 if (max_size > APPLDATA_MAX_REC_SIZE) {
174 pr_err("Maximum OS record size %i exceeds the maximum "
175 "record size %i\n", max_size, APPLDATA_MAX_REC_SIZE);
176 rc = -ENOMEM;
177 goto out;
178 }
179
180 appldata_os_data = kzalloc(max_size, GFP_KERNEL | GFP_DMA);
181 if (appldata_os_data == NULL) {
182 rc = -ENOMEM;
183 goto out;
184 }
185
186 appldata_os_data->per_cpu_size = sizeof(struct appldata_os_per_cpu);
187 appldata_os_data->cpu_offset = offsetof(struct appldata_os_data,
188 os_cpu);
189
190 ops.data = appldata_os_data;
191 ops.callback = &appldata_get_os_data;
192 rc = appldata_register_ops(&ops);
193 if (rc != 0)
194 kfree(appldata_os_data);
195 out:
196 return rc;
197 }
198
199 /*
200 * appldata_os_exit()
201 *
202 * unregister ops
203 */
appldata_os_exit(void)204 static void __exit appldata_os_exit(void)
205 {
206 appldata_unregister_ops(&ops);
207 kfree(appldata_os_data);
208 }
209
210
211 module_init(appldata_os_init);
212 module_exit(appldata_os_exit);
213
214 MODULE_LICENSE("GPL");
215 MODULE_AUTHOR("Gerald Schaefer");
216 MODULE_DESCRIPTION("Linux-VM Monitor Stream, OS statistics");
217