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 <asm/appldata.h>
21#include <asm/smp.h>
22
23#include "appldata.h"
24
25
26#define LOAD_INT(x) ((x) >> FSHIFT)
27#define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100)
28
29/*
30 * OS data
31 *
32 * This is accessed as binary data by z/VM. If changes to it can't be avoided,
33 * the structure version (product ID, see appldata_base.c) needs to be changed
34 * as well and all documentation and z/VM applications using it must be
35 * updated.
36 *
37 * The record layout is documented in the Linux for zSeries Device Drivers
38 * book:
39 * http://oss.software.ibm.com/developerworks/opensource/linux390/index.shtml
40 */
41struct appldata_os_per_cpu {
42	u32 per_cpu_user;	/* timer ticks spent in user mode   */
43	u32 per_cpu_nice;	/* ... spent with modified priority */
44	u32 per_cpu_system;	/* ... spent in kernel mode         */
45	u32 per_cpu_idle;	/* ... spent in idle mode           */
46
47	/* New in 2.6 */
48	u32 per_cpu_irq;	/* ... spent in interrupts          */
49	u32 per_cpu_softirq;	/* ... spent in softirqs            */
50	u32 per_cpu_iowait;	/* ... spent while waiting for I/O  */
51
52	/* New in modification level 01 */
53	u32 per_cpu_steal;	/* ... stolen by hypervisor	    */
54	u32 cpu_id;		/* number of this CPU		    */
55} __attribute__((packed));
56
57struct appldata_os_data {
58	u64 timestamp;
59	u32 sync_count_1;	/* after VM collected the record data, */
60	u32 sync_count_2;	/* sync_count_1 and sync_count_2 should be the
61				   same. If not, the record has been updated on
62				   the Linux side while VM was collecting the
63				   (possibly corrupt) data */
64
65	u32 nr_cpus;		/* number of (virtual) CPUs        */
66	u32 per_cpu_size;	/* size of the per-cpu data struct */
67	u32 cpu_offset;		/* offset of the first per-cpu data struct */
68
69	u32 nr_running;		/* number of runnable threads      */
70	u32 nr_threads;		/* number of threads               */
71	u32 avenrun[3];		/* average nr. of running processes during */
72				/* the last 1, 5 and 15 minutes */
73
74	/* New in 2.6 */
75	u32 nr_iowait;		/* number of blocked threads
76				   (waiting for I/O)               */
77
78	/* per cpu data */
79	struct appldata_os_per_cpu os_cpu[0];
80} __attribute__((packed));
81
82static struct appldata_os_data *appldata_os_data;
83
84static struct appldata_ops ops = {
85	.name	   = "os",
86	.record_nr = APPLDATA_RECORD_OS_ID,
87	.owner	   = THIS_MODULE,
88	.mod_lvl   = {0xF0, 0xF1},		/* EBCDIC "01" */
89};
90
91
92/*
93 * appldata_get_os_data()
94 *
95 * gather OS data
96 */
97static void appldata_get_os_data(void *data)
98{
99	int i, j, rc;
100	struct appldata_os_data *os_data;
101	unsigned int new_size;
102
103	os_data = data;
104	os_data->sync_count_1++;
105
106	os_data->nr_threads = nr_threads;
107	os_data->nr_running = nr_running();
108	os_data->nr_iowait  = nr_iowait();
109	os_data->avenrun[0] = avenrun[0] + (FIXED_1/200);
110	os_data->avenrun[1] = avenrun[1] + (FIXED_1/200);
111	os_data->avenrun[2] = avenrun[2] + (FIXED_1/200);
112
113	j = 0;
114	for_each_online_cpu(i) {
115		os_data->os_cpu[j].per_cpu_user =
116			cputime_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_USER]);
117		os_data->os_cpu[j].per_cpu_nice =
118			cputime_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_NICE]);
119		os_data->os_cpu[j].per_cpu_system =
120			cputime_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_SYSTEM]);
121		os_data->os_cpu[j].per_cpu_idle =
122			cputime_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IDLE]);
123		os_data->os_cpu[j].per_cpu_irq =
124			cputime_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IRQ]);
125		os_data->os_cpu[j].per_cpu_softirq =
126			cputime_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_SOFTIRQ]);
127		os_data->os_cpu[j].per_cpu_iowait =
128			cputime_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IOWAIT]);
129		os_data->os_cpu[j].per_cpu_steal =
130			cputime_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_STEAL]);
131		os_data->os_cpu[j].cpu_id = i;
132		j++;
133	}
134
135	os_data->nr_cpus = j;
136
137	new_size = sizeof(struct appldata_os_data) +
138		   (os_data->nr_cpus * sizeof(struct appldata_os_per_cpu));
139	if (ops.size != new_size) {
140		if (ops.active) {
141			rc = appldata_diag(APPLDATA_RECORD_OS_ID,
142					   APPLDATA_START_INTERVAL_REC,
143					   (unsigned long) ops.data, new_size,
144					   ops.mod_lvl);
145			if (rc != 0)
146				pr_err("Starting a new OS data collection "
147				       "failed with rc=%d\n", rc);
148
149			rc = appldata_diag(APPLDATA_RECORD_OS_ID,
150					   APPLDATA_STOP_REC,
151					   (unsigned long) ops.data, ops.size,
152					   ops.mod_lvl);
153			if (rc != 0)
154				pr_err("Stopping a faulty OS data "
155				       "collection failed with rc=%d\n", rc);
156		}
157		ops.size = new_size;
158	}
159	os_data->timestamp = get_tod_clock();
160	os_data->sync_count_2++;
161}
162
163
164/*
165 * appldata_os_init()
166 *
167 * init data, register ops
168 */
169static int __init appldata_os_init(void)
170{
171	int rc, max_size;
172
173	max_size = sizeof(struct appldata_os_data) +
174		   (num_possible_cpus() * sizeof(struct appldata_os_per_cpu));
175	if (max_size > APPLDATA_MAX_REC_SIZE) {
176		pr_err("Maximum OS record size %i exceeds the maximum "
177		       "record size %i\n", max_size, APPLDATA_MAX_REC_SIZE);
178		rc = -ENOMEM;
179		goto out;
180	}
181
182	appldata_os_data = kzalloc(max_size, GFP_KERNEL | GFP_DMA);
183	if (appldata_os_data == NULL) {
184		rc = -ENOMEM;
185		goto out;
186	}
187
188	appldata_os_data->per_cpu_size = sizeof(struct appldata_os_per_cpu);
189	appldata_os_data->cpu_offset   = offsetof(struct appldata_os_data,
190							os_cpu);
191
192	ops.data = appldata_os_data;
193	ops.callback  = &appldata_get_os_data;
194	rc = appldata_register_ops(&ops);
195	if (rc != 0)
196		kfree(appldata_os_data);
197out:
198	return rc;
199}
200
201/*
202 * appldata_os_exit()
203 *
204 * unregister ops
205 */
206static void __exit appldata_os_exit(void)
207{
208	appldata_unregister_ops(&ops);
209	kfree(appldata_os_data);
210}
211
212
213module_init(appldata_os_init);
214module_exit(appldata_os_exit);
215
216MODULE_LICENSE("GPL");
217MODULE_AUTHOR("Gerald Schaefer");
218MODULE_DESCRIPTION("Linux-VM Monitor Stream, OS statistics");
219