xref: /openbmc/linux/arch/s390/appldata/appldata_os.c (revision 7aacf86b)
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 #define LOAD_INT(x) ((x) >> FSHIFT)
29 #define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100)
30 
31 /*
32  * OS data
33  *
34  * This is accessed as binary data by z/VM. If changes to it can't be avoided,
35  * the structure version (product ID, see appldata_base.c) needs to be changed
36  * as well and all documentation and z/VM applications using it must be
37  * updated.
38  *
39  * The record layout is documented in the Linux for zSeries Device Drivers
40  * book:
41  * http://oss.software.ibm.com/developerworks/opensource/linux390/index.shtml
42  */
43 struct appldata_os_per_cpu {
44 	u32 per_cpu_user;	/* timer ticks spent in user mode   */
45 	u32 per_cpu_nice;	/* ... spent with modified priority */
46 	u32 per_cpu_system;	/* ... spent in kernel mode         */
47 	u32 per_cpu_idle;	/* ... spent in idle mode           */
48 
49 	/* New in 2.6 */
50 	u32 per_cpu_irq;	/* ... spent in interrupts          */
51 	u32 per_cpu_softirq;	/* ... spent in softirqs            */
52 	u32 per_cpu_iowait;	/* ... spent while waiting for I/O  */
53 
54 	/* New in modification level 01 */
55 	u32 per_cpu_steal;	/* ... stolen by hypervisor	    */
56 	u32 cpu_id;		/* number of this CPU		    */
57 } __attribute__((packed));
58 
59 struct appldata_os_data {
60 	u64 timestamp;
61 	u32 sync_count_1;	/* after VM collected the record data, */
62 	u32 sync_count_2;	/* sync_count_1 and sync_count_2 should be the
63 				   same. If not, the record has been updated on
64 				   the Linux side while VM was collecting the
65 				   (possibly corrupt) data */
66 
67 	u32 nr_cpus;		/* number of (virtual) CPUs        */
68 	u32 per_cpu_size;	/* size of the per-cpu data struct */
69 	u32 cpu_offset;		/* offset of the first per-cpu data struct */
70 
71 	u32 nr_running;		/* number of runnable threads      */
72 	u32 nr_threads;		/* number of threads               */
73 	u32 avenrun[3];		/* average nr. of running processes during */
74 				/* the last 1, 5 and 15 minutes */
75 
76 	/* New in 2.6 */
77 	u32 nr_iowait;		/* number of blocked threads
78 				   (waiting for I/O)               */
79 
80 	/* per cpu data */
81 	struct appldata_os_per_cpu os_cpu[0];
82 } __attribute__((packed));
83 
84 static struct appldata_os_data *appldata_os_data;
85 
86 static struct appldata_ops ops = {
87 	.name	   = "os",
88 	.record_nr = APPLDATA_RECORD_OS_ID,
89 	.owner	   = THIS_MODULE,
90 	.mod_lvl   = {0xF0, 0xF1},		/* EBCDIC "01" */
91 };
92 
93 
94 /*
95  * appldata_get_os_data()
96  *
97  * gather OS data
98  */
99 static void appldata_get_os_data(void *data)
100 {
101 	int i, j, rc;
102 	struct appldata_os_data *os_data;
103 	unsigned int new_size;
104 
105 	os_data = data;
106 	os_data->sync_count_1++;
107 
108 	os_data->nr_threads = nr_threads;
109 	os_data->nr_running = nr_running();
110 	os_data->nr_iowait  = nr_iowait();
111 	os_data->avenrun[0] = avenrun[0] + (FIXED_1/200);
112 	os_data->avenrun[1] = avenrun[1] + (FIXED_1/200);
113 	os_data->avenrun[2] = avenrun[2] + (FIXED_1/200);
114 
115 	j = 0;
116 	for_each_online_cpu(i) {
117 		os_data->os_cpu[j].per_cpu_user =
118 			nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_USER]);
119 		os_data->os_cpu[j].per_cpu_nice =
120 			nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_NICE]);
121 		os_data->os_cpu[j].per_cpu_system =
122 			nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_SYSTEM]);
123 		os_data->os_cpu[j].per_cpu_idle =
124 			nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IDLE]);
125 		os_data->os_cpu[j].per_cpu_irq =
126 			nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IRQ]);
127 		os_data->os_cpu[j].per_cpu_softirq =
128 			nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_SOFTIRQ]);
129 		os_data->os_cpu[j].per_cpu_iowait =
130 			nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IOWAIT]);
131 		os_data->os_cpu[j].per_cpu_steal =
132 			nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_STEAL]);
133 		os_data->os_cpu[j].cpu_id = i;
134 		j++;
135 	}
136 
137 	os_data->nr_cpus = j;
138 
139 	new_size = sizeof(struct appldata_os_data) +
140 		   (os_data->nr_cpus * sizeof(struct appldata_os_per_cpu));
141 	if (ops.size != new_size) {
142 		if (ops.active) {
143 			rc = appldata_diag(APPLDATA_RECORD_OS_ID,
144 					   APPLDATA_START_INTERVAL_REC,
145 					   (unsigned long) ops.data, new_size,
146 					   ops.mod_lvl);
147 			if (rc != 0)
148 				pr_err("Starting a new OS data collection "
149 				       "failed with rc=%d\n", rc);
150 
151 			rc = appldata_diag(APPLDATA_RECORD_OS_ID,
152 					   APPLDATA_STOP_REC,
153 					   (unsigned long) ops.data, ops.size,
154 					   ops.mod_lvl);
155 			if (rc != 0)
156 				pr_err("Stopping a faulty OS data "
157 				       "collection failed with rc=%d\n", rc);
158 		}
159 		ops.size = new_size;
160 	}
161 	os_data->timestamp = get_tod_clock();
162 	os_data->sync_count_2++;
163 }
164 
165 
166 /*
167  * appldata_os_init()
168  *
169  * init data, register ops
170  */
171 static int __init appldata_os_init(void)
172 {
173 	int rc, max_size;
174 
175 	max_size = sizeof(struct appldata_os_data) +
176 		   (num_possible_cpus() * sizeof(struct appldata_os_per_cpu));
177 	if (max_size > APPLDATA_MAX_REC_SIZE) {
178 		pr_err("Maximum OS record size %i exceeds the maximum "
179 		       "record size %i\n", max_size, APPLDATA_MAX_REC_SIZE);
180 		rc = -ENOMEM;
181 		goto out;
182 	}
183 
184 	appldata_os_data = kzalloc(max_size, GFP_KERNEL | GFP_DMA);
185 	if (appldata_os_data == NULL) {
186 		rc = -ENOMEM;
187 		goto out;
188 	}
189 
190 	appldata_os_data->per_cpu_size = sizeof(struct appldata_os_per_cpu);
191 	appldata_os_data->cpu_offset   = offsetof(struct appldata_os_data,
192 							os_cpu);
193 
194 	ops.data = appldata_os_data;
195 	ops.callback  = &appldata_get_os_data;
196 	rc = appldata_register_ops(&ops);
197 	if (rc != 0)
198 		kfree(appldata_os_data);
199 out:
200 	return rc;
201 }
202 
203 /*
204  * appldata_os_exit()
205  *
206  * unregister ops
207  */
208 static void __exit appldata_os_exit(void)
209 {
210 	appldata_unregister_ops(&ops);
211 	kfree(appldata_os_data);
212 }
213 
214 
215 module_init(appldata_os_init);
216 module_exit(appldata_os_exit);
217 
218 MODULE_LICENSE("GPL");
219 MODULE_AUTHOR("Gerald Schaefer");
220 MODULE_DESCRIPTION("Linux-VM Monitor Stream, OS statistics");
221