xref: /openbmc/linux/arch/s390/appldata/appldata_os.c (revision a1e58bbd)
1 /*
2  * arch/s390/appldata/appldata_os.c
3  *
4  * Data gathering module for Linux-VM Monitor Stream, Stage 1.
5  * Collects misc. OS related data (CPU utilization, running processes).
6  *
7  * Copyright (C) 2003,2006 IBM Corporation, IBM Deutschland Entwicklung GmbH.
8  *
9  * Author: Gerald Schaefer <gerald.schaefer@de.ibm.com>
10  */
11 
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/slab.h>
15 #include <linux/errno.h>
16 #include <linux/kernel_stat.h>
17 #include <linux/netdevice.h>
18 #include <linux/sched.h>
19 #include <asm/appldata.h>
20 #include <asm/smp.h>
21 
22 #include "appldata.h"
23 
24 
25 #define MY_PRINT_NAME	"appldata_os"		/* for debug messages, etc. */
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  */
41 struct 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 
57 struct 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 
82 static struct appldata_os_data *appldata_os_data;
83 
84 static 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 static inline void appldata_print_debug(struct appldata_os_data *os_data)
93 {
94 	int a0, a1, a2, i;
95 
96 	P_DEBUG("--- OS - RECORD ---\n");
97 	P_DEBUG("nr_threads   = %u\n", os_data->nr_threads);
98 	P_DEBUG("nr_running   = %u\n", os_data->nr_running);
99 	P_DEBUG("nr_iowait    = %u\n", os_data->nr_iowait);
100 	P_DEBUG("avenrun(int) = %8x / %8x / %8x\n", os_data->avenrun[0],
101 		os_data->avenrun[1], os_data->avenrun[2]);
102 	a0 = os_data->avenrun[0];
103 	a1 = os_data->avenrun[1];
104 	a2 = os_data->avenrun[2];
105 	P_DEBUG("avenrun(float) = %d.%02d / %d.%02d / %d.%02d\n",
106 		LOAD_INT(a0), LOAD_FRAC(a0), LOAD_INT(a1), LOAD_FRAC(a1),
107 		LOAD_INT(a2), LOAD_FRAC(a2));
108 
109 	P_DEBUG("nr_cpus = %u\n", os_data->nr_cpus);
110 	for (i = 0; i < os_data->nr_cpus; i++) {
111 		P_DEBUG("cpu%u : user = %u, nice = %u, system = %u, "
112 			"idle = %u, irq = %u, softirq = %u, iowait = %u, "
113 			"steal = %u\n",
114 				os_data->os_cpu[i].cpu_id,
115 				os_data->os_cpu[i].per_cpu_user,
116 				os_data->os_cpu[i].per_cpu_nice,
117 				os_data->os_cpu[i].per_cpu_system,
118 				os_data->os_cpu[i].per_cpu_idle,
119 				os_data->os_cpu[i].per_cpu_irq,
120 				os_data->os_cpu[i].per_cpu_softirq,
121 				os_data->os_cpu[i].per_cpu_iowait,
122 				os_data->os_cpu[i].per_cpu_steal);
123 	}
124 
125 	P_DEBUG("sync_count_1 = %u\n", os_data->sync_count_1);
126 	P_DEBUG("sync_count_2 = %u\n", os_data->sync_count_2);
127 	P_DEBUG("timestamp    = %lX\n", os_data->timestamp);
128 }
129 
130 /*
131  * appldata_get_os_data()
132  *
133  * gather OS data
134  */
135 static void appldata_get_os_data(void *data)
136 {
137 	int i, j, rc;
138 	struct appldata_os_data *os_data;
139 	unsigned int new_size;
140 
141 	os_data = data;
142 	os_data->sync_count_1++;
143 
144 	os_data->nr_threads = nr_threads;
145 	os_data->nr_running = nr_running();
146 	os_data->nr_iowait  = nr_iowait();
147 	os_data->avenrun[0] = avenrun[0] + (FIXED_1/200);
148 	os_data->avenrun[1] = avenrun[1] + (FIXED_1/200);
149 	os_data->avenrun[2] = avenrun[2] + (FIXED_1/200);
150 
151 	j = 0;
152 	for_each_online_cpu(i) {
153 		os_data->os_cpu[j].per_cpu_user =
154 			cputime_to_jiffies(kstat_cpu(i).cpustat.user);
155 		os_data->os_cpu[j].per_cpu_nice =
156 			cputime_to_jiffies(kstat_cpu(i).cpustat.nice);
157 		os_data->os_cpu[j].per_cpu_system =
158 			cputime_to_jiffies(kstat_cpu(i).cpustat.system);
159 		os_data->os_cpu[j].per_cpu_idle =
160 			cputime_to_jiffies(kstat_cpu(i).cpustat.idle);
161 		os_data->os_cpu[j].per_cpu_irq =
162 			cputime_to_jiffies(kstat_cpu(i).cpustat.irq);
163 		os_data->os_cpu[j].per_cpu_softirq =
164 			cputime_to_jiffies(kstat_cpu(i).cpustat.softirq);
165 		os_data->os_cpu[j].per_cpu_iowait =
166 			cputime_to_jiffies(kstat_cpu(i).cpustat.iowait);
167 		os_data->os_cpu[j].per_cpu_steal =
168 			cputime_to_jiffies(kstat_cpu(i).cpustat.steal);
169 		os_data->os_cpu[j].cpu_id = i;
170 		j++;
171 	}
172 
173 	os_data->nr_cpus = j;
174 
175 	new_size = sizeof(struct appldata_os_data) +
176 		   (os_data->nr_cpus * sizeof(struct appldata_os_per_cpu));
177 	if (ops.size != new_size) {
178 		if (ops.active) {
179 			rc = appldata_diag(APPLDATA_RECORD_OS_ID,
180 					   APPLDATA_START_INTERVAL_REC,
181 					   (unsigned long) ops.data, new_size,
182 					   ops.mod_lvl);
183 			if (rc != 0) {
184 				P_ERROR("os: START NEW DIAG 0xDC failed, "
185 					"return code: %d, new size = %i\n", rc,
186 					new_size);
187 				P_INFO("os: stopping old record now\n");
188 			} else
189 				P_INFO("os: new record size = %i\n", new_size);
190 
191 			rc = appldata_diag(APPLDATA_RECORD_OS_ID,
192 					   APPLDATA_STOP_REC,
193 					   (unsigned long) ops.data, ops.size,
194 					   ops.mod_lvl);
195 			if (rc != 0)
196 				P_ERROR("os: STOP OLD DIAG 0xDC failed, "
197 					"return code: %d, old size = %i\n", rc,
198 					ops.size);
199 			else
200 				P_INFO("os: old record size = %i stopped\n",
201 					ops.size);
202 		}
203 		ops.size = new_size;
204 	}
205 	os_data->timestamp = get_clock();
206 	os_data->sync_count_2++;
207 #ifdef APPLDATA_DEBUG
208 	appldata_print_debug(os_data);
209 #endif
210 }
211 
212 
213 /*
214  * appldata_os_init()
215  *
216  * init data, register ops
217  */
218 static int __init appldata_os_init(void)
219 {
220 	int rc, max_size;
221 
222 	max_size = sizeof(struct appldata_os_data) +
223 		   (NR_CPUS * sizeof(struct appldata_os_per_cpu));
224 	if (max_size > APPLDATA_MAX_REC_SIZE) {
225 		P_ERROR("Max. size of OS record = %i, bigger than maximum "
226 			"record size (%i)\n", max_size, APPLDATA_MAX_REC_SIZE);
227 		rc = -ENOMEM;
228 		goto out;
229 	}
230 	P_DEBUG("max. sizeof(os) = %i, sizeof(os_cpu) = %lu\n", max_size,
231 		sizeof(struct appldata_os_per_cpu));
232 
233 	appldata_os_data = kzalloc(max_size, GFP_DMA);
234 	if (appldata_os_data == NULL) {
235 		P_ERROR("No memory for %s!\n", ops.name);
236 		rc = -ENOMEM;
237 		goto out;
238 	}
239 
240 	appldata_os_data->per_cpu_size = sizeof(struct appldata_os_per_cpu);
241 	appldata_os_data->cpu_offset   = offsetof(struct appldata_os_data,
242 							os_cpu);
243 	P_DEBUG("cpu offset = %u\n", appldata_os_data->cpu_offset);
244 
245 	ops.data = appldata_os_data;
246 	ops.callback  = &appldata_get_os_data;
247 	rc = appldata_register_ops(&ops);
248 	if (rc != 0) {
249 		P_ERROR("Error registering ops, rc = %i\n", rc);
250 		kfree(appldata_os_data);
251 	} else {
252 		P_DEBUG("%s-ops registered!\n", ops.name);
253 	}
254 out:
255 	return rc;
256 }
257 
258 /*
259  * appldata_os_exit()
260  *
261  * unregister ops
262  */
263 static void __exit appldata_os_exit(void)
264 {
265 	appldata_unregister_ops(&ops);
266 	kfree(appldata_os_data);
267 	P_DEBUG("%s-ops unregistered!\n", ops.name);
268 }
269 
270 
271 module_init(appldata_os_init);
272 module_exit(appldata_os_exit);
273 
274 MODULE_LICENSE("GPL");
275 MODULE_AUTHOR("Gerald Schaefer");
276 MODULE_DESCRIPTION("Linux-VM Monitor Stream, OS statistics");
277