1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Data gathering module for Linux-VM Monitor Stream, Stage 1. 4 * Collects misc. OS related data (CPU utilization, running processes). 5 * 6 * Copyright IBM Corp. 2003, 2006 7 * 8 * Author: Gerald Schaefer <gerald.schaefer@de.ibm.com> 9 */ 10 11 #define KMSG_COMPONENT "appldata" 12 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 13 14 #include <linux/module.h> 15 #include <linux/init.h> 16 #include <linux/slab.h> 17 #include <linux/errno.h> 18 #include <linux/kernel_stat.h> 19 #include <linux/netdevice.h> 20 #include <linux/sched.h> 21 #include <linux/sched/loadavg.h> 22 #include <linux/sched/stat.h> 23 #include <asm/appldata.h> 24 #include <asm/smp.h> 25 26 #include "appldata.h" 27 28 29 #define LOAD_INT(x) ((x) >> FSHIFT) 30 #define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100) 31 32 /* 33 * OS data 34 * 35 * This is accessed as binary data by z/VM. If changes to it can't be avoided, 36 * the structure version (product ID, see appldata_base.c) needs to be changed 37 * as well and all documentation and z/VM applications using it must be 38 * updated. 39 * 40 * The record layout is documented in the Linux for zSeries Device Drivers 41 * book: 42 * http://oss.software.ibm.com/developerworks/opensource/linux390/index.shtml 43 */ 44 struct appldata_os_per_cpu { 45 u32 per_cpu_user; /* timer ticks spent in user mode */ 46 u32 per_cpu_nice; /* ... spent with modified priority */ 47 u32 per_cpu_system; /* ... spent in kernel mode */ 48 u32 per_cpu_idle; /* ... spent in idle mode */ 49 50 /* New in 2.6 */ 51 u32 per_cpu_irq; /* ... spent in interrupts */ 52 u32 per_cpu_softirq; /* ... spent in softirqs */ 53 u32 per_cpu_iowait; /* ... spent while waiting for I/O */ 54 55 /* New in modification level 01 */ 56 u32 per_cpu_steal; /* ... stolen by hypervisor */ 57 u32 cpu_id; /* number of this CPU */ 58 } __attribute__((packed)); 59 60 struct appldata_os_data { 61 u64 timestamp; 62 u32 sync_count_1; /* after VM collected the record data, */ 63 u32 sync_count_2; /* sync_count_1 and sync_count_2 should be the 64 same. If not, the record has been updated on 65 the Linux side while VM was collecting the 66 (possibly corrupt) data */ 67 68 u32 nr_cpus; /* number of (virtual) CPUs */ 69 u32 per_cpu_size; /* size of the per-cpu data struct */ 70 u32 cpu_offset; /* offset of the first per-cpu data struct */ 71 72 u32 nr_running; /* number of runnable threads */ 73 u32 nr_threads; /* number of threads */ 74 u32 avenrun[3]; /* average nr. of running processes during */ 75 /* the last 1, 5 and 15 minutes */ 76 77 /* New in 2.6 */ 78 u32 nr_iowait; /* number of blocked threads 79 (waiting for I/O) */ 80 81 /* per cpu data */ 82 struct appldata_os_per_cpu os_cpu[0]; 83 } __attribute__((packed)); 84 85 static struct appldata_os_data *appldata_os_data; 86 87 static struct appldata_ops ops = { 88 .name = "os", 89 .record_nr = APPLDATA_RECORD_OS_ID, 90 .owner = THIS_MODULE, 91 .mod_lvl = {0xF0, 0xF1}, /* EBCDIC "01" */ 92 }; 93 94 95 /* 96 * appldata_get_os_data() 97 * 98 * gather OS data 99 */ 100 static void appldata_get_os_data(void *data) 101 { 102 int i, j, rc; 103 struct appldata_os_data *os_data; 104 unsigned int new_size; 105 106 os_data = data; 107 os_data->sync_count_1++; 108 109 os_data->nr_threads = nr_threads; 110 os_data->nr_running = nr_running(); 111 os_data->nr_iowait = nr_iowait(); 112 os_data->avenrun[0] = avenrun[0] + (FIXED_1/200); 113 os_data->avenrun[1] = avenrun[1] + (FIXED_1/200); 114 os_data->avenrun[2] = avenrun[2] + (FIXED_1/200); 115 116 j = 0; 117 for_each_online_cpu(i) { 118 os_data->os_cpu[j].per_cpu_user = 119 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_USER]); 120 os_data->os_cpu[j].per_cpu_nice = 121 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_NICE]); 122 os_data->os_cpu[j].per_cpu_system = 123 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_SYSTEM]); 124 os_data->os_cpu[j].per_cpu_idle = 125 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IDLE]); 126 os_data->os_cpu[j].per_cpu_irq = 127 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IRQ]); 128 os_data->os_cpu[j].per_cpu_softirq = 129 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_SOFTIRQ]); 130 os_data->os_cpu[j].per_cpu_iowait = 131 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_IOWAIT]); 132 os_data->os_cpu[j].per_cpu_steal = 133 nsecs_to_jiffies(kcpustat_cpu(i).cpustat[CPUTIME_STEAL]); 134 os_data->os_cpu[j].cpu_id = i; 135 j++; 136 } 137 138 os_data->nr_cpus = j; 139 140 new_size = sizeof(struct appldata_os_data) + 141 (os_data->nr_cpus * sizeof(struct appldata_os_per_cpu)); 142 if (ops.size != new_size) { 143 if (ops.active) { 144 rc = appldata_diag(APPLDATA_RECORD_OS_ID, 145 APPLDATA_START_INTERVAL_REC, 146 (unsigned long) ops.data, new_size, 147 ops.mod_lvl); 148 if (rc != 0) 149 pr_err("Starting a new OS data collection " 150 "failed with rc=%d\n", rc); 151 152 rc = appldata_diag(APPLDATA_RECORD_OS_ID, 153 APPLDATA_STOP_REC, 154 (unsigned long) ops.data, ops.size, 155 ops.mod_lvl); 156 if (rc != 0) 157 pr_err("Stopping a faulty OS data " 158 "collection failed with rc=%d\n", rc); 159 } 160 ops.size = new_size; 161 } 162 os_data->timestamp = get_tod_clock(); 163 os_data->sync_count_2++; 164 } 165 166 167 /* 168 * appldata_os_init() 169 * 170 * init data, register ops 171 */ 172 static int __init appldata_os_init(void) 173 { 174 int rc, max_size; 175 176 max_size = sizeof(struct appldata_os_data) + 177 (num_possible_cpus() * sizeof(struct appldata_os_per_cpu)); 178 if (max_size > APPLDATA_MAX_REC_SIZE) { 179 pr_err("Maximum OS record size %i exceeds the maximum " 180 "record size %i\n", max_size, APPLDATA_MAX_REC_SIZE); 181 rc = -ENOMEM; 182 goto out; 183 } 184 185 appldata_os_data = kzalloc(max_size, GFP_KERNEL | GFP_DMA); 186 if (appldata_os_data == NULL) { 187 rc = -ENOMEM; 188 goto out; 189 } 190 191 appldata_os_data->per_cpu_size = sizeof(struct appldata_os_per_cpu); 192 appldata_os_data->cpu_offset = offsetof(struct appldata_os_data, 193 os_cpu); 194 195 ops.data = appldata_os_data; 196 ops.callback = &appldata_get_os_data; 197 rc = appldata_register_ops(&ops); 198 if (rc != 0) 199 kfree(appldata_os_data); 200 out: 201 return rc; 202 } 203 204 /* 205 * appldata_os_exit() 206 * 207 * unregister ops 208 */ 209 static void __exit appldata_os_exit(void) 210 { 211 appldata_unregister_ops(&ops); 212 kfree(appldata_os_data); 213 } 214 215 216 module_init(appldata_os_init); 217 module_exit(appldata_os_exit); 218 219 MODULE_LICENSE("GPL"); 220 MODULE_AUTHOR("Gerald Schaefer"); 221 MODULE_DESCRIPTION("Linux-VM Monitor Stream, OS statistics"); 222