arch/s390/appldata/appldata_os.c

   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         .ctl_nr    = CTL_APPLDATA_OS,
  86         .name      = "os",
  87         .record_nr = APPLDATA_RECORD_OS_ID,
  88         .owner     = THIS_MODULE,
  89         .mod_lvl   = {0xF0, 0xF1},              /* EBCDIC "01" */
  90 };
  91
  92
  93 static inline void appldata_print_debug(struct appldata_os_data *os_data)
  94 {
  95         int a0, a1, a2, i;
  96
  97         P_DEBUG("--- OS - RECORD ---\n");
  98         P_DEBUG("nr_threads   = %u\n", os_data->nr_threads);
  99         P_DEBUG("nr_running   = %u\n", os_data->nr_running);
 100         P_DEBUG("nr_iowait    = %u\n", os_data->nr_iowait);
 101         P_DEBUG("avenrun(int) = %8x / %8x / %8x\n", os_data->avenrun[0],
 102                 os_data->avenrun[1], os_data->avenrun[2]);
 103         a0 = os_data->avenrun[0];
 104         a1 = os_data->avenrun[1];
 105         a2 = os_data->avenrun[2];
 106         P_DEBUG("avenrun(float) = %d.%02d / %d.%02d / %d.%02d\n",
 107                 LOAD_INT(a0), LOAD_FRAC(a0), LOAD_INT(a1), LOAD_FRAC(a1),
 108                 LOAD_INT(a2), LOAD_FRAC(a2));
 109
 110         P_DEBUG("nr_cpus = %u\n", os_data->nr_cpus);
 111         for (i = 0; i < os_data->nr_cpus; i++) {
 112                 P_DEBUG("cpu%u : user = %u, nice = %u, system = %u, "
 113                         "idle = %u, irq = %u, softirq = %u, iowait = %u, "
 114                         "steal = %u\n",
 115                                 os_data->os_cpu[i].cpu_id,
 116                                 os_data->os_cpu[i].per_cpu_user,
 117                                 os_data->os_cpu[i].per_cpu_nice,
 118                                 os_data->os_cpu[i].per_cpu_system,
 119                                 os_data->os_cpu[i].per_cpu_idle,
 120                                 os_data->os_cpu[i].per_cpu_irq,
 121                                 os_data->os_cpu[i].per_cpu_softirq,
 122                                 os_data->os_cpu[i].per_cpu_iowait,
 123                                 os_data->os_cpu[i].per_cpu_steal);
 124         }
 125
 126         P_DEBUG("sync_count_1 = %u\n", os_data->sync_count_1);
 127         P_DEBUG("sync_count_2 = %u\n", os_data->sync_count_2);
 128         P_DEBUG("timestamp    = %lX\n", os_data->timestamp);
 129 }
 130
 131 /*
 132  * appldata_get_os_data()
 133  *
 134  * gather OS data
 135  */
 136 static void appldata_get_os_data(void *data)
 137 {
 138         int i, j, rc;
 139         struct appldata_os_data *os_data;
 140         unsigned int new_size;
 141
 142         os_data = data;
 143         os_data->sync_count_1++;
 144
 145         os_data->nr_threads = nr_threads;
 146         os_data->nr_running = nr_running();
 147         os_data->nr_iowait  = nr_iowait();
 148         os_data->avenrun[0] = avenrun[0] + (FIXED_1/200);
 149         os_data->avenrun[1] = avenrun[1] + (FIXED_1/200);
 150         os_data->avenrun[2] = avenrun[2] + (FIXED_1/200);
 151
 152         j = 0;
 153         for_each_online_cpu(i) {
 154                 os_data->os_cpu[j].per_cpu_user =
 155                         cputime_to_jiffies(kstat_cpu(i).cpustat.user);
 156                 os_data->os_cpu[j].per_cpu_nice =
 157                         cputime_to_jiffies(kstat_cpu(i).cpustat.nice);
 158                 os_data->os_cpu[j].per_cpu_system =
 159                         cputime_to_jiffies(kstat_cpu(i).cpustat.system);
 160                 os_data->os_cpu[j].per_cpu_idle =
 161                         cputime_to_jiffies(kstat_cpu(i).cpustat.idle);
 162                 os_data->os_cpu[j].per_cpu_irq =
 163                         cputime_to_jiffies(kstat_cpu(i).cpustat.irq);
 164                 os_data->os_cpu[j].per_cpu_softirq =
 165                         cputime_to_jiffies(kstat_cpu(i).cpustat.softirq);
 166                 os_data->os_cpu[j].per_cpu_iowait =
 167                         cputime_to_jiffies(kstat_cpu(i).cpustat.iowait);
 168                 os_data->os_cpu[j].per_cpu_steal =
 169                         cputime_to_jiffies(kstat_cpu(i).cpustat.steal);
 170                 os_data->os_cpu[j].cpu_id = i;
 171                 j++;
 172         }
 173
 174         os_data->nr_cpus = j;
 175
 176         new_size = sizeof(struct appldata_os_data) +
 177                    (os_data->nr_cpus * sizeof(struct appldata_os_per_cpu));
 178         if (ops.size != new_size) {
 179                 if (ops.active) {
 180                         rc = appldata_diag(APPLDATA_RECORD_OS_ID,
 181                                            APPLDATA_START_INTERVAL_REC,
 182                                            (unsigned long) ops.data, new_size,
 183                                            ops.mod_lvl);
 184                         if (rc != 0) {
 185                                 P_ERROR("os: START NEW DIAG 0xDC failed, "
 186                                         "return code: %d, new size = %i\n", rc,
 187                                         new_size);
 188                                 P_INFO("os: stopping old record now\n");
 189                         } else
 190                                 P_INFO("os: new record size = %i\n", new_size);
 191
 192                         rc = appldata_diag(APPLDATA_RECORD_OS_ID,
 193                                            APPLDATA_STOP_REC,
 194                                            (unsigned long) ops.data, ops.size,
 195                                            ops.mod_lvl);
 196                         if (rc != 0)
 197                                 P_ERROR("os: STOP OLD DIAG 0xDC failed, "
 198                                         "return code: %d, old size = %i\n", rc,
 199                                         ops.size);
 200                         else
 201                                 P_INFO("os: old record size = %i stopped\n",
 202                                         ops.size);
 203                 }
 204                 ops.size = new_size;
 205         }
 206         os_data->timestamp = get_clock();
 207         os_data->sync_count_2++;
 208 #ifdef APPLDATA_DEBUG
 209         appldata_print_debug(os_data);
 210 #endif
 211 }
 212
 213
 214 /*
 215  * appldata_os_init()
 216  *
 217  * init data, register ops
 218  */
 219 static int __init appldata_os_init(void)
 220 {
 221         int rc, max_size;
 222
 223         max_size = sizeof(struct appldata_os_data) +
 224                    (NR_CPUS * sizeof(struct appldata_os_per_cpu));
 225         if (max_size > APPLDATA_MAX_REC_SIZE) {
 226                 P_ERROR("Max. size of OS record = %i, bigger than maximum "
 227                         "record size (%i)\n", max_size, APPLDATA_MAX_REC_SIZE);
 228                 rc = -ENOMEM;
 229                 goto out;
 230         }
 231         P_DEBUG("max. sizeof(os) = %i, sizeof(os_cpu) = %lu\n", max_size,
 232                 sizeof(struct appldata_os_per_cpu));
 233
 234         appldata_os_data = kzalloc(max_size, GFP_DMA);
 235         if (appldata_os_data == NULL) {
 236                 P_ERROR("No memory for %s!\n", ops.name);
 237                 rc = -ENOMEM;
 238                 goto out;
 239         }
 240
 241         appldata_os_data->per_cpu_size = sizeof(struct appldata_os_per_cpu);
 242         appldata_os_data->cpu_offset   = offsetof(struct appldata_os_data,
 243                                                         os_cpu);
 244         P_DEBUG("cpu offset = %u\n", appldata_os_data->cpu_offset);
 245
 246         ops.data = appldata_os_data;
 247         ops.callback  = &appldata_get_os_data;
 248         rc = appldata_register_ops(&ops);
 249         if (rc != 0) {
 250                 P_ERROR("Error registering ops, rc = %i\n", rc);
 251                 kfree(appldata_os_data);
 252         } else {
 253                 P_DEBUG("%s-ops registered!\n", ops.name);
 254         }
 255 out:
 256         return rc;
 257 }
 258
 259 /*
 260  * appldata_os_exit()
 261  *
 262  * unregister ops
 263  */
 264 static void __exit appldata_os_exit(void)
 265 {
 266         appldata_unregister_ops(&ops);
 267         kfree(appldata_os_data);
 268         P_DEBUG("%s-ops unregistered!\n", ops.name);
 269 }
 270
 271
 272 module_init(appldata_os_init);
 273 module_exit(appldata_os_exit);
 274
 275 MODULE_LICENSE("GPL");
 276 MODULE_AUTHOR("Gerald Schaefer");
 277 MODULE_DESCRIPTION("Linux-VM Monitor Stream, OS statistics");