bsd/dev/dtrace/lockstat.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25
  26 /* #pragma ident        "@(#)lockstat.c 1.12    08/01/16 SMI" */
  27
  28
  29 #ifdef KERNEL
  30 #ifndef _KERNEL
  31 #define _KERNEL /* Solaris vs. Darwin */
  32 #endif
  33 #endif
  34
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/errno.h>
  38 #include <sys/stat.h>
  39 #include <sys/ioctl.h>
  40 #include <sys/conf.h>
  41 #include <sys/fcntl.h>
  42 #include <miscfs/devfs/devfs.h>
  43
  44 #include <sys/dtrace.h>
  45 #include <sys/dtrace_impl.h>
  46
  47 #include <sys/dtrace_glue.h>
  48
  49 #include <sys/lockstat.h>
  50
  51 #include <kern/processor.h>
  52
  53 #define membar_producer dtrace_membar_producer
  54
  55 /*
  56  * Hot patch values, x86
  57  */
  58 #if defined(__i386__) || defined(__x86_64__)
  59 #define NOP     0x90
  60 #define RET     0xc3
  61 #define LOCKSTAT_AFRAMES 1
  62 #else
  63 #error "not ported to this architecture"
  64 #endif
  65
  66
  67 typedef struct lockstat_probe {
  68         const char      *lsp_func;
  69         const char      *lsp_name;
  70         int             lsp_probe;
  71         dtrace_id_t     lsp_id;
  72 } lockstat_probe_t;
  73
  74 lockstat_probe_t lockstat_probes[] =
  75 {
  76 #if defined(__i386__) || defined(__x86_64__)
  77         /* Not implemented yet on PPC... */
  78         { LS_LCK_MTX_LOCK,      LSA_ACQUIRE,    LS_LCK_MTX_LOCK_ACQUIRE, DTRACE_IDNONE },
  79         { LS_LCK_MTX_LOCK,      LSA_SPIN,       LS_LCK_MTX_LOCK_SPIN, DTRACE_IDNONE },
  80         { LS_LCK_MTX_TRY_LOCK,  LSA_ACQUIRE,    LS_LCK_MTX_TRY_LOCK_ACQUIRE, DTRACE_IDNONE },
  81         { LS_LCK_MTX_TRY_SPIN_LOCK, LSA_ACQUIRE, LS_LCK_MTX_TRY_SPIN_LOCK_ACQUIRE, DTRACE_IDNONE },
  82         { LS_LCK_MTX_UNLOCK,    LSA_RELEASE,    LS_LCK_MTX_UNLOCK_RELEASE, DTRACE_IDNONE },
  83         { LS_LCK_MTX_EXT_LOCK,  LSA_ACQUIRE,    LS_LCK_MTX_EXT_LOCK_ACQUIRE, DTRACE_IDNONE },
  84         { LS_LCK_MTX_EXT_LOCK,  LSA_SPIN,       LS_LCK_MTX_EXT_LOCK_SPIN, DTRACE_IDNONE },
  85         { LS_LCK_MTX_EXT_TRY_LOCK, LSA_ACQUIRE, LS_LCK_MTX_TRY_EXT_LOCK_ACQUIRE, DTRACE_IDNONE },
  86         { LS_LCK_MTX_UNLOCK,    LSA_RELEASE,    LS_LCK_MTX_EXT_UNLOCK_RELEASE, DTRACE_IDNONE },
  87         { LS_LCK_MTX_LOCK_SPIN_LOCK,    LSA_ACQUIRE,    LS_LCK_MTX_LOCK_SPIN_ACQUIRE, DTRACE_IDNONE },
  88 #endif
  89         { LS_LCK_MTX_LOCK,      LSA_BLOCK,      LS_LCK_MTX_LOCK_BLOCK, DTRACE_IDNONE },
  90         { LS_LCK_MTX_EXT_LOCK,  LSA_BLOCK,      LS_LCK_MTX_EXT_LOCK_BLOCK, DTRACE_IDNONE },
  91
  92         { LS_LCK_RW_LOCK_SHARED,        LSR_ACQUIRE,    LS_LCK_RW_LOCK_SHARED_ACQUIRE, DTRACE_IDNONE },
  93         { LS_LCK_RW_LOCK_SHARED,        LSR_BLOCK,      LS_LCK_RW_LOCK_SHARED_BLOCK, DTRACE_IDNONE },
  94         { LS_LCK_RW_LOCK_SHARED,        LSR_SPIN,       LS_LCK_RW_LOCK_SHARED_SPIN, DTRACE_IDNONE },
  95         { LS_LCK_RW_LOCK_EXCL,          LSR_ACQUIRE,    LS_LCK_RW_LOCK_EXCL_ACQUIRE, DTRACE_IDNONE },
  96         { LS_LCK_RW_LOCK_EXCL,          LSR_BLOCK,      LS_LCK_RW_LOCK_EXCL_BLOCK, DTRACE_IDNONE },
  97         { LS_LCK_RW_LOCK_EXCL,          LSR_SPIN,       LS_LCK_RW_LOCK_EXCL_SPIN, DTRACE_IDNONE },
  98         { LS_LCK_RW_DONE,               LSR_RELEASE,    LS_LCK_RW_DONE_RELEASE, DTRACE_IDNONE },
  99         { LS_LCK_RW_TRY_LOCK_SHARED,    LSR_ACQUIRE,    LS_LCK_RW_TRY_LOCK_SHARED_ACQUIRE, DTRACE_IDNONE },
 100         { LS_LCK_RW_TRY_LOCK_EXCL,      LSR_ACQUIRE,    LS_LCK_RW_TRY_LOCK_EXCL_ACQUIRE, DTRACE_IDNONE },
 101         { LS_LCK_RW_LOCK_SHARED_TO_EXCL, LSR_UPGRADE,   LS_LCK_RW_LOCK_SHARED_TO_EXCL_UPGRADE, DTRACE_IDNONE },
 102         { LS_LCK_RW_LOCK_SHARED_TO_EXCL,        LSR_BLOCK,      LS_LCK_RW_LOCK_SHARED_TO_EXCL_BLOCK, DTRACE_IDNONE },
 103         { LS_LCK_RW_LOCK_SHARED_TO_EXCL,        LSR_SPIN,       LS_LCK_RW_LOCK_SHARED_TO_EXCL_SPIN, DTRACE_IDNONE },
 104         { LS_LCK_RW_LOCK_EXCL_TO_SHARED,        LSR_DOWNGRADE,  LS_LCK_RW_LOCK_EXCL_TO_SHARED_DOWNGRADE, DTRACE_IDNONE },
 105
 106
 107 #ifdef  LATER
 108         /* Interlock and spinlock measurements would be nice, but later */
 109         { LS_LCK_SPIN_LOCK,     LSS_ACQUIRE,    LS_LCK_SPIN_LOCK_ACQUIRE, DTRACE_IDNONE },
 110         { LS_LCK_SPIN_LOCK,     LSS_SPIN,       LS_LCK_SPIN_LOCK_SPIN, DTRACE_IDNONE },
 111         { LS_LCK_SPIN_UNLOCK,   LSS_RELEASE,    LS_LCK_SPIN_UNLOCK_RELEASE, DTRACE_IDNONE },
 112
 113         { LS_LCK_RW_LOCK_EXCL_TO_SHARED,        LSA_ILK_SPIN,   LS_LCK_RW_LOCK_EXCL_TO_SHARED_ILK_SPIN, DTRACE_IDNONE },
 114         { LS_LCK_MTX_LOCK,      LSA_ILK_SPIN,   LS_LCK_MTX_LOCK_ILK_SPIN, DTRACE_IDNONE },
 115         { LS_LCK_MTX_EXT_LOCK,  LSA_ILK_SPIN,   LS_LCK_MTX_EXT_LOCK_ILK_SPIN, DTRACE_IDNONE },
 116         { LS_LCK_RW_TRY_LOCK_EXCL,      LSA_ILK_SPIN,   LS_LCK_RW_TRY_LOCK_EXCL_ILK_SPIN, DTRACE_IDNONE },
 117         { LS_LCK_RW_TRY_LOCK_SHARED,    LSA_SPIN,       LS_LCK_RW_TRY_LOCK_SHARED_SPIN, DTRACE_IDNONE },
 118 #endif
 119
 120         { NULL, NULL, 0, 0 }
 121 };
 122
 123 dtrace_id_t lockstat_probemap[LS_NPROBES];
 124
 125 #if CONFIG_DTRACE
 126 extern void lck_mtx_lock_lockstat_patch_point(void);
 127 extern void lck_mtx_try_lock_lockstat_patch_point(void);
 128 extern void lck_mtx_try_lock_spin_lockstat_patch_point(void);
 129 extern void lck_mtx_unlock_lockstat_patch_point(void);
 130 extern void lck_mtx_lock_ext_lockstat_patch_point(void);
 131 extern void lck_mtx_ext_unlock_lockstat_patch_point(void);
 132
 133 extern void lck_rw_lock_shared_lockstat_patch_point(void);
 134 extern void lck_rw_lock_exclusive_lockstat_patch_point(void);
 135 extern void lck_rw_lock_shared_to_exclusive_lockstat_patch_point(void);
 136 extern void lck_rw_try_lock_shared_lockstat_patch_point(void);
 137 extern void lck_rw_try_lock_exclusive_lockstat_patch_point(void);
 138 extern void lck_mtx_lock_spin_lockstat_patch_point(void);
 139 #endif /* CONFIG_DTRACE */
 140
 141 vm_offset_t *assembly_probes[] = {
 142 #if CONFIG_DTRACE
 143 #if defined(__i386__) || defined(__x86_64__)
 144         /*
 145          * On x86 these points are better done via hot patches, which ensure
 146          * there is zero overhead when not in use.  On x86 these patch points
 147          * are swapped between the return instruction and a no-op, with the
 148          * Dtrace call following the return.
 149          */
 150         (vm_offset_t *) lck_mtx_lock_lockstat_patch_point,
 151         (vm_offset_t *) lck_mtx_try_lock_lockstat_patch_point,
 152         (vm_offset_t *) lck_mtx_try_lock_spin_lockstat_patch_point,
 153         (vm_offset_t *) lck_mtx_unlock_lockstat_patch_point,
 154         (vm_offset_t *) lck_mtx_lock_ext_lockstat_patch_point,
 155         (vm_offset_t *) lck_mtx_ext_unlock_lockstat_patch_point,
 156         (vm_offset_t *) lck_rw_lock_shared_lockstat_patch_point,
 157         (vm_offset_t *) lck_rw_lock_exclusive_lockstat_patch_point,
 158         (vm_offset_t *) lck_rw_lock_shared_to_exclusive_lockstat_patch_point,
 159         (vm_offset_t *) lck_rw_try_lock_shared_lockstat_patch_point,
 160         (vm_offset_t *) lck_rw_try_lock_exclusive_lockstat_patch_point,
 161         (vm_offset_t *) lck_mtx_lock_spin_lockstat_patch_point,
 162 #else
 163         (vm_offset_t *) lck_mtx_unlock_lockstat_patch_point,
 164 #endif
 165 #endif /* CONFIG_DTRACE */
 166         NULL
 167 };
 168 /*
 169  * Hot patch switches back and forth the probe points between NOP and RET.
 170  * The argument indicates whether the probe point is on or off.
 171  */
 172 #if defined(__APPLE__)
 173 static
 174 #endif /* __APPLE__ */
 175 void lockstat_hot_patch(boolean_t active)
 176 {
 177 #pragma unused(active)
 178         int i;
 179
 180
 181         for (i = 0; assembly_probes[i]; i++) {
 182 #if defined(__i386__) || defined(__x86_64__)
 183                 uint8_t instr;
 184                 instr = (active ? NOP : RET );
 185                 (void) ml_nofault_copy( (vm_offset_t)&instr, *(assembly_probes[i]),
 186                                                                 sizeof(instr));
 187 #endif
 188         }
 189 }
 190
 191
 192
 193 void (*lockstat_probe)(dtrace_id_t, uint64_t, uint64_t,
 194                                     uint64_t, uint64_t, uint64_t);
 195
 196 static dev_info_t       *lockstat_devi; /* saved in xxattach() for xxinfo() */
 197 static dtrace_provider_id_t lockstat_id;
 198
 199 /*ARGSUSED*/
 200 static int
 201 lockstat_enable(void *arg, dtrace_id_t id, void *parg)
 202 {
 203 #pragma unused(arg) /* __APPLE__ */
 204
 205         lockstat_probe_t *probe = parg;
 206
 207         ASSERT(!lockstat_probemap[probe->lsp_probe]);
 208
 209         lockstat_probemap[probe->lsp_probe] = id;
 210         membar_producer();
 211
 212         lockstat_hot_patch(TRUE);
 213         membar_producer();
 214         return(0);
 215
 216 }
 217
 218 /*ARGSUSED*/
 219 static void
 220 lockstat_disable(void *arg, dtrace_id_t id, void *parg)
 221 {
 222 #pragma unused(arg, id) /* __APPLE__ */
 223
 224         lockstat_probe_t *probe = parg;
 225         int i;
 226
 227         ASSERT(lockstat_probemap[probe->lsp_probe]);
 228
 229         lockstat_probemap[probe->lsp_probe] = 0;
 230         lockstat_hot_patch(FALSE);
 231         membar_producer();
 232
 233         /*
 234          * See if we have any probes left enabled.
 235          */
 236         for (i = 0; i < LS_NPROBES; i++) {
 237                 if (lockstat_probemap[i]) {
 238                         /*
 239                          * This probe is still enabled.  We don't need to deal
 240                          * with waiting for all threads to be out of the
 241                          * lockstat critical sections; just return.
 242                          */
 243                         return;
 244                 }
 245         }
 246
 247 }
 248
 249 /*ARGSUSED*/
 250 static void
 251 lockstat_provide(void *arg, const dtrace_probedesc_t *desc)
 252 {
 253 #pragma unused(arg, desc) /* __APPLE__ */
 254
 255         int i = 0;
 256
 257         for (i = 0; lockstat_probes[i].lsp_func != NULL; i++) {
 258                 lockstat_probe_t *probe = &lockstat_probes[i];
 259
 260                 if (dtrace_probe_lookup(lockstat_id, "mach_kernel",
 261                     probe->lsp_func, probe->lsp_name) != 0)
 262                         continue;
 263
 264                 ASSERT(!probe->lsp_id);
 265                 probe->lsp_id = dtrace_probe_create(lockstat_id,
 266                     "mach_kernel", probe->lsp_func, probe->lsp_name,
 267                     LOCKSTAT_AFRAMES, probe);
 268         }
 269 }
 270
 271
 272 /*ARGSUSED*/
 273 static void
 274 lockstat_destroy(void *arg, dtrace_id_t id, void *parg)
 275 {
 276 #pragma unused(arg, id) /* __APPLE__ */
 277
 278         lockstat_probe_t *probe = parg;
 279
 280         ASSERT(!lockstat_probemap[probe->lsp_probe]);
 281         probe->lsp_id = 0;
 282 }
 283
 284 static dtrace_pattr_t lockstat_attr = {
 285 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 286 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
 287 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
 288 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 289 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 290 };
 291
 292 static dtrace_pops_t lockstat_pops = {
 293         lockstat_provide,
 294         NULL,
 295         lockstat_enable,
 296         lockstat_disable,
 297         NULL,
 298         NULL,
 299         NULL,
 300         NULL,
 301         NULL,
 302         lockstat_destroy
 303 };
 304
 305 static int
 306 lockstat_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
 307 {
 308         switch (cmd) {
 309         case DDI_ATTACH:
 310                 break;
 311         case DDI_RESUME:
 312                 return (DDI_SUCCESS);
 313         default:
 314                 return (DDI_FAILURE);
 315         }
 316
 317         if (ddi_create_minor_node(devi, "lockstat", S_IFCHR, 0,
 318             DDI_PSEUDO, 0) == DDI_FAILURE ||
 319             dtrace_register("lockstat", &lockstat_attr, DTRACE_PRIV_KERNEL,
 320             NULL, &lockstat_pops, NULL, &lockstat_id) != 0) {
 321                 ddi_remove_minor_node(devi, NULL);
 322                 return (DDI_FAILURE);
 323         }
 324
 325         lockstat_probe = dtrace_probe;
 326         membar_producer();
 327
 328         ddi_report_dev(devi);
 329         lockstat_devi = devi;
 330         return (DDI_SUCCESS);
 331 }
 332
 333 d_open_t _lockstat_open;
 334
 335 int _lockstat_open(dev_t dev, int flags, int devtype, struct proc *p)
 336 {
 337 #pragma unused(dev,flags,devtype,p)
 338         return 0;
 339 }
 340
 341 #define LOCKSTAT_MAJOR  -24 /* let the kernel pick the device number */
 342
 343 /*
 344  * A struct describing which functions will get invoked for certain
 345  * actions.
 346  */
 347 static struct cdevsw lockstat_cdevsw =
 348 {
 349         _lockstat_open,         /* open */
 350         eno_opcl,                       /* close */
 351         eno_rdwrt,                      /* read */
 352         eno_rdwrt,                      /* write */
 353         eno_ioctl,                      /* ioctl */
 354         (stop_fcn_t *)nulldev, /* stop */
 355         (reset_fcn_t *)nulldev, /* reset */
 356         NULL,                           /* tty's */
 357         eno_select,                     /* select */
 358         eno_mmap,                       /* mmap */
 359         eno_strat,                      /* strategy */
 360         eno_getc,                       /* getc */
 361         eno_putc,                       /* putc */
 362         0                                       /* type */
 363 };
 364
 365 static int gLockstatInited = 0;
 366
 367 void lockstat_init( void );
 368
 369 void lockstat_init( void )
 370 {
 371         if (0 == gLockstatInited)
 372         {
 373                 int majdevno = cdevsw_add(LOCKSTAT_MAJOR, &lockstat_cdevsw);
 374
 375                 if (majdevno < 0) {
 376                         printf("lockstat_init: failed to allocate a major number!\n");
 377                         gLockstatInited = 0;
 378                         return;
 379                 }
 380
 381                 lockstat_attach( (dev_info_t    *)(uintptr_t)majdevno, DDI_ATTACH );
 382                 gLockstatInited = 1;
 383         } else
 384                 panic("lockstat_init: called twice!\n");
 385 }
 386 #undef LOCKSTAT_MAJOR