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(__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 typedef struct lockstat_probe {
  67         const char      *lsp_func;
  68         const char      *lsp_name;
  69         int             lsp_probe;
  70         dtrace_id_t     lsp_id;
  71 } lockstat_probe_t;
  72
  73 lockstat_probe_t lockstat_probes[] =
  74 {
  75 #if defined(__x86_64__)
  76         /* Only provide implemented probes for each architecture  */
  77         { LS_LCK_MTX_LOCK,      LSA_ACQUIRE,    LS_LCK_MTX_LOCK_ACQUIRE, DTRACE_IDNONE },
  78         { LS_LCK_MTX_LOCK,      LSA_SPIN,       LS_LCK_MTX_LOCK_SPIN, DTRACE_IDNONE },
  79         { LS_LCK_MTX_LOCK,      LSA_BLOCK,      LS_LCK_MTX_LOCK_BLOCK, 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_LOCK,  LSA_BLOCK,      LS_LCK_MTX_EXT_LOCK_BLOCK, DTRACE_IDNONE },
  86 //      { LS_LCK_MTX_EXT_TRY_LOCK, LSA_ACQUIRE, LS_LCK_MTX_TRY_EXT_LOCK_ACQUIRE, DTRACE_IDNONE },
  87         { LS_LCK_MTX_EXT_UNLOCK,   LSA_RELEASE, LS_LCK_MTX_EXT_UNLOCK_RELEASE, DTRACE_IDNONE },
  88         { LS_LCK_MTX_LOCK_SPIN_LOCK,    LSA_ACQUIRE,    LS_LCK_MTX_LOCK_SPIN_ACQUIRE, DTRACE_IDNONE },
  89         { LS_LCK_RW_LOCK_SHARED,        LSR_ACQUIRE,    LS_LCK_RW_LOCK_SHARED_ACQUIRE, DTRACE_IDNONE },
  90         { LS_LCK_RW_LOCK_SHARED,        LSR_BLOCK,      LS_LCK_RW_LOCK_SHARED_BLOCK, DTRACE_IDNONE },
  91         { LS_LCK_RW_LOCK_SHARED,        LSR_SPIN,       LS_LCK_RW_LOCK_SHARED_SPIN, DTRACE_IDNONE },
  92         { LS_LCK_RW_LOCK_EXCL,          LSR_ACQUIRE,    LS_LCK_RW_LOCK_EXCL_ACQUIRE, DTRACE_IDNONE },
  93         { LS_LCK_RW_LOCK_EXCL,          LSR_BLOCK,      LS_LCK_RW_LOCK_EXCL_BLOCK, DTRACE_IDNONE },
  94         { LS_LCK_RW_LOCK_EXCL,          LSR_SPIN,       LS_LCK_RW_LOCK_EXCL_SPIN, DTRACE_IDNONE },
  95         { LS_LCK_RW_DONE,               LSR_RELEASE,    LS_LCK_RW_DONE_RELEASE, DTRACE_IDNONE },
  96         { LS_LCK_RW_TRY_LOCK_SHARED,    LSR_ACQUIRE,    LS_LCK_RW_TRY_LOCK_SHARED_ACQUIRE, DTRACE_IDNONE },
  97         { LS_LCK_RW_TRY_LOCK_EXCL,      LSR_ACQUIRE,    LS_LCK_RW_TRY_LOCK_EXCL_ACQUIRE, DTRACE_IDNONE },
  98         { LS_LCK_RW_LOCK_SHARED_TO_EXCL, LSR_UPGRADE,   LS_LCK_RW_LOCK_SHARED_TO_EXCL_UPGRADE, DTRACE_IDNONE },
  99         { LS_LCK_RW_LOCK_SHARED_TO_EXCL,        LSR_SPIN,       LS_LCK_RW_LOCK_SHARED_TO_EXCL_SPIN, DTRACE_IDNONE },
 100         { LS_LCK_RW_LOCK_SHARED_TO_EXCL,        LSR_BLOCK,      LS_LCK_RW_LOCK_SHARED_TO_EXCL_BLOCK, DTRACE_IDNONE },
 101         { LS_LCK_RW_LOCK_EXCL_TO_SHARED,        LSR_DOWNGRADE,  LS_LCK_RW_LOCK_EXCL_TO_SHARED_DOWNGRADE, DTRACE_IDNONE },
 102 #endif
 103 #ifdef  LATER
 104         /* Interlock and spinlock measurements would be nice, but later */
 105         { LS_LCK_SPIN_LOCK,     LSS_ACQUIRE,    LS_LCK_SPIN_LOCK_ACQUIRE, DTRACE_IDNONE },
 106         { LS_LCK_SPIN_LOCK,     LSS_SPIN,       LS_LCK_SPIN_LOCK_SPIN, DTRACE_IDNONE },
 107         { LS_LCK_SPIN_UNLOCK,   LSS_RELEASE,    LS_LCK_SPIN_UNLOCK_RELEASE, DTRACE_IDNONE },
 108
 109         { LS_LCK_RW_LOCK_EXCL_TO_SHARED,        LSA_ILK_SPIN,   LS_LCK_RW_LOCK_EXCL_TO_SHARED_ILK_SPIN, DTRACE_IDNONE },
 110         { LS_LCK_MTX_LOCK,      LSA_ILK_SPIN,   LS_LCK_MTX_LOCK_ILK_SPIN, DTRACE_IDNONE },
 111         { LS_LCK_MTX_EXT_LOCK,  LSA_ILK_SPIN,   LS_LCK_MTX_EXT_LOCK_ILK_SPIN, DTRACE_IDNONE },
 112         { LS_LCK_RW_TRY_LOCK_EXCL,      LSA_ILK_SPIN,   LS_LCK_RW_TRY_LOCK_EXCL_ILK_SPIN, DTRACE_IDNONE },
 113         { LS_LCK_RW_TRY_LOCK_SHARED,    LSA_SPIN,       LS_LCK_RW_TRY_LOCK_SHARED_SPIN, DTRACE_IDNONE },
 114 #endif
 115
 116         { NULL, NULL, 0, 0 }
 117 };
 118
 119 dtrace_id_t lockstat_probemap[LS_NPROBES];
 120
 121 #if CONFIG_DTRACE
 122 #if defined(__x86_64__)
 123 extern void lck_mtx_lock_lockstat_patch_point(void);
 124 extern void lck_mtx_try_lock_lockstat_patch_point(void);
 125 extern void lck_mtx_try_lock_spin_lockstat_patch_point(void);
 126 extern void lck_mtx_unlock_lockstat_patch_point(void);
 127 extern void lck_mtx_lock_ext_lockstat_patch_point(void);
 128 extern void lck_mtx_ext_unlock_lockstat_patch_point(void);
 129 extern void lck_rw_lock_shared_lockstat_patch_point(void);
 130 extern void lck_rw_lock_exclusive_lockstat_patch_point(void);
 131 extern void lck_rw_lock_shared_to_exclusive_lockstat_patch_point(void);
 132 extern void lck_rw_try_lock_shared_lockstat_patch_point(void);
 133 extern void lck_rw_try_lock_exclusive_lockstat_patch_point(void);
 134 extern void lck_mtx_lock_spin_lockstat_patch_point(void);
 135 #endif
 136
 137 #endif /* CONFIG_DTRACE */
 138
 139 typedef struct lockstat_assembly_probe {
 140         int lsap_probe;
 141         vm_offset_t * lsap_patch_point;
 142 } lockstat_assembly_probe_t;
 143
 144
 145         lockstat_assembly_probe_t assembly_probes[] =
 146         {
 147 #if CONFIG_DTRACE
 148 #if defined(__x86_64__)
 149                 /*
 150                  * On x86 these points are better done via hot patches, which ensure
 151                  * there is zero overhead when not in use.  On x86 these patch points
 152                  * are swapped between the return instruction and a no-op, with the
 153                  * Dtrace call following the return.
 154                  */
 155                 { LS_LCK_MTX_LOCK_ACQUIRE,              (vm_offset_t *) lck_mtx_lock_lockstat_patch_point },
 156                 { LS_LCK_MTX_TRY_LOCK_ACQUIRE,          (vm_offset_t *) lck_mtx_try_lock_lockstat_patch_point },
 157                 { LS_LCK_MTX_TRY_SPIN_LOCK_ACQUIRE,     (vm_offset_t *) lck_mtx_try_lock_spin_lockstat_patch_point },
 158                 { LS_LCK_MTX_UNLOCK_RELEASE,            (vm_offset_t *) lck_mtx_unlock_lockstat_patch_point },
 159                 { LS_LCK_MTX_EXT_LOCK_ACQUIRE,          (vm_offset_t *) lck_mtx_lock_ext_lockstat_patch_point },
 160                 { LS_LCK_MTX_EXT_UNLOCK_RELEASE,        (vm_offset_t *) lck_mtx_ext_unlock_lockstat_patch_point },
 161                 { LS_LCK_RW_LOCK_SHARED_ACQUIRE,        (vm_offset_t *) lck_rw_lock_shared_lockstat_patch_point },
 162                 { LS_LCK_RW_LOCK_EXCL_ACQUIRE,          (vm_offset_t *) lck_rw_lock_exclusive_lockstat_patch_point },
 163                 { LS_LCK_RW_LOCK_SHARED_TO_EXCL_UPGRADE,(vm_offset_t *) lck_rw_lock_shared_to_exclusive_lockstat_patch_point },
 164                 { LS_LCK_RW_TRY_LOCK_SHARED_ACQUIRE,    (vm_offset_t *) lck_rw_try_lock_shared_lockstat_patch_point },
 165                 { LS_LCK_RW_TRY_LOCK_EXCL_ACQUIRE,      (vm_offset_t *) lck_rw_try_lock_exclusive_lockstat_patch_point },
 166                 { LS_LCK_MTX_LOCK_SPIN_ACQUIRE,         (vm_offset_t *) lck_mtx_lock_spin_lockstat_patch_point },
 167 #endif
 168                 /* No assembly patch points for ARM */
 169 #endif /* CONFIG_DTRACE */
 170                 { LS_LCK_INVALID, NULL }
 171 };
 172
 173
 174 /*
 175  * APPLE NOTE:
 176  * Hot patch is used to manipulate probe points by swapping between
 177  * no-op and return instructions.
 178  * The active flag indicates whether the probe point will turn on or off.
 179  *      on == plant a NOP and thus fall through to the probe call
 180  *     off == plant a RET and thus avoid the probe call completely
 181  * The ls_probe identifies which probe we will patch.
 182  */
 183 static
 184 void lockstat_hot_patch(boolean_t active, int ls_probe)
 185 {
 186 #pragma unused(active)
 187         int i;
 188
 189         /*
 190          * Loop through entire table, in case there are
 191          * multiple patch points per probe.
 192          */
 193         for (i = 0; assembly_probes[i].lsap_patch_point; i++) {
 194                 if (ls_probe == assembly_probes[i].lsap_probe)
 195 #if defined(__x86_64__)
 196                 {
 197                         uint8_t instr;
 198                         instr = (active ? NOP : RET );
 199                         (void) ml_nofault_copy( (vm_offset_t)&instr, *(assembly_probes[i].lsap_patch_point),
 200                                                                 sizeof(instr));
 201                 }
 202 #endif
 203         } /* for */
 204 }
 205
 206 void (*lockstat_probe)(dtrace_id_t, uint64_t, uint64_t,
 207                                     uint64_t, uint64_t, uint64_t);
 208
 209
 210 /*
 211  * APPLE NOTE:
 212  * This wrapper is used only by assembler hot patched probes.
 213  */
 214 void
 215 lockstat_probe_wrapper(int probe, uintptr_t lp, int rwflag)
 216 {
 217         dtrace_id_t id;
 218         id = lockstat_probemap[probe];
 219         if (id != 0)
 220         {
 221                 (*lockstat_probe)(id, (uintptr_t)lp, (uint64_t)rwflag, 0,0,0);
 222         }
 223 }
 224
 225 static dev_info_t       *lockstat_devi; /* saved in xxattach() for xxinfo() */
 226 static dtrace_provider_id_t lockstat_id;
 227
 228 /*ARGSUSED*/
 229 static int
 230 lockstat_enable(void *arg, dtrace_id_t id, void *parg)
 231 {
 232 #pragma unused(arg) /* __APPLE__ */
 233
 234         lockstat_probe_t *probe = parg;
 235
 236         ASSERT(!lockstat_probemap[probe->lsp_probe]);
 237
 238         lockstat_probemap[probe->lsp_probe] = id;
 239         membar_producer();
 240
 241         lockstat_hot_patch(TRUE, probe->lsp_probe);
 242         membar_producer();
 243         return(0);
 244
 245 }
 246
 247 /*ARGSUSED*/
 248 static void
 249 lockstat_disable(void *arg, dtrace_id_t id, void *parg)
 250 {
 251 #pragma unused(arg, id) /* __APPLE__ */
 252
 253         lockstat_probe_t *probe = parg;
 254         int i;
 255
 256         ASSERT(lockstat_probemap[probe->lsp_probe]);
 257
 258         lockstat_probemap[probe->lsp_probe] = 0;
 259         lockstat_hot_patch(FALSE, probe->lsp_probe);
 260         membar_producer();
 261
 262         /*
 263          * See if we have any probes left enabled.
 264          */
 265         for (i = 0; i < LS_NPROBES; i++) {
 266                 if (lockstat_probemap[i]) {
 267                         /*
 268                          * This probe is still enabled.  We don't need to deal
 269                          * with waiting for all threads to be out of the
 270                          * lockstat critical sections; just return.
 271                          */
 272                         return;
 273                 }
 274         }
 275
 276 }
 277
 278 /*ARGSUSED*/
 279 static void
 280 lockstat_provide(void *arg, const dtrace_probedesc_t *desc)
 281 {
 282 #pragma unused(arg, desc) /* __APPLE__ */
 283
 284         int i = 0;
 285
 286         for (i = 0; lockstat_probes[i].lsp_func != NULL; i++) {
 287                 lockstat_probe_t *probe = &lockstat_probes[i];
 288
 289                 if (dtrace_probe_lookup(lockstat_id, "mach_kernel",
 290                     probe->lsp_func, probe->lsp_name) != 0)
 291                         continue;
 292
 293                 ASSERT(!probe->lsp_id);
 294                 probe->lsp_id = dtrace_probe_create(lockstat_id,
 295                     "mach_kernel", probe->lsp_func, probe->lsp_name,
 296                     LOCKSTAT_AFRAMES, probe);
 297         }
 298 }
 299
 300
 301 /*ARGSUSED*/
 302 static void
 303 lockstat_destroy(void *arg, dtrace_id_t id, void *parg)
 304 {
 305 #pragma unused(arg, id) /* __APPLE__ */
 306
 307         lockstat_probe_t *probe = parg;
 308
 309         ASSERT(!lockstat_probemap[probe->lsp_probe]);
 310         probe->lsp_id = 0;
 311 }
 312
 313 static dtrace_pattr_t lockstat_attr = {
 314 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 315 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
 316 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
 317 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 318 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 319 };
 320
 321 static dtrace_pops_t lockstat_pops = {
 322         lockstat_provide,
 323         NULL,
 324         lockstat_enable,
 325         lockstat_disable,
 326         NULL,
 327         NULL,
 328         NULL,
 329         NULL,
 330         NULL,
 331         lockstat_destroy
 332 };
 333
 334 static int
 335 lockstat_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
 336 {
 337         switch (cmd) {
 338         case DDI_ATTACH:
 339                 break;
 340         case DDI_RESUME:
 341                 return (DDI_SUCCESS);
 342         default:
 343                 return (DDI_FAILURE);
 344         }
 345
 346         if (ddi_create_minor_node(devi, "lockstat", S_IFCHR, 0,
 347             DDI_PSEUDO, 0) == DDI_FAILURE ||
 348             dtrace_register("lockstat", &lockstat_attr, DTRACE_PRIV_KERNEL,
 349             NULL, &lockstat_pops, NULL, &lockstat_id) != 0) {
 350                 ddi_remove_minor_node(devi, NULL);
 351                 return (DDI_FAILURE);
 352         }
 353
 354         lockstat_probe = dtrace_probe;
 355         membar_producer();
 356
 357         ddi_report_dev(devi);
 358         lockstat_devi = devi;
 359         return (DDI_SUCCESS);
 360 }
 361
 362 d_open_t _lockstat_open;
 363
 364 int _lockstat_open(dev_t dev, int flags, int devtype, struct proc *p)
 365 {
 366 #pragma unused(dev,flags,devtype,p)
 367         return 0;
 368 }
 369
 370 #define LOCKSTAT_MAJOR  -24 /* let the kernel pick the device number */
 371
 372 /*
 373  * A struct describing which functions will get invoked for certain
 374  * actions.
 375  */
 376 static struct cdevsw lockstat_cdevsw =
 377 {
 378         _lockstat_open,         /* open */
 379         eno_opcl,                       /* close */
 380         eno_rdwrt,                      /* read */
 381         eno_rdwrt,                      /* write */
 382         eno_ioctl,                      /* ioctl */
 383         (stop_fcn_t *)nulldev, /* stop */
 384         (reset_fcn_t *)nulldev, /* reset */
 385         NULL,                           /* tty's */
 386         eno_select,                     /* select */
 387         eno_mmap,                       /* mmap */
 388         eno_strat,                      /* strategy */
 389         eno_getc,                       /* getc */
 390         eno_putc,                       /* putc */
 391         0                                       /* type */
 392 };
 393
 394 static int gLockstatInited = 0;
 395
 396 void lockstat_init( void );
 397
 398 void lockstat_init( void )
 399 {
 400         if (0 == gLockstatInited)
 401         {
 402                 int majdevno = cdevsw_add(LOCKSTAT_MAJOR, &lockstat_cdevsw);
 403
 404                 if (majdevno < 0) {
 405                         printf("lockstat_init: failed to allocate a major number!\n");
 406                         gLockstatInited = 0;
 407                         return;
 408                 }
 409
 410                 lockstat_attach( (dev_info_t    *)(uintptr_t)majdevno, DDI_ATTACH );
 411                 gLockstatInited = 1;
 412         } else
 413                 panic("lockstat_init: called twice!\n");
 414 }
 415 #undef LOCKSTAT_MAJOR