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 2008 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 #elif   __ppc__
  63 #define NOP     0x60000000
  64 #define RET     0x4e800020      /* blr */
  65 #define LOCKSTAT_AFRAMES 2
  66 #else
  67 #error "not ported to this architecture"
  68 #endif
  69
  70
  71 typedef struct lockstat_probe {
  72         const char      *lsp_func;
  73         const char      *lsp_name;
  74         int             lsp_probe;
  75         dtrace_id_t     lsp_id;
  76 } lockstat_probe_t;
  77
  78 lockstat_probe_t lockstat_probes[] =
  79 {
  80 #if defined(__i386__) || defined(__x86_64__)
  81         /* Not implemented yet on PPC... */
  82         { LS_LCK_MTX_LOCK,      LSA_ACQUIRE,    LS_LCK_MTX_LOCK_ACQUIRE, DTRACE_IDNONE },
  83         { LS_LCK_MTX_LOCK,      LSA_SPIN,       LS_LCK_MTX_LOCK_SPIN, DTRACE_IDNONE },
  84         { LS_LCK_MTX_TRY_LOCK,  LSA_ACQUIRE,    LS_LCK_MTX_TRY_LOCK_ACQUIRE, DTRACE_IDNONE },
  85         { LS_LCK_MTX_TRY_SPIN_LOCK, LSA_ACQUIRE, LS_LCK_MTX_TRY_SPIN_LOCK_ACQUIRE, DTRACE_IDNONE },
  86         { LS_LCK_MTX_UNLOCK,    LSA_RELEASE,    LS_LCK_MTX_UNLOCK_RELEASE, DTRACE_IDNONE },
  87         { LS_LCK_MTX_EXT_LOCK,  LSA_ACQUIRE,    LS_LCK_MTX_EXT_LOCK_ACQUIRE, DTRACE_IDNONE },
  88         { LS_LCK_MTX_EXT_LOCK,  LSA_SPIN,       LS_LCK_MTX_EXT_LOCK_SPIN, DTRACE_IDNONE },
  89         { LS_LCK_MTX_EXT_TRY_LOCK, LSA_ACQUIRE, LS_LCK_MTX_TRY_EXT_LOCK_ACQUIRE, DTRACE_IDNONE },
  90         { LS_LCK_MTX_UNLOCK,    LSA_RELEASE,    LS_LCK_MTX_EXT_UNLOCK_RELEASE, DTRACE_IDNONE },
  91         { LS_LCK_MTX_LOCK_SPIN_LOCK,    LSA_ACQUIRE,    LS_LCK_MTX_LOCK_SPIN_ACQUIRE, DTRACE_IDNONE },
  92 #endif
  93         { LS_LCK_MTX_LOCK,      LSA_BLOCK,      LS_LCK_MTX_LOCK_BLOCK, DTRACE_IDNONE },
  94         { LS_LCK_MTX_EXT_LOCK,  LSA_BLOCK,      LS_LCK_MTX_EXT_LOCK_BLOCK, DTRACE_IDNONE },
  95
  96         { LS_LCK_RW_LOCK_SHARED,        LSR_ACQUIRE,    LS_LCK_RW_LOCK_SHARED_ACQUIRE, DTRACE_IDNONE },
  97         { LS_LCK_RW_LOCK_SHARED,        LSR_BLOCK,      LS_LCK_RW_LOCK_SHARED_BLOCK, DTRACE_IDNONE },
  98         { LS_LCK_RW_LOCK_SHARED,        LSR_SPIN,       LS_LCK_RW_LOCK_SHARED_SPIN, DTRACE_IDNONE },
  99         { LS_LCK_RW_LOCK_EXCL,          LSR_ACQUIRE,    LS_LCK_RW_LOCK_EXCL_ACQUIRE, DTRACE_IDNONE },
 100         { LS_LCK_RW_LOCK_EXCL,          LSR_BLOCK,      LS_LCK_RW_LOCK_EXCL_BLOCK, DTRACE_IDNONE },
 101         { LS_LCK_RW_LOCK_EXCL,          LSR_SPIN,       LS_LCK_RW_LOCK_EXCL_SPIN, DTRACE_IDNONE },
 102         { LS_LCK_RW_DONE,               LSR_RELEASE,    LS_LCK_RW_DONE_RELEASE, DTRACE_IDNONE },
 103         { LS_LCK_RW_TRY_LOCK_SHARED,    LSR_ACQUIRE,    LS_LCK_RW_TRY_LOCK_SHARED_ACQUIRE, DTRACE_IDNONE },
 104         { LS_LCK_RW_TRY_LOCK_EXCL,      LSR_ACQUIRE,    LS_LCK_RW_TRY_LOCK_EXCL_ACQUIRE, DTRACE_IDNONE },
 105         { LS_LCK_RW_LOCK_SHARED_TO_EXCL, LSR_UPGRADE,   LS_LCK_RW_LOCK_SHARED_TO_EXCL_UPGRADE, DTRACE_IDNONE },
 106         { LS_LCK_RW_LOCK_SHARED_TO_EXCL,        LSR_BLOCK,      LS_LCK_RW_LOCK_SHARED_TO_EXCL_BLOCK, DTRACE_IDNONE },
 107         { LS_LCK_RW_LOCK_SHARED_TO_EXCL,        LSR_SPIN,       LS_LCK_RW_LOCK_SHARED_TO_EXCL_SPIN, DTRACE_IDNONE },
 108         { LS_LCK_RW_LOCK_EXCL_TO_SHARED,        LSR_DOWNGRADE,  LS_LCK_RW_LOCK_EXCL_TO_SHARED_DOWNGRADE, DTRACE_IDNONE },
 109
 110
 111 #ifdef  LATER
 112         /* Interlock and spinlock measurements would be nice, but later */
 113         { LS_LCK_SPIN_LOCK,     LSS_ACQUIRE,    LS_LCK_SPIN_LOCK_ACQUIRE, DTRACE_IDNONE },
 114         { LS_LCK_SPIN_LOCK,     LSS_SPIN,       LS_LCK_SPIN_LOCK_SPIN, DTRACE_IDNONE },
 115         { LS_LCK_SPIN_UNLOCK,   LSS_RELEASE,    LS_LCK_SPIN_UNLOCK_RELEASE, DTRACE_IDNONE },
 116
 117         { LS_LCK_RW_LOCK_EXCL_TO_SHARED,        LSA_ILK_SPIN,   LS_LCK_RW_LOCK_EXCL_TO_SHARED_ILK_SPIN, DTRACE_IDNONE },
 118         { LS_LCK_MTX_LOCK,      LSA_ILK_SPIN,   LS_LCK_MTX_LOCK_ILK_SPIN, DTRACE_IDNONE },
 119         { LS_LCK_MTX_EXT_LOCK,  LSA_ILK_SPIN,   LS_LCK_MTX_EXT_LOCK_ILK_SPIN, DTRACE_IDNONE },
 120         { LS_LCK_RW_TRY_LOCK_EXCL,      LSA_ILK_SPIN,   LS_LCK_RW_TRY_LOCK_EXCL_ILK_SPIN, DTRACE_IDNONE },
 121         { LS_LCK_RW_TRY_LOCK_SHARED,    LSA_SPIN,       LS_LCK_RW_TRY_LOCK_SHARED_SPIN, DTRACE_IDNONE },
 122 #endif
 123
 124         { NULL, NULL, 0, 0 }
 125 };
 126
 127 dtrace_id_t lockstat_probemap[LS_NPROBES];
 128
 129 #if CONFIG_DTRACE
 130 extern void lck_mtx_lock_lockstat_patch_point(void);
 131 extern void lck_mtx_try_lock_lockstat_patch_point(void);
 132 extern void lck_mtx_try_lock_spin_lockstat_patch_point(void);
 133 extern void lck_mtx_unlock_lockstat_patch_point(void);
 134 extern void lck_mtx_lock_ext_lockstat_patch_point(void);
 135 extern void lck_mtx_ext_unlock_lockstat_patch_point(void);
 136
 137 extern void lck_rw_lock_shared_lockstat_patch_point(void);
 138 extern void lck_rw_lock_exclusive_lockstat_patch_point(void);
 139 extern void lck_rw_lock_shared_to_exclusive_lockstat_patch_point(void);
 140 extern void lck_rw_try_lock_shared_lockstat_patch_point(void);
 141 extern void lck_rw_try_lock_exclusive_lockstat_patch_point(void);
 142 extern void lck_mtx_lock_spin_lockstat_patch_point(void);
 143 #endif /* CONFIG_DTRACE */
 144
 145 vm_offset_t *assembly_probes[] = {
 146 #if CONFIG_DTRACE
 147 #if defined(__i386__) || defined(__x86_64__)
 148         /*
 149          * On x86 these points are better done via hot patches, which ensure
 150          * there is zero overhead when not in use.  On x86 these patch points
 151          * are swapped between the return instruction and a no-op, with the
 152          * Dtrace call following the return.
 153          */
 154         (vm_offset_t *) lck_mtx_lock_lockstat_patch_point,
 155         (vm_offset_t *) lck_mtx_try_lock_lockstat_patch_point,
 156         (vm_offset_t *) lck_mtx_try_lock_spin_lockstat_patch_point,
 157         (vm_offset_t *) lck_mtx_unlock_lockstat_patch_point,
 158         (vm_offset_t *) lck_mtx_lock_ext_lockstat_patch_point,
 159         (vm_offset_t *) lck_mtx_ext_unlock_lockstat_patch_point,
 160         (vm_offset_t *) lck_rw_lock_shared_lockstat_patch_point,
 161         (vm_offset_t *) lck_rw_lock_exclusive_lockstat_patch_point,
 162         (vm_offset_t *) lck_rw_lock_shared_to_exclusive_lockstat_patch_point,
 163         (vm_offset_t *) lck_rw_try_lock_shared_lockstat_patch_point,
 164         (vm_offset_t *) lck_rw_try_lock_exclusive_lockstat_patch_point,
 165         (vm_offset_t *) lck_mtx_lock_spin_lockstat_patch_point,
 166 #else
 167         (vm_offset_t *) lck_mtx_unlock_lockstat_patch_point,
 168 #endif
 169 #endif /* CONFIG_DTRACE */
 170         NULL
 171 };
 172 /*
 173  * Hot patch switches back and forth the probe points between NOP and RET.
 174  * The argument indicates whether the probe point is on or off.
 175  */
 176 #if defined(__APPLE__)
 177 static
 178 #endif /* __APPLE__ */
 179 void lockstat_hot_patch(boolean_t active)
 180 {
 181 #pragma unused(active)
 182         int i;
 183
 184
 185         for (i = 0; assembly_probes[i]; i++) {
 186 #if defined(__i386__) || defined(__x86_64__)
 187                 uint8_t instr;
 188                 instr = (active ? NOP : RET );
 189                 (void) ml_nofault_copy( (vm_offset_t)&instr, *(assembly_probes[i]),
 190                                                                 sizeof(instr));
 191 #endif
 192 #ifdef __ppc__
 193                 uint32_t instr;
 194                 instr = (active ? NOP : RET );
 195                 (void) ml_nofault_copy( (vm_offset_t)&instr, *(assembly_probes[i]), sizeof(instr));
 196 #endif
 197         }
 198 }
 199
 200
 201
 202 void (*lockstat_probe)(dtrace_id_t, uint64_t, uint64_t,
 203                                     uint64_t, uint64_t, uint64_t);
 204
 205 static dev_info_t       *lockstat_devi; /* saved in xxattach() for xxinfo() */
 206 static dtrace_provider_id_t lockstat_id;
 207
 208 /*ARGSUSED*/
 209 static void
 210 lockstat_enable(void *arg, dtrace_id_t id, void *parg)
 211 {
 212 #pragma unused(arg) /* __APPLE__ */
 213
 214         lockstat_probe_t *probe = parg;
 215
 216         ASSERT(!lockstat_probemap[probe->lsp_probe]);
 217
 218         lockstat_probemap[probe->lsp_probe] = id;
 219         membar_producer();
 220
 221         lockstat_hot_patch(TRUE);
 222         membar_producer();
 223
 224 }
 225
 226 /*ARGSUSED*/
 227 static void
 228 lockstat_disable(void *arg, dtrace_id_t id, void *parg)
 229 {
 230 #pragma unused(arg, id) /* __APPLE__ */
 231
 232         lockstat_probe_t *probe = parg;
 233         int i;
 234
 235         ASSERT(lockstat_probemap[probe->lsp_probe]);
 236
 237         lockstat_probemap[probe->lsp_probe] = 0;
 238         lockstat_hot_patch(FALSE);
 239         membar_producer();
 240
 241         /*
 242          * See if we have any probes left enabled.
 243          */
 244         for (i = 0; i < LS_NPROBES; i++) {
 245                 if (lockstat_probemap[i]) {
 246                         /*
 247                          * This probe is still enabled.  We don't need to deal
 248                          * with waiting for all threads to be out of the
 249                          * lockstat critical sections; just return.
 250                          */
 251                         return;
 252                 }
 253         }
 254
 255 }
 256
 257 /*ARGSUSED*/
 258 static void
 259 lockstat_provide(void *arg, const dtrace_probedesc_t *desc)
 260 {
 261 #pragma unused(arg, desc) /* __APPLE__ */
 262
 263         int i = 0;
 264
 265         for (i = 0; lockstat_probes[i].lsp_func != NULL; i++) {
 266                 lockstat_probe_t *probe = &lockstat_probes[i];
 267
 268                 if (dtrace_probe_lookup(lockstat_id, "mach_kernel",
 269                     probe->lsp_func, probe->lsp_name) != 0)
 270                         continue;
 271
 272                 ASSERT(!probe->lsp_id);
 273                 probe->lsp_id = dtrace_probe_create(lockstat_id,
 274                     "mach_kernel", probe->lsp_func, probe->lsp_name,
 275                     LOCKSTAT_AFRAMES, probe);
 276         }
 277 }
 278
 279
 280 /*ARGSUSED*/
 281 static void
 282 lockstat_destroy(void *arg, dtrace_id_t id, void *parg)
 283 {
 284 #pragma unused(arg, id) /* __APPLE__ */
 285
 286         lockstat_probe_t *probe = parg;
 287
 288         ASSERT(!lockstat_probemap[probe->lsp_probe]);
 289         probe->lsp_id = 0;
 290 }
 291
 292 static dtrace_pattr_t lockstat_attr = {
 293 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 294 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
 295 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
 296 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 297 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
 298 };
 299
 300 static dtrace_pops_t lockstat_pops = {
 301         lockstat_provide,
 302         NULL,
 303         lockstat_enable,
 304         lockstat_disable,
 305         NULL,
 306         NULL,
 307         NULL,
 308         NULL,
 309         NULL,
 310         lockstat_destroy
 311 };
 312
 313 static int
 314 lockstat_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
 315 {
 316         switch (cmd) {
 317         case DDI_ATTACH:
 318                 break;
 319         case DDI_RESUME:
 320                 return (DDI_SUCCESS);
 321         default:
 322                 return (DDI_FAILURE);
 323         }
 324
 325         if (ddi_create_minor_node(devi, "lockstat", S_IFCHR, 0,
 326             DDI_PSEUDO, 0) == DDI_FAILURE ||
 327             dtrace_register("lockstat", &lockstat_attr, DTRACE_PRIV_KERNEL,
 328             NULL, &lockstat_pops, NULL, &lockstat_id) != 0) {
 329                 ddi_remove_minor_node(devi, NULL);
 330                 return (DDI_FAILURE);
 331         }
 332
 333         lockstat_probe = dtrace_probe;
 334         membar_producer();
 335
 336         ddi_report_dev(devi);
 337         lockstat_devi = devi;
 338         return (DDI_SUCCESS);
 339 }
 340
 341 d_open_t _lockstat_open;
 342
 343 int _lockstat_open(dev_t dev, int flags, int devtype, struct proc *p)
 344 {
 345 #pragma unused(dev,flags,devtype,p)
 346         return 0;
 347 }
 348
 349 #define LOCKSTAT_MAJOR  -24 /* let the kernel pick the device number */
 350
 351 /*
 352  * A struct describing which functions will get invoked for certain
 353  * actions.
 354  */
 355 static struct cdevsw lockstat_cdevsw =
 356 {
 357         _lockstat_open,         /* open */
 358         eno_opcl,                       /* close */
 359         eno_rdwrt,                      /* read */
 360         eno_rdwrt,                      /* write */
 361         eno_ioctl,                      /* ioctl */
 362         (stop_fcn_t *)nulldev, /* stop */
 363         (reset_fcn_t *)nulldev, /* reset */
 364         NULL,                           /* tty's */
 365         eno_select,                     /* select */
 366         eno_mmap,                       /* mmap */
 367         eno_strat,                      /* strategy */
 368         eno_getc,                       /* getc */
 369         eno_putc,                       /* putc */
 370         0                                       /* type */
 371 };
 372
 373 static int gLockstatInited = 0;
 374
 375 void lockstat_init( void );
 376
 377 void lockstat_init( void )
 378 {
 379         if (0 == gLockstatInited)
 380         {
 381                 int majdevno = cdevsw_add(LOCKSTAT_MAJOR, &lockstat_cdevsw);
 382
 383                 if (majdevno < 0) {
 384                         printf("lockstat_init: failed to allocate a major number!\n");
 385                         gLockstatInited = 0;
 386                         return;
 387                 }
 388
 389                 lockstat_attach( (dev_info_t    *)(uintptr_t)majdevno, DDI_ATTACH );
 390                 gLockstatInited = 1;
 391         } else
 392                 panic("lockstat_init: called twice!\n");
 393 }
 394 #undef LOCKSTAT_MAJOR