dtrace.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright 2019 Joyent, Inc.
 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
 */

/*
 * DTrace - Dynamic Tracing for Solaris
 *
 * This is the implementation of the Solaris Dynamic Tracing framework
 * (DTrace).  The user-visible interface to DTrace is described at length in
 * the "Solaris Dynamic Tracing Guide".  The interfaces between the libdtrace
 * library, the in-kernel DTrace framework, and the DTrace providers are
 * described in the block comments in the <sys/dtrace.h> header file.  The
 * internal architecture of DTrace is described in the block comments in the
 * <sys/dtrace_impl.h> header file.  The comments contained within the DTrace
 * implementation very much assume mastery of all of these sources; if one has
 * an unanswered question about the implementation, one should consult them
 * first.
 *
 * The functions here are ordered roughly as follows:
 *
 *   - Probe context functions
 *   - Probe hashing functions
 *   - Non-probe context utility functions
 *   - Matching functions
 *   - Provider-to-Framework API functions
 *   - Probe management functions
 *   - DIF object functions
 *   - Format functions
 *   - Predicate functions
 *   - ECB functions
 *   - Buffer functions
 *   - Enabling functions
 *   - DOF functions
 *   - Anonymous enabling functions
 *   - Consumer state functions
 *   - Helper functions
 *   - Hook functions
 *   - Driver cookbook functions
 *
 * Each group of functions begins with a block comment labelled the "DTrace
 * [Group] Functions", allowing one to find each block by searching forward
 * on capital-f functions.
 */
#include <sys/errno.h>
#include <sys/stat.h>
#include <sys/modctl.h>
#include <sys/conf.h>
#include <sys/systm.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/cpuvar.h>
#include <sys/kmem.h>
#include <sys/strsubr.h>
#include <sys/sysmacros.h>
#include <sys/dtrace_impl.h>
#include <sys/atomic.h>
#include <sys/cmn_err.h>
#include <sys/mutex_impl.h>
#include <sys/rwlock_impl.h>
#include <sys/ctf_api.h>
#include <sys/panic.h>
#include <sys/priv_impl.h>
#include <sys/policy.h>
#include <sys/cred_impl.h>
#include <sys/procfs_isa.h>
#include <sys/taskq.h>
#include <sys/mkdev.h>
#include <sys/kdi.h>
#include <sys/zone.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include "strtolctype.h"

/*
 * DTrace Tunable Variables
 *
 * The following variables may be tuned by adding a line to /etc/system that
 * includes both the name of the DTrace module ("dtrace") and the name of the
 * variable.  For example:
 *
 *   set dtrace:dtrace_destructive_disallow = 1
 *
 * In general, the only variables that one should be tuning this way are those
 * that affect system-wide DTrace behavior, and for which the default behavior
 * is undesirable.  Most of these variables are tunable on a per-consumer
 * basis using DTrace options, and need not be tuned on a system-wide basis.
 * When tuning these variables, avoid pathological values; while some attempt
 * is made to verify the integrity of these variables, they are not considered
 * part of the supported interface to DTrace, and they are therefore not
 * checked comprehensively.  Further, these variables should not be tuned
 * dynamically via "mdb -kw" or other means; they should only be tuned via
 * /etc/system.
 */
int		dtrace_destructive_disallow = 0;
dtrace_optval_t	dtrace_nonroot_maxsize = (16 * 1024 * 1024);
size_t		dtrace_difo_maxsize = (256 * 1024);
dtrace_optval_t	dtrace_dof_maxsize = (8 * 1024 * 1024);
size_t		dtrace_statvar_maxsize = (16 * 1024);
size_t		dtrace_actions_max = (16 * 1024);
size_t		dtrace_retain_max = 1024;
dtrace_optval_t	dtrace_helper_actions_max = 1024;
dtrace_optval_t	dtrace_helper_providers_max = 32;
dtrace_optval_t	dtrace_dstate_defsize = (1 * 1024 * 1024);
size_t		dtrace_strsize_default = 256;
dtrace_optval_t	dtrace_cleanrate_default = 9900990;		/* 101 hz */
dtrace_optval_t	dtrace_cleanrate_min = 200000;			/* 5000 hz */
dtrace_optval_t	dtrace_cleanrate_max = (uint64_t)60 * NANOSEC;	/* 1/minute */
dtrace_optval_t	dtrace_aggrate_default = NANOSEC;		/* 1 hz */
dtrace_optval_t	dtrace_statusrate_default = NANOSEC;		/* 1 hz */
dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC;	 /* 6/minute */
dtrace_optval_t	dtrace_switchrate_default = NANOSEC;		/* 1 hz */
dtrace_optval_t	dtrace_nspec_default = 1;
dtrace_optval_t	dtrace_specsize_default = 32 * 1024;
dtrace_optval_t dtrace_stackframes_default = 20;
dtrace_optval_t dtrace_ustackframes_default = 20;
dtrace_optval_t dtrace_jstackframes_default = 50;
dtrace_optval_t dtrace_jstackstrsize_default = 512;
int		dtrace_msgdsize_max = 128;
hrtime_t	dtrace_chill_max = MSEC2NSEC(500);		/* 500 ms */
hrtime_t	dtrace_chill_interval = NANOSEC;		/* 1000 ms */
int		dtrace_devdepth_max = 32;
int		dtrace_err_verbose;
hrtime_t	dtrace_deadman_interval = NANOSEC;
hrtime_t	dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC;
hrtime_t	dtrace_deadman_user = (hrtime_t)30 * NANOSEC;
hrtime_t	dtrace_unregister_defunct_reap = (hrtime_t)60 * NANOSEC;

/*
 * DTrace External Variables
 *
 * As dtrace(7D) is a kernel module, any DTrace variables are obviously
 * available to DTrace consumers via the backtick (`) syntax.  One of these,
 * dtrace_zero, is made deliberately so:  it is provided as a source of
 * well-known, zero-filled memory.  While this variable is not documented,
 * it is used by some translators as an implementation detail.
 */
const char	dtrace_zero[256] = { 0 };	/* zero-filled memory */

/*
 * DTrace Internal Variables
 */
static dev_info_t	*dtrace_devi;		/* device info */
static vmem_t		*dtrace_arena;		/* probe ID arena */
static vmem_t		*dtrace_minor;		/* minor number arena */
static taskq_t		*dtrace_taskq;		/* task queue */
static dtrace_probe_t	**dtrace_probes;	/* array of all probes */
static int		dtrace_nprobes;		/* number of probes */
static dtrace_provider_t *dtrace_provider;	/* provider list */
static dtrace_meta_t	*dtrace_meta_pid;	/* user-land meta provider */
static int		dtrace_opens;		/* number of opens */
static int		dtrace_helpers;		/* number of helpers */
static int		dtrace_getf;		/* number of unpriv getf()s */
static void		*dtrace_softstate;	/* softstate pointer */
static dtrace_hash_t	*dtrace_bymod;		/* probes hashed by module */
static dtrace_hash_t	*dtrace_byfunc;		/* probes hashed by function */
static dtrace_hash_t	*dtrace_byname;		/* probes hashed by name */
static dtrace_toxrange_t *dtrace_toxrange;	/* toxic range array */
static int		dtrace_toxranges;	/* number of toxic ranges */
static int		dtrace_toxranges_max;	/* size of toxic range array */
static dtrace_anon_t	dtrace_anon;		/* anonymous enabling */
static kmem_cache_t	*dtrace_state_cache;	/* cache for dynamic state */
static uint64_t		dtrace_vtime_references; /* number of vtimestamp refs */
static kthread_t	*dtrace_panicked;	/* panicking thread */
static dtrace_ecb_t	*dtrace_ecb_create_cache; /* cached created ECB */
static dtrace_genid_t	dtrace_probegen;	/* current probe generation */
static dtrace_helpers_t *dtrace_deferred_pid;	/* deferred helper list */
static dtrace_enabling_t *dtrace_retained;	/* list of retained enablings */
static dtrace_genid_t	dtrace_retained_gen;	/* current retained enab gen */
static dtrace_dynvar_t	dtrace_dynhash_sink;	/* end of dynamic hash chains */
static int		dtrace_dynvar_failclean; /* dynvars failed to clean */

/*
 * DTrace Locking
 * DTrace is protected by three (relatively coarse-grained) locks:
 *
 * (1) dtrace_lock is required to manipulate essentially any DTrace state,
 *     including enabling state, probes, ECBs, consumer state, helper state,
 *     etc.  Importantly, dtrace_lock is _not_ required when in probe context;
 *     probe context is lock-free -- synchronization is handled via the
 *     dtrace_sync() cross call mechanism.
 *
 * (2) dtrace_provider_lock is required when manipulating provider state, or
 *     when provider state must be held constant.
 *
 * (3) dtrace_meta_lock is required when manipulating meta provider state, or
 *     when meta provider state must be held constant.
 *
 * The lock ordering between these three locks is dtrace_meta_lock before
 * dtrace_provider_lock before dtrace_lock.  (In particular, there are
 * several places where dtrace_provider_lock is held by the framework as it
 * calls into the providers -- which then call back into the framework,
 * grabbing dtrace_lock.)
 *
 * There are two other locks in the mix:  mod_lock and cpu_lock.  With respect
 * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical
 * role as a coarse-grained lock; it is acquired before both of these locks.
 * With respect to dtrace_meta_lock, its behavior is stranger:  cpu_lock must
 * be acquired _between_ dtrace_meta_lock and any other DTrace locks.
 * mod_lock is similar with respect to dtrace_provider_lock in that it must be
 * acquired _between_ dtrace_provider_lock and dtrace_lock.
 */
static kmutex_t		dtrace_lock;		/* probe state lock */
static kmutex_t		dtrace_provider_lock;	/* provider state lock */
static kmutex_t		dtrace_meta_lock;	/* meta-provider state lock */

/*
 * DTrace Provider Variables
 *
 * These are the variables relating to DTrace as a provider (that is, the
 * provider of the BEGIN, END, and ERROR probes).
 */
static dtrace_pattr_t	dtrace_provider_attr = {
{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
};

static void
dtrace_nullop_provide(void *arg __unused,
    const dtrace_probedesc_t *spec __unused)
{
}

static void
dtrace_nullop_module(void *arg __unused, struct modctl *mp __unused)
{
}

static void
dtrace_nullop(void *arg __unused, dtrace_id_t id __unused, void *parg __unused)
{
}

static int
dtrace_enable_nullop(void *arg __unused, dtrace_id_t id __unused,
    void *parg __unused)
{
	return (0);
}

static dtrace_pops_t	dtrace_provider_ops = {
	.dtps_provide = dtrace_nullop_provide,
	.dtps_provide_module = dtrace_nullop_module,
	.dtps_enable = dtrace_enable_nullop,
	.dtps_disable = dtrace_nullop,
	.dtps_suspend = dtrace_nullop,
	.dtps_resume = dtrace_nullop,
	.dtps_getargdesc = NULL,
	.dtps_getargval = NULL,
	.dtps_mode = NULL,
	.dtps_destroy = dtrace_nullop
};

static dtrace_id_t	dtrace_probeid_begin;	/* special BEGIN probe */
static dtrace_id_t	dtrace_probeid_end;	/* special END probe */
dtrace_id_t		dtrace_probeid_error;	/* special ERROR probe */

/*
 * DTrace Helper Tracing Variables
 *
 * These variables should be set dynamically to enable helper tracing.  The
 * only variables that should be set are dtrace_helptrace_enable (which should
 * be set to a non-zero value to allocate helper tracing buffers on the next
 * open of /dev/dtrace) and dtrace_helptrace_disable (which should be set to a
 * non-zero value to deallocate helper tracing buffers on the next close of
 * /dev/dtrace).  When (and only when) helper tracing is disabled, the
 * buffer size may also be set via dtrace_helptrace_bufsize.
 */
int			dtrace_helptrace_enable = 0;
int			dtrace_helptrace_disable = 0;
int			dtrace_helptrace_bufsize = 16 * 1024 * 1024;
uint32_t		dtrace_helptrace_nlocals;
static dtrace_helptrace_t *dtrace_helptrace_buffer;
static uint32_t		dtrace_helptrace_next = 0;
static int		dtrace_helptrace_wrapped = 0;

/*
 * DTrace Error Hashing
 *
 * On DEBUG kernels, DTrace will track the errors that has seen in a hash
 * table.  This is very useful for checking coverage of tests that are
 * expected to induce DIF or DOF processing errors, and may be useful for
 * debugging problems in the DIF code generator or in DOF generation .  The
 * error hash may be examined with the ::dtrace_errhash MDB dcmd.
 */
#ifdef DEBUG
static dtrace_errhash_t	dtrace_errhash[DTRACE_ERRHASHSZ];
static const char *dtrace_errlast;
static kthread_t *dtrace_errthread;
static kmutex_t dtrace_errlock;
#endif

/*
 * DTrace Macros and Constants
 *
 * These are various macros that are useful in various spots in the
 * implementation, along with a few random constants that have no meaning
 * outside of the implementation.  There is no real structure to this cpp
 * mishmash -- but is there ever?
 */
#define	DTRACE_HASHSTR(hash, probe)	\
	dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs)))

#define	DTRACE_HASHNEXT(hash, probe)	\
	(dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs)

#define	DTRACE_HASHPREV(hash, probe)	\
	(dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs)

#define	DTRACE_HASHEQ(hash, lhs, rhs)	\
	(strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \
	    *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0)

#define	DTRACE_AGGHASHSIZE_SLEW		17

#define	DTRACE_V4MAPPED_OFFSET		(sizeof (uint32_t) * 3)

/*
 * The key for a thread-local variable consists of the lower 61 bits of the
 * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL.
 * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never
 * equal to a variable identifier.  This is necessary (but not sufficient) to
 * assure that global associative arrays never collide with thread-local
 * variables.  To guarantee that they cannot collide, we must also define the
 * order for keying dynamic variables.  That order is:
 *
 *   [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ]
 *
 * Because the variable-key and the tls-key are in orthogonal spaces, there is
 * no way for a global variable key signature to match a thread-local key
 * signature.
 */
#define	DTRACE_TLS_THRKEY(where) { \
	uint_t intr = 0; \
	uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \
	for (; actv; actv >>= 1) \
		intr++; \
	ASSERT(intr < (1 << 3)); \
	(where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \
	    (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \
}

#define	DT_BSWAP_8(x)	((x) & 0xff)
#define	DT_BSWAP_16(x)	((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8))
#define	DT_BSWAP_32(x)	((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16))
#define	DT_BSWAP_64(x)	((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32))

#define	DT_MASK_LO 0x00000000FFFFFFFFULL

#define	DTRACE_STORE(type, tomax, offset, what) \
	*((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what);

#ifndef __x86
#define	DTRACE_ALIGNCHECK(addr, size, flags)				\
	if (addr & (size - 1)) {					\
		*flags |= CPU_DTRACE_BADALIGN;				\
		cpu_core[CPU->cpu_id].cpuc_dtrace_illval = addr;	\
		return (0);						\
	}
#else
#define	DTRACE_ALIGNCHECK(addr, size, flags)
#endif

/*
 * Test whether a range of memory starting at testaddr of size testsz falls
 * within the range of memory described by addr, sz.  We take care to avoid
 * problems with overflow and underflow of the unsigned quantities, and
 * disallow all negative sizes.  Ranges of size 0 are allowed.
 */
#define	DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \
	((testaddr) - (uintptr_t)(baseaddr) < (basesz) && \
	(testaddr) + (testsz) - (uintptr_t)(baseaddr) <= (basesz) && \
	(testaddr) + (testsz) >= (testaddr))

#define	DTRACE_RANGE_REMAIN(remp, addr, baseaddr, basesz)		\
do {									\
	if ((remp) != NULL) {						\
		*(remp) = (uintptr_t)(baseaddr) + (basesz) - (addr);	\
	}								\
_NOTE(CONSTCOND) } while (0)


/*
 * Test whether alloc_sz bytes will fit in the scratch region.  We isolate
 * alloc_sz on the righthand side of the comparison in order to avoid overflow
 * or underflow in the comparison with it.  This is simpler than the INRANGE
 * check above, because we know that the dtms_scratch_ptr is valid in the
 * range.  Allocations of size zero are allowed.
 */
#define	DTRACE_INSCRATCH(mstate, alloc_sz) \
	((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \
	(mstate)->dtms_scratch_ptr >= (alloc_sz))

#define	DTRACE_LOADFUNC(bits)						\
/*CSTYLED*/								\
uint##bits##_t								\
dtrace_load##bits(uintptr_t addr)					\
{									\
	size_t size = bits / NBBY;					\
	/*CSTYLED*/							\
	uint##bits##_t rval;						\
	int i;								\
	volatile uint16_t *flags = (volatile uint16_t *)		\
	    &cpu_core[CPU->cpu_id].cpuc_dtrace_flags;			\
									\
	DTRACE_ALIGNCHECK(addr, size, flags);				\
									\
	for (i = 0; i < dtrace_toxranges; i++) {			\
		if (addr >= dtrace_toxrange[i].dtt_limit)		\
			continue;					\
									\
		if (addr + size <= dtrace_toxrange[i].dtt_base)		\
			continue;					\
									\
		/*							\
		 * This address falls within a toxic region; return 0.	\
		 */							\
		*flags |= CPU_DTRACE_BADADDR;				\
		cpu_core[CPU->cpu_id].cpuc_dtrace_illval = addr;	\
		return (0);						\
	}								\
									\
	*flags |= CPU_DTRACE_NOFAULT;					\
	/*CSTYLED*/							\
	rval = *((volatile uint##bits##_t *)addr);			\
	*flags &= ~CPU_DTRACE_NOFAULT;					\
									\
	return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0);		\
}

#ifdef _LP64
#define	dtrace_loadptr	dtrace_load64
#else
#define	dtrace_loadptr	dtrace_load32
#endif

#define	DTRACE_DYNHASH_FREE	0
#define	DTRACE_DYNHASH_SINK	1
#define	DTRACE_DYNHASH_VALID	2

#define	DTRACE_MATCH_FAIL	-1
#define	DTRACE_MATCH_NEXT	0
#define	DTRACE_MATCH_DONE	1
#define	DTRACE_ANCHORED(probe)	((probe)->dtpr_func[0] != '\0')
#define	DTRACE_STATE_ALIGN	64

#define	DTRACE_FLAGS2FLT(flags)						\
	(((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR :		\
	((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP :		\
	((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO :		\
	((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV :		\
	((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV :		\
	((flags) & CPU_DTRACE_TUPOFLOW) ?  DTRACEFLT_TUPOFLOW :		\
	((flags) & CPU_DTRACE_BADALIGN) ?  DTRACEFLT_BADALIGN :		\
	((flags) & CPU_DTRACE_NOSCRATCH) ?  DTRACEFLT_NOSCRATCH :	\
	((flags) & CPU_DTRACE_BADSTACK) ?  DTRACEFLT_BADSTACK :		\
	DTRACEFLT_UNKNOWN)

#define	DTRACEACT_ISSTRING(act)						\
	((act)->dta_kind == DTRACEACT_DIFEXPR &&			\
	(act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING)

static size_t dtrace_strlen(const char *, size_t);
static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id);
static void dtrace_enabling_provide(dtrace_provider_t *);
static int dtrace_enabling_match(dtrace_enabling_t *, int *);
static void dtrace_enabling_matchall(void);
static void dtrace_enabling_reap(void);
static dtrace_state_t *dtrace_anon_grab(void);
static uint64_t dtrace_helper(int, dtrace_mstate_t *,
    dtrace_state_t *, uint64_t, uint64_t);
static dtrace_helpers_t *dtrace_helpers_create(proc_t *);
static void dtrace_buffer_drop(dtrace_buffer_t *);
static int dtrace_buffer_consumed(dtrace_buffer_t *, hrtime_t when);
static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t,
    dtrace_state_t *, dtrace_mstate_t *);
static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t,
    dtrace_optval_t);
static int dtrace_ecb_create_enable(dtrace_probe_t *, void *);
static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *);
static int dtrace_priv_proc(dtrace_state_t *, dtrace_mstate_t *);
static void dtrace_getf_barrier(void);
static int dtrace_canload_remains(uint64_t, size_t, size_t *,
    dtrace_mstate_t *, dtrace_vstate_t *);
static int dtrace_canstore_remains(uint64_t, size_t, size_t *,
    dtrace_mstate_t *, dtrace_vstate_t *);

/*
 * DTrace Probe Context Functions
 *
 * These functions are called from probe context.  Because probe context is
 * any context in which C may be called, arbitrarily locks may be held,
 * interrupts may be disabled, we may be in arbitrary dispatched state, etc.
 * As a result, functions called from probe context may only call other DTrace
 * support functions -- they may not interact at all with the system at large.
 * (Note that the ASSERT macro is made probe-context safe by redefining it in
 * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary
 * loads are to be performed from probe context, they _must_ be in terms of
 * the safe dtrace_load*() variants.
 *
 * Some functions in this block are not actually called from probe context;
 * for these functions, there will be a comment above the function reading
 * "Note:  not called from probe context."
 */
void
dtrace_panic(const char *format, ...)
{
	va_list alist;

	va_start(alist, format);
	dtrace_vpanic(format, alist);
	va_end(alist);
}

int
dtrace_assfail(const char *a, const char *f, int l)
{
	dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l);

	/*
	 * We just need something here that even the most clever compiler
	 * cannot optimize away.
	 */
	return (a[(uintptr_t)f]);
}

/*
 * Atomically increment a specified error counter from probe context.
 */
static void
dtrace_error(uint32_t *counter)
{
	/*
	 * Most counters stored to in probe context are per-CPU counters.
	 * However, there are some error conditions that are sufficiently
	 * arcane that they don't merit per-CPU storage.  If these counters
	 * are incremented concurrently on different CPUs, scalability will be
	 * adversely affected -- but we don't expect them to be white-hot in a
	 * correctly constructed enabling...
	 */
	uint32_t oval, nval;

	do {
		oval = *counter;

		if ((nval = oval + 1) == 0) {
			/*
			 * If the counter would wrap, set it to 1 -- assuring
			 * that the counter is never zero when we have seen
			 * errors.  (The counter must be 32-bits because we
			 * aren't guaranteed a 64-bit compare&swap operation.)
			 * To save this code both the infamy of being fingered
			 * by a priggish news story and the indignity of being
			 * the target of a neo-puritan witch trial, we're
			 * carefully avoiding any colorful description of the
			 * likelihood of this condition -- but suffice it to
			 * say that it is only slightly more likely than the
			 * overflow of predicate cache IDs, as discussed in
			 * dtrace_predicate_create().
			 */
			nval = 1;
		}
	} while (dtrace_cas32(counter, oval, nval) != oval);
}

/*
 * Use the DTRACE_LOADFUNC macro to define functions for each of loading a
 * uint8_t, a uint16_t, a uint32_t and a uint64_t.
 */
/* BEGIN CSTYLED */
DTRACE_LOADFUNC(8)
DTRACE_LOADFUNC(16)
DTRACE_LOADFUNC(32)
DTRACE_LOADFUNC(64)
/* END CSTYLED */

static int
dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate)
{
	if (dest < mstate->dtms_scratch_base)
		return (0);

	if (dest + size < dest)
		return (0);

	if (dest + size > mstate->dtms_scratch_ptr)
		return (0);

	return (1);
}

static int
dtrace_canstore_statvar(uint64_t addr, size_t sz, size_t *remain,
    dtrace_statvar_t **svars, int nsvars)
{
	int i;
	size_t maxglobalsize, maxlocalsize;

	if (nsvars == 0)
		return (0);

	maxglobalsize = dtrace_statvar_maxsize + sizeof (uint64_t);
	maxlocalsize = maxglobalsize * NCPU;

	for (i = 0; i < nsvars; i++) {
		dtrace_statvar_t *svar = svars[i];
		uint8_t scope;
		size_t size;

		if (svar == NULL || (size = svar->dtsv_size) == 0)
			continue;

		scope = svar->dtsv_var.dtdv_scope;

		/*
		 * We verify that our size is valid in the spirit of providing
		 * defense in depth:  we want to prevent attackers from using
		 * DTrace to escalate an orthogonal kernel heap corruption bug
		 * into the ability to store to arbitrary locations in memory.
		 */
		VERIFY((scope == DIFV_SCOPE_GLOBAL && size <= maxglobalsize) ||
		    (scope == DIFV_SCOPE_LOCAL && size <= maxlocalsize));

		if (DTRACE_INRANGE(addr, sz, svar->dtsv_data,
		    svar->dtsv_size)) {
			DTRACE_RANGE_REMAIN(remain, addr, svar->dtsv_data,
			    svar->dtsv_size);
			return (1);
		}
	}

	return (0);
}

/*
 * Check to see if the address is within a memory region to which a store may
 * be issued.  This includes the DTrace scratch areas, and any DTrace variable
 * region.  The caller of dtrace_canstore() is responsible for performing any
 * alignment checks that are needed before stores are actually executed.
 */
static int
dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
    dtrace_vstate_t *vstate)
{
	return (dtrace_canstore_remains(addr, sz, NULL, mstate, vstate));
}

/*
 * Implementation of dtrace_canstore which communicates the upper bound of the
 * allowed memory region.
 */
static int
dtrace_canstore_remains(uint64_t addr, size_t sz, size_t *remain,
    dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
{
	/*
	 * First, check to see if the address is in scratch space...
	 */
	if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base,
	    mstate->dtms_scratch_size)) {
		DTRACE_RANGE_REMAIN(remain, addr, mstate->dtms_scratch_base,
		    mstate->dtms_scratch_size);
		return (1);
	}

	/*
	 * Now check to see if it's a dynamic variable.  This check will pick
	 * up both thread-local variables and any global dynamically-allocated
	 * variables.
	 */
	if (DTRACE_INRANGE(addr, sz, vstate->dtvs_dynvars.dtds_base,
	    vstate->dtvs_dynvars.dtds_size)) {
		dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
		uintptr_t base = (uintptr_t)dstate->dtds_base +
		    (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t));
		uintptr_t chunkoffs;
		dtrace_dynvar_t *dvar;

		/*
		 * Before we assume that we can store here, we need to make
		 * sure that it isn't in our metadata -- storing to our
		 * dynamic variable metadata would corrupt our state.  For
		 * the range to not include any dynamic variable metadata,
		 * it must:
		 *
		 *	(1) Start above the hash table that is at the base of
		 *	the dynamic variable space
		 *
		 *	(2) Have a starting chunk offset that is beyond the
		 *	dtrace_dynvar_t that is at the base of every chunk
		 *
		 *	(3) Not span a chunk boundary
		 *
		 *	(4) Not be in the tuple space of a dynamic variable
		 *
		 */
		if (addr < base)
			return (0);

		chunkoffs = (addr - base) % dstate->dtds_chunksize;

		if (chunkoffs < sizeof (dtrace_dynvar_t))
			return (0);

		if (chunkoffs + sz > dstate->dtds_chunksize)
			return (0);

		dvar = (dtrace_dynvar_t *)((uintptr_t)addr - chunkoffs);

		if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE)
			return (0);

		if (chunkoffs < sizeof (dtrace_dynvar_t) +
		    ((dvar->dtdv_tuple.dtt_nkeys - 1) * sizeof (dtrace_key_t)))
			return (0);

		DTRACE_RANGE_REMAIN(remain, addr, dvar, dstate->dtds_chunksize);
		return (1);
	}

	/*
	 * Finally, check the static local and global variables.  These checks
	 * take the longest, so we perform them last.
	 */
	if (dtrace_canstore_statvar(addr, sz, remain,
	    vstate->dtvs_locals, vstate->dtvs_nlocals))
		return (1);

	if (dtrace_canstore_statvar(addr, sz, remain,
	    vstate->dtvs_globals, vstate->dtvs_nglobals))
		return (1);

	return (0);
}


/*
 * Convenience routine to check to see if the address is within a memory
 * region in which a load may be issued given the user's privilege level;
 * if not, it sets the appropriate error flags and loads 'addr' into the
 * illegal value slot.
 *
 * DTrace subroutines (DIF_SUBR_*) should use this helper to implement
 * appropriate memory access protection.
 */
static int
dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
    dtrace_vstate_t *vstate)
{
	return (dtrace_canload_remains(addr, sz, NULL, mstate, vstate));
}

/*
 * Implementation of dtrace_canload which communicates the upper bound of the
 * allowed memory region.
 */
static int
dtrace_canload_remains(uint64_t addr, size_t sz, size_t *remain,
    dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
{
	volatile uintptr_t *illval = &cpu_core[CPU->cpu_id].cpuc_dtrace_illval;
	file_t *fp;

	/*
	 * If we hold the privilege to read from kernel memory, then
	 * everything is readable.
	 */
	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) {
		DTRACE_RANGE_REMAIN(remain, addr, addr, sz);
		return (1);
	}

	/*
	 * You can obviously read that which you can store.
	 */
	if (dtrace_canstore_remains(addr, sz, remain, mstate, vstate))
		return (1);

	/*
	 * We're allowed to read from our own string table.
	 */
	if (DTRACE_INRANGE(addr, sz, mstate->dtms_difo->dtdo_strtab,
	    mstate->dtms_difo->dtdo_strlen)) {
		DTRACE_RANGE_REMAIN(remain, addr,
		    mstate->dtms_difo->dtdo_strtab,
		    mstate->dtms_difo->dtdo_strlen);
		return (1);
	}

	if (vstate->dtvs_state != NULL &&
	    dtrace_priv_proc(vstate->dtvs_state, mstate)) {
		proc_t *p;

		/*
		 * When we have privileges to the current process, there are
		 * several context-related kernel structures that are safe to
		 * read, even absent the privilege to read from kernel memory.
		 * These reads are safe because these structures contain only
		 * state that (1) we're permitted to read, (2) is harmless or
		 * (3) contains pointers to additional kernel state that we're
		 * not permitted to read (and as such, do not present an
		 * opportunity for privilege escalation).  Finally (and
		 * critically), because of the nature of their relation with
		 * the current thread context, the memory associated with these
		 * structures cannot change over the duration of probe context,
		 * and it is therefore impossible for this memory to be
		 * deallocated and reallocated as something else while it's
		 * being operated upon.
		 */
		if (DTRACE_INRANGE(addr, sz, curthread, sizeof (kthread_t))) {
			DTRACE_RANGE_REMAIN(remain, addr, curthread,
			    sizeof (kthread_t));
			return (1);
		}

		if ((p = curthread->t_procp) != NULL && DTRACE_INRANGE(addr,
		    sz, curthread->t_procp, sizeof (proc_t))) {
			DTRACE_RANGE_REMAIN(remain, addr, curthread->t_procp,
			    sizeof (proc_t));
			return (1);
		}

		if (curthread->t_cred != NULL && DTRACE_INRANGE(addr, sz,
		    curthread->t_cred, sizeof (cred_t))) {
			DTRACE_RANGE_REMAIN(remain, addr, curthread->t_cred,
			    sizeof (cred_t));
			return (1);
		}

		if (p != NULL && p->p_pidp != NULL && DTRACE_INRANGE(addr, sz,
		    &(p->p_pidp->pid_id), sizeof (pid_t))) {
			DTRACE_RANGE_REMAIN(remain, addr, &(p->p_pidp->pid_id),
			    sizeof (pid_t));
			return (1);
		}

		if (curthread->t_cpu != NULL && DTRACE_INRANGE(addr, sz,
		    curthread->t_cpu, offsetof(cpu_t, cpu_pause_thread))) {
			DTRACE_RANGE_REMAIN(remain, addr, curthread->t_cpu,
			    offsetof(cpu_t, cpu_pause_thread));
			return (1);
		}
	}

	if ((fp = mstate->dtms_getf) != NULL) {
		uintptr_t psz = sizeof (void *);
		vnode_t *vp;
		vnodeops_t *op;

		/*
		 * When getf() returns a file_t, the enabling is implicitly
		 * granted the (transient) right to read the returned file_t
		 * as well as the v_path and v_op->vnop_name of the underlying
		 * vnode.  These accesses are allowed after a successful
		 * getf() because the members that they refer to cannot change
		 * once set -- and the barrier logic in the kernel's closef()
		 * path assures that the file_t and its referenced vode_t
		 * cannot themselves be stale (that is, it impossible for
		 * either dtms_getf itself or its f_vnode member to reference
		 * freed memory).
		 */
		if (DTRACE_INRANGE(addr, sz, fp, sizeof (file_t))) {
			DTRACE_RANGE_REMAIN(remain, addr, fp, sizeof (file_t));
			return (1);
		}

		if ((vp = fp->f_vnode) != NULL) {
			size_t slen;

			if (DTRACE_INRANGE(addr, sz, &vp->v_path, psz)) {
				DTRACE_RANGE_REMAIN(remain, addr, &vp->v_path,
				    psz);
				return (1);
			}

			slen = strlen(vp->v_path) + 1;
			if (DTRACE_INRANGE(addr, sz, vp->v_path, slen)) {
				DTRACE_RANGE_REMAIN(remain, addr, vp->v_path,
				    slen);
				return (1);
			}

			if (DTRACE_INRANGE(addr, sz, &vp->v_op, psz)) {
				DTRACE_RANGE_REMAIN(remain, addr, &vp->v_op,
				    psz);
				return (1);
			}

			if ((op = vp->v_op) != NULL &&
			    DTRACE_INRANGE(addr, sz, &op->vnop_name, psz)) {
				DTRACE_RANGE_REMAIN(remain, addr,
				    &op->vnop_name, psz);
				return (1);
			}

			if (op != NULL && op->vnop_name != NULL &&
			    DTRACE_INRANGE(addr, sz, op->vnop_name,
			    (slen = strlen(op->vnop_name) + 1))) {
				DTRACE_RANGE_REMAIN(remain, addr,
				    op->vnop_name, slen);
				return (1);
			}
		}
	}

	DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV);
	*illval = addr;
	return (0);
}

/*
 * Convenience routine to check to see if a given string is within a memory
 * region in which a load may be issued given the user's privilege level;
 * this exists so that we don't need to issue unnecessary dtrace_strlen()
 * calls in the event that the user has all privileges.
 */
static int
dtrace_strcanload(uint64_t addr, size_t sz, size_t *remain,
    dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
{
	size_t rsize;

	/*
	 * If we hold the privilege to read from kernel memory, then
	 * everything is readable.
	 */
	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) {
		DTRACE_RANGE_REMAIN(remain, addr, addr, sz);
		return (1);
	}

	/*
	 * Even if the caller is uninterested in querying the remaining valid
	 * range, it is required to ensure that the access is allowed.
	 */
	if (remain == NULL) {
		remain = &rsize;
	}
	if (dtrace_canload_remains(addr, 0, remain, mstate, vstate)) {
		size_t strsz;
		/*
		 * Perform the strlen after determining the length of the
		 * memory region which is accessible.  This prevents timing
		 * information from being used to find NULs in memory which is
		 * not accessible to the caller.
		 */
		strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr,
		    MIN(sz, *remain));
		if (strsz <= *remain) {
			return (1);
		}
	}

	return (0);
}

/*
 * Convenience routine to check to see if a given variable is within a memory
 * region in which a load may be issued given the user's privilege level.
 */
static int
dtrace_vcanload(void *src, dtrace_diftype_t *type, size_t *remain,
    dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
{
	size_t sz;
	ASSERT(type->dtdt_flags & DIF_TF_BYREF);

	/*
	 * Calculate the max size before performing any checks since even
	 * DTRACE_ACCESS_KERNEL-credentialed callers expect that this function
	 * return the max length via 'remain'.
	 */
	if (type->dtdt_kind == DIF_TYPE_STRING) {
		dtrace_state_t *state = vstate->dtvs_state;

		if (state != NULL) {
			sz = state->dts_options[DTRACEOPT_STRSIZE];
		} else {
			/*
			 * In helper context, we have a NULL state; fall back
			 * to using the system-wide default for the string size
			 * in this case.
			 */
			sz = dtrace_strsize_default;
		}
	} else {
		sz = type->dtdt_size;
	}

	/*
	 * If we hold the privilege to read from kernel memory, then
	 * everything is readable.
	 */
	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) {
		DTRACE_RANGE_REMAIN(remain, (uintptr_t)src, src, sz);
		return (1);
	}

	if (type->dtdt_kind == DIF_TYPE_STRING) {
		return (dtrace_strcanload((uintptr_t)src, sz, remain, mstate,
		    vstate));
	}
	return (dtrace_canload_remains((uintptr_t)src, sz, remain, mstate,
	    vstate));
}

/*
 * Convert a string to a signed integer using safe loads.
 *
 * NOTE: This function uses various macros from strtolctype.h to manipulate
 * digit values, etc -- these have all been checked to ensure they make
 * no additional function calls.
 */
static int64_t
dtrace_strtoll(char *input, int base, size_t limit)
{
	uintptr_t pos = (uintptr_t)input;
	int64_t val = 0;
	int x;
	boolean_t neg = B_FALSE;
	char c, cc, ccc;
	uintptr_t end = pos + limit;

	/*
	 * Consume any whitespace preceding digits.
	 */
	while ((c = dtrace_load8(pos)) == ' ' || c == '\t')
		pos++;

	/*
	 * Handle an explicit sign if one is present.
	 */
	if (c == '-' || c == '+') {
		if (c == '-')
			neg = B_TRUE;
		c = dtrace_load8(++pos);
	}

	/*
	 * Check for an explicit hexadecimal prefix ("0x" or "0X") and skip it
	 * if present.
	 */
	if (base == 16 && c == '0' && ((cc = dtrace_load8(pos + 1)) == 'x' ||
	    cc == 'X') && isxdigit(ccc = dtrace_load8(pos + 2))) {
		pos += 2;
		c = ccc;
	}

	/*
	 * Read in contiguous digits until the first non-digit character.
	 */
	for (; pos < end && c != '\0' && lisalnum(c) && (x = DIGIT(c)) < base;
	    c = dtrace_load8(++pos))
		val = val * base + x;

	return (neg ? -val : val);
}

/*
 * Compare two strings using safe loads.
 */
static int
dtrace_strncmp(char *s1, char *s2, size_t limit)
{
	uint8_t c1, c2;
	volatile uint16_t *flags;

	if (s1 == s2 || limit == 0)
		return (0);

	flags = (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags;

	do {
		if (s1 == NULL) {
			c1 = '\0';
		} else {
			c1 = dtrace_load8((uintptr_t)s1++);
		}

		if (s2 == NULL) {
			c2 = '\0';
		} else {
			c2 = dtrace_load8((uintptr_t)s2++);
		}

		if (c1 != c2)
			return (c1 - c2);
	} while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT));

	return (0);
}

/*
 * Compute strlen(s) for a string using safe memory accesses.  The additional
 * len parameter is used to specify a maximum length to ensure completion.
 */
static size_t
dtrace_strlen(const char *s, size_t lim)
{
	uint_t len;

	for (len = 0; len != lim; len++) {
		if (dtrace_load8((uintptr_t)s++) == '\0')
			break;
	}

	return (len);
}

/*
 * Check if an address falls within a toxic region.
 */
static int
dtrace_istoxic(uintptr_t kaddr, size_t size)
{
	uintptr_t taddr, tsize;
	int i;

	for (i = 0; i < dtrace_toxranges; i++) {
		taddr = dtrace_toxrange[i].dtt_base;
		tsize = dtrace_toxrange[i].dtt_limit - taddr;

		if (kaddr - taddr < tsize) {
			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
			cpu_core[CPU->cpu_id].cpuc_dtrace_illval = kaddr;
			return (1);
		}

		if (taddr - kaddr < size) {
			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
			cpu_core[CPU->cpu_id].cpuc_dtrace_illval = taddr;
			return (1);
		}
	}

	return (0);
}

/*
 * Copy src to dst using safe memory accesses.  The src is assumed to be unsafe
 * memory specified by the DIF program.  The dst is assumed to be safe memory
 * that we can store to directly because it is managed by DTrace.  As with
 * standard bcopy, overlapping copies are handled properly.
 */
static void
dtrace_bcopy(const void *src, void *dst, size_t len)
{
	if (len != 0) {
		uint8_t *s1 = dst;
		const uint8_t *s2 = src;

		if (s1 <= s2) {
			do {
				*s1++ = dtrace_load8((uintptr_t)s2++);
			} while (--len != 0);
		} else {
			s2 += len;
			s1 += len;

			do {
				*--s1 = dtrace_load8((uintptr_t)--s2);
			} while (--len != 0);
		}
	}
}

/*
 * Copy src to dst using safe memory accesses, up to either the specified
 * length, or the point that a nul byte is encountered.  The src is assumed to
 * be unsafe memory specified by the DIF program.  The dst is assumed to be
 * safe memory that we can store to directly because it is managed by DTrace.
 * Unlike dtrace_bcopy(), overlapping regions are not handled.
 */
static void
dtrace_strcpy(const void *src, void *dst, size_t len)
{
	if (len != 0) {
		uint8_t *s1 = dst, c;
		const uint8_t *s2 = src;

		do {
			*s1++ = c = dtrace_load8((uintptr_t)s2++);
		} while (--len != 0 && c != '\0');
	}
}

/*
 * Copy src to dst, deriving the size and type from the specified (BYREF)
 * variable type.  The src is assumed to be unsafe memory specified by the DIF
 * program.  The dst is assumed to be DTrace variable memory that is of the
 * specified type; we assume that we can store to directly.
 */
static void
dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type, size_t limit)
{
	ASSERT(type->dtdt_flags & DIF_TF_BYREF);

	if (type->dtdt_kind == DIF_TYPE_STRING) {
		dtrace_strcpy(src, dst, MIN(type->dtdt_size, limit));
	} else {
		dtrace_bcopy(src, dst, MIN(type->dtdt_size, limit));
	}
}

/*
 * Compare s1 to s2 using safe memory accesses.  The s1 data is assumed to be
 * unsafe memory specified by the DIF program.  The s2 data is assumed to be
 * safe memory that we can access directly because it is managed by DTrace.
 */
static int
dtrace_bcmp(const void *s1, const void *s2, size_t len)
{
	volatile uint16_t *flags;

	flags = (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags;

	if (s1 == s2)
		return (0);

	if (s1 == NULL || s2 == NULL)
		return (1);

	if (s1 != s2 && len != 0) {
		const uint8_t *ps1 = s1;
		const uint8_t *ps2 = s2;

		do {
			if (dtrace_load8((uintptr_t)ps1++) != *ps2++)
				return (1);
		} while (--len != 0 && !(*flags & CPU_DTRACE_FAULT));
	}
	return (0);
}

/*
 * Zero the specified region using a simple byte-by-byte loop.  Note that this
 * is for safe DTrace-managed memory only.
 */
static void
dtrace_bzero(void *dst, size_t len)
{
	uchar_t *cp;

	for (cp = dst; len != 0; len--)
		*cp++ = 0;
}

static void
dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum)
{
	uint64_t result[2];

	result[0] = addend1[0] + addend2[0];
	result[1] = addend1[1] + addend2[1] +
	    (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0);

	sum[0] = result[0];
	sum[1] = result[1];
}

/*
 * Shift the 128-bit value in a by b. If b is positive, shift left.
 * If b is negative, shift right.
 */
static void
dtrace_shift_128(uint64_t *a, int b)
{
	uint64_t mask;

	if (b == 0)
		return;

	if (b < 0) {
		b = -b;
		if (b >= 64) {
			a[0] = a[1] >> (b - 64);
			a[1] = 0;
		} else {
			a[0] >>= b;
			mask = 1LL << (64 - b);
			mask -= 1;
			a[0] |= ((a[1] & mask) << (64 - b));
			a[1] >>= b;
		}
	} else {
		if (b >= 64) {
			a[1] = a[0] << (b - 64);
			a[0] = 0;
		} else {
			a[1] <<= b;
			mask = a[0] >> (64 - b);
			a[1] |= mask;
			a[0] <<= b;
		}
	}
}

/*
 * The basic idea is to break the 2 64-bit values into 4 32-bit values,
 * use native multiplication on those, and then re-combine into the
 * resulting 128-bit value.
 *
 * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) =
 *     hi1 * hi2 << 64 +
 *     hi1 * lo2 << 32 +
 *     hi2 * lo1 << 32 +
 *     lo1 * lo2
 */
static void
dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product)
{
	uint64_t hi1, hi2, lo1, lo2;
	uint64_t tmp[2];

	hi1 = factor1 >> 32;
	hi2 = factor2 >> 32;

	lo1 = factor1 & DT_MASK_LO;
	lo2 = factor2 & DT_MASK_LO;

	product[0] = lo1 * lo2;
	product[1] = hi1 * hi2;

	tmp[0] = hi1 * lo2;
	tmp[1] = 0;
	dtrace_shift_128(tmp, 32);
	dtrace_add_128(product, tmp, product);

	tmp[0] = hi2 * lo1;
	tmp[1] = 0;
	dtrace_shift_128(tmp, 32);
	dtrace_add_128(product, tmp, product);
}

/*
 * This privilege check should be used by actions and subroutines to
 * verify that the user credentials of the process that enabled the
 * invoking ECB match the target credentials
 */
static int
dtrace_priv_proc_common_user(dtrace_state_t *state)
{
	cred_t *cr, *s_cr = state->dts_cred.dcr_cred;

	/*
	 * We should always have a non-NULL state cred here, since if cred
	 * is null (anonymous tracing), we fast-path bypass this routine.
	 */
	ASSERT(s_cr != NULL);

	if ((cr = CRED()) != NULL &&
	    s_cr->cr_uid == cr->cr_uid &&
	    s_cr->cr_uid == cr->cr_ruid &&
	    s_cr->cr_uid == cr->cr_suid &&
	    s_cr->cr_gid == cr->cr_gid &&
	    s_cr->cr_gid == cr->cr_rgid &&
	    s_cr->cr_gid == cr->cr_sgid)
		return (1);

	return (0);
}

/*
 * This privilege check should be used by actions and subroutines to
 * verify that the zone of the process that enabled the invoking ECB
 * matches the target credentials
 */
static int
dtrace_priv_proc_common_zone(dtrace_state_t *state)
{
	cred_t *cr, *s_cr = state->dts_cred.dcr_cred;

	/*
	 * We should always have a non-NULL state cred here, since if cred
	 * is null (anonymous tracing), we fast-path bypass this routine.
	 */
	ASSERT(s_cr != NULL);

	if ((cr = CRED()) != NULL && s_cr->cr_zone == cr->cr_zone)
		return (1);

	return (0);
}

/*
 * This privilege check should be used by actions and subroutines to
 * verify that the process has not setuid or changed credentials.
 */
static int
dtrace_priv_proc_common_nocd()
{
	proc_t *proc;

	if ((proc = ttoproc(curthread)) != NULL &&
	    !(proc->p_flag & SNOCD))
		return (1);

	return (0);
}

static int
dtrace_priv_proc_destructive(dtrace_state_t *state, dtrace_mstate_t *mstate)
{
	int action = state->dts_cred.dcr_action;

	if (!(mstate->dtms_access & DTRACE_ACCESS_PROC))
		goto bad;

	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) &&
	    dtrace_priv_proc_common_zone(state) == 0)
		goto bad;

	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) &&
	    dtrace_priv_proc_common_user(state) == 0)
		goto bad;

	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) &&
	    dtrace_priv_proc_common_nocd() == 0)
		goto bad;

	return (1);

bad:
	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;

	return (0);
}

static int
dtrace_priv_proc_control(dtrace_state_t *state, dtrace_mstate_t *mstate)
{
	if (mstate->dtms_access & DTRACE_ACCESS_PROC) {
		if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL)
			return (1);

		if (dtrace_priv_proc_common_zone(state) &&
		    dtrace_priv_proc_common_user(state) &&
		    dtrace_priv_proc_common_nocd())
			return (1);
	}

	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;

	return (0);
}

static int
dtrace_priv_proc(dtrace_state_t *state, dtrace_mstate_t *mstate)
{
	if ((mstate->dtms_access & DTRACE_ACCESS_PROC) &&
	    (state->dts_cred.dcr_action & DTRACE_CRA_PROC))
		return (1);

	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;

	return (0);
}

static int
dtrace_priv_kernel(dtrace_state_t *state)
{
	if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL)
		return (1);

	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;

	return (0);
}

static int
dtrace_priv_kernel_destructive(dtrace_state_t *state)
{
	if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE)
		return (1);

	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;

	return (0);
}

/*
 * Determine if the dte_cond of the specified ECB allows for processing of
 * the current probe to continue.  Note that this routine may allow continued
 * processing, but with access(es) stripped from the mstate's dtms_access
 * field.
 */
static int
dtrace_priv_probe(dtrace_state_t *state, dtrace_mstate_t *mstate,
    dtrace_ecb_t *ecb)
{
	dtrace_probe_t *probe = ecb->dte_probe;
	dtrace_provider_t *prov = probe->dtpr_provider;
	dtrace_pops_t *pops = &prov->dtpv_pops;
	int mode = DTRACE_MODE_NOPRIV_DROP;

	ASSERT(ecb->dte_cond);

	if (pops->dtps_mode != NULL) {
		mode = pops->dtps_mode(prov->dtpv_arg,
		    probe->dtpr_id, probe->dtpr_arg);

		ASSERT(mode & (DTRACE_MODE_USER | DTRACE_MODE_KERNEL));
		ASSERT(mode & (DTRACE_MODE_NOPRIV_RESTRICT |
		    DTRACE_MODE_NOPRIV_DROP));
	}

	/*
	 * If the dte_cond bits indicate that this consumer is only allowed to
	 * see user-mode firings of this probe, check that the probe was fired
	 * while in a user context.  If that's not the case, use the policy
	 * specified by the provider to determine if we drop the probe or
	 * merely restrict operation.
	 */
	if (ecb->dte_cond & DTRACE_COND_USERMODE) {
		ASSERT(mode != DTRACE_MODE_NOPRIV_DROP);

		if (!(mode & DTRACE_MODE_USER)) {
			if (mode & DTRACE_MODE_NOPRIV_DROP)
				return (0);

			mstate->dtms_access &= ~DTRACE_ACCESS_ARGS;
		}
	}

	/*
	 * This is more subtle than it looks. We have to be absolutely certain
	 * that CRED() isn't going to change out from under us so it's only
	 * legit to examine that structure if we're in constrained situations.
	 * Currently, the only times we'll this check is if a non-super-user
	 * has enabled the profile or syscall providers -- providers that
	 * allow visibility of all processes. For the profile case, the check
	 * above will ensure that we're examining a user context.
	 */
	if (ecb->dte_cond & DTRACE_COND_OWNER) {
		cred_t *cr;
		cred_t *s_cr = state->dts_cred.dcr_cred;
		proc_t *proc;

		ASSERT(s_cr != NULL);

		if ((cr = CRED()) == NULL ||
		    s_cr->cr_uid != cr->cr_uid ||
		    s_cr->cr_uid != cr->cr_ruid ||
		    s_cr->cr_uid != cr->cr_suid ||
		    s_cr->cr_gid != cr->cr_gid ||
		    s_cr->cr_gid != cr->cr_rgid ||
		    s_cr->cr_gid != cr->cr_sgid ||
		    (proc = ttoproc(curthread)) == NULL ||
		    (proc->p_flag & SNOCD)) {
			if (mode & DTRACE_MODE_NOPRIV_DROP)
				return (0);

			mstate->dtms_access &= ~DTRACE_ACCESS_PROC;
		}
	}

	/*
	 * If our dte_cond is set to DTRACE_COND_ZONEOWNER and we are not
	 * in our zone, check to see if our mode policy is to restrict rather
	 * than to drop; if to restrict, strip away both DTRACE_ACCESS_PROC
	 * and DTRACE_ACCESS_ARGS
	 */
	if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) {
		cred_t *cr;
		cred_t *s_cr = state->dts_cred.dcr_cred;

		ASSERT(s_cr != NULL);

		if ((cr = CRED()) == NULL ||
		    s_cr->cr_zone->zone_id != cr->cr_zone->zone_id) {
			if (mode & DTRACE_MODE_NOPRIV_DROP)
				return (0);

			mstate->dtms_access &=
			    ~(DTRACE_ACCESS_PROC | DTRACE_ACCESS_ARGS);
		}
	}

	/*
	 * By merits of being in this code path at all, we have limited
	 * privileges.  If the provider has indicated that limited privileges
	 * are to denote restricted operation, strip off the ability to access
	 * arguments.
	 */
	if (mode & DTRACE_MODE_LIMITEDPRIV_RESTRICT)
		mstate->dtms_access &= ~DTRACE_ACCESS_ARGS;

	return (1);
}

/*
 * Note:  not called from probe context.  This function is called
 * asynchronously (and at a regular interval) from outside of probe context to
 * clean the dirty dynamic variable lists on all CPUs.  Dynamic variable
 * cleaning is explained in detail in <sys/dtrace_impl.h>.
 */
void
dtrace_dynvar_clean(dtrace_dstate_t *dstate)
{
	dtrace_dynvar_t *dirty;
	dtrace_dstate_percpu_t *dcpu;
	dtrace_dynvar_t **rinsep;
	int i, j, work = 0;

	for (i = 0; i < NCPU; i++) {
		dcpu = &dstate->dtds_percpu[i];
		rinsep = &dcpu->dtdsc_rinsing;

		/*
		 * If the dirty list is NULL, there is no dirty work to do.
		 */
		if (dcpu->dtdsc_dirty == NULL)
			continue;

		if (dcpu->dtdsc_rinsing != NULL) {
			/*
			 * If the rinsing list is non-NULL, then it is because
			 * this CPU was selected to accept another CPU's
			 * dirty list -- and since that time, dirty buffers
			 * have accumulated.  This is a highly unlikely
			 * condition, but we choose to ignore the dirty
			 * buffers -- they'll be picked up a future cleanse.
			 */
			continue;
		}

		if (dcpu->dtdsc_clean != NULL) {
			/*
			 * If the clean list is non-NULL, then we're in a
			 * situation where a CPU has done deallocations (we
			 * have a non-NULL dirty list) but no allocations (we
			 * also have a non-NULL clean list).  We can't simply
			 * move the dirty list into the clean list on this
			 * CPU, yet we also don't want to allow this condition
			 * to persist, lest a short clean list prevent a
			 * massive dirty list from being cleaned (which in
			 * turn could lead to otherwise avoidable dynamic
			 * drops).  To deal with this, we look for some CPU
			 * with a NULL clean list, NULL dirty list, and NULL
			 * rinsing list -- and then we borrow this CPU to
			 * rinse our dirty list.
			 */
			for (j = 0; j < NCPU; j++) {
				dtrace_dstate_percpu_t *rinser;

				rinser = &dstate->dtds_percpu[j];

				if (rinser->dtdsc_rinsing != NULL)
					continue;

				if (rinser->dtdsc_dirty != NULL)
					continue;

				if (rinser->dtdsc_clean != NULL)
					continue;

				rinsep = &rinser->dtdsc_rinsing;
				break;
			}

			if (j == NCPU) {
				/*
				 * We were unable to find another CPU that
				 * could accept this dirty list -- we are
				 * therefore unable to clean it now.
				 */
				dtrace_dynvar_failclean++;
				continue;
			}
		}

		work = 1;

		/*
		 * Atomically move the dirty list aside.
		 */
		do {
			dirty = dcpu->dtdsc_dirty;

			/*
			 * Before we zap the dirty list, set the rinsing list.
			 * (This allows for a potential assertion in
			 * dtrace_dynvar():  if a free dynamic variable appears
			 * on a hash chain, either the dirty list or the
			 * rinsing list for some CPU must be non-NULL.)
			 */
			*rinsep = dirty;
			dtrace_membar_producer();
		} while (dtrace_casptr(&dcpu->dtdsc_dirty,
		    dirty, NULL) != dirty);
	}

	if (!work) {
		/*
		 * We have no work to do; we can simply return.
		 */
		return;
	}

	dtrace_sync();

	for (i = 0; i < NCPU; i++) {
		dcpu = &dstate->dtds_percpu[i];

		if (dcpu->dtdsc_rinsing == NULL)
			continue;

		/*
		 * We are now guaranteed that no hash chain contains a pointer
		 * into this dirty list; we can make it clean.
		 */
		ASSERT(dcpu->dtdsc_clean == NULL);
		dcpu->dtdsc_clean = dcpu->dtdsc_rinsing;
		dcpu->dtdsc_rinsing = NULL;
	}

	/*
	 * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make
	 * sure that all CPUs have seen all of the dtdsc_clean pointers.
	 * This prevents a race whereby a CPU incorrectly decides that
	 * the state should be something other than DTRACE_DSTATE_CLEAN
	 * after dtrace_dynvar_clean() has completed.
	 */
	dtrace_sync();

	dstate->dtds_state = DTRACE_DSTATE_CLEAN;
}

/*
 * Depending on the value of the op parameter, this function looks-up,
 * allocates or deallocates an arbitrarily-keyed dynamic variable.  If an
 * allocation is requested, this function will return a pointer to a
 * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no
 * variable can be allocated.  If NULL is returned, the appropriate counter
 * will be incremented.
 */
dtrace_dynvar_t *
dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys,
    dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op,
    dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
{
	uint64_t hashval = DTRACE_DYNHASH_VALID;
	dtrace_dynhash_t *hash = dstate->dtds_hash;
	dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL;
	processorid_t me = CPU->cpu_id, cpu = me;
	dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me];
	size_t bucket, ksize;
	size_t chunksize = dstate->dtds_chunksize;
	uintptr_t kdata, lock, nstate;
	uint_t i;

	ASSERT(nkeys != 0);

	/*
	 * Hash the key.  As with aggregations, we use Jenkins' "One-at-a-time"
	 * algorithm.  For the by-value portions, we perform the algorithm in
	 * 16-bit chunks (as opposed to 8-bit chunks).  This speeds things up a
	 * bit, and seems to have only a minute effect on distribution.  For
	 * the by-reference data, we perform "One-at-a-time" iterating (safely)
	 * over each referenced byte.  It's painful to do this, but it's much
	 * better than pathological hash distribution.  The efficacy of the
	 * hashing algorithm (and a comparison with other algorithms) may be
	 * found by running the ::dtrace_dynstat MDB dcmd.
	 */
	for (i = 0; i < nkeys; i++) {
		if (key[i].dttk_size == 0) {
			uint64_t val = key[i].dttk_value;

			hashval += (val >> 48) & 0xffff;
			hashval += (hashval << 10);
			hashval ^= (hashval >> 6);

			hashval += (val >> 32) & 0xffff;
			hashval += (hashval << 10);
			hashval ^= (hashval >> 6);

			hashval += (val >> 16) & 0xffff;
			hashval += (hashval << 10);
			hashval ^= (hashval >> 6);

			hashval += val & 0xffff;
			hashval += (hashval << 10);
			hashval ^= (hashval >> 6);
		} else {
			/*
			 * This is incredibly painful, but it beats the hell
			 * out of the alternative.
			 */
			uint64_t j, size = key[i].dttk_size;
			uintptr_t base = (uintptr_t)key[i].dttk_value;

			if (!dtrace_canload(base, size, mstate, vstate))
				break;

			for (j = 0; j < size; j++) {
				hashval += dtrace_load8(base + j);
				hashval += (hashval << 10);
				hashval ^= (hashval >> 6);
			}
		}
	}

	if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT))
		return (NULL);

	hashval += (hashval << 3);
	hashval ^= (hashval >> 11);
	hashval += (hashval << 15);

	/*
	 * There is a remote chance (ideally, 1 in 2^31) that our hashval
	 * comes out to be one of our two sentinel hash values.  If this
	 * actually happens, we set the hashval to be a value known to be a
	 * non-sentinel value.
	 */
	if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK)
		hashval = DTRACE_DYNHASH_VALID;

	/*
	 * Yes, it's painful to do a divide here.  If the cycle count becomes
	 * important here, tricks can be pulled to reduce it.  (However, it's
	 * critical that hash collisions be kept to an absolute minimum;
	 * they're much more painful than a divide.)  It's better to have a
	 * solution that generates few collisions and still keeps things
	 * relatively simple.
	 */
	bucket = hashval % dstate->dtds_hashsize;

	if (op == DTRACE_DYNVAR_DEALLOC) {
		volatile uintptr_t *lockp = &hash[bucket].dtdh_lock;

		for (;;) {
			while ((lock = *lockp) & 1)
				continue;

			if (dtrace_casptr((void *)lockp,
			    (void *)lock, (void *)(lock + 1)) == (void *)lock)
				break;
		}

		dtrace_membar_producer();
	}

top:
	prev = NULL;
	lock = hash[bucket].dtdh_lock;

	dtrace_membar_consumer();

	start = hash[bucket].dtdh_chain;
	ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK ||
	    start->dtdv_hashval != DTRACE_DYNHASH_FREE ||
	    op != DTRACE_DYNVAR_DEALLOC));

	for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) {
		dtrace_tuple_t *dtuple = &dvar->dtdv_tuple;
		dtrace_key_t *dkey = &dtuple->dtt_key[0];

		if (dvar->dtdv_hashval != hashval) {
			if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) {
				/*
				 * We've reached the sink, and therefore the
				 * end of the hash chain; we can kick out of
				 * the loop knowing that we have seen a valid
				 * snapshot of state.
				 */
				ASSERT(dvar->dtdv_next == NULL);
				ASSERT(dvar == &dtrace_dynhash_sink);
				break;
			}

			if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) {
				/*
				 * We've gone off the rails:  somewhere along
				 * the line, one of the members of this hash
				 * chain was deleted.  Note that we could also
				 * detect this by simply letting this loop run
				 * to completion, as we would eventually hit
				 * the end of the dirty list.  However, we
				 * want to avoid running the length of the
				 * dirty list unnecessarily (it might be quite
				 * long), so we catch this as early as
				 * possible by detecting the hash marker.  In
				 * this case, we simply set dvar to NULL and
				 * break; the conditional after the loop will
				 * send us back to top.
				 */
				dvar = NULL;
				break;
			}

			goto next;
		}

		if (dtuple->dtt_nkeys != nkeys)
			goto next;

		for (i = 0; i < nkeys; i++, dkey++) {
			if (dkey->dttk_size != key[i].dttk_size)
				goto next; /* size or type mismatch */

			if (dkey->dttk_size != 0) {
				if (dtrace_bcmp(
				    (void *)(uintptr_t)key[i].dttk_value,
				    (void *)(uintptr_t)dkey->dttk_value,
				    dkey->dttk_size))
					goto next;
			} else {
				if (dkey->dttk_value != key[i].dttk_value)
					goto next;
			}
		}

		if (op != DTRACE_DYNVAR_DEALLOC)
			return (dvar);

		ASSERT(dvar->dtdv_next == NULL ||
		    dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE);

		if (prev != NULL) {
			ASSERT(hash[bucket].dtdh_chain != dvar);
			ASSERT(start != dvar);
			ASSERT(prev->dtdv_next == dvar);
			prev->dtdv_next = dvar->dtdv_next;
		} else {
			if (dtrace_casptr(&hash[bucket].dtdh_chain,
			    start, dvar->dtdv_next) != start) {
				/*
				 * We have failed to atomically swing the
				 * hash table head pointer, presumably because
				 * of a conflicting allocation on another CPU.
				 * We need to reread the hash chain and try
				 * again.
				 */
				goto top;
			}
		}

		dtrace_membar_producer();

		/*
		 * Now set the hash value to indicate that it's free.
		 */
		ASSERT(hash[bucket].dtdh_chain != dvar);
		dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;

		dtrace_membar_producer();

		/*
		 * Set the next pointer to point at the dirty list, and
		 * atomically swing the dirty pointer to the newly freed dvar.
		 */
		do {
			next = dcpu->dtdsc_dirty;
			dvar->dtdv_next = next;
		} while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next);

		/*
		 * Finally, unlock this hash bucket.
		 */
		ASSERT(hash[bucket].dtdh_lock == lock);
		ASSERT(lock & 1);
		hash[bucket].dtdh_lock++;

		return (NULL);
next:
		prev = dvar;
		continue;
	}

	if (dvar == NULL) {
		/*
		 * If dvar is NULL, it is because we went off the rails:
		 * one of the elements that we traversed in the hash chain
		 * was deleted while we were traversing it.  In this case,
		 * we assert that we aren't doing a dealloc (deallocs lock
		 * the hash bucket to prevent themselves from racing with
		 * one another), and retry the hash chain traversal.
		 */
		ASSERT(op != DTRACE_DYNVAR_DEALLOC);
		goto top;
	}

	if (op != DTRACE_DYNVAR_ALLOC) {
		/*
		 * If we are not to allocate a new variable, we want to
		 * return NULL now.  Before we return, check that the value
		 * of the lock word hasn't changed.  If it has, we may have
		 * seen an inconsistent snapshot.
		 */
		if (op == DTRACE_DYNVAR_NOALLOC) {
			if (hash[bucket].dtdh_lock != lock)
				goto top;
		} else {
			ASSERT(op == DTRACE_DYNVAR_DEALLOC);
			ASSERT(hash[bucket].dtdh_lock == lock);
			ASSERT(lock & 1);
			hash[bucket].dtdh_lock++;
		}

		return (NULL);
	}

	/*
	 * We need to allocate a new dynamic variable.  The size we need is the
	 * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the
	 * size of any auxiliary key data (rounded up to 8-byte alignment) plus
	 * the size of any referred-to data (dsize).  We then round the final
	 * size up to the chunksize for allocation.
	 */
	for (ksize = 0, i = 0; i < nkeys; i++)
		ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));

	/*
	 * This should be pretty much impossible, but could happen if, say,
	 * strange DIF specified the tuple.  Ideally, this should be an
	 * assertion and not an error condition -- but that requires that the
	 * chunksize calculation in dtrace_difo_chunksize() be absolutely
	 * bullet-proof.  (That is, it must not be able to be fooled by
	 * malicious DIF.)  Given the lack of backwards branches in DIF,
	 * solving this would presumably not amount to solving the Halting
	 * Problem -- but it still seems awfully hard.
	 */
	if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) +
	    ksize + dsize > chunksize) {
		dcpu->dtdsc_drops++;
		return (NULL);
	}

	nstate = DTRACE_DSTATE_EMPTY;

	do {
retry:
		free = dcpu->dtdsc_free;

		if (free == NULL) {
			dtrace_dynvar_t *clean = dcpu->dtdsc_clean;
			void *rval;

			if (clean == NULL) {
				/*
				 * We're out of dynamic variable space on
				 * this CPU.  Unless we have tried all CPUs,
				 * we'll try to allocate from a different
				 * CPU.
				 */
				switch (dstate->dtds_state) {
				case DTRACE_DSTATE_CLEAN: {
					void *sp = &dstate->dtds_state;

					if (++cpu >= NCPU)
						cpu = 0;

					if (dcpu->dtdsc_dirty != NULL &&
					    nstate == DTRACE_DSTATE_EMPTY)
						nstate = DTRACE_DSTATE_DIRTY;

					if (dcpu->dtdsc_rinsing != NULL)
						nstate = DTRACE_DSTATE_RINSING;

					dcpu = &dstate->dtds_percpu[cpu];

					if (cpu != me)
						goto retry;

					(void) dtrace_cas32(sp,
					    DTRACE_DSTATE_CLEAN, nstate);

					/*
					 * To increment the correct bean
					 * counter, take another lap.
					 */
					goto retry;
				}

				case DTRACE_DSTATE_DIRTY:
					dcpu->dtdsc_dirty_drops++;
					break;

				case DTRACE_DSTATE_RINSING:
					dcpu->dtdsc_rinsing_drops++;
					break;

				case DTRACE_DSTATE_EMPTY:
					dcpu->dtdsc_drops++;
					break;
				}

				DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP);
				return (NULL);
			}

			/*
			 * The clean list appears to be non-empty.  We want to
			 * move the clean list to the free list; we start by
			 * moving the clean pointer aside.
			 */
			if (dtrace_casptr(&dcpu->dtdsc_clean,
			    clean, NULL) != clean) {
				/*
				 * We are in one of two situations:
				 *
				 *  (a)	The clean list was switched to the
				 *	free list by another CPU.
				 *
				 *  (b)	The clean list was added to by the
				 *	cleansing cyclic.
				 *
				 * In either of these situations, we can
				 * just reattempt the free list allocation.
				 */
				goto retry;
			}

			ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE);

			/*
			 * Now we'll move the clean list to our free list.
			 * It's impossible for this to fail:  the only way
			 * the free list can be updated is through this
			 * code path, and only one CPU can own the clean list.
			 * Thus, it would only be possible for this to fail if
			 * this code were racing with dtrace_dynvar_clean().
			 * (That is, if dtrace_dynvar_clean() updated the clean
			 * list, and we ended up racing to update the free
			 * list.)  This race is prevented by the dtrace_sync()
			 * in dtrace_dynvar_clean() -- which flushes the
			 * owners of the clean lists out before resetting
			 * the clean lists.
			 */
			dcpu = &dstate->dtds_percpu[me];
			rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean);
			ASSERT(rval == NULL);
			goto retry;
		}

		dvar = free;
		new_free = dvar->dtdv_next;
	} while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free);

	/*
	 * We have now allocated a new chunk.  We copy the tuple keys into the
	 * tuple array and copy any referenced key data into the data space
	 * following the tuple array.  As we do this, we relocate dttk_value
	 * in the final tuple to point to the key data address in the chunk.
	 */
	kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys];
	dvar->dtdv_data = (void *)(kdata + ksize);
	dvar->dtdv_tuple.dtt_nkeys = nkeys;

	for (i = 0; i < nkeys; i++) {
		dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i];
		size_t kesize = key[i].dttk_size;

		if (kesize != 0) {
			dtrace_bcopy(
			    (const void *)(uintptr_t)key[i].dttk_value,
			    (void *)kdata, kesize);
			dkey->dttk_value = kdata;
			kdata += P2ROUNDUP(kesize, sizeof (uint64_t));
		} else {
			dkey->dttk_value = key[i].dttk_value;
		}

		dkey->dttk_size = kesize;
	}

	ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE);
	dvar->dtdv_hashval = hashval;
	dvar->dtdv_next = start;

	if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start)
		return (dvar);

	/*
	 * The cas has failed.  Either another CPU is adding an element to
	 * this hash chain, or another CPU is deleting an element from this
	 * hash chain.  The simplest way to deal with both of these cases
	 * (though not necessarily the most efficient) is to free our
	 * allocated block and re-attempt it all.  Note that the free is
	 * to the dirty list and _not_ to the free list.  This is to prevent
	 * races with allocators, above.
	 */
	dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;

	dtrace_membar_producer();

	do {
		free = dcpu->dtdsc_dirty;
		dvar->dtdv_next = free;
	} while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free);

	goto top;
}

/*ARGSUSED*/
static void
dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg)
{
	if ((int64_t)nval < (int64_t)*oval)
		*oval = nval;
}

/*ARGSUSED*/
static void
dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg)
{
	if ((int64_t)nval > (int64_t)*oval)
		*oval = nval;
}

static void
dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr)
{
	int i, zero = DTRACE_QUANTIZE_ZEROBUCKET;
	int64_t val = (int64_t)nval;

	if (val < 0) {
		for (i = 0; i < zero; i++) {
			if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) {
				quanta[i] += incr;
				return;
			}
		}
	} else {
		for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) {
			if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) {
				quanta[i - 1] += incr;
				return;
			}
		}

		quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr;
		return;
	}

	ASSERT(0);
}

static void
dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr)
{
	uint64_t arg = *lquanta++;
	int32_t base = DTRACE_LQUANTIZE_BASE(arg);
	uint16_t step = DTRACE_LQUANTIZE_STEP(arg);
	uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg);
	int32_t val = (int32_t)nval, level;

	ASSERT(step != 0);
	ASSERT(levels != 0);

	if (val < base) {
		/*
		 * This is an underflow.
		 */
		lquanta[0] += incr;
		return;
	}

	level = (val - base) / step;

	if (level < levels) {
		lquanta[level + 1] += incr;
		return;
	}

	/*
	 * This is an overflow.
	 */
	lquanta[levels + 1] += incr;
}

static int
dtrace_aggregate_llquantize_bucket(uint16_t factor, uint16_t low,
    uint16_t high, uint16_t nsteps, int64_t value)
{
	int64_t this = 1, last, next;
	int base = 1, order;

	ASSERT(factor <= nsteps);
	ASSERT(nsteps % factor == 0);

	for (order = 0; order < low; order++)
		this *= factor;

	/*
	 * If our value is less than our factor taken to the power of the
	 * low order of magnitude, it goes into the zeroth bucket.
	 */
	if (value < (last = this))
		return (0);

	for (this *= factor; order <= high; order++) {
		int nbuckets = this > nsteps ? nsteps : this;

		if ((next = this * factor) < this) {
			/*
			 * We should not generally get log/linear quantizations
			 * with a high magnitude that allows 64-bits to
			 * overflow, but we nonetheless protect against this
			 * by explicitly checking for overflow, and clamping
			 * our value accordingly.
			 */
			value = this - 1;
		}

		if (value < this) {
			/*
			 * If our value lies within this order of magnitude,
			 * determine its position by taking the offset within
			 * the order of magnitude, dividing by the bucket
			 * width, and adding to our (accumulated) base.
			 */
			return (base + (value - last) / (this / nbuckets));
		}

		base += nbuckets - (nbuckets / factor);
		last = this;
		this = next;
	}

	/*
	 * Our value is greater than or equal to our factor taken to the
	 * power of one plus the high magnitude -- return the top bucket.
	 */
	return (base);
}

static void
dtrace_aggregate_llquantize(uint64_t *llquanta, uint64_t nval, uint64_t incr)
{
	uint64_t arg = *llquanta++;
	uint16_t factor = DTRACE_LLQUANTIZE_FACTOR(arg);
	uint16_t low = DTRACE_LLQUANTIZE_LOW(arg);
	uint16_t high = DTRACE_LLQUANTIZE_HIGH(arg);
	uint16_t nsteps = DTRACE_LLQUANTIZE_NSTEP(arg);

	llquanta[dtrace_aggregate_llquantize_bucket(factor,
	    low, high, nsteps, nval)] += incr;
}

/*ARGSUSED*/
static void
dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg)
{
	data[0]++;
	data[1] += nval;
}

/*ARGSUSED*/
static void
dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg)
{
	int64_t snval = (int64_t)nval;
	uint64_t tmp[2];

	data[0]++;
	data[1] += nval;

	/*
	 * What we want to say here is:
	 *
	 * data[2] += nval * nval;
	 *
	 * But given that nval is 64-bit, we could easily overflow, so
	 * we do this as 128-bit arithmetic.
	 */
	if (snval < 0)
		snval = -snval;

	dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp);
	dtrace_add_128(data + 2, tmp, data + 2);
}

/*ARGSUSED*/
static void
dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg)
{
	*oval = *oval + 1;
}

/*ARGSUSED*/
static void
dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg)
{
	*oval += nval;
}

/*
 * Aggregate given the tuple in the principal data buffer, and the aggregating
 * action denoted by the specified dtrace_aggregation_t.  The aggregation
 * buffer is specified as the buf parameter.  This routine does not return
 * failure; if there is no space in the aggregation buffer, the data will be
 * dropped, and a corresponding counter incremented.
 */
static void
dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf,
    intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg)
{
	dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec;
	uint32_t i, ndx, size, fsize;
	uint32_t align = sizeof (uint64_t) - 1;
	dtrace_aggbuffer_t *agb;
	dtrace_aggkey_t *key;
	uint32_t hashval = 0, limit, isstr;
	caddr_t tomax, data, kdata;
	dtrace_actkind_t action;
	dtrace_action_t *act;
	uintptr_t offs;

	if (buf == NULL)
		return;

	if (!agg->dtag_hasarg) {
		/*
		 * Currently, only quantize() and lquantize() take additional
		 * arguments, and they have the same semantics:  an increment
		 * value that defaults to 1 when not present.  If additional
		 * aggregating actions take arguments, the setting of the
		 * default argument value will presumably have to become more
		 * sophisticated...
		 */
		arg = 1;
	}

	action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION;
	size = rec->dtrd_offset - agg->dtag_base;
	fsize = size + rec->dtrd_size;

	ASSERT(dbuf->dtb_tomax != NULL);
	data = dbuf->dtb_tomax + offset + agg->dtag_base;

	if ((tomax = buf->dtb_tomax) == NULL) {
		dtrace_buffer_drop(buf);
		return;
	}

	/*
	 * The metastructure is always at the bottom of the buffer.
	 */
	agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size -
	    sizeof (dtrace_aggbuffer_t));

	if (buf->dtb_offset == 0) {
		/*
		 * We just kludge up approximately 1/8th of the size to be
		 * buckets.  If this guess ends up being routinely
		 * off-the-mark, we may need to dynamically readjust this
		 * based on past performance.
		 */
		uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t);

		if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) <
		    (uintptr_t)tomax || hashsize == 0) {
			/*
			 * We've been given a ludicrously small buffer;
			 * increment our drop count and leave.
			 */
			dtrace_buffer_drop(buf);
			return;
		}

		/*
		 * And now, a pathetic attempt to try to get a an odd (or
		 * perchance, a prime) hash size for better hash distribution.
		 */
		if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3))
			hashsize -= DTRACE_AGGHASHSIZE_SLEW;

		agb->dtagb_hashsize = hashsize;
		agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb -
		    agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *));
		agb->dtagb_free = (uintptr_t)agb->dtagb_hash;

		for (i = 0; i < agb->dtagb_hashsize; i++)
			agb->dtagb_hash[i] = NULL;
	}

	ASSERT(agg->dtag_first != NULL);
	ASSERT(agg->dtag_first->dta_intuple);

	/*
	 * Calculate the hash value based on the key.  Note that we _don't_
	 * include the aggid in the hashing (but we will store it as part of
	 * the key).  The hashing algorithm is Bob Jenkins' "One-at-a-time"
	 * algorithm: a simple, quick algorithm that has no known funnels, and
	 * gets good distribution in practice.  The efficacy of the hashing
	 * algorithm (and a comparison with other algorithms) may be found by
	 * running the ::dtrace_aggstat MDB dcmd.
	 */
	for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
		i = act->dta_rec.dtrd_offset - agg->dtag_base;
		limit = i + act->dta_rec.dtrd_size;
		ASSERT(limit <= size);
		isstr = DTRACEACT_ISSTRING(act);

		for (; i < limit; i++) {
			hashval += data[i];
			hashval += (hashval << 10);
			hashval ^= (hashval >> 6);

			if (isstr && data[i] == '\0')
				break;
		}
	}

	hashval += (hashval << 3);
	hashval ^= (hashval >> 11);
	hashval += (hashval << 15);

	/*
	 * Yes, the divide here is expensive -- but it's generally the least
	 * of the performance issues given the amount of data that we iterate
	 * over to compute hash values, compare data, etc.
	 */
	ndx = hashval % agb->dtagb_hashsize;

	for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) {
		ASSERT((caddr_t)key >= tomax);
		ASSERT((caddr_t)key < tomax + buf->dtb_size);

		if (hashval != key->dtak_hashval || key->dtak_size != size)
			continue;

		kdata = key->dtak_data;
		ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size);

		for (act = agg->dtag_first; act->dta_intuple;
		    act = act->dta_next) {
			i = act->dta_rec.dtrd_offset - agg->dtag_base;
			limit = i + act->dta_rec.dtrd_size;
			ASSERT(limit <= size);
			isstr = DTRACEACT_ISSTRING(act);

			for (; i < limit; i++) {
				if (kdata[i] != data[i])
					goto next;

				if (isstr && data[i] == '\0')
					break;
			}
		}

		if (action != key->dtak_action) {
			/*
			 * We are aggregating on the same value in the same
			 * aggregation with two different aggregating actions.
			 * (This should have been picked up in the compiler,
			 * so we may be dealing with errant or devious DIF.)
			 * This is an error condition; we indicate as much,
			 * and return.
			 */
			DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
			return;
		}

		/*
		 * This is a hit:  we need to apply the aggregator to
		 * the value at this key.
		 */
		agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg);
		return;
next:
		continue;
	}

	/*
	 * We didn't find it.  We need to allocate some zero-filled space,
	 * link it into the hash table appropriately, and apply the aggregator
	 * to the (zero-filled) value.
	 */
	offs = buf->dtb_offset;
	while (offs & (align - 1))
		offs += sizeof (uint32_t);

	/*
	 * If we don't have enough room to both allocate a new key _and_
	 * its associated data, increment the drop count and return.
	 */
	if ((uintptr_t)tomax + offs + fsize >
	    agb->dtagb_free - sizeof (dtrace_aggkey_t)) {
		dtrace_buffer_drop(buf);
		return;
	}

	/*CONSTCOND*/
	ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1)));
	key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t));
	agb->dtagb_free -= sizeof (dtrace_aggkey_t);

	key->dtak_data = kdata = tomax + offs;
	buf->dtb_offset = offs + fsize;

	/*
	 * Now copy the data across.
	 */
	*((dtrace_aggid_t *)kdata) = agg->dtag_id;

	for (i = sizeof (dtrace_aggid_t); i < size; i++)
		kdata[i] = data[i];

	/*
	 * Because strings are not zeroed out by default, we need to iterate
	 * looking for actions that store strings, and we need to explicitly
	 * pad these strings out with zeroes.
	 */
	for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
		int nul;

		if (!DTRACEACT_ISSTRING(act))
			continue;

		i = act->dta_rec.dtrd_offset - agg->dtag_base;
		limit = i + act->dta_rec.dtrd_size;
		ASSERT(limit <= size);

		for (nul = 0; i < limit; i++) {
			if (nul) {
				kdata[i] = '\0';
				continue;
			}

			if (data[i] != '\0')
				continue;

			nul = 1;
		}
	}

	for (i = size; i < fsize; i++)
		kdata[i] = 0;

	key->dtak_hashval = hashval;
	key->dtak_size = size;
	key->dtak_action = action;
	key->dtak_next = agb->dtagb_hash[ndx];
	agb->dtagb_hash[ndx] = key;

	/*
	 * Finally, apply the aggregator.
	 */
	*((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial;
	agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg);
}

/*
 * Given consumer state, this routine finds a speculation in the INACTIVE
 * state and transitions it into the ACTIVE state.  If there is no speculation
 * in the INACTIVE state, 0 is returned.  In this case, no error counter is
 * incremented -- it is up to the caller to take appropriate action.
 */
static int
dtrace_speculation(dtrace_state_t *state)
{
	int i = 0;
	dtrace_speculation_state_t current;
	uint32_t *stat = &state->dts_speculations_unavail, count;

	while (i < state->dts_nspeculations) {
		dtrace_speculation_t *spec = &state->dts_speculations[i];

		current = spec->dtsp_state;

		if (current != DTRACESPEC_INACTIVE) {
			if (current == DTRACESPEC_COMMITTINGMANY ||
			    current == DTRACESPEC_COMMITTING ||
			    current == DTRACESPEC_DISCARDING)
				stat = &state->dts_speculations_busy;
			i++;
			continue;
		}

		if (dtrace_cas32((uint32_t *)&spec->dtsp_state,
		    current, DTRACESPEC_ACTIVE) == current)
			return (i + 1);
	}

	/*
	 * We couldn't find a speculation.  If we found as much as a single
	 * busy speculation buffer, we'll attribute this failure as "busy"
	 * instead of "unavail".
	 */
	do {
		count = *stat;
	} while (dtrace_cas32(stat, count, count + 1) != count);

	return (0);
}

/*
 * This routine commits an active speculation.  If the specified speculation
 * is not in a valid state to perform a commit(), this routine will silently do
 * nothing.  The state of the specified speculation is transitioned according
 * to the state transition diagram outlined in <sys/dtrace_impl.h>
 */
static void
dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu,
    dtrace_specid_t which)
{
	dtrace_speculation_t *spec;
	dtrace_buffer_t *src, *dest;
	uintptr_t daddr, saddr, dlimit, slimit;
	dtrace_speculation_state_t current, new;
	intptr_t offs;
	uint64_t timestamp;

	if (which == 0)
		return;

	if (which > state->dts_nspeculations) {
		cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
		return;
	}

	spec = &state->dts_speculations[which - 1];
	src = &spec->dtsp_buffer[cpu];
	dest = &state->dts_buffer[cpu];

	do {
		current = spec->dtsp_state;

		if (current == DTRACESPEC_COMMITTINGMANY)
			break;

		switch (current) {
		case DTRACESPEC_INACTIVE:
		case DTRACESPEC_DISCARDING:
			return;

		case DTRACESPEC_COMMITTING:
			/*
			 * This is only possible if we are (a) commit()'ing
			 * without having done a prior speculate() on this CPU
			 * and (b) racing with another commit() on a different
			 * CPU.  There's nothing to do -- we just assert that
			 * our offset is 0.
			 */
			ASSERT(src->dtb_offset == 0);
			return;

		case DTRACESPEC_ACTIVE:
			new = DTRACESPEC_COMMITTING;
			break;

		case DTRACESPEC_ACTIVEONE:
			/*
			 * This speculation is active on one CPU.  If our
			 * buffer offset is non-zero, we know that the one CPU
			 * must be us.  Otherwise, we are committing on a
			 * different CPU from the speculate(), and we must
			 * rely on being asynchronously cleaned.
			 */
			if (src->dtb_offset != 0) {
				new = DTRACESPEC_COMMITTING;
				break;
			}
			/*FALLTHROUGH*/

		case DTRACESPEC_ACTIVEMANY:
			new = DTRACESPEC_COMMITTINGMANY;
			break;

		default:
			ASSERT(0);
		}
	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
	    current, new) != current);

	/*
	 * We have set the state to indicate that we are committing this
	 * speculation.  Now reserve the necessary space in the destination
	 * buffer.
	 */
	if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset,
	    sizeof (uint64_t), state, NULL)) < 0) {
		dtrace_buffer_drop(dest);
		goto out;
	}

	/*
	 * We have sufficient space to copy the speculative buffer into the
	 * primary buffer.  First, modify the speculative buffer, filling
	 * in the timestamp of all entries with the current time.  The data
	 * must have the commit() time rather than the time it was traced,
	 * so that all entries in the primary buffer are in timestamp order.
	 */
	timestamp = dtrace_gethrtime();
	saddr = (uintptr_t)src->dtb_tomax;
	slimit = saddr + src->dtb_offset;
	while (saddr < slimit) {
		size_t size;
		dtrace_rechdr_t *dtrh = (dtrace_rechdr_t *)saddr;

		if (dtrh->dtrh_epid == DTRACE_EPIDNONE) {
			saddr += sizeof (dtrace_epid_t);
			continue;
		}
		ASSERT3U(dtrh->dtrh_epid, <=, state->dts_necbs);
		size = state->dts_ecbs[dtrh->dtrh_epid - 1]->dte_size;

		ASSERT3U(saddr + size, <=, slimit);
		ASSERT3U(size, >=, sizeof (dtrace_rechdr_t));
		ASSERT3U(DTRACE_RECORD_LOAD_TIMESTAMP(dtrh), ==, UINT64_MAX);

		DTRACE_RECORD_STORE_TIMESTAMP(dtrh, timestamp);

		saddr += size;
	}

	/*
	 * Copy the buffer across.  (Note that this is a
	 * highly subobtimal bcopy(); in the unlikely event that this becomes
	 * a serious performance issue, a high-performance DTrace-specific
	 * bcopy() should obviously be invented.)
	 */
	daddr = (uintptr_t)dest->dtb_tomax + offs;
	dlimit = daddr + src->dtb_offset;
	saddr = (uintptr_t)src->dtb_tomax;

	/*
	 * First, the aligned portion.
	 */
	while (dlimit - daddr >= sizeof (uint64_t)) {
		*((uint64_t *)daddr) = *((uint64_t *)saddr);

		daddr += sizeof (uint64_t);
		saddr += sizeof (uint64_t);
	}

	/*
	 * Now any left-over bit...
	 */
	while (dlimit - daddr)
		*((uint8_t *)daddr++) = *((uint8_t *)saddr++);

	/*
	 * Finally, commit the reserved space in the destination buffer.
	 */
	dest->dtb_offset = offs + src->dtb_offset;

out:
	/*
	 * If we're lucky enough to be the only active CPU on this speculation
	 * buffer, we can just set the state back to DTRACESPEC_INACTIVE.
	 */
	if (current == DTRACESPEC_ACTIVE ||
	    (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) {
		uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state,
		    DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE);

		ASSERT(rval == DTRACESPEC_COMMITTING);
	}

	src->dtb_offset = 0;
	src->dtb_xamot_drops += src->dtb_drops;
	src->dtb_drops = 0;
}

/*
 * This routine discards an active speculation.  If the specified speculation
 * is not in a valid state to perform a discard(), this routine will silently
 * do nothing.  The state of the specified speculation is transitioned
 * according to the state transition diagram outlined in <sys/dtrace_impl.h>
 */
static void
dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu,
    dtrace_specid_t which)
{
	dtrace_speculation_t *spec;
	dtrace_speculation_state_t current, new;
	dtrace_buffer_t *buf;

	if (which == 0)
		return;

	if (which > state->dts_nspeculations) {
		cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
		return;
	}

	spec = &state->dts_speculations[which - 1];
	buf = &spec->dtsp_buffer[cpu];

	do {
		current = spec->dtsp_state;

		switch (current) {
		case DTRACESPEC_INACTIVE:
		case DTRACESPEC_COMMITTINGMANY:
		case DTRACESPEC_COMMITTING:
		case DTRACESPEC_DISCARDING:
			return;

		case DTRACESPEC_ACTIVE:
		case DTRACESPEC_ACTIVEMANY:
			new = DTRACESPEC_DISCARDING;
			break;

		case DTRACESPEC_ACTIVEONE:
			if (buf->dtb_offset != 0) {
				new = DTRACESPEC_INACTIVE;
			} else {
				new = DTRACESPEC_DISCARDING;
			}
			break;

		default:
			ASSERT(0);
		}
	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
	    current, new) != current);

	buf->dtb_offset = 0;
	buf->dtb_drops = 0;
}

/*
 * Note:  not called from probe context.  This function is called
 * asynchronously from cross call context to clean any speculations that are
 * in the COMMITTINGMANY or DISCARDING states.  These speculations may not be
 * transitioned back to the INACTIVE state until all CPUs have cleaned the
 * speculation.
 */
static void
dtrace_speculation_clean_here(dtrace_state_t *state)
{
	dtrace_icookie_t cookie;
	processorid_t cpu = CPU->cpu_id;
	dtrace_buffer_t *dest = &state->dts_buffer[cpu];
	dtrace_specid_t i;

	cookie = dtrace_interrupt_disable();

	if (dest->dtb_tomax == NULL) {
		dtrace_interrupt_enable(cookie);
		return;
	}

	for (i = 0; i < state->dts_nspeculations; i++) {
		dtrace_speculation_t *spec = &state->dts_speculations[i];
		dtrace_buffer_t *src = &spec->dtsp_buffer[cpu];

		if (src->dtb_tomax == NULL)
			continue;

		if (spec->dtsp_state == DTRACESPEC_DISCARDING) {
			src->dtb_offset = 0;
			continue;
		}

		if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
			continue;

		if (src->dtb_offset == 0)
			continue;

		dtrace_speculation_commit(state, cpu, i + 1);
	}

	dtrace_interrupt_enable(cookie);
}

/*
 * Note:  not called from probe context.  This function is called
 * asynchronously (and at a regular interval) to clean any speculations that
 * are in the COMMITTINGMANY or DISCARDING states.  If it discovers that there
 * is work to be done, it cross calls all CPUs to perform that work;
 * COMMITMANY and DISCARDING speculations may not be transitioned back to the
 * INACTIVE state until they have been cleaned by all CPUs.
 */
static void
dtrace_speculation_clean(dtrace_state_t *state)
{
	int work = 0, rv;
	dtrace_specid_t i;

	for (i = 0; i < state->dts_nspeculations; i++) {
		dtrace_speculation_t *spec = &state->dts_speculations[i];

		ASSERT(!spec->dtsp_cleaning);

		if (spec->dtsp_state != DTRACESPEC_DISCARDING &&
		    spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
			continue;

		work++;
		spec->dtsp_cleaning = 1;
	}

	if (!work)
		return;

	dtrace_xcall(DTRACE_CPUALL,
	    (dtrace_xcall_t)dtrace_speculation_clean_here, state);

	/*
	 * We now know that all CPUs have committed or discarded their
	 * speculation buffers, as appropriate.  We can now set the state
	 * to inactive.
	 */
	for (i = 0; i < state->dts_nspeculations; i++) {
		dtrace_speculation_t *spec = &state->dts_speculations[i];
		dtrace_speculation_state_t current, new;

		if (!spec->dtsp_cleaning)
			continue;

		current = spec->dtsp_state;
		ASSERT(current == DTRACESPEC_DISCARDING ||
		    current == DTRACESPEC_COMMITTINGMANY);

		new = DTRACESPEC_INACTIVE;

		rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new);
		ASSERT(rv == current);
		spec->dtsp_cleaning = 0;
	}
}

/*
 * Called as part of a speculate() to get the speculative buffer associated
 * with a given speculation.  Returns NULL if the specified speculation is not
 * in an ACTIVE state.  If the speculation is in the ACTIVEONE state -- and
 * the active CPU is not the specified CPU -- the speculation will be
 * atomically transitioned into the ACTIVEMANY state.
 */
static dtrace_buffer_t *
dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid,
    dtrace_specid_t which)
{
	dtrace_speculation_t *spec;
	dtrace_speculation_state_t current, new;
	dtrace_buffer_t *buf;

	if (which == 0)
		return (NULL);

	if (which > state->dts_nspeculations) {
		cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
		return (NULL);
	}

	spec = &state->dts_speculations[which - 1];
	buf = &spec->dtsp_buffer[cpuid];

	do {
		current = spec->dtsp_state;

		switch (current) {
		case DTRACESPEC_INACTIVE:
		case DTRACESPEC_COMMITTINGMANY:
		case DTRACESPEC_DISCARDING:
			return (NULL);

		case DTRACESPEC_COMMITTING:
			ASSERT(buf->dtb_offset == 0);
			return (NULL);

		case DTRACESPEC_ACTIVEONE:
			/*
			 * This speculation is currently active on one CPU.
			 * Check the offset in the buffer; if it's non-zero,
			 * that CPU must be us (and we leave the state alone).
			 * If it's zero, assume that we're starting on a new
			 * CPU -- and change the state to indicate that the
			 * speculation is active on more than one CPU.
			 */
			if (buf->dtb_offset != 0)
				return (buf);

			new = DTRACESPEC_ACTIVEMANY;
			break;

		case DTRACESPEC_ACTIVEMANY:
			return (buf);

		case DTRACESPEC_ACTIVE:
			new = DTRACESPEC_ACTIVEONE;
			break;

		default:
			ASSERT(0);
		}
	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
	    current, new) != current);

	ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY);
	return (buf);
}

/*
 * Return a string.  In the event that the user lacks the privilege to access
 * arbitrary kernel memory, we copy the string out to scratch memory so that we
 * don't fail access checking.
 *
 * dtrace_dif_variable() uses this routine as a helper for various
 * builtin values such as 'execname' and 'probefunc.'
 */
uintptr_t
dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state,
    dtrace_mstate_t *mstate)
{
	uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
	uintptr_t ret;
	size_t strsz;

	/*
	 * The easy case: this probe is allowed to read all of memory, so
	 * we can just return this as a vanilla pointer.
	 */
	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
		return (addr);

	/*
	 * This is the tougher case: we copy the string in question from
	 * kernel memory into scratch memory and return it that way: this
	 * ensures that we won't trip up when access checking tests the
	 * BYREF return value.
	 */
	strsz = dtrace_strlen((char *)addr, size) + 1;

	if (mstate->dtms_scratch_ptr + strsz >
	    mstate->dtms_scratch_base + mstate->dtms_scratch_size) {
		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
		return (0);
	}

	dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr,
	    strsz);
	ret = mstate->dtms_scratch_ptr;
	mstate->dtms_scratch_ptr += strsz;
	return (ret);
}

/*
 * This function implements the DIF emulator's variable lookups.  The emulator
 * passes a reserved variable identifier and optional built-in array index.
 */
static uint64_t
dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v,
    uint64_t ndx)
{
	/*
	 * If we're accessing one of the uncached arguments, we'll turn this
	 * into a reference in the args array.
	 */
	if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) {
		ndx = v - DIF_VAR_ARG0;
		v = DIF_VAR_ARGS;
	}

	switch (v) {
	case DIF_VAR_ARGS:
		if (!(mstate->dtms_access & DTRACE_ACCESS_ARGS)) {
			cpu_core[CPU->cpu_id].cpuc_dtrace_flags |=
			    CPU_DTRACE_KPRIV;
			return (0);
		}

		ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS);
		if (ndx >= sizeof (mstate->dtms_arg) /
		    sizeof (mstate->dtms_arg[0])) {
			int aframes = mstate->dtms_probe->dtpr_aframes + 2;
			dtrace_provider_t *pv;
			uint64_t val;

			pv = mstate->dtms_probe->dtpr_provider;
			if (pv->dtpv_pops.dtps_getargval != NULL)
				val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg,
				    mstate->dtms_probe->dtpr_id,
				    mstate->dtms_probe->dtpr_arg, ndx, aframes);
			else
				val = dtrace_getarg(ndx, aframes);

			/*
			 * This is regrettably required to keep the compiler
			 * from tail-optimizing the call to dtrace_getarg().
			 * The condition always evaluates to true, but the
			 * compiler has no way of figuring that out a priori.
			 * (None of this would be necessary if the compiler
			 * could be relied upon to _always_ tail-optimize
			 * the call to dtrace_getarg() -- but it can't.)
			 */
			if (mstate->dtms_probe != NULL)
				return (val);

			ASSERT(0);
		}

		return (mstate->dtms_arg[ndx]);

	case DIF_VAR_UREGS: {
		klwp_t *lwp;

		if (!dtrace_priv_proc(state, mstate))
			return (0);

		if ((lwp = curthread->t_lwp) == NULL) {
			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
			cpu_core[CPU->cpu_id].cpuc_dtrace_illval = 0;
			return (0);
		}

		return (dtrace_getreg(lwp->lwp_regs, ndx));
	}

	case DIF_VAR_VMREGS: {
		uint64_t rval;

		if (!dtrace_priv_kernel(state))
			return (0);

		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);

		rval = dtrace_getvmreg(ndx,
		    &cpu_core[CPU->cpu_id].cpuc_dtrace_flags);

		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);

		return (rval);
	}

	case DIF_VAR_CURTHREAD:
		if (!dtrace_priv_proc(state, mstate))
			return (0);
		return ((uint64_t)(uintptr_t)curthread);

	case DIF_VAR_TIMESTAMP:
		if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) {
			mstate->dtms_timestamp = dtrace_gethrtime();
			mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP;
		}
		return (mstate->dtms_timestamp);

	case DIF_VAR_VTIMESTAMP:
		ASSERT(dtrace_vtime_references != 0);
		return (curthread->t_dtrace_vtime);

	case DIF_VAR_WALLTIMESTAMP:
		if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) {
			mstate->dtms_walltimestamp = dtrace_gethrestime();
			mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP;
		}
		return (mstate->dtms_walltimestamp);

	case DIF_VAR_IPL:
		if (!dtrace_priv_kernel(state))
			return (0);
		if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) {
			mstate->dtms_ipl = dtrace_getipl();
			mstate->dtms_present |= DTRACE_MSTATE_IPL;
		}
		return (mstate->dtms_ipl);

	case DIF_VAR_EPID:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID);
		return (mstate->dtms_epid);

	case DIF_VAR_ID:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return (mstate->dtms_probe->dtpr_id);

	case DIF_VAR_STACKDEPTH:
		if (!dtrace_priv_kernel(state))
			return (0);
		if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) {
			int aframes = mstate->dtms_probe->dtpr_aframes + 2;

			mstate->dtms_stackdepth = dtrace_getstackdepth(aframes);
			mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH;
		}
		return (mstate->dtms_stackdepth);

	case DIF_VAR_USTACKDEPTH:
		if (!dtrace_priv_proc(state, mstate))
			return (0);
		if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) {
			/*
			 * See comment in DIF_VAR_PID.
			 */
			if (DTRACE_ANCHORED(mstate->dtms_probe) &&
			    CPU_ON_INTR(CPU)) {
				mstate->dtms_ustackdepth = 0;
			} else {
				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
				mstate->dtms_ustackdepth =
				    dtrace_getustackdepth();
				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
			}
			mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH;
		}
		return (mstate->dtms_ustackdepth);

	case DIF_VAR_CALLER:
		if (!dtrace_priv_kernel(state))
			return (0);
		if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) {
			int aframes = mstate->dtms_probe->dtpr_aframes + 2;

			if (!DTRACE_ANCHORED(mstate->dtms_probe)) {
				/*
				 * If this is an unanchored probe, we are
				 * required to go through the slow path:
				 * dtrace_caller() only guarantees correct
				 * results for anchored probes.
				 */
				pc_t caller[2];

				dtrace_getpcstack(caller, 2, aframes,
				    (uint32_t *)(uintptr_t)mstate->dtms_arg[0]);
				mstate->dtms_caller = caller[1];
			} else if ((mstate->dtms_caller =
			    dtrace_caller(aframes)) == -1) {
				/*
				 * We have failed to do this the quick way;
				 * we must resort to the slower approach of
				 * calling dtrace_getpcstack().
				 */
				pc_t caller;

				dtrace_getpcstack(&caller, 1, aframes, NULL);
				mstate->dtms_caller = caller;
			}

			mstate->dtms_present |= DTRACE_MSTATE_CALLER;
		}
		return (mstate->dtms_caller);

	case DIF_VAR_UCALLER:
		if (!dtrace_priv_proc(state, mstate))
			return (0);

		if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) {
			uint64_t ustack[3];

			/*
			 * dtrace_getupcstack() fills in the first uint64_t
			 * with the current PID.  The second uint64_t will
			 * be the program counter at user-level.  The third
			 * uint64_t will contain the caller, which is what
			 * we're after.
			 */
			ustack[2] = 0;
			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
			dtrace_getupcstack(ustack, 3);
			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
			mstate->dtms_ucaller = ustack[2];
			mstate->dtms_present |= DTRACE_MSTATE_UCALLER;
		}

		return (mstate->dtms_ucaller);

	case DIF_VAR_PROBEPROV:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return (dtrace_dif_varstr(
		    (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name,
		    state, mstate));

	case DIF_VAR_PROBEMOD:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return (dtrace_dif_varstr(
		    (uintptr_t)mstate->dtms_probe->dtpr_mod,
		    state, mstate));

	case DIF_VAR_PROBEFUNC:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return (dtrace_dif_varstr(
		    (uintptr_t)mstate->dtms_probe->dtpr_func,
		    state, mstate));

	case DIF_VAR_PROBENAME:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return (dtrace_dif_varstr(
		    (uintptr_t)mstate->dtms_probe->dtpr_name,
		    state, mstate));

	case DIF_VAR_PID:
		if (!dtrace_priv_proc(state, mstate))
			return (0);

		/*
		 * Note that we are assuming that an unanchored probe is
		 * always due to a high-level interrupt.  (And we're assuming
		 * that there is only a single high level interrupt.)
		 */
		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
			return (pid0.pid_id);

		/*
		 * It is always safe to dereference one's own t_procp pointer:
		 * it always points to a valid, allocated proc structure.
		 * Further, it is always safe to dereference the p_pidp member
		 * of one's own proc structure.  (These are truisms becuase
		 * threads and processes don't clean up their own state --
		 * they leave that task to whomever reaps them.)
		 */
		return ((uint64_t)curthread->t_procp->p_pidp->pid_id);

	case DIF_VAR_PPID:
		if (!dtrace_priv_proc(state, mstate))
			return (0);

		/*
		 * See comment in DIF_VAR_PID.
		 */
		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
			return (pid0.pid_id);

		/*
		 * It is always safe to dereference one's own t_procp pointer:
		 * it always points to a valid, allocated proc structure.
		 * (This is true because threads don't clean up their own
		 * state -- they leave that task to whomever reaps them.)
		 */
		return ((uint64_t)curthread->t_procp->p_ppid);

	case DIF_VAR_TID:
		/*
		 * See comment in DIF_VAR_PID.
		 */
		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
			return (0);

		return ((uint64_t)curthread->t_tid);

	case DIF_VAR_EXECNAME:
		if (!dtrace_priv_proc(state, mstate))
			return (0);

		/*
		 * See comment in DIF_VAR_PID.
		 */
		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
			return ((uint64_t)(uintptr_t)p0.p_user.u_comm);

		/*
		 * It is always safe to dereference one's own t_procp pointer:
		 * it always points to a valid, allocated proc structure.
		 * (This is true because threads don't clean up their own
		 * state -- they leave that task to whomever reaps them.)
		 */
		return (dtrace_dif_varstr(
		    (uintptr_t)curthread->t_procp->p_user.u_comm,
		    state, mstate));

	case DIF_VAR_ZONENAME:
		if (!dtrace_priv_proc(state, mstate))
			return (0);

		/*
		 * See comment in DIF_VAR_PID.
		 */
		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
			return ((uint64_t)(uintptr_t)p0.p_zone->zone_name);

		/*
		 * It is always safe to dereference one's own t_procp pointer:
		 * it always points to a valid, allocated proc structure.
		 * (This is true because threads don't clean up their own
		 * state -- they leave that task to whomever reaps them.)
		 */
		return (dtrace_dif_varstr(
		    (uintptr_t)curthread->t_procp->p_zone->zone_name,
		    state, mstate));

	case DIF_VAR_UID:
		if (!dtrace_priv_proc(state, mstate))
			return (0);

		/*
		 * See comment in DIF_VAR_PID.
		 */
		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
			return ((uint64_t)p0.p_cred->cr_uid);

		/*
		 * It is always safe to dereference one's own t_procp pointer:
		 * it always points to a valid, allocated proc structure.
		 * (This is true because threads don't clean up their own
		 * state -- they leave that task to whomever reaps them.)
		 *
		 * Additionally, it is safe to dereference one's own process
		 * credential, since this is never NULL after process birth.
		 */
		return ((uint64_t)curthread->t_procp->p_cred->cr_uid);

	case DIF_VAR_GID:
		if (!dtrace_priv_proc(state, mstate))
			return (0);

		/*
		 * See comment in DIF_VAR_PID.
		 */
		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
			return ((uint64_t)p0.p_cred->cr_gid);

		/*
		 * It is always safe to dereference one's own t_procp pointer:
		 * it always points to a valid, allocated proc structure.
		 * (This is true because threads don't clean up their own
		 * state -- they leave that task to whomever reaps them.)
		 *
		 * Additionally, it is safe to dereference one's own process
		 * credential, since this is never NULL after process birth.
		 */
		return ((uint64_t)curthread->t_procp->p_cred->cr_gid);

	case DIF_VAR_ERRNO: {
		klwp_t *lwp;
		if (!dtrace_priv_proc(state, mstate))
			return (0);

		/*
		 * See comment in DIF_VAR_PID.
		 */
		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
			return (0);

		/*
		 * It is always safe to dereference one's own t_lwp pointer in
		 * the event that this pointer is non-NULL.  (This is true
		 * because threads and lwps don't clean up their own state --
		 * they leave that task to whomever reaps them.)
		 */
		if ((lwp = curthread->t_lwp) == NULL)
			return (0);