dtrace.c

Sun Solaris 10 中的 DTrace 组件的源代码。请参看: http://www.sun.com/software/solaris/observability.jsp

		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);
}

/*
 * 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 i)
{
	/*
	 * 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) {
		i = v - DIF_VAR_ARG0;
		v = DIF_VAR_ARGS;
	}

	switch (v) {
	case DIF_VAR_ARGS:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS);
		if (i >= 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, i, aframes);
			else
				val = dtrace_getarg(i, 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[i]);

	case DIF_VAR_UREGS: {
		klwp_t *lwp;

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

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

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

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

			if ((mstate->dtms_caller =
			    dtrace_caller(aframes)) == -1) {
				/*
				 * We have failed to do this the quick way;
				 * we must resort to the slow and painful
				 * 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_PROBEPROV:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return ((uint64_t)(uintptr_t)
		    mstate->dtms_probe->dtpr_provider->dtpv_name);

	case DIF_VAR_PROBEMOD:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return ((uint64_t)(uintptr_t)
		    mstate->dtms_probe->dtpr_mod);

	case DIF_VAR_PROBEFUNC:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return ((uint64_t)(uintptr_t)
		    mstate->dtms_probe->dtpr_func);

	case DIF_VAR_PROBENAME:
		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
		return ((uint64_t)(uintptr_t)
		    mstate->dtms_probe->dtpr_name);

	case DIF_VAR_PID:
		if (!dtrace_priv_proc(state))
			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_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))
			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 ((uint64_t)(uintptr_t)
		    curthread->t_procp->p_user.u_comm);

	case DIF_VAR_ZONENAME:
		if (!dtrace_priv_proc(state))
			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 ((uint64_t)(uintptr_t)
		    curthread->t_procp->p_zone->zone_name);

	default:
		DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
		return (0);
	}
}

/*
 * Emulate the execution of DTrace ID subroutines invoked by the call opcode.
 * Notice that we don't bother validating the proper number of arguments or
 * their types in the tuple stack.  This isn't needed because all argument
 * interpretation is safe because of our load safety -- the worst that can
 * happen is that a bogus program can obtain bogus results.
 */
static void
dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs,
    dtrace_key_t *tupregs, dtrace_mstate_t *mstate, dtrace_state_t *state)
{
	volatile uint16_t *flags = &cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
	volatile uintptr_t *illval = &cpu_core[CPU->cpu_id].cpuc_dtrace_illval;

	union {
		mutex_impl_t mi;
		uint64_t mx;
	} m;

	union {
		krwlock_t ri;
		uintptr_t rw;
	} r;

	switch (subr) {
	case DIF_SUBR_RAND:
		regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875;
		break;

	case DIF_SUBR_MUTEX_OWNED:
		m.mx = dtrace_load64(tupregs[0].dttk_value);
		if (MUTEX_TYPE_ADAPTIVE(&m.mi))
			regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER;
		else
			regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock);
		break;

	case DIF_SUBR_MUTEX_OWNER:
		m.mx = dtrace_load64(tupregs[0].dttk_value);
		if (MUTEX_TYPE_ADAPTIVE(&m.mi) &&
		    MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER)
			regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi);
		else
			regs[rd] = 0;
		break;

	case DIF_SUBR_MUTEX_TYPE_ADAPTIVE:
		m.mx = dtrace_load64(tupregs[0].dttk_value);
		regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi);
		break;

	case DIF_SUBR_MUTEX_TYPE_SPIN:
		m.mx = dtrace_load64(tupregs[0].dttk_value);
		regs[rd] = MUTEX_TYPE_SPIN(&m.mi);
		break;

	case DIF_SUBR_RW_READ_HELD: {
		uintptr_t tmp;

		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
		regs[rd] = _RW_READ_HELD(&r.ri, tmp);
		break;
	}

	case DIF_SUBR_RW_WRITE_HELD:
		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
		regs[rd] = _RW_WRITE_HELD(&r.ri);
		break;

	case DIF_SUBR_RW_ISWRITER:
		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
		regs[rd] = _RW_ISWRITER(&r.ri);
		break;

	case DIF_SUBR_BCOPY: {
		/*
		 * We need to be sure that the destination is in the scratch
		 * region -- no other region is allowed.
		 */
		uintptr_t src = tupregs[0].dttk_value;
		uintptr_t dest = tupregs[1].dttk_value;
		size_t size = tupregs[2].dttk_value;

		if (!dtrace_inscratch(dest, size, mstate)) {
			*flags |= CPU_DTRACE_BADADDR;
			*illval = regs[rd];
			break;
		}

		dtrace_bcopy((void *)src, (void *)dest, size);
		break;
	}

	case DIF_SUBR_ALLOCA:
	case DIF_SUBR_COPYIN: {
		uintptr_t dest = P2ROUNDUP(mstate->dtms