dev/dtrace/fasttrap.c

6  * You may not use this file except in compliance with the License.
13  * When distributing Covered Code, include this CDDL HEADER in each
87  * to a bit of scratch space reserved in the user thread's ulwp_t structure
93  * instruction set that writing the code to emulate these in the kernel is
120  * may be taken in functions invoked by the DTrace framework, but may not be
121  * held while calling functions in the DTrace framework.
153  * FASTTRAP_MAX_DEFAULT or the value for fasttrap-max-probes in the
201  * unique zone. In example, dtrace_probe_t is 72 bytes in size,
209  * fasttrap_probe_t's are variable in size. Some quick profiling has shown
337 	static volatile int in = 0;
338 	ASSERT(in == 0);
339 	in = 1;
423 	 * a second. This situation can occur in certain circumstances where
432 		/* The time value passed to dtrace_timeout is in nanos */
440 	in = 0;
459  * we have to remove all the instrumentation we had previously enabled in the
485 	 * because we're in fork().
521 	 * Free any ptss pages/entries in the child.
612 	 * on the generation in which this probe was last modified.
621 	 * this path in which case we'll have constructed the tracepoint we'd
649 		 * to worry about another thread being in the midst of
702 		 * Activate the tracepoint in the ISA-specific manner.
711 		 * Increment the count of the number of tracepoints active in
795 	 * If we somehow lost this tracepoint, we're in a world of hurt.
838 		 * If the current probe's tracepoint is in use, swap it
871 		 * Tag the modified probe with the generation in which it was
884 	 * tracepoint secure in the knowledge that no other thread is going to
897 		 * this process since it's in a completely unrecoverable
905 		 * in the victim process.
928 	 * Tag the modified probe with the generation in which it was changed.
1014 	 * If this probe's provider is retired (meaning it was valid in a
1016 	 * provider can't go away while we're in this code path.
1022 	 * If we can't find the process, it may be that we're in the context of
1023 	 * a fork in which the traced process is being born and we're copying
1043 		 * Confirm that curproc is indeed forking the process in which
1061 	 * time a probe has been enabled in this process, we need to allocate scratch
1075 	 * in their process's text.
1483 	 * of a vfork(2), and isn't a zombie (but may be in fork).
1514 	// Seems like OS X has no equivalent to crhold, even though it has a cr_ref field in ucred
1561 	 * the full provider name -- we keep it in pieces in the provider
1627 	 * corresponds to this process's hash chain in the provider hash
1673 	 * Mark the provider to be removed in our post-processing step, mark it
1683 	 * provider lock in callbacks invoked from the DTrace framework. The
1785 	 * dtrace_probe_lookup() and dtrace_probe_create() in the face of
1958 	 * expressed in 10 or fewer decimal digits. Make sure that we'll
2088 	 * dtrace_probe_lookup() and dtrace_probe_create() in the face of
2095 	 * APPLE NOTE: This is hideously expensive. See note in
2627 		 * fasttrap_probe_t's are variable in size. We use an array of zones to