This is the mail archive of the systemtap@sourceware.org mailing list for the systemtap project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH -tip 3/3] Add get_signal tracepoint


> Hmm, actually, trace_signal_send() doesn't record the return value.

Is that because it's called before the action really happens?
Is it important that it be called beforehand?  If it's called
afterwards, it's easy to pass the return value.

> So, what about trace_signal_overflow() for RT-signals and
> trace_signal_loss_info() for non-RT?

Really you can distinguish those just by looking at sig and info, so
perhaps a single tracepoint is enough.  I guess it really depends on what
filtering you would want and how inconvenient it is to have to apply that
filtering.  Having these two distinct tracepoints lets you trivially trace
only "silent information loss" without seeing the events where userland
gets full information (if applications are paying attention).

If you want to have a full suite of tracepoints where each one covers one
unambiguous corner of the semantics, then there are more than these just
for sending.  e.g. see below.

Thanks,
Roland


--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -838,8 +841,10 @@ static int __send_signal(int sig, struct
 
 	assert_spin_locked(&t->sighand->siglock);
 
-	if (!prepare_signal(sig, t, from_ancestor_ns))
+	if (!prepare_signal(sig, t, from_ancestor_ns)) {
+		trace_signal_generate_ignored(sig, group, info);
 		return 0;
+	}
 
 	pending = group ? &t->signal->shared_pending : &t->pending;
 	/*
@@ -847,8 +852,10 @@ static int __send_signal(int sig, struct
 	 * exactly one non-rt signal, so that we can get more
 	 * detailed information about the cause of the signal.
 	 */
-	if (legacy_queue(pending, sig))
+	if (legacy_queue(pending, sig)) {
+		trace_signal_generate_dropped_duplicate(sig, group, info);
 		return 0;
+	}
 	/*
 	 * fast-pathed signals for kernel-internal things like SIGSTOP
 	 * or SIGKILL.
@@ -896,12 +903,22 @@ static int __send_signal(int sig, struct
 			break;
 		}
 	} else if (!is_si_special(info)) {
-		if (sig >= SIGRTMIN && info->si_code != SI_USER)
-		/*
-		 * Queue overflow, abort.  We may abort if the signal was rt
-		 * and sent by user using something other than kill().
-		 */
+		if (sig >= SIGRTMIN && info->si_code != SI_USER) {
+			/*
+			 * Queue overflow, abort.  We may abort if the
+			 * signal was rt and sent by user using something
+			 * other than kill().
+			 */
+			trace_signal_generate_overflow_fail(sig, group, info);
 			return -EAGAIN;
+		} else {
+			/*
+			 * This is a silent loss of information.  We still
+			 * send the signal, but the *info bits are lost.
+			 */
+			trace_signal_generate_overflow_lose_info(sig, group,
+								 info);
+		}
 	}
 
 out_set:


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]