[Bug translator/20149] New: a function probe with a line number acts like a statement probe
jistone at redhat dot com
sourceware-bugzilla@sourceware.org
Thu May 26 18:13:00 GMT 2016
https://sourceware.org/bugzilla/show_bug.cgi?id=20149
Bug ID: 20149
Summary: a function probe with a line number acts like a
statement probe
Product: systemtap
Version: unspecified
Status: NEW
Severity: normal
Priority: P2
Component: translator
Assignee: systemtap at sourceware dot org
Reporter: jistone at redhat dot com
Target Milestone: ---
If you probe a function with a line number different than its entry, you'll get
a warning that a statement probe was probably what you wanted. Then it
*should* act like a function-entry probe anyway, but instead it proceeds like a
statement.
First a plain function probe:
$ stap -p2 -e 'probe process("stap").function("main@main.cxx") {next}'
# probes
process("/usr/local/bin/stap").function("main@../main.cxx:1123") /*
pc=.absolute+0xfc80 */ /* <-
process("/usr/local/bin/stap").function("main@../main.cxx:1123") */
Now change the line number:
$ stap -p2 -e 'probe process("stap").function("main@main.cxx:1223") {next}'
WARNING: For probing a particular line, use a .statement() probe, not
.function(): keyword at <input>:1:1
source: probe process("stap").function("main@main.cxx:1223") {next}
^
# probes
process("/usr/local/bin/stap").function("main@../main.cxx:1223") /*
pc=.absolute+0x1056e */ /* <-
process("/usr/local/bin/stap").function("main@../main.cxx:1223") */
So we got a warning, but the pc is clearly in the middle of the function
anyway, like a statement probe. We want function probes to always use the
entry!
This happens even if you explicitly write ".call", and even worse it happens
for ".return"! Return probes play tricks with the architectural return
address, which will probably not work correctly at all if you're not at the
beginning of the function. (e.g. the stack pointer will often move later)
--
You are receiving this mail because:
You are the assignee for the bug.
More information about the Systemtap
mailing list