Bug 14369 - Markers with $vars are reported "not found" when missing debuginfo
Summary: Markers with $vars are reported "not found" when missing debuginfo
Status: RESOLVED FIXED
Alias: None
Product: systemtap
Classification: Unclassified
Component: translator (show other bugs)
Version: unspecified
: P2 normal
Target Milestone: ---
Assignee: Unassigned
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2012-07-18 02:52 UTC by Josh Stone
Modified: 2012-08-18 01:53 UTC (History)
1 user (show)

See Also:
Host:
Target:
Build:
Last reconfirmed:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Josh Stone 2012-07-18 02:52:27 UTC
This is related to: https://bugzilla.redhat.com/show_bug.cgi?id=841057

As noted there: When we locate an SDT marker, stap takes different paths depending on the presence of $vars in the handler.  Without any $var, we just build the probe directly, and life is good.  With a $var, we decide we need debuginfo, and so that goes down the query_addr path.

Now, if there's anything wrong with the debuginfo, then we end up throwing up entirely with "semantic error: no match".  An unreachable $var is a problem, sure, but we should use the right error.  And this also means that @defined() can't work around this issue.

It doesn't even have to be corrupt debuginfo, as rhbz841057.  Even our sdt_asm test binary can show this, when you don't have DWARF in part of a file.

First, with no DWARF at all:
> $ gcc ../testsuite/systemtap.base/sdt_asm.S -o sdt_asm.x
> $ ./run-stap -e 'probe process("./sdt_asm.x").mark("*") { println($foo) }' -p2 
> WARNING: cannot find module /home/jistone/systemtap/build/sdt_asm.x debuginfo: No DWARF information found
> semantic error: while resolving probe point: identifier 'process' at <input>:1:7
>         source: probe process("./sdt_asm.x").mark("*") { println($foo) }
>                       ^
> 
> semantic error: no match
> Pass 2: analysis failed.  Try again with another '--vp 01' option.

With some DWARF added (but nothing meaningful near asm probes):
> $ gcc ../testsuite/systemtap.base/sdt_asm.S -o sdt_asm.x -g
> $ ./run-stap -e 'probe process("./sdt_asm.x").mark("*") { println($foo) }' -p2 
> semantic error: while resolving probe point: identifier 'process' at <input>:1:7
>         source: probe process("./sdt_asm.x").mark("*") { println($foo) }
>                       ^
> 
> semantic error: no match
> Pass 2: analysis failed.  Try again with another '--vp 01' option.

And just to sanity check that the probe points do exist:
> $ ./run-stap -e 'probe process("./sdt_asm.x").mark("*") { println(1) }' -p2 
> # probes
> process("/home/jistone/systemtap/build/sdt_asm.x").statement(0x4004ab) /* pc=.absolute+0x4ab */ /* <- process("./sdt_asm.x").mark("memory") = process("./sdt_asm.x").statement(0x4004ab) <- process("./sdt_asm.x").mark("*") */
> process("/home/jistone/systemtap/build/sdt_asm.x").statement(0x4004b1) /* pc=.absolute+0x4b1 */ /* <- process("./sdt_asm.x").mark("a") = process("./sdt_asm.x").statement(0x4004b1) <- process("./sdt_asm.x").mark("*") */
[...]

We ought to fall back on this direct method when the query_addr path fails, and then the $foo can fail later (if it even survives optimization).
Comment 1 Josh Stone 2012-07-18 22:21:47 UTC
(In reply to comment #0)
> We ought to fall back on this direct method when the query_addr path fails, and
> then the $foo can fail later (if it even survives optimization).

commit 73d53dd2213ce54a72db9b3bd029eb1f39760656

With this, we now get:

> $ gcc ../testsuite/systemtap.base/sdt_asm.S -o sdt_asm.x 
> $ ./run-stap -e 'probe process("./sdt_asm.x").mark("*") { println($foo) }' -p2
> WARNING: cannot find module /home/jistone/systemtap/build/sdt_asm.x debuginfo: No DWARF information found
> semantic error: target-symbol requires debuginfo: identifier '$foo' at <input>:1:50
>         source: probe process("./sdt_asm.x").mark("*") { println($foo) }
>                                                                  ^
> 
> Pass 2: analysis failed.  Try again with another '--vp 01' option.

> $ gcc ../testsuite/systemtap.base/sdt_asm.S -o sdt_asm.x -g
> $ ./run-stap -e 'probe process("./sdt_asm.x").mark("*") { println($foo) }' -p2
> semantic error: target-symbol requires debuginfo: identifier '$foo' at <input>:1:50
>         source: probe process("./sdt_asm.x").mark("*") { println($foo) }
>                                                                  ^
> 
> Pass 2: analysis failed.  Try again with another '--vp 01' option.

> $ ./run-stap -e 'probe process("./sdt_asm.x").mark("*") { println(@defined($foo)) }' -p2
> # probes
> process("/home/jistone/systemtap/build/sdt_asm.x").statement(0x4004ab) /* pc=.absolute+0x4ab */ /* <- process("./sdt_asm.x").mark("memory") = process("./sdt_asm.x").statement(0x4004ab) <- process("./sdt_asm.x").mark("*") */
> process("/home/jistone/systemtap/build/sdt_asm.x").statement(0x4004b1) /* pc=.absolute+0x4b1 */ /* <- process("./sdt_asm.x").mark("a") = process("./sdt_asm.x").statement(0x4004b1) <- process("./sdt_asm.x").mark("*") */


I'm not entirely thrilled with how the new error gets chained though, when full debuginfo is present:

> $ ./run-stap -e 'probe process("./stap").mark("*") { println($foo) }' -p2
> semantic error: unable to find local 'foo' near pc 0x41140f in cleanup ../main.cxx ( (alternatives: $s $rc): identifier '$foo' at <input>:1:45
>         source: probe process("./stap").mark("*") { println($foo) }
>                                                             ^
> 
> semantic error: target-symbol requires debuginfo: identifier '$foo' at :1:45
>         source: probe process("./stap").mark("*") { println($foo) }
>                                                             ^
> 
> Pass 2: analysis failed.  Try again with another '--vp 01' option.

Ideas on how to improve that are welcome...
Comment 2 Josh Stone 2012-08-18 01:53:04 UTC
(In reply to comment #1)
> I'm not entirely thrilled with how the new error gets chained though, when full
> debuginfo is present: [...] Ideas on how to improve that are welcome...

fche declared it Good Enough, so I'm closing this bug.