public inbox for systemtap@sourceware.org
 help / color / mirror / Atom feed
* [Bug uprobes/18171] New: setjmp/longjmp clashing with uprobes, causing SIGSEGV or SIGILL
@ 2015-03-28  0:27 fche at redhat dot com
  2015-03-28  0:31 ` [Bug uprobes/18171] setjmp/longjmp clashing with u[ret]probes, " fche at redhat dot com
                   ` (6 more replies)
  0 siblings, 7 replies; 8+ messages in thread
From: fche at redhat dot com @ 2015-03-28  0:27 UTC (permalink / raw)
  To: systemtap

https://sourceware.org/bugzilla/show_bug.cgi?id=18171

            Bug ID: 18171
           Summary: setjmp/longjmp clashing with uprobes, causing SIGSEGV
                    or SIGILL
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: uprobes
          Assignee: systemtap at sourceware dot org
          Reporter: fche at redhat dot com

The testsuite/systemtap.base/bz5274* test case now regularly fails on modern
kernels such as 3.18.9-200.fc21.x86_64, on at least i686 & x86-64.

% gcc -g -o bz5274 .../testsuite/systemtap.base/bz5274.c
% cat > bz5274a.stp << EOF
probe never
# ,process("bz5274").function("funca")
  ,process("bz5274").function("funca").return
# ,process("bz5274").function("funcb")
  ,process("bz5274").function("funcb").return
  ,process("bz5274").function("funcc")
# ,process("bz5274").function("funcc").return
  ,process("bz5274").function("funcd")
# ,process("bz5274").function("funcd").return
# ,process("bz5274").function("main")
  ,process("bz5274").function("main").return  # <-- out=SIGBUS vs in=SIGILL
{i++ }
global i
EOF
% sudo stap -t bz5274a.stp -c ./bz5274
In bz5274.c: funcb :20 : i=1.  Calling funcc
In bz5274.c: funcc :13 : i=1.  Calling funcd
In bz5274.c: funcd :7 : i=1.  Calling longjmp
In bz5274.c: funcb :20 : i=2.  Calling funcc
In bz5274.c: funcc :13 : i=2.  Calling funcd
In bz5274.c: funcd :7 : i=2.  Calling longjmp
In bz5274.c: funcb :20 : i=3.  Calling funcc
In bz5274.c: funcc :13 : i=3.  Calling funcd
In bz5274.c: funcd :7 : i=3.  Calling longjmp
In bz5274.c: funcb :20 : i=4.  Calling funcc
In bz5274.c: funcc :13 : i=4.  Calling funcd
In bz5274.c: funcd :7 : i=4.  Calling longjmp
WARNING: Child process exited with signal 11 (Segmentation fault)
i=0xa
----- probe hit report: 
process(".../testsuite/bz5274").function("funcb@.../systemtap.base/bz5274.c:18").return,
(bz5274a.stp:1:1), hits: 2 [...]
process(".../testsuite/bz5274").function("funcc@.../systemtap.base/bz5274.c:11"),
(bz5274a.stp:1:1), hits: 4 [...]
process(".../testsuite/bz5274").function("funcd@.../systemtap.base/bz5274.c:5"),
(bz5274a.stp:1:1), hits: 4 [...]
end, (<synthetic>:1:1), hits: 1, cycles: 4608min/4608avg/4608max, from: end,
index: 6
----- refresh report:

WARNING: .../staprun exited with status: 1
Pass 5: run failed.  [man error::pass5]

[same with -DSTP_ALIBI]

% coredumpctl (pick last $pid)
% sudo coredumpctl gdb $pid
Program terminated with signal SIGSEGV, Segmentation fault.
#0  0x0000000000400774 in funca (s=0xbffffffda1e80000 <error: Cannot access
memory at address 0xbffffffda1e80000>, env=0xb80000000abf)
    at .../bz5274.c:34

(gdb) info registers
rax            0x4    4
rbx            0x0    0
rcx            0x7fffffe4    2147483620
rdx            0x400752    4196178
rsi            0x4    4
rdi            0x7ffd3654d590    140725514982800
rbp            0x41ff894156415741    0x41ff894156415741
rsp            0x4007d8    0x4007d8 <__libc_csu_init+8>
r8             0x7ffd3654d550    140725514982736
r9             0x7ffd3654d570    140725514982768
r10            0x0    0
r11            0x246    582
r12            0x400570    4195696
r13            0x7ffd3654d740    140725514983232
r14            0x0    0
r15            0x0    0
rip            0x400774    0x400774 <funca+62>
eflags         0x10246    [ PF ZF IF RF ]
cs             0x33    51
ss             0x2b    43
ds             0x0    0
es             0x0    0
fs             0x0    0
gs             0x0    0


Note the corrupted $rsp value.


My attempts to reproduce this with "perf probe" failed, without exonerating the
kernel uprobes:

% stap -p2 bz5274a.stp) | sh .../scripts/stap2perf | sudo sh
[...]
  probe:bz5274_main_return                           [Tracepoint event]
  probe:bz5274_funcd                                 [Tracepoint event]
  probe:bz5274_funcc                                 [Tracepoint event]
  probe:bz5274_funcb_return                          [Tracepoint event]
  probe:bz5274_funca_return                          [Tracepoint event]

% sudo perf record -e 'probe:*' -aR .../bz5274
[...]
In bz5274.c: funcc :13 : i=4.  Calling funcd
In bz5274.c: funcd :7 : i=4.  Calling longjmp
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.888 MB perf.data (~38805 samples) ]

% sudo perf report
[...] "The perf.data file has no samples"

% sudo perf report -D
[...]
Aggregated stats:
           TOTAL events:       7357
            MMAP events:         96
            COMM events:        648
            EXIT events:          2
            FORK events:        334
           MMAP2 events:       6276
  FINISHED_ROUND events:          1
probe:bz5274_main_return stats:
probe:bz5274_funcd stats:
probe:bz5274_funcc stats:
probe:bz5274_funcb_return stats:
probe:bz5274_funca_return stats:

... indicates zero uprobes hits of any sort

-- 
You are receiving this mail because:
You are the assignee for the bug.

^ permalink raw reply	[flat|nested] 8+ messages in thread
end of thread, other threads:[~2018-10-18 18:39 UTC | newest]

Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-03-28  0:27 [Bug uprobes/18171] New: setjmp/longjmp clashing with uprobes, causing SIGSEGV or SIGILL fche at redhat dot com
2015-03-28  0:31 ` [Bug uprobes/18171] setjmp/longjmp clashing with u[ret]probes, " fche at redhat dot com
2015-03-30 14:17 ` mcermak at redhat dot com
2015-04-09 21:26 ` mjw at redhat dot com
2015-05-22 20:20 ` fche at redhat dot com
2017-10-11 10:44 ` mark at klomp dot org
2017-10-11 10:49 ` mjw at fedoraproject dot org
2018-10-18 18:39 ` fche at redhat dot com

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).