| Summary: | FAIL: gdb.reverse/step-reverse.exp: reverse step into fn call | ||
|---|---|---|---|
| Product: | gdb | Reporter: | Tom de Vries <vries> |
| Component: | gdb | Assignee: | Not yet assigned to anyone <unassigned> |
| Status: | NEW --- | ||
| Severity: | normal | ||
| Priority: | P2 | ||
| Version: | unknown | ||
| Target Milestone: | --- | ||
| Host: | Target: | ||
| Build: | Last reconfirmed: | ||
| Attachments: |
log file including debug infrun output (gcc 9.4.0 case)
log file including debug infrun output (gcc 12.2.1 case) Tentative patch |
||
Hmm, I'm starting to wonder if I mis-analyzed this. Looking at: ... stepi^M 78 a[5] = a[3] - a[4]; /* FINISH TEST */^M (gdb) PASS: gdb.reverse/step-reverse.exp: simple reverse stepi step^M 78 a[5] = a[3] - a[4]; /* FINISH TEST */^M (gdb) FAIL: gdb.reverse/step-reverse.exp: reverse step into fn call ... we see a step from line 78 to line ... 78. That looks like the problem. The decision to stop stepping happens here in this code in process_event_stop_test:
...
/* When stepping backward, stop at beginning of line range
(unless it's the function entry point, in which case
keep going back to the call point). */
CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
if (stop_pc == ecs->event_thread->control.step_range_start
&& stop_pc != ecs->stop_func_start
&& execution_direction == EXEC_REVERSE)
end_stepping_range (ecs);
else
keep_going (ecs);
return;
}
/* We stepped out of the stepping range. */
...
AFAICT, there just isn't enough information at this point to decide to stop stepping.
This code is part of an early-out that tries to do:
...
/* If stepping through a line, keep going if still within it.
...
As mentioned before, this failure shows up with gcc 9.4.0. It doesn't show up with gcc 12.2.1, because of a difference in generated code. In the 12.2.1 case we have 4 insns: ... 401208: 8b 55 cc mov -0x34(%rbp),%edx 40120b: 8b 45 d0 mov -0x30(%rbp),%eax 40120e: 29 c2 sub %eax,%edx 401210: 89 55 d4 mov %edx,-0x2c(%rbp) ... each with their line number entry: ... step-reverse.c 78 0x401208 x step-reverse.c 78 0x40120b x step-reverse.c 78 0x40120e x step-reverse.c 78 0x401210 x ... (In reply to Tom de Vries from comment #3) > As mentioned before, this failure shows up with gcc 9.4.0. > > It doesn't show up with gcc 12.2.1, because of a difference in generated > code. > > In the 12.2.1 case we have 4 insns: > ... > 401208: 8b 55 cc mov -0x34(%rbp),%edx > 40120b: 8b 45 d0 mov -0x30(%rbp),%eax > 40120e: 29 c2 sub %eax,%edx > 401210: 89 55 d4 mov %edx,-0x2c(%rbp) > ... > each with their line number entry: > ... > step-reverse.c 78 0x401208 x > step-reverse.c 78 0x40120b x > step-reverse.c 78 0x40120e x > step-reverse.c 78 0x401210 x > ... In the gcc 9.4.0 case, we have 5 insns: ... 401228: 8b 55 cc mov -0x34(%rbp),%edx 40122b: 8b 45 d0 mov -0x30(%rbp),%eax 40122e: 29 c2 sub %eax,%edx 401230: 89 d0 mov %edx,%eax 401232: 89 45 d4 mov %eax,-0x2c(%rbp) ... but still 4 line number entries: ... step-reverse.c 78 0x401228 x step-reverse.c 78 0x40122b x step-reverse.c 78 0x40122e x step-reverse.c 78 0x401232 x ... Note that the insn at 0x401230 doesn't have it's own line number entry. So, we are at line 79: ... (gdb) stepi^M 79 callee(); /* STEPI TEST */^M ... and do a stepi, which lands us, as expected at insn 0x401232: ... stepi^M 78 a[5] = a[3] - a[4]; /* FINISH TEST */^M (gdb) PASS: gdb.reverse/step-reverse.exp: simple reverse stepi print /x $pc^M $2 = 0x401232^M (gdb) PASS: gdb.reverse/step-reverse.exp: print /x $pc ... Then we do a step, which lands us, unexpectedly, at insn 0x40122e: ... step^M 78 a[5] = a[3] - a[4]; /* FINISH TEST */^M (gdb) FAIL: gdb.reverse/step-reverse.exp: reverse step into fn call print /x $pc^M $3 = 0x40122e^M (gdb) PASS: gdb.reverse/step-reverse.exp: print /x $pc ... It's stepping over the 0x401230 insn without line number entry that triggers the problem. Created attachment 14575 [details]
log file including debug infrun output (gcc 9.4.0 case)
$ gdb -q -batch outputs/gdb.reverse/step-reverse/step-reverse -ex "set trace-commands on" -ex start -ex record -ex "break 79" -ex continue -ex reverse-stepi -ex "set debug infrun 1" -ex reverse-step 2>&1 | tee log.txt
Created attachment 14576 [details]
log file including debug infrun output (gcc 12.2.1 case)
This passes gdb.reverse/*.exp:
...
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 181d961d80d..ba0b10de965 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -6900,11 +6900,12 @@ process_event_stop_test (struct execution_control_state *ecs)
if (stop_pc == ecs->event_thread->control.step_range_start
&& stop_pc != ecs->stop_func_start
&& execution_direction == EXEC_REVERSE)
- end_stepping_range (ecs);
+ ;
else
- keep_going (ecs);
-
- return;
+ {
+ keep_going (ecs);
+ return;
+ }
}
/* We stepped out of the stepping range. */
@@ -7401,6 +7402,8 @@ process_event_stop_test (struct execution_control_state *ecs)
"it's not the start of a statement");
}
}
+ else if (execution_direction == EXEC_REVERSE)
+ refresh_step_info = false;
/* We aren't done stepping.
...
Created attachment 14588 [details]
Tentative patch
|
[ Likewise for gdb.reverse/step-precsave.exp, which uses the same source file. ] On ubuntu 20.04.5 x86_64 (using gcc 9.4.0), I run into: ... (gdb) PASS: gdb.reverse/step-reverse.exp: simple reverse stepi step^M 78 a[5] = a[3] - a[4]; /* FINISH TEST */^M (gdb) FAIL: gdb.reverse/step-reverse.exp: reverse step into fn call step^M ... The basic problem here is that the forward execution goes like: ... $ ./build/gdb/gdb -q -batch ./build/gdb/testsuite/outputs/gdb.reverse/step-reverse/step-reverse -ex "b 78" -ex run -ex "set trace-commands on" -ex stepi -ex "stepi" -ex "stepi" -ex "stepi" -ex "stepi" Breakpoint 1 at 0x123b: file /home/vries/gdb/src/gdb/testsuite/gdb.reverse/step-reverse.c, line 78. Breakpoint 1, main () at /home/vries/gdb/src/gdb/testsuite/gdb.reverse/step-reverse.c:78 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 0x0000555555555243 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 79 callee(); /* STEPI TEST */ $ ... because we have: ... INDEX LINE ADDRESS IS-STMT PROLOGUE-END 26 75 0x0000555555555231 Y 27 78 0x000055555555523b Y 28 78 0x000055555555523e Y 29 78 0x0000555555555241 Y 30 78 0x0000555555555245 Y 31 79 0x0000555555555248 Y ... and when doing the backward execution, we stop at 0x0000555555555245 rather stepi-ing to 0x000055555555523b as is the intention. This simple test-case patch fixes it: ... diff --git a/gdb/testsuite/gdb.reverse/step-reverse.exp b/gdb/testsuite/gdb.reverse/step-reverse.exp index 27e4b175274..8ed4f6d0269 100644 --- a/gdb/testsuite/gdb.reverse/step-reverse.exp +++ b/gdb/testsuite/gdb.reverse/step-reverse.exp @@ -72,6 +72,8 @@ gdb_test_multiple "finish" "$test_message" { } } +set pc_finish_test [get_valueof /x {$pc} ""] + # stepi over flat code (no calls) set test_message "simple stepi" @@ -216,7 +218,13 @@ gdb_test_multiple "stepi" "$test_message" { exp_continue } -re "$stepi_location.* FINISH TEST.*$gdb_prompt $" { - pass "$test_message" + set pc [get_valueof /x {$pc} ""] + if { $pc != $pc_finish_test } { + send_gdb "stepi\n" + exp_continue + } else { + pass "$test_message" + } } -re "STEP INTO THIS CALL.*$gdb_prompt $" { fail "$test_message (too far)" ... (though it introduces a duplicate for the get_valueof). OTOH, this can be fixed by: ... diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c index 851852404eb..0e6ee8fe04d 100644 --- a/gdb/dwarf2/read.c +++ b/gdb/dwarf2/read.c @@ -20205,11 +20205,6 @@ dwarf_record_line_p (struct dwarf2_cu *cu, return 1; if (line != last_line) return 1; - /* Same line for the same file that we've seen already. - As a last check, for pr 17276, only record the line if the line - has never had a non-zero discriminator. */ - if (!line_has_non_zero_discriminator) - return 1; return 0; } ... which gets us: ... 25 75 0x0000555555555231 Y 26 78 0x000055555555523b Y 27 79 0x0000555555555248 Y ... and: ... Breakpoint 1, main () at step-reverse.c:78 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 0x000055555555523e 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 0x0000555555555241 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 0x0000555555555243 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 0x0000555555555245 78 a[5] = a[3] - a[4]; /* FINISH TEST */ +stepi 79 callee(); /* STEPI TEST */ ... So in that sense, it's similar to PR25911.