[testsuite] further robustify gdb.base/bp-permanent.exp
Commit Message
When I was testing the Nios II R2 patch I posted earlier today, I saw a
lot of failures from gdb.base/bp-permanent.exp. Upon investigation, I
realized that this was a due to a quirk of the simulator we have
available. Apparently, its execution engine maintains its own view of
the .text section and breakpoints set through the debug interface are
not visible to the running program. This means that the setup function
in the program is stashing away a copy of the original NOP instruction
at the breakpoint location, rather than the breakpoint instruction
inserted by GDB, and when the .exp file thinks it is writing a permanent
breakpoint in fact it is just telling the target to write a NOP to a
location that already contains one. :-S
Anyway, it is clear that if the program runs to termination instead of
stopping where GDB expects, none of the remaining tests are going to do
anything useful, so we might just as well report that the test is
unsupported and return.
OK to commit?
-Sandra
Comments
On 07/30/2015 05:05 AM, Sandra Loosemore wrote:
> When I was testing the Nios II R2 patch I posted earlier today, I saw a
> lot of failures from gdb.base/bp-permanent.exp. Upon investigation, I
> realized that this was a due to a quirk of the simulator we have
> available. Apparently, its execution engine maintains its own view of
> the .text section and breakpoints set through the debug interface are
> not visible to the running program. This means that the setup function
> in the program is stashing away a copy of the original NOP instruction
> at the breakpoint location, rather than the breakpoint instruction
> inserted by GDB, and when the .exp file thinks it is writing a permanent
> breakpoint in fact it is just telling the target to write a NOP to a
> location that already contains one. :-S
>
> Anyway, it is clear that if the program runs to termination instead of
> stopping where GDB expects, none of the remaining tests are going to do
> anything useful, so we might just as well report that the test is
> unsupported and return.
>
> OK to commit?
> + set test "permanent breakpoint causes random signal"
> + gdb_test_multiple "continue" $test {
> + -re "exited normally.*$gdb_prompt $" {
> + unsupported "failed to stop at permanent breakpoint"
> + return
> + }
> + -re "Program received signal SIGTRAP.*" {
This needs to end with "$gdb_prompt $" too.
OK with that change.
> + pass $test
> + }
> + }
>
> # Now set a breakpoint on top, thus creating a permanent breakpoint.
> gdb_breakpoint "$line_bp"
>
Thanks,
Pedro Alves
@@ -123,8 +123,21 @@ proc test {always_inserted sw_watchpoint} {
with_test_prefix "basics" {
# Run to the permanent breakpoint, just to make sure we've inserted it
# correctly.
- gdb_test "continue" "Program received signal SIGTRAP.*" \
- "permanent breakpoint causes random signal"
+ # If the target fails to stop, the remainder of the test will not work
+ # so just return. This can happen on some simulator targets where
+ # the running program doesn't see breakpoints that are visible to
+ # the execution engine, or where writes to the .text section are
+ # quietly ignored.
+ set test "permanent breakpoint causes random signal"
+ gdb_test_multiple "continue" $test {
+ -re "exited normally.*$gdb_prompt $" {
+ unsupported "failed to stop at permanent breakpoint"
+ return
+ }
+ -re "Program received signal SIGTRAP.*" {
+ pass $test
+ }
+ }
# Now set a breakpoint on top, thus creating a permanent breakpoint.
gdb_breakpoint "$line_bp"