This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH 2/2] Add test for fast tracepoint enable/disable
- From: Simon Marchi <simon dot marchi at ericsson dot com>
- To: Yao Qi <qiyaoltc at gmail dot com>
- Cc: <gdb-patches at sourceware dot org>
- Date: Thu, 31 Mar 2016 14:01:15 -0400
- Subject: Re: [PATCH 2/2] Add test for fast tracepoint enable/disable
- Authentication-results: sourceware.org; auth=none
- References: <1459351018-23718-1-git-send-email-simon dot marchi at ericsson dot com> <1459351018-23718-2-git-send-email-simon dot marchi at ericsson dot com> <86k2kirk18 dot fsf at gmail dot com>
On 16-03-31 06:58 AM, Yao Qi wrote:
> Simon Marchi <simon.marchi@ericsson.com> writes:
>
>> +# Check that the target supports trace.
>> +if { [gdb_compile_pthreads "$srcdir/$subdir/$srcfile" $binfile executable $options] != "" } {
>> + untested "Couldn't compile test program"
>> + return -1
>> +}
>
> Why do you compile program with pthread?
Good question. Because the setup for testing fast tracepoints is tedious, I always
use ftrace-lock.exp as a template, because it is relatively short. ftrace-lock.exp
uses gdb_compile_pthreads. I think it makes sense, because the IPA starts a helper
thread with pthread_create.
Note that it still builds fine with gdb_compile instead of gdb_compile_pthreads,
probably because libinproctrace.so has a DT_NEEDED for libpthreads.so, so it gets
pulled in. But using gdb_compile_pthreads should add -pthread to the CFLAGS, which
may impact compilation (e.g. choose re-entrant versions of functions).
However, I just noticed that the command line includes -lpthreads instead of -pthreads:
spawn gcc -Wl,--no-as-needed /home/emaisin/build/binutils-gdb/gdb/testsuite/../../../../src/binutils-gdb/gdb/testsuite/gdb.trace/ftrace-enable-disable.c \
/home/emaisin/build/binutils-gdb/gdb/testsuite/../gdbserver/libinproctrace.so -g \
-lpthread -lm -o \
/home/emaisin/build/binutils-gdb/gdb/testsuite/outputs/gdb.trace/ftrace-enable-disable/ftrace-enable-disable
I think that to be more "correct", we should try -pthread before -lpthread in
gdb_compile_pthreads...
>> +
>> +clean_restart ${testfile}
>> +
>> +if ![runto_main] {
>> + fail "Can't run to main to check for trace support"
>> + return -1
>> +}
>> +
>> +if ![gdb_target_supports_trace] {
>> + unsupported "target does not support trace"
>> + return -1
>> +}
>> +
>> +# Compile the test case with the in-process agent library.
>> +set libipa [get_in_proc_agent]
>> +gdb_load_shlibs $libipa
>> +
>> +lappend options shlib=$libipa
>> +
>> +if { [gdb_compile_pthreads "$srcdir/$subdir/$srcfile" $binfile executable $options] != "" } {
>> + untested "Couldn't compile test program with in-process agent library"
>> + return -1
>> +}
>
> Likewise.
Same response :)
>> +
>> +# This test makes sure that disabling and enabling fast tracepoints works
>> +# correctly.
>> +
>> +proc test_ftrace_enable_disable { } {
>
> Can you extend the test to cover normal tracepoint as well? It
> shouldn't be hard. You can just pass argument "type" to "proc
> test_trace_enable_disable" like,
>
> proc test_trace_enable_disable { type } {
>
> and the type can be "trace" and "ftrace".
I'll try to do that.
Thanks,
Simon