Failure of test 'basic gdbstub support'

Failure of test 'basic gdbstub support'

[Robert Henry]

[-- Attachment #1: Type: text/plain, Size: 1299 bytes --]

The newish test 'basic gdbstub support' fails for me on an out-of-the-box  build on a host x86_64.  (See below for the config.log head.)

Is this failure expected?  If so, where can I see that in the various CI engines you have running them?

In digging through the test driver python code in tests/tcg/multiarch/gdbstub/sha1.py I see that the test assumes that a breakpoint on the function SHA1Init is a breakpoint at the 1st assignment statement; the 1st next executes the 1st assignment statement, etc.

This is a very fragile assumption.  It depends on the compiler used to compile sha1.c; it depends on the optimization level; it depends on the accuracy of the pc mapping in the debug info; it depends on gdb.

Better would be to change SHA1Init to do its work, and then call another non-inlined function taking a context pointer, and then examine context->state[0] and context->state[1].

Thanks in advance

TEST    basic gdbstub support
make[2]: *** [/mnt/robhenry/qemu_robhenry_amd64/qemu/tests/tcg/multiarch/Makefile.target:51: run-gdbstub-sha1] Error 2


 QEMU configure log Tue 09 Jun 2020 02:45:06 PM PDT
# Configured with: '../configure' '--disable-sdl' '--enable-gtk' '--extra-ldflags=-L/usr/lib' '--enable-plugins' '--target-list=x86_64-softmmu x86_64-linux-user'


[-- Attachment #2: Type: text/html, Size: 3343 bytes --]

Re: Failure of test 'basic gdbstub support'

[Alex Bennée]

Robert Henry <robhenry@microsoft.com> writes:

> The newish test 'basic gdbstub support' fails for me on an out-of-the-box  build on a host x86_64.  (See below for the config.log head.)
>
> Is this failure expected?  If so, where can I see that in the various
> CI engines you have running them?

No it shouldn't fail - if the GDB doesn't attach we should be skipping
the test.

>
> In digging through the test driver python code in tests/tcg/multiarch/gdbstub/sha1.py I see that the test assumes that a breakpoint on the function SHA1Init is a breakpoint at the 1st assignment statement; the 1st next executes the 1st assignment statement, etc.
>
> This is a very fragile assumption.  It depends on the compiler used to
> compile sha1.c;

True - certainly for "native" built test images.

>it depends on the optimization level; it depends on the accuracy of the
>pc mapping in the debug info;

All test cases are built -O0 -g3 so should have full debug info.

>it depends on gdb.

This is out biggest problem at the moment because it seems support for
socket based debug is fairly new for gdb.

>
> Better would be to change SHA1Init to do its work, and then call
> another non-inlined function taking a context pointer, and then
> examine context->state[0] and context->state[1].

We are also trying to test the step and next functionality is working
here.

>
> Thanks in advance
>
> TEST    basic gdbstub support
> make[2]: ***
> [/mnt/robhenry/qemu_robhenry_amd64/qemu/tests/tcg/multiarch/Makefile.target:51:
> run-gdbstub-sha1] Error 2

Can you let me know what the contents of:

  cat tests/tcg/x86_64-linux-user/run-gdbstub-sha1.out

is please?
>
>  QEMU configure log Tue 09 Jun 2020 02:45:06 PM PDT
> # Configured with: '../configure' '--disable-sdl' '--enable-gtk' '--extra-ldflags=-L/usr/lib' '--enable-plugins' '--target-list=x86_64-softmmu x86_64-linux-user'


-- 
Alex Bennée