[PATCH] PowerPC: bp-permanent.exp, kill-after-signal fix

[Carl Love]

On Mon, 2022-05-02 at 09:55 -0500, will schmidt wrote:
> On Mon, 2022-05-02 at 15:32 +0100, Pedro Alves wrote:
> > On 2022-04-29 02:06, Carl Love via Gdb-patches wrote:
> > > GDB maintainers:
> > > 
> > > The gdb.base/bp-permanent.exp and the gdb.base/kill-after-signal
> > > tests
> > > fail on Power 10  The tests pass without the patch on Power 9 and
> > > Intel.  As stated in the commit log below, the tests have been
> > > run on
> > > Power 10 Linux version 5.14.0-70.9.1.el9_0.ppc64le and on a Power
> > > 9
> > > 5.4.0-96-generic kernels.  I have examined the code for the
> > > __kernel_start_sigtramp_rt64 function on both systems and the
> > > file is
> > > identical.  As far as I can tell, the failure is hardware
> > > specific.
> > > 
> > > The following patch fixes the issue on Power 10 without
> > > introducing any
> > > regression failures on Power 9 or Intel.
> > > 
> > > Please let me know if the patch is acceptable for mainline gdb. 
> > > Thanks.
> > > 
> > >                     Carl Love
> > > 
> > > ---------------------------------------------------------------
> > > -------
> > > PowerPC: bp-permanent.exp, kill-after-signal fix
> > > 
> > > The break point after the stepi on Intel is the entry point of
> > > the user
> > > signal handler function test_signal_handler.  The code at the
> > > break point
> > > looks like:
> > > 
> > >      0x<hex address> <test_signal_handler>: endbr64
> > > 
> > > On Power 10, the address where gdb stops after the stepi is in
> > > the kernel.
> > > The code at the breakpoint looks like:
> > > 
> > >   0x<hex address>  <__kernel_start_sigtramp_rt64>:	bctrl
> > > 
> > > Power 10 requires one more stepi to reach the user signal
> > > handler.
> > > 
> > > The tests run fine on Power 9.  The tests have been run on
> > > multiple Power 10
> > > systems.  The tests were done with newer and older Linux kernels
> > > and gcc
> > > compiler versions from the Power 9 system.  The tests fail
> > > identically on
> > > the two Power 10 systems but pass on the Power 9 system.
> > > 
> > > The two tests were run on the following PowerPC configurations:
> > > 
> > > Power 9, Ubuntu 20.04 LE, linux 5.4.0-96-generic,
> > > gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
> > > 
> > >   gdb.base/bp-permanent.exp 186 passes, no failures
> > >   gdb.base/kill-after-signal.exp 4 passes, no failures
> > > 
> > > Power 10, RHEL 9, Linux 5.14.0-70.9.1.el9_0.ppc64le,
> > > gcc (GCC) 11.2.1 20220127 (Red Hat 11.2.1-9)
> > >   gdb.base/bp-permanent.exp 182 passes, 4 failures
> > >   gdb.base/kill-after-signal.exp 3 passes, 1 failure
> > > 
> > > Power 10, SLE 15 SP3 , Linux 5.3.18-150300.59.63-default,
> > > gcc (SUSE Linux) 7.5.0
> > >   gdb.base/bp-permanent.exp 182 passes, 4 failures
> > >   gdb.base/kill-after-signal.exp 3 passes, 1 failure
> > > 
> > > The following patch fixes the tests on Power 10. The patch does
> > > not
> > > introduce regessions on Power 9 or Intel systems.
> > > ---
> > >  gdb/testsuite/gdb.base/bp-permanent.exp      |  8 ++++++++
> > >  gdb/testsuite/gdb.base/kill-after-signal.exp | 15
> > > ++++++++++++++-
> > >  2 files changed, 22 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/gdb/testsuite/gdb.base/bp-permanent.exp
> > > b/gdb/testsuite/gdb.base/bp-permanent.exp
> > > index 8be46e96238..f3f47e675ff 100644
> > > --- a/gdb/testsuite/gdb.base/bp-permanent.exp
> > > +++ b/gdb/testsuite/gdb.base/bp-permanent.exp
> > > @@ -258,8 +258,16 @@ proc test {always_inserted sw_watchpoint} {
> > >  		set test "single-step to handler"
> > >  		gdb_test_multiple "stepi" $test {
> > >  		    -re "Program received signal SIGTRAP.*$gdb_prompt
> > > $" {
> > > +			# Intel needs one stepi to get to the signal
> > > handler.
> > 
> > A bit odd to single out "Intel" here, when what this is is really
> > "architectures other than PowerPC".
> > But given Ulrich explained this isn't really about the hardware,
> > but instead the kernel version,
> > I think these comments should be updated to match reality better.
> 
> Interesting, I did not see the comment from Uli in my inbox here, but
> I
> did find it in the mailing list archives.
> 
> Uli stated:
> > I believe this is related to a kernel change (on Power), not the
> > hardware level as such.  The sigtramp trampoline was introduced
> > only recently.
> 
> I thought this (kernel source for handling signals) was investigated
> as
> part of the process to figure out the underlying cause of the
> issue.  
> I strongly agree the detail needs to be clarified.  :-)

I have investigated the Kernel signal handler
__kernel_start_sigtramp_rt64 in arch/powerpc/kernel/vdso64/sigtramp.S
where gdb stops.  Unfortunately, I don't see anything Power 10 specific
code paths.  I don't think it is a HW issue but clearly Power 10
specific.  I have not found anything in GLIBC that would explain it as 
there are no Power 10 specific code paths in that code.  

I agree it must be in the kernel somewhere, I just don't know where and
have yet to find it.  Unfortunately I don't know the kernel signal
handling code and am a bit lost as to where else to look.   It would be
nice to identify the specifc kernel change so we could add that to the
commit log description.  If anyone has thoughts as to where to look I
will look to see if I can find the change and add a reference in the
comit log before committing the patch.  Thanks for reviewing and
approving the patch.

                                  Carl