[RFC] Skip hardware breakpoint tests on OpenBSD

[RFC] Skip hardware breakpoint tests on OpenBSD

[Mark Kettenis]

The recently introduced skip_hw_breakpoint_tests procedure is a bit
flawed.  It lists hardware platforms that have hardware breakpoints,
but doesn't take into account that the OS running on those platforms
has to make that functionality available to GDB.  This isn't the case
on OpenBSD (and perhaps some other operating systems as well).  The
diff below makes sure the tests are skipped on OpenBSD.  However,
perhaps instead of whitelisting generic hardware platforms, we should
whitelis explicit hardware/OS combinations here?

Thoughts?

2011-03-09  Mark Kettenis  <kettenis@gnu.org>

        * lib/gdb.exp (skip_hw_breakpoint_tests): None of the OpenBSD
        targets actually implement hardware breakpoints, so return 1 for
        *-*-openbsd*.

Index: lib/gdb.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/lib/gdb.exp,v
retrieving revision 1.167
diff -u -p -r1.167 gdb.exp
--- lib/gdb.exp	9 Mar 2011 14:17:05 -0000	1.167
+++ lib/gdb.exp	9 Mar 2011 16:20:41 -0000
@@ -1795,6 +1795,11 @@ proc skip_hw_breakpoint_tests {} {
 	return 1
     }
 
+    # These targets don't support hardware breakpints
+    if { [istarget "*-*-openbsd*"] } {
+	return 1
+    }
+
     # These targets support hardware breakpoints natively
     if { [istarget "i?86-*-*"] 
 	 || [istarget "x86_64-*-*"]

Re: [RFC] Skip hardware breakpoint tests on OpenBSD

[Ulrich Weigand]

Mark Kettenis wrote:

> The recently introduced skip_hw_breakpoint_tests procedure is a bit
> flawed.  It lists hardware platforms that have hardware breakpoints,
> but doesn't take into account that the OS running on those platforms
> has to make that functionality available to GDB.  This isn't the case
> on OpenBSD (and perhaps some other operating systems as well).  The
> diff below makes sure the tests are skipped on OpenBSD.  However,
> perhaps instead of whitelisting generic hardware platforms, we should
> whitelis explicit hardware/OS combinations here?

I've just copied the existing tests that were spread throughout
various testcases into the new procedure ...

That said, I completely agree that this should really be specific
to the hardware/OS combinations.  The question is what exactly
these ought to be.  Looking through the config files, it would
appear that for the ia64, arm, powerpc and s390 architectures,
the only OS supporting watchpoints is Linux.  For the x86
architectures, the list gets a lot more complex:
   i[34567]86-*-go32*
   i[34567]86-*-msdosdjgpp*
   i[34567]86-*-cygwin*
   i[34567]86-*-mingw32*
   i[34567]86-*-freebsd*
   i[34567]86-*-kfreebsd*-gnu
   i[34567]86-*-linux*
   i[34567]86-*-solaris2.[6789]
   i[34567]86-*-solaris2.1[0-9]
   x86_64-*-mingw*
   x86_64-*-linux*
   x86_64-*-solaris2.1[0-9]*

While I don't really have access to any of those (except Linux)
for testing, it probably cannot hurt to replace the generic
x86 test with a test for those specific platforms; the worst
that could go wrong is that we forgot a platform.

B.t.w. it turns out that there are architectures that support
watchpoints that haven't been mentioned in the list at all:
   sparc-*-solaris2.[6789]
   sparc-*-solaris2.1[0-9]
   sparcv9-*-solaris2.[6789]
   sparcv9-*-solaris2.1[0-9]
   sparc64-*-solaris2.[6789]
   sparc64-*-solaris2.1[0-9]
   mips-sgi-irix5*
   mips-sgi-irix6*
   mips*-*-linux*

Here I'd argue it would be best for someone with access to the
platform to add them.

Thoughts?

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com

Re: [RFC] Skip hardware breakpoint tests on OpenBSD

[Eli Zaretskii]

> Date: Thu, 10 Mar 2011 00:58:44 +0100 (CET)
> From: "Ulrich Weigand" <uweigand@de.ibm.com>
> Cc: gdb-patches@sourceware.org
> 
> That said, I completely agree that this should really be specific
> to the hardware/OS combinations.  The question is what exactly
> these ought to be.  Looking through the config files, it would
> appear that for the ia64, arm, powerpc and s390 architectures,
> the only OS supporting watchpoints is Linux.  For the x86
> architectures, the list gets a lot more complex:
>    i[34567]86-*-go32*
>    i[34567]86-*-msdosdjgpp*
>    i[34567]86-*-cygwin*
>    i[34567]86-*-mingw32*
>    i[34567]86-*-freebsd*
>    i[34567]86-*-kfreebsd*-gnu
>    i[34567]86-*-linux*
>    i[34567]86-*-solaris2.[6789]
>    i[34567]86-*-solaris2.1[0-9]
>    x86_64-*-mingw*
>    x86_64-*-linux*
>    x86_64-*-solaris2.1[0-9]*

I can confirm that hardware-assisted breakpoints and watchpoints are
supported in the DJGPP (that includes go32 as well, btw) and the
MinGW32 builds.  This is as of GDB 7.2.

HTH

Re: [RFC] Skip hardware breakpoint tests on OpenBSD

[Ulrich Weigand]

Eli Zaretskii wrote:

> I can confirm that hardware-assisted breakpoints and watchpoints are
> supported in the DJGPP (that includes go32 as well, btw) and the
> MinGW32 builds.  This is as of GDB 7.2.

OK, thanks for verifying!

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com