Release 2.22: Next week ... (m68hc11 testsuite)

[James Murray]

On Tue, 2011-12-20 at 14:49 +0000, James Murray wrote:
> On Tue, 2011-12-20 at 14:55 +0100, Tristan Gingold wrote:
> > On Dec 20, 2011, at 2:51 PM, James Murray wrote:
>=20
> > > Any ideas on whether this is a real problem or if the test is never
> > > going to work on m68hc11

I've been working my way through the tests and feel like I'm getting a
better handle on things.

However, it appears that the ld testsuite is effectively skipped for the
main target. All the test files start similar to this:
#source: far-hc11.s
#as: -m68hc11
#ld: -m m68hc11elf
#objdump: -d --prefix-addresses -r
#target: m6811-*-* m6812-*-*

That means that when you build for the default target of m68hc11 they
are all ignored.

That's a big problem as it seems there's a long-standing critical bug
that they would have exposed.

The hc11/hc12 platform frequently use banked memory.
Later chips (hcs12, s12x) have built in call/rtc commands to do this,
but the earlier 11,12 chips had custom hardware and need specific
trampoline code to correctly jump there. This essential trampoline code
is not being emitted. I checked in my code, virgin CVS and back to 2.18.
All the same, missing.

At a glance, the trampoline generation code exists in BFD, but appears
to be broken somehow. This is rather beyond the scope of what I was
hoping to work on and dates back to the initial work that Stephane
Carrez contributed.

Also, regarding the build targets, I'm not fully sure what purpose it
serves to build for different sub-targets. It (now) makes more sense to
build for just the one target and specify the sub-target at runtime. I
don't plan on changing that presently though.

thoughts?

regards

James Murray