Re: [ARM] testsuite remaining errors

[Matthew Wahab <matthew dot wahab at foss dot arm dot com>]

On 19/11/15 23:08, Christophe Lyon wrote:
> On 18 November 2015 at 12:41, Matthew Wahab <matthew.wahab@foss.arm.com> wrote:
> > On 16/11/15 20:52, Christophe Lyon wrote:
> > > Hi,
> > >
> > > I ran the testsuite on a variety of arm-* configurations, and I
> > > noticed a few FAILs:

I've looked into the GAS failures:

* armv8-a+pan:

Labels appearing before .arm/.thumb directives.
The fix is to put the labels after the directives.

* FAIL: ARM WinCE basic tests (wince.d):

Needs -mccs added to the options.

This looks like a change in behaviour since
----
Author: Daniel Gutson <daniel.gutson@tallertechnologies.com>
Date:   Wed Mar 19 14:31:25 2014 +0000
    Add support for ARM assembler produced by CodeCompositor Studio.
----

Fixed by adjusting the test case.

* FAIL: ARM basic instructions (WinCE version) (wince_inst.d):

The output from arm-wince-pe is no longer different from the output of
other targets.

Fixed by copying the expected output from inst.d. I'd suggest deleting
this test but I still don't know anything about arm-wince-pe.

* blx-bl-convert:

The test uses the .type directive in a way that the manual says is only
valid for ELF targets. The test should be skipped for non-elf targets.

Also, option -mccs is needed to get the assembler to parse the file
without complaining about junk at the end of lines.

* ldst-offset0 (ARM load/store with 0 offset)
  thumb2_ldr_immediate_armv6t2 (Ldr immediate on armv6):

This test puts data (0x0) in the .text section which is
dissassembled by the arm-wince-pe objdump as andeq.

Fixed by skipping the test for coff targets. This follows
thumb2_ldr_immediate_armv6.d.

* armv7e-m+fpv5-d16 (Valid v7e-m+fpv5-d16):
* armv7e-m+fpv5-sp-d16 (Valid v7e-m+fpv5-sp-d16):

These are test files for the Cortex-M7 and the fpv5-sp-d16 and fpv5-d16
FPUs.

The patch that added this support is:
---
Author: Terry Guo <terry.guo@arm.com>
Date:   Fri Nov 21 11:31:37 2014 +0800
    Support ARM Cortex-M7
---

That patch updated elf32-arm but I don't think that the coff support was
added. (bfd/coff-arm.c seems to have had only refactoring changes since
2013.)

These tests should probably be skipped for coff targets.

* align2.s, align64.s, bundle-lock.s, bundle.s (bundle_align_mode):

Since
----
commit 7a5c933c7c2c3926b13a047c64f2af1d3ce8e1db
Author: Richard Earnshaw <Richard.Earnshaw@arm.com>
Date:   Fri Apr 24 15:51:49 2015 +0100
    [ARM]: Don't tail-pad over-aligned functions to the alignment boundary.
----

This is something going wrong with the section handling but I don't know
what. I'll open a bugzilla entry for it but won't be able to do any more
than that.

I'll send a patch to adjusts the other tests.

Matthew