Re: [1/2] C-SKY Port

[Hafiz Abid Qadeer <abid_qadeer at mentor dot com>]

On 21/10/18 04:22, Simon Marchi wrote:
> On 2018-10-20 11:04 p.m., Hafiz Abid Qadeer wrote:
>> On 20/10/18 19:55, Simon Marchi wrote:
>>> On 2018-07-25 6:54 a.m., Hafiz Abid Qadeer wrote:
>>
>>> Hi Hafiz,
>>>
>>> I noticed the "cooked_read" selftest fails with c-sky (ever since the c-sky support was added):
>>>
>>> (gdb) maintenance selftest cooked_read
>>> Running selftest regcache::cooked_read_test.
>>> ...
>>> Self test failed: arch csky: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> Self test failed: arch csky:ck510: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> Self test failed: arch csky:ck610: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> Self test failed: arch csky:ck801: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> Self test failed: arch csky:ck802: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> Self test failed: arch csky:ck803: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> Self test failed: arch csky:ck807: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> Self test failed: arch csky:ck810: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> Self test failed: arch csky:any: self-test failed at /home/simark/src/binutils-gdb/gdb/regcache.c:1697
>>> ...
>>> Self test failed: self-test failed at /home/simark/src/binutils-gdb/gdb/selftest-arch.c:86
>>> Ran 1 unit tests, 1 failed
>>>
>>> Could you take a look?
>> Hi Simon,
>> Thanks for letting me know. I am traveling at the moment. Will take a
>> look at it when I am back in a few days.
> 
> Ok, thanks!
> 
> A quick investigation shows it's because some raw registers are not in the
> save reggroup.  So csky should probably be added to that big if in the test:
> 
> 	  if (bfd_arch == bfd_arch_frv || bfd_arch == bfd_arch_h8300
> 	      || bfd_arch == bfd_arch_m32c || bfd_arch == bfd_arch_sh
> 	      || bfd_arch == bfd_arch_alpha || bfd_arch == bfd_arch_v850
> 	      || bfd_arch == bfd_arch_msp430 || bfd_arch == bfd_arch_mep
> 	      || bfd_arch == bfd_arch_mips || bfd_arch == bfd_arch_v850_rh850
> 	      || bfd_arch == bfd_arch_tic6x || bfd_arch == bfd_arch_mn10300
> 	      || bfd_arch == bfd_arch_rl78 || bfd_arch == bfd_arch_score
> 	      || bfd_arch == bfd_arch_riscv)
> 	    {
> 	      /* Raw registers.  If raw registers are not in save_reggroup,
> 		 their status are unknown.  */
> 	      if (gdbarch_register_reggroup_p (gdbarch, regnum, save_reggroup))
> 		SELF_CHECK (status == REG_VALID);
> 	      else
> 		SELF_CHECK (status == REG_UNKNOWN);
> 	    }
> 	  else
> 	    SELF_CHECK (status == REG_VALID);
> 

When I was reviewing the test results for csky, I saw this fail in
unittest.exp and fixed it by adding csky to the big if list as you
mentioned above. Somehow I missed that patch when upstreaming the port.
So if you are happy with that change then I will push it.

> But I'm tempted to replace all of that with simply;
> 
> 	  /* Raw registers.  If raw registers are not in save_reggroup,
> 	     their status are unknown.  */
> 	  if (gdbarch_register_reggroup_p (gdbarch, regnum, save_reggroup))
> 	    SELF_CHECK (status == REG_VALID);
> 	  else
> 	    SELF_CHECK (status == REG_UNKNOWN);
> 
> We won't have to maintain this big list, and I don't think we lose any testing
> coverage/safety.  For architectures for which all raw registers are in the
> save_reggroup (all arches not listed above), we will still always assert that
> status == REG_VALID.

I have not looked at this code much but your suggestion makes sense to me.

Regards,
-- 
Hafiz Abid Qadeer
Mentor Embedded/CodeSourcery