RE: [Patch, microblaze]: Port of Linux gdbserver

[Ajit Kumar Agarwal <ajit dot kumar dot agarwal at xilinx dot com>]

Hello Pedro:

-----Original Message-----
From: Pedro Alves [mailto:palves@redhat.com] 
Sent: Tuesday, September 16, 2014 10:34 PM
To: Ajit Kumar Agarwal; Michael Eager; Joel Brobecker
Cc: gdb-patches@sourceware.org; Vinod Kathail; Vidhumouli Hunsigida; Nagaraju Mekala
Subject: Re: [Patch, microblaze]: Port of Linux gdbserver

On 09/16/2014 07:41 AM, Ajit Kumar Agarwal wrote:
>>> >>This is identical to microblaze-with-stack-protect.c.  Both 
>>> >>specify tdesc_create_feature (result, "org.gnu.gdb.microblaze.core"); Why is this needed?
> This is needed as gdbserver code expects the register pc as "pc" instead of "rpc" for baremetel. The microblaze-linux-core.xml is changed from "rpc" to "pc" for gdbserver code to work.

>>This doesn't make much sense to me.  Can you expand please?  Why would you want the register to be named differently on Linux?  We've defined the >>org.gnu.gdb.microblaze.core with "rpc", presumably because that's what the architecture calls that core register.

>>Not reporting all the registers with the exact names GDB reports should be making GDB consider the description invalid.  Aren't you seeing that happen?

In Microblaze gdbserver code linux-microblaze-low.c we have are passing the "pc" in supply_register_by_name and the baremetal org.gnu.gdb.microblaze.core its been defined as "rpc". Due to this in regcache.c where the find_regno function compares "pc" passed with "rpc" and reports failures. That is why we have create microblaze-linux-core.xml to have "pc" instead of "rpc".

static void
microblaze_set_pc (struct regcache *regcache, CORE_ADDR pc)
{
  unsigned long newpc = pc;
  supply_register_by_name (regcache, "pc", &newpc);
}
/me looks at code

>>Oh, microblaze_gdbarch_init is incomplete...

  /* Check any target description for validity.  */
 >> if (tdesc_has_registers (tdesc))
    >>{
      >>const struct tdesc_feature *feature;
      >>int valid_p;
     >> int i;

      >>feature = tdesc_find_feature (tdesc,
           >>                         "org.gnu.gdb.microblaze.core");
      >>if (feature == NULL)
        >>return NULL;
      >>tdesc_data = tdesc_data_alloc ();

      >>valid_p = 1;
      >>for (i = 0; i < MICROBLAZE_NUM_CORE_REGS; i++)
       >> valid_p &= tdesc_numbered_register (feature, tdesc_data, i,
            >>                                microblaze_register_names[i]);
      >>feature = tdesc_find_feature (tdesc,
           >>                         "org.gnu.gdb.microblaze.stack-protect");
     >> if (feature != NULL)
       >> {
         >> valid_p = 1;
          >>valid_p &= tdesc_numbered_register (feature, tdesc_data,
               >>                               MICROBLAZE_SLR_REGNUM,
                    >>                          "rslr");
          >>valid_p &= tdesc_numbered_register (feature, tdesc_data,
               >>                               MICROBLAZE_SHR_REGNUM,
                    >>                          "rshr");
        >>}
     >>}

>>Note nothing is done with valid_p.  It's write-only.  Compare with other ports, like arm-tdep.c or mips-tdep.c.

Would look into this and will make the modification.
Thanks,
Pedro Alves