Re: Fixing Linux/SPARC

[Daniel Jacobowitz <drow at mvista dot com>]

On Sat, Nov 17, 2001 at 02:18:42PM -0500, Andrew Cagney wrote:
> >On Wed, Oct 17, 2001 at 04:07:52PM +0200, Mark Kettenis wrote:
> >
> >>Hi Dan & other interested folks,
> >>
> >>Something like the code below (completely untested, probably doesn't
> >>even compile) is needed to fix Linux/SPARC.
> >
> >
> >Compiles, after fixing the obvious typos.  I'm going to commit it to
> >the mainline as obvious, since it takes Linux/SPARC from not-building
> >to building; test results are abysmal, though.  It looks as if we can't
> >figure out the memory location variables are stored at correctly. 
> >They're being accessed in the unmapped 0x90000000 segment instead of
> >0x70000000, and not even at the right offsets.  I'll look in to it more
> >later.
> 
> Dan, can you please replace the calls to read_register_gen() with 
> regcache_collect().

Done.

Threads don't work on Sparc still, because prgregset_t is much smaller
than an elf_gregset_t, so we corrupt memory in thread_db_fetch_registers.
You know by now what my opinion of this problem is :)

It appears that, for Sparc at least, this was fixed on:
2000-03-21  Jakub Jelinek  <jakub@redhat.com>

My Sparc test machine runs a glibc older than that.  I'd like to to
come up with some autoconf test to either not compile at all when this
bug is present, disable thread support, or compile in some hack to make
the type the right size.  Any objection?

Actually, gross as it may be, something like this in thread-db.c ought
to do the trick:
union big_enough_gregset_t {
  prgregset_t p;
  gdb_gregset_t g;
};

and replacing uses of the prgregset with uses of big_enough_gregset_t.p
should fix the warnings without any autoconf checks.  How's that sound?

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

2001-11-17  Daniel Jacobowitz  <drow@mvista.com>

        * sparc-linux-nat.c (fill_gregset): Replace read_register_gen
        with regcache_collect.
        (fill_fpregset): Likewise.

Index: sparc-linux-nat.c
===================================================================
RCS file: /cvs/src/src/gdb/sparc-linux-nat.c,v
retrieving revision 1.1
diff -u -p -r1.1 sparc-linux-nat.c
--- sparc-linux-nat.c	2001/11/17 18:38:29	1.1
+++ sparc-linux-nat.c	2001/11/17 20:18:40
@@ -56,22 +56,22 @@ fill_gregset (elf_gregset_t *gregsetp, i
 
   for (i = G0_REGNUM; i <= I7_REGNUM; i++)
     if (regno == -1 || regno == i)
-      read_register_gen (i, (char *) (regp + (i - G0_REGNUM)));
+      regcache_collect (i, regp + (i - G0_REGNUM));
 
   if (regno == -1 || regno == PS_REGNUM)
-    read_register_gen (PS_REGNUM, (char *) (regp + 32));
+    regcache_collect (PS_REGNUM, regp + 32);
 
   if (regno == -1 || regno == PC_REGNUM)
-    read_register_gen (PC_REGNUM, (char *) (regp + 33));
+    regcache_collect (PC_REGNUM, regp + 33);
   if (regno == -1 || regno == NPC_REGNUM)
-    read_register_gen (NPC_REGNUM, (char *) (regp + 34));
+    regcache_collect (NPC_REGNUM, regp + 34);
   if (regno == -1 || regno == Y_REGNUM)
-    read_register_gen (Y_REGNUM, (char *) (regp + 35));
+    regcache_collect (Y_REGNUM, regp + 35);
 
   if (regno == -1 || regno == WIM_REGNUM)
-    read_register_gen (WIM_REGNUM, (char *) (regp + 36));
+    regcache_collect (WIM_REGNUM, regp + 36);
   if (regno == -1 || regno == TBR_REGNUM)
-    read_register_gen (TBR_REGNUM, (char *) (regp + 37));
+    regcache_collect (TBR_REGNUM, regp + 37);
 }
 
 void
@@ -92,8 +92,8 @@ fill_fpregset (elf_fpregset_t *fpregsetp
 
   for (i = FP0_REGNUM; i < FP0_REGNUM + 32; i++)
     if (regno == -1 || regno == i)
-      read_register_gen (i, (char *) &fpregsetp->pr_fr.pr_regs[i - FP0_REGNUM]);
+      regcache_collect (i, &fpregsetp->pr_fr.pr_regs[i - FP0_REGNUM]);
 
   if (regno == -1 || regno == FPS_REGNUM)
-    read_register_gen (FPS_REGNUM, (char *) &fpregsetp->pr_fsr);
+    regcache_collect (FPS_REGNUM, &fpregsetp->pr_fsr);
 }