Fix building on x86 OpenBSD 4.3

Fix building on x86 OpenBSD 4.3

[Pedro Alves]

[-- Attachment #1: Type: text/plain, Size: 766 bytes --]

Hi,

In my testing on several targets, I noticed that GDB isn't
building on x86 OpenBSD 4.3, due to the fact that struct
switchframe got its sf_ppl member removed, and GDB is trying
to access it.

I can't claim I know a think about this, but,

The patch that seems to have removed it, shows that is was always
0 just before removal:

 http://archive.netbsd.se/?ml=openbsd-tech&a=2007-09&m=5239543

And I can see it being set to != 0 in that files' history at:
 http://opengrok.creo.hu/openbsd/history/src/sys/arch/i386/i386/vm_machdep.c

So, I guess that the attached is the best to do, and I didn't
notice any breakage.  Of course, I don't even know what
triggers this code path.  :-)

Mark, perhaps you have a better fix in your local trees?

OK?

-- 
Pedro Alves

[-- Attachment #2: obsd_build_4.3.diff --]
[-- Type: text/x-diff, Size: 1775 bytes --]

2008-08-06  Pedro Alves  <pedro@codesourcery.com>

	* i386obsd-nat.c (i386obsd_supply_pcb): Remove handling of
	switchframe.sf_ppl != 0.

---
 gdb/i386obsd-nat.c |   21 ++++++---------------
 1 file changed, 6 insertions(+), 15 deletions(-)

Index: src/gdb/i386obsd-nat.c
===================================================================
--- src.orig/gdb/i386obsd-nat.c	2008-01-01 22:53:11.000000000 +0000
+++ src/gdb/i386obsd-nat.c	2008-08-06 16:00:41.000000000 +0100
@@ -61,21 +61,12 @@ i386obsd_supply_pcb (struct regcache *re
      checking if the saved interrupt priority level in the stack frame
      looks reasonable..  */
   read_memory (pcb->pcb_esp, (char *) &sf, sizeof sf);
-  if ((unsigned int) sf.sf_ppl < 0x100 && (sf.sf_ppl & 0xf) == 0)
-    {
-      /* Yes, we have a frame that matches cpu_switch().  */
-      pcb->pcb_esp += sizeof (struct switchframe);
-      regcache_raw_supply (regcache, I386_EDI_REGNUM, &sf.sf_edi);
-      regcache_raw_supply (regcache, I386_ESI_REGNUM, &sf.sf_esi);
-      regcache_raw_supply (regcache, I386_EBX_REGNUM, &sf.sf_ebx);
-      regcache_raw_supply (regcache, I386_EIP_REGNUM, &sf.sf_eip);
-    }
-  else
-    {
-      /* No, the pcb must have been last updated by savectx().  */
-      pcb->pcb_esp += 4;
-      regcache_raw_supply (regcache, I386_EIP_REGNUM, &sf);
-    }
+
+  pcb->pcb_esp += sizeof (struct switchframe);
+  regcache_raw_supply (regcache, I386_EDI_REGNUM, &sf.sf_edi);
+  regcache_raw_supply (regcache, I386_ESI_REGNUM, &sf.sf_esi);
+  regcache_raw_supply (regcache, I386_EBX_REGNUM, &sf.sf_ebx);
+  regcache_raw_supply (regcache, I386_EIP_REGNUM, &sf.sf_eip);
 
   regcache_raw_supply (regcache, I386_EBP_REGNUM, &pcb->pcb_ebp);
   regcache_raw_supply (regcache, I386_ESP_REGNUM, &pcb->pcb_esp);

Re: Fix building on x86 OpenBSD 4.3

[Mark Kettenis]

> From: Pedro Alves <pedro@codesourcery.com>
> Date: Wed, 6 Aug 2008 16:10:44 +0100
> 
> Hi,
> 
> In my testing on several targets, I noticed that GDB isn't
> building on x86 OpenBSD 4.3, due to the fact that struct
> switchframe got its sf_ppl member removed, and GDB is trying
> to access it.
> 
> I can't claim I know a think about this, but,
> 
> The patch that seems to have removed it, shows that is was always
> 0 just before removal:
> 
>  http://archive.netbsd.se/?ml=openbsd-tech&a=2007-09&m=5239543
> 
> And I can see it being set to != 0 in that files' history at:
>  http://opengrok.creo.hu/openbsd/history/src/sys/arch/i386/i386/vm_machdep.c
> 
> So, I guess that the attached is the best to do, and I didn't
> notice any breakage.  Of course, I don't even know what
> triggers this code path.  :-)
> 
> Mark, perhaps you have a better fix in your local trees?

Oops, yes I have.  Fixed this and never seemed to have committed the
fix when OpenBSD 4.3 was officially released.  Need to tweak it such
that the file compiles on older OpenBSD releases though.

Thanks for pointing this out.

Re: Fix building on x86 OpenBSD 4.3

[Mark Kettenis]

> From: Pedro Alves <pedro@codesourcery.com>
> Date: Wed, 6 Aug 2008 16:10:44 +0100
> 
> Hi,
> 
> In my testing on several targets, I noticed that GDB isn't
> building on x86 OpenBSD 4.3, due to the fact that struct
> switchframe got its sf_ppl member removed, and GDB is trying
> to access it.
> 
> I can't claim I know a think about this, but,
> 
> The patch that seems to have removed it, shows that is was always
> 0 just before removal:
> 
>  http://archive.netbsd.se/?ml=openbsd-tech&a=2007-09&m=5239543
> 
> And I can see it being set to != 0 in that files' history at:
>  http://opengrok.creo.hu/openbsd/history/src/sys/arch/i386/i386/vm_machdep.c
> 
> So, I guess that the attached is the best to do, and I didn't
> notice any breakage.  Of course, I don't even know what
> triggers this code path.  :-)
> 
> Mark, perhaps you have a better fix in your local trees?

Oops, yes I have.  Fixed this and never seemed to have committed the
fix when OpenBSD 4.3 was officially released.  Need to tweak it such
that the file compiles on older OpenBSD releases though.  I guess
that's what kept me from committing this.

Thanks for pointing this out, I'll fix it ASAP.

Mark

Re: Fix building on x86 OpenBSD 4.3

[Pedro Alves]

> > And I can see it being set to != 0 in that files' history at:

         ^^^ Of course, I meant *can't* here.  :-)

> http://opengrok.creo.hu/openbsd/history/src/sys/arch/i386/i386/vm_machdep.c

> Thanks for pointing this out, I'll fix it ASAP.

Thanks!

-- 
Pedro Alves

Re: Fix building on x86 OpenBSD 4.3

[Mark Kettenis]

> From: Pedro Alves <pedro@codesourcery.com>
> Date: Wed, 6 Aug 2008 16:32:57 +0100
> 
> > > And I can see it being set to != 0 in that files' history at:
> 
>          ^^^ Of course, I meant *can't* here.  :-)
> 
> > http://opengrok.creo.hu/openbsd/history/src/sys/arch/i386/i386/vm_machdep.c

Oh, it's well hidden.  It happens in assembly code in locore.s ;).

Re: Fix building on x86 OpenBSD 4.3

[Mark Kettenis]

> From: Pedro Alves <pedro@codesourcery.com>
> Date: Wed, 6 Aug 2008 16:32:57 +0100
> 
> > > And I can see it being set to != 0 in that files' history at:
> 
>          ^^^ Of course, I meant *can't* here.  :-)
> 
> > http://opengrok.creo.hu/openbsd/history/src/sys/arch/i386/i386/vm_machdep.c
> 
> > Thanks for pointing this out, I'll fix it ASAP.
> 
> Thanks!

I just comitted the attached patch.


Index: ChangeLog
from  Mark Kettenis  <kettenis@gnu.org>

	* i386obsd-nat.c (i386obsd_supply_pcb): Adjust for changes in
	OpenBSD 4.3.

Index: i386obsd-nat.c
===================================================================
RCS file: /cvs/src/src/gdb/i386obsd-nat.c,v
retrieving revision 1.11
diff -u -p -r1.11 i386obsd-nat.c
--- i386obsd-nat.c 1 Jan 2008 22:53:11 -0000 1.11
+++ i386obsd-nat.c 6 Aug 2008 19:34:38 -0000
@@ -60,10 +60,11 @@ i386obsd_supply_pcb (struct regcache *re
   /* Read the stack frame, and check its validity.  We do this by
      checking if the saved interrupt priority level in the stack frame
      looks reasonable..  */
-  read_memory (pcb->pcb_esp, (char *) &sf, sizeof sf);
-  if ((unsigned int) sf.sf_ppl < 0x100 && (sf.sf_ppl & 0xf) == 0)
+#ifdef PCB_SAVECTX
+  if ((pcb->pcb_flags & PCB_SAVECTX) == 0)
     {
       /* Yes, we have a frame that matches cpu_switch().  */
+      read_memory (pcb->pcb_esp, (char *) &sf, sizeof sf);
       pcb->pcb_esp += sizeof (struct switchframe);
       regcache_raw_supply (regcache, I386_EDI_REGNUM, &sf.sf_edi);
       regcache_raw_supply (regcache, I386_ESI_REGNUM, &sf.sf_esi);
@@ -71,9 +72,12 @@ i386obsd_supply_pcb (struct regcache *re
       regcache_raw_supply (regcache, I386_EIP_REGNUM, &sf.sf_eip);
     }
   else
+#endif
     {
       /* No, the pcb must have been last updated by savectx().  */
-      pcb->pcb_esp += 4;
+      pcb->pcb_esp = pcb->pcb_ebp;
+      pcb->pcb_ebp = read_memory_integer(pcb->pcb_esp, 4);
+      sf.sf_eip = read_memory_integer(pcb->pcb_esp + 4, 4);
       regcache_raw_supply (regcache, I386_EIP_REGNUM, &sf);
     }