GDB Agent for RTEMS on new powerpc/virtex BSP

[Robert S. Grimes]

Hi,

I'm trying to use Till Straumann's "GDB Agent for RTEMS" working on
RTEMS-4.7.99.2, using the powerpc/virtex BSP.  This BSP supports Xilinx
Virtex-4 and its embedded PPC-403 PowerPC.  This PPC/BSP combination
uses the "new exceptions" model, but has no floating point unit.  My
application is being developed using C++.

With a bit of help, I've been able to build a suitably patched gdb, and
with some help elsewhere, I've even got Insight to recognize the new
rtems-remote target.

I've also been able to build the agent, but that's about it.  There were
some major issues, mainly centered around the lack of FP.  But I did get
it to compile; attached is a patch file describing the changes.  (Note:
yes, a better symbol could, and should, be chosen).

When I attempt to connect, however, I get an exception 0.  Here is the
gdb session:

  (gdb) target rtems-remote 192.168.0.69:2159
  Remote debugging using 192.168.0.69:2159
  [New Thread  a010006]
  warning: no shared library support for this OS / ABI
  [Switching to Thread  a010006]
  helper_thread (Ignoring packet error, continuing...
  arg=Reply contains invalid hex digit 116
  ) at rtems-stub.c:621
  621                     sleep(1);
  Ignoring packet error, continuing...
  Ignoring packet error, continuing...
  warning: remote target does not support CRC operation - skipping...
  Remote communication error: Connection reset by peer.
  (gdb)

Meanwhile, the serial console shows this, which is displayed immediately
upon entry of the "target rtems-remote ..." command; the remainder of
the gdb output is displayed, a line or two at a time, a second or two apart.

  exception handler called for exception 0
       Next PC or Address of fault = C0276922
       Saved MSR = 61000000
       R0 = 0
       R1 = 0
       R2 = 0
       R3 = 0
       R4 = 0
       R5 = 0
       R6 = 0
       R7 = 0
       R8 = 0
       R9 = 0
       R10 = 0
       R11 = 0
       R12 = 0
       R13 = FFFE09
       R14 = 54FFFE02
       R15 = 40FFFE01
       R16 = D0000000
       R17 = 800000
       R18 = 1000000
       R19 = 0
       R20 = 0
       R21 = 0
       R22 = 0
       R23 = 3000000
       R24 = 1FFFFFF
       R25 = FF000000
       R26 = 0
       R27 = 0
       R28 = 0
       R29 = 820401
       R30 = 0
       R31 = 820C01
       CR = 0
       CTR = 3821401
       XER = 0
       LR = 821C01
       MSR = 0
       DAR = BA8C2B
  Stack Trace:
    IP: 0xC0276922, LR: 0x00821C01
  --^ 0x00000000

Here are the differences:

--- RTEMS_gdb_stub_1.4/rtems-gdb-stub-ppc-shared.c    Wed Sep 05
19:09:14 2007
+++ RTEMS_gdb_stub_1.4/rtems-gdb-stub-ppc-shared.rsg.c    Wed Sep 05
19:11:20 2007
@@ -123,11 +123,13 @@
     }
 
     if ( (tcb = get_tcb(msg->tid)) ) {
+#ifdef  RSG_USE_FP
         Context_Control_fp *fpc = tcb->fp_context;
         if ( fpc ) {
             memcpy(buf + FPR0_OFF, &fpc->f[0], 32*8 );
             memcpy(buf + FPSCR_OFF, &fpc->fpscr, 4);
         }
+#endif
         if (!f) {
             Frame        sfr = (Frame)tcb->Registers.gpr1;
             unsigned pcdummy = tcb->Registers.pc - 4;
@@ -162,12 +164,14 @@
     }
 
     if ( msg->tid && (tcb = get_tcb(msg->tid)) ) {
+#ifdef  RSG_USE_FP
         Context_Control_fp *fpc = tcb->fp_context;
         if ( fpc ) {
             memcpy(&fpc->f[0],   buf+FPR0_OFF,    32*8 );
             memset(&fpc->fpscr,  0,               sizeof(fpc->fpscr));
             memcpy(&fpc->fpscr,  buf + FPSCR_OFF, 4);
         }
+#endif
         if ( !f ) {
             /* setup TCB */
             memcpy(&tcb->Registers.gpr1,  buf + (GPR0_OFF + 4),        
2 *4);
@@ -240,16 +244,20 @@
 
     switch ( f->_EXC_number ) {
         case ASM_MACH_VECTOR     :
+#ifdef  RSG_USE_FP
             _BSP_clear_hostbridge_errors(1,0);
+#endif
             msg.sig = SIGBUS;
         break;
 
         case ASM_PROT_VECTOR     :
         case ASM_ISI_VECTOR      :
         case ASM_ALIGN_VECTOR    :
+#ifdef  RSG_USE_FP
         case ASM_IMISS_VECTOR    :
         case ASM_DLMISS_VECTOR   :
         case ASM_DSMISS_VECTOR   :
+#endif
             msg.sig = SIGSEGV;
         break;
 

I suspect my changes are way too simple-minded, but maybe I'm not
totally out in left field.  The first two changes skip copying of
non-existent FP registers (I think).  The ...hostbridge... call I don't
have a concrete clue about, but I am guessing that isn't necessary for
my BSP (could be wrong, of course!).  The last change - deleting the
undefined vectors in the switch statement - I would not expect to be a
problem.

With a lot of work, I can problem get this figured out, but I'm hoping
someone more expert than me might have some clues to save me some grief.

Thanks!
-Bob

BTW: It looks like Insight is going to work too, assuming I can get this
problem resolved, as it fails in the exact same way!  One of those
not-so-rare situations where a failure actually makes you feel good!  ;-)