patch for _2_4_devel KGDB in arch/ppc/kernel/ppc-stub.c

The following patch fixes 2 problems with KGDB:

a) On line 617, '&regs' should be 'regs' in the call to mem2hex().  The
resulting improper response packet confuses the GDB client resulting in
step/next and stepi/nexti GDB commands to "lose their grip" and allow
execution to continue unchecked.

b) When executing an 's' command, we don't want to start single stepping
until we *return* from the current exception that we're handling.
Therefore, the MSR_SE bit should not be set in the MSR until the rfi
from this exception is executed (i.e., only need to set that bit in
'regs->msr' and not do an actual 'mtmsr()' at this point).  If the
'mtmsr(msr | MSR_SE);' line is executed at this point, a single-step
exeption will occur inside the exception handler itself which causes
another exception and the 'printk(KERN_ERR "interrupt while in kgdb,
returning\n");' line at the beginning of handle_exception() to be
executed.  Not fatal but somewhat disturbing to see.

Mark
--


===== arch/ppc/kernel/ppc-stub.c 1.22 vs edited =====

--- 1.22/arch/ppc/kernel/ppc-stub.c     Fri Feb 28 10:26:48 2003
+++ edited/arch/ppc/kernel/ppc-stub.c   Tue Mar 11 16:55:19 2003

@@ -614,7 +614,7 @@
        *ptr++ = hexchars[SP_REGNUM >> 4];
        *ptr++ = hexchars[SP_REGNUM & 0xf];
        *ptr++ = ':';
-       ptr = mem2hex(((char *)&regs) + SP_REGNUM*4, ptr, 4);
+       ptr = mem2hex(((char *)regs) + SP_REGNUM*4, ptr, 4);
        *ptr++ = ';';
 #endif

@@ -795,7 +795,11 @@
                        mtmsr(msr);
 #else
                        regs->msr |= MSR_SE;
+                       /* Don't write MSR now, wait for exception return.
+                        * Otherwise, we'll start single stepping

immediately.
+                        * --MAG
                        mtmsr(msr | MSR_SE);
+                        */
 #endif
                        unlock_kernel();
                        kgdb_active = 0;


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/