Inlining of certain functions prevent R3000 kernel from booting...

["Steven J. Hill" <sjhill@cotw.com>]

Hello.

While trying to compile the latest CVS kernel, I notice that by not
inlining certain functions my kernel boots fine. If I put the inlines
back in, the kernel hangs and does not boot. Below are the three
functions that I am unable to inline:

****************************************************************
diff -urN -X cvs-exc.txt nino-linux/arch/mips/kernel/traps.c /opt/sgi-cvs/linux-
2.4/linux/arch/mips/kernel/traps.c
--- nino-linux/arch/mips/kernel/traps.c Wed Dec 12 17:42:53 2001
+++ /opt/sgi-cvs/linux-2.4/linux/arch/mips/kernel/traps.c       Sun Dec  2 05:34
:38 2001
@@ -100,7 +100,7 @@
 
 static struct task_struct *ll_task = NULL;
 
-static void simulate_ll(struct pt_regs *regp, unsigned int opcode)
+static inline void simulate_ll(struct pt_regs *regp, unsigned int opcode)
 {
        unsigned long value, *vaddr;
        long offset;
@@ -141,7 +141,7 @@
                send_sig(signal, current, 1);
 }
 
-static void simulate_sc(struct pt_regs *regp, unsigned int opcode)
+static inline void simulate_sc(struct pt_regs *regp, unsigned int opcode)
 {
        unsigned long *vaddr, reg;
        long offset;
diff -urN -X cvs-exc.txt nino-linux/arch/mips/kernel/unaligned.c /opt/sgi-cvs/li
nux-2.4/linux/arch/mips/kernel/unaligned.c
--- nino-linux/arch/mips/kernel/unaligned.c     Wed Dec 12 17:47:27 2001
+++ /opt/sgi-cvs/linux-2.4/linux/arch/mips/kernel/unaligned.c   Sat Dec  1 00:04
:49 2001
@@ -96,7 +96,7 @@
        if ((long)(~(pc) & ((a) | ((a)+(s)))) < 0)      \
                goto sigbus;
 
-static void emulate_load_store_insn(struct pt_regs *regs,
+static inline void emulate_load_store_insn(struct pt_regs *regs,
                                            unsigned long addr, unsigned long pc
)
 {
        union mips_instruction insn;
****************************************************************

I attempted to do a 'objdump -d' on the kernels with and without
the inline statements to see if any registers might be getting
hammered. I was not able to see anything obvious, but I did find
it difficult to decipher things once inlining was enabled. I have
placed the unstripped kernels in a tarball at:

    ftp://ftp.cotw.com/MIPS/mips-r3k-inline-kernels.tar.bz2

The cross toolchain was made up of:

    binutils-2.11.92.0.10
    gcc-3.0.2
    glibc-2.2.3

Is anyone else having problems like this? I am doing the Philips Nino port
which has a PR31700 (R3000A core) chip in it. Thanks.

-Steve

-- 
 Steven J. Hill - Embedded SW Engineer