About PLAT_TRAMPOLINE_STUFF_LINE

["Nori, Soma Sekhar" <nsekhar@xxxxxx>]

Hi,

I am porting 2.6.10 (kernel.org) onto a 4kec based board.

What should be the value of PLAT_TRAMPOLINE_STUFF_LINE 
(include/asm-mips/cpu-features.h) for 4kec?

If I do not define a cpu-features-overrides.h for my board, this macro
is 
getting set to 0 and as a result signalling code in kernel 
(arch/mips/kernel/signal.c) seems to break. 
All my userspace programs using signals are seg faulting.

Here is the faulting code:

<code>
static void inline setup_frame(struct k_sigaction * ka, struct pt_regs
*regs,
	int signr, sigset_t *set)
{
	struct sigframe *frame;
	int err = 0;

	frame = get_sigframe(ka, regs, sizeof(*frame));
	if (!access_ok(VERIFY_WRITE, frame, sizeof (*frame)))
		goto give_sigsegv;

	/*
	 * Set up the return code ...
	 *
	 *         li      v0, __NR_sigreturn
	 *         syscall
	 */
	if (PLAT_TRAMPOLINE_STUFF_LINE)
		__builtin_memset(frame->sf_code, '0',
		                 PLAT_TRAMPOLINE_STUFF_LINE);
	err |= __put_user(0x24020000 + __NR_sigreturn, frame->sf_code +
0);
	err |= __put_user(0x0000000c                 , frame->sf_code +
1);
	flush_cache_sigtramp((unsigned long) frame->sf_code);

	err |= setup_sigcontext(regs, &frame->sf_sc);
	err |= __copy_to_user(&frame->sf_mask, set, sizeof(*set));
	if (err)
		goto give_sigsegv;
    
    ...
</code>    
        

With PLAT_TRAMPOLINE_STUFF_LINE set to 0, get_sigframe always returns 0
and setup_frame sends a SIGSEGV because __put_user returns an error
value.

When I override the value of PLAT_TRAMPOLINE_STUFF_LINE to 16 (the cache
line size of the 4kec), the signalling code seems to work just fine. 
(None of my userspace programs crash anymore).

This macro does not seem to be overridden for most boards, so 0 must be 
a valid value for atleast some MIPS CPUs. 

Am I right in changing the value of PLAT_TRAMPOLINE_STUFF_LINE to 16?

A google search on this macro did not yield much information.

Any help/pointers regarding this is greatly appreciated.

Thanks,
Sekhar Nori.