On Mon, Mar 03, 2003 at 07:21:37PM +0900, Yoichi Yuasa wrote:
> I found a bug in simulate_ll and simulate_sc.
> The board that uses ll/sc emulation is not started.
>
> When send_sig in these, in order not to operate the value of EPC correctly,
> simulate_* happens continuously.
>
> The previous patches were not perfect, I changed more.
> Please apply these patches to CVS tree.
As previously mentioned there were some problems with your fix, so I
wrote an alternative fix which is attached below. It's untested because
I don't have any ll/sc-less test platform.
Ralf
Index: arch/mips/kernel/traps.c
===================================================================
RCS file: /home/cvs/linux/arch/mips/kernel/traps.c,v
retrieving revision 1.99.2.41
diff -u -r1.99.2.41 traps.c
--- arch/mips/kernel/traps.c 10 Feb 2003 22:50:48 -0000 1.99.2.41
+++ arch/mips/kernel/traps.c 4 Mar 2003 20:32:10 -0000
@@ -134,13 +134,14 @@
ll_bit = 0;
}
ll_task = current;
+
regs->regs[(opcode & RT) >> 16] = value;
compute_return_epc(regs);
return;
sig:
- send_sig(signal, current, 1);
+ force_sig(signal, current);
}
static inline void simulate_sc(struct pt_regs *regs, unsigned int opcode)
@@ -172,19 +173,21 @@
}
if (ll_bit == 0 || ll_task != current) {
regs->regs[reg] = 0;
- goto sig;
+ return;
}
- if (put_user(regs->regs[reg], vaddr))
+ if (put_user(regs->regs[reg], vaddr)) {
signal = SIGSEGV;
- else
- regs->regs[reg] = 1;
+ goto sig;
+ }
+
+ regs->regs[reg] = 1;
compute_return_epc(regs);
return;
sig:
- send_sig(signal, current, 1);
+ force_sig(signal, current);
}
/*
