Re: MC Bus Error / tcp_ack_saw_tstamp Was: IP28 Installation Success report

[tsbogend@alpha.franken.de (Thomas Bogendoerfer)]

On Thu, Dec 27, 2007 at 10:04:01AM +0100, Florian Lohoff wrote:
> On Sun, Dec 23, 2007 at 08:54:42PM +0100, Florian Lohoff wrote:
> > I thought an installation success report is sometimes nice to have:
> 
> Linux ip28 2.6.24-rc5-g8b3ba06b-dirty #21 Tue Dec 18 12:48:29 CET 2007 mips64 GNU/Linux
> 
> After ~10 days uptime and multiple gcc builds. Logging in via ssh 
> and issueing an "ls" in the build directory:
> 
> MC Bus Error

looks like

        /* GIO errors are fatal */
        if (gio_err_stat & GIO_ERRMASK)
                goto mips_be_fatal;

in ip28_be_interrupt is too strict. Load speculations to addresses
in the GIO address range where no device responds, will probably always
produce bus timeout. It might be worth to use 
check_addr_in_insn(cpu_err_addr, regs)) before killing the machine.

So

        if ((gio_err_stat & GIO_ERRMASK) &&
             check_addr_in_insn(gio_err_addr, regs))
                        goto mips_be_fatal;

might be better.

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessary a
good idea.                                                [ RFC1925, 2.3 ]