Re: Dynamic nop selection breaks boot on Geode LX

[Steven Rostedt <rostedt@goodmis.org>]

On Sun, 2010-10-03 at 17:32 +0100, Daniel Drake wrote:
> On 3 October 2010 06:50, Borislav Petkov <bp@alien8.de> wrote:
> > maybe because the Geode doesn't have both the
> > P6_NOP5 and the NOP with 4 0x66 prefixes:
> > http://kerneltrap.org/mailarchive/linux-kernel/2010/8/27/4612336
> >
> > and for some reason the trap can't find the fixup address. You say
> > "hangs" so you don't even get an "invalid opcode" OOPS?
> 
> The XO doesn't have standard VGA, so it is difficult to debug such
> early crashes. This is crashing so early that kernel messages don't
> even start to get sent over serial.
> 
> To debug these things, I checkpoint the code with various calls which
> send individual characters over serial:
> 
> static void log_serial(char c)
> {
>    while ((inb(0x3fd) & 0x20) == 0) ;
>    outb(c, 0x3f8);
>    while ((inb(0x3fd) & 0x40) == 0) ;
> }

Did you try earlyprintk? That is, on the kernel command line add:

	earlyprintk=ttyS0,115200

or whatever the tty and baud rate is. This will start printing out the
serial right at kernel startup. Even before printk is configured.

-- Steve

> 
> So, I'm not really sure if/how it crashed or oops'd. However, I can
> confirm that panic() does not get reached, since I put a character log
> in there and it doesn't get sent. Let me know if you want me to put
> character logging in other places.
> 
> I applied your two patches by hand and it doesn't solve the issue,
> because the init_amd_k6() code is called long after
> arch_init_ideal_nop5()
>