Re: 2.5.45 odd deref in serial_in

[Russell King]

On Mon, Nov 04, 2002 at 11:27:28PM -0500, Zwane Mwaikambo wrote:
> 0xc023b428 <serial_in+24>:      je     0xc023b461 <serial_in+81>
> 0xc023b42a <serial_in+26>:      cmp    $0x2,%eax
> 0xc023b42d <serial_in+29>:      je     0xc023b440 <serial_in+48>
> 0xc023b42f <serial_in+31>:      mov    0x8(%ebx),%eax
> 0xc023b432 <serial_in+34>:      add    %eax,%edx
> 0xc023b434 <serial_in+36>:      in     (%dx),%al
> 
> eax: 00000000   ebx: 81acc5f0   ecx: 00000000   edx: 00000005
> 
> ...
> 	default:
> 		return inb(up->port.iobase + offset); <--
> 	}

Ok, if I'm reading this correctly:

offset = %edx
up->port.iobase = 0x8(%ebx)
up = %ebx

To get to this return statement, we would have had to execute:

static _INLINE_ unsigned int serial_in(struct uart_8250_port *up, int offset)
{
        offset <<= up->port.regshift;

        switch (up->port.iotype) {

which also dereferences "up".  So something may have corrupted %ebx
between executing that switch statement and executing the inb().

Could the NMI handler be corrupting %ebx ?

-- 
Russell King ([e-mail])                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html