Re: [PATCH] set membase in serial8250_request_port

[Russell King]

On Mon, Oct 04, 2004 at 12:54:22PM -0600, Alex Williamson wrote:
> 
>    I'm running into a problem that seems to be caused by this really old
> changeset:
> 
> http://linux.bkbits.net:8080/linux-2.5/cset@3d9f67f2BWvXiLsZCFwD-8s_E9AN6A
> 
> When I run 'setserial /dev/ttyS1 uart 16450' on an ia64 system w/ MMIO
> UARTs, I get a NAT consumption oops from the kernel.  The problem is
> that this code path calls serial8250_release_port() where the membase
> gets cleared.  However, the subsequent call to serial8250_request_port()
> doesn't restore membase, causing a read from a bad address.  I don't see
> many users of the UPF_IOREMAP flag, so I think the solution is to simply
> make the remap case symmetric to the unmap case.  Patch below.  Thanks,

Mostly correct reasoning, but the solution is wrong.  Consider what
happens if we call request_port where we have set mapbase and pre-
initialised membase for a memory mapped port (eg, PCI card.)

This would cause us to re-ioremap the mapbase, which is wrong.  We
must obey the UPF_IOREMAP flag here.  Note also that this fix you're
reverting will break 8250 for PPC people...

Could you give further information about the problem you're seeing?
Bear in mind that I know precisely zero about ia64 oopsen so you'll
probably have to explain it to me in detail.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/