cpm_uart_port_map not initialised before serial console setup

cpm_uart_port_map not initialised before serial console setup

[Laurent Pinchart]

[-- Attachment #1: Type: text/plain, Size: 593 bytes --]

Hi everybody,

while trying to use SCC1 as a serial console, I found a bug in the cpm_uart 
driver.

The cpm_uart_port_map table is initialised by cpm_uart_count() which is called 
in cpm_uart_init() at module_init() time. cpm_uart_console_setup(), called at 
console_initcall() time, accesses cpm_uart_port_map, leading to a crash when 
using any serial port except SMC1 as the serial console.

I attached a very simple patch to fix the problem, but it might be subject to 
race conditions. Could anyone familiar with the cpm_uart driver have a look 
at it ?

Best regards,

Laurent Pinchart

[-- Attachment #2: cpm_uart.patch --]
[-- Type: text/x-diff, Size: 403 bytes --]

diff --git a/drivers/serial/cpm_uart/cpm_uart_core.c b/drivers/serial/cpm_uart/cpm_uart_core.c
index a632897..a3c30f5 100644
--- a/drivers/serial/cpm_uart/cpm_uart_core.c
+++ b/drivers/serial/cpm_uart/cpm_uart_core.c
@@ -1240,6 +1240,7 @@ static struct console cpm_scc_uart_conso
 
 int __init cpm_uart_console_init(void)
 {
+	cpm_uart_count();
 	register_console(&cpm_scc_uart_console);
 	return 0;
 }

Re: cpm_uart_port_map not initialised before serial console setup

[Vitaly Bordug]

On Thu, 13 Jul 2006 16:55:35 +0200
Laurent Pinchart <laurent.pinchart@tbox.biz> wrote:

> Hi everybody,
> 
> while trying to use SCC1 as a serial console, I found a bug in the cpm_uart 
> driver.
> 
> The cpm_uart_port_map table is initialised by cpm_uart_count() which is called 
> in cpm_uart_init() at module_init() time. cpm_uart_console_setup(), called at 
> console_initcall() time, accesses cpm_uart_port_map, leading to a crash when 
> using any serial port except SMC1 as the serial console.
> 
> I attached a very simple patch to fix the problem, but it might be subject to 
> race conditions. Could anyone familiar with the cpm_uart driver have a look 
> at it ?
> 
Ugh, you're right.

But I think we need to finally get rid of the stupid count/port_map creation based on #ifdefs.
I'll have a look at it immediately as time permits
 

-- 
Sincerely, 
Vitaly