problem in serial_ns16550.c

[Jean-Christophe PLAGNIOL-VILLARD]

On 17:52 Wed 03 Aug     , Sascha Hauer wrote:
> On Wed, Aug 03, 2011 at 07:35:18PM +0400, Antony Pavlov wrote:
> > Hi!
> > 
> > IMHO there is a problem in ns16550_probe() (see
> > drivers/serial/serial_ns16550.c:243).
> > 
> > There is the construction:
> > ------
> > if (!(dev->resource[0].flags & IORESOURCE_MEM_TYPE_MASK) &&
> >             ((plat->reg_read == NULL) || (plat->reg_write == NULL)))
> >                 return -EINVAL;
> > ------
> > 
> > Imagine creation of typical serial port:
> > ------
> > static struct NS16550_plat plat = {
> >         .clock = 1843200,
> > };
> > 
> > ...
> > 
> > add_ns16550_device(-1, UART_ADDR, 8, IORESOURCE_MEM_8BIT, &plat);
> > ------
> > 
> > Here we have plat.reg_read == NULL, plat.reg_write == NULL.
> > Usage of add_ns16550_device will make
> > dev->resource[0].flags & IORESOURCE_MEM_TYPE_MASK == IORESOURCE_MEM_8BIT.
> > 
> > But take into account this (see include/linux/ioport.h):
> > ------
> > #define IORESOURCE_MEM_TYPE_MASK        (3<<3)
> > #define IORESOURCE_MEM_8BIT             (0<<3)
> > ------
> > 
> > So IORESOURCE_MEM_8BIT == 0 (sic!)
> > 
> > A son tour, !(dev->resource[0].flags & IORESOURCE_MEM_TYPE_MASK) give
> > true, if flags select IORESOURCE_MEM_8BIT.
> > 
> > As a result, if add_ns16550_device() take IORESOURCE_MEM_8BIT, and
> > plat->reg_read == NULL, plat->reg_write == NULL then ns16550_probe()
> > will return -EINVAL.
> 
> Ok, then we have to check for the existence of plat->reg_read/write
> first. If the exist, we have to skip further IORESOURCE_MEM_xBIT checks.
> If they don't exist we do whatever IORESOURCE_MEM_xBIT indicates.
> 
> That only leaves the problem that if a user registers a ns16550 driver
> without reg_read/write and does not care about resource types either
> the driver defaults to 8 bit mmio accesses, but I think we can live with
> that.
that was my idea at the beginning

Best Regards,
J.