[1/9] macintosh/via-macii: Access autopoll_devs when inside lock

The interrupt handler should be excluded when accessing the autopoll_devs
variable.

Fixes: d95fd5fce88f0 ("m68k: Mac II ADB fixes") # v5.0+
Tested-by: Stan Johnson <userm57@yahoo.com>
Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
---
 drivers/macintosh/via-macii.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

Hi,

On Sun, Jun 28, 2020 at 02:23:12PM +1000, Finn Thain wrote:
> The interrupt handler should be excluded when accessing the autopoll_devs
> variable.
> 

I am quite baffled by this patch. Other than adding an unnecessary lock /
unlock sequence, accessing a variable (which is derived from another
variable) from inside or outside a lock does not make a difference.
If autopoll_devs = devs & 0xfffe is 0 inside the lock, it will just
as much be 0 outside the lock, and vice versa.

Can you explain this in some more detail ? Not that is matters much since
the change already made it into mainline, but I would like to understand
what if anything I am missing here.

Thanks,
Guenter

> Fixes: d95fd5fce88f0 ("m68k: Mac II ADB fixes") # v5.0+
> Tested-by: Stan Johnson <userm57@yahoo.com>
> Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
> ---
>  drivers/macintosh/via-macii.c | 9 +++------
>  1 file changed, 3 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/macintosh/via-macii.c b/drivers/macintosh/via-macii.c
> index ac824d7b2dcfc..6aa903529570d 100644
> --- a/drivers/macintosh/via-macii.c
> +++ b/drivers/macintosh/via-macii.c
> @@ -270,15 +270,12 @@ static int macii_autopoll(int devs)
>  	unsigned long flags;
>  	int err = 0;
>  
> +	local_irq_save(flags);
> +
>  	/* bit 1 == device 1, and so on. */
>  	autopoll_devs = devs & 0xFFFE;
>  
> -	if (!autopoll_devs)
> -		return 0;
> -
> -	local_irq_save(flags);
> -
> -	if (current_req == NULL) {
> +	if (autopoll_devs && !current_req) {
>  		/* Send a Talk Reg 0. The controller will repeatedly transmit
>  		 * this as long as it is idle.
>  		 */

On Sun, 9 Aug 2020, Guenter Roeck wrote:

> Hi,
> 
> On Sun, Jun 28, 2020 at 02:23:12PM +1000, Finn Thain wrote:
> > The interrupt handler should be excluded when accessing the 
> > autopoll_devs variable.
> > 
> 
> I am quite baffled by this patch. Other than adding an unnecessary lock 
> / unlock sequence,

The new lock/unlock sequence means that the expression (autopoll_devs && 
!current_req) can be understood to be atomic. That makes it easier for me 
to follow (being that both variables are shared state).

> accessing a variable (which is derived from another variable) from 
> inside or outside a lock does not make a difference. If autopoll_devs = 
> devs & 0xfffe is 0 inside the lock, it will just as much be 0 outside 
> the lock, and vice versa.
> 
> Can you explain this in some more detail ? Not that is matters much 
> since the change already made it into mainline, but I would like to 
> understand what if anything I am missing here.
> 

I think the new code is more readable and is obviously free of race 
conditions. It's not obvious to me why the old code was free of race 
conditions but if you can easily establish that by inspection then you are 
a better auditor than I am. Regardless, I'll stick with "Keep It Simple, 
Stupid".