Re: [PATCH v3 3/3] counter: 104-quad-8: Utilize helper functions to handle PR, FLAG and PSC

[William Breathitt Gray]

On Tue, Apr 11, 2023 at 05:43:49PM +0300, Andy Shevchenko wrote:
> On Tue, Apr 11, 2023 at 10:05:25AM -0400, William Breathitt Gray wrote:
> > On Tue, Apr 11, 2023 at 04:50:03PM +0300, Andy Shevchenko wrote:
> > > On Mon, Apr 10, 2023 at 10:03:13AM -0400, William Breathitt Gray wrote:
> > > > The Preset Register (PR), Flag Register (FLAG), and Filter Clock
> > > > Prescaler (PSC) have common usage patterns. Wrap up such usage into
> > > > dedicated functions to improve code clarity.
> 
> ...
> 
> > > >  	*val = 0;
> > > 
> > > Is not needed now as always being initialized by below call.
> > 
> > The regmap_noinc_read() call only reads the number of bytes requested.
> > Since we request 3 bytes, the upper bytes of the u64 val remain
> > uninitialized, so that is why we need to set *val = 0. This isn't
> > immediately clear in the code, so I can add a comment to make it
> > explicit.
> 
> Hmm...
> Since we are using byte array for the value, can we actually use
> memset() to show that explicitly? Perhaps in that way it will be more visible?
> 
> > > >  	spin_lock_irqsave(&priv->lock, irqflags);
> > > >  
> > > >  	iowrite8(SELECT_RLD | RESET_BP | TRANSFER_CNTR_TO_OL, &chan->control);
> > > > -
> > > > -	for (i = 0; i < 3; i++)
> > > > -		*val |= (unsigned long)ioread8(&chan->data) << (8 * i);
> > > > +	ioread8_rep(&chan->data, val, 3);
> 
> But hold on, wouldn't this have an endianess issue? The call fills in LE,
> while here you use the CPU order.
> 
> > > >  	spin_unlock_irqrestore(&priv->lock, irqflags);
> 
> That said, I think you should have something like
> 
> 	u8 value[3];
> 
> 	ioread8_rep(..., value, sizeof(value));
> 
> 	*val = get_unaligned_le24(value);
> 
> -- 
> With Best Regards,
> Andy Shevchenko

Yes, I think you're right, that solves both problems at once so I'll use
get_aligned_le24() to set *val.

William Breathitt Gray

Attachment: signature.asc
Description: PGP signature