Message-ID: <20250407194937.097b7709@jic23-huawei>
Date: Mon, 7 Apr 2025 19:49:37 +0100
From: Jonathan Cameron <jic23@...nel.org>
To: Matti Vaittinen <mazziesaccount@...il.com>
Cc: David Lechner <dlechner@...libre.com>, Matti Vaittinen
 <matti.vaittinen@...rohmeurope.com>, Lars-Peter Clausen <lars@...afoo.de>,
 Rob Herring <robh@...nel.org>, Krzysztof Kozlowski <krzk+dt@...nel.org>,
 Conor Dooley <conor+dt@...nel.org>, Nuno Sa <nuno.sa@...log.com>, Javier
 Carrasco <javier.carrasco.cruz@...il.com>, linux-iio@...r.kernel.org,
 devicetree@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 2/7] iio: adc: ti-adc128s052: Simplify using
 be16_to_cpu()

On Mon, 7 Apr 2025 08:23:07 +0300
Matti Vaittinen <mazziesaccount@...il.com> wrote:

> On 05/04/2025 20:29, Jonathan Cameron wrote:
> > On Thu, 3 Apr 2025 08:16:43 +0300
> > Matti Vaittinen <mazziesaccount@...il.com> wrote:
> >   
> >> On 03/04/2025 00:04, David Lechner wrote:  
> >>> On 4/2/25 1:09 AM, Matti Vaittinen wrote:  
> >>>> The register data is 12-bit big-endian data. Use be16_to_cpu() to do
> >>>> the conversion, and simple bitwise AND for masking to make it more
> >>>> obvious.
> >>>>
> >>>> Signed-off-by: Matti Vaittinen <mazziesaccount@...il.com>
> >>>> ---
> >>>> Revision history:
> >>>> v1 => v2:
> >>>>    - Fix commit msg to reflect the fact there was no bug
> >>>>    - Drop Fixes tag
> >>>>    - Use union for rx / tx buffer to avoid casting
> >>>>    - Keep the shared message protected by the mutex
> >>>> ---
> >>>>    drivers/iio/adc/ti-adc128s052.c | 18 ++++++++++--------
> >>>>    1 file changed, 10 insertions(+), 8 deletions(-)
> >>>>
> >>>> diff --git a/drivers/iio/adc/ti-adc128s052.c b/drivers/iio/adc/ti-adc128s052.c
> >>>> index a456ea78462f..3e69a5fce010 100644
> >>>> --- a/drivers/iio/adc/ti-adc128s052.c
> >>>> +++ b/drivers/iio/adc/ti-adc128s052.c
> >>>> @@ -28,32 +28,34 @@ struct adc128 {
> >>>>    	struct regulator *reg;
> >>>>    	struct mutex lock;
> >>>>    
> >>>> -	u8 buffer[2] __aligned(IIO_DMA_MINALIGN);
> >>>> +	union {
> >>>> +		__be16 rx_buffer;
> >>>> +		u8 tx_buffer[2];  
> > As below. Maybe
> > 		__be16 buffer16;
> > 		u8 buffer[2];  
> 
> Ok.
> 
> >>>> +	} __aligned(IIO_DMA_MINALIGN);
> >>>>    };
> >>>>    
> >>>>    static int adc128_adc_conversion(struct adc128 *adc, u8 channel)
> >>>>    {
> >>>>    	int ret;
> >>>> +	char *msg = &adc->tx_buffer[0];
> >>>>    
> >>>>    	mutex_lock(&adc->lock);
> >>>>    
> >>>> -	adc->buffer[0] = channel << 3;
> >>>> -	adc->buffer[1] = 0;
> >>>> +	msg[0] = channel << 3;
> >>>> +	msg[1] = 0;
> >>>>    
> >>>> -	ret = spi_write(adc->spi, &adc->buffer, 2);
> >>>> +	ret = spi_write(adc->spi, msg, sizeof(adc->tx_buffer));  
> > 
> > I'd get rid of msg as it's now just confusing given we are
> > using the sizeof() here.  
> 
> Ok.
> 
> >>>>    	if (ret < 0) {
> >>>>    		mutex_unlock(&adc->lock);
> >>>>    		return ret;
> >>>>    	}
> >>>>    
> >>>> -	ret = spi_read(adc->spi, &adc->buffer, 2);
> >>>> -
> >>>> +	ret = spi_read(adc->spi, &adc->rx_buffer, 2);  
> > 
> > sizeof(adc->rx_buffer)  
> 
> I was thinking of this but went with raw 2 - because we need to read 
> exactly 2 bytes from the device. Sizeof buffer is matter of software 
> where as the 2 bytes is dictated by the device. (Sure the size of buffer 
> needs to be large enough).
> 
> I don't care it that much though, so I can go with the sizeof() if 
> that's what you prefer. Just explaining that the '2' here was a 
> conscious choice :)

Hmm. If we have a case where we read less than 2 bytes into that buffer
then fair enough.  Otherwise it's correctly sized so having sizeof(buffer)
and having to check that size in only one place is a tiny bit preferable.


> 
> >>>>    	mutex_unlock(&adc->lock);
> >>>> -
> >>>>    	if (ret < 0)
> >>>>    		return ret;
> >>>>    
> >>>> -	return ((adc->buffer[0] << 8 | adc->buffer[1]) & 0xFFF);
> >>>> +	return be16_to_cpu(adc->rx_buffer) & 0xFFF;  
> >>>
> >>>
> >>> The cast isn't exactly beautiful, but this would save a lot of
> >>> lines of diff and a few lines of code by avoiding the need for
> >>> the union and the local msg variable.
> >>>
> >>> 	return be16_to_cpup((__be16 *)adc->buffer) & 0xFFF;  
> > 
> > The cast only works because we have forced the alignment for DMA safety.
> > That to me is a little fragile.
> > 
> > You could do get_unaligned_be16() which doesn't need the cast then carry
> > on using the original buffer.  
> >>
> >> Thanks again for the review David :)
> >>
> >> I am unsure which way to go. I kind of like having the __be16 in the
> >> struct, as it immediately yells "data from device is big-endian". OTOH,
> >> I've never loved unions (and, it silences the above "yelling" quite a
> >> bit). I still think this might be the first time I really see a valid
> >> use-case for an union :) And, you're right this adds more lines,
> >> besides, the cast doesn't look that ugly to me. Yet, I originally had a
> >> cast probably as simple as this (and I also had the __be16 in the
> >> struct), and Jonathan suggested using union to avoid it...
> >>
> >> At the end of the day, I suppose I am Okay with any of these 3
> >> approaches. Original cast, union or this cast you suggest. Jonathan, any
> >> preferences on your side?  
> > 
> > Majority of the diff is really about renaming buffer to tx_buffer.
> > Could just not bother doing that and instead have buffer and buffer16
> > as the two union elements. With msg gone as suggested above, then the diff
> > becomes only a few lines and you get to keep the nicety of it being either
> > a pair of u8s or a __be16.  
> 
> I was tempted to try using the spi_write_then_read() - but I suppose 
> this may be kind of a hot path.

Given it's small, I doubt it would make any noticeable difference in
performance.

> 
> I'll go with (not)renaming the buffer and dropping the msg, to squeeze 
> the diff.

Works for me.

J
> 
> Yours,
> 	-- Matti

Powered by blists - more mailing lists