On 10/5/23 20:01, Jonathan Cameron wrote:
> On Thu, 5 Oct 2023 13:59:22 +0300
> Dumitru Ceclan <mitrutzceclan@xxxxxxxxx> wrote:
>
>> The AD7173 family offer a complete integrated Sigma-Delta ADC solution
>> which can be used in high precision, low noise single channel
>> applications or higher speed multiplexed applications. The Sigma-Delta
>> ADC is intended primarily for measurement of signals close to DC but also
>> delivers outstanding performance with input bandwidths out to ~10kHz.
>>
...
>> + case IIO_CHAN_INFO_RAW:
>> + ret = ad_sigma_delta_single_conversion(indio_dev, chan, val);
>
> It's fairly usual for it to be safe to grab a single conversion when the
> buffered mode might be enabled. Do you need an iio_device_claim_direct_mode()
> here?
>
ad_sigma_delta_single_conversion() calls iio_device_claim_direct_mode()
>> +static int ad7173_update_scan_mode(struct iio_dev *indio_dev,
>> + const unsigned long *scan_mask)
>> +{
>> + struct ad7173_state *st = iio_priv(indio_dev);
>> + int i, ret = 0;
>> +
>> + iio_device_claim_direct_mode(indio_dev);
>
> This looks wrong.
> Firstly iio_device_claim_direct_mode() can fail so you always have
> to check the return value. If it does fail and you then call
> iio_release_direct_mode() it is unbalanced release of a mutex.
>
> Secondly update_scan_mode is only called as part of buffer setup
> and there should be no races around that (and the mutex this
> tries to grab is already held.
> https://elixir.bootlin.com/linux/latest/source/drivers/iio/industrialio-buffer.c#L1265
> )
>
> If you are protecting something device specific (rather than
> the mode) then a device specific lock should be taken.
>
The use of a lock was inspired from ad7124, but from looking at it the only use it has
was to protect concurrent access of the device setup from write_raw (that now uses ...direct_mode())
I think it's best to drop this lock.
>> + chan[chan_index].differential = fwnode_property_read_bool(child, "bipolar");
>
> bipolar doesn't normally == differential.
> You can have unipolar differential (just that you can't get a negative answer)
> Perhaps just a terminology thing?
>
This device supports only differential channels. Here, the differential flag is used to show
if bipolar coding should be used.
>> + st->info = device_get_match_data(dev);
>> + if (!st->info)
>> + return -ENODEV;
> This works for the cases of DT and ACPI but not for anyone just
> using the spi_device_id table.
> There is spi_device_get_match_data() to cover all options.
>
I could not find the spi_device_get_match_data() function in the repo.
It appears however as a suggestion from Andy Shevchenko in a thread:
https://www.mail-archive.com/linux-kernel@xxxxxxxxxxxxxxx/msg2382960.html
Is this it?
