Re: [PATCH v2 1/2] iio: gyro: Add driver support for ADXRS290

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On 22/07/20 3:53 pm, Andy Shevchenko wrote:
On Wed, Jul 22, 2020 at 12:40 PM Nishant Malpani
<nish.malpani25@xxxxxxxxx> wrote:
On 22/07/20 3:08 am, Andy Shevchenko wrote:
On Tue, Jul 21, 2020 at 11:35 PM Nishant Malpani
<nish.malpani25@xxxxxxxxx> wrote:
On 22/07/20 1:16 am, Andy Shevchenko wrote:

...

Can't you declare table as const int?

I'm not sure I understand you completely here; do you mean const int *?
So, an array of alternate integer and fractional parts? I suppose that's
possible but we'd be introducing unwanted complexity I feel - for
example, currently the index of the 3db frequency in the table is used
to directly map & set bits in the filter register corresponding to that
frequency but with the approach you share, we'd have to apply a
transformation (div by 2) to set the same bits in the filter register.
Do you think the added complexity justifies the removal of the casting?

It was a question. If you think it is too much, don't change :-)
 > ...

+       /* max transition time to measurement mode */
+       msleep_interruptible(ADXRS290_MAX_TRANSITION_TIME_MS);

I'm not sure what the point of interruptible variant here?

I referred Documentation/timers/timers-howto.rst for this.
My reasoning was shaped to use the interruptible variant because the
transition settles in a time *less than* 100ms and since 100ms is quite
a huge time to sleep, it should be interrupted in case a signal arrives.

This is probe of the device,
What are the expectations here?

I fail to understand why this can't be used in the probe() but perhaps
in a routine to standby/resume. Could you please elaborate?

I didn't say it can not be used, what I'm asking is what are the
expectations of the interruptible part here.
In other words what is the benefit that makes you choose this over
plain msleep().

Oh, sorry for I misunderstood.

I cannot think of anything more to add to the reasoning that I explained earlier; in that case, I'll fall back to msleep() (in v3) unless someone else comes with a strong point in msleep_interruptible()'s favor.

With regards,
Nishant Malpani




[Index of Archives]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Input]     [Linux Kernel]     [Linux SCSI]     [X.org]

  Powered by Linux