Re: [v2 3/3] hwmon: Add Aspeed ast2600 TACH support

[Guenter Roeck]

On Thu, Nov 03, 2022 at 05:40:44AM +0000, Billy Tsai wrote:
> On 2022/11/3, 12:30 PM, "Guenter Roeck" <groeck7@xxxxxxxxx on behalf of linux@xxxxxxxxxxxx> wrote:
> 
>     On Thu, Nov 03, 2022 at 03:52:59AM +0000, Billy Tsai wrote:
>     > > 
>     > > Can't I use a min/max RPM to let the driver know a reasonable timeout/polling period when
>     > > the driver is trying to get RPM?
>     > > Beacause that our tach controller have the falg to indicates the hardware detected the change
>     > > in the input signal. I need the proper timout to rule out slow RPMs.
> 
>     > If the chip measures the fan speed continuously, why would that ever be a
>     > problem, and why wait in the first place instead of just taking the most
>     > recent result ?
> 
>     > Pretty much every other driver is doing that, so I really don't understand
>     > why that would not work here.
> 
> When the fan speed drop from a very fast RPM to a very slow RPM. Especially when it is close to stopping.
> The most recent result will be no meaningful value. The slower RPM needs more time to sample it. E.g., If
> we want to measure the fan with 600 RPM, the controller needs at least 100ms. During this time period, we
> will always get the wrong value. So, our tach controller have the flag to avoid this problem:
> TACH_ASPEED_VALUE_UPDATE: tach value updated since last read
> This flag will be set when the controller detected the change of the signal and clear by read it.
> In order to use this flag, the controller needs the proper timeout based on minimum RPM to avoid waiting forever.
> 

I am not going to accept this patch as-is. If userspace wants to have
values accurate down to ms, this kind of approach is just wrong. Users
will have to live with the fact that measurements may be a bit (in the 
< 1 second range) out of date. Many older drivers even implement code
which avoids reading registers again for a second or longer. Older
temperature sensors may take several seconds to provide new readings.
That is not a reason to block userspace until a new value is available.
I do not see that as a problem. In my opinion it is much more of a
problem if the driver returns a completely bad value such as 0 or even
an error code because its software parameters did not match reality and
the driver didn't wait long enough for a new value. That would be _much_
worse than providing a value which is a few 100 ms out of date, and your
code is vulnerable to that problem.

Besides, for a fan to reduce its speed that quickly, it has to be manually
stopped. Normally fans take several seconds to stop if power is taken away
completely. Your code is adding a lot of complexity (and unnecessary
attributes) for no good reason.

Guenter