linux-kernel - Re: [PATCH 0/2] Add settle time support to mux-gpio

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <f056ed07-46d4-7f75-f6af-29f843c37d15@axentia.se>
Date:   Wed, 6 Oct 2021 14:24:06 +0200
From:   Peter Rosin <peda@...ntia.se>
To:     Vincent Whitchurch <vincent.whitchurch@...s.com>,
        devicetree@...r.kernel.org
Cc:     kernel@...s.com, robh+dt@...nel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 0/2] Add settle time support to mux-gpio

Hi!

On 2021-10-04 17:36, Vincent Whitchurch wrote:
> On one of our boards we use gpio-mux with iio-mux to read voltages using an
> ADC from a few different channels, and on this board the input voltage needs
> some time to stabilize after a switch of the mux.
> 
> This series add devicetree and driver support for this kind of hardware which
> requries a settle time after muxing.

It feels backwards to hide a universally useful thing like this fsleep in
drivers/mux/gpio.c. I think it belongs in drivers/iio/multiplexer/iio-mux.c
instead. Because the sleep is needed for the analog parts of the actual mux,
not the digital parts of the mux-controller.

However, currently the iio-mux does not know when the mux changes state (and
it can change state from "underneath" iio-mux by some other driver using the
same mux-control to drive some other mux).

But, fixing that so that the iio-mux knows if mux_control_select changes
the state (e.g. returning 1 instead of 0 on state changes) does not fix
this problem. Because the mux-control, again, can be used by some other
driver that changed its state right before the iio-mux selected it without
then needing a state change. And that could potentially happen quicker
than the prescribed fsleep.

So, fixing it needs some kind of new api that returns when the mux-control
changed its state last, e.g.
	ret = mux_control_select_stamp(mux, state, &stamp);
that does the same thing as mux_control_select, but also fills in a time
stamp for when the mux changed state.

Another similar option is to add an extra delay argument, e.g.
	ret = mux_control_select_delay(mux, stamp, delay_us);
that instead makes the call not return before the delay has passed since
the last state change, regardless who changed state. The mux-control would
need to keep track of when the last state changes happened of course, just
as in the above _select_stamp variant.

I think I like the last _select_delay variant best.

Cheers,
Peter