From: swarren@wwwdotorg.org (Stephen Warren)
To: linux-arm-kernel@lists.infradead.org
Subject: [PATCH v2] pinctrl: reserve pins when states are activated
Date: Mon, 22 Oct 2012 14:30:05 -0600 [thread overview]
Message-ID: <5085ACCD.3040203@wwwdotorg.org> (raw)
In-Reply-To: <CACRpkdZazjMoSzkr5t15-GeCnQKCJDsZyvT=KVMyw5XhjDybGQ@mail.gmail.com>
On 10/22/2012 02:21 AM, Linus Walleij wrote:
> On Fri, Oct 19, 2012 at 8:10 PM, Tony Lindgren <tony@atomide.com> wrote:
>> [Me]
>>> Instead: let use reserve the pins when the state is activated
>>> and drop them when the state is disabled, i.e. when we move to
>>> another state. This way different devices/functions can use the
>>> same pins at different times.
>>
>> Hmm doesn't this mean that we are now doing lots of extra
>> reserving and dropping of pins? Performance is important from
>> latency point of view for cases where we need to remux pins
>> constantly runtime PM.
>
> It is only done in case the pinmux state is switched in runtime
> suspend/resume, so it's e.g. possible to just alter the pin config.
>
> But in general what you say is true.
>
> We used to to the same thing by having drivers call
> pinctrl_get()/pinctrl_put() in this case instead, but that went
> away with the introduction of states, so we cannot encode
> different pin sets with say
> pinctrl_get(dev, "foo")/pinctrl_get(dev, "bar")
> anymore since there is only one pinctrl handle per device,
> but multiple states.
>
> If this turns out to be a severe performance bottleneck, I
> suggest to add some additional constraint API, like
> pinctrl_set_pinmux_homegeneous_pinsets(true) that will
> at runtime select whether the pin allocation is done when
> getting the pinctrl handle instead.
That API sounds like something system-wide, which seems like it would be
rather presumptuous (CPU/SoC support code couldn't execute it, since
that would presume a facet of all board designs that could change in the
future). Even a driver shouldn't be assuming this; it can't know what
boards it gets used in ahead of time.
Instead, it seems like the map registration code could easily look at
all states defined for a device, and determine if the set of pins/groups
used by those states was identical, and switch between up-front or
dynamic registration as needed by the specific map entries.
WARNING: multiple messages have this Message-ID (diff)
From: Stephen Warren <swarren@wwwdotorg.org>
To: Linus Walleij <linus.walleij@linaro.org>
Cc: Tony Lindgren <tony@atomide.com>,
Linus Walleij <linus.walleij@stericsson.com>,
linux-kernel@vger.kernel.org,
linux-arm-kernel@lists.infradead.org,
Stephen Warren <swarren@nvidia.com>,
Anmar Oueja <anmar.oueja@linaro.org>,
Patrice Chotard <patrice.chotard@st.com>,
Jean Nicolas Graux <jean-nicolas.graux@stericsson.com>,
Loic Pallardy <loic.pallardy@st.com>
Subject: Re: [PATCH v2] pinctrl: reserve pins when states are activated
Date: Mon, 22 Oct 2012 14:30:05 -0600 [thread overview]
Message-ID: <5085ACCD.3040203@wwwdotorg.org> (raw)
In-Reply-To: <CACRpkdZazjMoSzkr5t15-GeCnQKCJDsZyvT=KVMyw5XhjDybGQ@mail.gmail.com>
On 10/22/2012 02:21 AM, Linus Walleij wrote:
> On Fri, Oct 19, 2012 at 8:10 PM, Tony Lindgren <tony@atomide.com> wrote:
>> [Me]
>>> Instead: let use reserve the pins when the state is activated
>>> and drop them when the state is disabled, i.e. when we move to
>>> another state. This way different devices/functions can use the
>>> same pins at different times.
>>
>> Hmm doesn't this mean that we are now doing lots of extra
>> reserving and dropping of pins? Performance is important from
>> latency point of view for cases where we need to remux pins
>> constantly runtime PM.
>
> It is only done in case the pinmux state is switched in runtime
> suspend/resume, so it's e.g. possible to just alter the pin config.
>
> But in general what you say is true.
>
> We used to to the same thing by having drivers call
> pinctrl_get()/pinctrl_put() in this case instead, but that went
> away with the introduction of states, so we cannot encode
> different pin sets with say
> pinctrl_get(dev, "foo")/pinctrl_get(dev, "bar")
> anymore since there is only one pinctrl handle per device,
> but multiple states.
>
> If this turns out to be a severe performance bottleneck, I
> suggest to add some additional constraint API, like
> pinctrl_set_pinmux_homegeneous_pinsets(true) that will
> at runtime select whether the pin allocation is done when
> getting the pinctrl handle instead.
That API sounds like something system-wide, which seems like it would be
rather presumptuous (CPU/SoC support code couldn't execute it, since
that would presume a facet of all board designs that could change in the
future). Even a driver shouldn't be assuming this; it can't know what
boards it gets used in ahead of time.
Instead, it seems like the map registration code could easily look at
all states defined for a device, and determine if the set of pins/groups
used by those states was identical, and switch between up-front or
dynamic registration as needed by the specific map entries.
next prev parent reply other threads:[~2012-10-22 20:30 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-10-19 13:05 [PATCH v2] pinctrl: reserve pins when states are activated Linus Walleij
2012-10-19 13:05 ` Linus Walleij
2012-10-19 14:51 ` Jean-Nicolas GRAUX
2012-10-19 14:51 ` Jean-Nicolas GRAUX
2012-10-19 16:18 ` Stephen Warren
2012-10-19 16:18 ` Stephen Warren
2012-10-19 18:10 ` Tony Lindgren
2012-10-19 18:10 ` Tony Lindgren
2012-10-22 8:21 ` Linus Walleij
2012-10-22 8:21 ` Linus Walleij
2012-10-22 19:07 ` Tony Lindgren
2012-10-22 19:07 ` Tony Lindgren
2012-10-23 8:58 ` Linus Walleij
2012-10-23 8:58 ` Linus Walleij
2012-10-22 20:30 ` Stephen Warren [this message]
2012-10-22 20:30 ` Stephen Warren
2012-10-23 9:31 ` Linus Walleij
2012-10-23 9:31 ` Linus Walleij
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=5085ACCD.3040203@wwwdotorg.org \
--to=swarren@wwwdotorg.org \
--cc=linux-arm-kernel@lists.infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.