public inbox for linux-arm-kernel@lists.infradead.org 
 help / color / mirror / Atom feed
From: santosh.shilimkar@ti•com (Santosh Shilimkar)
To: linux-arm-kernel@lists•infradead.org
Subject: [PATCH v2 17/18] ARM: OMAP4+: CPUidle: Add OMAP5 idle driver support
Date: Fri, 5 Apr 2013 15:11:32 +0530	[thread overview]
Message-ID: <515E9C4C.3020709@ti.com> (raw)
In-Reply-To: <87li8ymb84.fsf@linaro.org>

On Thursday 04 April 2013 11:25 PM, Kevin Hilman wrote:
> Santosh Shilimkar <santosh.shilimkar@ti•com> writes:
> 
>> On Thursday 04 April 2013 02:55 AM, Kevin Hilman wrote:
>>> Santosh Shilimkar <santosh.shilimkar@ti•com> writes:
>>>
>>>> The OMAP5 idle driver can re-use OMAP4 CPUidle driver implementation thanks
>>>> to compatible MPUSS design.
>>>>
>>>> Though unlike OMAP4, on OMAP5 devices, MPUSS CSWR (Close Switch
>>>> Retention) power states can be achieved with respective power states
>>>> on CPU0 and CPU1 power domain. This mode was broken on OMAP4 devices
>>>> because of hardware limitation.
>>>
>>> I'm a bit confused by the description here.
>>>
>>> I gather from the code that this means that CPU0 and CPU1 can hit CSWR
>>> independently, correct?
>>>
>> They can be programmed independently without any ordering(like
>> CPU0 last etc), but the actual transition to the low power CSWR
>> happens together. Till that the first CPU hit WFI remains in WFI
>> state waiting for other CPU to hit WFI and then both transition
>> together.
>> Completely managed inside hardware.
> 
> OK, elaborating this in the changelog would be helpful.  Use the "will I
> understand this changelog in a year" rule to see if the changelog is
> detailed enough.  Or better, "will Kevin understand this changelog in a
> year."  (hint: the answer is usually no.)  ;)
> 
:-) I added above description in change-log.

>>>> Also there is no CPU low power entry order requirement like
>>>> master CPU etc for CSWR C-state, which is icing on the cake. 
>>>
>>> Even on secure devices?
>>>
>> Yes. On secure devices too. Actually since we don't loose context,
>> secure entry/exit doesn't come into picture.
>>
>>>> Code makes use of voting scheme for cluster low power state to support
>>>> MPUSS CSWR C-state.
>>>
>>> The voting scheme and associated locking should be better described
>>> here, or commented in the code itself.
>>>
>> You are right. It deserves some description.
>>
>>>> Supported OMAP5 CPUidle C-states:
>>>>         C1 - CPU0 ON(WFI) + CPU1 ON(WFI) + MPUSS ON
>>>>         C2 - CPU0 CSWR + CPU1 CSWR + MPUSS CSWR
>>>>         C3 - CPU0 OFF + CPU1 OFF + MPUSS OSWR
>>>>
>>>> Acked-by: Nishanth Menon <nm@ti•com>
>>>> Signed-off-by: Santosh Shilimkar <santosh.shilimkar@ti•com>
>>>
>>> [...]
>>>
>>>> +static int omap_enter_idle_smp(struct cpuidle_device *dev,
>>>> +			struct cpuidle_driver *drv,
>>>> +			int index)
>>>> +{
>>>> +	struct idle_statedata *cx = state_ptr + index;
>>>> +	int cpu_id = smp_processor_id();
>>>> +	unsigned long flag;
>>>> +
>>>> +	clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu_id);
>>>
>>> I think the CPUidle core handles the broadcast notification now.
>>>
>> Not in mainline yet. And those patches came after my patches and
>> I don't wanted un-necessary merge dependency, I avoided it. Its trivial
>> though to drop if from here once the idle next is merged.
> 
> OK.
> 
> I believe that stuff is already queued, no?  Maybe ahave this as an
> add-on separate patch that can be used for your loal testing, but does
> not go upstream.
> 
Will do.

> I would only include this if you're sure the other series is not going
> upstream.
>
It might go upstream so I will manually apply the patch or pull a branch
if available.
 
>>>> +	raw_spin_lock_irqsave(&mpu_lock, flag);
>>>> +	cx->mpu_state_vote++;
>>>
>>> How about using an atomic_t and atomic_inc()/atomic_dec() instead, which
>>> will avoid the need for a spinlock.
>>>
>> Spin lock is not just for the vote variable. I had atomics opps in first
>> version I gave it to product team. But they found a race condition in
>> where MPU power state was getting overwritten by other CPU.
>>
>>> Even with that, it still seems potentially racy.  Do you need a memory
>>> barrier here to be sure that any changes from another CPU are visible
>>> here, and vice versa?
>>>
>> With locks, you don't need barriers since the updated copy is guaranteed.
> 
> It's guaranteed because the spinlock implementation uses barriers. 
> 
Yeah.

>> Can you please elaborate on potential race ? I have given pretty hard
>> thought and didn't see any race which can be exposed with locks in place.
> 
> I was referring to using atomic ops.  With atomic ops, you'd need an
> explicit barrier (which you're getting inside the spinlock
> implementation)
> 
> That being said, I have not thought through all the corner cases as you
> have, so I'll defer to your choice (as long as it's well documented.)
> If you decide to stick with spinlocks, be sure to describe all of what
> the spinlock is protecting, and why.
> 
Have updated the change-log as well as code to elaborate the lock use.

Thanks for review.

Regards,
Santosh

  reply	other threads:[~2013-04-05  9:41 UTC|newest]

Thread overview: 64+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2013-03-25 10:04 [PATCH v2 00/18] ARM: OMAP5: PM: Add MPUSS suspend and CPUidle support Santosh Shilimkar
2013-03-25 10:04 ` [PATCH v2 01/18] ARM: OMAP4+: PM: Consolidate MPU subsystem PM code for re-use Santosh Shilimkar
2013-04-03 19:44   ` Kevin Hilman
2013-04-04 11:32     ` Santosh Shilimkar
2013-03-25 10:04 ` [PATCH v2 02/18] ARM: OMAP5: PM: Update CPU context register offset Santosh Shilimkar
2013-03-25 10:04 ` [PATCH v2 03/18] ARM: OMAP4+: PM: Consolidate and use OMAP4 PM code for OMAP5 Santosh Shilimkar
2013-04-03 20:20   ` Kevin Hilman
2013-04-04 11:51     ` Santosh Shilimkar
2013-04-04 11:55       ` Santosh Shilimkar
2013-03-25 10:04 ` [PATCH v2 04/18] ARM: OMAP5: PM: Set MPUSS-EMIF clock-domain static dependency Santosh Shilimkar
2013-03-25 10:04 ` [PATCH v2 05/18] ARM: OMAP5: PM: Enables ES2 PM mode by default Santosh Shilimkar
2013-04-03 20:25   ` Kevin Hilman
2013-04-04 12:02     ` Santosh Shilimkar
2013-03-25 10:04 ` [PATCH v2 06/18] ARM: OMAP5: PM: Enable Mercury retention mode on CPUx powerdomains Santosh Shilimkar
2013-04-03 20:31   ` Kevin Hilman
2013-04-04 12:08     ` Santosh Shilimkar
2013-03-25 10:04 ` [PATCH v2 07/18] ARM: OMAP5: Add init_late() hook to enable pm initialization Santosh Shilimkar
2013-04-03 20:33   ` Kevin Hilman
2013-04-04 12:28     ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 08/18] ARM: OMAP5: PM: Add CPU power off in hotplug path Santosh Shilimkar
2013-04-03 20:49   ` Kevin Hilman
2013-04-04 13:23     ` Santosh Shilimkar
2013-04-04 17:31       ` Kevin Hilman
2013-04-05  9:04         ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 09/18] ARM: OMAP4+: PM: Restore CPU power state to ON with clockdomain force wakeup method Santosh Shilimkar
2013-04-03 20:54   ` Kevin Hilman
2013-04-04 13:37     ` Santosh Shilimkar
2013-04-04 17:42       ` Kevin Hilman
2013-04-05  9:07         ` Santosh Shilimkar
2013-04-05 11:58           ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 10/18] ARM: OMAP5: PM: Add MPU Open Switch Retention support Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 11/18] ARM: OMAP5: PM: Add L2 memory power down support Santosh Shilimkar
2013-04-03 20:58   ` Kevin Hilman
2013-04-04 13:46     ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 12/18] ARM: OMAP4: CPUidle: Avoid double idle driver registration Santosh Shilimkar
2013-04-03 21:03   ` Kevin Hilman
2013-04-04 13:47     ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 13/18] ARM: OMAP: CPUidle: Unregister drivere on device registration failure Santosh Shilimkar
2013-04-03 21:03   ` Kevin Hilman
2013-04-04 13:48     ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 14/18] ARM: OMAP4: CPUidle: Make C-state description field more precise Santosh Shilimkar
2013-04-03 21:05   ` Kevin Hilman
2013-04-04 13:48     ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 15/18] ARM: OMAP4+: CPUidle: Consolidate idle driver for OMAP5 support Santosh Shilimkar
2013-04-03 21:10   ` Kevin Hilman
2013-04-04 14:04     ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 16/18] ARM: OMAP4+: CPUidle: Deprecate use of omap4_mpuss_read_prev_context_state() Santosh Shilimkar
2013-04-03 21:37   ` Kevin Hilman
2013-04-04 13:59     ` Santosh Shilimkar
2013-03-25 10:05 ` [PATCH v2 17/18] ARM: OMAP4+: CPUidle: Add OMAP5 idle driver support Santosh Shilimkar
2013-04-03 21:25   ` Kevin Hilman
2013-04-04 14:16     ` Santosh Shilimkar
2013-04-04 17:55       ` Kevin Hilman
2013-04-05  9:41         ` Santosh Shilimkar [this message]
2013-03-25 10:05 ` [PATCH v2 18/18] ARM: OMAP5: PM: handle device instance for warm reset Santosh Shilimkar
2013-03-25 11:46 ` [PATCH v2 00/18] ARM: OMAP5: PM: Add MPUSS suspend and CPUidle support Lokesh Vutla
2013-03-25 12:10   ` Santosh Shilimkar
2013-03-25 12:27 ` Sourav Poddar
2013-03-25 12:47   ` Rajendra Nayak
2013-03-25 13:00     ` Sourav Poddar
2013-04-03 22:52 ` Kevin Hilman
2013-04-04 14:34   ` Santosh Shilimkar
2013-04-04 16:49     ` Santosh Shilimkar
2013-04-04 17:57       ` Kevin Hilman

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=515E9C4C.3020709@ti.com \
    --to=santosh.shilimkar@ti$(echo .)com \
    --cc=linux-arm-kernel@lists$(echo .)infradead.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox