[PATCH 0/3] ARM: OMAP2+: hwmod: RTC: Add lock and unlock hooks
t-kristo at ti.com
Thu Jul 16 05:03:50 PDT 2015
On 07/16/2015 01:13 PM, Paul Walmsley wrote:
> On Thu, 16 Jul 2015, Tero Kristo wrote:
>> On 07/16/2015 03:15 AM, Paul Walmsley wrote:
>>> On Tue, 14 Jul 2015, Tero Kristo wrote:
>>>> On 07/14/2015 01:09 PM, Lokesh Vutla wrote:
>>>>> On Wednesday 10 June 2015 02:56 PM, Lokesh Vutla wrote:
>>>>>> Some IP blocks like RTC, needs an additional unlocking mechanism for
>>>>>> writing to its registers. This patch adds optional lock and unlock
>>>>>> function pointers to the IP block's hwmod data which gets executed
>>>>>> before and after writing into IP sysconfig register.
>>>>>> And also hook lock and unlock functions to AMx3xx, DRA7 RTC hwmod
>>>>>> so that sysconfig registers are updated properly.
>>>>> ping on this series.
>>>>> Thanks and regards,
>>>> It is also racy, as there is no locking in place to avoid concurrent
>>>> access to
>>>> the lock/unlock registers across hwmod+driver.
>>> I don't see the race. Where is it?
>> See drivers/rtc/rtc-omap.c, am3352_rtc_unlock and am3352_rtc_lock.
>> That code is accessing the exact same registers.
> I guess my question is, when is it possible that code could race with the
> hwmod code for the same device?
Hmm yea I think you are right, this only gets potentially called within
pm_runtime_get/put_sync for RTC.
The current sequence is highly inefficient though, as we are doing
multiple lock/unlock operations to the RTC from multiple sources. See
following rtcwake trace on am43xx-gp-evm as an example.
/ # rtcwake -s 4 -m mem
[ 7.425322] am3352_rtc_unlock
[ 7.428330] am3352_rtc_lock
[ 7.431139] am3352_rtc_unlock
[ 7.434116] am3352_rtc_lock
wakeup from "mem" at Sat Jan 1 00:00:11 2000
[ 7.448549] PM: Syncing filesystems ... done.
[ 7.455425] Freezing user space processes ... (elapsed 0.001 seconds)
[ 7.463738] Freezing remaining freezable tasks ... (elapsed 0.001
[ 7.472532] Suspending console(s) (use no_console_suspend to debug)
[ 7.481878] am3352_rtc_unlock
[ 7.481889] am3352_rtc_lock
[ 7.482307] PM: suspend of devices complete after 2.713 msecs
[ 7.483479] PM: late suspend of devices complete after 1.153 msecs
[ 7.484727] omap_hwmod_rtc_unlock
[ 7.484733] omap_hwmod_rtc_lock
[ 7.485182] PM: noirq suspend of devices complete after 1.685 msecs
[ 7.485190] Disabling non-boot CPUs ...
[ 7.485199] PM: Successfully put all powerdomains to target state
[ 7.485199] PM: Wakeup source RTC Alarm
[ 7.499853] PM: noirq resume of devices complete after 14.558 msecs
[ 7.500047] am3352_rtc_unlock
[ 7.500052] am3352_rtc_lock
[ 7.500123] am3352_rtc_unlock
[ 7.500128] am3352_rtc_lock
[ 7.501019] PM: early resume of devices complete after 0.809 msecs
[ 7.501464] am3352_rtc_unlock
[ 7.501472] am3352_rtc_lock
[ 7.558046] PM: resume of devices complete after 57.007 msecs
[ 7.638807] Restarting tasks ... done.
[ 7.643173] am3352_rtc_unlock
[ 7.646162] am3352_rtc_lock
But, I guess this is for some interested party to optimize if needed,
and it is mostly an issue with the RTC driver itself.
More information about the linux-arm-kernel