linux-kernel - Re: [PATCH] update clocksource raw_time in timekeeping

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

Message-ID: <af3ea28a0909100107q328d1288x8d677d4d5f79f728@mail.gmail.com>
Date:	Thu, 10 Sep 2009 16:07:10 +0800
From:	ye janboe <janboe.ye@...il.com>
To:	john stultz <johnstul@...ibm.com>
Cc:	zippel@...ux-m68k.org, akpm@...ux-foundation.org, mingo@...e.hu,
	linux-kernel <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] update clocksource raw_time in timekeeping_suspend

hi, John

Thanks for your comments.

After sent this patch, I realize that this patch exposes the hardware
detail ugly in common code.

In embed system, user space apps need to have a method to get the
right time which will not be impacted by NTP and suspend.

Yes, you are right. I want to add sleep_length to the raw time and
user space apps could get the right time after suspend.

Is this right semantics of CLOCK_MONOTONIC_RAW?

Janboe
2009/9/10 john stultz <johnstul@...ibm.com>:
> On Wed, 2009-09-09 at 15:35 +0800, ye janboe wrote:
>> after resume from suspend, raw_time is not updated in
>> timekeeping_suspend. CLOCK_MONOTONIC_RAW could not get the real hw
>> time.
>> This patch fix this issue.
>
> Hmm.. I'll admit suspend probably was less considered with
> CLOCK_MONOTONIC_RAW, so the semantics aren't well established.
>
> However, I do think we want CLOCK_MONOTONIC_RAW to at-least closely map
> to CLOCK_MONOTONIC (but *not* be NTP adjusted). I think that is what
> folks would most likely expect.
>
> However, that isn't what this patch seems to do.
>
> Over suspend, I believe all hardware counters reset, so this patch would
> seem to try to subtract the value back.
>
> This sort of makes sense for something like the TSC, which never wraps,
> so the raw_time would be set back to a tranlation of the actual TSC
> counter,  but for other clocksources like the ACPI PM, it would only
> subtract at most 5 seconds. So this leaks hardware specific detail in an
> ugly way.
>
> Instead I suspect the most intuitive change would be to add in the
> sleep_length to the raw time. This keeps CLOCK_MONOTONIC_RAW behaving
> similarly to CLOCK_MONOTONIC, which I believe makes it more useful for
> folks using CLOCK_MONOTONIC_RAW for things like tuning time
> synchronization.
>
> But let me know more why you chose this implementation and maybe that
> will show some better insight in to how you expect it to behave.
>
> thanks
> -john
>
>
>
>> Signed-off-by: janboe <janboe.ye@...il.com>
>> ---
>>  kernel/time/timekeeping.c |    6 ++++++
>>  1 files changed, 6 insertions(+), 0 deletions(-)
>>
>> diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
>> index e8c77d9..8420b85 100644
>> --- a/kernel/time/timekeeping.c
>> +++ b/kernel/time/timekeeping.c
>> @@ -331,6 +331,8 @@ static unsigned long timekeeping_suspend_time;
>>  static int timekeeping_resume(struct sys_device *dev)
>>  {
>>         unsigned long flags;
>> +       s64 nsec;
>> +       cycle_t last_cycle, cycle_delta;
>>         unsigned long now = read_persistent_clock();
>>
>>         clocksource_resume();
>> @@ -346,8 +348,12 @@ static int timekeeping_resume(struct sys_device *dev)
>>         }
>>         update_xtime_cache(0);
>>         /* re-base the last cycle value */
>> +       last_cycle = clock->cycle_last;
>>         clock->cycle_last = 0;
>>         clock->cycle_last = clocksource_read(clock);
>> +       cycle_delta = clock->cycle_last - last_cycle;
>> +       nsec = cyc2ns(clock, cycle_delta);
>> +       timespec_add_ns(&clock->raw_time, nsec);
>>         clock->error = 0;
>>         timekeeping_suspended = 0;
>>         write_sequnlock_irqrestore(&xtime_lock, flags);
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/