Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Rolf Eike Beer]

After upgrade to 5.11 get this multiple times per second on my C8000:

[   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than allowed 
by the 'jiffies' clock's max_cycles value (10): time overflow danger
[   36.998705]          timekeeping: Your kernel is sick, but tries to cope by 
capping time updates

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Rolf Eike Beer]

Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
> After upgrade to 5.11 get this multiple times per second on my C8000:
> 
> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
> danger [   36.998705]          timekeeping: Your kernel is sick, but tries
> to cope by capping time updates

Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade attempt, I'll 
retry once I have built a proper 5.11 kernel.

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Rolf Eike Beer]

[-- Attachment #1: Type: text/plain, Size: 1115 bytes --]

Am Montag, 1. März 2021, 17:49:42 CET schrieb Rolf Eike Beer:
> Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
> > After upgrade to 5.11 get this multiple times per second on my C8000:
> > 
> > [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
> > allowed by the 'jiffies' clock's max_cycles value (10): time overflow
> > danger [   36.998705]          timekeeping: Your kernel is sick, but tries
> > to cope by capping time updates
> 
> Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade attempt,
> I'll retry once I have built a proper 5.11 kernel.

Ok, it's there also in 5.11.2:

[    0.000000] rcu: RCU calculated value of scheduler-enlistment delay is 25 
jiffies.
[    0.000000] NR_IRQS: 128
[    0.000003] sched_clock: 64 bits at 1000MHz, resolution 1ns, wraps every 
4398046511103ns
[    0.106687] WARNING: timekeeping: Cycle offset (26) is larger than allowed 
by the 'jiffies' clock's max_cycles value (10): time overflow danger
[    0.106698]          timekeeping: Your kernel is sick, but tries to cope by 
capping time updates


[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Helge Deller]

On 3/1/21 7:44 PM, Rolf Eike Beer wrote:
> Am Montag, 1. März 2021, 17:49:42 CET schrieb Rolf Eike Beer:
>> Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
>>> After upgrade to 5.11 get this multiple times per second on my C8000:
>>>
>>> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
>>> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
>>> danger [   36.998705]          timekeeping: Your kernel is sick, but tries
>>> to cope by capping time updates

I know I have seen this at least once with a 32-bit kernel in qemu as well....

>> Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade attempt,
>> I'll retry once I have built a proper 5.11 kernel.
>
> Ok, it's there also in 5.11.2:

You don't see it in 5.11, but in 5.11.2.
Sadly none of the changes between those versions seem related
to this problem.

Do you still see this?
I'd like to get it anaylzed/fixed.

Helge


>
> [    0.000000] rcu: RCU calculated value of scheduler-enlistment delay is 25
> jiffies.
> [    0.000000] NR_IRQS: 128
> [    0.000003] sched_clock: 64 bits at 1000MHz, resolution 1ns, wraps every
> 4398046511103ns
> [    0.106687] WARNING: timekeeping: Cycle offset (26) is larger than allowed
> by the 'jiffies' clock's max_cycles value (10): time overflow danger
> [    0.106698]          timekeeping: Your kernel is sick, but tries to cope by
> capping time updates
>

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Rolf Eike Beer]

[-- Attachment #1: Type: text/plain, Size: 1149 bytes --]

Am Mittwoch, 3. März 2021, 15:29:42 CET schrieb Helge Deller:
> On 3/1/21 7:44 PM, Rolf Eike Beer wrote:
> > Am Montag, 1. März 2021, 17:49:42 CET schrieb Rolf Eike Beer:
> >> Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
> >>> After upgrade to 5.11 get this multiple times per second on my C8000:
> >>> 
> >>> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
> >>> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
> >>> danger [   36.998705]          timekeeping: Your kernel is sick, but
> >>> tries
> >>> to cope by capping time updates
> 
> I know I have seen this at least once with a 32-bit kernel in qemu as
> well....
> >> Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade attempt,
> >> I'll retry once I have built a proper 5.11 kernel.
> > 
> > Ok, it's there also in 5.11.2:
> You don't see it in 5.11, but in 5.11.2.
> Sadly none of the changes between those versions seem related
> to this problem.
> 
> Do you still see this?
> I'd like to get it anaylzed/fixed.

I don't see it in 5.10.4, but in 5.11.2. I have not tried 5.11.0.

Eike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[John David Anglin]

On 2021-03-03 9:43 a.m., Rolf Eike Beer wrote:
> I don't see it in 5.10.4, but in 5.11.2. I have not tried 5.11.0.
I have tried almost every 5.10.x version and haven't seen it on rp3440 or c8000.

Dave

-- 
John David Anglin  dave.anglin@bell.net

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Rolf Eike Beer]

[-- Attachment #1: Type: text/plain, Size: 1101 bytes --]

Am Mittwoch, 3. März 2021, 15:29:42 CET schrieb Helge Deller:
> On 3/1/21 7:44 PM, Rolf Eike Beer wrote:
> > Am Montag, 1. März 2021, 17:49:42 CET schrieb Rolf Eike Beer:
> >> Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
> >>> After upgrade to 5.11 get this multiple times per second on my C8000:
> >>> 
> >>> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
> >>> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
> >>> danger [   36.998705]          timekeeping: Your kernel is sick, but
> >>> tries
> >>> to cope by capping time updates
> 
> I know I have seen this at least once with a 32-bit kernel in qemu as
> well....
> >> Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade attempt,
> >> I'll retry once I have built a proper 5.11 kernel.
> > 
> > Ok, it's there also in 5.11.2:
> You don't see it in 5.11, but in 5.11.2.
> Sadly none of the changes between those versions seem related
> to this problem.
> 
> Do you still see this?
> I'd like to get it anaylzed/fixed.

Me too. What do you need?

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Helge Deller]

On 3/14/21 10:47 AM, Rolf Eike Beer wrote:
> Am Mittwoch, 3. März 2021, 15:29:42 CET schrieb Helge Deller:
>> On 3/1/21 7:44 PM, Rolf Eike Beer wrote:
>>> Am Montag, 1. März 2021, 17:49:42 CET schrieb Rolf Eike Beer:
>>>> Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
>>>>> After upgrade to 5.11 get this multiple times per second on my C8000:
>>>>>
>>>>> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
>>>>> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
>>>>> danger [   36.998705]          timekeeping: Your kernel is sick, but
>>>>> tries
>>>>> to cope by capping time updates
>>
>> I know I have seen this at least once with a 32-bit kernel in qemu as
>> well....
>>>> Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade attempt,
>>>> I'll retry once I have built a proper 5.11 kernel.
>>>
>>> Ok, it's there also in 5.11.2:
>> You don't see it in 5.11, but in 5.11.2.
>> Sadly none of the changes between those versions seem related
>> to this problem.
>>
>> Do you still see this?
>> I'd like to get it anaylzed/fixed.
>
> Me too. What do you need?

I actually don't know.
First of all it would be great if we could reproduce it.
Right now I don't see this issue any longer, so I have nowhere to start from.

Helge

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Rolf Eike Beer]

[-- Attachment #1: Type: text/plain, Size: 1514 bytes --]

Am Sonntag, 14. März 2021, 12:16:11 CET schrieben Sie:
> On 3/14/21 10:47 AM, Rolf Eike Beer wrote:
> > Am Mittwoch, 3. März 2021, 15:29:42 CET schrieb Helge Deller:
> >> On 3/1/21 7:44 PM, Rolf Eike Beer wrote:
> >>> Am Montag, 1. März 2021, 17:49:42 CET schrieb Rolf Eike Beer:
> >>>> Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
> >>>>> After upgrade to 5.11 get this multiple times per second on my C8000:
> >>>>> 
> >>>>> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
> >>>>> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
> >>>>> danger [   36.998705]          timekeeping: Your kernel is sick, but
> >>>>> tries
> >>>>> to cope by capping time updates
> >> 
> >> I know I have seen this at least once with a 32-bit kernel in qemu as
> >> well....
> >> 
> >>>> Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade
> >>>> attempt,
> >>>> I'll retry once I have built a proper 5.11 kernel.
> >>> 
> >>> Ok, it's there also in 5.11.2:
> >> You don't see it in 5.11, but in 5.11.2.
> >> Sadly none of the changes between those versions seem related
> >> to this problem.
> >> 
> >> Do you still see this?
> >> I'd like to get it anaylzed/fixed.
> > 
> > Me too. What do you need?
> 
> I actually don't know.
> First of all it would be great if we could reproduce it.
> Right now I don't see this issue any longer, so I have nowhere to start
> from.

I get it every time if I boot that kernel.

Eike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Rolf Eike Beer]

Am 2021-03-14 13:08, schrieb Rolf Eike Beer:
> Am Sonntag, 14. März 2021, 12:16:11 CET schrieben Sie:
>> On 3/14/21 10:47 AM, Rolf Eike Beer wrote:
>> > Am Mittwoch, 3. März 2021, 15:29:42 CET schrieb Helge Deller:
>> >> On 3/1/21 7:44 PM, Rolf Eike Beer wrote:
>> >>> Am Montag, 1. März 2021, 17:49:42 CET schrieb Rolf Eike Beer:
>> >>>> Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
>> >>>>> After upgrade to 5.11 get this multiple times per second on my C8000:
>> >>>>>
>> >>>>> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
>> >>>>> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
>> >>>>> danger [   36.998705]          timekeeping: Your kernel is sick, but
>> >>>>> tries
>> >>>>> to cope by capping time updates
>> >>
>> >> I know I have seen this at least once with a 32-bit kernel in qemu as
>> >> well....
>> >>
>> >>>> Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade
>> >>>> attempt,
>> >>>> I'll retry once I have built a proper 5.11 kernel.
>> >>>
>> >>> Ok, it's there also in 5.11.2:
>> >> You don't see it in 5.11, but in 5.11.2.
>> >> Sadly none of the changes between those versions seem related
>> >> to this problem.
>> >>
>> >> Do you still see this?
>> >> I'd like to get it anaylzed/fixed.
>> >
>> > Me too. What do you need?
>> 
>> I actually don't know.
>> First of all it would be great if we could reproduce it.
>> Right now I don't see this issue any longer, so I have nowhere to 
>> start
>> from.
> 
> I get it every time if I boot that kernel.

I still see it in 5.13.12, so I'm in the bad situation now that I don't 
have a working kernel on that machine anymore and need to think about 
how to restore it. While the latter is exactly my problem I would still 
love to see the kernel problem solved.

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[John David Anglin]

On 2021-08-26 10:30 a.m., Rolf Eike Beer wrote:
> Am 2021-03-14 13:08, schrieb Rolf Eike Beer:
>> Am Sonntag, 14. März 2021, 12:16:11 CET schrieben Sie:
>>> On 3/14/21 10:47 AM, Rolf Eike Beer wrote:
>>> > Am Mittwoch, 3. März 2021, 15:29:42 CET schrieb Helge Deller:
>>> >> On 3/1/21 7:44 PM, Rolf Eike Beer wrote:
>>> >>> Am Montag, 1. März 2021, 17:49:42 CET schrieb Rolf Eike Beer:
>>> >>>> Am Montag, 1. März 2021, 17:25:18 CET schrieb Rolf Eike Beer:
>>> >>>>> After upgrade to 5.11 get this multiple times per second on my C8000:
>>> >>>>>
>>> >>>>> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
>>> >>>>> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
>>> >>>>> danger [   36.998705]          timekeeping: Your kernel is sick, but
>>> >>>>> tries
>>> >>>>> to cope by capping time updates
>>> >>
>>> >> I know I have seen this at least once with a 32-bit kernel in qemu as
>>> >> well....
>>> >>
>>> >>>> Not 5.11, but 5.10.11. 5.10.4 is fine. It could be a bad upgrade
>>> >>>> attempt,
>>> >>>> I'll retry once I have built a proper 5.11 kernel.
>>> >>>
>>> >>> Ok, it's there also in 5.11.2:
>>> >> You don't see it in 5.11, but in 5.11.2.
>>> >> Sadly none of the changes between those versions seem related
>>> >> to this problem.
>>> >>
>>> >> Do you still see this?
>>> >> I'd like to get it anaylzed/fixed.
>>> >
>>> > Me too. What do you need?
>>>
>>> I actually don't know.
>>> First of all it would be great if we could reproduce it.
>>> Right now I don't see this issue any longer, so I have nowhere to start
>>> from.
>>
>> I get it every time if I boot that kernel.
>
> I still see it in 5.13.12, so I'm in the bad situation now that I don't have a working kernel on that machine anymore and need to think about
> how to restore it. While the latter is exactly my problem I would still love to see the kernel problem solved.
I don't see this either.  So, it probably has to do with your config.

Since it fails consistently on your system, you could use git bisect to find the change which introduced problem.

Dave

-- 
John David Anglin  dave.anglin@bell.net

Re: Cycle offset is larger than allowed by the 'jiffies' clock's max_cycles value

[Rolf Eike Beer]

[-- Attachment #1: Type: text/plain, Size: 836 bytes --]

Am Montag, 1. März 2021, 17:25:18 CEST schrieb Rolf Eike Beer:
> After upgrade to 5.11 get this multiple times per second on my C8000:
> 
> [   36.998702] WARNING: timekeeping: Cycle offset (29) is larger than
> allowed by the 'jiffies' clock's max_cycles value (10): time overflow
> danger [   36.998705]          timekeeping: Your kernel is sick, but tries
> to cope by capping time updates

This has nothing to do with the kernel version. For whatever reason 
CONFIG_DEBUG_TIMEKEEPING was flipped on in the newer versions, no idea why and 
how I did that.

However, that "may have a (very small) performance impact" in the Kconfig 
isn't exactly true. When output through my 9600 serial line it slows the 
kernel down to a crawl, so much that it doesn't even boot.

So, everything is fine for now.

Regards,

Eike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 195 bytes --]