Bug 92481

Summary:	Recent patches to ntp.c and time.c prevent correction by ptpd2
Product:	Timers	Reporter:	George Joseph (george.joseph)
Component:	Other	Assignee:	john stultz (john.stultz)
Status:	CLOSED PATCH_ALREADY_AVAILABLE
Severity:	high	CC:	arekm, embedded
Priority:	P1
Hardware:	ARM
OS:	Linux
Kernel Version:	3.18.5	Subsystem:
Regression:	No	Bisected commit-id:
Attachments:	Patch to fix false positives on 32bit systems

Description George Joseph 2015-02-02 04:10:07 UTC

Hardware: imx6q-cubox-i  ( 6 units )

After upgrading from 3.18.4 to 3.18.5, ptpd2 in slave mode was pegging at -500ppm slew rate (the max) while trying to keep the system clock in sync.  On 3.18.4, the drift correction was always about -50ppm.

With the following 2 commits reverted, ptpd2 is back to about a 50ppm drift correction.

time: settimeofday: Validate the values of tv from usercommit 5e5aeb4367b450a28f447f6d5ab57d8f2ab16a5f upstream.
Verify that the frequency value from userspace is valid and makes sense.

time: adjtimex: Validate the ADJ_FREQUENCY values
commit 5e5aeb4367b450a28f447f6d5ab57d8f2ab16a5f upstream.
Verify that the frequency value from userspace is valid and makes sense.

Comment 1 john stultz 2015-02-02 17:50:20 UTC

Thanks for the report!

Could you confirm that it is only the "time: adjtimex: Validate the ADJ_FREQUENCY values" patch that needs to be reverted?

thanks
-john

Comment 2 john stultz 2015-02-02 19:08:29 UTC

Created attachment 165631 [details]
Patch to fix false positives on 32bit systems

Can you try this patch ontop of the affected kernel to validate it resolves the issue for you?

Comment 3 George Joseph 2015-02-02 19:17:59 UTC

(In reply to john stultz from comment #2)
> Created attachment 165631 [details]
> Patch to fix false positives on 32bit systems
> 
> Can you try this patch ontop of the affected kernel to validate it resolves
> the issue for you?

Funny, I was just checking the value of (LONG_MIN / PPM_SCALE) and (LONG_MIN / PPM_SCALE) and they were evaluating to -32 and 32. :)

So, yep... using LLONG gives the correct -140737488355 to 140737488355 range.

Thanks for looking at this so quickly!

Comment 4 George Joseph 2015-02-02 19:18:38 UTC

Oh, and just to confirm the unstated...  ptpd is now able to maintain the clock.

Comment 5 john stultz 2015-02-02 20:07:20 UTC

Willing to provide a Tested-by: tag?

Comment 6 George Joseph 2015-02-02 20:34:15 UTC

(In reply to john stultz from comment #5)
> Willing to provide a Tested-by: tag?
Sure...

Tested-by: George Joseph <george.joseph@fairview5.com>

Comment 7 john stultz 2015-02-02 20:43:37 UTC

Thanks! I'll be sending the patch out later today.

Comment 8 john stultz 2015-02-19 18:02:11 UTC

Apologies for this being a bit slow to make it upstream. The patch has made it to the -tip tree, and hopefully will be submitted to Linus soon.

Comment 9 embedded 2015-02-28 22:57:46 UTC

Backporting the patch into 3.10.67 and above solves the "adjtime: invalid argument" problem with openntpd.

Comment 10 john stultz 2015-08-31 18:35:42 UTC

Patch has been merged for awhile. Closing this out. Let me know if there are any continued issues.