On 07/08/2012 08:09 AM, Lingzhu Xiang wrote: > A bug in the hrtimer subsystem caused all TIMER_ABSTIME CLOCK_REALTIME > timers to expire one second early during and after leap seconds, thus > such sub-second timers expire immediately, causing load spikes if used > in fast loops. Calling clock_was_set() reverts this state. See [1]. > > [1]: http://lwn.net/Articles/504658/. > > This patch is a regression test for the bug, based on the test case > provided by John Stultz from the above link, and also tries to recover > from inconsistency caused by reproduction of the bug.
A sample test run looks like this: leapsec_timer 0 TINFO : test start at 2012-07-08 08:02:32.996771295 +0800 leapsec_timer 0 TINFO : now is 2012-07-08 08:02:32.996929883 +0800 leapsec_timer 0 TINFO : sleep till 2012-07-08 08:02:33.996929883 +0800 leapsec_timer 0 TINFO : now is 2012-07-08 08:02:33.996975957 +0800 leapsec_timer 0 TINFO : hrtimer early expiration is not detected. leapsec_timer 0 TINFO : scheduling leap second 2012-07-09 08:00:00.000000000 +0800 leapsec_timer 0 TINFO : setting time to 2012-07-09 07:59:58.000000000 +0800 leapsec_timer 0 TINFO : 2012-07-09 07:59:58.000080000 +0800 adjtimex: clock synchronized leapsec_timer 0 TINFO : 2012-07-09 07:59:58.000101000 +0800 adjtimex: clock synchronized leapsec_timer 0 TINFO : 2012-07-09 07:59:58.000119000 +0800 adjtimex: insert leap second leapsec_timer 0 TINFO : 2012-07-09 07:59:58.000138000 +0800 adjtimex: insert leap second leapsec_timer 0 TINFO : 2012-07-09 07:59:58.500174000 +0800 adjtimex: insert leap second leapsec_timer 0 TINFO : 2012-07-09 07:59:59.000246000 +0800 adjtimex: insert leap second leapsec_timer 0 TINFO : 2012-07-09 07:59:59.500320000 +0800 adjtimex: insert leap second leapsec_timer 0 TINFO : 2012-07-09 07:59:59.000368000 +0800 adjtimex: leap second in progress leapsec_timer 0 TINFO : 2012-07-09 07:59:59.500403000 +0800 adjtimex: leap second in progress leapsec_timer 0 TINFO : 2012-07-09 08:00:00.000432000 +0800 adjtimex: leap second has occurred leapsec_timer 0 TINFO : 2012-07-09 08:00:00.500457000 +0800 adjtimex: leap second has occurred leapsec_timer 0 TINFO : 2012-07-09 08:00:01.000475000 +0800 adjtimex: leap second has occurred leapsec_timer 0 TINFO : 2012-07-09 08:00:01.500489000 +0800 adjtimex: leap second has occurred leapsec_timer 0 TINFO : 2012-07-09 08:00:02.000506000 +0800 adjtimex: leap second has occurred leapsec_timer 0 TINFO : now is 2012-07-09 08:00:02.500524432 +0800 leapsec_timer 0 TINFO : sleep till 2012-07-09 08:00:03.500524432 +0800 leapsec_timer 0 TINFO : now is 2012-07-09 08:00:02.500550236 +0800 leapsec_timer 1 TFAIL : hrtimer early expiration is detected. leapsec_timer 0 TINFO : 2012-07-09 08:00:02.500561000 +0800 adjtimex: clock synchronized leapsec_timer 0 TINFO : 2012-07-09 08:00:02.500568000 +0800 adjtimex: clock synchronized leapsec_timer 0 TINFO : restoring time to 2012-07-08 08:02:37.497346606 +0800 leapsec_timer 0 TINFO : now is 2012-07-08 08:02:37.497351161 +0800 leapsec_timer 0 TINFO : sleep till 2012-07-08 08:02:38.497351161 +0800 leapsec_timer 0 TINFO : now is 2012-07-08 08:02:38.497391392 +0800 leapsec_timer 0 TINFO : hrtimer early expiration is not detected. As can be observed, the first timer before the leap second and third timer after resetting time state worked as expected (early expiration not detected), but the second timer after the leap second before resetting time state returned unexpectedly early. Thus the above bug was reproduced. Note that this test case might cause kernel freeze. Running with -c 5 will highly increase the likelihood of that. ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Ltp-list mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ltp-list
