From mboxrd@z Thu Jan 1 00:00:00 1970 From: khilman@deeprootsystems.com (Kevin Hilman) Date: Fri, 19 Oct 2012 16:58:36 -0700 Subject: [RFC] ARM: sched_clock: update epoch_cyc on resume In-Reply-To: (Barry Song's message of "Tue, 24 Jul 2012 14:43:28 +0800") References: <1342567672-29071-1-git-send-email-ccross@android.com> Message-ID: <87r4ouggk3.fsf@deeprootsystems.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Barry Song <21cnbao@gmail.com> writes: > 2012/7/18 Colin Cross : >> Many clocks that are used to provide sched_clock will reset during >> suspend. If read_sched_clock returns 0 after suspend, sched_clock will >> appear to jump forward. This patch resets cd.epoch_cyc to the current >> value of read_sched_clock during resume, which causes sched_clock() just >> after suspend to return the same value as sched_clock() just before >> suspend. >> >> In addition, during the window where epoch_ns has been updated before >> suspend, but epoch_cyc has not been updated after suspend, it is unknown >> whether the clock has reset or not, and sched_clock() could return a >> bogus value. Add a suspended flag, and return the pre-suspend epoch_ns >> value during this period. > > Acked-by: Barry Song <21cnbao@gmail.com> > > this patch should also fix the issue that: > 1. launch some rt threads, rt threads sleep before suspend > 2. repeat to suspend/resume > 3. after resuming, waking up rt threads > > repeat 1-3 again and again, sometimes all rt threads will hang after > resuming due to wrong sched_clock will make sched_rt think rt_time is > much more than rt_runtime (default 950ms in 1s). then rt threads will > lost cpu timeslot to run since the 95% threshold is there. Re-visiting this in light of a related problem. I've run into a similar issue where IRQ threads are prevented from running during resume becase the RT throttling kicks because RT runtime is accumulated during suspend. Using the 'needs_suspend' version fixes this problem too. However, because of the RT throttling issue, it seems like *all* platforms should be using the 'needs_suspend' version always. But, as already pointed out, that makes the timed printk output during suspend/resume rather unhelpful. Having to choose between useful printk times during suspend/resume and functioning IRQ threads during suspend/resume isn't a choice I want to make. I'd rather have both. Any ideas? Kevin