I think the patchset is from https://lore.kernel.org/all/20211006111852.1514359-1-bigeasy@xxxxxxxxxxxxx/ Now we use RT5.10. So we port the patches from the v5.10-rt-rebase. But it only use 3 patches compared to original one. https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/log/?h=v5.10-rt-rebase irq_work: Also rcuwait for !IRQ_WORK_HARD_IRQ on PREEMPT_RT Sebastian Andrzej Siewior 2 -2/+9 irq_work: Handle some irq_work in a per-CPU thread on PREEMPT_RT Sebastian Andrzej Siewior 3 -43/+106 irq_work: Allow irq_work_sync() to sleep if irq_work() no IRQ support. Sebastian Andrzej Siewior 2 -1/+19 In V5.15-rt-rebase, it uses 4 patches. https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/log/?h=v5.15-rt-rebase irq_work: Also rcuwait for !IRQ_WORK_HARD_IRQ on PREEMPT_RT Sebastian Andrzej Siewior 2 -2/+9 irq_work: Handle some irq_work in a per-CPU thread on PREEMPT_RT Sebastian Andrzej Siewior 1 -12/+106 irq_work: Allow irq_work_sync() to sleep if irq_work() no IRQ support. Sebastian Andrzej Siewior 2 -0/+13 sched/rt: Annotate the RT balancing logic irqwork as IRQ_WORK_HARD_IRQ Sebastian Andrzej Siewior 1 -1/+1 But I also try port those 4 patches to RT5.10. I also can reproduce this issue. After I try to set the value -1 to sched_rt_runtime_us and run the test, I got the same result (screen freeze) root@intel-corei7-64:/proc/sys/kernel# cat sched_rt_runtime_us 950000 root@intel-corei7-64:/proc/sys/kernel# echo -1 > sched_rt_runtime_us root@intel-corei7-64:/proc/sys/kernel# cat sched_rt_runtime_us -1 Angus -----Original Message----- From: Ahmed S. Darwish <a.darwish@xxxxxxxxxxxxx> Sent: Tuesday, March 29, 2022 1:22 PM To: Chen, Angus <angus.chen@xxxxxxxxx> Cc: linux-rt-users@xxxxxxxxxxxxxxx; bigeasy@xxxxxxxxxxxxx; Chang, Junxiao <junxiao.chang@xxxxxxxxx>; Ung, Teng En <teng.en.ung@xxxxxxxxx>; tglx@xxxxxxxxxxxxx Subject: Re: RT kernel video playback issue (screen freeze) Am Tue, Mar 29, 2022, schrieb Chen, Angus: > > When we playing a H265 file on our Tiger Lake platform (OS : Yocto, > CPU : Intel® CoreT i7-1185GRE Processor (12M Cache, up to 4.40 GHz)), > we encounter a system hang/freeze issue (without enabling GuC > submitting) > > We use the following command to reproduce this issue. > [gst-launch snippet]... > > We also can reproduce this issue on upstream v5.15.27-rt35. > > The dmesg log from UART port. After the log showed "the sched: RT > throttling activated", the screen freeze. > First, are you running a custom RT workload that has gone rogue; e.g., your RT application got stuck in an infinite loop? Second, were you by any chance running gst-launch with RT priorities? Anyway, before running the gst-launch snippet above, can you also try first, under root: echo -1 > /proc/sys/kernel/sched_rt_runtime_us and see if it helps? > > > After bisecting, we found that it was related to the following > patchset > > https://lkml.org/lkml/2021/10/6/367 > In general, please avoid using https://lkml.org. It's a private service that obscures the unique email identifier, Message-ID, within its URL. Use the now official "https://lore.kernel.org/"; instead, which keeps the Message-ID field explicit. Anyway, I always got a Cloudflare-404 when accessing the lkml.org link above, sorry. Good luck, -- Ahmed S. Darwish Linutronix GmbH
