On 05/11/25 14:59, Peter Zijlstra wrote: > On Wed, Nov 05, 2025 at 02:47:39PM +0100, Andrea Righi wrote: > > On Wed, Oct 29, 2025 at 08:08:37PM +0100, Andrea Righi wrote: > > > sched_ext tasks can be starved by long-running RT tasks, especially since > > > RT throttling was replaced by deadline servers to boost only SCHED_NORMAL > > > tasks. > > > > > > Several users in the community have reported issues with RT stalling > > > sched_ext tasks. This is fairly common on distributions or environments > > > where applications like video compositors, audio services, etc. run as RT > > > tasks by default. > > > > > > Example trace (showing a per-CPU kthread stalled due to the sway Wayland > > > compositor running as an RT task): > > > > > > runnable task stall (kworker/0:0[106377] failed to run for 5.043s) > > > ... > > > CPU 0 : nr_run=3 flags=0xd cpu_rel=0 ops_qseq=20646200 pnt_seq=45388738 > > > curr=sway[994] class=rt_sched_class > > > R kworker/0:0[106377] -5043ms > > > scx_state/flags=3/0x1 dsq_flags=0x0 ops_state/qseq=0/0 > > > sticky/holding_cpu=-1/-1 dsq_id=0x8000000000000002 dsq_vtime=0 > > > slice=20000000 > > > cpus=01 > > > > > > This is often perceived as a bug in the BPF schedulers, but in reality > > > schedulers can't do much: RT tasks run outside their control and can > > > potentially consume 100% of the CPU bandwidth. > > > > > > Fix this by adding a sched_ext deadline server, so that sched_ext tasks > > > are > > > also boosted and do not suffer starvation. > > > > > > Two kselftests are also provided to verify the starvation fixes and > > > bandwidth allocation is correct. > > > > Peter, Juri, this has now been tested quite extensively on our side and > > we're considering applying these patches to Tejun's sched_ext branch. > > > > Do you have any objections or concerns? > > Yeah, I want to finish this first: > > > https://lkml.kernel.org/r/[email protected] > > Because as is, the whole dl_server stuff isn't quite right. >
And I'm spending time on "[PATCH 04/11] sched/deadline: Add support to initialize and remove dl_server bandwidth" which I am still not 100% sure is correct (or that is correct how we handle setting runtime to 0 for fair_server today). Apologies, had some travelling and pto, but should be able to write something more about it in the next few days. Thanks, Juri
