linux/include
Tejun Heo 61debc251c sched_ext: Use per-CPU DSQs instead of per-node global DSQs in bypass mode
Bypass mode routes tasks through fallback dispatch queues. Originally a single
global DSQ, b7b3b2dbae ("sched_ext: Split the global DSQ per NUMA node")
changed this to per-node DSQs to resolve NUMA-related livelocks.

Dan Schatzberg found per-node DSQs can still livelock when many threads are
pinned to different small CPU subsets: each CPU must scan many incompatible
tasks to find runnable ones, causing severe contention with high CPU counts.

Switch to per-CPU bypass DSQs. Each task queues on its current CPU. Default
idle CPU selection and direct dispatch handle most cases well.

This introduces a failure mode when tasks concentrate on one CPU in
over-saturated systems. If the BPF scheduler severely skews placement before
triggering bypass, that CPU's queue may be too long to drain, causing RCU
stalls. A load balancer in a future patch will address this. The bypass DSQ is
separate from local DSQ to enable load balancing: local DSQs use rq locks,
preventing efficient scanning and transfer across CPUs, especially problematic
when systems are already contended.

v2: Clarified why bypass DSQ is separate from local DSQ (Andrea Righi).

Reported-by: Dan Schatzberg <schatzberg.dan@gmail.com>
Reviewed-by: Dan Schatzberg <schatzberg.dan@gmail.com>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
2025-11-12 06:43:44 -10:00
..
acpi More power management updates for 6.18-rc1 2025-10-07 09:39:51 -07:00
asm-generic hyperv-next for v6.18 2025-10-07 08:40:15 -07:00
clocksource clocksource/drivers/arm_arch_timer_mmio: Switch over to standalone driver 2025-09-23 12:31:50 +02:00
crypto This update includes the following changes: 2025-10-04 14:59:29 -07:00
cxl
drm drm/gpuvm: Fix kernel-doc warning for drm_gpuvm_map_req.map 2025-10-15 18:37:05 +02:00
dt-bindings There's a bunch of patches here across drivers/clk/ to migrate drivers to use 2025-10-07 09:28:37 -07:00
hyperv hyperv: Remove the spurious null directive line 2025-10-02 21:21:24 +00:00
keys KEYS: trusted_tpm1: Move private functionality out of public header 2025-09-27 21:05:06 +03:00
kunit linux_kselftest-kunit-6.18-rc1 2025-10-01 19:15:11 -07:00
kvm KVM: arm64: Kill leftovers of ad-hoc timer userspace access 2025-10-13 14:42:41 +01:00
linux sched_ext: Use per-CPU DSQs instead of per-node global DSQs in bypass mode 2025-11-12 06:43:44 -10:00
math-emu
media media: v4l2-common: Add legacy camera sensor clock helper 2025-09-09 15:59:19 +02:00
memory
misc
net net/ip6_tunnel: Prevent perpetual tunnel growth 2025-10-13 17:43:46 -07:00
pcmcia
ras
rdma
rv kernel-6.18-rc1.clone3 2025-09-29 10:36:50 -07:00
scsi SCSI misc on 20251002 2025-10-03 19:17:48 -07:00
soc There's a bunch of patches here across drivers/clk/ to migrate drivers to use 2025-10-07 09:28:37 -07:00
sound ASoC: tas2781: Support more newly-released amplifiers tas58xx in the driver 2025-10-13 11:08:09 +01:00
target
trace dma-mapping fixes for Linux 6.18: 2025-10-07 12:48:06 -07:00
uapi ARM: 2025-10-18 07:07:14 -10:00
ufs scsi: ufs: core: Include UTP error in INT_FATAL_ERRORS 2025-09-30 16:10:29 -04:00
vdso Updates for the VDSO subsystem: 2025-09-30 16:58:21 -07:00
video
xen xen: replace XENFEAT_auto_translated_physmap with xen_pv_domain() 2025-09-08 17:01:36 +02:00
Kbuild