File scx.changes of Package scx
-------------------------------------------------------------------
Thu Apr 03 07:49:22 UTC 2025 - fredrik.lonnegren@suse.com
- Update to version 1.0.11:
* scxtop: Disable bpf conditional attachment
* Add scxctl into tools
* scx_bpfland: Fix idle detection when SMT is disabled
* scx_layered: add node spread growth algo
* scx_p2dq: Add config for minimum queue size for pick2
* scx_p2dq: Add config for min runs for pick2
* scx_p2dq: Add stats for pick2 on dispatch/select paths
* scx_p2dq: Allow disabling pick2 load balancing on dispatch
* scx_p2dq: Add option to only load balance max DSQ tasks
* scx_flash: simplify scheduling policy
* scx_p2dq: Add helper to scale vtime by task weight
* scxtop: Fix DSQ initialization
* scx_layered: add protected layer flag
* scx_p2dq: Add min_nr_queued_pick2 check on pick_two_llc_ctx
* scx_bpfland: handle per-CPU tasks preempted by a higher scheduling class
* scx_lavd: Respect the core preference and some more.
* scx_layered: add cpus_pct cfg param
* scx_flash: mitigate stalls from higher priority scheduling classes
* scx_p2dq: Add idle resume QoS config
* scx_layered: Add idle qos config
* scxtop: adding hw_pressure event trace.
* scx_layered: add node spread reverse, swap it with node spread
* scx_layered: make nodespread account for smt
* scx_p2dq: Fix vtime handling in enqueue path
* scxtop: Add checks when updating DSQ latency
* scx_layered: More optimal core allocation
* scxtop: Add fork event tracing
* scx_p2dq: Cleanup pick_two_cpu selection
* scx_p2dq: Update default scheduler config
* scxtop: trace on_sched_exec event.
* scx_layered: add exclude match rule
* scx_lavd: Make the idle CPU selection compute domain aware.
* scx_lavd: Enable SCX_OPS_ALLOW_QUEUED_WAKEUP and misc.
* scx_p2dq: Fix vtime accounting
* scxtop: Add key bindings to toggle hw pressure
* scxtop: Add exit event tracing
* scx_bpfland: Enable SCX_OPS_ALLOW_QUEUED_WAKEUP
* scx_bpfland: Introduce --idle-resume-us
* scx_p2dq: Add support for SCX_OPS_ALLOW_QUEUED_WAKEUP
* scxtop: fix clock_sync_failure errors
* scxtop: adding cpu hotplug event on exit trace event.
* scx_layered: add NodeSpreadRandom growth algo
* scx_layered: Fix accounting, reporting and CPU allocation fluctuations
* scx_lavd: Better control and handling of per-CPU tasks.
* scx_bpfland: Improve NUMA node locality
* scx_p2dq: Fix stalls caused by improper handling of affinitized tasks
* scx_bpfland: Introduce an option to disable SMT awareness
* scx_lavd: Fix incorrect task affinity testing and more.
* scx_bpfland: Fix direct dispatch migration
* scx_p2dq: Fix affinitized task handling
* scx_lavd: Revise task affinity testing.
* scx_p2dq: Fix handling for two LLCs
* scx_p2dq: Update default load balancing config
* scx_layered: remove extra call to test_and_clear_cpu in pick_idle_from
* scx_chaos: init with basic functionality
* scx_p2dq: Improve locality for affinitized tasks
* scx_p2dq: Avoid aggressive return on cpu selection
* scx_bpfland: NUMA load balancer
-------------------------------------------------------------------
Fri Mar 07 10:17:37 UTC 2025 - fredrik.lonnegren@suse.com
- remove fix-aarch64.patch: fixed upstream
- Update to version 1.0.10:
* scxtop: adding one perf event type L1-dcache-load-misses.
* scx_p2dq: Add scx_p2dq scheduler
* scxtop: track sched_waking events
* scx_p2dq: Fix vtime handling and cleanup
* scx_layered: making unexpected gpu error smoother.
* scx_layered: Remove CPU shuffling when idle CPU is not found in layered_select_cpu()
* scx_layered: enable creating scx appimage
* scx_wd40: Introduce wd40 scheduler
* scxtop: make q exit help if in help instead of quitting
* scxtop: track number of dropped events and render it
* scx_p2dq: Add perf boost to interactive tasks
* scx_layered: Update enqueue path to use direct dispatch into foreign local DSQs
* scxtop: Add softirq events
* scx_layered: Fix pick_idle_cpu() sometimes returning a disallowed CPU
* scx_lavd: Avoid scx_bpf_kick_cpu(SCX_KICK_PREEMPT) for self-preemption
* scx_p2dq: Update load balancing
* scx_wd40: add cpumask library
* scx_layered: Enable queued wakeup optimization
* scx_loader: Add p2dq scheduler
* scx_lavd: Fix the performance regression in recent changes.
* scxtop: propose to give the ability to focus on one process.
* scx: Don't panic on missing enums in C schedulers
* scx_p2dq: Fix verifier issue on big/little topology
* scx_p2dq: Add autoslice tuning
* scxtop: add experimental_long_tail_tracing
* scx_p2dq: Update pick2 migration logic
* scx_bpfland: disable auto local dispatch for migration-disabled tasks
* scxtop: Add IPI events
* scxtop: __bpf_ringbuf_reserve() failure set to ENOMEM instead.
* scx_layered: add debug logging for gpu support and fix UsingGpu matcher
* scx_p2dq: Add initial wake sync support
* scxtop: Make tokio runtime options configurable
* scx_layered: change sysinfo version to 0.30
* scx_utils: use clsuter_id to detect if big.LITTLE CPUs
* scx_bpfland: avoid creating redundant cache domains
* scx_layered: fix gpu matcher bugs
* scxtop: Add localization on number formatting
* scx_bpfland: reduce task migrations
* sched_ext: Add NUMA-aware idle CPU helpers
* scx_bpfland: Always re-enqueue tasks that are running on a busy SMT core
* scxtop: Add config file
* scx_utils: gpu topology adding nearest available devices.
* scx_bpfland: Skip idle CPU selection when system is fully busy
* scx_bpfland; use scx_bpf_now() consistently
* scx_bpfland: fix non-SMT CPU busy state detection
* scx_lavd: Use tracepoints for futex for reliable tracing
* scx_lavd: Fix BPF verifier errors in old kernels.
* scx_p2dq: Handle invalid load sum calculations
* scxtop: Add keymap configuration
* scxtop: add logging to file with panics
* scxtop: clamp gauge ratio
* scxtop: use smartstr with SSO for comms
* scxtop: Add separate subcommand for TUI and shell completion
* scx_layered: fix gpu matcher
* scxtop: Set correct max on LLC sparkline view
* scxtop: Add EventDispatchManager for handling events
* scx_bpfland: NUMA-aware scheduling
* scxtop: Fix max sparkline value across NUMA nodes
* scx_layered: add stats tracking when antistall is ran
* layered: update documentation for recently added options
* scx_utils: using TopologyLevel::Internal nearest gpu discovery mode.
* scxtop: adding gpu_mem_total tracepoint.
* scxtop: adding cpu hotplug (enter) event tracepoint.
* scxtop: Conditionally attach GPU probes
* scx_flash: drop user lock priority boosting
* scx_bpfland: fix cache domain detection
* scxtop: Filter sched_switch events less aggressively
* scxtop: Allow attaching hardware specific perf events
* scxtop: stop traces with bpf timer
* scxtop: always send bpf event to terminate trace
* scx_layered: expand kprobes used for gpu support
* scxtop: Add trace command
* scx_utils, scx_lavd: Enable futex holder boosting conditionally.
* scx_bpfland: Fix is_wake_sync()
* scx_utils: Fix kernels without CONFIG_DEBUG_FS
* scxtop: making few kernel probes conditional.
* scheds: Introduce scx_tickless
* scx_layered: Improve kthread handling
* scx_layered: Implement IsKthread match
-------------------------------------------------------------------
Tue Feb 11 08:54:08 UTC 2025 - Fredrik Lönnegren <fredrik.lonnegren@suse.com>
- add fix-aarch64.patch: Add patches for building for aarch64
* https://github.com/sched-ext/scx/issues/1320
-------------------------------------------------------------------
Wed Feb 05 08:41:57 UTC 2025 - fredrik.lonnegren@suse.com
- Update to version 1.0.9:
* versions: Synchronize crate versions to the latest used
* scx_lavd: Use BPF CO-RE for SCX_ENQ_CPU_SELECTED
* scx_layered: make gpu poll interval configurable
* scx_prev: a simple scheduler tested on OLTP workloads
* scx_lavd: Don't try preemption for greedy tasks
* scxtop: Add trace options
* scx_central: alert for 'silly' values for the chosen cpu from command line.
* scx_lavd: Preemption when ops.select_cpu() was skipped
* scxtop: Add support for generating perfetto traces
* scx_bpfland: enable task preemption
* scx_layered: add gpu matcher
* rust schedulers: for sibling cpu using static dispatch instead.
* scx_bpfland: reduce heap allocation with energy profile/primary domain.
* scx_bpfland: Mark --local-kthreads/-k as experimental
* scx_loader: Update performance profiles for scx_bpfland
* scxtop: Fix layout for sparklines
* scxtop: Add scx_stats support
* scx_lavd: Conder the hardware resource sharing level in CPU ordering
* scxtop: Fill render area by setting max events
* scxtop: Add event to default list after selection
* scx_layered: fixes for MATCH_SCXCMD_JOIN
* scxtop: Fix scroll on page up
* scxtop: update help display with last changes (perf event scrolling).
* scx_bpfland: Introduce --cpufreq to enable CPU frequency scaling
* scx_bpfland: Introduce --local-pcpu to prioritize per-CPU tasks
* scx_bpfland: Get rid of the interactive task classification via nvcsw
* scx_bpfland: Allow tasks to overflow primary domain more aggressively
* scx_bpfland: Virtual deadline with partial execution vruntime
* scx_bpfland: Do not scale slice lag with the task weight
* scx_bpfland: Set max performance level with balance_performance
* scx_lavd: Direct dispatch at ops.enqueue()
* scx_lavd: Do not reserve an idle cpu at ops.enqueue()
* scx_loader: Avoid race condition with start_scheduler
* scx_lavd: Use time_delta() for clock difference calculation
* scxtop: Add ability to select perf events
* scxtop: proposing to add bus-cycles event.
* scx_layered: add 'join layer' command
* scx_central: fix cpu affinity corrupted allocation (glibc).
* scx_rusty: fix custom PartialOrd implementation build warning.
* scx_layered: Fix dump_layer_cpumask()
* scx_rustland_core: fix static mutable reference usage warning (BUF).
* scxtop: Collect uncore frequency
* scx_rusty: use lb_domain naming for load balancing domains
* scxtop: Fix node barchart
* scx_bpfland: avoid starvation of per-CPU tasks
* scxtop: Fix attach on newer kernels
* scxtop: Add CPU frequency to default sparkline view
* scxtop: Add barchart to node view
* scx_bpfland: re-enqueue tasks when a CPU is taken by a higher sched class
* scx_bpfland: give queued tasks a chance to run on the same CPU
* scx_bpfland: do not dispatch per-CPU kthreads directly by default
* scx_bpfland: proactively wake idle CPUs after task enqueue
* scx_layered: Add pid namespace layer matching
* scx_bpfland: do not allow per-CPU kthread to preempt other tasks
* scxtop: Add scheduler view to llc/node views
* scx_lavd: Add a fast path for a migration-disabled task in pick_idle_cpu()
* scxtop: Add scxtop tui
* scx_flash: prevent scheduling bubbles on kernels >= 6.12
* scx_rusty: prevent scheduling bubbles on kernels >= 6.12
* scx_bpfland: use nvcsw metrics only for cpufreq scaling
* scx_bpfland: evaluate deadline based on vruntime and task's sum_vruntime
* scx_bpfland: consider sum_runtime instead of avg_runtime
* scx_bpfland: never actively wake-up idle CPUs
* scx_flash: Option for userspace lock boosting
* scx_bpfland: support !CONFIG_SMP kernels
* scx_rusty: use arena allocations instead of maps for tasks
* scx_rusty: use map to manage task masks
* scx_layered: Fix multplication overflows
* scx_layered: Improve open layer prot[_preempt] accounting
* scx_layered: Don't apply preempt protection between two open layers
* scx_layered: Disable owner protection for preempt layers
* scx_rusty: remove per-task temp mask and replace with per-CPU singleton
* scx_layered: Implement layer property disallow_preempt_after_us
* scx_lavd: Prioritize a migration-disabled task
* scx_lavd: Ensure to check all compute domains when task stealing
* scx_lavd: Kick an idle cpu as early as possible on ops.select_cpu()
* scx_lavd: Change the default time slice to 5 msec
* scx_lavd: Add a fast path to pick_idle_cpu() when overloaded
* scx_lavd: Turn on frequency scaling in performance mode
* scx_lavd: Properly reinitialize cpumask for autopilot mode
* scx_rustland_core: Select any idle CPU when task affinity changes
-------------------------------------------------------------------
Tue Dec 17 09:35:57 UTC 2024 - fredrik.lonnegren@suse.com
- Update to version 1.0.8:
* scx_layered: Put all tasks with custom affinities into the hi fallback DSQs
* scx_bpfland: do not rely on scx_utils::autopower
* scx_layered: Make low fallback DSQs useful and other changes
* scx_layered: handle nr_to_free calculation
* scx_layered: Implement empty LLC draining
* scx_layered: queued_runtime tracking bug fix and disable xllc_mig_min_us
* scx_bpfland: server workload improvements
* scx_loader: add mode for server-oriented workloads
* scx_layered: Improve layer core growth algos
* scx_layered: Make layered work in pid namespaces
* scx_loader: fix recursion of type convertion
* scx_layered: Add pid namespace layer matching
-------------------------------------------------------------------
Sat Dec 07 13:29:51 UTC 2024 - fredrik.lonnegren@suse.com
- Update to version 1.0.7:
* scx_layered: Add fallback DSQ cost accounting
* scx_layered: Fix verifier issue when tracing
* scx_lavd: update cur_logical_clk atomically
* scx_loader: tune scx_bpfland default options
* scx_layered: Fix verifier issue on older kernels
* scx_layered: Use layer idle_smt option
* scx_layered: Add fallback costs to dump
* scx_layered: Work around older kernels choking on function calls
* scx_layered: add timer antistall
* scx_layered: Fix error in dispatch consumption
* scx_layered: Add helper for layer slice duration
* scx_layered: Fix cost accounting for fallback dsqs
* scx_layered: Consume from local LLCs for dispatch
* scheds: introduce scx_flash
* scx_layered: Perf improvements and a bug fix
* scx_layered: Don't limit antistall execution to layered_cpumask
* scx_layered: Fix verifier issues on older kernels
* scx_layered: Cleanups around topology handling
* scx_lavd: Factor the task's runtime more aggressively in a deadline
calculation
* scx_layered: Work around verification failure in antistall_set()
* scx_loader: add scx_flash as supported scheduler
* scx_layered: Add netdev IRQ balancing
* scx_layered: Use PROG_RUN for cpumask updates
* scx_lavd: Optimize the layout of struct task_ctx
* scx_layered: Remove high fallback dsq budget check
* scx_lavd: Do not exclude exiting tasks
* scx_rustland_core: proactively wake up CPUs when selected by user space
* scx_layered: Reimplement CPU allocation and some other fixes
* scx_rusty: Fix verifier errors on older kernels
* scx_layered: select_cpu() fixes and updates
* scx_lavd: Boost time slice more generously
* scx_lavd: Optimize performance criticality calculation
* scx_lavd: Optimize the cpuc_ctx layout for cache friendliness
* scx_lavd: Optimize preemption
* scx_bpfland: restart on energy profile change
* scx_lavd: add tunables for adjusting time slices
* scx_layered: Track owned/open execution times and per-LLC-layer stats
* scx_loader: restart scheduler upon fail
* scx_lavd: Limit the slice extension of a lock holder
* scx_layered: Implement in-layer execution protection to replace cost based
fairness
* scx_bpfland: dump cache_id_map in ascending order
* scx_layered: Fix idle selection on big/little
* scx_layered: Prioritize sched userspace and fix owned execution protection
* scx_layered: State tracking updates and layer sizing related fixes
* scx_lavd: Load balancing across compute domains
* scx_layered: Deprecate idle_smt layer config
* scx_lavd: Fetch active profile from power-profiles-daemon dbus interface
when using autopower
* scx_layered: Make vtime_now per-LLC
* scx_layered: Implement layer config xllc_mig_min_us
* scx_utils: Extend topology to support Snapdragon X Elite
* scx_rusty: Refactor access operation of nodemask
* scx_stats,scx_lavd: describe non-nested stats meta
* scx_lavd: Reset per-CPU preemption information when a CPU is released
* scx_layered: Fix vtime and llc_id handling bugs
* scx_layered: Fix a silly bug in select_cpu() and implement per-layer fifo
mode
-------------------------------------------------------------------
Mon Nov 18 15:49:21 UTC 2024 - fredrik.lonnegren@suse.com
- Update to version 1.0.6:
* scx_lavd: support CPU hotplug correctly
* scx_rusty: fix single dom short-circuit
* scx_layered: Add per layer time slices to stats
* scx_layered: Refactor topology algorithms to a separate module
* scx_layered: Add per layer weights
* scx_bpfland: rework lowlatency mode
* scx_layered: Add big cpumask
* scx_layered: Use idle smt mask for idle selection
* scx_layered: Improve perf on non topo aware paths
* scx_rusty: Fix BPF crash during CPU hotplug
* scx_layered: Update idle topology selection order
* scx_layered: lighten/reduce nested loops in layered dispatch
* scx_layered: Add stress-ng example layer
* scx_layered: Make stress-ng non exclusive in example
* scx_bpfland: prevent per-CPU DSQ stall with per-CPU kthreads
* scx_layered: enable configuring layer iteration when no topo
* [rusty] Fix load stats when host is under-utilized
* scx_layered: Update CI to show stats
* layered: split dispatch into no_topo version
* scx_layered: Rename load_adj statistic
* layered: attempt to work steal from own llc before others
* scx_layered: Fix verifier errors
* scx_layered: Cleanup debug messages
* scx_bpfland: fix cpumask initialization error
* scx_layered: Change default DSQ iter algo
* scx_layered: Cleanup topology preempt path
* scx_layered: Implement reverse weight DSQ algorithm
* scx_layered: make default value for disable_topology dynamic
* scx_lavd: mitigate the lock holder preemption problem
* scx_rustland_core: use handle_mm_fault kprobe
* scx_bpfland: drop per-cpu DSQs
* scx_lavd: do not inspect scx_lavd process itself
* scx_layered: Remove layer iteration
* scx_mitosis: handle enqueue() on !wakeup
* scx_mitosis: Handle pinned tasks
* scx_layered: add RandomTopo layer growth algorithm
* scx_layered: make disable_topology arg require equals
* scx_bpfland: rework lowlatency mode
* scx_layered: Fix crash on aarch64 due to unavailable cache id file
* scx_lavd: add missing reset_lock_futex_boost()
* scx_rustland: Adjust task's vruntime budget based on latency weight
* scx_layered: fix exit_task ctx lookup err
* scx_lavd: fix/work around a verifier error
* scx_lavd: various optimizations for more consistent performance
* scx_loader: introduce configuration
* scx_layered: Add monitor
* scx_layered: Fix declarations in timer
* scx_lavd: fix uninitialized memory access at comp_preemption_info()
* scx_layered: support verifying on older kernels and fix logic
* scx_layered: Add cost accounting
* scx_lavd: optimize preemption
* scx_layered: Refactor dispatch
* scx_lavd: tuning and optimizing latency criticality calculation
* scx_layered: Add layer name to bpf
* scx_layered: Add layer CPU cost to dump
* scx_lavd: Correct the type of taskc within lavd_dispatch()
* scx_lavd: optimize ops.dispatch() and DSQ placement
* scx_layered: Fix dump output format
* scx_layered: Add additional drain to fallback DSQs
* scx_layered: Fix trace format
* scx_layered: point costc to global struct when initializing budgets
* scx_lavd: entirely drop kernel lock tracing
* scx_rusty: Restore push domain tasks when no task found
* scx_lavd: fix a performance regression bug
(perf bench -f simple sched messaging)
-------------------------------------------------------------------
Tue Oct 29 18:02:44 UTC 2024 - Fredrik Lönnegren <fredrik.lonnegren@suse.com>
- Move scx-<version>.tar.zst to buildtime service.
-------------------------------------------------------------------
Tue Oct 22 08:48:50 UTC 2024 - Fredrik Lönnegren <fredrik.lonnegren@suse.com>
- Add devel package that includes the common headers from the upstream
repository.
-------------------------------------------------------------------
Thu Oct 3 19:00:46 UTC 2024 - Fredrik Lönnegren <fredrik.lonnegren@suse.com>
- Update to v1.0.5:
* scx_rustland_core: introduce topology awareness
* scx_layered: clean up Layer::new layer_growth_algo
* scx_lavd: improve greedy ratio calculation and more
* scx_rustland_core: move includes back to the lib section
* scx_bpfland: use sum_exec_runtime to evaluate task's used time slice
* scx_layered: Fix typo in stats
* scx_layered: Pass layer spec for core growth algo
* scx_rusty: init domains when calculating averages
* scx_layered: Add random layer growth algo by
* scx_layered: Add topology aware core growth selection
* scx_layered: Add stats for XNUMA/XLLC migrations
* scx_lavd: add a short circuit for the case of no turbo core
* scx_lavd: boost the latency-criticality of kernel threads
* scx_bpfland: refine idle CPU selection logic
* scx_layered: add round robin growth strategy
* scxstats_to_openmetrics: fix format string
* scx_rustland_core: improve idle CPU selection API and logic
* scx_rustland_core: prevent CI failures
* scx_rustland_core: Access the returned value of saturating_sub()
* scx_layered: Refactor match_layer() and implement helper function to access
cpumask within bpf_cpumask
* scx_bpfland: Remove the usage of cast_mask in bpfland_enqueue
* scx_lavd: consider waker's CPU when ops.select_cpu()
* scx_layered: Add a hi fallback dsq per llc
* scx_layered: Add Big/Little core growth algos
* scx_layered: Add topology aware preemption
* scx_lavd: find a victim cpu for preemption within task's compute domain
* scx_layered: Add waker stats per layer
* scx_layered: Cleanup dump format
* scx_lavd: propagate waker's latency criticality to its wakee
* scx_layered: Restrict preemption to layer cpumask
* scx_layered: Make layered idle CPU selection topology aware
* scx_rustland_core: fix mm stall
* scx_loader: Add initial automatic scheduler switching via --monitor-no-dbus
* scx_layered: Add layer growth algo to layer bpf config
* Sync from kernel and re-enable scx_flatcg and scx_pair
* scx_layered: Fix idle core selection
* scx_loader: Add systemd service and on-DBUS launch
* scx_lavd: more accurately determine the performance criticality threshold
* scx_lavd: fix two potential bugs
* scx_layered: Fix cache initialization
* Revert "scx_rustland_core: prevent deadlock with per-CPU DSQs and CPU
* scx_bpfland small fixes and improvements
* scx_stats: Implement macro #stat_doc to autogen doc from stat desc
* scx_lavd: Fix typo
