scx_bypass_lb_{donee,resched}_cpumask were file-scope statics shared by all
scheduler instances. With CONFIG_EXT_SUB_SCHED, multiple sched instances
each arm their own bypass_lb_timer; concurrent bypass_lb_node() calls RMW
the global cpumasks with no lock, corrupting donee/resched decisions.
Move the cpumasks into struct scx_sched, allocate them alongside the timer
in scx_alloc_and_add_sched(), free them in scx_sched_free_rcu_work().
Fixes: 95d1df610c ("sched_ext: Implement load balancer for bypass mode")
Cc: stable@vger.kernel.org # v6.19+
Reported-by: Chris Mason <clm@meta.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Move enforcement of SCX context-sensitive kfunc restrictions from per-task
runtime kf_mask checks to BPF verifier-time filtering, using the BPF core's
struct_ops context information.
A shared .filter callback is attached to each context-sensitive BTF set
and consults a per-op allow table (scx_kf_allow_flags[]) indexed by SCX
ops member offset. Disallowed calls are now rejected at program load time
instead of at runtime.
The old model split reachability across two places: each SCX_CALL_OP*()
set bits naming its op context, and each kfunc's scx_kf_allowed() check
OR'd together the bits it accepted. A kfunc was callable when those two
masks overlapped. The new model transposes the result to the caller side -
each op's allow flags directly list the kfunc groups it may call. The old
bit assignments were:
Call-site bits:
ops.select_cpu = ENQUEUE | SELECT_CPU
ops.enqueue = ENQUEUE
ops.dispatch = DISPATCH
ops.cpu_release = CPU_RELEASE
Kfunc-group accepted bits:
enqueue group = ENQUEUE | DISPATCH
select_cpu group = SELECT_CPU | ENQUEUE
dispatch group = DISPATCH
cpu_release group = CPU_RELEASE
Intersecting them yields the reachability now expressed directly by
scx_kf_allow_flags[]:
ops.select_cpu -> SELECT_CPU | ENQUEUE
ops.enqueue -> SELECT_CPU | ENQUEUE
ops.dispatch -> ENQUEUE | DISPATCH
ops.cpu_release -> CPU_RELEASE
Unlocked ops carried no kf_mask bits and reached only unlocked kfuncs;
that maps directly to UNLOCKED in the new table.
Equivalence was checked by walking every (op, kfunc-group) combination
across SCX ops, SYSCALL, and non-SCX struct_ops callers against the old
scx_kf_allowed() runtime checks. With two intended exceptions (see below),
all combinations reach the same verdict; disallowed calls are now caught at
load time instead of firing scx_error() at runtime.
scx_bpf_dsq_move_set_slice() and scx_bpf_dsq_move_set_vtime() are
exceptions: they have no runtime check at all, but the new filter rejects
them from ops outside dispatch/unlocked. The affected cases are nonsensical
- the values these setters store are only read by
scx_bpf_dsq_move{,_vtime}(), which is itself restricted to
dispatch/unlocked, so a setter call from anywhere else was already dead
code.
Runtime scx_kf_mask enforcement is left in place by this patch and removed
in a follow-up.
Original-patch-by: Juntong Deng <juntong.deng@outlook.com>
Original-patch-by: Cheng-Yang Chou <yphbchou0911@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
scx_kf_allowed_if_unlocked() uses !current->scx.kf_mask as a proxy for "no
SCX-tracked lock held". kf_mask is removed in a follow-up patch, so its two
callers - select_cpu_from_kfunc() and scx_dsq_move() - need another basis.
Add a new bool scx_rq.in_select_cpu, set across the SCX_CALL_OP_TASK_RET
that invokes ops.select_cpu(), to capture the one case where SCX itself
holds no lock but try_to_wake_up() holds @p's pi_lock. Together with
scx_locked_rq(), it expresses the same accepted-context set.
select_cpu_from_kfunc() needs a runtime test because it has to take
different locking paths depending on context. Open-code as a three-way
branch. The unlocked branch takes raw_spin_lock_irqsave(&p->pi_lock)
directly - pi_lock alone is enough for the fields the kfunc reads, and is
lighter than task_rq_lock().
scx_dsq_move() doesn't really need a runtime test - its accepted contexts
could be enforced at verifier load time. But since the runtime state is
already there and using it keeps the upcoming load-time filter simpler, just
write it the same way: (scx_locked_rq() || in_select_cpu) &&
!kf_allowed(DISPATCH).
scx_kf_allowed_if_unlocked() is deleted with the conversions.
No semantic change.
v2: s/No functional change/No semantic change/ - the unlocked path now acquires
pi_lock instead of the heavier task_rq_lock() (Andrea Righi).
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
SCX_ENQ_IMMED makes enqueue to local DSQs succeed only if the task can start
running immediately. Otherwise, the task is re-enqueued through ops.enqueue().
This provides tighter control but requires specifying the flag on every
insertion.
Add SCX_OPS_ALWAYS_ENQ_IMMED ops flag. When set, SCX_ENQ_IMMED is
automatically applied to all local DSQ enqueues including through
scx_bpf_dsq_move_to_local().
scx_qmap is updated with -I option to test the feature and -F option for
IMMED stress testing which forces every Nth enqueue to a busy local DSQ.
v2: - Cover scx_bpf_dsq_move_to_local() path (now has enq_flags via ___v2).
- scx_qmap: Remove sched_switch and cpu_release handlers (superseded by
kernel-side wakeup_preempt_scx()). Add -F for IMMED stress testing.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Add SCX_ENQ_IMMED enqueue flag for local DSQ insertions. Once a task is
dispatched with IMMED, it either gets on the CPU immediately and stays on it,
or gets reenqueued back to the BPF scheduler. It will never linger on a local
DSQ behind other tasks or on a CPU taken by a higher-priority class.
rq_is_open() uses rq->next_class to determine whether the rq is available,
and wakeup_preempt_scx() triggers reenqueue when a higher-priority class task
arrives. These capture all higher class preemptions. Combined with reenqueue
points in the dispatch path, all cases where an IMMED task would not execute
immediately are covered.
SCX_TASK_IMMED persists in p->scx.flags until the next fresh enqueue, so the
guarantee survives SAVE/RESTORE cycles. If preempted while running,
put_prev_task_scx() reenqueues through ops.enqueue() with
SCX_TASK_REENQ_PREEMPTED instead of silently placing the task back on the
local DSQ.
This enables tighter scheduling latency control by preventing tasks from
piling up on local DSQs. It also enables opportunistic CPU sharing across
sub-schedulers - without this, a sub-scheduler can stuff the local DSQ of a
shared CPU, making it difficult for others to use.
v2: - Rewrite is_curr_done() as rq_is_open() using rq->next_class and
implement wakeup_preempt_scx() to achieve complete coverage of all
cases where IMMED tasks could get stranded.
- Track IMMED persistently in p->scx.flags and reenqueue
preempted-while-running tasks through ops.enqueue().
- Bound deferred reenq cycles (SCX_REENQ_LOCAL_MAX_REPEAT).
- Misc renames, documentation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
scx_disable() directly called kthread_queue_work() which can acquire
worker->lock, pi_lock and rq->__lock. This made scx_disable() unsafe to
call while holding locks that conflict with this chain - in particular,
scx_claim_exit() calls scx_disable() for each descendant while holding
scx_sched_lock, which nests inside rq->__lock in scx_bypass().
The error path (scx_vexit()) was already bouncing through irq_work to
avoid this issue. Generalize the pattern to all scx_disable() calls by
always going through irq_work. irq_work_queue() is lockless and safe to
call from any context, and the actual kthread_queue_work() call happens
in the irq_work handler outside any locks.
Rename error_irq_work to disable_irq_work to reflect the broader usage.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Add a dedicated scx_dump_lock and per-sched dump_disabled flag so that
debug dumping can be safely disabled during sched teardown without
relying on scx_sched_lock. This is a prep for the next patch which
decouples the sysrq dump path from scx_sched_lock to resolve a lock
ordering issue.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
While running scx_flatcg, dmesg prints "SCX_OPS_HAS_CGROUP_WEIGHT is
deprecated and a noop", in code, SCX_OPS_HAS_CGROUP_WEIGHT has been
marked as DEPRECATED, and will be removed on 6.18. Now it's time to do it.
Signed-off-by: Zhao Mengmeng <zhaomengmeng@kylinos.cn>
Signed-off-by: Tejun Heo <tj@kernel.org>
SCX_ENQ_REENQ indicates that a task is being re-enqueued but doesn't tell the
BPF scheduler why. Add SCX_TASK_REENQ_REASON flags using bits 12-13 of
p->scx.flags to communicate the reason during ops.enqueue():
- NONE: Not being reenqueued
- KFUNC: Reenqueued by scx_bpf_dsq_reenq() and friends
More reasons will be added.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Add infrastructure to pass flags through the deferred reenqueue path.
reenq_local() now takes a reenq_flags parameter, and scx_sched_pcpu gains a
deferred_reenq_local_flags field to accumulate flags from multiple
scx_bpf_dsq_reenq() calls before processing. No flags are defined yet.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Wrap the deferred_reenq_local_node list_head into struct
scx_deferred_reenq_local. More fields will be added and this allows using a
shorthand pointer to access them.
No functional change.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
The deferred reenqueue local mechanism uses an llist (lockless list) for
collecting schedulers that need their local DSQs re-enqueued. Convert to a
regular list protected by a raw_spinlock.
The llist was used for its lockless properties, but the upcoming changes to
support remote reenqueue require more complex list operations that are
difficult to implement correctly with lockless data structures. A spinlock-
protected regular list provides the necessary flexibility.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Global DSQs are currently stored as an array of scx_dispatch_q pointers,
one per NUMA node. To allow adding more per-node data structures, wrap the
global DSQ in scx_sched_pnode and replace global_dsqs with pnode array.
NUMA-aware allocation is maintained. No functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Reviewed-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Add rhashtable-based lookup for sub-schedulers indexed by cgroup_id to
enable efficient scheduler discovery in preparation for multiple scheduler
support. The hash table allows quick lookup of the appropriate scheduler
instance when processing tasks from different cgroups.
This extends scx_link_sched() to register sub-schedulers in the hash table
and scx_unlink_sched() to remove them. A new scx_find_sub_sched() function
provides the lookup interface.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
scx_bpf_reenqueue_local() currently re-enqueues all tasks on the local DSQ
regardless of which sub-scheduler owns them. With multiple sub-schedulers,
each should only re-enqueue tasks it owns or are owned by its descendants.
Replace the per-rq boolean flag with a lock-free linked list to track
per-scheduler reenqueue requests. Filter tasks in reenq_local() using
hierarchical ownership checks and block deferrals during bypass to prevent
use on dead schedulers.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Add a back pointer from scx_sched_pcpu to scx_sched. This will be used by
the next patch to make scx_bpf_reenqueue_local() sub-sched aware.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Currently, the watchdog checks all tasks as if they are all on scx_root.
Move scx_watchdog_timeout inside scx_sched and make check_rq_for_timeouts()
use the timeout from the scx_sched associated with each task.
refresh_watchdog() is added, which determines the timer interval as half of
the shortest watchdog timeouts of all scheds and arms or disarms it as
necessary. Every scx_sched instance has equivalent or better detection
latency while sharing the same timer.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
scx_dsp_ctx and scx_dsp_max_batch are global variables used in the dispatch
path. In prepration for multiple scheduler support, move the former into
scx_sched_pcpu and the latter into scx_sched. No user-visible behavior
changes intended.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
When a sub-scheduler enters bypass mode, its tasks must be scheduled by an
ancestor to guarantee forward progress. Tasks from bypassing descendants are
queued in the bypass DSQs of the nearest non-bypassing ancestor, or the root
scheduler if all ancestors are bypassing. This requires coordination between
bypassing schedulers and their hosts.
Add bypass_enq_target_dsq() to find the correct bypass DSQ by walking up the
hierarchy until reaching a non-bypassing ancestor. When a sub-scheduler starts
bypassing, all its runnable tasks are re-enqueued after scx_bypassing() is set,
ensuring proper migration to ancestor bypass DSQs.
Update scx_dispatch_sched() to handle hosting bypassed descendants. When a
scheduler is not bypassing but has bypassing descendants, it must schedule both
its own tasks and bypassed descendant tasks. A simple policy is implemented
where every Nth dispatch attempt (SCX_BYPASS_HOST_NTH=2) consumes from the
bypass DSQ. A fallback consumption is also added at the end of dispatch to
ensure bypassed tasks make progress even when normal scheduling is idle.
Update enable_bypass_dsp() and disable_bypass_dsp() to increment
bypass_dsp_enable_depth on both the bypassing scheduler and its parent host,
ensuring both can detect that bypass dispatch is active through
bypass_dsp_enabled().
Add SCX_EV_SUB_BYPASS_DISPATCH event counter to track scheduling of bypassed
descendant tasks.
v2: Fix comment typos (Andrea).
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
The bypass_depth field tracks nesting of bypass operations but is also used
to determine whether the bypass dispatch path should be active. With
hierarchical scheduling, child schedulers may need to activate their parent's
bypass dispatch path without affecting the parent's bypass_depth, requiring
separation of these concerns.
Add bypass_dsp_enable_depth and bypass_dsp_claim to independently control
bypass dispatch path activation. The new enable_bypass_dsp() and
disable_bypass_dsp() functions manage this state with proper claim semantics
to prevent races. The bypass dispatch path now only activates when
bypass_dsp_enabled() returns true, which checks the new enable_depth counter.
The disable operation is carefully ordered after all tasks are moved out of
bypass DSQs to ensure they are drained before the dispatch path is disabled.
During scheduler teardown, disable_bypass_dsp() is called explicitly to ensure
cleanup even if bypass mode was never entered normally.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
In preparation of multiple scheduler support, make bypass state
per-scx_sched. Move scx_bypass_depth, bypass_timestamp and bypass_lb_timer
from globals into scx_sched. Move SCX_RQ_BYPASSING from rq to scx_sched_pcpu
as SCX_SCHED_PCPU_BYPASSING.
scx_bypass() now takes @sch and scx_rq_bypassing(rq) is replaced with
scx_bypassing(sch, cpu). All callers updated.
scx_bypassed_for_enable existed to balance the global scx_bypass_depth when
enable failed. Now that bypass_depth is per-scheduler, the counter is
destroyed along with the scheduler on enable failure. Remove
scx_bypassed_for_enable.
As all tasks currently use the root scheduler, there's no observable behavior
change.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
To support bypass mode for sub-schedulers, move bypass_dsq from struct scx_rq
to struct scx_sched_pcpu. Add bypass_dsq() helper. Move bypass_dsq
initialization from init_sched_ext_class() to scx_alloc_and_attach_sched().
bypass_lb_cpu() now takes a CPU number instead of rq pointer. All callers
updated. No behavior change as all tasks use the root scheduler.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
The abort state was tracked in the global scx_aborting flag which was used to
break out of potential live-lock scenarios when an error occurs. With
hierarchical scheduling, each scheduler instance must track its own abort
state independently so that an aborting scheduler doesn't interfere with
others.
Move the aborting flag into struct scx_sched and update all access sites. The
early initialization check in scx_root_enable() that warned about residual
aborting state is no longer needed as each scheduler instance now starts with
a clean state.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
The default time slice was stored in the global scx_slice_dfl variable which
was dynamically modified when entering and exiting bypass mode. With
hierarchical scheduling, each scheduler instance needs its own default slice
configuration so that bypass operations on one scheduler don't affect others.
Move slice_dfl into struct scx_sched and update all access sites. The bypass
logic now modifies the root scheduler's slice_dfl. At task initialization in
init_scx_entity(), use the SCX_SLICE_DFL constant directly since the task may
not yet be associated with a specific scheduler.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Add checks to enforce scheduling authority boundaries when multiple
schedulers are present:
1. In scx_dsq_insert_preamble() and the dispatch retry path, ignore attempts
to insert tasks that the scheduler doesn't own, counting them via
SCX_EV_INSERT_NOT_OWNED. As BPF schedulers are allowed to ignore
dequeues, such attempts can occur legitimately during sub-scheduler
enabling when tasks move between schedulers. The counter helps distinguish
normal cases from scheduler bugs.
2. For scx_bpf_dsq_insert_vtime() and scx_bpf_select_cpu_and(), error out
when sub-schedulers are attached. These functions lack the aux__prog
parameter needed to identify the calling scheduler, so they cannot be used
safely with multiple schedulers. BPF programs should use the arg-wrapped
versions (__scx_bpf_dsq_insert_vtime() and __scx_bpf_select_cpu_and())
instead.
These checks ensure that with multiple concurrent schedulers, scheduler
identity can be properly determined and unauthorized task operations are
prevented or tracked.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
In preparation for multiple scheduler support, introduce scx_prog_sched()
accessor which returns the scx_sched instance associated with a BPF program.
The association is determined via the special KF_IMPLICIT_ARGS kfunc
parameter, which provides access to bpf_prog_aux. This aux can be used to
retrieve the struct_ops (sched_ext_ops) that the program is associated with,
and from there, the corresponding scx_sched instance.
For compatibility, when ops.sub_attach is not implemented (older schedulers
without sub-scheduler support), unassociated programs fall back to scx_root.
A warning is logged once per scheduler for such programs.
As scx_root is still the only scheduler, this shouldn't introduce
user-visible behavior changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
In preparation of multiple scheduler support, add p->scx.sched which points
to the scx_sched instance that the task is scheduled by, which is currently
always scx_root. Add scx_task_sched[_rcu]() accessors which return the
associated scx_sched of the specified task and replace the raw scx_root
dereferences with it where applicable. scx_task_on_sched() is also added to
test whether a given task is on the specified sched.
As scx_root is still the only scheduler, this shouldn't introduce
user-visible behavior changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
A system often runs multiple workloads especially in multi-tenant server
environments where a system is split into partitions servicing separate
more-or-less independent workloads each requiring an application-specific
scheduler. To support such and other use cases, sched_ext is in the process
of growing multiple scheduler support.
When partitioning a system in terms of CPUs for such use cases, an
oft-taken approach is hard partitioning the system using cpuset. While it
would be possible to tie sched_ext multiple scheduler support to cpuset
partitions, such an approach would have fundamental limitations stemming
from the lack of dynamism and flexibility.
Users often don't care which specific CPUs are assigned to which workload
and want to take advantage of optimizations which are enabled by running
workloads on a larger machine - e.g. opportunistic over-commit, improving
latency critical workload characteristics while maintaining bandwidth
fairness, employing control mechanisms based on different criteria than
on-CPU time for e.g. flexible memory bandwidth isolation, packing similar
parts from different workloads on same L3s to improve cache efficiency,
and so on.
As this sort of dynamic behaviors are impossible or difficult to implement
with hard partitioning, sched_ext is implementing cgroup sub-sched support
where schedulers can be attached to the cgroup hierarchy and a parent
scheduler is responsible for controlling the CPUs that each child can use
at any given moment. This makes CPU distribution dynamically controlled by
BPF allowing high flexibility.
This patch adds the skeletal sched_ext cgroup sub-sched support:
- sched_ext_ops.sub_cgroup_id and .sub_attach/detach() are added. Non-zero
sub_cgroup_id indicates that the scheduler is to be attached to the
identified cgroup. A sub-sched is attached to the cgroup iff the nearest
ancestor scheduler implements .sub_attach() and grants the attachment. Max
nesting depth is limited by SCX_SUB_MAX_DEPTH.
- When a scheduler exits, all its descendant schedulers are exited
together. Also, cgroup.scx_sched added which points to the effective
scheduler instance for the cgroup. This is updated on scheduler
init/exit and inherited on cgroup online. When a cgroup is offlined, the
attached scheduler is automatically exited.
- Sub-sched support is gated on CONFIG_EXT_SUB_SCHED which is
automatically enabled if both SCX and cgroups are enabled. Sub-sched
support is not tied to the CPU controller but rather the cgroup
hierarchy itself. This is intentional as the support for cpu.weight and
cpu.max based resource control is orthogonal to sub-sched support. Note
that CONFIG_CGROUPS around cgroup subtree iteration support for
scx_task_iter is replaced with CONFIG_EXT_SUB_SCHED for consistency.
- This allows loading sub-scheds and most framework operations such as
propagating disable down the hierarchy work. However, sub-scheds are not
operational yet and all tasks stay with the root sched. This will serve
as the basis for building up full sub-sched support.
- DSQs point to the scx_sched they belong to.
- scx_qmap is updated to allow attachment of sub-scheds and also serving
as sub-scheds.
- scx_is_descendant() is added but not yet used in this patch. It is used by
later changes in the series and placed here as this is where the function
belongs.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
The task ownership state machine in sched_ext is quite hard to follow
from the code alone. The interaction of ownership states, memory
ordering rules and cross-CPU "lock dancing" makes the overall model
subtle.
Extend the documentation next to scx_ops_state to provide a more
structured and self-contained description of the state transitions and
their synchronization rules.
The new reference should make the code easier to reason about and
maintain and can help future contributors understand the overall
task-ownership workflow.
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
SCX_EFLAG_INITIALIZED is the sole member of enum scx_exit_flags with no
explicit value, so the compiler assigns it 0. This makes the bitwise OR
in scx_ops_init() a no-op:
sch->exit_info->flags |= SCX_EFLAG_INITIALIZED; /* |= 0 */
As a result, BPF schedulers cannot distinguish whether ops.init()
completed successfully by inspecting exit_info->flags.
Assign the value 1LLU << 0 so the flag is actually set.
Fixes: f3aec2adce ("sched_ext: Add SCX_EFLAG_INITIALIZED to indicate successful ops.init()")
Signed-off-by: David Carlier <devnexen@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently, ops.dequeue() is only invoked when the sched_ext core knows
that a task resides in BPF-managed data structures, which causes it to
miss scheduling property change events. In addition, ops.dequeue()
callbacks are completely skipped when tasks are dispatched to non-local
DSQs from ops.select_cpu(). As a result, BPF schedulers cannot reliably
track task state.
Fix this by guaranteeing that each task entering the BPF scheduler's
custody triggers exactly one ops.dequeue() call when it leaves that
custody, whether the exit is due to a dispatch (regular or via a core
scheduling pick) or to a scheduling property change (e.g.
sched_setaffinity(), sched_setscheduler(), set_user_nice(), NUMA
balancing, etc.).
BPF scheduler custody concept: a task is considered to be in the BPF
scheduler's custody when the scheduler is responsible for managing its
lifecycle. This includes tasks dispatched to user-created DSQs or stored
in the BPF scheduler's internal data structures from ops.enqueue().
Custody ends when the task is dispatched to a terminal DSQ (such as the
local DSQ or %SCX_DSQ_GLOBAL), selected by core scheduling, or removed
due to a property change.
Tasks directly dispatched to terminal DSQs bypass the BPF scheduler
entirely and are never in its custody. Terminal DSQs include:
- Local DSQs (%SCX_DSQ_LOCAL or %SCX_DSQ_LOCAL_ON): per-CPU queues
where tasks go directly to execution.
- Global DSQ (%SCX_DSQ_GLOBAL): the built-in fallback queue where the
BPF scheduler is considered "done" with the task.
As a result, ops.dequeue() is not invoked for tasks directly dispatched
to terminal DSQs.
To identify dequeues triggered by scheduling property changes, introduce
the new ops.dequeue() flag %SCX_DEQ_SCHED_CHANGE: when this flag is set,
the dequeue was caused by a scheduling property change.
New ops.dequeue() semantics:
- ops.dequeue() is invoked exactly once when the task leaves the BPF
scheduler's custody, in one of the following cases:
a) regular dispatch: a task dispatched to a user DSQ or stored in
internal BPF data structures is moved to a terminal DSQ
(ops.dequeue() called without any special flags set),
b) core scheduling dispatch: core-sched picks task before dispatch
(ops.dequeue() called with %SCX_DEQ_CORE_SCHED_EXEC flag set),
c) property change: task properties modified before dispatch,
(ops.dequeue() called with %SCX_DEQ_SCHED_CHANGE flag set).
This allows BPF schedulers to:
- reliably track task ownership and lifecycle,
- maintain accurate accounting of managed tasks,
- update internal state when tasks change properties.
Cc: Tejun Heo <tj@kernel.org>
Cc: Emil Tsalapatis <emil@etsalapatis.com>
Cc: Kuba Piecuch <jpiecuch@google.com>
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
In bypass mode, tasks are queued on per-CPU bypass DSQs. While this works well
in most cases, there is a failure mode where a BPF scheduler can skew task
placement severely before triggering bypass in highly over-saturated systems.
If most tasks end up concentrated on a few CPUs, those CPUs can accumulate
queues that are too long to drain in a reasonable time, leading to RCU stalls
and hung tasks.
Implement a simple timer-based load balancer that redistributes tasks across
CPUs within each NUMA node. The balancer runs periodically (default 500ms,
tunable via bypass_lb_intv_us module parameter) and moves tasks from overloaded
CPUs to underloaded ones.
When moving tasks between bypass DSQs, the load balancer holds nested DSQ locks
to avoid dropping and reacquiring the donor DSQ lock on each iteration, as
donor DSQs can be very long and highly contended. Add the SCX_ENQ_NESTED flag
and use raw_spin_lock_nested() in dispatch_enqueue() to support this. The load
balancer timer function reads scx_bypass_depth locklessly to check whether
bypass mode is active. Use WRITE_ONCE() when updating scx_bypass_depth to pair
with the READ_ONCE() in the timer function.
This has been tested on a 192 CPU dual socket AMD EPYC machine with ~20k
runnable tasks running scx_cpu0. As scx_cpu0 queues all tasks to CPU0, almost
all tasks end up on CPU0 creating severe imbalance. Without the load balancer,
disabling the scheduler can lead to RCU stalls and hung tasks, taking a very
long time to complete. With the load balancer, disable completes in about a
second.
The load balancing operation can be monitored using the sched_ext_bypass_lb
tracepoint and disabled by setting bypass_lb_intv_us to 0.
v2: Lock both rq and DSQ in bypass_lb_cpu() and use dispatch_dequeue_locked()
to prevent races with dispatch_dequeue() (Andrea Righi).
Cc: Andrea Righi <arighi@nvidia.com>
Cc: Dan Schatzberg <schatzberg.dan@gmail.com>
Cc: Emil Tsalapatis <etsal@meta.com>
Reviewed_by: Emil Tsalapatis <emil@etsalapatis.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
The warned bitfields in struct scx_sched are updated racily from concurrent
CPUs causing RMW races, which is fine for these boolean warning flags. Add a
comment marking this area to prevent future fields that can't tolerate racy
updates from being added here.
Signed-off-by: Tejun Heo <tj@kernel.org>
SCX_KICK_WAIT is used to synchronously wait for the target CPU to complete
a reschedule and can be used to implement operations like core scheduling.
This used to be implemented by scx_next_task_picked() incrementing pnt_seq,
which was always called when a CPU picks the next task to run, allowing
SCX_KICK_WAIT to reliably wait for the target CPU to enter the scheduler and
pick the next task.
However, commit b999e365c2 ("sched_ext: Replace scx_next_task_picked()
with switch_class()") replaced scx_next_task_picked() with the
switch_class() callback, which is only called when switching between sched
classes. This broke SCX_KICK_WAIT because pnt_seq would no longer be
reliably incremented unless the previous task was SCX and the next task was
not.
This fix leverages commit 4c95380701 ("sched/ext: Fold balance_scx() into
pick_task_scx()") which refactored the pick path making put_prev_task_scx()
the natural place to track task switches for SCX_KICK_WAIT. The fix moves
pnt_seq increment to put_prev_task_scx() and also increments it in
pick_task_scx() to handle cases where the same task is re-selected, whether
by BPF scheduler decision or slice refill. The semantics: If the current
task on the target CPU is SCX, SCX_KICK_WAIT waits until the CPU enters the
scheduling path. This provides sufficient guarantee for use cases like core
scheduling while keeping the operation self-contained within SCX.
v2: - Also increment pnt_seq in pick_task_scx() to handle same-task
re-selection (Andrea Righi).
- Use smp_cond_load_acquire() for the busy-wait loop for better
architecture optimization (Peter Zijlstra).
Reported-by: Wen-Fang Liu <liuwenfang@honor.com>
Link: http://lkml.kernel.org/r/228ebd9e6ed3437996dffe15735a9caa@honor.com
Cc: Peter Zijlstra <peterz@infradead.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Implement the missing cgroup_set_idle() callback that was marked as a
TODO. This allows BPF schedulers to be notified when a cgroup's idle
state changes, enabling them to adjust their scheduling behavior
accordingly.
The implementation follows the same pattern as other cgroup callbacks
like cgroup_set_weight() and cgroup_set_bandwidth(). It checks if the
BPF scheduler has implemented the callback and invokes it with the
appropriate parameters.
Fixes a spelling error in the cgroup_set_bandwidth() documentation.
tj: s/scx_cgroup_rwsem/scx_cgroup_ops_rwsem/ to fix build breakage.
Signed-off-by: zhidao su <soolaugust@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
ops.exit() may be called even if the loading failed before ops.init()
finishes successfully. This is because ops.exit() allows rich exit info
communication. Add SCX_EFLAG_INITIALIZED flag to scx_exit_info.flags to
indicate whether ops.init() finished successfully.
This enables BPF schedulers to distinguish between exit scenarios and
handle cleanup appropriately based on initialization state.
Acked-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Convert warned_zero_slice and warned_deprecated_rq in scx_sched struct to
single-bit bitfields. While this doesn't reduce struct size immediately,
it prepares for future bitfield additions.
v2: Update patch description.
Acked-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
scx_bpf_cpu_rq() works on an unlocked rq which generally isn't safe.
For the common use-cases scx_bpf_locked_rq() and
scx_bpf_cpu_curr() work, so add a deprecation warning
to scx_bpf_cpu_rq() so it can eventually be removed.
Signed-off-by: Christian Loehle <christian.loehle@arm.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
scx_sched.event_stats_cpu is the percpu counters that are used to track
stats. Introduce struct scx_sched_pcpu and move the counters inside. This
will ease adding more per-cpu fields. No functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Andrea Righi <arighi@nvidia.com>
There currently isn't a place to place SCX-internal types and accessors to
be shared between ext.c and ext_idle.c. Create kernel/sched/ext_internal.h
and move internal type and accessor definitions there. This trims ext.c a
bit and makes future additions easier. Pure code reorganization. No
functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Andrea Righi <arighi@nvidia.com>
