When a running task is migrated to another CPU, the stop_task is used to
preempt the running task and migrate it. This, expectedly, invokes
ops.cpu_release(). If the BPF scheduler then calls
scx_bpf_reenqueue_local(), it re-enqueues all tasks on the local DSQ
including the task which is being migrated.
This creates an unnecessary re-enqueue of a task which is about to be
deactivated and re-activated for migration anyway. It can also cause
confusion for the BPF scheduler as scx_bpf_task_cpu() of the task and its
allowed CPUs may not agree while migration is pending.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 245254f708 ("sched_ext: Implement sched_ext_ops.cpu_acquire/release()")
Acked-by: David Vernet <void@manifault.com>
scx_bpf_reenqueue_local() is used to re-enqueue tasks on the local DSQ from
ops.cpu_release(). Because the BPF scheduler may dispatch tasks to the same
local DSQ, to avoid processing the same tasks repeatedly, it first takes the
number of queued tasks and processes the task at the head of the queue that
number of times.
This is incorrect as a task can be dispatched to the same local DSQ with
SCX_ENQ_HEAD. Such a task will be processed repeatedly until the count is
exhausted and the succeeding tasks won't be processed at all.
Fix it by first moving all candidate tasks to a private list and then
processing that list. While at it, remove the WARNs. They're rather
superflous as later steps will check them anyway.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 245254f708 ("sched_ext: Implement sched_ext_ops.cpu_acquire/release()")
Acked-by: David Vernet <void@manifault.com>
Daniel Borkmann says:
====================
pull-request: bpf-next 2024-07-08
The following pull-request contains BPF updates for your *net-next* tree.
We've added 102 non-merge commits during the last 28 day(s) which contain
a total of 127 files changed, 4606 insertions(+), 980 deletions(-).
The main changes are:
1) Support resilient split BTF which cuts down on duplication and makes BTF
as compact as possible wrt BTF from modules, from Alan Maguire & Eduard Zingerman.
2) Add support for dumping kfunc prototypes from BTF which enables both detecting
as well as dumping compilable prototypes for kfuncs, from Daniel Xu.
3) Batch of s390x BPF JIT improvements to add support for BPF arena and to implement
support for BPF exceptions, from Ilya Leoshkevich.
4) Batch of riscv64 BPF JIT improvements in particular to add 12-argument support
for BPF trampolines and to utilize bpf_prog_pack for the latter, from Pu Lehui.
5) Extend BPF test infrastructure to add a CHECKSUM_COMPLETE validation option
for skbs and add coverage along with it, from Vadim Fedorenko.
6) Inline bpf_get_current_task/_btf() helpers in the arm64 BPF JIT which gives
a small 1% performance improvement in micro-benchmarks, from Puranjay Mohan.
7) Extend the BPF verifier to track the delta between linked registers in order
to better deal with recent LLVM code optimizations, from Alexei Starovoitov.
8) Fix bpf_wq_set_callback_impl() kfunc signature where the third argument should
have been a pointer to the map value, from Benjamin Tissoires.
9) Extend BPF selftests to add regular expression support for test output matching
and adjust some of the selftest when compiled under gcc, from Cupertino Miranda.
10) Simplify task_file_seq_get_next() and remove an unnecessary loop which always
iterates exactly once anyway, from Dan Carpenter.
11) Add the capability to offload the netfilter flowtable in XDP layer through
kfuncs, from Florian Westphal & Lorenzo Bianconi.
12) Various cleanups in networking helpers in BPF selftests to shave off a few
lines of open-coded functions on client/server handling, from Geliang Tang.
13) Properly propagate prog->aux->tail_call_reachable out of BPF verifier, so
that x86 JIT does not need to implement detection, from Leon Hwang.
14) Fix BPF verifier to add a missing check_func_arg_reg_off() to prevent an
out-of-bounds memory access for dynpointers, from Matt Bobrowski.
15) Fix bpf_session_cookie() kfunc to return __u64 instead of long pointer as
it might lead to problems on 32-bit archs, from Jiri Olsa.
16) Enhance traffic validation and dynamic batch size support in xsk selftests,
from Tushar Vyavahare.
bpf-next-for-netdev
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (102 commits)
selftests/bpf: DENYLIST.aarch64: Remove fexit_sleep
selftests/bpf: amend for wrong bpf_wq_set_callback_impl signature
bpf: helpers: fix bpf_wq_set_callback_impl signature
libbpf: Add NULL checks to bpf_object__{prev_map,next_map}
selftests/bpf: Remove exceptions tests from DENYLIST.s390x
s390/bpf: Implement exceptions
s390/bpf: Change seen_reg to a mask
bpf: Remove unnecessary loop in task_file_seq_get_next()
riscv, bpf: Optimize stack usage of trampoline
bpf, devmap: Add .map_alloc_check
selftests/bpf: Remove arena tests from DENYLIST.s390x
selftests/bpf: Add UAF tests for arena atomics
selftests/bpf: Introduce __arena_global
s390/bpf: Support arena atomics
s390/bpf: Enable arena
s390/bpf: Support address space cast instruction
s390/bpf: Support BPF_PROBE_MEM32
s390/bpf: Land on the next JITed instruction after exception
s390/bpf: Introduce pre- and post- probe functions
s390/bpf: Get rid of get_probe_mem_regno()
...
====================
Link: https://patch.msgid.link/20240708221438.10974-1-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
perf_pending_irq() invokes perf_event_wakeup() and __perf_pending_irq().
The former is in charge of waking any tasks which waits to be woken up
while the latter disables perf-events.
The irq_work perf_pending_irq(), while this an irq_work, the callback
is invoked in thread context on PREEMPT_RT. This is needed because all
the waking functions (wake_up_all(), kill_fasync()) acquire sleep locks
which must not be used with disabled interrupts.
Disabling events, as done by __perf_pending_irq(), expects a hardirq
context and disabled interrupts. This requirement is not fulfilled on
PREEMPT_RT.
Split functionality based on perf_event::pending_disable into irq_work
named `pending_disable_irq' and invoke it in hardirq context on
PREEMPT_RT. Rename the split out callback to perf_pending_disable().
Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Marco Elver <elver@google.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: https://lore.kernel.org/r/20240704170424.1466941-8-bigeasy@linutronix.de
perf_pending_task() is invoked in task context and disables preemption
because perf_swevent_get_recursion_context() used to access per-CPU
variables. The other reason is to create a RCU read section while
accessing the perf_event.
The recursion counter is no longer a per-CPU accounter so disabling
preemption is no longer required. The RCU section is needed and must be
created explicit.
Replace the preemption-disable section with a explicit RCU-read section.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Marco Elver <elver@google.com>
Link: https://lore.kernel.org/r/20240704170424.1466941-7-bigeasy@linutronix.de
The swevent_htable::recursion counter is used to avoid creating an
swevent while an event is processed to avoid recursion. The counter is
per-CPU and preemption must be disabled to have a stable counter.
perf_pending_task() disables preemption to access the counter and then
signal. This is problematic on PREEMPT_RT because sending a signal uses
a spinlock_t which must not be acquired in atomic on PREEMPT_RT because
it becomes a sleeping lock.
The atomic context can be avoided by moving the counter into the
task_struct. There is a 4 byte hole between futex_state (usually always
on) and the following perf pointer (perf_event_ctxp). After the
recursion lost some weight it fits perfectly.
Move swevent_htable::recursion into task_struct.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Marco Elver <elver@google.com>
Link: https://lore.kernel.org/r/20240704170424.1466941-6-bigeasy@linutronix.de
There are four recursion counter, one for each context. The type of the
counter is `int' but the counter is used as `bool' since it is only
incremented if zero.
The main goal here is to shrink the whole struct into 32bit int which
can later be added task_struct into an existing hole.
Reduce the type of the recursion counter to an unsigned char, keep the
increment/ decrement operation.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Marco Elver <elver@google.com>
Link: https://lore.kernel.org/r/20240704170424.1466941-5-bigeasy@linutronix.de
A signal is delivered by raising irq_work() which works from any context
including NMI. irq_work() can be delayed if the architecture does not
provide an interrupt vector. In order not to lose a signal, the signal
is injected via task_work during event_sched_out().
Instead going via irq_work, the signal could be added directly via
task_work. The signal is sent to current and can be enqueued on its
return path to userland.
Queue signal via task_work and consider possible NMI context. Remove
perf_event::pending_sigtrap and and use perf_event::pending_work
instead.
Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Marco Elver <elver@google.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: https://lore.kernel.org/r/20240704170424.1466941-4-bigeasy@linutronix.de
Adding task_work from NMI context requires the following:
- The kasan_record_aux_stack() is not NMU safe and must be avoided.
- Using TWA_RESUME is NMI safe. If the NMI occurs while the CPU is in
userland then it will continue in userland and not invoke the `work'
callback.
Add TWA_NMI_CURRENT as an additional notify mode. In this mode skip
kasan and use irq_work in hardirq-mode to for needed interrupt. Set
TIF_NOTIFY_RESUME within the irq_work callback due to k[ac]san
instrumentation in test_and_set_bit() which does not look NMI safe in
case of a report.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20240704170424.1466941-3-bigeasy@linutronix.de
Only if perf_event::pending_sigtrap is zero, the irq_work accounted by
increminging perf_event::nr_pending. The member perf_event::pending_addr
might be overwritten by a subsequent event if the signal was not yet
delivered and is expected. The irq_work will not be enqeueued again
because it has a check to be only enqueued once.
Move irq_work_queue() to where the counter is incremented and
perf_event::pending_sigtrap is set to make it more obvious that the
irq_work is scheduled once.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Marco Elver <elver@google.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: https://lore.kernel.org/r/20240704170424.1466941-2-bigeasy@linutronix.de
The perf pending task work is never waited upon the matching event
release. In the case of a child event, released via free_event()
directly, this can potentially result in a leaked event, such as in the
following scenario that doesn't even require a weak IRQ work
implementation to trigger:
schedule()
prepare_task_switch()
=======> <NMI>
perf_event_overflow()
event->pending_sigtrap = ...
irq_work_queue(&event->pending_irq)
<======= </NMI>
perf_event_task_sched_out()
event_sched_out()
event->pending_sigtrap = 0;
atomic_long_inc_not_zero(&event->refcount)
task_work_add(&event->pending_task)
finish_lock_switch()
=======> <IRQ>
perf_pending_irq()
//do nothing, rely on pending task work
<======= </IRQ>
begin_new_exec()
perf_event_exit_task()
perf_event_exit_event()
// If is child event
free_event()
WARN(atomic_long_cmpxchg(&event->refcount, 1, 0) != 1)
// event is leaked
Similar scenarios can also happen with perf_event_remove_on_exec() or
simply against concurrent perf_event_release().
Fix this with synchonizing against the possibly remaining pending task
work while freeing the event, just like is done with remaining pending
IRQ work. This means that the pending task callback neither need nor
should hold a reference to the event, preventing it from ever beeing
freed.
Fixes: 517e6a301f ("perf: Fix perf_pending_task() UaF")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20240621091601.18227-5-frederic@kernel.org
When a task is scheduled out, pending sigtrap deliveries are deferred
to the target task upon resume to userspace via task_work.
However failures while adding an event's callback to the task_work
engine are ignored. And since the last call for events exit happen
after task work is eventually closed, there is a small window during
which pending sigtrap can be queued though ignored, leaking the event
refcount addition such as in the following scenario:
TASK A
-----
do_exit()
exit_task_work(tsk);
<IRQ>
perf_event_overflow()
event->pending_sigtrap = pending_id;
irq_work_queue(&event->pending_irq);
</IRQ>
=========> PREEMPTION: TASK A -> TASK B
event_sched_out()
event->pending_sigtrap = 0;
atomic_long_inc_not_zero(&event->refcount)
// FAILS: task work has exited
task_work_add(&event->pending_task)
[...]
<IRQ WORK>
perf_pending_irq()
// early return: event->oncpu = -1
</IRQ WORK>
[...]
=========> TASK B -> TASK A
perf_event_exit_task(tsk)
perf_event_exit_event()
free_event()
WARN(atomic_long_cmpxchg(&event->refcount, 1, 0) != 1)
// leak event due to unexpected refcount == 2
As a result the event is never released while the task exits.
Fix this with appropriate task_work_add()'s error handling.
Fixes: 517e6a301f ("perf: Fix perf_pending_task() UaF")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20240621091601.18227-4-frederic@kernel.org
Apparently despite it being marked inline, the compiler
may not inline __down_write_common() which makes it difficult
to identify the cause of lock contention, as the wchan of the
blocked function will always be listed as __down_write_common().
So add __always_inline annotation to the common function (as
well as the inlined helper callers) to force it to be inlined
so a more useful blocking function will be listed (via wchan).
This mirrors commit 92cc5d00a4 ("locking/rwsem: Add
__always_inline annotation to __down_read_common() and inlined
callers") which did the same for __down_read_common.
I sort of worry that I'm playing wack-a-mole here, and talking
with compiler people, they tell me inline means nothing, which
makes me want to cry a little. So I'm wondering if we need to
replace all the inlines with __always_inline, or remove them
because either we mean something by it, or not.
Fixes: c995e638cc ("locking/rwsem: Fold __down_{read,write}*()")
Reported-by: Tim Murray <timmurray@google.com>
Signed-off-by: John Stultz <jstultz@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Link: https://lkml.kernel.org/r/20240709060831.495366-1-jstultz@google.com
The test thread will start N benchmark kthreads and then schedule out
until the test time finished and notify the benchmark kthreads to stop.
The benchmark kthreads will keep running until notified to stop.
There's a problem with current implementation when the benchmark
kthreads number is equal to the CPUs on a non-preemptible kernel:
since the scheduler will balance the kthreads across the CPUs and
when the test time's out the test thread won't get a chance to be
scheduled on any CPU then cannot notify the benchmark kthreads to stop.
This can be easily reproduced on a VM (simulated with 16 CPUs) with
PREEMPT_VOLUNTARY:
estuary:/mnt$ ./dma_map_benchmark -t 16 -s 1
rcu: INFO: rcu_sched self-detected stall on CPU
rcu: 10-...!: (5221 ticks this GP) idle=ed24/1/0x4000000000000000 softirq=142/142 fqs=0
rcu: (t=5254 jiffies g=-559 q=45 ncpus=16)
rcu: rcu_sched kthread starved for 5255 jiffies! g-559 f0x0 RCU_GP_WAIT_FQS(5) ->state=0x0 ->cpu=12
rcu: Unless rcu_sched kthread gets sufficient CPU time, OOM is now expected behavior.
rcu: RCU grace-period kthread stack dump:
task:rcu_sched state:R running task stack:0 pid:16 tgid:16 ppid:2 flags:0x00000008
Call trace
__switch_to+0xec/0x138
__schedule+0x2f8/0x1080
schedule+0x30/0x130
schedule_timeout+0xa0/0x188
rcu_gp_fqs_loop+0x128/0x528
rcu_gp_kthread+0x1c8/0x208
kthread+0xec/0xf8
ret_from_fork+0x10/0x20
Sending NMI from CPU 10 to CPUs 0:
NMI backtrace for cpu 0
CPU: 0 PID: 332 Comm: dma-map-benchma Not tainted 6.10.0-rc1-vanilla-LSE #8
Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
pstate: 20400005 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : arm_smmu_cmdq_issue_cmdlist+0x218/0x730
lr : arm_smmu_cmdq_issue_cmdlist+0x488/0x730
sp : ffff80008748b630
x29: ffff80008748b630 x28: 0000000000000000 x27: ffff80008748b780
x26: 0000000000000000 x25: 000000000000bc70 x24: 000000000001bc70
x23: ffff0000c12af080 x22: 0000000000010000 x21: 000000000000ffff
x20: ffff80008748b700 x19: ffff0000c12af0c0 x18: 0000000000010000
x17: 0000000000000001 x16: 0000000000000040 x15: ffffffffffffffff
x14: 0001ffffffffffff x13: 000000000000ffff x12: 00000000000002f1
x11: 000000000001ffff x10: 0000000000000031 x9 : ffff800080b6b0b8
x8 : ffff0000c2a48000 x7 : 000000000001bc71 x6 : 0001800000000000
x5 : 00000000000002f1 x4 : 01ffffffffffffff x3 : 000000000009aaf1
x2 : 0000000000000018 x1 : 000000000000000f x0 : ffff0000c12af18c
Call trace:
arm_smmu_cmdq_issue_cmdlist+0x218/0x730
__arm_smmu_tlb_inv_range+0xe0/0x1a8
arm_smmu_iotlb_sync+0xc0/0x128
__iommu_dma_unmap+0x248/0x320
iommu_dma_unmap_page+0x5c/0xe8
dma_unmap_page_attrs+0x38/0x1d0
map_benchmark_thread+0x118/0x2c0
kthread+0xec/0xf8
ret_from_fork+0x10/0x20
Solve this by adding scheduling point in the kthread loop,
so if there're other threads in the system they may have
a chance to run, especially the thread to notify the test
end. However this may degrade the test concurrency so it's
recommended to run this on an idle system.
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Acked-by: Barry Song <baohua@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
DSQs are very opaque in the consumption path. The BPF scheduler has no way
of knowing which tasks are being considered and which is picked. This patch
adds BPF DSQ iterator.
- Allows iterating tasks queued on a DSQ in the dispatch order or reverse
from anywhere using bpf_for_each(scx_dsq) or calling the iterator kfuncs
directly.
- Has ordering guarantee where only tasks which were already queued when the
iteration started are visible and consumable during the iteration.
v5: - Add a comment to the naked list_empty(&dsq->list) test in
consume_dispatch_q() to explain the reasoning behind the lockless test
and by extension why nldsq_next_task() isn't used there.
- scx_qmap changes separated into its own patch.
v4: - bpf_iter_scx_dsq_new() declaration in common.bpf.h was using the wrong
type for the last argument (bool rev instead of u64 flags). Fix it.
v3: - Alexei pointed out that the iterator is too big to allocate on stack.
Added a prep patch to reduce the size of the cursor. Now
bpf_iter_scx_dsq is 48 bytes and bpf_iter_scx_dsq_kern is 40 bytes on
64bit.
- u32_before() comparison factored out.
v2: - scx_bpf_consume_task() is separated out into a separate patch.
- DSQ seq and iter flags don't need to be u64. Use u32.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: David Vernet <dvernet@meta.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: bpf@vger.kernel.org
struct scx_dsq_node contains two data structure nodes to link the containing
task to a DSQ and a flags field that is protected by the lock of the
associated DSQ. One reason why they are grouped into a struct is to use the
type independently as a cursor node when iterating tasks on a DSQ. However,
when iterating, the cursor only needs to be linked on the FIFO list and the
rb_node part ends up inflating the size of the iterator data structure
unnecessarily making it potentially too expensive to place it on stack.
Take ->priq and ->flags out of scx_dsq_node and put them in sched_ext_entity
as ->dsq_priq and ->dsq_flags, respectively. scx_dsq_node is renamed to
scx_dsq_list_node and the field names are renamed accordingly. This will
help implementing DSQ task iterator that can be allocated on stack.
No functional change intended.
Signed-off-by: Tejun Heo <tj@kernel.org>
Suggested-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: David Vernet <void@manifault.com>
While sched_ext was out of tree, everything sched_ext specific which can be
put in kernel/sched/ext.h was put there to ease forward porting. However,
kernel/sched/sched.h is the better location for some of them. Relocate.
- struct sched_enq_and_set_ctx, sched_deq_and_put_task() and
sched_enq_and_set_task().
- scx_enabled() and scx_switched_all().
- for_active_class_range() and for_each_active_class(). sched_class
declarations are moved above the class iterators for this.
No functional changes intended.
Signed-off-by: Tejun Heo <tj@kernel.org>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: David Vernet <void@manifault.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
For flexibility, sched_ext allows the BPF scheduler to select the CPU to
execute a task on at dispatch time so that e.g. a queue can be shared across
multiple CPUs. To enable this, the dispatch path is executed from balance()
so that a dispatched task can be hot-migrated to its target CPU. This means
that sched_ext needs its balance() method invoked before every
pick_next_task() even when the CPU is waking up from SCHED_IDLE.
for_balance_class_range() defined in kernel/sched/ext.h implements this
selective iteration promotion. However, the indirection obfuscates more than
helps. Open code the iteration promotion in put_prev_task_balance() and
remove for_balance_class_range().
No functional changes intended.
Signed-off-by: Tejun Heo <tj@kernel.org>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: David Vernet <void@manifault.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
- scx_ops_cpu_preempt is only used in kernel/sched/ext.c and doesn't need to
be global. Make it static.
- Relocate task_on_scx() so that the inline functions are located together.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: David Vernet <void@manifault.com>
sched_domains regulate the load balancing for sched_classes. A machine can
be partitioned into multiple sections that are not load-balanced across
using either isolcpus= boot param or cpuset partitions. In such cases, tasks
that are in one partition are expected to stay within that partition.
cpuset configured partitions are always reflected in each member task's
cpumask. As SCX always honors the task cpumasks, the BPF scheduler is
automatically in compliance with the configured partitions.
However, for isolcpus= domain isolation, the isolated CPUs are simply
omitted from the top-level sched_domain[s] without further restrictions on
tasks' cpumasks, so, for example, a task currently running in an isolated
CPU may have more CPUs in its allowed cpumask while expected to remain on
the same CPU.
There is no straightforward way to enforce this partitioning preemptively on
BPF schedulers and erroring out after a violation can be surprising.
isolcpus= domain isolation is being replaced with cpuset partitions anyway,
so keep it simple and simply disallow loading a BPF scheduler if isolcpus=
domain isolation is in effect.
Signed-off-by: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20240626082342.GY31592@noisy.programming.kicks-ass.net
Cc: David Vernet <void@manifault.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Frederic Weisbecker <frederic@kernel.org>
When initializing p->scx.weight, scx_ops_enable_task() wasn't considering
whether the task is SCHED_IDLE. Update it to use WEIGHT_IDLEPRIO as the
source weight for SCHED_IDLE tasks. This leaves reweight_task_scx() the sole
user of set_task_scx_weight(). Open code it. @weight is going to be provided
by sched core in the future anyway.
v2: Use the newly available @lw->weight to set @p->scx.weight in
reweight_task_scx().
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: David Vernet <void@manifault.com>
Cc: Peter Zijlstra <peterz@infradead.org>
sched_fork() returns with -EAGAIN if dl_prio(@p). a7a9fc5492 ("sched_ext:
Add boilerplate for extensible scheduler class") added scx_pre_fork() call
before it and then scx_cancel_fork() on the exit path. This is silly as the
dl_prio() block can just be moved above the scx_pre_fork() call.
Move the dl_prio() block above the scx_pre_fork() call and remove the now
unnecessary scx_cancel_fork() invocation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: David Vernet <void@manifault.com>
rq contains many useful fields to implement a custom scheduler. For
example, various clock signals like clock_task and clock_pelt can be
used to track load. It also contains stats in other sched_classes, which
are useful to drive scheduling decisions in ext.
tj: Put the new helper below scx_bpf_task_*() helpers.
Signed-off-by: Hongyan Xia <hongyan.xia2@arm.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
d329605287 ("sched/fair: set_load_weight() must also call reweight_task()
for SCHED_IDLE tasks") applied to sched/core changes how reweight_task() is
called causing conflicts with e83edbf88f ("sched: Add
sched_class->reweight_task()"). Resolve the conflicts by taking
set_load_weight() changes from d329605287 and updating
sched_class->reweight_task() to take pointer to struct load_weight instead
of int prio.
Signed-off-by: Tejun Heo<tj@kernel.org>
I realized this while having a map containing both a struct bpf_timer and
a struct bpf_wq: the third argument provided to the bpf_wq callback is
not the struct bpf_wq pointer itself, but the pointer to the value in
the map.
Which means that the users need to double cast the provided "value" as
this is not a struct bpf_wq *.
This is a change of API, but there doesn't seem to be much users of bpf_wq
right now, so we should be able to go with this right now.
Fixes: 81f1d7a583 ("bpf: wq: add bpf_wq_set_callback_impl")
Signed-off-by: Benjamin Tissoires <bentiss@kernel.org>
Link: https://lore.kernel.org/r/20240708-fix-wq-v2-1-667e5c9fbd99@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Currently, kprobe event checks whether the target symbol name is unique
or not, so that it does not put a probe on an unexpected place. But this
skips the check if the target is on a module because the module may not
be loaded.
To fix this issue, this patch checks the number of probe target symbols
in a target module when the module is loaded. If the probe is not on the
unique name symbols in the module, it will be rejected at that point.
Note that the symbol which has a unique name in the target module,
it will be accepted even if there are same-name symbols in the
kernel or other modules,
Link: https://lore.kernel.org/all/172016348553.99543.2834679315611882137.stgit@devnote2/
Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
The PWQ allocation and WQ enlistment are not within the same lock-held
critical section; therefore, their states can become out of sync when
the user modifies the unbound mask or if CPU hotplug events occur in
the interim since those operations only update the WQs that are already
in the list.
Make the PWQ allocation and WQ enlistment atomic.
Signed-off-by: Lai Jiangshan <jiangshan.ljs@antgroup.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
For early wq allocation, rescuer initialization is the last step of the
creation of a new wq. Make the behavior the same for all allocations.
Prepare for initializing rescuer's affinities with the default pwq's
affinities.
Prepare for moving the whole workqueue initializing procedure into
wq_pool_mutex and cpu hotplug locks.
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Lai Jiangshan <jiangshan.ljs@antgroup.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
workqueue creation includes adding it to the workqueue list.
Prepare for moving the whole workqueue initializing procedure into
wq_pool_mutex and cpu hotplug locks.
Signed-off-by: Lai Jiangshan <jiangshan.ljs@antgroup.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
In the function cgroup_base_stat_cputime_show, there are five
instances of #ifdef, which makes the code not concise.
To address this, add the function cgroup_force_idle_show
to make the code more succinct.
Signed-off-by: Chen Ridong <chenridong@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
When rcu_barrier() calls rcu_rdp_cpu_online() and observes a CPU off
rnp->qsmaskinitnext, it means that all accesses from the offline CPU
preceding the CPUHP_TEARDOWN_CPU are visible to RCU barrier, including
callbacks expiration and counter updates.
However interrupts can still fire after stop_machine() re-enables
interrupts and before rcutree_report_cpu_dead(). The related accesses
happening between CPUHP_TEARDOWN_CPU and rnp->qsmaskinitnext clearing
are _NOT_ guaranteed to be seen by rcu_barrier() without proper
ordering, especially when callbacks are invoked there to the end, making
rcutree_migrate_callback() bypass barrier_lock.
The following theoretical race example can make rcu_barrier() hang:
CPU 0 CPU 1
----- -----
//cpu_down()
smpboot_park_threads()
//ksoftirqd is parked now
<IRQ>
rcu_sched_clock_irq()
invoke_rcu_core()
do_softirq()
rcu_core()
rcu_do_batch()
// callback storm
// rcu_do_batch() returns
// before completing all
// of them
// do_softirq also returns early because of
// timeout. It defers to ksoftirqd but
// it's parked
</IRQ>
stop_machine()
take_cpu_down()
rcu_barrier()
spin_lock(barrier_lock)
// observes rcu_segcblist_n_cbs(&rdp->cblist) != 0
<IRQ>
do_softirq()
rcu_core()
rcu_do_batch()
//completes all pending callbacks
//smp_mb() implied _after_ callback number dec
</IRQ>
rcutree_report_cpu_dead()
rnp->qsmaskinitnext &= ~rdp->grpmask;
rcutree_migrate_callback()
// no callback, early return without locking
// barrier_lock
//observes !rcu_rdp_cpu_online(rdp)
rcu_barrier_entrain()
rcu_segcblist_entrain()
// Observe rcu_segcblist_n_cbs(rsclp) == 0
// because no barrier between reading
// rnp->qsmaskinitnext and rsclp->len
rcu_segcblist_add_len()
smp_mb__before_atomic()
// will now observe the 0 count and empty
// list, but too late, we enqueue regardless
WRITE_ONCE(rsclp->len, rsclp->len + v);
// ignored barrier callback
// rcu barrier stall...
This could be solved with a read memory barrier, enforcing the message
passing between rnp->qsmaskinitnext and rsclp->len, matching the full
memory barrier after rsclp->len addition in rcu_segcblist_add_len()
performed at the end of rcu_do_batch().
However the rcu_barrier() is complicated enough and probably doesn't
need too many more subtleties. CPU down is a slowpath and the
barrier_lock seldom contended. Solve the issue with unconditionally
locking the barrier_lock on rcutree_migrate_callbacks(). This makes sure
that either rcu_barrier() sees the empty queue or its entrained
callback will be migrated.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_sync structure's ->gp_count field is always accessed under the
protection of that same structure's ->rss_lock field, with the exception
of a pair of WARN_ON_ONCE() calls just prior to acquiring that lock in
functions rcu_sync_exit() and rcu_sync_dtor(). These lockless accesses
are unnecessary and impair KCSAN's ability to catch bugs that might be
inserted via other lockless accesses.
This commit therefore moves those WARN_ON_ONCE() calls under the lock.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
If a CPU is running either a userspace application or a guest OS in
nohz_full mode, it is possible for a system call to occur just as an
RCU grace period is starting. If that CPU also has the scheduling-clock
tick enabled for any reason (such as a second runnable task), and if the
system was booted with rcutree.use_softirq=0, then RCU can add insult to
injury by awakening that CPU's rcuc kthread, resulting in yet another
task and yet more OS jitter due to switching to that task, running it,
and switching back.
In addition, in the common case where that system call is not of
excessively long duration, awakening the rcuc task is pointless.
This pointlessness is due to the fact that the CPU will enter an extended
quiescent state upon returning to the userspace application or guest OS.
In this case, the rcuc kthread cannot do anything that the main RCU
grace-period kthread cannot do on its behalf, at least if it is given
a few additional milliseconds (for example, given the time duration
specified by rcutree.jiffies_till_first_fqs, give or take scheduling
delays).
This commit therefore adds a rcutree.nohz_full_patience_delay kernel
boot parameter that specifies the grace period age (in milliseconds,
rounded to jiffies) before which RCU will refrain from awakening the
rcuc kthread. Preliminary experimentation suggests a value of 1000,
that is, one second. Increasing rcutree.nohz_full_patience_delay will
increase grace-period latency and in turn increase memory footprint,
so systems with constrained memory might choose a smaller value.
Systems with less-aggressive OS-jitter requirements might choose the
default value of zero, which keeps the traditional immediate-wakeup
behavior, thus avoiding increases in grace-period latency.
[ paulmck: Apply Leonardo Bras feedback. ]
Link: https://lore.kernel.org/all/20240328171949.743211-1-leobras@redhat.com/
Reported-by: Leonardo Bras <leobras@redhat.com>
Suggested-by: Leonardo Bras <leobras@redhat.com>
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Leonardo Bras <leobras@redhat.com>
The default aux_watermark is half the AUX area buffer size. In general,
on a 64-bit architecture, the AUX area buffer size could be a bigger than
fits in a 32-bit type, but the calculation does not allow for that
possibility.
However the aux_watermark value is recorded in a u32, so should not be
more than U32_MAX either.
Fix by doing the calculation in a correctly sized type, and limiting the
result to U32_MAX.
Fixes: d68e6799a5 ("perf: Cap allocation order at aux_watermark")
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20240624201101.60186-7-adrian.hunter@intel.com
nr_pages is unsigned long but gets passed to rb_alloc_aux() as an int,
and is stored as an int.
Only power-of-2 values are accepted, so if nr_pages is a 64_bit value, it
will be passed to rb_alloc_aux() as zero.
That is not ideal because:
1. the value is incorrect
2. rb_alloc_aux() is at risk of misbehaving, although it manages to
return -ENOMEM in that case, it is a result of passing zero to get_order()
even though the get_order() result is documented to be undefined in that
case.
Fix by simply validating the maximum supported value in the first place.
Use -ENOMEM error code for consistency with the current error code that
is returned in that case.
Fixes: 45bfb2e504 ("perf: Add AUX area to ring buffer for raw data streams")
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20240624201101.60186-6-adrian.hunter@intel.com
When a task's weight is being changed, set_load_weight() is called with
@update_load set. As weight changes aren't trivial for the fair class,
set_load_weight() calls fair.c::reweight_task() for fair class tasks.
However, set_load_weight() first tests task_has_idle_policy() on entry and
skips calling reweight_task() for SCHED_IDLE tasks. This is buggy as
SCHED_IDLE tasks are just fair tasks with a very low weight and they would
incorrectly skip load, vlag and position updates.
Fix it by updating reweight_task() to take struct load_weight as idle weight
can't be expressed with prio and making set_load_weight() call
reweight_task() for SCHED_IDLE tasks too when @update_load is set.
Fixes: 9059393e4e ("sched/fair: Use reweight_entity() for set_user_nice()")
Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org # v4.15+
Link: http://lkml.kernel.org/r/20240624102331.GI31592@noisy.programming.kicks-ass.net
The current code loops over the psi_states only to call a helper which
then resolves back to the action needed for each state using a switch
statement. That is effectively creating a double indirection of a kind
which, given how all the states need to be explicitly listed and handled
anyway, we can simply remove. Both the for loop and the switch statement
that is.
The benefit is both in the code size and CPU time spent in this function.
YMMV but on my Steam Deck, while in a game, the patch makes the CPU usage
go from ~2.4% down to ~1.2%. Text size at the same time went from 0x323 to
0x2c1.
Signed-off-by: Tvrtko Ursulin <tursulin@ursulin.net>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Chengming Zhou <chengming.zhou@linux.dev>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: https://lkml.kernel.org/r/20240625135000.38652-1-tursulin@igalia.com
