The more intense grace-period processing resulting from the 50x RCU
Tasks Trace grace-period speedups exposed the following race condition:
o Task A running on CPU 0 executes rcu_read_lock_trace(),
entering a read-side critical section.
o When Task A eventually invokes rcu_read_unlock_trace()
to exit its read-side critical section, this function
notes that the ->trc_reader_special.s flag is zero and
and therefore invoke wil set ->trc_reader_nesting to zero
using WRITE_ONCE(). But before that happens...
o The RCU Tasks Trace grace-period kthread running on some other
CPU interrogates Task A, but this fails because this task is
currently running. This kthread therefore sends an IPI to CPU 0.
o CPU 0 receives the IPI, and thus invokes trc_read_check_handler().
Because Task A has not yet cleared its ->trc_reader_nesting
counter, this function sees that Task A is still within its
read-side critical section. This function therefore sets the
->trc_reader_nesting.b.need_qs flag, AKA the .need_qs flag.
Except that Task A has already checked the .need_qs flag, which
is part of the ->trc_reader_special.s flag. The .need_qs flag
therefore remains set until Task A's next rcu_read_unlock_trace().
o Task A now invokes synchronize_rcu_tasks_trace(), which cannot
start a new grace period until the current grace period completes.
And thus cannot return until after that time.
But Task A's .need_qs flag is still set, which prevents the current
grace period from completing. And because Task A is blocked, it
will never execute rcu_read_unlock_trace() until its call to
synchronize_rcu_tasks_trace() returns.
We are therefore deadlocked.
This race is improbable, but 80 hours of rcutorture made it happen twice.
The race was possible before the grace-period speedup, but roughly 50x
less probable. Several thousand hours of rcutorture would have been
necessary to have a reasonable chance of making this happen before this
50x speedup.
This commit therefore eliminates this deadlock by setting
->trc_reader_nesting to a large negative number before checking the
.need_qs and zeroing (or decrementing with respect to its initial
value) ->trc_reader_nesting. For its part, the IPI handler's
trc_read_check_handler() function adds a check for negative values,
deferring evaluation of the task in this case. Taken together, these
changes avoid this deadlock scenario.
Fixes: 276c410448 ("rcu-tasks: Split ->trc_reader_need_end")
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: <bpf@vger.kernel.org>
Cc: <stable@vger.kernel.org> # 5.7.x
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When CONFIG_BPF_SYSCALL is not set, but CONFIG_BPF_JIT=y
the kernel build fails:
In file included from ../kernel/bpf/trampoline.c:11:
../kernel/bpf/trampoline.c: In function ‘bpf_trampoline_update’:
../kernel/bpf/trampoline.c:220:39: error: ‘call_rcu_tasks_trace’ undeclared
../kernel/bpf/trampoline.c: In function ‘__bpf_prog_enter_sleepable’:
../kernel/bpf/trampoline.c:411:2: error: implicit declaration of function ‘rcu_read_lock_trace’
../kernel/bpf/trampoline.c: In function ‘__bpf_prog_exit_sleepable’:
../kernel/bpf/trampoline.c:416:2: error: implicit declaration of function ‘rcu_read_unlock_trace’
This is due to:
obj-$(CONFIG_BPF_JIT) += trampoline.o
obj-$(CONFIG_BPF_JIT) += dispatcher.o
There is a number of functions that arch/x86/net/bpf_jit_comp.c is
using from these two files, but none of them will be used when
only cBPF is on (which is the case for BPF_SYSCALL=n BPF_JIT=y).
Add rcu_trace functions to rcupdate_trace.h. The JITed code won't execute them
and BPF trampoline logic won't be used without BPF_SYSCALL.
Fixes: 1e6c62a882 ("bpf: Introduce sleepable BPF programs")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Link: https://lore.kernel.org/bpf/20200831155155.62754-1-alexei.starovoitov@gmail.com
The synchronize_rcu_tasks_trace() header comment incorrectly claims that
any number of things delimit RCU Tasks Trace read-side critical sections,
when in fact only rcu_read_lock_trace() and rcu_read_unlock_trace() do so.
This commit therefore fixes this comment, and, while in the area, fixes
a typo in the rcu_read_lock_trace() header comment.
Reported-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit provides a new TASKS_TRACE_RCU_READ_MB Kconfig option that
enables use of read-side memory barriers by both rcu_read_lock_trace()
and rcu_read_unlock_trace() when the are executed with the
current->trc_reader_special.b.need_mb flag set. This flag is currently
never set. Doing that is the subject of a later commit.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit splits ->trc_reader_need_end by using the rcu_special union.
This change permits readers to check to see if a memory barrier is
required without any added overhead in the common case where no such
barrier is required. This commit also adds the read-side checking.
Later commits will add the machinery to properly set the new
->trc_reader_special.b.need_mb field.
This commit also makes rcu_read_unlock_trace_special() tolerate nested
read-side critical sections within interrupt and NMI handlers.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Because RCU does not watch exception early-entry/late-exit, idle-loop,
or CPU-hotplug execution, protection of tracing and BPF operations is
needlessly complicated. This commit therefore adds a variant of
Tasks RCU that:
o Has explicit read-side markers to allow finite grace periods in
the face of in-kernel loops for PREEMPT=n builds. These markers
are rcu_read_lock_trace() and rcu_read_unlock_trace().
o Protects code in the idle loop, exception entry/exit, and
CPU-hotplug code paths. In this respect, RCU-tasks trace is
similar to SRCU, but with lighter-weight readers.
o Avoids expensive read-side instruction, having overhead similar
to that of Preemptible RCU.
There are of course downsides:
o The grace-period code can send IPIs to CPUs, even when those
CPUs are in the idle loop or in nohz_full userspace. This is
mitigated by later commits.
o It is necessary to scan the full tasklist, much as for Tasks RCU.
o There is a single callback queue guarded by a single lock,
again, much as for Tasks RCU. However, those early use cases
that request multiple grace periods in quick succession are
expected to do so from a single task, which makes the single
lock almost irrelevant. If needed, multiple callback queues
can be provided using any number of schemes.
Perhaps most important, this variant of RCU does not affect the vanilla
flavors, rcu_preempt and rcu_sched. The fact that RCU Tasks Trace
readers can operate from idle, offline, and exception entry/exit in no
way enables rcu_preempt and rcu_sched readers to do so.
The memory ordering was outlined here:
https://lore.kernel.org/lkml/20200319034030.GX3199@paulmck-ThinkPad-P72/
This effort benefited greatly from off-list discussions of BPF
requirements with Alexei Starovoitov and Andrii Nakryiko. At least
some of the on-list discussions are captured in the Link: tags below.
In addition, KCSAN was quite helpful in finding some early bugs.
Link: https://lore.kernel.org/lkml/20200219150744.428764577@infradead.org/
Link: https://lore.kernel.org/lkml/87mu8p797b.fsf@nanos.tec.linutronix.de/
Link: https://lore.kernel.org/lkml/20200225221305.605144982@linutronix.de/
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Andrii Nakryiko <andriin@fb.com>
[ paulmck: Apply feedback from Steve Rostedt and Joel Fernandes. ]
[ paulmck: Decrement trc_n_readers_need_end upon IPI failure. ]
[ paulmck: Fix locking issue reported by rcutorture. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
