The recursion check in __bpf_prog_enter and __bpf_prog_exit
leaves some (not inlined) functions unprotected:
In __bpf_prog_enter:
- migrate_disable is called before prog->active is checked
In __bpf_prog_exit:
- migrate_enable,rcu_read_unlock_strict are called after
prog->active is decreased
When attaching trampoline to them we get panic like:
traps: PANIC: double fault, error_code: 0x0
double fault: 0000 [#1] SMP PTI
RIP: 0010:__bpf_prog_enter+0x4/0x50
...
Call Trace:
<IRQ>
bpf_trampoline_6442466513_0+0x18/0x1000
migrate_disable+0x5/0x50
__bpf_prog_enter+0x9/0x50
bpf_trampoline_6442466513_0+0x18/0x1000
migrate_disable+0x5/0x50
__bpf_prog_enter+0x9/0x50
bpf_trampoline_6442466513_0+0x18/0x1000
migrate_disable+0x5/0x50
__bpf_prog_enter+0x9/0x50
bpf_trampoline_6442466513_0+0x18/0x1000
migrate_disable+0x5/0x50
...
Fixing this by adding deny list of btf ids for tracing
programs and checking btf id during program verification.
Adding above functions to this list.
Suggested-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210429114712.43783-1-jolsa@kernel.org
Add a kconfig knob which allows for unprivileged bpf to be disabled by default.
If set, the knob sets /proc/sys/kernel/unprivileged_bpf_disabled to value of 2.
This still allows a transition of 2 -> {0,1} through an admin. Similarly,
this also still keeps 1 -> {1} behavior intact, so that once set to permanently
disabled, it cannot be undone aside from a reboot.
We've also added extra2 with max of 2 for the procfs handler, so that an admin
still has a chance to toggle between 0 <-> 2.
Either way, as an additional alternative, applications can make use of CAP_BPF
that we added a while ago.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/74ec548079189e4e4dffaeb42b8987bb3c852eee.1620765074.git.daniel@iogearbox.net
Right now, all core BPF related options are scattered in different Kconfig
locations mainly due to historic reasons. Moving forward, lets add a proper
subsystem entry under ...
General setup --->
BPF subsystem --->
... in order to have all knobs in a single location and thus ease BPF related
configuration. Networking related bits such as sockmap are out of scope for
the general setup and therefore better suited to remain in net/Kconfig.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/f23f58765a4d59244ebd8037da7b6a6b2fb58446.1620765074.git.daniel@iogearbox.net
RTC drivers used to leave .set_alarm() NULL in order to signal the RTC
device doesn't support alarms. The drivers are now clearing the
RTC_FEATURE_ALARM bit for that purpose in order to keep the rtc_class_ops
structure const. So now, .set_alarm() is set unconditionally and this
possibly causes the alarmtimer code to select an RTC device that doesn't
support alarms.
Test RTC_FEATURE_ALARM instead of relying on ops->set_alarm to determine
whether alarms are available.
Fixes: 7ae41220ef ("rtc: introduce features bitfield")
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20210511014516.563031-1-alexandre.belloni@bootlin.com
Only the very first page of BPF ringbuf that contains consumer position
counter is supposed to be mapped as writeable by user-space. Producer
position is read-only and can be modified only by the kernel code. BPF ringbuf
data pages are read-only as well and are not meant to be modified by
user-code to maintain integrity of per-record headers.
This patch allows to map only consumer position page as writeable and
everything else is restricted to be read-only. remap_vmalloc_range()
internally adds VM_DONTEXPAND, so all the established memory mappings can't be
extended, which prevents any future violations through mremap()'ing.
Fixes: 457f44363a ("bpf: Implement BPF ring buffer and verifier support for it")
Reported-by: Ryota Shiga (Flatt Security)
Reported-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
A BPF program might try to reserve a buffer larger than the ringbuf size.
If the consumer pointer is way ahead of the producer, that would be
successfully reserved, allowing the BPF program to read or write out of
the ringbuf allocated area.
Reported-by: Ryota Shiga (Flatt Security)
Fixes: 457f44363a ("bpf: Implement BPF ring buffer and verifier support for it")
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Fix a bug in the verifier's scalar32_min_max_*() functions which leads to
incorrect tracking of 32 bit bounds for the simulation of and/or/xor bitops.
When both the src & dst subreg is a known constant, then the assumption is
that scalar_min_max_*() will take care to update bounds correctly. However,
this is not the case, for example, consider a register R2 which has a tnum
of 0xffffffff00000000, meaning, lower 32 bits are known constant and in this
case of value 0x00000001. R2 is then and'ed with a register R3 which is a
64 bit known constant, here, 0x100000002.
What can be seen in line '10:' is that 32 bit bounds reach an invalid state
where {u,s}32_min_value > {u,s}32_max_value. The reason is scalar32_min_max_*()
delegates 32 bit bounds updates to scalar_min_max_*(), however, that really
only takes place when both the 64 bit src & dst register is a known constant.
Given scalar32_min_max_*() is intended to be designed as closely as possible
to scalar_min_max_*(), update the 32 bit bounds in this situation through
__mark_reg32_known() which will set all {u,s}32_{min,max}_value to the correct
constant, which is 0x00000000 after the fix (given 0x00000001 & 0x00000002 in
32 bit space). This is possible given var32_off already holds the final value
as dst_reg->var_off is updated before calling scalar32_min_max_*().
Before fix, invalid tracking of R2:
[...]
9: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,smin_value=-9223372036854775807 (0x8000000000000001),smax_value=9223372032559808513 (0x7fffffff00000001),umin_value=1,umax_value=0xffffffff00000001,var_off=(0x1; 0xffffffff00000000),s32_min_value=1,s32_max_value=1,u32_min_value=1,u32_max_value=1) R3_w=inv4294967298 R10=fp0
9: (5f) r2 &= r3
10: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,smin_value=0,smax_value=4294967296 (0x100000000),umin_value=0,umax_value=0x100000000,var_off=(0x0; 0x100000000),s32_min_value=1,s32_max_value=0,u32_min_value=1,u32_max_value=0) R3_w=inv4294967298 R10=fp0
[...]
After fix, correct tracking of R2:
[...]
9: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,smin_value=-9223372036854775807 (0x8000000000000001),smax_value=9223372032559808513 (0x7fffffff00000001),umin_value=1,umax_value=0xffffffff00000001,var_off=(0x1; 0xffffffff00000000),s32_min_value=1,s32_max_value=1,u32_min_value=1,u32_max_value=1) R3_w=inv4294967298 R10=fp0
9: (5f) r2 &= r3
10: R0_w=inv1337 R1=ctx(id=0,off=0,imm=0) R2_w=inv(id=0,smin_value=0,smax_value=4294967296 (0x100000000),umin_value=0,umax_value=0x100000000,var_off=(0x0; 0x100000000),s32_min_value=0,s32_max_value=0,u32_min_value=0,u32_max_value=0) R3_w=inv4294967298 R10=fp0
[...]
Fixes: 3f50f132d8 ("bpf: Verifier, do explicit ALU32 bounds tracking")
Fixes: 2921c90d47 ("bpf: Fix a verifier failure with xor")
Reported-by: Manfred Paul (@_manfp)
Reported-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
After the introduction of the cgroup.kill there is only one call site
of cgroup_task_freeze() left: cgroup_exit(). cgroup_task_freeze() is
currently taking rcu_read_lock() to read task's cgroup flags, but
because it's always called with css_set_lock locked, the rcu protection
is excessive.
Simplify the code by inlining cgroup_task_freeze().
v2: fix build
Signed-off-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
RCU priority boosting cannot do anything unless there is at least one
task blocking the current RCU grace period that was preempted within
the RCU read-side critical section that it still resides in. However,
the current rcu_torture_boost_failed() code will count this as an RCU
priority-boosting failure if there were no CPUs blocking the current
grace period. This situation can happen (for example) if the last CPU
blocking the current grace period was subjected to vCPU preemption,
which is always a risk for rcutorture guest OSes.
This commit therefore causes rcu_torture_boost_failed() to refrain from
reporting failure unless there is at least one task blocking the current
RCU grace period that was preempted within the RCU read-side critical
section that it still resides in.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Add comments to synchronize_rcu() and friends that point to
Documentation/RCU/Design/Memory-Ordering/Tree-RCU-Memory-Ordering.rst.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Although there are trace events for RCU grace periods, these are only
enabled in CONFIG_RCU_TRACE=y kernels. This commit therefore marks
rcu_gp_cleanup() noinline in order to provide a function that can be
traced that is invoked near the end of each grace period.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Kernels built with CONFIG_RCU_STRICT_GRACE_PERIOD=y can experience
significant lock contention due to RCU's resulting focus on ending grace
periods as soon as possible. This is OK, but only if there are not very
many CPUs. This commit therefore puts this Kconfig option off-limits
to systems with more than four CPUs.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, show_rcu_gp_kthreads() only dumps rcu_node structures that
have outdated ideas of the current grace-period number. This commit
also dumps those that are in any way blocking the current grace period.
This helps diagnose RCU priority boosting failures.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When any CPU comes online, it checks to see if an RCU-boost kthread has
already been created for that CPU's leaf rcu_node structure, and if
not, it creates one. Unfortunately, it also verifies that this leaf
rcu_node structure actually has at least one online CPU, and if not,
it declines to create the kthread. Although this behavior makes sense
during early boot, especially on systems that claim far more CPUs than
they actually have, it makes no sense for the first CPU to come online
for a given rcu_node structure. There is no point in checking because
we know there is a CPU on its way in.
The problem is that timing differences can cause this incoming CPU to not
yet be reflected in the various bit masks even at rcutree_online_cpu()
time, and there is no chance at rcutree_prepare_cpu() time. Plus it
would be better to create the RCU-boost kthread at rcutree_prepare_cpu()
to handle the case where the CPU is involved in an RCU priority inversion
very shortly after it comes online.
This commit therefore moves the checking to rcu_prepare_kthreads(), which
is called only at early boot, when the check is appropriate. In addition,
it makes rcutree_prepare_cpu() invoke rcu_spawn_one_boost_kthread(), which
no longer does any checking for online CPUs.
With this change, RCU priority boosting tests now pass for short rcutorture
runs, even with single-CPU leaf rcu_node structures.
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Scott Wood <swood@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds each rcu_node structure's ->qsmask and "bBEG" output
indicating whether: (1) There is a boost kthread, (2) A reader needs
to be (or is in the process of being) boosted, (3) A reader is blocking
an expedited grace period, and (4) A reader is blocking a normal grace
period. This helps diagnose RCU priority boosting failures.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
If another lockdep report runs concurrently with an RCU lockdep report
from RCU_LOCKDEP_WARN(), the following sequence of events can occur:
1. debug_lockdep_rcu_enabled() sees that lockdep is enabled
when called from (say) synchronize_rcu().
2. Lockdep is disabled by a concurrent lockdep report.
3. debug_lockdep_rcu_enabled() evaluates its lockdep-expression
argument, for example, lock_is_held(&rcu_bh_lock_map).
4. Because lockdep is now disabled, lock_is_held() plays it safe and
returns the constant 1.
5. But in this case, the constant 1 is not safe, because invoking
synchronize_rcu() under rcu_read_lock_bh() is disallowed.
6. debug_lockdep_rcu_enabled() wrongly invokes lockdep_rcu_suspicious(),
resulting in a false-positive splat.
This commit therefore changes RCU_LOCKDEP_WARN() to check
debug_lockdep_rcu_enabled() after checking the lockdep expression,
so that any "safe" returns from lock_is_held() are rejected by
debug_lockdep_rcu_enabled(). This requires memory ordering, which is
supplied by READ_ONCE(debug_locks). The resulting volatile accesses
prevent the compiler from reordering and the fact that only one variable
is being accessed prevents the underlying hardware from reordering.
The combination works for IA64, which can reorder reads to the same
location, but this is defeated by the volatile accesses, which compile
to load instructions that provide ordering.
Reported-by: syzbot+dde0cc33951735441301@syzkaller.appspotmail.com
Reported-by: Matthew Wilcox <willy@infradead.org>
Reported-by: syzbot+88e4f02896967fe1ab0d@syzkaller.appspotmail.com
Reported-by: Thomas Gleixner <tglx@linutronix.de>
Suggested-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The reason that lockdep_rcu_suspicious() prints the value of debug_locks
is because a value of zero indicates a likely false positive. This can
work, but is a bit obtuse. This commit therefore explicitly calls out
the possibility of a false positive.
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds ->gp_max to show_rcu_gp_kthreads() output in order to
better diagnose RCU priority boosting failures.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds ->rt_priority and ->gp_start to show_rcu_gp_kthreads()
output in order to better diagnose RCU priority boosting failures.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, rcu_spawn_core_kthreads() is invoked via an early_initcall(),
which works, except that rcu_spawn_gp_kthread() is also invoked via an
early_initcall() and rcu_spawn_core_kthreads() relies on adjustments to
kthread_prio that are carried out by rcu_spawn_gp_kthread(). There is
no guaranttee of ordering among early_initcall() handlers, and thus no
guarantee that kthread_prio will be properly checked and range-limited
at the time that rcu_spawn_core_kthreads() needs it.
In most cases, this bug is harmless. After all, the only reason that
rcu_spawn_gp_kthread() adjusts the value of kthread_prio is if the user
specified a nonsensical value for this boot parameter, which experience
indicates is rare.
Nevertheless, a bug is a bug. This commit therefore causes the
rcu_spawn_core_kthreads() function to be invoked directly from
rcu_spawn_gp_kthread() after any needed adjustments to kthread_prio have
been carried out.
Fixes: 48d07c04b4 ("rcu: Enable elimination of Tree-RCU softirq processing")
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit cleans up some comments and code in kernel/rcu/tree.c.
Signed-off-by: Zhouyi Zhou <zhouzhouyi@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Commit 9ee01e0f69 ("x86/entry: Clean up idtentry_enter/exit()
leftovers") left the rcu_irq_exit_preempt() in place in order to avoid
conflicts with the -rcu tree. Now that this change has long since hit
mainline, this commit removes the no-longer-used rcu_irq_exit_preempt()
function.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
func_states_equal makes a very short lived allocation for idmap,
probably because it's too large to fit on the stack. However the
function is called quite often, leading to a lot of alloc / free
churn. Replace the temporary allocation with dedicated scratch
space in struct bpf_verifier_env.
Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Link: https://lore.kernel.org/bpf/20210429134656.122225-4-lmb@cloudflare.com
Resizing and copying stack and reference tracking state currently
does a lot of kfree / kmalloc when the size of the tracked set changes.
The logic in copy_*_state and realloc_*_state is also hard to follow.
Refactor this into two core functions. copy_array copies from a source
into a destination. It avoids reallocation by taking the allocated
size of the destination into account via ksize(). The function is
essentially krealloc_array, with the difference that the contents of
dst are not preserved. realloc_array changes the size of an array and
zeroes newly allocated items. Contrary to krealloc both functions don't
free the destination if the size is zero. Instead we rely on free_func_state
to clean up.
realloc_stack_state is renamed to grow_stack_state to better convey
that it never shrinks the stack state.
Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210429134656.122225-2-lmb@cloudflare.com
To make the purpose of the code more apparent, this commit moves the
tests of mem_dump_obj() to a new rcu_torture_mem_dump_obj() function
and calls it from rcu_torture_cleanup().
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
It will frequently be the case that rcu_torture_boost() will get a
->start_gp_poll() cookie that needs almost all of the current grace period
plus an additional grace period to elapse before ->poll_gp_state() will
return true. It is quite possible that the current grace period will have
(say) two seconds of stall by a CPU failing to pass through a quiescent
state, followed by 300 milliseconds of delay due to a preempted reader.
The next grace period might suffer only one second of stall by a CPU,
followed by another 300 milliseconds of delay due to a preempted reader.
This is an example of RCU priority boosting doing its job, but the full
elapsed time of 3.6 seconds exceeds the 3.5-second limit. In addition,
there is no CPU stall in force at the 3.5-second mark, so this would
nevertheless currently be counted as an RCU priority boosting failure.
This commit therefore avoids this sort of false positive by resetting
the gp_state_time timestamp any time that the current grace period is
being blocked by a CPU. This results in extremely frequent calls to
the ->check_boost_failed() function, so this commit provides a lockless
fastpath that is selected by supplying a NULL CPU-number pointer.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, rcu_torture_boost() runs CPU-bound at real-time priority
to force RCU priority inversions. It then checks that grace periods
progress during this CPU-bound time. If grace periods fail to progress,
it reports and RCU priority boosting failure.
However, it is possible (and sometimes does happen) that the grace period
fails to progress due to a CPU failing to pass through a quiescent state
for an extended time period (3.5 seconds by default). This can happen
due to vCPU preemption, long-running interrupts, and much else besides.
There is nothing that RCU priority boosting can do about these situations,
and so they should not be counted as RCU priority boosting failures.
This commit therefore checks for CPUs (as opposed to preempted tasks)
holding up a grace period, and flags the resulting RCU priority boosting
failures, but does not splat nor count them as errors. It does rate-limit
them to avoid flooding the console log.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
It is possible that a delayed grace period that rcu_torture_boost()
was polling for ended while rcu_torture_boost_failed() was printing the
failure splat. It would be good to know when this happens. This commit
therefore has rcu_torture_boost_failed() recheck the grace period after
printing the splat, and printing a message indicating whether or not
the grace period has ended.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit consolidates two loops in rcu_torture_boost(), one of which
counts the number of boost-test episodes and the other of which computes
the start time of the next episode, into one loop that does both with but
a single acquisition of boost_mutex. This means that the count of the
number of boost-test episodes is incremented after an episode completes
rather than before it starts, but it also avoids the over-counting that
was possible previously.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
If an rcu_torture_boost() kthread determines that its grace period
has not yet ended, it invokes rcu_torture_boost_failed() which checks
whether enough time has elapsed for this to be considered a failure of
RCU priority boosting, and, if so, flags the error.
Unfortunately, that kthread might be preempted for some seconds between
the time that it checks the grace period and the time that it checks the
time. This delay can result in a false positive, featuring a complaint
that a particular grace period has not ended, followed by a diagnostic
dump featuring a much later grace period.
This commit avoids these false positives by rechecking for the end of
the grace period after the time check.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, rcutorture's testing of RCU priority boosting insists not
only that grace periods complete, but also that callbacks be invoked.
Although this is in fact what the user would want, ensuring that there
is sufficient CPU bandwidth devoted to callback execution is in fact
the user's responsibility. One could argue that rcutorture can take on
that responsibility, which is true in theory. But in practice, ensuring
sufficient CPU bandwidth to ksoftirqd, any rcuc kthreads, and any rcuo
kthreads is not particularly consistent with rcutorture's main job,
that of stress-testing RCU. In addition, if the system administrator
(say) makes very poor choices when pinning rcuo kthreads and then runs
rcutorture, there really isn't much rcutorture can do.
Besides, RCU priority boosting only boosts lagging readers, not all the
machinery required to invoke callbacks in a timely fashion.
This commit therefore switches rcutorture's evaluation of RCU priority
boosting from callback execution to grace-period completion by using
the new start_poll_synchronize_rcu() and poll_state_synchronize_rcu()
functions. When rcutorture is built in (as in when there is no innocent
workload to inconvenience), the ksoftirqd ktheads are boosted to real-time
priority 2 in order to allow timeouts to work properly in the face of
rcutorture's testing of RCU priority boosting.
Indeed, it is not as easy as it looks to create a reliable test of RCU
priority boosting without destroying the rest of the kernel!
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds a (*readlock_held)() function pointer to the
rcu_torture_ops structure in order to make the rcu_torture_one_read()
function's rcu_dereference_check() lockdep expression more appropriate
for a given run.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds scale_type of acqrel, lock, and lock-irq to
test acquisition and release. Note that the refscale.nreaders=1
module parameter is required if you wish to test uncontended locking.
In contrast, acqrel uses a per-CPU variable, so should be just fine with
large values of the refscale.nreaders=1 module parameter.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds a block comment that gives a high-level overview of
how RCU Rude grace periods progress. It also gives an overview of the
memory ordering.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds a block comment that gives a high-level overview of how
RCU tasks grace periods progress. It also adds a note about how exiting
tasks are handled, plus it gives an overview of the memory ordering.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
An srcu_struct structure that is initialized before rcu_init_geometry()
will have its srcu_node hierarchy based on CONFIG_NR_CPUS. Once
rcu_init_geometry() is called, this hierarchy is compressed as needed
for the actual maximum number of CPUs for this system.
Later on, that srcu_struct structure is confused, sometimes referring
to its initial CONFIG_NR_CPUS-based hierarchy, and sometimes instead
to the new num_possible_cpus() hierarchy. For example, each of its
->mynode fields continues to reference the original leaf rcu_node
structures, some of which might no longer exist. On the other hand,
srcu_for_each_node_breadth_first() traverses to the new node hierarchy.
There are at least two bad possible outcomes to this:
1) a) A callback enqueued early on an srcu_data structure (call it
*sdp) is recorded pending on sdp->mynode->srcu_data_have_cbs in
srcu_funnel_gp_start() with sdp->mynode pointing to a deep leaf
(say 3 levels).
b) The grace period ends after rcu_init_geometry() shrinks the
nodes level to a single one. srcu_gp_end() walks through the new
srcu_node hierarchy without ever reaching the old leaves so the
callback is never executed.
This is easily reproduced on an 8 CPUs machine with CONFIG_NR_CPUS >= 32
and "rcupdate.rcu_self_test=1". The srcu_barrier() after early tests
verification never completes and the boot hangs:
[ 5413.141029] INFO: task swapper/0:1 blocked for more than 4915 seconds.
[ 5413.147564] Not tainted 5.12.0-rc4+ #28
[ 5413.151927] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 5413.159753] task:swapper/0 state:D stack: 0 pid: 1 ppid: 0 flags:0x00004000
[ 5413.168099] Call Trace:
[ 5413.170555] __schedule+0x36c/0x930
[ 5413.174057] ? wait_for_completion+0x88/0x110
[ 5413.178423] schedule+0x46/0xf0
[ 5413.181575] schedule_timeout+0x284/0x380
[ 5413.185591] ? wait_for_completion+0x88/0x110
[ 5413.189957] ? mark_held_locks+0x61/0x80
[ 5413.193882] ? mark_held_locks+0x61/0x80
[ 5413.197809] ? _raw_spin_unlock_irq+0x24/0x50
[ 5413.202173] ? wait_for_completion+0x88/0x110
[ 5413.206535] wait_for_completion+0xb4/0x110
[ 5413.210724] ? srcu_torture_stats_print+0x110/0x110
[ 5413.215610] srcu_barrier+0x187/0x200
[ 5413.219277] ? rcu_tasks_verify_self_tests+0x50/0x50
[ 5413.224244] ? rdinit_setup+0x2b/0x2b
[ 5413.227907] rcu_verify_early_boot_tests+0x2d/0x40
[ 5413.232700] do_one_initcall+0x63/0x310
[ 5413.236541] ? rdinit_setup+0x2b/0x2b
[ 5413.240207] ? rcu_read_lock_sched_held+0x52/0x80
[ 5413.244912] kernel_init_freeable+0x253/0x28f
[ 5413.249273] ? rest_init+0x250/0x250
[ 5413.252846] kernel_init+0xa/0x110
[ 5413.256257] ret_from_fork+0x22/0x30
2) An srcu_struct structure that is initialized before rcu_init_geometry()
and used afterward will always have stale rdp->mynode references,
resulting in callbacks to be missed in srcu_gp_end(), just like in
the previous scenario.
This commit therefore causes init_srcu_struct_nodes to initialize the
geometry, if needed. This ensures that the srcu_node hierarchy is
properly built and distributed from the get-go.
Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Once srcu_init() is called, the SRCU core will make use of delayed
workqueues, which rely on timers. However init_timers() is called
several steps after rcu_init(). This means that a call_srcu() after
rcu_init() but before init_timers() would find itself within a dangerously
uninitialized timer core.
This commit therefore creates a separate call to srcu_init() after
init_timer() completes, which ensures that we stay in early SRCU mode
until timers are safe(r).
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Pre-srcu_init() invocations of call_srcu() initialize the srcu_struct
structure in question, so there is no need to check this initialization
in srcu_init() when initiating grace periods for srcu_struct structures
that had early call_srcu() invocations. This commit therefore drops
the calls to check_init_srcu_struct() in srcu_init().
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Because alloc_percpu() zeroes out the allocated memory, there is no need
to zero-fill newly allocated per-CPU memory. This commit therefore removes
the loop zeroing the ->srcu_lock_count and ->srcu_unlock_count arrays
from init_srcu_struct_nodes(). This is the only use of that function's
is_static parameter, which this commit also removes.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently each CPU has its own ->nocb_timer queued when the nocb_gp
wakeup must be deferred. This approach has many drawbacks, compared to
a solution based on a single timer per NOCB group:
* There are a lot of timers to maintain.
* The per-rdp ->nocb_lock must be held to queue and cancel the timer
and this lock can already be heavily contended.
* One timer firing doesn't cancel the other timers in the same group:
- These other timers can thus cause spurious wakeups
- Each rdp that queued a timer must lock both ->nocb_lock and then
->nocb_gp_lock upon exit from the kernel to idle/user/guest mode.
* We can't cancel all of them if we detect an unflushed bypass in
nocb_gp_wait(). In fact currently we only ever cancel the ->nocb_timer
of the leader group.
* The leader group's nocb_timer is cancelled without locking ->nocb_lock
in nocb_gp_wait(). This currently appears to be safe but is an
accident waiting to happen.
* Since the timer acquires ->nocb_lock, it requires extra care in the
NOCB (de-)offloading process, requiring that it be either enabled or
disabled and then flushed.
This commit instead uses the rcuog kthread's CPU's ->nocb_timer instead.
It is protected by nocb_gp_lock, which is _way_ less contended and
remains so even after this change. As a matter of fact, the nocb_timer
almost never fires and the deferred wakeup is mostly carried out upon
idle/user/guest entry. Now the early check performed at this point in
do_nocb_deferred_wakeup() is done on rdp_gp->nocb_defer_wakeup, which
is of course racy. However, this raciness is harmless because we only
need the guarantee that the timer is queued if we were the last one to
queue it. Any other situation (another CPU has queued it and we either
see it or not) is fine.
This solves all the issues listed above.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently we have three functions which depend on each other.
Two of them are quite tiny and the last one where the most
work is done. All of them are related to queuing RCU batches
to reclaim objects after a GP.
1. kfree_rcu_monitor(). It consist of few lines. It acquires a spin-lock
and calls kfree_rcu_drain_unlock().
2. kfree_rcu_drain_unlock(). It also consists of few lines of code. It
calls queue_kfree_rcu_work() to queue the batch. If this fails,
it rearms the monitor work to try again later.
3. queue_kfree_rcu_work(). This provides the bulk of the functionality,
attempting to start a new batch to free objects after a GP.
Since there are no external users of functions [2] and [3], both
can eliminated by moving all logic directly into [1], which both
shrinks and simplifies the code.
Also replace comments which start with "/*" to "//" format to make it
unified across the file.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The kvfree_rcu() function now defers allocations in the common
case due to the fact that there is no lockless access to the
memory-allocator caches/pools. In addition, in CONFIG_PREEMPT_NONE=y
and in CONFIG_PREEMPT_VOLUNTARY=y kernels, there is no reliable way to
determine if spinlocks are held. As a result, allocation is deferred in
the common case, and the two-argument form of kvfree_rcu() thus uses the
"channel 3" queue through all the rcu_head structures. This channel
is called referred to as the emergency case in comments, and these
comments are now obsolete.
This commit therefore updates these comments to reflect the new
common-case nature of such emergencies.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Replace an open-coded version of the kfree_rcu_monitor() function body
with a call to that function.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Before attempting to start a new batch the "monitor_todo" variable is
set to "false" and set back to "true" when a previous RCU batch is still
in progress. This is at best confusing.
Thus change this variable to "false" only when a new batch has been
successfully queued, otherwise, just leave it be.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_scheduler_active flag is set to RCU_SCHEDULER_RUNNING once the
scheduler is up and running. That signal is used in order to check
and queue a "monitor work" to reclaim freed objects (if there are any)
during early boot. This flag is used by kvfree_rcu() to determine when
work can safely be queued, at which point memory passed to earlier
invocations of kvfree_rcu() can be processed.
However, only "krcp->head" is checked for objects that need to be
released, and there are now two more, namely, "krcp->bkvhead[0]" and
"krcp->bkvhead[1]". Therefore, check these two additional channels.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
nr_bkv_objs is a count of the objects in the kvfree_rcu page cache.
Accessing it requires holding the ->lock. Switch to READ_ONCE() and
WRITE_ONCE() macros to provide lockless access to this counter.
This lockless access is used for the shrinker.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Add a drain_page_cache() function to drain a per-cpu page cache.
The reason behind of it is a system can run into a low memory
condition, in that case a page shrinker can ask for its users
to free their caches in order to get extra memory available for
other needs in a system.
When a system hits such condition, a page cache is drained for
all CPUs in a system. By default a page cache work is delayed
with 5 seconds interval until a memory pressure disappears, if
needed it can be changed. See a rcu_delay_page_cache_fill_msec
module parameter.
Co-developed-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Zqiang <qiang.zhang@windriver.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
