The walk through the cgroup hierarchy during the enqueue/dequeue of a task
is split in 2 distinct parts for throttled cfs_rq without any added value
but making code less readable.
Change the code ordering such that everything related to a cfs_rq
(throttled or not) will be done in the same loop.
In addition, the same steps ordering is used when updating a cfs_rq:
- update_load_avg
- update_cfs_group
- update *h_nr_running
This reordering enables the use of h_nr_running in PELT algorithm.
No functional and performance changes are expected and have been noticed
during tests.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: "Dietmar Eggemann <dietmar.eggemann@arm.com>"
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Phil Auld <pauld@redhat.com>
Cc: Hillf Danton <hdanton@sina.com>
Link: https://lore.kernel.org/r/20200224095223.13361-5-mgorman@techsingularity.net
Commit 219ca39427 ("audit: use union for audit_field values since
they are mutually exclusive") combined a number of separate fields in
the audit_field struct into a single union. Generally this worked
just fine because they are generally mutually exclusive.
Unfortunately in audit_data_to_entry() the overlap can be a problem
when a specific error case is triggered that causes the error path
code to attempt to cleanup an audit_field struct and the cleanup
involves attempting to free a stored LSM string (the lsm_str field).
Currently the code always has a non-NULL value in the
audit_field.lsm_str field as the top of the for-loop transfers a
value into audit_field.val (both .lsm_str and .val are part of the
same union); if audit_data_to_entry() fails and the audit_field
struct is specified to contain a LSM string, but the
audit_field.lsm_str has not yet been properly set, the error handling
code will attempt to free the bogus audit_field.lsm_str value that
was set with audit_field.val at the top of the for-loop.
This patch corrects this by ensuring that the audit_field.val is only
set when needed (it is cleared when the audit_field struct is
allocated with kcalloc()). It also corrects a few other issues to
ensure that in case of error the proper error code is returned.
Cc: stable@vger.kernel.org
Fixes: 219ca39427 ("audit: use union for audit_field values since they are mutually exclusive")
Reported-by: syzbot+1f4d90ead370d72e450b@syzkaller.appspotmail.com
Signed-off-by: Paul Moore <paul@paul-moore.com>
Pull irq fixes from Thomas Gleixner:
"Two fixes for the irq core code which are follow ups to the recent MSI
fixes:
- The WARN_ON which was put into the MSI setaffinity callback for
paranoia reasons actually triggered via a callchain which escaped
when all the possible ways to reach that code were analyzed.
The proc/irq/$N/*affinity interfaces have a quirk which came in
when ALPHA moved to the generic interface: In case that the written
affinity mask does not contain any online CPU it calls into ALPHAs
magic auto affinity setting code.
A few years later this mechanism was also made available to x86 for
no good reasons and in a way which circumvents all sanity checks
for interrupts which cannot have their affinity set from process
context on X86 due to the way the X86 interrupt delivery works.
It would be possible to make this work properly, but there is no
point in doing so. If the interrupt is not yet started then the
affinity setting has no effect and if it is started already then it
is already assigned to an online CPU so there is no point to
randomly move it to some other CPU. Just return EINVAL as the code
has done before that change forever.
- The new MSI quirk bit in the irq domain flags turned out to be
already occupied, which escaped the author and the reviewers
because the already in use bits were 0,6,2,3,4,5 listed in that
order.
That bit 6 was simply overlooked because the ordering was straight
forward linear otherwise. So the new bit ended up being a
duplicate.
Fix it up by switching the oddball 6 to the obvious 1"
* tag 'irq-urgent-2020-02-22' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
genirq/irqdomain: Make sure all irq domain flags are distinct
genirq/proc: Reject invalid affinity masks (again)
Pull s390 fixes from Vasily Gorbik:
- Remove ieee_emulation_warnings sysctl which is a dead code.
- Avoid triggering rebuild of the kernel during make install.
- Enable protected virtualization guest support in default configs.
- Fix cio_ignore seq_file .next function to increase position index.
And use kobj_to_dev instead of container_of in cio code.
- Fix storage block address lists to contain absolute addresses in qdio
code.
- Few clang warnings and spelling fixes.
* tag 's390-5.6-4' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
s390/qdio: fill SBALEs with absolute addresses
s390/qdio: fill SL with absolute addresses
s390: remove obsolete ieee_emulation_warnings
s390: make 'install' not depend on vmlinux
s390/kaslr: Fix casts in get_random
s390/mm: Explicitly compare PAGE_DEFAULT_KEY against zero in storage_key_init_range
s390/pkey/zcrypt: spelling s/crytp/crypt/
s390/cio: use kobj_to_dev() API
s390/defconfig: enable CONFIG_PROTECTED_VIRTUALIZATION_GUEST
s390/cio: cio_ignore_proc_seq_next should increase position index
Daniel Borkmann says:
====================
pull-request: bpf-next 2020-02-21
The following pull-request contains BPF updates for your *net-next* tree.
We've added 25 non-merge commits during the last 4 day(s) which contain
a total of 33 files changed, 2433 insertions(+), 161 deletions(-).
The main changes are:
1) Allow for adding TCP listen sockets into sock_map/hash so they can be used
with reuseport BPF programs, from Jakub Sitnicki.
2) Add a new bpf_program__set_attach_target() helper for adding libbpf support
to specify the tracepoint/function dynamically, from Eelco Chaudron.
3) Add bpf_read_branch_records() BPF helper which helps use cases like profile
guided optimizations, from Daniel Xu.
4) Enable bpf_perf_event_read_value() in all tracing programs, from Song Liu.
5) Relax BTF mandatory check if only used for libbpf itself e.g. to process
BTF defined maps, from Andrii Nakryiko.
6) Move BPF selftests -mcpu compilation attribute from 'probe' to 'v3' as it has
been observed that former fails in envs with low memlock, from Yonghong Song.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 736b46027e ("net: Add ID (if needed) to sock_reuseport and expose
reuseport_lock") has introduced lazy generation of reuseport group IDs that
survive group resize.
By comparing the identifier we check if BPF reuseport program is not trying
to select a socket from a BPF map that belongs to a different reuseport
group than the one the packet is for.
Because SOCKARRAY used to be the only BPF map type that can be used with
reuseport BPF, it was possible to delay the generation of reuseport group
ID until a socket from the group was inserted into BPF map for the first
time.
Now that SOCK{MAP,HASH} can be used with reuseport BPF we have two options,
either generate the reuseport ID on map update, like SOCKARRAY does, or
allocate an ID from the start when reuseport group gets created.
This patch takes the latter approach to keep sockmap free of calls into
reuseport code. This streamlines the reuseport_id access as its lifetime
now matches the longevity of reuseport object.
The cost of this simplification, however, is that we allocate reuseport IDs
for all SO_REUSEPORT users. Even those that don't use SOCKARRAY in their
setups. With the way identifiers are currently generated, we can have at
most S32_MAX reuseport groups, which hopefully is sufficient. If we ever
get close to the limit, we can switch an u64 counter like sk_cookie.
Another change is that we now always call into SOCKARRAY logic to unlink
the socket from the map when unhashing or closing the socket. Previously we
did it only when at least one socket from the group was in a BPF map.
It is worth noting that this doesn't conflict with sockmap tear-down in
case a socket is in a SOCK{MAP,HASH} and belongs to a reuseport
group. sockmap tear-down happens first:
prot->unhash
`- tcp_bpf_unhash
|- tcp_bpf_remove
| `- while (sk_psock_link_pop(psock))
| `- sk_psock_unlink
| `- sock_map_delete_from_link
| `- __sock_map_delete
| `- sock_map_unref
| `- sk_psock_put
| `- sk_psock_drop
| `- rcu_assign_sk_user_data(sk, NULL)
`- inet_unhash
`- reuseport_detach_sock
`- bpf_sk_reuseport_detach
`- WRITE_ONCE(sk->sk_user_data, NULL)
Suggested-by: Martin Lau <kafai@fb.com>
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200218171023.844439-10-jakub@cloudflare.com
SOCKMAP & SOCKHASH now support storing references to listening
sockets. Nothing keeps us from using these map types a collection of
sockets to select from in BPF reuseport programs. Whitelist the map types
with the bpf_sk_select_reuseport helper.
The restriction that the socket has to be a member of a reuseport group
still applies. Sockets in SOCKMAP/SOCKHASH that don't have sk_reuseport_cb
set are not a valid target and we signal it with -EINVAL.
The main benefit from this change is that, in contrast to
REUSEPORT_SOCKARRAY, SOCK{MAP,HASH} don't impose a restriction that a
listening socket can be just one BPF map at the same time.
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200218171023.844439-9-jakub@cloudflare.com
Pull networking fixes from David Miller:
1) Limit xt_hashlimit hash table size to avoid OOM or hung tasks, from
Cong Wang.
2) Fix deadlock in xsk by publishing global consumer pointers when NAPI
is finished, from Magnus Karlsson.
3) Set table field properly to RT_TABLE_COMPAT when necessary, from
Jethro Beekman.
4) NLA_STRING attributes are not necessary NULL terminated, deal wiht
that in IFLA_ALT_IFNAME. From Eric Dumazet.
5) Fix checksum handling in atlantic driver, from Dmitry Bezrukov.
6) Handle mtu==0 devices properly in wireguard, from Jason A.
Donenfeld.
7) Fix several lockdep warnings in bonding, from Taehee Yoo.
8) Fix cls_flower port blocking, from Jason Baron.
9) Sanitize internal map names in libbpf, from Toke Høiland-Jørgensen.
10) Fix RDMA race in qede driver, from Michal Kalderon.
11) Fix several false lockdep warnings by adding conditions to
list_for_each_entry_rcu(), from Madhuparna Bhowmik.
12) Fix sleep in atomic in mlx5 driver, from Huy Nguyen.
13) Fix potential deadlock in bpf_map_do_batch(), from Yonghong Song.
14) Hey, variables declared in switch statement before any case
statements are not initialized. I learn something every day. Get
rids of this stuff in several parts of the networking, from Kees
Cook.
* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (99 commits)
bnxt_en: Issue PCIe FLR in kdump kernel to cleanup pending DMAs.
bnxt_en: Improve device shutdown method.
net: netlink: cap max groups which will be considered in netlink_bind()
net: thunderx: workaround BGX TX Underflow issue
ionic: fix fw_status read
net: disable BRIDGE_NETFILTER by default
net: macb: Properly handle phylink on at91rm9200
s390/qeth: fix off-by-one in RX copybreak check
s390/qeth: don't warn for napi with 0 budget
s390/qeth: vnicc Fix EOPNOTSUPP precedence
openvswitch: Distribute switch variables for initialization
net: ip6_gre: Distribute switch variables for initialization
net: core: Distribute switch variables for initialization
udp: rehash on disconnect
net/tls: Fix to avoid gettig invalid tls record
bpf: Fix a potential deadlock with bpf_map_do_batch
bpf: Do not grab the bucket spinlock by default on htab batch ops
ice: Wait for VF to be reset/ready before configuration
ice: Don't tell the OS that link is going down
ice: Don't reject odd values of usecs set by user
...
Currently, if rcu_barrier() returns too soon, the test waits 100ms and
then does another instance of the test. However, if rcu_barrier() were
to have waited for more than 100ms too short a time, this could cause
the test's rcu_head structures to be reused while they were still on
RCU's callback lists. This can result in knock-on errors that obscure
the original rcu_barrier() test failure.
This commit therefore adds code that attempts to wait until all of
the test's callbacks have been invoked. Of course, if RCU completely
lost track of the corresponding rcu_head structures, this wait could be
forever. This commit therefore also complains if this attempted recovery
takes more than one second, and it also gives up when the test ends.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, rcu_torture_barrier_cbs() posts callbacks from whatever CPU
it is running on, which means that all these kthreads might well be
posting from the same CPU, which would drastically reduce the effectiveness
of this test. This commit therefore uses IPIs to make the callbacks be
posted from the corresponding CPU (given by local variable myid).
If the IPI fails (which can happen if the target CPU is offline or does
not exist at all), the callback is posted on whatever CPU is currently
running.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
During changes to kfree_rcu() code, we often check the amount of free
memory. As an alternative to checking this manually, this commit adds a
measurement in the test itself. It measures four times during the test
for available memory, digitally filters these measurements to produce a
running average with a weight of 0.5, and compares this digitally filtered
value with the amount of available memory at the beginning of the test.
Something like the following is printed at the end of the run:
Total time taken by all kfree'ers: 6369738407 ns, loops: 10000, batches: 764, memory footprint: 216MB
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcutorture global variable rcu_torture_current is accessed locklessly,
so it must use the RCU pointer load/store primitives. This commit
therefore adds several that were missed.
This data race was reported by KCSAN. Not appropriate for backporting due
to failure being unlikely and due to this being used only by rcutorture.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcutorture rcu_torture_count and rcu_torture_batch per-CPU variables
are read locklessly, so this commit adds the READ_ONCE() to a load in
order to avoid various types of compiler vandalism^Woptimization.
This data race was reported by KCSAN. Not appropriate for backporting
due to failure being unlikely and due to this being rcutorture.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_fwd_cb_nodelay variable suppresses excessively long read-side
delays while carrying out an rcutorture forward-progress test. As such,
it is accessed both by readers and updaters, and most of the accesses
therefore use *_ONCE(). Except for one in rcu_read_delay(), which this
commit fixes.
This data race was reported by KCSAN. Not appropriate for backporting
due to this being rcutorture.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The ->rtort_pipe_count field in the rcu_torture structure checks for
too-short grace periods, and is therefore read by rcutorture's readers
while being updated by rcutorture's writers. This commit therefore
adds the needed READ_ONCE() and WRITE_ONCE() invocations.
This data race was reported by KCSAN. Not appropriate for backporting
due to failure being unlikely and due to this being rcutorture.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
In theory, RCU-hotplug operations are supposed to work as soon as there
is more than one CPU online. However, in practice, in normal production
there is no way to make them happen until userspace is up and running.
Besides which, on smaller systems, rcutorture doesn't start doing hotplug
operations until 30 seconds after the start of boot, which on most
systems also means the better part of 30 seconds after the end of boot.
This commit therefore provides a new torture.disable_onoff_at_boot kernel
boot parameter that suppresses CPU-hotplug torture operations until
about the time that init is spawned.
Of course, if you know of a need for boottime CPU-hotplug operations,
then you should avoid passing this argument to any of the torture tests.
You might also want to look at the splats linked to below.
Link: https://lore.kernel.org/lkml/20191206185208.GA25636@paulmck-ThinkPad-P72/
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
In normal production, an excessively long wait on a grace period
(synchronize_rcu(), for example) at boottime is often just as bad
as at any other time. In fact, given the desire for fast boot, any
sort of long wait at boot is a bad idea. However, heavy rcutorture
testing on large hyperthreaded systems can generate such long waits
during boot as a matter of course. This commit therefore causes
the rcupdate.rcu_cpu_stall_suppress_at_boot kernel boot parameter to
suppress reporting of bootime bad-sequence warning due to excessively
long grace-period waits.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
In normal production, an RCU CPU stall warning at boottime is often
just as bad as at any other time. In fact, given the desire for fast
boot, any sort of long-term stall at boot is a bad idea. However,
heavy rcutorture testing on large hyperthreaded systems can generate
boottime RCU CPU stalls as a matter of course. This commit therefore
provides a kernel boot parameter that suppresses reporting of boottime
RCU CPU stall warnings and similarly of rcutorture writer stalls.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
During boot, CPU hotplug is often disabled, for example by PCI probing.
On large systems that take substantial time to boot, this can result
in spurious RCU_HOTPLUG errors. This commit therefore forgives any
boottime -EBUSY CPU-hotplug failures by adjusting counters to pretend
that the corresponding attempt never happened. A non-splat record
of the failed attempt is emitted to the console with the added string
"(-EBUSY forgiven during boot)".
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Additional rcutorture aggression can result in, believe it or not,
boot times in excess of three minutes on large hyperthreaded systems.
This is long enough for rcutorture to decide to do some callback flooding,
which seems a bit excessive given that userspace cannot have started
until long after boot, and it is userspace that does the real-world
callback flooding. Worse yet, because Tiny RCU lacks forward-progress
functionality, the looping-in-the-kernel tests can also be problematic
during early boot.
This commit therefore causes rcutorture to hold off on callback
flooding until about the time that init is spawned, and the same
for looping-in-the-kernel tests for Tiny RCU.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Some larger systems can take in excess of 50 seconds to complete their
early boot initcalls prior to spawing init. This does not in any way
help the forward-progress judgments of built-in rcutorture (when
rcutorture is built as a module, the insmod or modprobe command normally
cannot happen until some time after boot completes). This commit
therefore suppresses such complaints until about the time that init
is spawned.
This also includes a fix to a stupid error located by kbuild test robot.
[ paulmck: Apply kbuild test robot feedback. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
[ paulmck: Fix to nohz_full slow-expediting recovery logic, per bpetkov. ]
[ paulmck: Restrict splat to CONFIG_PREEMPT_RT=y kernels and simplify. ]
Tested-by: Borislav Petkov <bp@alien8.de>
A read of the srcu_struct structure's ->srcu_gp_seq field should not
need READ_ONCE() when that structure's ->lock is held. Except that this
lock is not always held when updating this field. This commit therefore
acquires the lock around updates and removes a now-unneeded READ_ONCE().
This data race was reported by KCSAN.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
[ paulmck: Switch from READ_ONCE() to lock per Peter Zilstra question. ]
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
The srcu_struct structure's ->srcu_idx field is accessed locklessly,
so reads must use READ_ONCE(). This commit therefore adds the needed
READ_ONCE() invocation where it was missed.
This data race was reported by KCSAN. Not appropriate for backporting
due to failure being unlikely.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The srcu_struct structure's ->srcu_gp_seq_needed_exp field is accessed
locklessly, so updates must use WRITE_ONCE(). This commit therefore
adds the needed WRITE_ONCE() invocations.
This data race was reported by KCSAN. Not appropriate for backporting
due to failure being unlikely.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The srcu_node structure's ->srcu_gp_seq_needed_exp field is accessed
locklessly, so updates must use WRITE_ONCE(). This commit therefore
adds the needed WRITE_ONCE() invocations.
This data race was reported by KCSAN. Not appropriate for backporting
due to failure being unlikely.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Sparse reports a warning at exit_tasks_rcu_finish(void)
|warning: context imbalance in exit_tasks_rcu_finish()
|- wrong count at exit
To fix this, this commit adds a __releases(&tasks_rcu_exit_srcu).
Given that exit_tasks_rcu_finish() does actually call __srcu_read_lock(),
this not only fixes the warning but also improves on the readability of
the code.
Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Sparse reports a warning at exit_tasks_rcu_start(void)
|warning: context imbalance in exit_tasks_rcu_start() - wrong count at exit
To fix this, this commit adds an __acquires(&tasks_rcu_exit_srcu).
Given that exit_tasks_rcu_start() does actually call __srcu_read_lock(),
this not only fixes the warning but also improves on the readability of
the code.
Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
The RCU tasks list of callbacks, rcu_tasks_cbs_head, is sampled locklessly
by rcu_tasks_kthread() when waiting for work to do. This commit therefore
applies READ_ONCE() to that lockless sampling and WRITE_ONCE() to the
single potential store outside of rcu_tasks_kthread.
This data race was reported by KCSAN. Not appropriate for backporting
due to failure being unlikely.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The __call_rcu() function's header comment refers to a cpu argument
that no longer exists, and the comment of the return path from
rcu_nocb_try_bypass() ignores the non-no-CBs CPU case. This commit
therefore update both comments.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit fixes a spelling mistake in a pr_info() message.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
RCU priority boosting currently is not applied until the grace period
is at least 250 milliseconds old (or the number of milliseconds specified
by the CONFIG_RCU_BOOST_DELAY Kconfig option). Although this has worked
well, it can result in OOM under conditions of RCU callback flooding.
One can argue that the real-time systems using RCU priority boosting
should carefully avoid RCU callback flooding, but one can just as well
argue that an OOM is a rather obnoxious error message.
This commit therefore disables the RCU priority boosting delay when
there are excessive numbers of callbacks queued.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
In default configutions, RCU currently waits at least 100 milliseconds
before asking cond_resched() and/or resched_rcu() for help seeking
quiescent states to end a grace period. But 100 milliseconds can be
one good long time during an RCU callback flood, for example, as can
happen when user processes repeatedly open and close files in a tight
loop. These 100-millisecond gaps in successive grace periods during a
callback flood can result in excessive numbers of callbacks piling up,
unnecessarily increasing memory footprint.
This commit therefore asks cond_resched() and/or resched_rcu() for help
as early as the first FQS scan when at least one of the CPUs has more
than 20,000 callbacks queued, a number that can be changed using the new
rcutree.qovld kernel boot parameter. An auxiliary qovld_calc variable
is used to avoid acquisition of locks that have not yet been initialized.
Early tests indicate that this reduces the RCU-callback memory footprint
during rcutorture floods by from 50% to 4x, depending on configuration.
Reported-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Reported-by: Tejun Heo <tj@kernel.org>
[ paulmck: Fix bug located by Qian Cai. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Tested-by: Dexuan Cui <decui@microsoft.com>
Tested-by: Qian Cai <cai@lca.pw>
The rcu_data structure's ->core_needs_qs field does not necessarily get
cleared in a timely fashion after the corresponding CPUs' quiescent state
has been reported. From a functional viewpoint, no harm done, but this
can result in excessive invocation of RCU core processing, as witnessed
by the kernel test robot, which saw greatly increased softirq overhead.
This commit therefore restores the rcu_report_qs_rdp() function's
clearing of this field, but only when running on the corresponding CPU.
Cases where some other CPU reports the quiescent state (for example, on
behalf of an idle CPU) are handled by setting this field appropriately
within the __note_gp_changes() function's end-of-grace-period checks.
This handling is carried out regardless of whether the end of a grace
period actually happened, thus handling the case where a CPU goes non-idle
after a quiescent state is reported on its behalf, but before the grace
period ends. This fix also avoids cross-CPU updates to ->core_needs_qs,
While in the area, this commit changes the __note_gp_changes() need_gp
variable's name to need_qs because it is a quiescent state that is needed
from the CPU in question.
Fixes: ed93dfc6bc ("rcu: Confine ->core_needs_qs accesses to the corresponding CPU")
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
If CPU hotplug testing is enabled, a lock might appear to be maximally
unfair just because one of the CPUs was offline almost all the time.
This commit therefore forgives unfairness if CPU hotplug testing was
enabled.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Both lock_torture_writer() and lock_torture_reader() use the "static"
keyword on their DEFINE_TORTURE_RANDOM(rand) declarations, which means
that a single instance of a random-number generator are shared among all
the writers and another is shared among all the readers. Unfortunately,
this random-number generator was not designed for concurrent access.
This commit therefore removes both "static" keywords so that each reader
and each writer gets its own random-number generator.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The __torture_print_stats() function in locktorture.c carefully
initializes local variable "min" to statp[0].n_lock_acquired, but
then compares it to statp[i].n_lock_fail. Given that the .n_lock_fail
field should normally be zero, and given the initialization, it seems
reasonable to display the maximum and minimum number acquisitions
instead of miscomputing the maximum and minimum number of failures.
This commit therefore switches from failures to acquisitions.
And this turns out to be not only a day-zero bug, but entirely my
own fault. I hate it when that happens!
Fixes: 0af3fe1efa ("locktorture: Add a lock-torture kernel module")
Reported-by: Will Deacon <will@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Acked-by: Will Deacon <will@kernel.org>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Peter Zijlstra <peterz@infradead.org>
The event is given three parameters, first one is the name
of RCU flavour, second one is the number of elements in array
for free and last one is an address of the array holding
pointers to be freed by the kfree_bulk() function.
To enable the trace event your kernel has to be build with
CONFIG_RCU_TRACE=y, after that it is possible to track the
events using ftrace subsystem.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The kfree_rcu() logic can be improved further by using kfree_bulk()
interface along with "basic batching support" introduced earlier.
The are at least two advantages of using "bulk" interface:
- in case of large number of kfree_rcu() requests kfree_bulk()
reduces the per-object overhead caused by calling kfree()
per-object.
- reduces the number of cache-misses due to "pointer chasing"
between objects which can be far spread between each other.
This approach defines a new kfree_rcu_bulk_data structure that
stores pointers in an array with a specific size. Number of entries
in that array depends on PAGE_SIZE making kfree_rcu_bulk_data
structure to be exactly one page.
Since it deals with "block-chain" technique there is an extra
need in dynamic allocation when a new block is required. Memory
is allocated with GFP_NOWAIT | __GFP_NOWARN flags, i.e. that
allows to skip direct reclaim under low memory condition to
prevent stalling and fails silently under high memory pressure.
The "emergency path" gets maintained when a system is run out of
memory. In that case objects are linked into regular list.
The "rcuperf" was run to analyze this change in terms of memory
consumption and kfree_bulk() throughput.
1) Testing on the Intel(R) Xeon(R) W-2135 CPU @ 3.70GHz, 12xCPUs
with following parameters:
kfree_loops=200000 kfree_alloc_num=1000 kfree_rcu_test=1 kfree_vary_obj_size=1
dev.2020.01.10a branch
Default / CONFIG_SLAB
53607352517 ns, loops: 200000, batches: 1885, memory footprint: 1248MB
53529637912 ns, loops: 200000, batches: 1921, memory footprint: 1193MB
53570175705 ns, loops: 200000, batches: 1929, memory footprint: 1250MB
Patch / CONFIG_SLAB
23981587315 ns, loops: 200000, batches: 810, memory footprint: 1219MB
23879375281 ns, loops: 200000, batches: 822, memory footprint: 1190MB
24086841707 ns, loops: 200000, batches: 794, memory footprint: 1380MB
Default / CONFIG_SLUB
51291025022 ns, loops: 200000, batches: 1713, memory footprint: 741MB
51278911477 ns, loops: 200000, batches: 1671, memory footprint: 719MB
51256183045 ns, loops: 200000, batches: 1719, memory footprint: 647MB
Patch / CONFIG_SLUB
50709919132 ns, loops: 200000, batches: 1618, memory footprint: 456MB
50736297452 ns, loops: 200000, batches: 1633, memory footprint: 507MB
50660403893 ns, loops: 200000, batches: 1628, memory footprint: 429MB
in case of CONFIG_SLAB there is double increase in performance and
slightly higher memory usage. As for CONFIG_SLUB, the performance
figures are better together with lower memory usage.
2) Testing on the HiKey-960, arm64, 8xCPUs with below parameters:
CONFIG_SLAB=y
kfree_loops=200000 kfree_alloc_num=1000 kfree_rcu_test=1
102898760401 ns, loops: 200000, batches: 5822, memory footprint: 158MB
89947009882 ns, loops: 200000, batches: 6715, memory footprint: 115MB
rcuperf shows approximately ~12% better throughput in case of
using "bulk" interface. The "drain logic" or its RCU callback
does the work faster that leads to better throughput.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Tested-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, nocb_gp_wait() unconditionally complains if there is a
callback not already associated with a grace period. This assumes that
either there was no such callback initially on the one hand, or that
the rcu_advance_cbs() function assigned all such callbacks to a grace
period on the other. However, in theory there are some situations that
would prevent rcu_advance_cbs() from assigning all of the callbacks.
This commit therefore checks for unassociated callbacks immediately after
rcu_advance_cbs() returns, while the corresponding rcu_node structure's
->lock is still held. If there are unassociated callbacks at that point,
the subsequent WARN_ON_ONCE() is disabled.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The ->nocb_lock lockdep assertion is currently guarded by cpu_online(),
which is incorrect for no-CBs CPUs, whose callback lists must be
protected by ->nocb_lock regardless of whether or not the corresponding
CPU is online. This situation could result in failure to detect bugs
resulting from failing to hold ->nocb_lock for offline CPUs.
This commit therefore removes the cpu_online() guard.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit reworks the atomic_cmpxchg() loop in rcu_eqs_special_set()
to do only the initial read from the current CPU's rcu_data structure's
->dynticks field explicitly. On subsequent passes, this value is instead
retained from the failing atomic_cmpxchg() operation.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Sparse reports warning at rcu_nocb_bypass_unlock()
warning: context imbalance in rcu_nocb_bypass_unlock() - unexpected unlock
The root cause is a missing annotation of rcu_nocb_bypass_unlock()
which causes the warning.
This commit therefore adds the missing __releases(&rdp->nocb_bypass_lock)
annotation.
Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Acked-by: Boqun Feng <boqun.feng@gmail.com>
Sparse reports warning at rcu_nocb_bypass_lock()
|warning: context imbalance in rcu_nocb_bypass_lock() - wrong count at exit
To fix this, this commit adds an __acquires(&rdp->nocb_bypass_lock).
Given that rcu_nocb_bypass_lock() does actually call raw_spin_lock()
when raw_spin_trylock() fails, this not only fixes the warning but also
improves on the readability of the code.
Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently rcu_check_gp_start_stall() complains if a grace period takes
too long to start, where "too long" is roughly one RCU CPU stall-warning
interval. This has worked well, but there are some debugging Kconfig
options (such as CONFIG_EFI_PGT_DUMP=y) that can make booting take a
very long time, so much so that the stall-warning interval has expired
before RCU's grace-period kthread has even been spawned.
This commit therefore resets the rcu_state.gp_req_activity and
rcu_state.gp_activity timestamps just before the grace-period kthread
is spawned, and modifies the checks and adds ordering to ensure that
if rcu_check_gp_start_stall() sees that the grace-period kthread
has been spawned, that it will also see the resets applied to the
rcu_state.gp_req_activity and rcu_state.gp_activity timestamps.
Reported-by: Qian Cai <cai@lca.pw>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
[ paulmck: Fix whitespace issues reported by Qian Cai. ]
Tested-by: Qian Cai <cai@lca.pw>
[ paulmck: Simplify grace-period wakeup check per Steve Rostedt feedback. ]
The rcu_barrier_callback() function does an atomic_dec_and_test(), and
if it is the last CPU to check in, does the required wakeup. Either way,
it does an event trace. Unfortunately, this is susceptible to the
following sequence of events:
o CPU 0 invokes rcu_barrier_callback(), but atomic_dec_and_test()
says that it is not last. But at this point, CPU 0 is delayed,
perhaps due to an NMI, SMI, or vCPU preemption.
o CPU 1 invokes rcu_barrier_callback(), and atomic_dec_and_test()
says that it is last. So CPU 1 traces completion and does
the needed wakeup.
o The awakened rcu_barrier() function does cleanup and releases
rcu_state.barrier_mutex.
o Another CPU now acquires rcu_state.barrier_mutex and starts
another round of rcu_barrier() processing, including updating
rcu_state.barrier_sequence.
o CPU 0 gets its act back together and does its tracing. Except
that rcu_state.barrier_sequence has already been updated, so
its tracing is incorrect and probably quite confusing.
(Wait! Why did this CPU check in twice for one rcu_barrier()
invocation???)
This commit therefore causes rcu_barrier_callback() to take a
snapshot of the value of rcu_state.barrier_sequence before invoking
atomic_dec_and_test(), thus guaranteeing that the event-trace output
is sensible, even if the timing of the event-trace output might still
be confusing. (Wait! Why did the old rcu_barrier() complete before
all of its CPUs checked in???) But being that this is RCU, only so much
confusion can reasonably be eliminated.
This data race was reported by KCSAN. Not appropriate for backporting
due to failure being unlikely and due to the mild consequences of the
failure, namely a confusing event trace.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
