Fix: move wait loop increment before first conditional block
The fix "Fix: high cpu usage in synchronize_rcu with long RCU read-side
C.S." has an imperfection in urcu.c and urcu-qsbr.c: when incrementing
the wait loop counter for the last time, the first conditional branch is
not taken, but the following conditionals are, and they assume the first
conditional has been taken.
Within urcu.c (urcu-mb, urcu-membarrier and urcu-signal), and
urcu-qsbr.c, this will simply skip the first wait_gp() call, without any
noticeable ill side-effect.
Fix: high cpu usage in synchronize_rcu with long RCU read-side C.S.
We noticed that with this kind of scenario:
- application using urcu-mb, urcu-membarrier, urcu-signal, or urcu-bp,
- long RCU read-side critical sections, caused by e.g. long network I/O
system calls,
- other short lived RCU critical sections running in other threads,
- very frequent invocation of call_rcu to enqueue callbacks,
lead to abnormally high CPU usage within synchronize_rcu() in the
call_rcu worker threads.
Inspection of the code gives us the answer: in urcu.c, we expect that if
we need to wait on a futex (wait_gp()), we expect to be able to end the
grace period within the next loop, having been notified by a
rcu_read_unlock(). However, this is not always the case: we can very
well be awakened by a rcu_read_unlock() executed on a thread running
short-lived RCU read-side critical sections, while the long-running RCU
read-side C.S. is still active. We end up in a situation where we
busy-wait for a very long time, because the counter is !=
RCU_QS_ACTIVE_ATTEMPTS until a 32-bit overflow happens (or more likely,
until we complete the grace period). We need to change the wait_loops ==
RCU_QS_ACTIVE_ATTEMPTS check into an inequality to use wait_gp() for
every attempts beyond RCU_QS_ACTIVE_ATTEMPTS loops.
urcu-bp.c also has this issue. Moreover, it uses usleep() rather than
poll() when dealing with long-running RCU read-side critical sections.
Turn the usleep 1000us (1ms) into a poll of 10ms. One of the advantage
of using poll() rather than usleep() is that it does not interact with
SIGALRM.
urcu-qsbr.c already checks for wait_loops >= RCU_QS_ACTIVE_ATTEMPTS, so
it is not affected by this issue.
Looking into these loops, however, shows that overflow of the loop
counter, although unlikely, would bring us back to a situation of high
cpu usage (a negative value well below RCU_QS_ACTIVE_ATTEMPTS).
Therefore, change the counter behavior so it stops incrementing when it
reaches RCU_QS_ACTIVE_ATTEMPTS, to eliminate overflow.
Fix: urcu-bp interaction with threads vs constructors/destructors
Add a reference counter for threads using urcu-bp, thus ensuring that
even if the urcu destructor is executed before each thread using RCU
read-side critical sections exit, those threads will not see a corrupted
thread list.
Also, don't use URCU_TLS() within urcu_bp_thread_exit_notifier(). It
appears that this is racy (although this was probably due to the issue
fixed by reference counting). Anyway, play safe, and pass the rcu_key
received as parameter instead.
Those issues only reproduce when threads are still active when the
urcu-bp destructor is called.
Clang 3.3 with -O2 optimisations is especially picky about arithmetic
on NULL pointers. This undefined behavior is turned into optimized out
NULL checks by clang 3.3. Fix the undefined behavior by checking against
the pointer directly, without going back and forth around NULL with
pointer arithmetic.
Keep ABI compatible with already compiled LGPL applications
Applications with _LGPL_SOURCE defined that were compiled against bogus
tls-compat.h header *and* which are using multiple urcu flavors
concurrently need to be told that they need to be recompiled against a
fixed tls-compat.h header. Detect these usage, and abort() with a
message error on stderr.
This is needed for stable-0.7 and stable-0.8 branches of userspace RCU
only. The upcoming 0.9 will bump the soname version, and therefore does
not have to care about this aspect of ABI compatibility.
Vladimir Nikulichev noticed crashes when using this setup. The problem
can be pinpointed to a missing macro expansion in urcu/tls-compat.h:
looking at the output of
nm tests/unit/.libs/test_urcu_multiflavor :
U __tls_access_rcu_reader
this seems to be the issue. We're missing macro expansion in
tls-compat.h. With this commit, it becomes:
U __tls_access_rcu_reader_bp
U __tls_access_rcu_reader_mb
U __tls_access_rcu_reader_memb
U __tls_access_rcu_reader_sig
Please note that this affects an unusual configuration of userspace RCU
(with TLS pthread key fallback), needed for some BSD that don't support
compiler TLS. Strictly speaking, this requires bumping the URCU library
soname version major number, because it breaks the ABI presented to
applications on those unusual configurations.
A following commit will handle the ABI migration: for stable releases
(stable-0.7 and stable-0.8 branches), the ABI is kept compatible, and
bogus usage are detected. For the upcoming stable-0.9, the soname will
simply be bumped.
Reported-by: Vladimir Nikulichev <nvs@tbricks.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
When compiling code using the rcu_xchg_pointer() family of functions,
with the following define:
#define URCU_INLINE_SMALL_FUNCTIONS
prior to including urcu headers, when compiling with gcc with
-Wsign-compare and -Wextra, gcc warns about:
urcu-xchg.c: In function ‘reload’:
urcu-xchg.c:19:1: warning: ordered comparison of pointer with integer zero [-Wextra]
urcu-xchg.c:19:1: warning: signed and unsigned type in conditional expression [-Wsign-compare]
For the "ordered comparison of pointer with integer zero" warning, fix
this by comparing (type) -1 against (type) 0 instead of just 0, so if
"type" is a pointer type, this pointer type will be applied to the right
operand too, thus fixing the warning.
For the "signed and unsigned type in conditional expression" warning, we
need caa_cast_long_keep_sign() to always evaluate to the same type
signedness. In order to do so, when we need to sign-extend the value,
cast it to unsigned long after first casting it to long.
Reported-by: Stephen Hemminger <stephen@networkplumber.org> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
* Dmitri Shubin <sbn@tbricks.com> wrote:
> Shouldn't the condition in line 94 actually be
>
> 94 #if (!defined(BUILD_QSBR_LIB) && !defined(RCU_DEBUG))
>
> So when RCU_DEBUG is _not_ defined we get static inlines for
> rcu_read_{,un}lock() ?
Indeed!
Reported-by: Dmitri Shubin <sbn@tbricks.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
This fixes an issue that appears after this recent urcu-bp fix is
applied:
Fix: urcu-bp: Bulletproof RCU arena resize bug
Prior to this fix, on Linux at least, the behavior was to allocate
(and leak) one memory map region per reader thread. It worked, except
for the unfortunate leak. The fact that it worked, even though not the
way we had intended it to, is is why testing did not raise any red flag.
That state of affairs has prevailed for a long time, but it was
side-tracking some issues. After fixing the underlying bug that was
causing the memory map leak, another issue appears.
The garbage collection scheme reclaiming the thread tracking structures
in urcu-bp fails in stress tests to due a bug in glibc (tested against
glibc 2.13 and 2.17). Under this workload, on a 2-core/hyperthreaded i7:
./test_urcu_bp 40 4 10
we can easily trigger a segmentation fault in the pthread_kill() code.
Program terminated with signal 11, Segmentation fault.
Backtrace:
#0 __pthread_kill (threadid=140723681437440, signo=0) at ../nptl/sysdeps/unix/sysv/linux/pthread_kill.c:42
42 ../nptl/sysdeps/unix/sysv/linux/pthread_kill.c: No such file or directory.
(gdb) bt full
#0 __pthread_kill (threadid=140723681437440, signo=0) at ../nptl/sysdeps/unix/sysv/linux/pthread_kill.c:42
__x = <optimized out>
pd = 0x7ffcc90b2700
tid = <optimized out>
val = <optimized out>
#1 0x0000000000403009 in rcu_gc_registry () at ../../urcu-bp.c:437
tid = 140723681437440
ret = 0
chunk = 0x7ffcca0b8000
rcu_reader_reg = 0x7ffcca0b8120
__PRETTY_FUNCTION__ = "rcu_gc_registry"
#2 0x0000000000402b9c in synchronize_rcu_bp () at ../../urcu-bp.c:230
cur_snap_readers = {next = 0x7ffcb4888cc0, prev = 0x7ffcb4888cc0}
qsreaders = {next = 0x7ffcb4888cd0, prev = 0x7ffcb4888cd0}
newmask = {__val = {18446744067267100671, 18446744073709551615 <repeats 15 times>}}
oldmask = {__val = {0, 140723337334144, 0, 0, 0, 140723690351643, 0, 140723127058464, 4, 0, 140723698253920, 140723693868864, 4096, 140723690370432, 140723698253920, 140723059951840}}
ret = 0
__PRETTY_FUNCTION__ = "synchronize_rcu_bp"
#3 0x0000000000401803 in thr_writer (_count=0x76b2f0) at test_urcu_bp.c:223
count = 0x76b2f0
new = 0x7ffca80008c0
old = 0x7ffca40008c0
#4 0x00007ffcc9c83f8e in start_thread (arg=0x7ffcb4889700) at pthread_create.c:311
__res = <optimized out>
pd = 0x7ffcb4889700
now = <optimized out>
unwind_buf = {cancel_jmp_buf = {{jmp_buf = {140723337336576, 6546223316613858487, 0, 140723698253920, 140723693868864, 4096, -6547756131873848137,
-6547872135220034377}, mask_was_saved = 0}}, priv = {pad = {0x0, 0x0, 0x0, 0x0}, data = {prev = 0x0, cleanup = 0x0, canceltype = 0}}}
not_first_call = 0
pagesize_m1 = <optimized out>
sp = <optimized out>
freesize = <optimized out>
__PRETTY_FUNCTION__ = "start_thread"
#5 0x00007ffcc99ade1d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:113
It appears that the memory backing the thread information can be
relinquished by NPTL concurrently with execution of pthread_kill()
targeting an already joined thread and cause this segfault. We were
using pthread_kill(tid, 0) to discover if the target thread was alive or
not, as documented in pthread_kill(3):
If sig is 0, then no signal is sent, but error checking is still per‐
formed; this can be used to check for the existence of a thread ID.
but it appears that the glibc implementation is racy.
Instead of using the racy pthread_kill implementation, implement cleanup
using a pthread_key destroy notifier for a dummy key. This notifier is
called for each thread exit and destroy.
It is not correct to move the registry address range, since there are
external references from reader threads. This will trigger on workloads
with many threads.
Typically, on Linux, mremap can expand the existing range, which is OK.
However, if there is not enough space around the existing range, it may
try to map it at a different address, which is incorrect.
It is more likely that this bug will be observed on operating systems
where urcu uses the mmap/munmap fallback instead of mremap.
Moreover, prior to commit:
"Fix: urcu-bp: Bulletproof RCU arena resize bug"
this issue was hidden by the fact that each thread ended up with their
own memory mapping (leaked), on Linux at least.
compat_futex.c has one instance included in each urcu shared object, as
well as within some of the test applications. However, it is expected
that an entire program interact with the same lock and completion
variables. Therefore, define them as globally visible, but weak, so the
entire program agree on which object should be used.
Reported-by: Vladimir Nikulichev <nvs@tbricks.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
compat_arch_x86.c is linked into many .so and even into test programs.
The basic problem with this is that it contains a statically defined
mutex, which will fail to protect concurrent use of this compat code by
different shared objects.
Fix this by defining both the mutex (now called __urcu_x86_compat_mutex)
and __rcu_cas_avail as weak symbols. Therefore, the first symbol that
gets loaded in a program will by used by everyone.
Reported-by: Vladimir Nikulichev <nvs@tbricks.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> While trying to use the BP flavor of RCU I ran into random crashes. I
> tracked it down to issues with resizing of the BP RCU memory pool.
>
> The problem is in the urcu-bp.c file in the resize_arena() function.
> On successful allocation / remapping the len member of the
> registry_arena struct is never set anywhere function. On the second
> resize of the arena the code in resize_arena() still thinks the
> previous size is equal to the original mapping size. I've fixed this
> issue locally by just adding the following code at the bottom of
> resize_arena().
Good catch !!
However, I think your fix misses one case: if we happen to re-use the
same region, we want to update the length too.
Fix: hash table growth (for small tables) should be limited
Buckets with many entries encountered in a hash table could cause it to
grow to a large size, beyond the scope for which this mechanism is
expected to play a role when node accounting is available. Indeed, when
the hash table grows to larger size, split-counter node accounting is
expected to deal with resize/shrink rather than relying on an heuristic
based on the largest bucket size.
This is fixing an issue where we see hash tables sometimes reaching 65k
entries index (65536*8 = 524288 bytes) for a workload limited to adding
1000 entries and then removing all of them, done in a loop (random
keys).
Fix: Use a filled signal mask to disable all signals
Changelog from David Pelton's original patch:
While using lttng-ust with an application that was calling fork()
with pending signals, I found that all signals were getting unmasked
shortly before the underlying call to fork(). After some
investigation, I found that the rcu_bp_before_fork() function was
unmasking all signals. Based on the comments for this function, it
should be masking all signals. Inspection of the rest of the code
in urcu-bp.c revealed the same pattern in two other functions.
This patch changes the code to use a filled signal mask to disable
all signals. The change to rcu_bp_before_fork() addressed the
problem I was seeing while using lttng-ust. The changes to the
other two functions appear to fix other instances of the same
problem.
Updates by Mathieu Desnoyers:
- Use SIG_BLOCK instead of SIG_SETMASK when setting a filled mask. This
has the same behavior in this case (since we're blocking all signals),
but is semantically neater: if we ever some signals from that mask,
we'd like to to a union with the signal mask already blocked by the
application.
- Also fix incorrect signal masking in compat_arch_x86.c.
Reported-by: David Pelton <dpelton@ciena.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Some sparc Debian setups advertise a "sparc" host cpu (rather than
sparc64).
In all cases, I think it should be safe to add a "sparc" entry to
userspace RCU configure.ac upstream, e.g.
[sparc], [ARCHTYPE="sparc64"],
in the event someone would launch the build on an environment not
supporting sparc v9, the build would fail because the 32-bit compiler
would not be able to generate sparc v9 instructions (unless
explicitely instructed to do so by the -m32 -Wa,-Av9a flags).
Fix hurd-i386: move cpuset tests outside of sched_setaffinity conditional
Comment about introduction of cpuset.h within urcu tests:
> Unfortunately it doesn't work, because sched_setaffinity is for now
> just a fail-stub on hurd-i386, and thus configure considers it as
> missing, and thus the CPU_SET test is disabled completely.
>
> I however guess you could just disable defining your own cpu_set_t
> when !HAVE_SCHED_SETAFFINITY, since it is probably used only for using
> sched_setaffinity.
Fix by moving cpu_set_t, CPU_SET and CPU_ZERO tests outside of the
sched_setaffinity conditional.
Reported-by: Samuel Thibault <sthibault@debian.org> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Fix tests: finer-grained use of CPU_SET, CPU_ZERO and cpu_set_t
Noticed build failure at
https://buildd.debian.org/status/package.php?p=liburcu :
Tail of log for liburcu on hurd-i386:
test_urcu.c:110:0: warning: "CPU_SET" redefined [enabled by default]
In file included from /usr/include/pthread/pthread.h:50:0,
from /usr/include/pthread.h:2,
from test_urcu.c:26:
/usr/include/sched.h:80:0: note: this is the location of the previous definition
make[3]: *** [test_urcu.o] Error 1
make[2]: *** [all-recursive] Error 1
make[1]: *** [all] Error 2
dh_auto_build: make -j1 returned exit code 2
make: *** [build-arch] Error 2
dpkg-buildpackage: error: debian/rules build-arch gave error exit status 2
make[3]: Entering directory `/build/buildd-liburcu_0.7.6-1-hurd-i386-wGBAtt/liburcu-0.7.6/tests'
CC test_urcu.o
make[3]: Leaving directory `/build/buildd-liburcu_0.7.6-1-hurd-i386-wGBAtt/liburcu-0.7.6/tests'
make[2]: Leaving directory `/build/buildd-liburcu_0.7.6-1-hurd-i386-wGBAtt/liburcu-0.7.6'
Fix build on architectures with HAVE_SCHED_GETCPU but without HAVE_SYSCONF
Noticed on: https://buildd.debian.org/status/package.php?p=liburcu
Tail of log for liburcu on kfreebsd-amd64:
CC urcu.lo
In file included from urcu.c:450:0:
urcu-call-rcu-impl.h:145:12: error: static declaration of 'sched_getcpu' follows non-static declaration
In file included from /usr/include/sched.h:43:0,
from /usr/include/pthread.h:20,
from urcu.c:30:
/usr/include/x86_64-kfreebsd-gnu/bits/sched.h:65:12: note: previous declaration of 'sched_getcpu' was here
make[3]: *** [urcu.lo] Error 1
make[3]: Leaving directory `/build/buildd-liburcu_0.7.6-1-kfreebsd-amd64-nnkICd/liburcu-0.7.6'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/build/buildd-liburcu_0.7.6-1-kfreebsd-amd64-nnkICd/liburcu-0.7.6'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/build/buildd-liburcu_0.7.6-1-kfreebsd-amd64-nnkICd/liburcu-0.7.6'
dh_auto_build: make -j1 returned exit code 2
make: *** [build-arch] Error 2
Tail of log for liburcu on kfreebsd-i386:
CC urcu.lo
In file included from urcu.c:450:0:
urcu-call-rcu-impl.h:145:12: error: static declaration of 'sched_getcpu' follows non-static declaration
In file included from /usr/include/sched.h:43:0,
from /usr/include/pthread.h:20,
from urcu.c:30:
/usr/include/i386-kfreebsd-gnu/bits/sched.h:65:12: note: previous declaration of 'sched_getcpu' was here
make[3]: *** [urcu.lo] Error 1
make[3]: Leaving directory `/build/buildd-liburcu_0.7.6-1-kfreebsd-i386-sWzNKU/liburcu-0.7.6'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/build/buildd-liburcu_0.7.6-1-kfreebsd-i386-sWzNKU/liburcu-0.7.6'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/build/buildd-liburcu_0.7.6-1-kfreebsd-i386-sWzNKU/liburcu-0.7.6'
dh_auto_build: make -j1 returned exit code 2
make: *** [build-arch] Error 2
CMM_STORE_SHARED(x, v) is a macro that really acts like an assignment
expression, e.g.:
x = v;
but internally also has "mc" barriers (useful for cache-incoherent
architectures).
The issue here is that (x = v) can evaluate to "v", but very often we're
not interested to use the assignment expression result. When we have an
explicit assignment, the compiler won't complain that the result of this
expression is unused, but given that the added barrier requires that we
make this macro evaluate explicitly to a value, clang complains.
Fix this by adding "_v = _v" at the last line of the macro, thus
performing what would appear like an effect-less assignment, but
actually tricks clang into thinking we are evaluating to an assignment
expression, thus suppressing the warning.
I've had a report of someone running into issues with the RCU lock-free
hash table by embedding the struct cds_lfht_node into a packed structure
by mistake, thus not respecting alignment requirements stated in
urcu/rculfhash.h. Assertions on "replace" and "add" operations should
catch this, but I notice that we should add assertions on the
REMOVAL_OWNER_FLAG to cover all possible misalignments.
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Discourage use of pthread_atfork() for call_rcu handlers
Discourage use of glibc pthread_atfork() for call_rcu handlers due to
its inappropriate assumptions about single-threadedness while pthread
atfork handlers are executing. This results in hangs within the glibc
memory allocator.
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Fix call_rcu fork handling by putting all call_rcu threads in a
quiescent state before fork (paused state), and unpausing them when the
parent returns from fork.
On the child, everything will run fine as long as we don't issue fork()
from a call_rcu callback.
Side-note: pthread_atfork is not appropriate when using with multithread
and malloc/free. The glibc malloc implementation sadly expects that all
malloc/free are executed from the context of a single thread while
pthread atfork handlers are running, which leads to interesting hang in
glibc.
Fix TLS detection: test with linker, add --disable-compiler-tls
NetBSD 5.1 and older, as well as Darwin, succeed to compile code
containing TLS, but cannot link it. Test with linker in addition to
compiler for TLS support.
Also add a --disable-compiler-tls configure option to allow users to
force using the pthread getspecific fall back.
Cleanup: cast pthread_self() return value to unsigned long
pthread_t can map to other things that unsigned long (e.g. pointer).
Cast it to unsigned long for debug printing and for debug delay random
value purposes.
Fallback mechanism not working on platform where TLS is unsupported
The CONFIG_RCU_TLS entry in config.h.in is defined by default to "TLS".
This has the unfortunate consequence of defining CONFIG_RCU_TLS on
platform where TLS is unsupported and effectively disabling the pthread
based fallback mechanism. This macro should be #undef by default and the
AX_TLS m4 macro will properly detect if TLS is supported.
Signed-off-by: Christian Babeux <christian.babeux@efficios.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
* Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> * Lai Jiangshan (laijs@cn.fujitsu.com) wrote:
> > test code:
> > ./tests/test_urcu_lfs 100 10 10
> >
> > bug produce rate > 60%
> >
> > {{{
> > I didn't see any bug when "./tests/test_urcu_lfs 10 10 10" Or
> +"./tests/test_urcu_lfs 100 100 10"
> > But I just test it about 5 times
> > }}}
> >
> > 4cores*1threads: Intel(R) Core(TM) i5 CPU 760
> > RCU_MB (no time to test for other rcu type)
> > test commit: 768fba83676f49eb73fd1d8ad452016a84c5ec2a
> >
> > I didn't see any bug when "./tests/test_urcu_mb 10 100 10"
> >
> > Sorry, I tried, but I failed to find out the root cause currently.
>
> I think I managed to narrow down the issue:
>
> 1) the master branch does not reproduce it, but commit
> 768fba83676f49eb73fd1d8ad452016a84c5ec2a repdroduces it about 50% of the
> time.
>
> 2) the main change between 768fba83676f49eb73fd1d8ad452016a84c5ec2a and
> current master (f94061a3df4c9eab9ac869a19e4228de54771fcb) is call_rcu
> moving to wfcqueue.
>
> 3) the bug always arise, for me, at the end of the 10 seconds.
> However, it might be simply due to the fact that most of the memory
> get freed at the end of program execution.
>
> 4) I've been able to get a backtrace, and it looks like we have some
> call_rcu callback-invokation threads still working while
> call_rcu_data_free() is invoked. In the backtrace, call_rcu_data_free()
> is nicely waiting for the next thread to stop, and during that time,
> two callback-invokation threads are invoking callbacks (and one of
> them triggers the segfault).
>
> So I expect that commit
>
> commit 5161f31e09ce33dd79afad8d08a2372fbf1c4fbe
> Author: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> Date: Tue Sep 25 10:50:49 2012 -0500
>
> call_rcu: use wfcqueue, eliminate false-sharing
>
> Eliminate false-sharing between call_rcu (enqueuer) and worker threads
> on the queue head and tail.
>
> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
>
> Could have managed to fix the issue, or change the timing enough that it
> does not reproduces. I'll continue investigating.
The bug was in call rcu. It is not required for master, because we fixed
it while moving to wfcqueue. We were erroneously writing to the head
field of the default call_rcu_data rather than tail.
The conditions to reproduce this bug:
1) setup per-cpu callback-invokation threads,
2) use call_rcu
3) call call_rcu_data_free() while there are still some pending
callbacks that have not yet been executed by the callback-invokation
threads,
4) we then get corruption due to the "default" callback invokation
that walks through a corrupted queue.
Ensure that read-side functions meet 10-line LGPL criterion
This commit ensures that all read-side functions meet the 10-line LGPL
criterion that permits them to be expanded directly into non-LGPL code,
without function-call instructions. It also documents this as the
intent.
[ paulmck: Spelling fixes called out by Josh Triplett and name
change called out by Mathieu Desnoyers (_rcu_read_lock_help() ->
_rcu_read_lock_update(). ]
[ Mathieu Desnoyers: _rcu_read_unlock_help renamed to
_rcu_read_unlock_update_and_wakeup, and spelling fix for
preced -> precede. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Lai Jiangshan [Thu, 9 Aug 2012 14:24:38 +0000 (10:24 -0400)]
urcu: move busy-wait code and name it ___cds_wfq_node_sync_next()
This code which waits for a node's next pointer until it appears, will
be used many times, move it to a help function and name it
___cds_wfq_node_sync_next().
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Lai Jiangshan [Thu, 9 Aug 2012 14:19:14 +0000 (10:19 -0400)]
urcu: fix compat_futex_noasync()
This patch fix two critical problems in the compatibility fallback of
compact_futex_noasync():
1) compat_futex_cond is not bound to any @uaddr, it services all @uaddr,
if you wakeup only one thread(pthread_cond_signal), the @uaddr of
this waking thread and the @uaddr of the woken-up thread may be different.
The woken-up thread will very probably go to sleep again
because his own condition is not true.
*And* this waking thread(FUTEX_WAKE) wake up NOTHING.
2) If the caller want to wake up all waiting threads, he will use INT_MAX
for @val, and:
for (i = 0; i < INT_MAX; i++)
pthread_cond_signal(&compat_futex_cond);
becomes almost infinity loop.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
[ Edit by Mathieu Desnoyers: add explanations about supported
MIPS architectures, extracted from conversation with Ralf Baechle:
* Supported architectures
Ralf Baechle (edited by Mathieu Desnoyers):
This code works on all MIPS architecture variants. The memory barrier
instruction, SYNC, was introduced for MIPS II. The original MIPS I
instruction set doesn't have SYNC nor SMP support in the processor
architecture itself so SMP required horrible kludges in the system
hardware. I think it's safe to say that Linux/MIPS will never support
any of these MIPS I SMP systems. In the unlikely case this happens
anyway, we have a (Linux) kernel emulation of the SYNC instruction.
Voila - full binary compatibility across all MIPS processors and the
oldest hardware pays the performance penalty.
* Choice of barrier for cmm_mb()/cmm_rmb()/cmm_wmb()
Ralf Baechle:
"RMI (aka Netlogic and now Broadcom) XLR processor cores can be
configured to permit LD-LD, LD-ST, ST-LD and ST-ST reordering; default
is only ST-ST reordering. To allow Linux to eventually enable full
reordering cmm_mb(), cmm_rmb() and cmm_wmb() all should perform SYNC
and a compiler barrier."
* No-op choice for cmm_read_barrier_depends():
Ralf Baechle:
"Technically there is nothing in the MIPS architecture spec that would
keep a MIPS implementation from reordering as freely as an Alpha or
even more liberally. In practice most do strong ordering. However
there is no MIPS implementation that makes full use of all the rope
provided. So in theory a paranoid implementation of
cmm_read_barrier_depends() for MIPS should perform a SYNC. In reality
it's not necessary and no sane MIPS core designer would implement
something that would design a core that need a non-empty
cmm_read_barrier_depends(). The reason why my patch had an empty one
is that I was using the Alpha code as a template."
Mathieu Desnoyers:
Moreover, the Linux kernel chooses a no-op for MIPS
read_barrier_depends() implementation, so any MIPS architecture that
would be as weak as Alpha would break the Linux kernel before breaking
the userspace RCU library.
* No need to put ".set noreorder" in cmm_mb() inline assembly:
Ralf Baechle:
"Certain instructions such as SYNC won't get reordered." ]
> I tried to build the latest urcu (git master e51500) on a Centos 6.2 box, and got:
>
> jscott@dxi0-62:~/src/userspace-rcu$ make -j4
> CDPATH="${ZSH_VERSION+.}:" && cd . && /bin/sh /users/jscott/src/userspace-rcu/config/missing --run aclocal-1.11 -I
> +config
> CDPATH="${ZSH_VERSION+.}:" && cd . && /bin/sh /users/jscott/src/userspace-rcu/config/missing --run autoconf
> cd . && /bin/sh /users/jscott/src/userspace-rcu/config/missing --run automake-1.11 --foreign
> configure:4010: error: possibly undefined macro: m4_ifnblank
> If this token and others are legitimate, please use m4_pattern_allow.
> See the Autoconf documentation.
> make: *** [configure] Error 1
> make: *** Waiting for unfinished jobs....
>
> Some digging showed that the macro m4_ifnblank requires autoconf 2.64. Centos 6.2 has autoconf 2.63. :(
>
> I just worked around it by reverting commit a767fd locally, then I can build fine.
Reported-by: John Steele Scott <toojays@toojays.net> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> CC rcutorture_urcu-urcutorture.o
> In file included from urcutorture.c:9:
> api.h: In function '__smp_thread_id':
> api.h:160: warning: cast from pointer to integer of different size
> api.h:160: warning: cast from pointer to integer of different size
> api.h: In function 'wait_thread':
> api.h:210: warning: cast from pointer to integer of different size
> api.h:210: warning: cast from pointer to integer of different size
Commit 4d0d66bb795d1ed938e11a97a4e5f71326e20c71, implementing
tls-compat.h for pthread TLS compatibility, adds a prefix in front of
each TLS symbol (__tls_*). However, some of these symbols are exported
by the URCU library (e.g. rcu_reader_mb, defined in urcu.c as
"rcu_reader", which is overloaded by the urcu/map/urcu.h) to
applications. Therefore, this breaks binary compatibility with 0.6.x
versions of the library. This is not intended, and therefore is a bug,
so we remove this __tls_* prefix from the variables declared, defined
and referenced to through the tls-compat.h API for compilers supporting
"__thread".
NetBSD 5 implements a mremap with a different semantic. Rename our
wrapper symbol name so it does not clash with the NetBSD 5 symbol.
Eventually, we could envision doing a special-case that uses the NetBSD
5 version instead of the fallback, but let's first get it working before
going into optimization land.
Suggested-by: Marek Vavruša <marek.vavrusa@nic.cz> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
We choose to provide full memory barriers before and after successful
hash table update operations. Eventually, new API with weaker semantic
can be added, but let's make the basic API as fool-proof as possible.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Reviewed-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Use cmm_smp_mb__before_uatomic_or() prior to the uatomic_or() in
_rcu_lfht_del() to ensure correct memory barrier semantic when we relax
(in the future) the barrier implementation of some architectures.
These currently translate into simple compiler barriers on all
architectures, but the and/or/add/sub/inc/dec uatomics do not provide
memory ordering guarantees (only uatomic_add_return, uatomic_sub_return,
uatomic_xchg, and uatomic_cmpxchg provides full memory barrier
guarantees before and after the atomic operations).
T0 T1
replace A by B
del A
read A->next
-> check REMOVED flag, not set yet.
read A->next
-> check REMOVED flag, not set yet.
cmpxchg A->next to set REMOVED flag
-> cmpxchg succeeds
uatomic_or to set REMOVED flag
uatomic_xchg to atomically set the REMOVAL_OWNER flag
-> first to set the flag.
Replace returns node -> free(A) Del success -> free(A)
With this race, we have a double-free.
The problem with the replace code is that it does not set the
"REMOVAL_OWNER" flag.
Test case to reproduce the bug:
test_urcu_hash 0 2 20 -A -s -M 1 -N 1 -O 1
(2 threads, doing replace/del, with a hash table that has only a single
key for all values). After just a couple of seconds, either the program
hangs, or, more often, it does:
rculfhash: replace unneeded rcu_dereference by CMM_LOAD_SHARED
The difference between the two is that CMM_LOAD_SHARED() does not imply
a read barrier between the read and following uses of the data pointed
to by the pointer read.
All sites that only use the pointer load for its bits (never
dereference) don't need the read barrier implied by rcu_dereference.
Reviewed-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
avoid empty statement.
-------------------------
if (condition)
dbg_printf() /* forget ";", but compiler say nothing if dbg_printf() is empty */
statement;
-------------------------
also add printf format check.
(we can use gcc extention "__printf(1, 2)" to declare a dummy inline function
to do the check, but I use "printf()" directly here)
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
projects / urcu.git / log
