Commit 4c41c69e changed the way the coroutine pool is sized because for
virtio-blk devices with a large queue size and heavy I/O, it was just
too small and caused coroutines to be deleted and reallocated soon
afterwards. The change made the size dynamic based on the number of
queues and the queue size of virtio-blk devices.
There are two important numbers here: Slightly simplified, when a
coroutine terminates, it is generally stored in the global release pool
up to a certain pool size, and if the pool is full, it is freed.
Conversely, when allocating a new coroutine, the coroutines in the
release pool are reused if the pool already has reached a certain
minimum size (the batch size), otherwise we allocate new coroutines.
The problem after commit 4c41c69e is that it not only increases the
maximum pool size (which is the intended effect), but also the batch
size for reusing coroutines (which is a bug). It means that in cases
with many devices and/or a large queue size (which defaults to the
number of vcpus for virtio-blk-pci), many thousand coroutines could be
sitting in the release pool without being reused.
This is not only a waste of memory and allocations, but it actually
makes the QEMU process likely to hit the vm.max_map_count limit on Linux
because each coroutine requires two mappings (its stack and the guard
page for the stack), causing it to abort() in qemu_alloc_stack() because
when the limit is hit, mprotect() starts to fail with ENOMEM.
In order to fix the problem, change the batch size back to 64 to avoid
uselessly accumulating coroutines in the release pool, but keep the
dynamic maximum pool size so that coroutines aren't freed too early
in heavy I/O scenarios.
Note that this fix doesn't strictly make it impossible to hit the limit,
but this would only happen if most of the coroutines are actually in use
at the same time, not just sitting in a pool. This is the same behaviour
as we already had before commit 4c41c69e. Fully preventing this would
require allowing qemu_coroutine_create() to return an error, but it
doesn't seem to be a scenario that people hit in practice.
Cc: qemu-stable@nongnu.org
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2079938
Fixes: 4c41c69e05
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20220510151020.105528-3-kwolf@redhat.com>
Tested-by: Hiroki Narukawa <hnarukaw@yahoo-corp.jp>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It's true that these functions currently affect the batch size in which
coroutines are reused (i.e. moved from the global release pool to the
allocation pool of a specific thread), but this is a bug and will be
fixed in a separate patch.
In fact, the comment in the header file already just promises that it
influences the pool size, so reflect this in the name of the functions.
As a nice side effect, the shorter function name makes some line
wrapping unnecessary.
Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20220510151020.105528-2-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Thread-Local Storage variables cannot be used directly from coroutine
code because the compiler may optimize TLS variable accesses across
qemu_coroutine_yield() calls. When the coroutine is re-entered from
another thread the TLS variables from the old thread must no longer be
used.
Use QEMU_DEFINE_STATIC_CO_TLS() for the current and leader variables.
The alloc_pool QSLIST needs a typedef so the return value of
get_ptr_alloc_pool() can be stored in a local variable.
One example of why this code is necessary: a coroutine that yields
before calling qemu_coroutine_create() to create another coroutine is
affected by the TLS issue.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220307153853.602859-3-stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Coroutine pool size was 64 from long ago, and the basis was organized in the commit message in 4d68e86b.
At that time, virtio-blk queue-size and num-queue were not configuable, and equivalent values were 128 and 1.
Coroutine pool size 64 was fine then.
Later queue-size and num-queue got configuable, and default values were increased.
Coroutine pool with size 64 exhausts frequently with random disk IO in new size, and slows down.
This commit adjusts coroutine pool size adaptively with new values.
This commit adds 64 by default, but now coroutine is not only for block devices,
and is not too much burdon comparing with new default.
pool size of 128 * vCPUs.
Signed-off-by: Hiroki Narukawa <hnarukaw@yahoo-corp.jp>
Message-id: 20220214115302.13294-2-hnarukaw@yahoo-corp.jp
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
clang's C11 atomic_fetch_*() functions only take a C11 atomic type
pointer argument. QEMU uses direct types (int, etc) and this causes a
compiler error when a QEMU code calls these functions in a source file
that also included <stdatomic.h> via a system header file:
$ CC=clang CXX=clang++ ./configure ... && make
../util/async.c:79:17: error: address argument to atomic operation must be a pointer to _Atomic type ('unsigned int *' invalid)
Avoid using atomic_*() names in QEMU's atomic.h since that namespace is
used by <stdatomic.h>. Prefix QEMU's APIs with 'q' so that atomic.h
and <stdatomic.h> can co-exist. I checked /usr/include on my machine and
searched GitHub for existing "qatomic_" users but there seem to be none.
This patch was generated using:
$ git grep -h -o '\<atomic\(64\)\?_[a-z0-9_]\+' include/qemu/atomic.h | \
sort -u >/tmp/changed_identifiers
$ for identifier in $(</tmp/changed_identifiers); do
sed -i "s%\<$identifier\>%q$identifier%g" \
$(git grep -I -l "\<$identifier\>")
done
I manually fixed line-wrap issues and misaligned rST tables.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200923105646.47864-1-stefanha@redhat.com>
No header includes qemu-common.h after this commit, as prescribed by
qemu-common.h's file comment.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20190523143508.25387-5-armbru@redhat.com>
[Rebased with conflicts resolved automatically, except for
include/hw/arm/xlnx-zynqmp.h hw/arm/nrf51_soc.c hw/arm/msf2-soc.c
block/qcow2-refcount.c block/qcow2-cluster.c block/qcow2-cache.c
target/arm/cpu.h target/lm32/cpu.h target/m68k/cpu.h target/mips/cpu.h
target/moxie/cpu.h target/nios2/cpu.h target/openrisc/cpu.h
target/riscv/cpu.h target/tilegx/cpu.h target/tricore/cpu.h
target/unicore32/cpu.h target/xtensa/cpu.h; bsd-user/main.c and
net/tap-bsd.c fixed up]
bdrv_do_drained_begin/end() assume that they are called with the
AioContext lock of bs held. If we call drain functions from a coroutine
with the AioContext lock held, we yield and schedule a BH to move out of
coroutine context. This means that the lock for the home context of the
coroutine is released and must be re-acquired in the bottom half.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
qemu_aio_coroutine_enter() is (indirectly) called recursively when
processing co_queue_wakeup. This can lead to stack exhaustion.
This patch rewrites co_queue_wakeup in an iterative fashion (instead of
recursive) with bounded memory usage to prevent stack exhaustion.
qemu_co_queue_run_restart() is inlined into qemu_aio_coroutine_enter()
and the qemu_coroutine_enter() call is turned into a loop to avoid
recursion.
There is one change that is worth mentioning: Previously, when
coroutine A queued coroutine B, qemu_co_queue_run_restart() entered
coroutine B from coroutine A. If A was terminating then it would still
stay alive until B yielded. After this patch B is entered by A's parent
so that a A can be deleted immediately if it is terminating.
It is safe to make this change since B could never interact with A if it
was terminating anyway.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 20180322152834.12656-3-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The previous patch fixed a race condition, in which there were
coroutines being executing doubly, or after coroutine deletion.
We can detect common scenarios when this happens, and print an error
message and abort before we corrupt memory / data, or segfault.
This patch will abort if an attempt to enter a coroutine is made while
it is currently pending execution, either in a specific AioContext bh,
or pending execution via a timer. It will also abort if a coroutine
is scheduled, before a prior scheduled run has occurred.
We cannot rely on the existing co->caller check for recursive re-entry
to catch this, as the coroutine may run and exit with
COROUTINE_TERMINATE before the scheduled coroutine executes.
(This is the scenario that was occurring and fixed in the previous
patch).
This patch also re-orders the Coroutine struct elements in an attempt to
optimize caching.
Signed-off-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Submission of requests on linux aio is a bit tricky and can lead to
requests completions on submission path:
44713c9e85 ("linux-aio: Handle io_submit() failure gracefully")
0ed93d84ed ("linux-aio: process completions from ioq_submit()")
That means that any coroutine which has been yielded in order to wait
for completion can be resumed from submission path and be eventually
terminated (freed).
The following use-after-free crash was observed when IO throttling
was enabled:
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7f5813dff700 (LWP 56417)]
virtqueue_unmap_sg (elem=0x7f5804009a30, len=1, vq=<optimized out>) at virtio.c:252
(gdb) bt
#0 virtqueue_unmap_sg (elem=0x7f5804009a30, len=1, vq=<optimized out>) at virtio.c:252
^^^^^^^^^^^^^^
remember the address
#1 virtqueue_fill (vq=0x5598b20d21b0, elem=0x7f5804009a30, len=1, idx=0) at virtio.c:282
#2 virtqueue_push (vq=0x5598b20d21b0, elem=elem@entry=0x7f5804009a30, len=<optimized out>) at virtio.c:308
#3 virtio_blk_req_complete (req=req@entry=0x7f5804009a30, status=status@entry=0 '\000') at virtio-blk.c:61
#4 virtio_blk_rw_complete (opaque=<optimized out>, ret=0) at virtio-blk.c:126
#5 blk_aio_complete (acb=0x7f58040068d0) at block-backend.c:923
#6 coroutine_trampoline (i0=<optimized out>, i1=<optimized out>) at coroutine-ucontext.c:78
(gdb) p * elem
$8 = {index = 77, out_num = 2, in_num = 1,
in_addr = 0x7f5804009ad8, out_addr = 0x7f5804009ae0,
in_sg = 0x0, out_sg = 0x7f5804009a50}
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
'in_sg' and 'out_sg' are invalid.
e.g. it is impossible that 'in_sg' is zero,
instead its value must be equal to:
(gdb) p/x 0x7f5804009ad8 + sizeof(elem->in_addr[0]) + 2 * sizeof(elem->out_addr[0])
$26 = 0x7f5804009af0
Seems 'elem' was corrupted. Meanwhile another thread raised an abort:
Thread 12 (Thread 0x7f57f2ffd700 (LWP 56426)):
#0 raise () from /lib/x86_64-linux-gnu/libc.so.6
#1 abort () from /lib/x86_64-linux-gnu/libc.so.6
#2 qemu_coroutine_enter (co=0x7f5804009af0) at qemu-coroutine.c:113
#3 qemu_co_queue_run_restart (co=0x7f5804009a30) at qemu-coroutine-lock.c:60
#4 qemu_coroutine_enter (co=0x7f5804009a30) at qemu-coroutine.c:119
^^^^^^^^^^^^^^^^^^
WTF?? this is equal to elem from crashed thread
#5 qemu_co_queue_run_restart (co=0x7f57e7f16ae0) at qemu-coroutine-lock.c:60
#6 qemu_coroutine_enter (co=0x7f57e7f16ae0) at qemu-coroutine.c:119
#7 qemu_co_queue_run_restart (co=0x7f5807e112a0) at qemu-coroutine-lock.c:60
#8 qemu_coroutine_enter (co=0x7f5807e112a0) at qemu-coroutine.c:119
#9 qemu_co_queue_run_restart (co=0x7f5807f17820) at qemu-coroutine-lock.c:60
#10 qemu_coroutine_enter (co=0x7f5807f17820) at qemu-coroutine.c:119
#11 qemu_co_queue_run_restart (co=0x7f57e7f18e10) at qemu-coroutine-lock.c:60
#12 qemu_coroutine_enter (co=0x7f57e7f18e10) at qemu-coroutine.c:119
#13 qemu_co_enter_next (queue=queue@entry=0x5598b1e742d0) at qemu-coroutine-lock.c:106
#14 timer_cb (blk=0x5598b1e74280, is_write=<optimized out>) at throttle-groups.c:419
Crash can be explained by access of 'co' object from the loop inside
qemu_co_queue_run_restart():
while ((next = QSIMPLEQ_FIRST(&co->co_queue_wakeup))) {
QSIMPLEQ_REMOVE_HEAD(&co->co_queue_wakeup, co_queue_next);
^^^^^^^^^^^^^^^^^^^^
on each iteration 'co' is accessed,
but 'co' can be already freed
qemu_coroutine_enter(next);
}
When 'next' coroutine is resumed (entered) it can in its turn resume
'co', and eventually free it. That's why we see 'co' (which was freed)
has the same address as 'elem' from the first backtrace.
The fix is obvious: use temporary queue and do not touch coroutine after
first qemu_coroutine_enter() is invoked.
The issue is quite rare and happens every ~12 hours on very high IO
and CPU load (building linux kernel with -j512 inside guest) when IO
throttling is enabled. With the fix applied guest is running ~35 hours
and is still alive so far.
Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20170601160847.23720-1-roman.penyaev@profitbricks.com
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Fam Zheng <famz@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: qemu-devel@nongnu.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
It's a variant of qemu_coroutine_enter with an explicit AioContext
parameter.
Signed-off-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Running a very small critical section on pthread_mutex_t and CoMutex
shows that pthread_mutex_t is much faster because it doesn't actually
go to sleep. What happens is that the critical section is shorter
than the latency of entering the kernel and thus FUTEX_WAIT always
fails. With CoMutex there is no such latency but you still want to
avoid wait and wakeup. So introduce it artificially.
This only works with one waiters; because CoMutex is fair, it will
always have more waits and wakeups than a pthread_mutex_t.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 20170213181244.16297-3-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
aio_co_wake provides the infrastructure to start a coroutine on a "home"
AioContext. It will be used by CoMutex and CoQueue, so that coroutines
don't jump from one context to another when they go to sleep on a
mutex or waitqueue. However, it can also be used as a more efficient
alternative to one-shot bottom halves, and saves the effort of tracking
which AioContext a coroutine is running on.
aio_co_schedule is the part of aio_co_wake that starts a coroutine
on a remove AioContext, but it is also useful to implement e.g.
bdrv_set_aio_context callbacks.
The implementation of aio_co_schedule is based on a lock-free
multiple-producer, single-consumer queue. The multiple producers use
cmpxchg to add to a LIFO stack. The consumer (a per-AioContext bottom
half) grabs all items added so far, inverts the list to make it FIFO,
and goes through it one item at a time until it's empty. The data
structure was inspired by OSv, which uses it in the very code we'll
"port" to QEMU for the thread-safe CoMutex.
Most of the new code is really tests.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 20170213135235.12274-3-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
In the context of asynchronous work, if we have a worker coroutine that
didn't yield, the parent coroutine cannot be reentered because it hasn't
yielded yet. In this case we don't even have to reenter the parent
because it will see that the work is already done and won't even yield.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
See the doc comments for a description of this new coroutine API.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1474989516-18255-2-git-send-email-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
A coroutine that takes a lock must also release it again. If the
coroutine terminates without having released all its locks, it's buggy
and we'll probably run into a deadlock sooner or later. Make sure that
we don't get such cases.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
In practice the entry argument is always known at creation time, and
it is confusing that sometimes qemu_coroutine_enter is used with a
non-NULL argument to re-enter a coroutine (this happens in
block/sheepdog.c and tests/test-coroutine.c). So pass the opaque value
at creation time, for consistency with e.g. aio_bh_new.
Mostly done with the following semantic patch:
@ entry1 @
expression entry, arg, co;
@@
- co = qemu_coroutine_create(entry);
+ co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry2 @
expression entry, arg;
identifier co;
@@
- Coroutine *co = qemu_coroutine_create(entry);
+ Coroutine *co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry3 @
expression entry, arg;
@@
- qemu_coroutine_enter(qemu_coroutine_create(entry), arg);
+ qemu_coroutine_enter(qemu_coroutine_create(entry, arg));
@ reentry @
expression co;
@@
- qemu_coroutine_enter(co, NULL);
+ qemu_coroutine_enter(co);
except for the aforementioned few places where the semantic patch
stumbled (as expected) and for test_co_queue, which would otherwise
produce an uninitialized variable warning.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
CoQueue do not need to remove any element but the head of the list;
processing is always strictly FIFO. Therefore, the simpler singly-linked
QSIMPLEQ can be used instead.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.
This commit was created with scripts/clean-includes.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1454089805-5470-6-git-send-email-peter.maydell@linaro.org
The coroutine files are currently referenced by the block-obj-y
variable. The coroutine functionality though is already used by
more than just the block code. eg migration code uses coroutine
yield. In the future the I/O channel code will also use the
coroutine yield functionality. Since the coroutine code is nicely
self-contained it can be easily built as part of the libqemuutil.a
library, making it widely available.
The headers are also moved into include/qemu, instead of the
include/block directory, since they are now part of the util
codebase, and the impl was never in the block/ directory
either.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Kevin Wolf
Stefan Hajnoczi
Hiroki Narukawa
Markus Armbruster
Jeff Cody
Roman Pen
Fam Zheng
Paolo Bonzini
Peter Maydell
Daniel P. Berrange