qemu/block/linux-aio.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

508 lines
14 KiB
C
Raw Normal View History

/*
* Linux native AIO support.
*
* Copyright (C) 2009 IBM, Corp.
* Copyright (C) 2009 Red Hat, Inc.
*
* This work is licensed under the terms of the GNU GPL, version 2 or later.
* See the COPYING file in the top-level directory.
*/
#include "qemu/osdep.h"
#include "block/aio.h"
#include "qemu/queue.h"
#include "block/block.h"
#include "block/raw-aio.h"
#include "qemu/event_notifier.h"
#include "qemu/coroutine.h"
#include "qemu/defer-call.h"
#include "qapi/error.h"
#include "sysemu/block-backend.h"
/* Only used for assertions. */
#include "qemu/coroutine_int.h"
#include <libaio.h>
/*
* Queue size (per-device).
*
* XXX: eventually we need to communicate this to the guest and/or make it
* tunable by the guest. If we get more outstanding requests at a time
* than this we will get EAGAIN from io_submit which is communicated to
* the guest as an I/O error.
*/
#define MAX_EVENTS 1024
/* Maximum number of requests in a batch. (default value) */
#define DEFAULT_MAX_BATCH 32
struct qemu_laiocb {
Coroutine *co;
LinuxAioState *ctx;
struct iocb iocb;
ssize_t ret;
size_t nbytes;
QEMUIOVector *qiov;
bool is_read;
QSIMPLEQ_ENTRY(qemu_laiocb) next;
};
typedef struct {
unsigned int in_queue;
unsigned int in_flight;
bool blocked;
QSIMPLEQ_HEAD(, qemu_laiocb) pending;
} LaioQueue;
struct LinuxAioState {
AioContext *aio_context;
io_context_t ctx;
EventNotifier e;
/* No locking required, only accessed from AioContext home thread */
LaioQueue io_q;
QEMUBH *completion_bh;
int event_idx;
int event_max;
};
static void ioq_submit(LinuxAioState *s);
static inline ssize_t io_event_ret(struct io_event *ev)
{
return (ssize_t)(((uint64_t)ev->res2 << 32) | ev->res);
}
/*
* Completes an AIO request.
*/
static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
{
int ret;
ret = laiocb->ret;
if (ret != -ECANCELED) {
if (ret == laiocb->nbytes) {
ret = 0;
} else if (ret >= 0) {
/* Short reads mean EOF, pad with zeros. */
if (laiocb->is_read) {
qemu_iovec_memset(laiocb->qiov, ret, 0,
laiocb->qiov->size - ret);
} else {
ret = -ENOSPC;
}
}
}
laiocb->ret = ret;
/*
* If the coroutine is already entered it must be in ioq_submit() and
* will notice laio->ret has been filled in when it eventually runs
* later. Coroutines cannot be entered recursively so avoid doing
* that!
*/
assert(laiocb->co->ctx == laiocb->ctx->aio_context);
if (!qemu_coroutine_entered(laiocb->co)) {
aio_co_wake(laiocb->co);
}
}
/**
* aio_ring buffer which is shared between userspace and kernel.
*
* This copied from linux/fs/aio.c, common header does not exist
* but AIO exists for ages so we assume ABI is stable.
*/
struct aio_ring {
unsigned id; /* kernel internal index number */
unsigned nr; /* number of io_events */
unsigned head; /* Written to by userland or by kernel. */
unsigned tail;
unsigned magic;
unsigned compat_features;
unsigned incompat_features;
unsigned header_length; /* size of aio_ring */
misc: Replace zero-length arrays with flexible array member (automatic) Description copied from Linux kernel commit from Gustavo A. R. Silva (see [3]): --v-- description start --v-- The current codebase makes use of the zero-length array language extension to the C90 standard, but the preferred mechanism to declare variable-length types such as these ones is a flexible array member [1], introduced in C99: struct foo { int stuff; struct boo array[]; }; By making use of the mechanism above, we will get a compiler warning in case the flexible array does not occur last in the structure, which will help us prevent some kind of undefined behavior bugs from being unadvertenly introduced [2] to the Linux codebase from now on. --^-- description end --^-- Do the similar housekeeping in the QEMU codebase (which uses C99 since commit 7be41675f7cb). All these instances of code were found with the help of the following Coccinelle script: @@ identifier s, m, a; type t, T; @@ struct s { ... t m; - T a[0]; + T a[]; }; @@ identifier s, m, a; type t, T; @@ struct s { ... t m; - T a[0]; + T a[]; } QEMU_PACKED; [1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=76497732932f [3] https://git.kernel.org/pub/scm/linux/kernel/git/gustavoars/linux.git/commit/?id=17642a2fbd2c1 Inspired-by: Gustavo A. R. Silva <gustavo@embeddedor.com> Reviewed-by: David Hildenbrand <david@redhat.com> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2020-03-04 18:38:15 +03:00
struct io_event io_events[];
};
/**
* io_getevents_peek:
* @ctx: AIO context
* @events: pointer on events array, output value
* Returns the number of completed events and sets a pointer
* on events array. This function does not update the internal
* ring buffer, only reads head and tail. When @events has been
* processed io_getevents_commit() must be called.
*/
static inline unsigned int io_getevents_peek(io_context_t ctx,
struct io_event **events)
{
struct aio_ring *ring = (struct aio_ring *)ctx;
unsigned int head = ring->head, tail = ring->tail;
unsigned int nr;
nr = tail >= head ? tail - head : ring->nr - head;
*events = ring->io_events + head;
/* To avoid speculative loads of s->events[i] before observing tail.
Paired with smp_wmb() inside linux/fs/aio.c: aio_complete(). */
smp_rmb();
return nr;
}
/**
* io_getevents_commit:
* @ctx: AIO context
* @nr: the number of events on which head should be advanced
*
* Advances head of a ring buffer.
*/
static inline void io_getevents_commit(io_context_t ctx, unsigned int nr)
{
struct aio_ring *ring = (struct aio_ring *)ctx;
if (nr) {
ring->head = (ring->head + nr) % ring->nr;
}
}
/**
* io_getevents_advance_and_peek:
* @ctx: AIO context
* @events: pointer on events array, output value
* @nr: the number of events on which head should be advanced
*
* Advances head of a ring buffer and returns number of elements left.
*/
static inline unsigned int
io_getevents_advance_and_peek(io_context_t ctx,
struct io_event **events,
unsigned int nr)
{
io_getevents_commit(ctx, nr);
return io_getevents_peek(ctx, events);
}
/**
* qemu_laio_process_completions:
* @s: AIO state
*
* Fetches completed I/O requests and invokes their callbacks.
*
* The function is somewhat tricky because it supports nested event loops, for
* example when a request callback invokes aio_poll(). In order to do this,
* indices are kept in LinuxAioState. Function schedules BH completion so it
* can be called again in a nested event loop. When there are no events left
* to complete the BH is being canceled.
*/
static void qemu_laio_process_completions(LinuxAioState *s)
{
struct io_event *events;
defer_call_begin();
/* Reschedule so nested event loops see currently pending completions */
qemu_bh_schedule(s->completion_bh);
while ((s->event_max = io_getevents_advance_and_peek(s->ctx, &events,
s->event_idx))) {
for (s->event_idx = 0; s->event_idx < s->event_max; ) {
struct iocb *iocb = events[s->event_idx].obj;
struct qemu_laiocb *laiocb =
container_of(iocb, struct qemu_laiocb, iocb);
laiocb->ret = io_event_ret(&events[s->event_idx]);
/* Change counters one-by-one because we can be nested. */
s->io_q.in_flight--;
s->event_idx++;
qemu_laio_process_completion(laiocb);
}
}
qemu_bh_cancel(s->completion_bh);
/* If we are nested we have to notify the level above that we are done
* by setting event_max to zero, upper level will then jump out of it's
* own `for` loop. If we are the last all counters dropped to zero. */
s->event_max = 0;
s->event_idx = 0;
defer_call_end();
}
static void qemu_laio_process_completions_and_submit(LinuxAioState *s)
{
qemu_laio_process_completions(s);
if (!QSIMPLEQ_EMPTY(&s->io_q.pending)) {
ioq_submit(s);
}
}
static void qemu_laio_completion_bh(void *opaque)
{
LinuxAioState *s = opaque;
qemu_laio_process_completions_and_submit(s);
}
static void qemu_laio_completion_cb(EventNotifier *e)
{
LinuxAioState *s = container_of(e, LinuxAioState, e);
if (event_notifier_test_and_clear(&s->e)) {
qemu_laio_process_completions_and_submit(s);
}
}
static bool qemu_laio_poll_cb(void *opaque)
{
EventNotifier *e = opaque;
LinuxAioState *s = container_of(e, LinuxAioState, e);
struct io_event *events;
aio-posix: split poll check from ready handler Adaptive polling measures the execution time of the polling check plus handlers called when a polled event becomes ready. Handlers can take a significant amount of time, making it look like polling was running for a long time when in fact the event handler was running for a long time. For example, on Linux the io_submit(2) syscall invoked when a virtio-blk device's virtqueue becomes ready can take 10s of microseconds. This can exceed the default polling interval (32 microseconds) and cause adaptive polling to stop polling. By excluding the handler's execution time from the polling check we make the adaptive polling calculation more accurate. As a result, the event loop now stays in polling mode where previously it would have fallen back to file descriptor monitoring. The following data was collected with virtio-blk num-queues=2 event_idx=off using an IOThread. Before: 168k IOPS, IOThread syscalls: 9837.115 ( 0.020 ms): IO iothread1/620155 io_submit(ctx_id: 140512552468480, nr: 16, iocbpp: 0x7fcb9f937db0) = 16 9837.158 ( 0.002 ms): IO iothread1/620155 write(fd: 103, buf: 0x556a2ef71b88, count: 8) = 8 9837.161 ( 0.001 ms): IO iothread1/620155 write(fd: 104, buf: 0x556a2ef71b88, count: 8) = 8 9837.163 ( 0.001 ms): IO iothread1/620155 ppoll(ufds: 0x7fcb90002800, nfds: 4, tsp: 0x7fcb9f1342d0, sigsetsize: 8) = 3 9837.164 ( 0.001 ms): IO iothread1/620155 read(fd: 107, buf: 0x7fcb9f939cc0, count: 512) = 8 9837.174 ( 0.001 ms): IO iothread1/620155 read(fd: 105, buf: 0x7fcb9f939cc0, count: 512) = 8 9837.176 ( 0.001 ms): IO iothread1/620155 read(fd: 106, buf: 0x7fcb9f939cc0, count: 512) = 8 9837.209 ( 0.035 ms): IO iothread1/620155 io_submit(ctx_id: 140512552468480, nr: 32, iocbpp: 0x7fca7d0cebe0) = 32 174k IOPS (+3.6%), IOThread syscalls: 9809.566 ( 0.036 ms): IO iothread1/623061 io_submit(ctx_id: 140539805028352, nr: 32, iocbpp: 0x7fd0cdd62be0) = 32 9809.625 ( 0.001 ms): IO iothread1/623061 write(fd: 103, buf: 0x5647cfba5f58, count: 8) = 8 9809.627 ( 0.002 ms): IO iothread1/623061 write(fd: 104, buf: 0x5647cfba5f58, count: 8) = 8 9809.663 ( 0.036 ms): IO iothread1/623061 io_submit(ctx_id: 140539805028352, nr: 32, iocbpp: 0x7fd0d0388b50) = 32 Notice that ppoll(2) and eventfd read(2) syscalls are eliminated because the IOThread stays in polling mode instead of falling back to file descriptor monitoring. As usual, polling is not implemented on Windows so this patch ignores the new io_poll_read() callback in aio-win32.c. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Stefano Garzarella <sgarzare@redhat.com> Message-id: 20211207132336.36627-2-stefanha@redhat.com [Fixed up aio_set_event_notifier() calls in tests/unit/test-fdmon-epoll.c added after this series was queued. --Stefan] Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2021-12-07 16:23:31 +03:00
return io_getevents_peek(s->ctx, &events);
}
static void qemu_laio_poll_ready(EventNotifier *opaque)
{
EventNotifier *e = opaque;
LinuxAioState *s = container_of(e, LinuxAioState, e);
qemu_laio_process_completions_and_submit(s);
}
static void ioq_init(LaioQueue *io_q)
{
QSIMPLEQ_INIT(&io_q->pending);
io_q->in_queue = 0;
io_q->in_flight = 0;
io_q->blocked = false;
}
static void ioq_submit(LinuxAioState *s)
{
int ret, len;
struct qemu_laiocb *aiocb;
struct iocb *iocbs[MAX_EVENTS];
QSIMPLEQ_HEAD(, qemu_laiocb) completed;
do {
if (s->io_q.in_flight >= MAX_EVENTS) {
break;
}
len = 0;
QSIMPLEQ_FOREACH(aiocb, &s->io_q.pending, next) {
iocbs[len++] = &aiocb->iocb;
if (s->io_q.in_flight + len >= MAX_EVENTS) {
break;
}
}
ret = io_submit(s->ctx, len, iocbs);
if (ret == -EAGAIN) {
break;
}
if (ret < 0) {
/* Fail the first request, retry the rest */
aiocb = QSIMPLEQ_FIRST(&s->io_q.pending);
QSIMPLEQ_REMOVE_HEAD(&s->io_q.pending, next);
s->io_q.in_queue--;
aiocb->ret = ret;
qemu_laio_process_completion(aiocb);
continue;
}
s->io_q.in_flight += ret;
s->io_q.in_queue -= ret;
aiocb = container_of(iocbs[ret - 1], struct qemu_laiocb, iocb);
QSIMPLEQ_SPLIT_AFTER(&s->io_q.pending, aiocb, next, &completed);
} while (ret == len && !QSIMPLEQ_EMPTY(&s->io_q.pending));
s->io_q.blocked = (s->io_q.in_queue > 0);
linux-aio: process completions from ioq_submit() In order to reduce completion latency it makes sense to harvest completed requests ASAP. Very fast backend device can complete requests just after submission, so it is worth trying to check ring buffer in order to peek completed requests directly after io_submit() has been called. Indeed, this patch reduces the completions latencies and increases the overall throughput, e.g. the following is the percentiles of number of completed requests at once: 1th 10th 20th 30th 40th 50th 60th 70th 80th 90th 99.99th Before 2 4 42 112 128 128 128 128 128 128 128 After 1 1 4 14 33 45 47 48 50 51 108 That means, that before the current patch is applied the ring buffer is observed as full (128 requests were consumed at once) in 60% of calls. After patch is applied the distribution of number of completed requests is "smoother" and the queue (requests in-flight) is almost never full. The fio read results are the following (write results are almost the same and are not showed here): Before ------ job: (groupid=0, jobs=8): err= 0: pid=2227: Tue Jul 19 11:29:50 2016 Description : [Emulation of Storage Server Access Pattern] read : io=54681MB, bw=1822.7MB/s, iops=179779, runt= 30001msec slat (usec): min=172, max=16883, avg=338.35, stdev=109.66 clat (usec): min=1, max=21977, avg=1051.45, stdev=299.29 lat (usec): min=317, max=22521, avg=1389.83, stdev=300.73 clat percentiles (usec): | 1.00th=[ 346], 5.00th=[ 596], 10.00th=[ 708], 20.00th=[ 852], | 30.00th=[ 932], 40.00th=[ 996], 50.00th=[ 1048], 60.00th=[ 1112], | 70.00th=[ 1176], 80.00th=[ 1256], 90.00th=[ 1384], 95.00th=[ 1496], | 99.00th=[ 1800], 99.50th=[ 1928], 99.90th=[ 2320], 99.95th=[ 2672], | 99.99th=[ 4704] bw (KB /s): min=205229, max=553181, per=12.50%, avg=233278.26, stdev=18383.51 After ------ job: (groupid=0, jobs=8): err= 0: pid=2220: Tue Jul 19 11:31:51 2016 Description : [Emulation of Storage Server Access Pattern] read : io=57637MB, bw=1921.2MB/s, iops=189529, runt= 30002msec slat (usec): min=169, max=20636, avg=329.61, stdev=124.18 clat (usec): min=2, max=19592, avg=988.78, stdev=251.04 lat (usec): min=381, max=21067, avg=1318.42, stdev=243.58 clat percentiles (usec): | 1.00th=[ 310], 5.00th=[ 580], 10.00th=[ 748], 20.00th=[ 876], | 30.00th=[ 908], 40.00th=[ 948], 50.00th=[ 1012], 60.00th=[ 1064], | 70.00th=[ 1080], 80.00th=[ 1128], 90.00th=[ 1224], 95.00th=[ 1288], | 99.00th=[ 1496], 99.50th=[ 1608], 99.90th=[ 1960], 99.95th=[ 2256], | 99.99th=[ 5408] bw (KB /s): min=212149, max=390160, per=12.49%, avg=245746.04, stdev=11606.75 Throughput increased from 1822MB/s to 1921MB/s, average completion latencies decreased from 1051us to 988us. Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com> Message-id: 1468931263-32667-4-git-send-email-roman.penyaev@profitbricks.com Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Paolo Bonzini <pbonzini@redhat.com> Cc: qemu-devel@nongnu.org Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-19 15:27:43 +03:00
if (s->io_q.in_flight) {
/* We can try to complete something just right away if there are
* still requests in-flight. */
qemu_laio_process_completions(s);
/*
* Even we have completed everything (in_flight == 0), the queue can
* have still pended requests (in_queue > 0). We do not attempt to
* repeat submission to avoid IO hang. The reason is simple: s->e is
* still set and completion callback will be called shortly and all
* pended requests will be submitted from there.
*/
}
}
static uint64_t laio_max_batch(LinuxAioState *s, uint64_t dev_max_batch)
{
uint64_t max_batch = s->aio_context->aio_max_batch ?: DEFAULT_MAX_BATCH;
/*
* AIO context can be shared between multiple block devices, so
* `dev_max_batch` allows reducing the batch size for latency-sensitive
* devices.
*/
max_batch = MIN_NON_ZERO(dev_max_batch, max_batch);
/* limit the batch with the number of available events */
max_batch = MIN_NON_ZERO(MAX_EVENTS - s->io_q.in_flight, max_batch);
return max_batch;
}
static void laio_deferred_fn(void *opaque)
{
LinuxAioState *s = opaque;
if (!s->io_q.blocked && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
ioq_submit(s);
}
}
static int laio_do_submit(int fd, struct qemu_laiocb *laiocb, off_t offset,
int type, uint64_t dev_max_batch)
{
LinuxAioState *s = laiocb->ctx;
struct iocb *iocbs = &laiocb->iocb;
QEMUIOVector *qiov = laiocb->qiov;
switch (type) {
case QEMU_AIO_WRITE:
io_prep_pwritev(iocbs, fd, qiov->iov, qiov->niov, offset);
avoid TABs in files that only contain a few Most files that have TABs only contain a handful of them. Change them to spaces so that we don't confuse people. disas, standard-headers, linux-headers and libdecnumber are imported from other projects and probably should be exempted from the check. Outside those, after this patch the following files still contain both 8-space and TAB sequences at the beginning of the line. Many of them have a majority of TABs, or were initially committed with all tabs. bsd-user/i386/target_syscall.h bsd-user/x86_64/target_syscall.h crypto/aes.c hw/audio/fmopl.c hw/audio/fmopl.h hw/block/tc58128.c hw/display/cirrus_vga.c hw/display/xenfb.c hw/dma/etraxfs_dma.c hw/intc/sh_intc.c hw/misc/mst_fpga.c hw/net/pcnet.c hw/sh4/sh7750.c hw/timer/m48t59.c hw/timer/sh_timer.c include/crypto/aes.h include/disas/bfd.h include/hw/sh4/sh.h libdecnumber/decNumber.c linux-headers/asm-generic/unistd.h linux-headers/linux/kvm.h linux-user/alpha/target_syscall.h linux-user/arm/nwfpe/double_cpdo.c linux-user/arm/nwfpe/fpa11_cpdt.c linux-user/arm/nwfpe/fpa11_cprt.c linux-user/arm/nwfpe/fpa11.h linux-user/flat.h linux-user/flatload.c linux-user/i386/target_syscall.h linux-user/ppc/target_syscall.h linux-user/sparc/target_syscall.h linux-user/syscall.c linux-user/syscall_defs.h linux-user/x86_64/target_syscall.h slirp/cksum.c slirp/if.c slirp/ip.h slirp/ip_icmp.c slirp/ip_icmp.h slirp/ip_input.c slirp/ip_output.c slirp/mbuf.c slirp/misc.c slirp/sbuf.c slirp/socket.c slirp/socket.h slirp/tcp_input.c slirp/tcpip.h slirp/tcp_output.c slirp/tcp_subr.c slirp/tcp_timer.c slirp/tftp.c slirp/udp.c slirp/udp.h target/cris/cpu.h target/cris/mmu.c target/cris/op_helper.c target/sh4/helper.c target/sh4/op_helper.c target/sh4/translate.c tcg/sparc/tcg-target.inc.c tests/tcg/cris/check_addo.c tests/tcg/cris/check_moveq.c tests/tcg/cris/check_swap.c tests/tcg/multiarch/test-mmap.c ui/vnc-enc-hextile-template.h ui/vnc-enc-zywrle.h util/envlist.c util/readline.c The following have only TABs: bsd-user/i386/target_signal.h bsd-user/sparc64/target_signal.h bsd-user/sparc64/target_syscall.h bsd-user/sparc/target_signal.h bsd-user/sparc/target_syscall.h bsd-user/x86_64/target_signal.h crypto/desrfb.c hw/audio/intel-hda-defs.h hw/core/uboot_image.h hw/sh4/sh7750_regnames.c hw/sh4/sh7750_regs.h include/hw/cris/etraxfs_dma.h linux-user/alpha/termbits.h linux-user/arm/nwfpe/fpopcode.h linux-user/arm/nwfpe/fpsr.h linux-user/arm/syscall_nr.h linux-user/arm/target_signal.h linux-user/cris/target_signal.h linux-user/i386/target_signal.h linux-user/linux_loop.h linux-user/m68k/target_signal.h linux-user/microblaze/target_signal.h linux-user/mips64/target_signal.h linux-user/mips/target_signal.h linux-user/mips/target_syscall.h linux-user/mips/termbits.h linux-user/ppc/target_signal.h linux-user/sh4/target_signal.h linux-user/sh4/termbits.h linux-user/sparc64/target_syscall.h linux-user/sparc/target_signal.h linux-user/x86_64/target_signal.h linux-user/x86_64/termbits.h pc-bios/optionrom/optionrom.h slirp/mbuf.h slirp/misc.h slirp/sbuf.h slirp/tcp.h slirp/tcp_timer.h slirp/tcp_var.h target/i386/svm.h target/sparc/asi.h target/xtensa/core-dc232b/xtensa-modules.inc.c target/xtensa/core-dc233c/xtensa-modules.inc.c target/xtensa/core-de212/core-isa.h target/xtensa/core-de212/xtensa-modules.inc.c target/xtensa/core-fsf/xtensa-modules.inc.c target/xtensa/core-sample_controller/core-isa.h target/xtensa/core-sample_controller/xtensa-modules.inc.c target/xtensa/core-test_kc705_be/core-isa.h target/xtensa/core-test_kc705_be/xtensa-modules.inc.c tests/tcg/cris/check_abs.c tests/tcg/cris/check_addc.c tests/tcg/cris/check_addcm.c tests/tcg/cris/check_addoq.c tests/tcg/cris/check_bound.c tests/tcg/cris/check_ftag.c tests/tcg/cris/check_int64.c tests/tcg/cris/check_lz.c tests/tcg/cris/check_openpf5.c tests/tcg/cris/check_sigalrm.c tests/tcg/cris/crisutils.h tests/tcg/cris/sys.c tests/tcg/i386/test-i386-ssse3.c ui/vgafont.h Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Message-Id: <20181213223737.11793-3-pbonzini@redhat.com> Reviewed-by: Aleksandar Markovic <amarkovic@wavecomp.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Wainer dos Santos Moschetta <wainersm@redhat.com> Acked-by: Richard Henderson <richard.henderson@linaro.org> Acked-by: Eric Blake <eblake@redhat.com> Acked-by: David Gibson <david@gibson.dropbear.id.au> Reviewed-by: Stefan Markovic <smarkovic@wavecomp.com> Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2018-12-14 01:37:37 +03:00
break;
case QEMU_AIO_ZONE_APPEND:
io_prep_pwritev(iocbs, fd, qiov->iov, qiov->niov, offset);
break;
case QEMU_AIO_READ:
io_prep_preadv(iocbs, fd, qiov->iov, qiov->niov, offset);
avoid TABs in files that only contain a few Most files that have TABs only contain a handful of them. Change them to spaces so that we don't confuse people. disas, standard-headers, linux-headers and libdecnumber are imported from other projects and probably should be exempted from the check. Outside those, after this patch the following files still contain both 8-space and TAB sequences at the beginning of the line. Many of them have a majority of TABs, or were initially committed with all tabs. bsd-user/i386/target_syscall.h bsd-user/x86_64/target_syscall.h crypto/aes.c hw/audio/fmopl.c hw/audio/fmopl.h hw/block/tc58128.c hw/display/cirrus_vga.c hw/display/xenfb.c hw/dma/etraxfs_dma.c hw/intc/sh_intc.c hw/misc/mst_fpga.c hw/net/pcnet.c hw/sh4/sh7750.c hw/timer/m48t59.c hw/timer/sh_timer.c include/crypto/aes.h include/disas/bfd.h include/hw/sh4/sh.h libdecnumber/decNumber.c linux-headers/asm-generic/unistd.h linux-headers/linux/kvm.h linux-user/alpha/target_syscall.h linux-user/arm/nwfpe/double_cpdo.c linux-user/arm/nwfpe/fpa11_cpdt.c linux-user/arm/nwfpe/fpa11_cprt.c linux-user/arm/nwfpe/fpa11.h linux-user/flat.h linux-user/flatload.c linux-user/i386/target_syscall.h linux-user/ppc/target_syscall.h linux-user/sparc/target_syscall.h linux-user/syscall.c linux-user/syscall_defs.h linux-user/x86_64/target_syscall.h slirp/cksum.c slirp/if.c slirp/ip.h slirp/ip_icmp.c slirp/ip_icmp.h slirp/ip_input.c slirp/ip_output.c slirp/mbuf.c slirp/misc.c slirp/sbuf.c slirp/socket.c slirp/socket.h slirp/tcp_input.c slirp/tcpip.h slirp/tcp_output.c slirp/tcp_subr.c slirp/tcp_timer.c slirp/tftp.c slirp/udp.c slirp/udp.h target/cris/cpu.h target/cris/mmu.c target/cris/op_helper.c target/sh4/helper.c target/sh4/op_helper.c target/sh4/translate.c tcg/sparc/tcg-target.inc.c tests/tcg/cris/check_addo.c tests/tcg/cris/check_moveq.c tests/tcg/cris/check_swap.c tests/tcg/multiarch/test-mmap.c ui/vnc-enc-hextile-template.h ui/vnc-enc-zywrle.h util/envlist.c util/readline.c The following have only TABs: bsd-user/i386/target_signal.h bsd-user/sparc64/target_signal.h bsd-user/sparc64/target_syscall.h bsd-user/sparc/target_signal.h bsd-user/sparc/target_syscall.h bsd-user/x86_64/target_signal.h crypto/desrfb.c hw/audio/intel-hda-defs.h hw/core/uboot_image.h hw/sh4/sh7750_regnames.c hw/sh4/sh7750_regs.h include/hw/cris/etraxfs_dma.h linux-user/alpha/termbits.h linux-user/arm/nwfpe/fpopcode.h linux-user/arm/nwfpe/fpsr.h linux-user/arm/syscall_nr.h linux-user/arm/target_signal.h linux-user/cris/target_signal.h linux-user/i386/target_signal.h linux-user/linux_loop.h linux-user/m68k/target_signal.h linux-user/microblaze/target_signal.h linux-user/mips64/target_signal.h linux-user/mips/target_signal.h linux-user/mips/target_syscall.h linux-user/mips/termbits.h linux-user/ppc/target_signal.h linux-user/sh4/target_signal.h linux-user/sh4/termbits.h linux-user/sparc64/target_syscall.h linux-user/sparc/target_signal.h linux-user/x86_64/target_signal.h linux-user/x86_64/termbits.h pc-bios/optionrom/optionrom.h slirp/mbuf.h slirp/misc.h slirp/sbuf.h slirp/tcp.h slirp/tcp_timer.h slirp/tcp_var.h target/i386/svm.h target/sparc/asi.h target/xtensa/core-dc232b/xtensa-modules.inc.c target/xtensa/core-dc233c/xtensa-modules.inc.c target/xtensa/core-de212/core-isa.h target/xtensa/core-de212/xtensa-modules.inc.c target/xtensa/core-fsf/xtensa-modules.inc.c target/xtensa/core-sample_controller/core-isa.h target/xtensa/core-sample_controller/xtensa-modules.inc.c target/xtensa/core-test_kc705_be/core-isa.h target/xtensa/core-test_kc705_be/xtensa-modules.inc.c tests/tcg/cris/check_abs.c tests/tcg/cris/check_addc.c tests/tcg/cris/check_addcm.c tests/tcg/cris/check_addoq.c tests/tcg/cris/check_bound.c tests/tcg/cris/check_ftag.c tests/tcg/cris/check_int64.c tests/tcg/cris/check_lz.c tests/tcg/cris/check_openpf5.c tests/tcg/cris/check_sigalrm.c tests/tcg/cris/crisutils.h tests/tcg/cris/sys.c tests/tcg/i386/test-i386-ssse3.c ui/vgafont.h Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Message-Id: <20181213223737.11793-3-pbonzini@redhat.com> Reviewed-by: Aleksandar Markovic <amarkovic@wavecomp.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Wainer dos Santos Moschetta <wainersm@redhat.com> Acked-by: Richard Henderson <richard.henderson@linaro.org> Acked-by: Eric Blake <eblake@redhat.com> Acked-by: David Gibson <david@gibson.dropbear.id.au> Reviewed-by: Stefan Markovic <smarkovic@wavecomp.com> Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2018-12-14 01:37:37 +03:00
break;
case QEMU_AIO_FLUSH:
io_prep_fdsync(iocbs, fd);
break;
/* Currently Linux kernel does not support other operations */
default:
fprintf(stderr, "%s: invalid AIO request type 0x%x.\n",
__func__, type);
return -EIO;
}
io_set_eventfd(&laiocb->iocb, event_notifier_get_fd(&s->e));
QSIMPLEQ_INSERT_TAIL(&s->io_q.pending, laiocb, next);
s->io_q.in_queue++;
if (!s->io_q.blocked) {
if (s->io_q.in_queue >= laio_max_batch(s, dev_max_batch)) {
ioq_submit(s);
} else {
defer_call(laio_deferred_fn, s);
}
}
return 0;
}
int coroutine_fn laio_co_submit(int fd, uint64_t offset, QEMUIOVector *qiov,
int type, uint64_t dev_max_batch)
{
int ret;
AioContext *ctx = qemu_get_current_aio_context();
struct qemu_laiocb laiocb = {
.co = qemu_coroutine_self(),
.nbytes = qiov ? qiov->size : 0,
.ctx = aio_get_linux_aio(ctx),
linux-aio: process completions from ioq_submit() In order to reduce completion latency it makes sense to harvest completed requests ASAP. Very fast backend device can complete requests just after submission, so it is worth trying to check ring buffer in order to peek completed requests directly after io_submit() has been called. Indeed, this patch reduces the completions latencies and increases the overall throughput, e.g. the following is the percentiles of number of completed requests at once: 1th 10th 20th 30th 40th 50th 60th 70th 80th 90th 99.99th Before 2 4 42 112 128 128 128 128 128 128 128 After 1 1 4 14 33 45 47 48 50 51 108 That means, that before the current patch is applied the ring buffer is observed as full (128 requests were consumed at once) in 60% of calls. After patch is applied the distribution of number of completed requests is "smoother" and the queue (requests in-flight) is almost never full. The fio read results are the following (write results are almost the same and are not showed here): Before ------ job: (groupid=0, jobs=8): err= 0: pid=2227: Tue Jul 19 11:29:50 2016 Description : [Emulation of Storage Server Access Pattern] read : io=54681MB, bw=1822.7MB/s, iops=179779, runt= 30001msec slat (usec): min=172, max=16883, avg=338.35, stdev=109.66 clat (usec): min=1, max=21977, avg=1051.45, stdev=299.29 lat (usec): min=317, max=22521, avg=1389.83, stdev=300.73 clat percentiles (usec): | 1.00th=[ 346], 5.00th=[ 596], 10.00th=[ 708], 20.00th=[ 852], | 30.00th=[ 932], 40.00th=[ 996], 50.00th=[ 1048], 60.00th=[ 1112], | 70.00th=[ 1176], 80.00th=[ 1256], 90.00th=[ 1384], 95.00th=[ 1496], | 99.00th=[ 1800], 99.50th=[ 1928], 99.90th=[ 2320], 99.95th=[ 2672], | 99.99th=[ 4704] bw (KB /s): min=205229, max=553181, per=12.50%, avg=233278.26, stdev=18383.51 After ------ job: (groupid=0, jobs=8): err= 0: pid=2220: Tue Jul 19 11:31:51 2016 Description : [Emulation of Storage Server Access Pattern] read : io=57637MB, bw=1921.2MB/s, iops=189529, runt= 30002msec slat (usec): min=169, max=20636, avg=329.61, stdev=124.18 clat (usec): min=2, max=19592, avg=988.78, stdev=251.04 lat (usec): min=381, max=21067, avg=1318.42, stdev=243.58 clat percentiles (usec): | 1.00th=[ 310], 5.00th=[ 580], 10.00th=[ 748], 20.00th=[ 876], | 30.00th=[ 908], 40.00th=[ 948], 50.00th=[ 1012], 60.00th=[ 1064], | 70.00th=[ 1080], 80.00th=[ 1128], 90.00th=[ 1224], 95.00th=[ 1288], | 99.00th=[ 1496], 99.50th=[ 1608], 99.90th=[ 1960], 99.95th=[ 2256], | 99.99th=[ 5408] bw (KB /s): min=212149, max=390160, per=12.49%, avg=245746.04, stdev=11606.75 Throughput increased from 1822MB/s to 1921MB/s, average completion latencies decreased from 1051us to 988us. Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com> Message-id: 1468931263-32667-4-git-send-email-roman.penyaev@profitbricks.com Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Paolo Bonzini <pbonzini@redhat.com> Cc: qemu-devel@nongnu.org Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-19 15:27:43 +03:00
.ret = -EINPROGRESS,
.is_read = (type == QEMU_AIO_READ),
.qiov = qiov,
};
ret = laio_do_submit(fd, &laiocb, offset, type, dev_max_batch);
if (ret < 0) {
return ret;
}
linux-aio: process completions from ioq_submit() In order to reduce completion latency it makes sense to harvest completed requests ASAP. Very fast backend device can complete requests just after submission, so it is worth trying to check ring buffer in order to peek completed requests directly after io_submit() has been called. Indeed, this patch reduces the completions latencies and increases the overall throughput, e.g. the following is the percentiles of number of completed requests at once: 1th 10th 20th 30th 40th 50th 60th 70th 80th 90th 99.99th Before 2 4 42 112 128 128 128 128 128 128 128 After 1 1 4 14 33 45 47 48 50 51 108 That means, that before the current patch is applied the ring buffer is observed as full (128 requests were consumed at once) in 60% of calls. After patch is applied the distribution of number of completed requests is "smoother" and the queue (requests in-flight) is almost never full. The fio read results are the following (write results are almost the same and are not showed here): Before ------ job: (groupid=0, jobs=8): err= 0: pid=2227: Tue Jul 19 11:29:50 2016 Description : [Emulation of Storage Server Access Pattern] read : io=54681MB, bw=1822.7MB/s, iops=179779, runt= 30001msec slat (usec): min=172, max=16883, avg=338.35, stdev=109.66 clat (usec): min=1, max=21977, avg=1051.45, stdev=299.29 lat (usec): min=317, max=22521, avg=1389.83, stdev=300.73 clat percentiles (usec): | 1.00th=[ 346], 5.00th=[ 596], 10.00th=[ 708], 20.00th=[ 852], | 30.00th=[ 932], 40.00th=[ 996], 50.00th=[ 1048], 60.00th=[ 1112], | 70.00th=[ 1176], 80.00th=[ 1256], 90.00th=[ 1384], 95.00th=[ 1496], | 99.00th=[ 1800], 99.50th=[ 1928], 99.90th=[ 2320], 99.95th=[ 2672], | 99.99th=[ 4704] bw (KB /s): min=205229, max=553181, per=12.50%, avg=233278.26, stdev=18383.51 After ------ job: (groupid=0, jobs=8): err= 0: pid=2220: Tue Jul 19 11:31:51 2016 Description : [Emulation of Storage Server Access Pattern] read : io=57637MB, bw=1921.2MB/s, iops=189529, runt= 30002msec slat (usec): min=169, max=20636, avg=329.61, stdev=124.18 clat (usec): min=2, max=19592, avg=988.78, stdev=251.04 lat (usec): min=381, max=21067, avg=1318.42, stdev=243.58 clat percentiles (usec): | 1.00th=[ 310], 5.00th=[ 580], 10.00th=[ 748], 20.00th=[ 876], | 30.00th=[ 908], 40.00th=[ 948], 50.00th=[ 1012], 60.00th=[ 1064], | 70.00th=[ 1080], 80.00th=[ 1128], 90.00th=[ 1224], 95.00th=[ 1288], | 99.00th=[ 1496], 99.50th=[ 1608], 99.90th=[ 1960], 99.95th=[ 2256], | 99.99th=[ 5408] bw (KB /s): min=212149, max=390160, per=12.49%, avg=245746.04, stdev=11606.75 Throughput increased from 1822MB/s to 1921MB/s, average completion latencies decreased from 1051us to 988us. Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com> Message-id: 1468931263-32667-4-git-send-email-roman.penyaev@profitbricks.com Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Paolo Bonzini <pbonzini@redhat.com> Cc: qemu-devel@nongnu.org Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-19 15:27:43 +03:00
if (laiocb.ret == -EINPROGRESS) {
qemu_coroutine_yield();
}
return laiocb.ret;
}
void laio_detach_aio_context(LinuxAioState *s, AioContext *old_context)
{
aio_set_event_notifier(old_context, &s->e, NULL, NULL, NULL);
qemu_bh_delete(s->completion_bh);
s->aio_context = NULL;
}
void laio_attach_aio_context(LinuxAioState *s, AioContext *new_context)
{
s->aio_context = new_context;
s->completion_bh = aio_bh_new(new_context, qemu_laio_completion_bh, s);
aio_set_event_notifier(new_context, &s->e,
qemu_laio_completion_cb,
aio-posix: split poll check from ready handler Adaptive polling measures the execution time of the polling check plus handlers called when a polled event becomes ready. Handlers can take a significant amount of time, making it look like polling was running for a long time when in fact the event handler was running for a long time. For example, on Linux the io_submit(2) syscall invoked when a virtio-blk device's virtqueue becomes ready can take 10s of microseconds. This can exceed the default polling interval (32 microseconds) and cause adaptive polling to stop polling. By excluding the handler's execution time from the polling check we make the adaptive polling calculation more accurate. As a result, the event loop now stays in polling mode where previously it would have fallen back to file descriptor monitoring. The following data was collected with virtio-blk num-queues=2 event_idx=off using an IOThread. Before: 168k IOPS, IOThread syscalls: 9837.115 ( 0.020 ms): IO iothread1/620155 io_submit(ctx_id: 140512552468480, nr: 16, iocbpp: 0x7fcb9f937db0) = 16 9837.158 ( 0.002 ms): IO iothread1/620155 write(fd: 103, buf: 0x556a2ef71b88, count: 8) = 8 9837.161 ( 0.001 ms): IO iothread1/620155 write(fd: 104, buf: 0x556a2ef71b88, count: 8) = 8 9837.163 ( 0.001 ms): IO iothread1/620155 ppoll(ufds: 0x7fcb90002800, nfds: 4, tsp: 0x7fcb9f1342d0, sigsetsize: 8) = 3 9837.164 ( 0.001 ms): IO iothread1/620155 read(fd: 107, buf: 0x7fcb9f939cc0, count: 512) = 8 9837.174 ( 0.001 ms): IO iothread1/620155 read(fd: 105, buf: 0x7fcb9f939cc0, count: 512) = 8 9837.176 ( 0.001 ms): IO iothread1/620155 read(fd: 106, buf: 0x7fcb9f939cc0, count: 512) = 8 9837.209 ( 0.035 ms): IO iothread1/620155 io_submit(ctx_id: 140512552468480, nr: 32, iocbpp: 0x7fca7d0cebe0) = 32 174k IOPS (+3.6%), IOThread syscalls: 9809.566 ( 0.036 ms): IO iothread1/623061 io_submit(ctx_id: 140539805028352, nr: 32, iocbpp: 0x7fd0cdd62be0) = 32 9809.625 ( 0.001 ms): IO iothread1/623061 write(fd: 103, buf: 0x5647cfba5f58, count: 8) = 8 9809.627 ( 0.002 ms): IO iothread1/623061 write(fd: 104, buf: 0x5647cfba5f58, count: 8) = 8 9809.663 ( 0.036 ms): IO iothread1/623061 io_submit(ctx_id: 140539805028352, nr: 32, iocbpp: 0x7fd0d0388b50) = 32 Notice that ppoll(2) and eventfd read(2) syscalls are eliminated because the IOThread stays in polling mode instead of falling back to file descriptor monitoring. As usual, polling is not implemented on Windows so this patch ignores the new io_poll_read() callback in aio-win32.c. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Stefano Garzarella <sgarzare@redhat.com> Message-id: 20211207132336.36627-2-stefanha@redhat.com [Fixed up aio_set_event_notifier() calls in tests/unit/test-fdmon-epoll.c added after this series was queued. --Stefan] Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2021-12-07 16:23:31 +03:00
qemu_laio_poll_cb,
qemu_laio_poll_ready);
}
LinuxAioState *laio_init(Error **errp)
{
int rc;
LinuxAioState *s;
s = g_malloc0(sizeof(*s));
rc = event_notifier_init(&s->e, false);
if (rc < 0) {
error_setg_errno(errp, -rc, "failed to initialize event notifier");
goto out_free_state;
}
rc = io_setup(MAX_EVENTS, &s->ctx);
if (rc < 0) {
error_setg_errno(errp, -rc, "failed to create linux AIO context");
goto out_close_efd;
}
ioq_init(&s->io_q);
return s;
out_close_efd:
event_notifier_cleanup(&s->e);
out_free_state:
g_free(s);
return NULL;
}
void laio_cleanup(LinuxAioState *s)
{
event_notifier_cleanup(&s->e);
if (io_destroy(s->ctx) != 0) {
fprintf(stderr, "%s: destroy AIO context %p failed\n",
__func__, &s->ctx);
}
g_free(s);
}
bool laio_has_fdsync(int fd)
{
struct iocb cb;
struct iocb *cbs[] = {&cb, NULL};
io_context_t ctx = 0;
io_setup(1, &ctx);
/* check if host kernel supports IO_CMD_FDSYNC */
io_prep_fdsync(&cb, fd);
int ret = io_submit(ctx, 1, cbs);
io_destroy(ctx);
return (ret == -EINVAL) ? false : true;
}