OpenE2K
/
qemu-e2k
13
4
Fork 0
Code Issues 4 Pull Requests Projects 2 Releases Activity

-----BEGIN PGP SIGNATURE----- 

Version: GnuPG v1
 
 iQEcBAABAgAGBQJVevNrAAoJEJykq7OBq3PIn3wH/iTkoxK+8c6vHY4gAyW/3Vel
 JAWyDULliLg/mJiyDyz+9hbyb3FmDsQ7Gve5laFCqSES8IChe/fAoR2dXAapHbJm
 VroBii+C4MIakAjB1eSGVTMeJVgyq0SxCbEgo6MSgCNRg8pH/ne4kGMn0elsyMQm
 VrxqsoW1qda0gLbybu7uRKmdZlb5Eklk0zQmBlzd25j3xgMXmo7ZcXW0d7dCi/xd
 ot01aPNHWKy2FyYsI+N1SMagCedjn0FFCHMMc08W34olUOAschV+Croxi6gupbTl
 RdlYNnB6BBYLgT7mxYuKKd20r4Q8Si3E5YjIsYaXIskbCOxDuryKboyheMXWqdA=
 =mx5x
 -----END PGP SIGNATURE-----

Merge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging

# gpg: Signature made Fri Jun 12 15:57:47 2015 BST using RSA key ID 81AB73C8
# gpg: Good signature from "Stefan Hajnoczi <stefanha@redhat.com>"
# gpg:                 aka "Stefan Hajnoczi <stefanha@gmail.com>"

* remotes/stefanha/tags/block-pull-request:
  qemu-iotests: expand test 093 to support group throttling
  throttle: Update throttle infrastructure copyright
  throttle: add the name of the ThrottleGroup to BlockDeviceInfo
  throttle: acquire the ThrottleGroup lock in bdrv_swap()
  throttle: Add throttle group support
  throttle: Add throttle group infrastructure tests
  throttle: Add throttle group infrastructure
  throttle: Extract timers from ThrottleState into a separate structure
  raw-posix: Fix .bdrv_co_get_block_status() for unaligned image size
  Revert "iothread: release iothread around aio_poll"

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This commit is contained in:
Peter Maydell 2015-06-12 18:04:14 +01:00
parent 0a2df857a7 2db33f88d2
commit 8aeaa055f5
21 changed files with 958 additions and 223 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
async.c
View File

@ -280,6 +280,12 @@ static void aio_timerlist_notify(void *opaque)
aio_notify(opaque);
}
static void aio_rfifolock_cb(void *opaque)
{
/* Kick owner thread in case they are blocked in aio_poll() */
aio_notify(opaque);
}
AioContext *aio_context_new(Error **errp)
{
int ret;
@ -297,7 +303,7 @@ AioContext *aio_context_new(Error **errp)
event_notifier_test_and_clear);
ctx->thread_pool = NULL;
qemu_mutex_init(&ctx->bh_lock);
rfifolock_init(&ctx->lock, NULL, NULL);
rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx);
timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
return ctx;

38
block.c
View File

@ -36,6 +36,7 @@
#include "qmp-commands.h"
#include "qemu/timer.h"
#include "qapi-event.h"
#include "block/throttle-groups.h"
#ifdef CONFIG_BSD
#include <sys/types.h>
@ -1822,12 +1823,18 @@ static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
bs_dest->enable_write_cache = bs_src->enable_write_cache;
/* i/o throttled req */
memcpy(&bs_dest->throttle_state,
&bs_src->throttle_state,
sizeof(ThrottleState));
bs_dest->throttle_state = bs_src->throttle_state,
bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
bs_dest->pending_reqs[0] = bs_src->pending_reqs[0];
bs_dest->pending_reqs[1] = bs_src->pending_reqs[1];
bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0];
bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1];
bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
memcpy(&bs_dest->round_robin,
&bs_src->round_robin,
sizeof(bs_dest->round_robin));
memcpy(&bs_dest->throttle_timers,
&bs_src->throttle_timers,
sizeof(ThrottleTimers));
/* r/w error */
bs_dest->on_read_error = bs_src->on_read_error;
@ -1881,12 +1888,21 @@ void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
QTAILQ_REMOVE(&graph_bdrv_states, bs_old, node_list);
}
/* If the BlockDriverState is part of a throttling group acquire
* its lock since we're going to mess with the protected fields.
* Otherwise there's no need to worry since no one else can touch
* them. */
if (bs_old->throttle_state) {
throttle_group_lock(bs_old);
}
/* bs_new must be unattached and shouldn't have anything fancy enabled */
assert(!bs_new->blk);
assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
assert(bs_new->job == NULL);
assert(bs_new->io_limits_enabled == false);
assert(!throttle_have_timer(&bs_new->throttle_state));
assert(bs_new->throttle_state == NULL);
assert(!throttle_timers_are_initialized(&bs_new->throttle_timers));
tmp = *bs_new;
*bs_new = *bs_old;
@ -1903,7 +1919,13 @@ void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
/* Check a few fields that should remain attached to the device */
assert(bs_new->job == NULL);
assert(bs_new->io_limits_enabled == false);
assert(!throttle_have_timer(&bs_new->throttle_state));
assert(bs_new->throttle_state == NULL);
assert(!throttle_timers_are_initialized(&bs_new->throttle_timers));
/* Release the ThrottleGroup lock */
if (bs_old->throttle_state) {
throttle_group_unlock(bs_old);
}
/* insert the nodes back into the graph node list if needed */
if (bs_new->node_name[0] != '\0') {
@ -3691,7 +3713,7 @@ void bdrv_detach_aio_context(BlockDriverState *bs)
}
if (bs->io_limits_enabled) {
throttle_detach_aio_context(&bs->throttle_state);
throttle_timers_detach_aio_context(&bs->throttle_timers);
}
if (bs->drv->bdrv_detach_aio_context) {
bs->drv->bdrv_detach_aio_context(bs);
@ -3727,7 +3749,7 @@ void bdrv_attach_aio_context(BlockDriverState *bs,
bs->drv->bdrv_attach_aio_context(bs, new_context);
}
if (bs->io_limits_enabled) {
throttle_attach_aio_context(&bs->throttle_state, new_context);
throttle_timers_attach_aio_context(&bs->throttle_timers, new_context);
}
QLIST_FOREACH(ban, &bs->aio_notifiers, list) {

1
block/Makefile.objs
View File

@ -10,6 +10,7 @@ block-obj-$(CONFIG_WIN32) += raw-win32.o win32-aio.o
block-obj-$(CONFIG_POSIX) += raw-posix.o
block-obj-$(CONFIG_LINUX_AIO) += linux-aio.o
block-obj-y += null.o mirror.o io.o
block-obj-y += throttle-groups.o
block-obj-y += nbd.o nbd-client.o sheepdog.o
block-obj-$(CONFIG_LIBISCSI) += iscsi.o

75
block/io.c
View File

@ -23,9 +23,9 @@
*/
#include "trace.h"
#include "sysemu/qtest.h"
#include "block/blockjob.h"
#include "block/block_int.h"
#include "block/throttle-groups.h"
#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
@ -65,7 +65,7 @@ void bdrv_set_io_limits(BlockDriverState *bs,
{
int i;
throttle_config(&bs->throttle_state, cfg);
throttle_group_config(bs, cfg);
for (i = 0; i < 2; i++) {
qemu_co_enter_next(&bs->throttled_reqs[i]);
@ -95,72 +95,33 @@ static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
void bdrv_io_limits_disable(BlockDriverState *bs)
{
bs->io_limits_enabled = false;
bdrv_start_throttled_reqs(bs);
throttle_destroy(&bs->throttle_state);
}
static void bdrv_throttle_read_timer_cb(void *opaque)
{
BlockDriverState *bs = opaque;
qemu_co_enter_next(&bs->throttled_reqs[0]);
}
static void bdrv_throttle_write_timer_cb(void *opaque)
{
BlockDriverState *bs = opaque;
qemu_co_enter_next(&bs->throttled_reqs[1]);
throttle_group_unregister_bs(bs);
}
/* should be called before bdrv_set_io_limits if a limit is set */
void bdrv_io_limits_enable(BlockDriverState *bs)
void bdrv_io_limits_enable(BlockDriverState *bs, const char *group)
{
int clock_type = QEMU_CLOCK_REALTIME;
if (qtest_enabled()) {
/* For testing block IO throttling only */
clock_type = QEMU_CLOCK_VIRTUAL;
}
assert(!bs->io_limits_enabled);
throttle_init(&bs->throttle_state,
bdrv_get_aio_context(bs),
clock_type,
bdrv_throttle_read_timer_cb,
bdrv_throttle_write_timer_cb,
bs);
throttle_group_register_bs(bs, group);
bs->io_limits_enabled = true;
}
/* This function makes an IO wait if needed
*
* @nb_sectors: the number of sectors of the IO
* @is_write: is the IO a write
*/
static void bdrv_io_limits_intercept(BlockDriverState *bs,
unsigned int bytes,
bool is_write)
void bdrv_io_limits_update_group(BlockDriverState *bs, const char *group)
{
/* does this io must wait */
bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
/* if must wait or any request of this type throttled queue the IO */
if (must_wait ||
!qemu_co_queue_empty(&bs->throttled_reqs[is_write])) {
qemu_co_queue_wait(&bs->throttled_reqs[is_write]);
}
/* the IO will be executed, do the accounting */
throttle_account(&bs->throttle_state, is_write, bytes);
/* if the next request must wait -> do nothing */
if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
/* this bs is not part of any group */
if (!bs->throttle_state) {
return;
}
/* else queue next request for execution */
qemu_co_queue_next(&bs->throttled_reqs[is_write]);
/* this bs is a part of the same group than the one we want */
if (!g_strcmp0(throttle_group_get_name(bs), group)) {
return;
}
/* need to change the group this bs belong to */
bdrv_io_limits_disable(bs);
bdrv_io_limits_enable(bs, group);
}
void bdrv_setup_io_funcs(BlockDriver *bdrv)
@ -967,7 +928,7 @@ static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
/* throttling disk I/O */
if (bs->io_limits_enabled) {
bdrv_io_limits_intercept(bs, bytes, false);
throttle_group_co_io_limits_intercept(bs, bytes, false);
}
/* Align read if necessary by padding qiov */
@ -1297,7 +1258,7 @@ static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
/* throttling disk I/O */
if (bs->io_limits_enabled) {
bdrv_io_limits_intercept(bs, bytes, true);
throttle_group_co_io_limits_intercept(bs, bytes, true);
}
/*

8
block/qapi.c
View File

@ -24,6 +24,7 @@
#include "block/qapi.h"
#include "block/block_int.h"
#include "block/throttle-groups.h"
#include "block/write-threshold.h"
#include "qmp-commands.h"
#include "qapi-visit.h"
@ -65,7 +66,9 @@ BlockDeviceInfo *bdrv_block_device_info(BlockDriverState *bs, Error **errp)
if (bs->io_limits_enabled) {
ThrottleConfig cfg;
throttle_get_config(&bs->throttle_state, &cfg);
throttle_group_get_config(bs, &cfg);
info->bps = cfg.buckets[THROTTLE_BPS_TOTAL].avg;
info->bps_rd = cfg.buckets[THROTTLE_BPS_READ].avg;
info->bps_wr = cfg.buckets[THROTTLE_BPS_WRITE].avg;
@ -90,6 +93,9 @@ BlockDeviceInfo *bdrv_block_device_info(BlockDriverState *bs, Error **errp)
info->has_iops_size = cfg.op_size;
info->iops_size = cfg.op_size;
info->has_group = true;
info->group = g_strdup(throttle_group_get_name(bs));
}
info->write_threshold = bdrv_write_threshold_get(bs);

5
block/raw-posix.c
View File

@ -1848,8 +1848,9 @@ static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
*pnum = nb_sectors;
ret = BDRV_BLOCK_DATA;
} else if (data == start) {
/* On a data extent, compute sectors to the end of the extent. */
*pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
/* On a data extent, compute sectors to the end of the extent,
* possibly including a partial sector at EOF. */
*pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
ret = BDRV_BLOCK_DATA;
} else {
/* On a hole, compute sectors to the beginning of the next extent. */

496
block/throttle-groups.c Normal file
View File

@ -0,0 +1,496 @@
/*
* QEMU block throttling group infrastructure
*
* Copyright (C) Nodalink, EURL. 2014
* Copyright (C) Igalia, S.L. 2015
*
* Authors:
* Benoît Canet <benoit.canet@nodalink.com>
* Alberto Garcia <berto@igalia.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation; either version 2 or
* (at your option) version 3 of the License.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, see <http://www.gnu.org/licenses/>.
*/
#include "block/throttle-groups.h"
#include "qemu/queue.h"
#include "qemu/thread.h"
#include "sysemu/qtest.h"
/* The ThrottleGroup structure (with its ThrottleState) is shared
* among different BlockDriverState and it's independent from
* AioContext, so in order to use it from different threads it needs
* its own locking.
*
* This locking is however handled internally in this file, so it's
* mostly transparent to outside users (but see the documentation in
* throttle_groups_lock()).
*
* The whole ThrottleGroup structure is private and invisible to
* outside users, that only use it through its ThrottleState.
*
* In addition to the ThrottleGroup structure, BlockDriverState has
* fields that need to be accessed by other members of the group and
* therefore also need to be protected by this lock. Once a BDS is
* registered in a group those fields can be accessed by other threads
* any time.
*
* Again, all this is handled internally and is mostly transparent to
* the outside. The 'throttle_timers' field however has an additional
* constraint because it may be temporarily invalid (see for example
* bdrv_set_aio_context()). Therefore in this file a thread will
* access some other BDS's timers only after verifying that that BDS
* has throttled requests in the queue.
*/
typedef struct ThrottleGroup {
char *name; /* This is constant during the lifetime of the group */
QemuMutex lock; /* This lock protects the following four fields */
ThrottleState ts;
QLIST_HEAD(, BlockDriverState) head;
BlockDriverState *tokens[2];
bool any_timer_armed[2];
/* These two are protected by the global throttle_groups_lock */
unsigned refcount;
QTAILQ_ENTRY(ThrottleGroup) list;
} ThrottleGroup;
static QemuMutex throttle_groups_lock;
static QTAILQ_HEAD(, ThrottleGroup) throttle_groups =
QTAILQ_HEAD_INITIALIZER(throttle_groups);
/* Increments the reference count of a ThrottleGroup given its name.
*
* If no ThrottleGroup is found with the given name a new one is
* created.
*
* @name: the name of the ThrottleGroup
* @ret: the ThrottleGroup
*/
static ThrottleGroup *throttle_group_incref(const char *name)
{
ThrottleGroup *tg = NULL;
ThrottleGroup *iter;
qemu_mutex_lock(&throttle_groups_lock);
/* Look for an existing group with that name */
QTAILQ_FOREACH(iter, &throttle_groups, list) {
if (!strcmp(name, iter->name)) {
tg = iter;
break;
}
}
/* Create a new one if not found */
if (!tg) {
tg = g_new0(ThrottleGroup, 1);
tg->name = g_strdup(name);
qemu_mutex_init(&tg->lock);
throttle_init(&tg->ts);
QLIST_INIT(&tg->head);
QTAILQ_INSERT_TAIL(&throttle_groups, tg, list);
}
tg->refcount++;
qemu_mutex_unlock(&throttle_groups_lock);
return tg;
}
/* Decrease the reference count of a ThrottleGroup.
*
* When the reference count reaches zero the ThrottleGroup is
* destroyed.
*
* @tg: The ThrottleGroup to unref
*/
static void throttle_group_unref(ThrottleGroup *tg)
{
qemu_mutex_lock(&throttle_groups_lock);
if (--tg->refcount == 0) {
QTAILQ_REMOVE(&throttle_groups, tg, list);
qemu_mutex_destroy(&tg->lock);
g_free(tg->name);
g_free(tg);
}
qemu_mutex_unlock(&throttle_groups_lock);
}
/* Get the name from a BlockDriverState's ThrottleGroup. The name (and
* the pointer) is guaranteed to remain constant during the lifetime
* of the group.
*
* @bs: a BlockDriverState that is member of a throttling group
* @ret: the name of the group.
*/
const char *throttle_group_get_name(BlockDriverState *bs)
{
ThrottleGroup *tg = container_of(bs->throttle_state, ThrottleGroup, ts);
return tg->name;
}
/* Return the next BlockDriverState in the round-robin sequence,
* simulating a circular list.
*
* This assumes that tg->lock is held.
*
* @bs: the current BlockDriverState
* @ret: the next BlockDriverState in the sequence
*/
static BlockDriverState *throttle_group_next_bs(BlockDriverState *bs)
{
ThrottleState *ts = bs->throttle_state;
ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
BlockDriverState *next = QLIST_NEXT(bs, round_robin);
if (!next) {
return QLIST_FIRST(&tg->head);
}
return next;
}
/* Return the next BlockDriverState in the round-robin sequence with
* pending I/O requests.
*
* This assumes that tg->lock is held.
*
* @bs: the current BlockDriverState
* @is_write: the type of operation (read/write)
* @ret: the next BlockDriverState with pending requests, or bs
* if there is none.
*/
static BlockDriverState *next_throttle_token(BlockDriverState *bs,
bool is_write)
{
ThrottleGroup *tg = container_of(bs->throttle_state, ThrottleGroup, ts);
BlockDriverState *token, *start;
start = token = tg->tokens[is_write];
/* get next bs round in round robin style */
token = throttle_group_next_bs(token);
while (token != start && !token->pending_reqs[is_write]) {
token = throttle_group_next_bs(token);
}
/* If no IO are queued for scheduling on the next round robin token
* then decide the token is the current bs because chances are
* the current bs get the current request queued.
*/
if (token == start && !token->pending_reqs[is_write]) {
token = bs;
}
return token;
}
/* Check if the next I/O request for a BlockDriverState needs to be
* throttled or not. If there's no timer set in this group, set one
* and update the token accordingly.
*
* This assumes that tg->lock is held.
*
* @bs: the current BlockDriverState
* @is_write: the type of operation (read/write)
* @ret: whether the I/O request needs to be throttled or not
*/
static bool throttle_group_schedule_timer(BlockDriverState *bs,
bool is_write)
{
ThrottleState *ts = bs->throttle_state;
ThrottleTimers *tt = &bs->throttle_timers;
ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
bool must_wait;
/* Check if any of the timers in this group is already armed */
if (tg->any_timer_armed[is_write]) {
return true;
}
must_wait = throttle_schedule_timer(ts, tt, is_write);
/* If a timer just got armed, set bs as the current token */
if (must_wait) {
tg->tokens[is_write] = bs;
tg->any_timer_armed[is_write] = true;
}
return must_wait;
}
/* Look for the next pending I/O request and schedule it.
*
* This assumes that tg->lock is held.
*
* @bs: the current BlockDriverState
* @is_write: the type of operation (read/write)
*/
static void schedule_next_request(BlockDriverState *bs, bool is_write)
{
ThrottleGroup *tg = container_of(bs->throttle_state, ThrottleGroup, ts);
bool must_wait;
BlockDriverState *token;
/* Check if there's any pending request to schedule next */
token = next_throttle_token(bs, is_write);
if (!token->pending_reqs[is_write]) {
return;
}
/* Set a timer for the request if it needs to be throttled */
must_wait = throttle_group_schedule_timer(token, is_write);
/* If it doesn't have to wait, queue it for immediate execution */
if (!must_wait) {
/* Give preference to requests from the current bs */
if (qemu_in_coroutine() &&
qemu_co_queue_next(&bs->throttled_reqs[is_write])) {
token = bs;
} else {
ThrottleTimers *tt = &token->throttle_timers;
int64_t now = qemu_clock_get_ns(tt->clock_type);
timer_mod(tt->timers[is_write], now + 1);
tg->any_timer_armed[is_write] = true;
}
tg->tokens[is_write] = token;
}
}
/* Check if an I/O request needs to be throttled, wait and set a timer
* if necessary, and schedule the next request using a round robin
* algorithm.
*
* @bs: the current BlockDriverState
* @bytes: the number of bytes for this I/O
* @is_write: the type of operation (read/write)
*/
void coroutine_fn throttle_group_co_io_limits_intercept(BlockDriverState *bs,
unsigned int bytes,
bool is_write)
{
bool must_wait;
BlockDriverState *token;
ThrottleGroup *tg = container_of(bs->throttle_state, ThrottleGroup, ts);
qemu_mutex_lock(&tg->lock);
/* First we check if this I/O has to be throttled. */
token = next_throttle_token(bs, is_write);
must_wait = throttle_group_schedule_timer(token, is_write);
/* Wait if there's a timer set or queued requests of this type */
if (must_wait || bs->pending_reqs[is_write]) {
bs->pending_reqs[is_write]++;
qemu_mutex_unlock(&tg->lock);
qemu_co_queue_wait(&bs->throttled_reqs[is_write]);
qemu_mutex_lock(&tg->lock);
bs->pending_reqs[is_write]--;
}
/* The I/O will be executed, so do the accounting */
throttle_account(bs->throttle_state, is_write, bytes);
/* Schedule the next request */
schedule_next_request(bs, is_write);
qemu_mutex_unlock(&tg->lock);
}
/* Update the throttle configuration for a particular group. Similar
* to throttle_config(), but guarantees atomicity within the
* throttling group.
*
* @bs: a BlockDriverState that is member of the group
* @cfg: the configuration to set
*/
void throttle_group_config(BlockDriverState *bs, ThrottleConfig *cfg)
{
ThrottleTimers *tt = &bs->throttle_timers;
ThrottleState *ts = bs->throttle_state;
ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
qemu_mutex_lock(&tg->lock);
throttle_config(ts, tt, cfg);
/* throttle_config() cancels the timers */
tg->any_timer_armed[0] = tg->any_timer_armed[1] = false;
qemu_mutex_unlock(&tg->lock);
}
/* Get the throttle configuration from a particular group. Similar to
* throttle_get_config(), but guarantees atomicity within the
* throttling group.
*
* @bs: a BlockDriverState that is member of the group
* @cfg: the configuration will be written here
*/
void throttle_group_get_config(BlockDriverState *bs, ThrottleConfig *cfg)
{
ThrottleState *ts = bs->throttle_state;
ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
qemu_mutex_lock(&tg->lock);
throttle_get_config(ts, cfg);
qemu_mutex_unlock(&tg->lock);
}
/* ThrottleTimers callback. This wakes up a request that was waiting
* because it had been throttled.
*
* @bs: the BlockDriverState whose request had been throttled
* @is_write: the type of operation (read/write)
*/
static void timer_cb(BlockDriverState *bs, bool is_write)
{
ThrottleState *ts = bs->throttle_state;
ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
bool empty_queue;
/* The timer has just been fired, so we can update the flag */
qemu_mutex_lock(&tg->lock);
tg->any_timer_armed[is_write] = false;
qemu_mutex_unlock(&tg->lock);
/* Run the request that was waiting for this timer */
empty_queue = !qemu_co_enter_next(&bs->throttled_reqs[is_write]);
/* If the request queue was empty then we have to take care of
* scheduling the next one */
if (empty_queue) {
qemu_mutex_lock(&tg->lock);
schedule_next_request(bs, is_write);
qemu_mutex_unlock(&tg->lock);
}
}
static void read_timer_cb(void *opaque)
{
timer_cb(opaque, false);
}
static void write_timer_cb(void *opaque)
{
timer_cb(opaque, true);
}
/* Register a BlockDriverState in the throttling group, also
* initializing its timers and updating its throttle_state pointer to
* point to it. If a throttling group with that name does not exist
* yet, it will be created.
*
* @bs: the BlockDriverState to insert
* @groupname: the name of the group
*/
void throttle_group_register_bs(BlockDriverState *bs, const char *groupname)
{
int i;
ThrottleGroup *tg = throttle_group_incref(groupname);
int clock_type = QEMU_CLOCK_REALTIME;
if (qtest_enabled()) {
/* For testing block IO throttling only */
clock_type = QEMU_CLOCK_VIRTUAL;
}
bs->throttle_state = &tg->ts;
qemu_mutex_lock(&tg->lock);
/* If the ThrottleGroup is new set this BlockDriverState as the token */
for (i = 0; i < 2; i++) {
if (!tg->tokens[i]) {
tg->tokens[i] = bs;
}
}
QLIST_INSERT_HEAD(&tg->head, bs, round_robin);
throttle_timers_init(&bs->throttle_timers,
bdrv_get_aio_context(bs),
clock_type,
read_timer_cb,
write_timer_cb,
bs);
qemu_mutex_unlock(&tg->lock);
}
/* Unregister a BlockDriverState from its group, removing it from the
* list, destroying the timers and setting the throttle_state pointer
* to NULL.
*
* The group will be destroyed if it's empty after this operation.
*
* @bs: the BlockDriverState to remove
*/
void throttle_group_unregister_bs(BlockDriverState *bs)
{
ThrottleGroup *tg = container_of(bs->throttle_state, ThrottleGroup, ts);
int i;
qemu_mutex_lock(&tg->lock);
for (i = 0; i < 2; i++) {
if (tg->tokens[i] == bs) {
BlockDriverState *token = throttle_group_next_bs(bs);
/* Take care of the case where this is the last bs in the group */
if (token == bs) {
token = NULL;
}
tg->tokens[i] = token;
}
}
/* remove the current bs from the list */
QLIST_REMOVE(bs, round_robin);
throttle_timers_destroy(&bs->throttle_timers);
qemu_mutex_unlock(&tg->lock);
throttle_group_unref(tg);
bs->throttle_state = NULL;
}
/* Acquire the lock of this throttling group.
*
* You won't normally need to use this. None of the functions from the
* ThrottleGroup API require you to acquire the lock since all of them
* deal with it internally.
*
* This should only be used in exceptional cases when you want to
* access the protected fields of a BlockDriverState directly
* (e.g. bdrv_swap()).
*
* @bs: a BlockDriverState that is member of the group
*/
void throttle_group_lock(BlockDriverState *bs)
{
ThrottleGroup *tg = container_of(bs->throttle_state, ThrottleGroup, ts);
qemu_mutex_lock(&tg->lock);
}
/* Release the lock of this throttling group.
*
* See the comments in throttle_group_lock().
*/
void throttle_group_unlock(BlockDriverState *bs)
{
ThrottleGroup *tg = container_of(bs->throttle_state, ThrottleGroup, ts);
qemu_mutex_unlock(&tg->lock);
}
static void throttle_groups_init(void)
{
qemu_mutex_init(&throttle_groups_lock);
}
block_init(throttle_groups_init);

38
blockdev.c
View File

@ -34,6 +34,7 @@
#include "sysemu/blockdev.h"
#include "hw/block/block.h"
#include "block/blockjob.h"
#include "block/throttle-groups.h"
#include "monitor/monitor.h"
#include "qemu/option.h"
#include "qemu/config-file.h"
@ -357,6 +358,7 @@ static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
const char *id;
bool has_driver_specific_opts;
BlockdevDetectZeroesOptions detect_zeroes;
const char *throttling_group;
/* Check common options by copying from bs_opts to opts, all other options
* stay in bs_opts for processing by bdrv_open(). */
@ -459,6 +461,8 @@ static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
cfg.op_size = qemu_opt_get_number(opts, "throttling.iops-size", 0);
throttling_group = qemu_opt_get(opts, "throttling.group");
if (!check_throttle_config(&cfg, &error)) {
error_propagate(errp, error);
goto early_err;
@ -547,7 +551,10 @@ static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
/* disk I/O throttling */
if (throttle_enabled(&cfg)) {
bdrv_io_limits_enable(bs);
if (!throttling_group) {
throttling_group = blk_name(blk);
}
bdrv_io_limits_enable(bs, throttling_group);
bdrv_set_io_limits(bs, &cfg);
}
@ -711,6 +718,8 @@ DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
{ "iops_size", "throttling.iops-size" },
{ "group", "throttling.group" },
{ "readonly", "read-only" },
};
@ -1951,7 +1960,9 @@ void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
bool has_iops_wr_max,
int64_t iops_wr_max,
bool has_iops_size,
int64_t iops_size, Error **errp)
int64_t iops_size,
bool has_group,
const char *group, Error **errp)
{
ThrottleConfig cfg;
BlockDriverState *bs;
@ -2004,14 +2015,19 @@ void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
aio_context = bdrv_get_aio_context(bs);
aio_context_acquire(aio_context);
if (!bs->io_limits_enabled && throttle_enabled(&cfg)) {
bdrv_io_limits_enable(bs);
} else if (bs->io_limits_enabled && !throttle_enabled(&cfg)) {
bdrv_io_limits_disable(bs);
}
if (bs->io_limits_enabled) {
if (throttle_enabled(&cfg)) {
/* Enable I/O limits if they're not enabled yet, otherwise
* just update the throttling group. */
if (!bs->io_limits_enabled) {
bdrv_io_limits_enable(bs, has_group ? group : device);
} else if (has_group) {
bdrv_io_limits_update_group(bs, group);
}
/* Set the new throttling configuration */
bdrv_set_io_limits(bs, &cfg);
} else if (bs->io_limits_enabled) {
/* If all throttling settings are set to 0, disable I/O limits */
bdrv_io_limits_disable(bs);
}
aio_context_release(aio_context);
@ -3188,6 +3204,10 @@ QemuOptsList qemu_common_drive_opts = {
.name = "throttling.iops-size",
.type = QEMU_OPT_NUMBER,
.help = "when limiting by iops max size of an I/O in bytes",
},{
.name = "throttling.group",
.type = QEMU_OPT_STRING,
.help = "name of the block throttling group",
},{
.name = "copy-on-read",
.type = QEMU_OPT_BOOL,

10
hmp.c
View File

@ -400,7 +400,8 @@ static void print_block_info(Monitor *mon, BlockInfo *info,
" iops_max=%" PRId64
" iops_rd_max=%" PRId64
" iops_wr_max=%" PRId64
" iops_size=%" PRId64 "\n",
" iops_size=%" PRId64
" group=%s\n",
inserted->bps,
inserted->bps_rd,
inserted->bps_wr,
@ -413,7 +414,8 @@ static void print_block_info(Monitor *mon, BlockInfo *info,
inserted->iops_max,
inserted->iops_rd_max,
inserted->iops_wr_max,
inserted->iops_size);
inserted->iops_size,
inserted->group);
}
if (verbose) {
@ -1357,7 +1359,9 @@ void hmp_block_set_io_throttle(Monitor *mon, const QDict *qdict)
false,
0,
false, /* No default I/O size */
0, &err);
0,
false,
NULL, &err);
hmp_handle_error(mon, &err);
}

3
include/block/block.h
View File

@ -173,8 +173,9 @@ void bdrv_stats_print(Monitor *mon, const QObject *data);
void bdrv_info_stats(Monitor *mon, QObject **ret_data);
/* disk I/O throttling */
void bdrv_io_limits_enable(BlockDriverState *bs);
void bdrv_io_limits_enable(BlockDriverState *bs, const char *group);
void bdrv_io_limits_disable(BlockDriverState *bs);
void bdrv_io_limits_update_group(BlockDriverState *bs, const char *group);
void bdrv_init(void);
void bdrv_init_with_whitelist(void);

7
include/block/block_int.h
View File

@ -379,9 +379,14 @@ struct BlockDriverState {
unsigned int serialising_in_flight;
/* I/O throttling */
ThrottleState throttle_state;
CoQueue throttled_reqs[2];
bool io_limits_enabled;
/* The following fields are protected by the ThrottleGroup lock.
* See the ThrottleGroup documentation for details. */
ThrottleState *throttle_state;
ThrottleTimers throttle_timers;
unsigned pending_reqs[2];
QLIST_ENTRY(BlockDriverState) round_robin;
/* I/O stats (display with "info blockstats"). */
BlockAcctStats stats;

46
include/block/throttle-groups.h Normal file
View File

@ -0,0 +1,46 @@
/*
* QEMU block throttling group infrastructure
*
* Copyright (C) Nodalink, EURL. 2014
* Copyright (C) Igalia, S.L. 2015
*
* Authors:
* Benoît Canet <benoit.canet@nodalink.com>
* Alberto Garcia <berto@igalia.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation; either version 2 or
* (at your option) version 3 of the License.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, see <http://www.gnu.org/licenses/>.
*/
#ifndef THROTTLE_GROUPS_H
#define THROTTLE_GROUPS_H
#include "qemu/throttle.h"
#include "block/block_int.h"
const char *throttle_group_get_name(BlockDriverState *bs);
void throttle_group_config(BlockDriverState *bs, ThrottleConfig *cfg);
void throttle_group_get_config(BlockDriverState *bs, ThrottleConfig *cfg);
void throttle_group_register_bs(BlockDriverState *bs, const char *groupname);
void throttle_group_unregister_bs(BlockDriverState *bs);
void coroutine_fn throttle_group_co_io_limits_intercept(BlockDriverState *bs,
unsigned int bytes,
bool is_write);
void throttle_group_lock(BlockDriverState *bs);
void throttle_group_unlock(BlockDriverState *bs);
#endif

46
include/qemu/throttle.h
View File

@ -1,10 +1,12 @@
/*
* QEMU throttling infrastructure
*
* Copyright (C) Nodalink, SARL. 2013
* Copyright (C) Nodalink, EURL. 2013-2014
* Copyright (C) Igalia, S.L. 2015
*
* Author:
* Benoît Canet <benoit.canet@irqsave.net>
* Authors:
* Benoît Canet <benoit.canet@nodalink.com>
* Alberto Garcia <berto@igalia.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
@ -65,14 +67,17 @@ typedef struct ThrottleConfig {
typedef struct ThrottleState {
ThrottleConfig cfg; /* configuration */
int64_t previous_leak; /* timestamp of the last leak done */
QEMUTimer * timers[2]; /* timers used to do the throttling */
} ThrottleState;
typedef struct ThrottleTimers {
QEMUTimer *timers[2]; /* timers used to do the throttling */
QEMUClockType clock_type; /* the clock used */
/* Callbacks */
QEMUTimerCB *read_timer_cb;
QEMUTimerCB *write_timer_cb;
void *timer_opaque;
} ThrottleState;
} ThrottleTimers;
/* operations on single leaky buckets */
void throttle_leak_bucket(LeakyBucket *bkt, int64_t delta);
@ -86,20 +91,23 @@ bool throttle_compute_timer(ThrottleState *ts,
int64_t *next_timestamp);
/* init/destroy cycle */
void throttle_init(ThrottleState *ts,
AioContext *aio_context,
QEMUClockType clock_type,
void (read_timer)(void *),
void (write_timer)(void *),
void *timer_opaque);
void throttle_init(ThrottleState *ts);
void throttle_destroy(ThrottleState *ts);
void throttle_timers_init(ThrottleTimers *tt,
AioContext *aio_context,
QEMUClockType clock_type,
QEMUTimerCB *read_timer_cb,
QEMUTimerCB *write_timer_cb,
void *timer_opaque);
void throttle_detach_aio_context(ThrottleState *ts);
void throttle_timers_destroy(ThrottleTimers *tt);
void throttle_attach_aio_context(ThrottleState *ts, AioContext *new_context);
void throttle_timers_detach_aio_context(ThrottleTimers *tt);
bool throttle_have_timer(ThrottleState *ts);
void throttle_timers_attach_aio_context(ThrottleTimers *tt,
AioContext *new_context);
bool throttle_timers_are_initialized(ThrottleTimers *tt);
/* configuration */
bool throttle_enabled(ThrottleConfig *cfg);
@ -108,12 +116,16 @@ bool throttle_conflicting(ThrottleConfig *cfg);
bool throttle_is_valid(ThrottleConfig *cfg);
void throttle_config(ThrottleState *ts, ThrottleConfig *cfg);
void throttle_config(ThrottleState *ts,
ThrottleTimers *tt,
ThrottleConfig *cfg);
void throttle_get_config(ThrottleState *ts, ThrottleConfig *cfg);
/* usage */
bool throttle_schedule_timer(ThrottleState *ts, bool is_write);
bool throttle_schedule_timer(ThrottleState *ts,
ThrottleTimers *tt,
bool is_write);
void throttle_account(ThrottleState *ts, bool is_write, uint64_t size);

11
iothread.c
View File

@ -31,14 +31,21 @@ typedef ObjectClass IOThreadClass;
static void *iothread_run(void *opaque)
{
IOThread *iothread = opaque;
bool blocking;
qemu_mutex_lock(&iothread->init_done_lock);
iothread->thread_id = qemu_get_thread_id();
qemu_cond_signal(&iothread->init_done_cond);
qemu_mutex_unlock(&iothread->init_done_lock);
while (!atomic_read(&iothread->stopping)) {
aio_poll(iothread->ctx, true);
while (!iothread->stopping) {
aio_context_acquire(iothread->ctx);
blocking = true;
while (!iothread->stopping && aio_poll(iothread->ctx, blocking)) {
/* Progress was made, keep going */
blocking = false;
}
aio_context_release(iothread->ctx);
}
return NULL;
}

29
qapi/block-core.json
View File

@ -259,6 +259,8 @@
#
# @iops_size: #optional an I/O size in bytes (Since 1.7)
#
# @group: #optional throttle group name (Since 2.4)
#
# @cache: the cache mode used for the block device (since: 2.3)
#
# @write_threshold: configured write threshold for the device.
@ -278,7 +280,7 @@
'*bps_max': 'int', '*bps_rd_max': 'int',
'*bps_wr_max': 'int', '*iops_max': 'int',
'*iops_rd_max': 'int', '*iops_wr_max': 'int',
'*iops_size': 'int', 'cache': 'BlockdevCacheInfo',
'*iops_size': 'int', '*group': 'str', 'cache': 'BlockdevCacheInfo',
'write_threshold': 'int' } }
##
@ -1062,6 +1064,27 @@
#
# Change I/O throttle limits for a block drive.
#
# Since QEMU 2.4, each device with I/O limits is member of a throttle
# group.
#
# If two or more devices are members of the same group, the limits
# will apply to the combined I/O of the whole group in a round-robin
# fashion. Therefore, setting new I/O limits to a device will affect
# the whole group.
#
# The name of the group can be specified using the 'group' parameter.
# If the parameter is unset, it is assumed to be the current group of
# that device. If it's not in any group yet, the name of the device
# will be used as the name for its group.
#
# The 'group' parameter can also be used to move a device to a
# different group. In this case the limits specified in the parameters
# will be applied to the new group only.
#
# I/O limits can be disabled by setting all of them to 0. In this case
# the device will be removed from its group and the rest of its
# members will no be affected. The 'group' parameter is ignored.
#
# @device: The name of the device
#
# @bps: total throughput limit in bytes per second
@ -1090,6 +1113,8 @@
#
# @iops_size: #optional an I/O size in bytes (Since 1.7)
#
# @group: #optional throttle group name (Since 2.4)
#
# Returns: Nothing on success
# If @device is not a valid block device, DeviceNotFound
#
@ -1101,7 +1126,7 @@
'*bps_max': 'int', '*bps_rd_max': 'int',
'*bps_wr_max': 'int', '*iops_max': 'int',
'*iops_rd_max': 'int', '*iops_wr_max': 'int',
'*iops_size': 'int' } }
'*iops_size': 'int', '*group': 'str' } }
##
# @block-stream:

1
qemu-options.hx
View File

@ -468,6 +468,7 @@ DEF("drive", HAS_ARG, QEMU_OPTION_drive,
" [[,bps_max=bm]|[[,bps_rd_max=rm][,bps_wr_max=wm]]]\n"
" [[,iops_max=im]|[[,iops_rd_max=irm][,iops_wr_max=iwm]]]\n"
" [[,iops_size=is]]\n"
" [[,group=g]]\n"
" use 'file' as a drive image\n", QEMU_ARCH_ALL)
STEXI
@item -drive @var{option}[,@var{option}[,@var{option}[,...]]]

3
qmp-commands.hx
View File

@ -1853,7 +1853,7 @@ EQMP
{
.name = "block_set_io_throttle",
.args_type = "device:B,bps:l,bps_rd:l,bps_wr:l,iops:l,iops_rd:l,iops_wr:l,bps_max:l?,bps_rd_max:l?,bps_wr_max:l?,iops_max:l?,iops_rd_max:l?,iops_wr_max:l?,iops_size:l?",
.args_type = "device:B,bps:l,bps_rd:l,bps_wr:l,iops:l,iops_rd:l,iops_wr:l,bps_max:l?,bps_rd_max:l?,bps_wr_max:l?,iops_max:l?,iops_rd_max:l?,iops_wr_max:l?,iops_size:l?,group:s?",
.mhandler.cmd_new = qmp_marshal_input_block_set_io_throttle,
},
@ -1879,6 +1879,7 @@ Arguments:
- "iops_rd_max": read I/O operations max (json-int)
- "iops_wr_max": write I/O operations max (json-int)
- "iops_size": I/O size in bytes when limiting (json-int)
- "group": throttle group name (json-string)
Example:

93
tests/qemu-iotests/093
View File

@ -3,6 +3,7 @@
# Tests for IO throttling
#
# Copyright (C) 2015 Red Hat, Inc.
# Copyright (C) 2015 Igalia, S.L.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
@ -22,6 +23,7 @@ import iotests
class ThrottleTestCase(iotests.QMPTestCase):
test_img = "null-aio://"
max_drives = 3
def blockstats(self, device):
result = self.vm.qmp("query-blockstats")
@ -32,26 +34,31 @@ class ThrottleTestCase(iotests.QMPTestCase):
raise Exception("Device not found for blockstats: %s" % device)
def setUp(self):
self.vm = iotests.VM().add_drive(self.test_img)
self.vm = iotests.VM()
for i in range(0, self.max_drives):
self.vm.add_drive(self.test_img)
self.vm.launch()
def tearDown(self):
self.vm.shutdown()
def do_test_throttle(self, seconds, params):
def do_test_throttle(self, ndrives, seconds, params):
def check_limit(limit, num):
# IO throttling algorithm is discrete, allow 10% error so the test
# is more robust
return limit == 0 or \
(num < seconds * limit * 1.1
and num > seconds * limit * 0.9)
(num < seconds * limit * 1.1 / ndrives
and num > seconds * limit * 0.9 / ndrives)
nsec_per_sec = 1000000000
params['device'] = 'drive0'
params['group'] = 'test'
result = self.vm.qmp("block_set_io_throttle", conv_keys=False, **params)
self.assert_qmp(result, 'return', {})
# Set the I/O throttling parameters to all drives
for i in range(0, ndrives):
params['device'] = 'drive%d' % i
result = self.vm.qmp("block_set_io_throttle", conv_keys=False, **params)
self.assert_qmp(result, 'return', {})
# Set vm clock to a known value
ns = seconds * nsec_per_sec
@ -66,32 +73,60 @@ class ThrottleTestCase(iotests.QMPTestCase):
params['iops'] / 2,
params['iops_rd'])
rd_nr *= seconds * 2
rd_nr /= ndrives
wr_nr = max(params['bps'] / rq_size / 2,
params['bps_wr'] / rq_size,
params['iops'] / 2,
params['iops_wr'])
wr_nr *= seconds * 2
for i in range(rd_nr):
self.vm.hmp_qemu_io("drive0", "aio_read %d %d" % (i * rq_size, rq_size))
for i in range(wr_nr):
self.vm.hmp_qemu_io("drive0", "aio_write %d %d" % (i * rq_size, rq_size))
wr_nr /= ndrives
start_rd_bytes, start_rd_iops, start_wr_bytes, start_wr_iops = self.blockstats('drive0')
# Send I/O requests to all drives
for i in range(rd_nr):
for drive in range(0, ndrives):
self.vm.hmp_qemu_io("drive%d" % drive, "aio_read %d %d" %
(i * rq_size, rq_size))
for i in range(wr_nr):
for drive in range(0, ndrives):
self.vm.hmp_qemu_io("drive%d" % drive, "aio_write %d %d" %
(i * rq_size, rq_size))
# We'll store the I/O stats for each drive in these arrays
start_rd_bytes = [0] * ndrives
start_rd_iops = [0] * ndrives
start_wr_bytes = [0] * ndrives
start_wr_iops = [0] * ndrives
end_rd_bytes = [0] * ndrives
end_rd_iops = [0] * ndrives
end_wr_bytes = [0] * ndrives
end_wr_iops = [0] * ndrives
# Read the stats before advancing the clock
for i in range(0, ndrives):
start_rd_bytes[i], start_rd_iops[i], start_wr_bytes[i], \
start_wr_iops[i] = self.blockstats('drive%d' % i)
self.vm.qtest("clock_step %d" % ns)
end_rd_bytes, end_rd_iops, end_wr_bytes, end_wr_iops = self.blockstats('drive0')
rd_bytes = end_rd_bytes - start_rd_bytes
rd_iops = end_rd_iops - start_rd_iops
wr_bytes = end_wr_bytes - start_wr_bytes
wr_iops = end_wr_iops - start_wr_iops
# Read the stats after advancing the clock
for i in range(0, ndrives):
end_rd_bytes[i], end_rd_iops[i], end_wr_bytes[i], \
end_wr_iops[i] = self.blockstats('drive%d' % i)
self.assertTrue(check_limit(params['bps'], rd_bytes + wr_bytes))
self.assertTrue(check_limit(params['bps_rd'], rd_bytes))
self.assertTrue(check_limit(params['bps_wr'], wr_bytes))
self.assertTrue(check_limit(params['iops'], rd_iops + wr_iops))
self.assertTrue(check_limit(params['iops_rd'], rd_iops))
self.assertTrue(check_limit(params['iops_wr'], wr_iops))
# Check that the I/O is within the limits and evenly distributed
for i in range(0, ndrives):
rd_bytes = end_rd_bytes[i] - start_rd_bytes[i]
rd_iops = end_rd_iops[i] - start_rd_iops[i]
wr_bytes = end_wr_bytes[i] - start_wr_bytes[i]
wr_iops = end_wr_iops[i] - start_wr_iops[i]
self.assertTrue(check_limit(params['bps'], rd_bytes + wr_bytes))
self.assertTrue(check_limit(params['bps_rd'], rd_bytes))
self.assertTrue(check_limit(params['bps_wr'], wr_bytes))
self.assertTrue(check_limit(params['iops'], rd_iops + wr_iops))
self.assertTrue(check_limit(params['iops_rd'], rd_iops))
self.assertTrue(check_limit(params['iops_wr'], wr_iops))
def test_all(self):
params = {"bps": 4096,
@ -101,11 +136,13 @@ class ThrottleTestCase(iotests.QMPTestCase):
"iops_rd": 10,
"iops_wr": 10,
}
# Pick each out of all possible params and test
for tk in params:
limits = dict([(k, 0) for k in params])
limits[tk] = params[tk]
self.do_test_throttle(5, limits)
# Repeat the test with different numbers of drives
for ndrives in range(1, self.max_drives + 1):
# Pick each out of all possible params and test
for tk in params:
limits = dict([(k, 0) for k in params])
limits[tk] = params[tk] * ndrives
self.do_test_throttle(ndrives, 5, limits)
class ThrottleTestCoroutine(ThrottleTestCase):
test_img = "null-co://"

19
tests/test-aio.c
View File

@ -107,7 +107,6 @@ static void test_notify(void)
typedef struct {
QemuMutex start_lock;
EventNotifier notifier;
bool thread_acquired;
} AcquireTestData;
@ -119,8 +118,6 @@ static void *test_acquire_thread(void *opaque)
qemu_mutex_lock(&data->start_lock);
qemu_mutex_unlock(&data->start_lock);
g_usleep(500000);
event_notifier_set(&data->notifier);
aio_context_acquire(ctx);
aio_context_release(ctx);
@ -129,19 +126,20 @@ static void *test_acquire_thread(void *opaque)
return NULL;
}
static void dummy_notifier_read(EventNotifier *n)
static void dummy_notifier_read(EventNotifier *unused)
{
event_notifier_test_and_clear(n);
g_assert(false); /* should never be invoked */
}
static void test_acquire(void)
{
QemuThread thread;
EventNotifier notifier;
AcquireTestData data;
/* Dummy event notifier ensures aio_poll() will block */
event_notifier_init(&data.notifier, false);
aio_set_event_notifier(ctx, &data.notifier, dummy_notifier_read);
event_notifier_init(&notifier, false);
aio_set_event_notifier(ctx, &notifier, dummy_notifier_read);
g_assert(!aio_poll(ctx, false)); /* consume aio_notify() */
qemu_mutex_init(&data.start_lock);
@ -155,13 +153,12 @@ static void test_acquire(void)
/* Block in aio_poll(), let other thread kick us and acquire context */
aio_context_acquire(ctx);
qemu_mutex_unlock(&data.start_lock); /* let the thread run */
g_assert(aio_poll(ctx, true));
g_assert(!data.thread_acquired);
g_assert(!aio_poll(ctx, true));
aio_context_release(ctx);
qemu_thread_join(&thread);
aio_set_event_notifier(ctx, &data.notifier, NULL);
event_notifier_cleanup(&data.notifier);
aio_set_event_notifier(ctx, &notifier, NULL);
event_notifier_cleanup(&notifier);
g_assert(data.thread_acquired);
}

163
tests/test-throttle.c
View File

@ -1,10 +1,12 @@
/*
* Throttle infrastructure tests
*
* Copyright Nodalink, SARL. 2013
* Copyright Nodalink, EURL. 2013-2014
* Copyright Igalia, S.L. 2015
*
* Authors:
* Benoît Canet <benoit.canet@irqsave.net>
* Benoît Canet <benoit.canet@nodalink.com>
* Alberto Garcia <berto@igalia.com>
*
* This work is licensed under the terms of the GNU LGPL, version 2 or later.
* See the COPYING.LIB file in the top-level directory.
@ -15,11 +17,13 @@
#include "block/aio.h"
#include "qemu/throttle.h"
#include "qemu/error-report.h"
#include "block/throttle-groups.h"
static AioContext *ctx;
static LeakyBucket bkt;
static ThrottleConfig cfg;
static ThrottleState ts;
static ThrottleTimers tt;
/* useful function */
static bool double_cmp(double x, double y)
@ -103,17 +107,19 @@ static void test_init(void)
{
int i;
/* fill the structure with crap */
/* fill the structures with crap */
memset(&ts, 1, sizeof(ts));
memset(&tt, 1, sizeof(tt));
/* init the structure */
throttle_init(&ts, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
/* init structures */
throttle_init(&ts);
throttle_timers_init(&tt, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
/* check initialized fields */
g_assert(ts.clock_type == QEMU_CLOCK_VIRTUAL);
g_assert(ts.timers[0]);
g_assert(ts.timers[1]);
g_assert(tt.clock_type == QEMU_CLOCK_VIRTUAL);
g_assert(tt.timers[0]);
g_assert(tt.timers[1]);
/* check other fields where cleared */
g_assert(!ts.previous_leak);
@ -124,17 +130,18 @@ static void test_init(void)
g_assert(!ts.cfg.buckets[i].level);
}
throttle_destroy(&ts);
throttle_timers_destroy(&tt);
}
static void test_destroy(void)
{
int i;
throttle_init(&ts, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
throttle_destroy(&ts);
throttle_init(&ts);
throttle_timers_init(&tt, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
throttle_timers_destroy(&tt);
for (i = 0; i < 2; i++) {
g_assert(!ts.timers[i]);
g_assert(!tt.timers[i]);
}
}
@ -170,11 +177,12 @@ static void test_config_functions(void)
orig_cfg.op_size = 1;
throttle_init(&ts, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
throttle_init(&ts);
throttle_timers_init(&tt, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
/* structure reset by throttle_init previous_leak should be null */
g_assert(!ts.previous_leak);
throttle_config(&ts, &orig_cfg);
throttle_config(&ts, &tt, &orig_cfg);
/* has previous leak been initialized by throttle_config ? */
g_assert(ts.previous_leak);
@ -182,7 +190,7 @@ static void test_config_functions(void)
/* get back the fixed configuration */
throttle_get_config(&ts, &final_cfg);
throttle_destroy(&ts);
throttle_timers_destroy(&tt);
g_assert(final_cfg.buckets[THROTTLE_BPS_TOTAL].avg == 153);
g_assert(final_cfg.buckets[THROTTLE_BPS_READ].avg == 56);
@ -323,43 +331,47 @@ static void test_is_valid(void)
static void test_have_timer(void)
{
/* zero the structure */
/* zero structures */
memset(&ts, 0, sizeof(ts));
memset(&tt, 0, sizeof(tt));
/* no timer set should return false */
g_assert(!throttle_have_timer(&ts));
g_assert(!throttle_timers_are_initialized(&tt));
/* init the structure */
throttle_init(&ts, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
/* init structures */
throttle_init(&ts);
throttle_timers_init(&tt, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
/* timer set by init should return true */
g_assert(throttle_have_timer(&ts));
g_assert(throttle_timers_are_initialized(&tt));
throttle_destroy(&ts);
throttle_timers_destroy(&tt);
}
static void test_detach_attach(void)
{
/* zero the structure */
/* zero structures */
memset(&ts, 0, sizeof(ts));
memset(&tt, 0, sizeof(tt));
/* init the structure */
throttle_init(&ts, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
throttle_init(&ts);
throttle_timers_init(&tt, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
/* timer set by init should return true */
g_assert(throttle_have_timer(&ts));
g_assert(throttle_timers_are_initialized(&tt));
/* timer should no longer exist after detaching */
throttle_detach_aio_context(&ts);
g_assert(!throttle_have_timer(&ts));
throttle_timers_detach_aio_context(&tt);
g_assert(!throttle_timers_are_initialized(&tt));
/* timer should exist again after attaching */
throttle_attach_aio_context(&ts, ctx);
g_assert(throttle_have_timer(&ts));
throttle_timers_attach_aio_context(&tt, ctx);
g_assert(throttle_timers_are_initialized(&tt));
throttle_destroy(&ts);
throttle_timers_destroy(&tt);
}
static bool do_test_accounting(bool is_ops, /* are we testing bps or ops */
@ -387,9 +399,10 @@ static bool do_test_accounting(bool is_ops, /* are we testing bps or ops */
cfg.op_size = op_size;
throttle_init(&ts, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
throttle_config(&ts, &cfg);
throttle_init(&ts);
throttle_timers_init(&tt, ctx, QEMU_CLOCK_VIRTUAL,
read_timer_cb, write_timer_cb, &ts);
throttle_config(&ts, &tt, &cfg);
/* account a read */
throttle_account(&ts, false, size);
@ -414,7 +427,7 @@ static bool do_test_accounting(bool is_ops, /* are we testing bps or ops */
return false;
}
throttle_destroy(&ts);
throttle_timers_destroy(&tt);
return true;
}
@ -490,23 +503,80 @@ static void test_accounting(void)
(64.0 / 13)));
}
static void test_groups(void)
{
ThrottleConfig cfg1, cfg2;
BlockDriverState *bdrv1, *bdrv2, *bdrv3;
bdrv1 = bdrv_new();
bdrv2 = bdrv_new();
bdrv3 = bdrv_new();
g_assert(bdrv1->throttle_state == NULL);
g_assert(bdrv2->throttle_state == NULL);
g_assert(bdrv3->throttle_state == NULL);
throttle_group_register_bs(bdrv1, "bar");
throttle_group_register_bs(bdrv2, "foo");
throttle_group_register_bs(bdrv3, "bar");
g_assert(bdrv1->throttle_state != NULL);
g_assert(bdrv2->throttle_state != NULL);
g_assert(bdrv3->throttle_state != NULL);
g_assert(!strcmp(throttle_group_get_name(bdrv1), "bar"));
g_assert(!strcmp(throttle_group_get_name(bdrv2), "foo"));
g_assert(bdrv1->throttle_state == bdrv3->throttle_state);
/* Setting the config of a group member affects the whole group */
memset(&cfg1, 0, sizeof(cfg1));
cfg1.buckets[THROTTLE_BPS_READ].avg = 500000;
cfg1.buckets[THROTTLE_BPS_WRITE].avg = 285000;
cfg1.buckets[THROTTLE_OPS_READ].avg = 20000;
cfg1.buckets[THROTTLE_OPS_WRITE].avg = 12000;
throttle_group_config(bdrv1, &cfg1);
throttle_group_get_config(bdrv1, &cfg1);
throttle_group_get_config(bdrv3, &cfg2);
g_assert(!memcmp(&cfg1, &cfg2, sizeof(cfg1)));
cfg2.buckets[THROTTLE_BPS_READ].avg = 4547;
cfg2.buckets[THROTTLE_BPS_WRITE].avg = 1349;
cfg2.buckets[THROTTLE_OPS_READ].avg = 123;
cfg2.buckets[THROTTLE_OPS_WRITE].avg = 86;
throttle_group_config(bdrv3, &cfg1);
throttle_group_get_config(bdrv1, &cfg1);
throttle_group_get_config(bdrv3, &cfg2);
g_assert(!memcmp(&cfg1, &cfg2, sizeof(cfg1)));
throttle_group_unregister_bs(bdrv1);
throttle_group_unregister_bs(bdrv2);
throttle_group_unregister_bs(bdrv3);
g_assert(bdrv1->throttle_state == NULL);
g_assert(bdrv2->throttle_state == NULL);
g_assert(bdrv3->throttle_state == NULL);
}
int main(int argc, char **argv)
{
GSource *src;
Error *local_error = NULL;
init_clocks();
qemu_init_main_loop(&local_error);
ctx = qemu_get_aio_context();
ctx = aio_context_new(&local_error);
if (!ctx) {
error_report("Failed to create AIO Context: '%s'",
error_get_pretty(local_error));
error_free(local_error);
local_error ? error_get_pretty(local_error) :
"Failed to initialize the QEMU main loop");
if (local_error) {
error_free(local_error);
}
exit(1);
}
src = aio_get_g_source(ctx);
g_source_attach(src, NULL);
g_source_unref(src);
bdrv_init();
do {} while (g_main_context_iteration(NULL, false));
@ -523,6 +593,7 @@ int main(int argc, char **argv)
g_test_add_func("/throttle/config/is_valid", test_is_valid);
g_test_add_func("/throttle/config_functions", test_config_functions);
g_test_add_func("/throttle/accounting", test_accounting);
g_test_add_func("/throttle/groups", test_groups);
return g_test_run();
}

81
util/throttle.c
View File

@ -1,10 +1,12 @@
/*
* QEMU throttling infrastructure
*
* Copyright (C) Nodalink, SARL. 2013
* Copyright (C) Nodalink, EURL. 2013-2014
* Copyright (C) Igalia, S.L. 2015
*
* Author:
* Benoît Canet <benoit.canet@irqsave.net>
* Authors:
* Benoît Canet <benoit.canet@nodalink.com>
* Alberto Garcia <berto@igalia.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
@ -159,29 +161,36 @@ bool throttle_compute_timer(ThrottleState *ts,
}
/* Add timers to event loop */
void throttle_attach_aio_context(ThrottleState *ts, AioContext *new_context)
void throttle_timers_attach_aio_context(ThrottleTimers *tt,
AioContext *new_context)
{
ts->timers[0] = aio_timer_new(new_context, ts->clock_type, SCALE_NS,
ts->read_timer_cb, ts->timer_opaque);
ts->timers[1] = aio_timer_new(new_context, ts->clock_type, SCALE_NS,
ts->write_timer_cb, ts->timer_opaque);
tt->timers[0] = aio_timer_new(new_context, tt->clock_type, SCALE_NS,
tt->read_timer_cb, tt->timer_opaque);
tt->timers[1] = aio_timer_new(new_context, tt->clock_type, SCALE_NS,
tt->write_timer_cb, tt->timer_opaque);
}
/* To be called first on the ThrottleState */
void throttle_init(ThrottleState *ts,
AioContext *aio_context,
QEMUClockType clock_type,
QEMUTimerCB *read_timer_cb,
QEMUTimerCB *write_timer_cb,
void *timer_opaque)
void throttle_init(ThrottleState *ts)
{
memset(ts, 0, sizeof(ThrottleState));
}
ts->clock_type = clock_type;
ts->read_timer_cb = read_timer_cb;
ts->write_timer_cb = write_timer_cb;
ts->timer_opaque = timer_opaque;
throttle_attach_aio_context(ts, aio_context);
/* To be called first on the ThrottleTimers */
void throttle_timers_init(ThrottleTimers *tt,
AioContext *aio_context,
QEMUClockType clock_type,
QEMUTimerCB *read_timer_cb,
QEMUTimerCB *write_timer_cb,
void *timer_opaque)
{
memset(tt, 0, sizeof(ThrottleTimers));
tt->clock_type = clock_type;
tt->read_timer_cb = read_timer_cb;
tt->write_timer_cb = write_timer_cb;
tt->timer_opaque = timer_opaque;
throttle_timers_attach_aio_context(tt, aio_context);
}
/* destroy a timer */
@ -195,25 +204,25 @@ static void throttle_timer_destroy(QEMUTimer **timer)
}
/* Remove timers from event loop */
void throttle_detach_aio_context(ThrottleState *ts)
void throttle_timers_detach_aio_context(ThrottleTimers *tt)
{
int i;
for (i = 0; i < 2; i++) {
throttle_timer_destroy(&ts->timers[i]);
throttle_timer_destroy(&tt->timers[i]);
}
}
/* To be called last on the ThrottleState */
void throttle_destroy(ThrottleState *ts)
/* To be called last on the ThrottleTimers */
void throttle_timers_destroy(ThrottleTimers *tt)
{
throttle_detach_aio_context(ts);
throttle_timers_detach_aio_context(tt);
}
/* is any throttling timer configured */
bool throttle_have_timer(ThrottleState *ts)
bool throttle_timers_are_initialized(ThrottleTimers *tt)
{
if (ts->timers[0]) {
if (tt->timers[0]) {
return true;
}
@ -324,9 +333,12 @@ static void throttle_cancel_timer(QEMUTimer *timer)
/* Used to configure the throttle
*
* @ts: the throttle state we are working on
* @tt: the throttle timers we use in this aio context
* @cfg: the config to set
*/
void throttle_config(ThrottleState *ts, ThrottleConfig *cfg)
void throttle_config(ThrottleState *ts,
ThrottleTimers *tt,
ThrottleConfig *cfg)
{
int i;
@ -336,10 +348,10 @@ void throttle_config(ThrottleState *ts, ThrottleConfig *cfg)
throttle_fix_bucket(&ts->cfg.buckets[i]);
}
ts->previous_leak = qemu_clock_get_ns(ts->clock_type);
ts->previous_leak = qemu_clock_get_ns(tt->clock_type);
for (i = 0; i < 2; i++) {
throttle_cancel_timer(ts->timers[i]);
throttle_cancel_timer(tt->timers[i]);
}
}
@ -358,12 +370,15 @@ void throttle_get_config(ThrottleState *ts, ThrottleConfig *cfg)
*
* NOTE: this function is not unit tested due to it's usage of timer_mod
*
* @tt: the timers structure
* @is_write: the type of operation (read/write)
* @ret: true if the timer has been scheduled else false
*/
bool throttle_schedule_timer(ThrottleState *ts, bool is_write)
bool throttle_schedule_timer(ThrottleState *ts,
ThrottleTimers *tt,
bool is_write)
{
int64_t now = qemu_clock_get_ns(ts->clock_type);
int64_t now = qemu_clock_get_ns(tt->clock_type);
int64_t next_timestamp;
bool must_wait;
@ -378,12 +393,12 @@ bool throttle_schedule_timer(ThrottleState *ts, bool is_write)
}
/* request throttled and timer pending -> do nothing */
if (timer_pending(ts->timers[is_write])) {
if (timer_pending(tt->timers[is_write])) {
return true;
}
/* request throttled and timer not pending -> arm timer */
timer_mod(ts->timers[is_write], next_timestamp);
timer_mod(tt->timers[is_write], next_timestamp);
return true;
}