@@ -1822,15 +1822,18 @@ 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];
+ 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));
- 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;
/* r/w error */
bs_dest->on_read_error = bs_src->on_read_error;
@@ -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, &bs->throttle_timers, cfg);
+ throttle_group_config(bs, cfg);
for (i = 0; i < 2; i++) {
qemu_co_enter_next(&bs->throttled_reqs[i]);
@@ -95,76 +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_timers_destroy(&bs->throttle_timers);
-}
-
-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);
- throttle_timers_init(&bs->throttle_timers,
- 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,
- &bs->throttle_timers,
- 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]);
+ /* this bs is not part of any group */
+ if (!bs->throttle_state) {
+ return;
}
- /* 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, &bs->throttle_timers,
- 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;
}
- /* else queue next request for execution */
- qemu_co_queue_next(&bs->throttled_reqs[is_write]);
+ /* 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)
@@ -971,7 +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 */
@@ -1301,7 +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);
}
/*
@@ -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;
@@ -23,6 +23,9 @@
*/
#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
@@ -160,6 +163,153 @@ static BlockDriverState *throttle_group_next_bs(BlockDriverState *bs)
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.
@@ -195,9 +345,49 @@ void throttle_group_get_config(BlockDriverState *bs, ThrottleConfig *cfg)
qemu_mutex_unlock(&tg->lock);
}
-/* Register a BlockDriverState in the throttling group, also updating
- * its throttle_state pointer to point to it. If a throttling group
- * with that name does not exist yet, it will be created.
+/* 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
@@ -206,6 +396,12 @@ 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;
@@ -218,11 +414,20 @@ void throttle_group_register_bs(BlockDriverState *bs, const char *groupname)
}
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 and setting the throttle_state pointer to NULL.
+ * list, destroying the timers and setting the throttle_state pointer
+ * to NULL.
*
* The group will be destroyed if it's empty after this operation.
*
@@ -247,6 +452,7 @@ void throttle_group_unregister_bs(BlockDriverState *bs)
/* 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);
@@ -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;
@@ -2004,16 +2015,21 @@ 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)) {
+ 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);
}
- if (bs->io_limits_enabled) {
- bdrv_set_io_limits(bs, &cfg);
- }
-
aio_context_release(aio_context);
}
@@ -3189,6 +3205,10 @@ QemuOptsList qemu_common_drive_opts = {
.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,
.help = "copy read data from backing file into image file",
@@ -1356,7 +1356,9 @@ 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);
}
@@ -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);
@@ -379,10 +379,13 @@ struct BlockDriverState {
unsigned int serialising_in_flight;
/* I/O throttling */
- ThrottleState throttle_state;
- ThrottleTimers throttle_timers;
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"). */
@@ -36,4 +36,8 @@ 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);
+
#endif
@@ -1062,6 +1062,27 @@
#
# 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 +1111,8 @@
#
# @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 +1124,7 @@
'*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:
@@ -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}[,...]]]
@@ -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: