* [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support
@ 2015-04-10 13:35 Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 1/8] throttle: Extract timers from ThrottleState into a separate structure Alberto Garcia
` (7 more replies)
0 siblings, 8 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Stefan Hajnoczi
V7:
- Make block_set_io_throttle() keep a device in the same group if the
parameter is omitted, and rewrite the code a bit to make it clearer
and slightly more efficient.
- Update the documentation of block_set_io_throttle() to explain in
detail how it works.
- Improve the iotest 093 to test scenarios with several devices in the
same group.
- Update version number to 2.4 in the 'Since:' fields.
V6: https://lists.gnu.org/archive/html/qemu-devel/2015-03/msg06036.html
- Use qemu_init_main_loop() in the throttling tests to make sure that
all BDS have an AioContext attached.
V5: https://lists.gnu.org/archive/html/qemu-devel/2015-03/msg05990.html
- Fixed and expanded the tests. There's a new one that checks that
configuration changes affect the whole throttling group.
- Refactored the code a bit: new function schedule_next_request()
- any_timer_armed is now reset in throttle_group_config(), it was
causing qemu-iotest 093 to fail.
- Documentation improvements
V4: https://lists.gnu.org/archive/html/qemu-devel/2015-03/msg05558.html
- All functions from the ThrottleGroup API now receive a BlockDriverState.
- The creation/destruction of ThrottleTimers is now handled internally.
- bdrv_io_limits_enable() no longer accepts NULL as a valid group name.
- No member of a group can access someone else's throttled_reqs queues
anymore.
- No member of a group can access someone else's timers unless they
have queued requests.
- After a timer is fired make sure that there was actually a request
in the queue, and schedule a new one otherwise.
- Protect bdrv_swap() by locking the throttling group.
- throttle_group_co_io_limits_intercept() is now marked as a coroutine
function.
- Documentation updates.
Regards,
Berto
Alberto Garcia (7):
throttle: Add throttle group infrastructure
throttle: Add throttle group infrastructure tests
throttle: Add throttle group support
throttle: acquire the ThrottleGroup lock in bdrv_swap()
throttle: add the name of the ThrottleGroup to BlockDeviceInfo
throttle: Update throttle infrastructure copyright
qemu-iotests: expand test 093 to support group throttling
Benoît Canet (1):
throttle: Extract timers from ThrottleState into a separate structure
block.c | 101 ++++-----
block/Makefile.objs | 1 +
block/qapi.c | 8 +-
block/throttle-groups.c | 489 ++++++++++++++++++++++++++++++++++++++++
blockdev.c | 38 +++-
hmp.c | 10 +-
include/block/block.h | 3 +-
include/block/block_int.h | 7 +-
include/block/throttle-groups.h | 46 ++++
include/qemu/throttle.h | 46 ++--
qapi/block-core.json | 29 ++-
qemu-options.hx | 1 +
qmp-commands.hx | 3 +-
tests/qemu-iotests/093 | 93 +++++---
tests/test-throttle.c | 161 +++++++++----
util/throttle.c | 81 ++++---
16 files changed, 919 insertions(+), 198 deletions(-)
create mode 100644 block/throttle-groups.c
create mode 100644 include/block/throttle-groups.h
--
2.1.4
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 1/8] throttle: Extract timers from ThrottleState into a separate structure
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
@ 2015-04-10 13:35 ` Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 2/8] throttle: Add throttle group infrastructure Alberto Garcia
` (6 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Benoît Canet, Stefan Hajnoczi
From: Benoît Canet <benoit.canet@nodalink.com>
Group throttling will share ThrottleState between multiple bs.
As a consequence the ThrottleState will be accessed by multiple aio
context.
Timers are tied to their aio context so they must go out of the
ThrottleState structure.
This commit paves the way for each bs of a common ThrottleState to
have its own timer.
Signed-off-by: Benoit Canet <benoit.canet@nodalink.com>
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
block.c | 35 ++++++++++++--------
include/block/block_int.h | 1 +
include/qemu/throttle.h | 38 ++++++++++++++--------
tests/test-throttle.c | 82 ++++++++++++++++++++++++++---------------------
util/throttle.c | 73 ++++++++++++++++++++++++-----------------
5 files changed, 135 insertions(+), 94 deletions(-)
diff --git a/block.c b/block.c
index f2f8ae7..28508ef 100644
--- a/block.c
+++ b/block.c
@@ -130,7 +130,7 @@ void bdrv_set_io_limits(BlockDriverState *bs,
{
int i;
- throttle_config(&bs->throttle_state, cfg);
+ throttle_config(&bs->throttle_state, &bs->throttle_timers, cfg);
for (i = 0; i < 2; i++) {
qemu_co_enter_next(&bs->throttled_reqs[i]);
@@ -163,7 +163,7 @@ void bdrv_io_limits_disable(BlockDriverState *bs)
bdrv_start_throttled_reqs(bs);
- throttle_destroy(&bs->throttle_state);
+ throttle_timers_destroy(&bs->throttle_timers);
}
static void bdrv_throttle_read_timer_cb(void *opaque)
@@ -182,12 +182,13 @@ static void bdrv_throttle_write_timer_cb(void *opaque)
void bdrv_io_limits_enable(BlockDriverState *bs)
{
assert(!bs->io_limits_enabled);
- throttle_init(&bs->throttle_state,
- bdrv_get_aio_context(bs),
- QEMU_CLOCK_VIRTUAL,
- bdrv_throttle_read_timer_cb,
- bdrv_throttle_write_timer_cb,
- bs);
+ throttle_init(&bs->throttle_state);
+ throttle_timers_init(&bs->throttle_timers,
+ bdrv_get_aio_context(bs),
+ QEMU_CLOCK_VIRTUAL,
+ bdrv_throttle_read_timer_cb,
+ bdrv_throttle_write_timer_cb,
+ bs);
bs->io_limits_enabled = true;
}
@@ -201,7 +202,9 @@ static void bdrv_io_limits_intercept(BlockDriverState *bs,
bool is_write)
{
/* does this io must wait */
- bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
+ 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 ||
@@ -214,7 +217,8 @@ static void bdrv_io_limits_intercept(BlockDriverState *bs,
/* if the next request must wait -> do nothing */
- if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
+ if (throttle_schedule_timer(&bs->throttle_state, &bs->throttle_timers,
+ is_write)) {
return;
}
@@ -2090,6 +2094,9 @@ static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
memcpy(&bs_dest->throttle_state,
&bs_src->throttle_state,
sizeof(ThrottleState));
+ 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;
@@ -2151,7 +2158,7 @@ void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
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(!throttle_timers_are_initialized(&bs_new->throttle_timers));
tmp = *bs_new;
*bs_new = *bs_old;
@@ -2168,7 +2175,7 @@ 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(!throttle_timers_are_initialized(&bs_new->throttle_timers));
/* insert the nodes back into the graph node list if needed */
if (bs_new->node_name[0] != '\0') {
@@ -5858,7 +5865,7 @@ 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);
@@ -5894,7 +5901,7 @@ 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) {
diff --git a/include/block/block_int.h b/include/block/block_int.h
index dccb092..c88a062 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -377,6 +377,7 @@ struct BlockDriverState {
/* I/O throttling */
ThrottleState throttle_state;
+ ThrottleTimers throttle_timers;
CoQueue throttled_reqs[2];
bool io_limits_enabled;
diff --git a/include/qemu/throttle.h b/include/qemu/throttle.h
index b890613..2c560db 100644
--- a/include/qemu/throttle.h
+++ b/include/qemu/throttle.h
@@ -65,14 +65,17 @@ 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 +89,23 @@ 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_timers_init(ThrottleTimers *tt,
+ AioContext *aio_context,
+ QEMUClockType clock_type,
+ QEMUTimerCB *read_timer_cb,
+ QEMUTimerCB *write_timer_cb,
+ void *timer_opaque);
-void throttle_destroy(ThrottleState *ts);
+void throttle_timers_destroy(ThrottleTimers *tt);
-void throttle_detach_aio_context(ThrottleState *ts);
+void throttle_timers_detach_aio_context(ThrottleTimers *tt);
-void throttle_attach_aio_context(ThrottleState *ts, AioContext *new_context);
+void throttle_timers_attach_aio_context(ThrottleTimers *tt,
+ AioContext *new_context);
-bool throttle_have_timer(ThrottleState *ts);
+bool throttle_timers_are_initialized(ThrottleTimers *tt);
/* configuration */
bool throttle_enabled(ThrottleConfig *cfg);
@@ -108,12 +114,16 @@ 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);
diff --git a/tests/test-throttle.c b/tests/test-throttle.c
index d8ba415..458f577 100644
--- a/tests/test-throttle.c
+++ b/tests/test-throttle.c
@@ -20,6 +20,7 @@ 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 +104,19 @@ 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 +127,18 @@ 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 +174,12 @@ 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 +187,7 @@ 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 +328,47 @@ 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 +396,10 @@ 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 +424,7 @@ 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;
}
diff --git a/util/throttle.c b/util/throttle.c
index f976ac7..d76a48e 100644
--- a/util/throttle.c
+++ b/util/throttle.c
@@ -159,29 +159,36 @@ 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));
+}
+
+/* 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));
- 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);
+ 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 +202,25 @@ 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 +331,12 @@ 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 +346,10 @@ 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 +368,15 @@ 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 +391,12 @@ 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;
}
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 2/8] throttle: Add throttle group infrastructure
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 1/8] throttle: Extract timers from ThrottleState into a separate structure Alberto Garcia
@ 2015-04-10 13:35 ` Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 3/8] throttle: Add throttle group infrastructure tests Alberto Garcia
` (5 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Stefan Hajnoczi
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
block/Makefile.objs | 1 +
block/throttle-groups.c | 261 ++++++++++++++++++++++++++++++++++++++++
include/block/block_int.h | 1 +
include/block/throttle-groups.h | 39 ++++++
4 files changed, 302 insertions(+)
create mode 100644 block/throttle-groups.c
create mode 100644 include/block/throttle-groups.h
diff --git a/block/Makefile.objs b/block/Makefile.objs
index db2933e..1e61ce0 100644
--- a/block/Makefile.objs
+++ b/block/Makefile.objs
@@ -10,6 +10,7 @@ block-obj-$(CONFIG_WIN32) += raw-win32.o win32-aio.o
block-obj-$(CONFIG_POSIX) += raw-posix.o
block-obj-$(CONFIG_LINUX_AIO) += linux-aio.o
block-obj-y += null.o mirror.o
+block-obj-y += throttle-groups.o
block-obj-y += nbd.o nbd-client.o sheepdog.o
block-obj-$(CONFIG_LIBISCSI) += iscsi.o
diff --git a/block/throttle-groups.c b/block/throttle-groups.c
new file mode 100644
index 0000000..352077f
--- /dev/null
+++ b/block/throttle-groups.c
@@ -0,0 +1,261 @@
+/*
+ * 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"
+
+/* 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
+ * transparent to outside users.
+ *
+ * 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;
+}
+
+/* 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);
+}
+
+/* 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.
+ *
+ * @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);
+
+ 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);
+ qemu_mutex_unlock(&tg->lock);
+}
+
+/* Unregister a BlockDriverState from its group, removing it from the
+ * list 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);
+ qemu_mutex_unlock(&tg->lock);
+
+ throttle_group_unref(tg);
+ bs->throttle_state = NULL;
+}
+
+static void throttle_groups_init(void)
+{
+ qemu_mutex_init(&throttle_groups_lock);
+}
+
+block_init(throttle_groups_init);
diff --git a/include/block/block_int.h b/include/block/block_int.h
index c88a062..f0896e9 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -380,6 +380,7 @@ struct BlockDriverState {
ThrottleTimers throttle_timers;
CoQueue throttled_reqs[2];
bool io_limits_enabled;
+ QLIST_ENTRY(BlockDriverState) round_robin;
/* I/O stats (display with "info blockstats"). */
BlockAcctStats stats;
diff --git a/include/block/throttle-groups.h b/include/block/throttle-groups.h
new file mode 100644
index 0000000..b966ec7
--- /dev/null
+++ b/include/block/throttle-groups.h
@@ -0,0 +1,39 @@
+/*
+ * 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);
+
+#endif
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 3/8] throttle: Add throttle group infrastructure tests
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 1/8] throttle: Extract timers from ThrottleState into a separate structure Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 2/8] throttle: Add throttle group infrastructure Alberto Garcia
@ 2015-04-10 13:35 ` Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 4/8] throttle: Add throttle group support Alberto Garcia
` (4 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Stefan Hajnoczi
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
tests/test-throttle.c | 79 ++++++++++++++++++++++++++++++++++++++++++++-------
1 file changed, 69 insertions(+), 10 deletions(-)
diff --git a/tests/test-throttle.c b/tests/test-throttle.c
index 458f577..aa0e297 100644
--- a/tests/test-throttle.c
+++ b/tests/test-throttle.c
@@ -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.
@@ -15,6 +17,7 @@
#include "block/aio.h"
#include "qemu/throttle.h"
#include "qemu/error-report.h"
+#include "block/throttle-groups.h"
static AioContext *ctx;
static LeakyBucket bkt;
@@ -500,23 +503,78 @@ 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);
do {} while (g_main_context_iteration(NULL, false));
@@ -533,6 +591,7 @@ 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();
}
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 4/8] throttle: Add throttle group support
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
` (2 preceding siblings ...)
2015-04-10 13:35 ` [Qemu-devel] [PATCH 3/8] throttle: Add throttle group infrastructure tests Alberto Garcia
@ 2015-04-10 13:35 ` Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 5/8] throttle: acquire the ThrottleGroup lock in bdrv_swap() Alberto Garcia
` (3 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Stefan Hajnoczi
The throttle group support use a cooperative round robin scheduling
algorithm.
The principles of the algorithm are simple:
- Each BDS of the group is used as a token in a circular way.
- The active BDS computes if a wait must be done and arms the right
timer.
- If a wait must be done the token timer will be armed so the token
will become the next active BDS.
Signed-off-by: Alberto Garcia <berto@igalia.com>
---
block.c | 83 +++++-----------
block/qapi.c | 5 +-
block/throttle-groups.c | 207 +++++++++++++++++++++++++++++++++++++++-
blockdev.c | 38 ++++++--
hmp.c | 4 +-
include/block/block.h | 3 +-
include/block/block_int.h | 7 +-
include/block/throttle-groups.h | 4 +
qapi/block-core.json | 25 ++++-
qemu-options.hx | 1 +
qmp-commands.hx | 3 +-
11 files changed, 302 insertions(+), 78 deletions(-)
diff --git a/block.c b/block.c
index 28508ef..5ca972f 100644
--- a/block.c
+++ b/block.c
@@ -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>
@@ -130,7 +131,7 @@ 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]);
@@ -160,70 +161,33 @@ 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)
{
assert(!bs->io_limits_enabled);
- throttle_init(&bs->throttle_state);
- throttle_timers_init(&bs->throttle_timers,
- bdrv_get_aio_context(bs),
- QEMU_CLOCK_VIRTUAL,
- 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);
}
size_t bdrv_opt_mem_align(BlockDriverState *bs)
@@ -2091,15 +2055,18 @@ 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;
@@ -3170,7 +3137,7 @@ 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 */
@@ -3411,7 +3378,7 @@ 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);
}
/*
diff --git a/block/qapi.c b/block/qapi.c
index 8a19aed..65af057 100644
--- a/block/qapi.c
+++ b/block/qapi.c
@@ -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"
@@ -63,7 +64,9 @@ BlockDeviceInfo *bdrv_block_device_info(BlockDriverState *bs)
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;
diff --git a/block/throttle-groups.c b/block/throttle-groups.c
index 352077f..f909f50 100644
--- a/block/throttle-groups.c
+++ b/block/throttle-groups.c
@@ -23,6 +23,8 @@
*/
#include "block/throttle-groups.h"
+#include "qemu/queue.h"
+#include "qemu/thread.h"
/* The ThrottleGroup structure (with its ThrottleState) is shared
* among different BlockDriverState and it's independent from
@@ -160,6 +162,153 @@ 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 +344,49 @@ 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
@@ -218,11 +407,20 @@ 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),
+ QEMU_CLOCK_VIRTUAL,
+ 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 +445,7 @@ 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);
diff --git a/blockdev.c b/blockdev.c
index fbb3a79..a712479 100644
--- a/blockdev.c
+++ b/blockdev.c
@@ -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" },
};
@@ -1887,7 +1896,9 @@ 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;
@@ -1940,14 +1951,19 @@ 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);
@@ -3017,6 +3033,10 @@ 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",
diff --git a/hmp.c b/hmp.c
index f31ae27..50f30f2 100644
--- a/hmp.c
+++ b/hmp.c
@@ -1272,7 +1272,9 @@ 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);
}
diff --git a/include/block/block.h b/include/block/block.h
index 4c57d63..e0d53f3 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -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);
diff --git a/include/block/block_int.h b/include/block/block_int.h
index f0896e9..4e09825 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -376,10 +376,13 @@ 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"). */
diff --git a/include/block/throttle-groups.h b/include/block/throttle-groups.h
index b966ec7..322139a 100644
--- a/include/block/throttle-groups.h
+++ b/include/block/throttle-groups.h
@@ -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
diff --git a/qapi/block-core.json b/qapi/block-core.json
index 7873084..dce44fe 100644
--- a/qapi/block-core.json
+++ b/qapi/block-core.json
@@ -962,6 +962,27 @@
#
# 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
@@ -990,6 +1011,8 @@
#
# @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
#
@@ -1001,7 +1024,7 @@
'*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:
diff --git a/qemu-options.hx b/qemu-options.hx
index 319d971..7d6b2ec 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -464,6 +464,7 @@ 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}[,...]]]
diff --git a/qmp-commands.hx b/qmp-commands.hx
index 3a42ad0..ce897ff 100644
--- a/qmp-commands.hx
+++ b/qmp-commands.hx
@@ -1730,7 +1730,7 @@ 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,
},
@@ -1756,6 +1756,7 @@ 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:
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 5/8] throttle: acquire the ThrottleGroup lock in bdrv_swap()
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
` (3 preceding siblings ...)
2015-04-10 13:35 ` [Qemu-devel] [PATCH 4/8] throttle: Add throttle group support Alberto Garcia
@ 2015-04-10 13:35 ` Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 6/8] throttle: add the name of the ThrottleGroup to BlockDeviceInfo Alberto Garcia
` (2 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Stefan Hajnoczi
bdrv_swap() touches the fields of a BlockDriverState that are
protected by the ThrottleGroup lock. Although those fields end up in
their original place, they are temporarily swapped in the process,
so there's a chance that an operation on a member of the same group
happening on a different thread can try to use them.
Signed-off-by: Alberto Garcia <berto@igalia.com>
---
block.c | 15 +++++++++++++++
block/throttle-groups.c | 31 ++++++++++++++++++++++++++++++-
include/block/throttle-groups.h | 3 +++
3 files changed, 48 insertions(+), 1 deletion(-)
diff --git a/block.c b/block.c
index 5ca972f..5643082 100644
--- a/block.c
+++ b/block.c
@@ -2120,11 +2120,20 @@ 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(bs_new->throttle_state == NULL);
assert(!throttle_timers_are_initialized(&bs_new->throttle_timers));
tmp = *bs_new;
@@ -2142,8 +2151,14 @@ 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(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') {
QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_new, node_list);
diff --git a/block/throttle-groups.c b/block/throttle-groups.c
index f909f50..20a510a 100644
--- a/block/throttle-groups.c
+++ b/block/throttle-groups.c
@@ -32,7 +32,8 @@
* its own locking.
*
* This locking is however handled internally in this file, so it's
- * transparent to outside users.
+ * 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.
@@ -452,6 +453,34 @@ void throttle_group_unregister_bs(BlockDriverState *bs)
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);
diff --git a/include/block/throttle-groups.h b/include/block/throttle-groups.h
index 322139a..fab113f 100644
--- a/include/block/throttle-groups.h
+++ b/include/block/throttle-groups.h
@@ -40,4 +40,7 @@ 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
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 6/8] throttle: add the name of the ThrottleGroup to BlockDeviceInfo
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
` (4 preceding siblings ...)
2015-04-10 13:35 ` [Qemu-devel] [PATCH 5/8] throttle: acquire the ThrottleGroup lock in bdrv_swap() Alberto Garcia
@ 2015-04-10 13:35 ` Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 7/8] throttle: Update throttle infrastructure copyright Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 8/8] qemu-iotests: expand test 093 to support group throttling Alberto Garcia
7 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Stefan Hajnoczi
Signed-off-by: Alberto Garcia <berto@igalia.com>
---
block/qapi.c | 3 +++
hmp.c | 6 ++++--
qapi/block-core.json | 4 +++-
3 files changed, 10 insertions(+), 3 deletions(-)
diff --git a/block/qapi.c b/block/qapi.c
index 65af057..ee823d1 100644
--- a/block/qapi.c
+++ b/block/qapi.c
@@ -91,6 +91,9 @@ BlockDeviceInfo *bdrv_block_device_info(BlockDriverState *bs)
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);
diff --git a/hmp.c b/hmp.c
index 50f30f2..759fb86 100644
--- a/hmp.c
+++ b/hmp.c
@@ -375,7 +375,8 @@ 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,
@@ -388,7 +389,8 @@ 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) {
diff --git a/qapi/block-core.json b/qapi/block-core.json
index dce44fe..4de8304 100644
--- a/qapi/block-core.json
+++ b/qapi/block-core.json
@@ -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' } }
##
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 7/8] throttle: Update throttle infrastructure copyright
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
` (5 preceding siblings ...)
2015-04-10 13:35 ` [Qemu-devel] [PATCH 6/8] throttle: add the name of the ThrottleGroup to BlockDeviceInfo Alberto Garcia
@ 2015-04-10 13:35 ` Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 8/8] qemu-iotests: expand test 093 to support group throttling Alberto Garcia
7 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Stefan Hajnoczi
Signed-off-by: Alberto Garcia <berto@igalia.com>
---
include/qemu/throttle.h | 8 +++++---
util/throttle.c | 8 +++++---
2 files changed, 10 insertions(+), 6 deletions(-)
diff --git a/include/qemu/throttle.h b/include/qemu/throttle.h
index 2c560db..5af76f0 100644
--- a/include/qemu/throttle.h
+++ b/include/qemu/throttle.h
@@ -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
diff --git a/util/throttle.c b/util/throttle.c
index d76a48e..706c131 100644
--- a/util/throttle.c
+++ b/util/throttle.c
@@ -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
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 8/8] qemu-iotests: expand test 093 to support group throttling
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
` (6 preceding siblings ...)
2015-04-10 13:35 ` [Qemu-devel] [PATCH 7/8] throttle: Update throttle infrastructure copyright Alberto Garcia
@ 2015-04-10 13:35 ` Alberto Garcia
7 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-04-10 13:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Kevin Wolf, Alberto Garcia, Stefan Hajnoczi
This patch improves the test by attaching a different number of drives
to the VM and putting them in the same throttling group. The test
verifies that the I/O is evenly distributed among all members of the
group, and that the limits are enforced.
By default the test is repeated 3 times with 1, 2 and 3 drives, but
the maximum number of simultaneous drives is configurable.
Signed-off-by: Alberto Garcia <berto@igalia.com>
---
tests/qemu-iotests/093 | 93 +++++++++++++++++++++++++++++++++++---------------
1 file changed, 65 insertions(+), 28 deletions(-)
diff --git a/tests/qemu-iotests/093 b/tests/qemu-iotests/093
index b9096a5..c0e9e2b 100755
--- a/tests/qemu-iotests/093
+++ b/tests/qemu-iotests/093
@@ -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
+ wr_nr /= ndrives
+
+ # Send I/O requests to all drives
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))
+ for drive in range(0, ndrives):
+ self.vm.hmp_qemu_io("drive%d" % drive, "aio_read %d %d" %
+ (i * rq_size, rq_size))
- start_rd_bytes, start_rd_iops, start_wr_bytes, start_wr_iops = self.blockstats('drive0')
+ 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
- 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))
+ # 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)
+
+ # 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://"
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [Qemu-devel] [PATCH 1/8] throttle: Extract timers from ThrottleState into a separate structure
2015-05-19 12:24 [Qemu-devel] [PATCH v8 0/8] Block Throttle Group Support Alberto Garcia
@ 2015-05-19 12:24 ` Alberto Garcia
0 siblings, 0 replies; 10+ messages in thread
From: Alberto Garcia @ 2015-05-19 12:24 UTC (permalink / raw)
To: qemu-devel
Cc: Kevin Wolf, Alberto Garcia, qemu-block, Stefan Hajnoczi,
Benoît Canet
From: Benoît Canet <benoit.canet@nodalink.com>
Group throttling will share ThrottleState between multiple bs.
As a consequence the ThrottleState will be accessed by multiple aio
context.
Timers are tied to their aio context so they must go out of the
ThrottleState structure.
This commit paves the way for each bs of a common ThrottleState to
have its own timer.
Signed-off-by: Benoit Canet <benoit.canet@nodalink.com>
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
block.c | 11 ++++---
block/io.c | 24 ++++++++------
include/block/block_int.h | 1 +
include/qemu/throttle.h | 38 ++++++++++++++--------
tests/test-throttle.c | 82 ++++++++++++++++++++++++++---------------------
util/throttle.c | 73 ++++++++++++++++++++++++-----------------
6 files changed, 135 insertions(+), 94 deletions(-)
diff --git a/block.c b/block.c
index 7904098..e41ba10 100644
--- a/block.c
+++ b/block.c
@@ -1814,6 +1814,9 @@ static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
memcpy(&bs_dest->throttle_state,
&bs_src->throttle_state,
sizeof(ThrottleState));
+ 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;
@@ -1875,7 +1878,7 @@ void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
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(!throttle_timers_are_initialized(&bs_new->throttle_timers));
tmp = *bs_new;
*bs_new = *bs_old;
@@ -1892,7 +1895,7 @@ 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(!throttle_timers_are_initialized(&bs_new->throttle_timers));
/* insert the nodes back into the graph node list if needed */
if (bs_new->node_name[0] != '\0') {
@@ -3668,7 +3671,7 @@ 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);
@@ -3704,7 +3707,7 @@ 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) {
diff --git a/block/io.c b/block/io.c
index 1ce62c4..1f20c9c 100644
--- a/block/io.c
+++ b/block/io.c
@@ -65,7 +65,7 @@ void bdrv_set_io_limits(BlockDriverState *bs,
{
int i;
- throttle_config(&bs->throttle_state, cfg);
+ throttle_config(&bs->throttle_state, &bs->throttle_timers, cfg);
for (i = 0; i < 2; i++) {
qemu_co_enter_next(&bs->throttled_reqs[i]);
@@ -98,7 +98,7 @@ void bdrv_io_limits_disable(BlockDriverState *bs)
bdrv_start_throttled_reqs(bs);
- throttle_destroy(&bs->throttle_state);
+ throttle_timers_destroy(&bs->throttle_timers);
}
static void bdrv_throttle_read_timer_cb(void *opaque)
@@ -123,12 +123,13 @@ void bdrv_io_limits_enable(BlockDriverState *bs)
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_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);
bs->io_limits_enabled = true;
}
@@ -142,7 +143,9 @@ static void bdrv_io_limits_intercept(BlockDriverState *bs,
bool is_write)
{
/* does this io must wait */
- bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
+ 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 ||
@@ -155,7 +158,8 @@ static void bdrv_io_limits_intercept(BlockDriverState *bs,
/* if the next request must wait -> do nothing */
- if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
+ if (throttle_schedule_timer(&bs->throttle_state, &bs->throttle_timers,
+ is_write)) {
return;
}
diff --git a/include/block/block_int.h b/include/block/block_int.h
index db29b74..b05ffe2 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -377,6 +377,7 @@ struct BlockDriverState {
/* I/O throttling */
ThrottleState throttle_state;
+ ThrottleTimers throttle_timers;
CoQueue throttled_reqs[2];
bool io_limits_enabled;
diff --git a/include/qemu/throttle.h b/include/qemu/throttle.h
index b890613..2c560db 100644
--- a/include/qemu/throttle.h
+++ b/include/qemu/throttle.h
@@ -65,14 +65,17 @@ 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 +89,23 @@ 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_timers_init(ThrottleTimers *tt,
+ AioContext *aio_context,
+ QEMUClockType clock_type,
+ QEMUTimerCB *read_timer_cb,
+ QEMUTimerCB *write_timer_cb,
+ void *timer_opaque);
-void throttle_destroy(ThrottleState *ts);
+void throttle_timers_destroy(ThrottleTimers *tt);
-void throttle_detach_aio_context(ThrottleState *ts);
+void throttle_timers_detach_aio_context(ThrottleTimers *tt);
-void throttle_attach_aio_context(ThrottleState *ts, AioContext *new_context);
+void throttle_timers_attach_aio_context(ThrottleTimers *tt,
+ AioContext *new_context);
-bool throttle_have_timer(ThrottleState *ts);
+bool throttle_timers_are_initialized(ThrottleTimers *tt);
/* configuration */
bool throttle_enabled(ThrottleConfig *cfg);
@@ -108,12 +114,16 @@ 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);
diff --git a/tests/test-throttle.c b/tests/test-throttle.c
index d8ba415..458f577 100644
--- a/tests/test-throttle.c
+++ b/tests/test-throttle.c
@@ -20,6 +20,7 @@ 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 +104,19 @@ 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 +127,18 @@ 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 +174,12 @@ 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 +187,7 @@ 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 +328,47 @@ 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 +396,10 @@ 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 +424,7 @@ 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;
}
diff --git a/util/throttle.c b/util/throttle.c
index f976ac7..d76a48e 100644
--- a/util/throttle.c
+++ b/util/throttle.c
@@ -159,29 +159,36 @@ 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));
+}
+
+/* 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));
- 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);
+ 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 +202,25 @@ 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 +331,12 @@ 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 +346,10 @@ 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 +368,15 @@ 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 +391,12 @@ 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;
}
--
2.1.4
^ permalink raw reply related [flat|nested] 10+ messages in thread
end of thread, other threads:[~2015-05-19 12:25 UTC | newest]
Thread overview: 10+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-04-10 13:35 [Qemu-devel] [PATCH v7 0/8] Block Throttle Group Support Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 1/8] throttle: Extract timers from ThrottleState into a separate structure Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 2/8] throttle: Add throttle group infrastructure Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 3/8] throttle: Add throttle group infrastructure tests Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 4/8] throttle: Add throttle group support Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 5/8] throttle: acquire the ThrottleGroup lock in bdrv_swap() Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 6/8] throttle: add the name of the ThrottleGroup to BlockDeviceInfo Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 7/8] throttle: Update throttle infrastructure copyright Alberto Garcia
2015-04-10 13:35 ` [Qemu-devel] [PATCH 8/8] qemu-iotests: expand test 093 to support group throttling Alberto Garcia
2015-05-19 12:24 [Qemu-devel] [PATCH v8 0/8] Block Throttle Group Support Alberto Garcia
2015-05-19 12:24 ` [Qemu-devel] [PATCH 1/8] throttle: Extract timers from ThrottleState into a separate structure Alberto Garcia
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.