Re: [PATCH v15 4/7] io-throttle controller infrastructure

From: Andrea Righi
Date: Tue Apr 28 2009 - 10:51:21 EST


On Tue, Apr 28, 2009 at 10:43:51AM +0200, Andrea Righi wrote:
> This is the core of the io-throttle kernel infrastructure. It creates
> the basic interfaces to the cgroup subsystem and implements the I/O
> measurement and throttling functionality.

Subject: io-throttle: correctly throttle O_DIRECT reads

There's a bug in the latest io-throttle patchset: the IO generated by
O_DIRECT reads is correctly accounted, but tasks doing direct IO are not
correctly throttled.

The following fix apply the correct behaviour, throttling the tasks that
are doing O_DIRECT reads directly, instead of delaying their IO
requests.

[ This patch must be applied on top of io-throttle v15 ]

Signed-off-by: Andrea Righi <righi.andrea@xxxxxxxxx>
---
block/blk-io-throttle.c | 21 ++++++++++++---------
1 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/block/blk-io-throttle.c b/block/blk-io-throttle.c
index 380a21a..8dc2c93 100644
--- a/block/blk-io-throttle.c
+++ b/block/blk-io-throttle.c
@@ -803,12 +803,21 @@ cgroup_io_throttle(struct bio *bio, struct block_device *bdev, ssize_t bytes)
* cgroup. If they're different we're doing writeback IO and we can't
* throttle the current task directly.
*/
- if (!is_in_dio())
+ if (!is_in_dio()) {
+ /*
+ * We're not doing O_DIRECT: find the source of this IO
+ * request.
+ */
iot = get_iothrottle_from_bio(bio);
+ }
rcu_read_lock();
curr_iot = task_to_iothrottle(current);
- if (curr_iot != iot) {
- css_get(&curr_iot->css);
+ if (iot == NULL) {
+ /* IO occurs in the same context of the current task */
+ iot = curr_iot;
+ css_get(&iot->css);
+ }
+ if (iot != curr_iot) {
/*
* IO occurs in a different context of the current task
* (writeback IO).
@@ -819,10 +828,6 @@ cgroup_io_throttle(struct bio *bio, struct block_device *bdev, ssize_t bytes)
*/
can_sleep = 0;
}
- if (iot == NULL) {
- /* IO occurs in the same context of the current task */
- iot = curr_iot;
- }
/* Apply IO throttling */
iothrottle_evaluate_sleep(&s, iot, bdev, bytes);
sleep = max(s.bw_sleep, s.iops_sleep);
@@ -831,8 +836,6 @@ cgroup_io_throttle(struct bio *bio, struct block_device *bdev, ssize_t bytes)
if (unlikely(sleep && can_sleep))
iothrottle_acct_stat(iot, bdev, type, sleep);
css_put(&iot->css);
- if (curr_iot != iot)
- css_put(&curr_iot->css);
rcu_read_unlock();
if (unlikely(sleep && can_sleep)) {
/* Throttle the current task directly */
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/