Messages in this thread Patch in this message | | | From | Artem Savkov <> | Subject | [PATCH v3 1/2] timer: add a function to adjust timeouts to be upper bound | Date | Wed, 30 Mar 2022 10:20:45 +0200 |
| |
Current timer wheel implementation is optimized for performance and energy usage but lacks in precision. This, normally, is not a problem as most timers that use timer wheel are used for timeouts and thus rarely expire, instead they often get canceled or modified before expiration. Even when they don't, expiring a bit late is not an issue for timeout timers.
TCP keepalive timer is a special case, it's aim is to prevent timeouts, so triggering earlier rather than later is desired behavior. In a reported case the user had a 3600s keepalive timer for preventing firewall disconnects (on a 3650s interval). They observed keepalive timers coming in up to four minutes late, causing unexpected disconnects.
This commit adds upper_bound_timeout() function that takes a relative timeout and adjusts it based on timer wheel granularity so that supplied value effectively becomes an upper bound for the timer.
This was previously discussed here: https://lore.kernel.org/all/20210302001054.4qgrvnkltvkgikzr@treble/T/#u
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com> Signed-off-by: Artem Savkov <asavkov@redhat.com> --- include/linux/timer.h | 1 + kernel/time/timer.c | 92 +++++++++++++++++++++++++++++++++++++------ 2 files changed, 80 insertions(+), 13 deletions(-)
diff --git a/include/linux/timer.h b/include/linux/timer.h index fda13c9d1256c..b209d31d543f0 100644 --- a/include/linux/timer.h +++ b/include/linux/timer.h @@ -168,6 +168,7 @@ static inline int timer_pending(const struct timer_list * timer) return !hlist_unhashed_lockless(&timer->entry); } +extern unsigned long upper_bound_timeout(unsigned long timeout); extern void add_timer_on(struct timer_list *timer, int cpu); extern int del_timer(struct timer_list * timer); extern int mod_timer(struct timer_list *timer, unsigned long expires); diff --git a/kernel/time/timer.c b/kernel/time/timer.c index 85f1021ad4595..76d4f26c991be 100644 --- a/kernel/time/timer.c +++ b/kernel/time/timer.c @@ -507,28 +507,38 @@ static inline unsigned calc_index(unsigned long expires, unsigned lvl, return LVL_OFFS(lvl) + (expires & LVL_MASK); } -static int calc_wheel_index(unsigned long expires, unsigned long clk, - unsigned long *bucket_expiry) +static inline int get_wheel_lvl(unsigned long delta) { - unsigned long delta = expires - clk; - unsigned int idx; - if (delta < LVL_START(1)) { - idx = calc_index(expires, 0, bucket_expiry); + return 0; } else if (delta < LVL_START(2)) { - idx = calc_index(expires, 1, bucket_expiry); + return 1; } else if (delta < LVL_START(3)) { - idx = calc_index(expires, 2, bucket_expiry); + return 2; } else if (delta < LVL_START(4)) { - idx = calc_index(expires, 3, bucket_expiry); + return 3; } else if (delta < LVL_START(5)) { - idx = calc_index(expires, 4, bucket_expiry); + return 4; } else if (delta < LVL_START(6)) { - idx = calc_index(expires, 5, bucket_expiry); + return 5; } else if (delta < LVL_START(7)) { - idx = calc_index(expires, 6, bucket_expiry); + return 6; } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) { - idx = calc_index(expires, 7, bucket_expiry); + return 7; + } + + return -1; +} + +static int calc_wheel_index(unsigned long expires, unsigned long clk, + unsigned long *bucket_expiry) +{ + unsigned long delta = expires - clk; + unsigned int idx; + int lvl = get_wheel_lvl(delta); + + if (lvl >= 0) { + idx = calc_index(expires, lvl, bucket_expiry); } else if ((long) delta < 0) { idx = clk & LVL_MASK; *bucket_expiry = clk; @@ -545,6 +555,62 @@ static int calc_wheel_index(unsigned long expires, unsigned long clk, return idx; } +/** + * upper_bound_timeout - return granularity-adjusted timeout + * @timeout: timeout value in jiffies + * + * This function return supplied timeout adjusted based on timer wheel + * granularity effectively making supplied value an upper bound at which the + * timer will expire. + */ +unsigned long upper_bound_timeout(unsigned long timeout) +{ + int lvl = get_wheel_lvl(timeout); + + if (lvl < 0) { + if ((long) timeout < 0) { + /* + * This will expire immediately so no adjustment + * needed. + */ + return timeout; + } else { + if (timeout > WHEEL_TIMEOUT_CUTOFF) + timeout = WHEEL_TIMEOUT_CUTOFF; + lvl = LVL_DEPTH - 1; + } + } + + if (timeout - LVL_GRAN(lvl) < LVL_START(lvl)) { + /* + * Beginning of each level is a special case, we can't just + * subtract LVL_GRAN(lvl) because then timeout ends up on a + * previous level and is guaranteed to fire off early. We can't + * mitigate this completely, but we can try to minimize the + * margin. + */ + if (timeout - LVL_GRAN(lvl - 1) < LVL_START(lvl)) { + /* + * If timeout is within previous level's granularity + * adjust timeout using that level's granularity. + */ + return timeout - LVL_GRAN(lvl - 1); + } else { + /* + * If LVL_GRAN(lvl - 1) < timeout < LVL_GRAN(lvl) + * the best we can do is to set timeout to the end of + * previous level. This means that the timer will + * trigger early. In worst case it will be _at least_ + * (LVL_GRAN(lvl) - LVL_GRAN(lvl -1)) jiffies early. + */ + return LVL_START(lvl) - 1; + } + } else { + return timeout - LVL_GRAN(lvl); + } +} +EXPORT_SYMBOL(upper_bound_timeout); + static void trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer) { -- 2.34.1
| |