From 0d6526f82c3cdefcca47f73f5fc08dc6f335eac6 Mon Sep 17 00:00:00 2001 From: Tim Chen Date: Mon, 21 Sep 2026 17:37:22 -0700 Subject: sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling bug alb_break_llc() decides whether to break LLC preference during active load balance. It does so by testing that every runnable fair task on the source rq prefers its LLC: env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable But the two counters cover different sets. nr_pref_llc_running is updated in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued, so it follows queued tasks. h_nr_runnable is updated in set_delayed()/ clear_delayed() and drops delay-dequeued tasks. So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks, alb_break_llc() returns false, and active balance is free to pull a task off its preferred LLC. Active balance only moves runnable tasks, and this is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB skips the per-task test in can_migrate_task(). The runnable set is the one we want. Fix it on the counter side. A task should be counted in nr_pref_llc_running exactly while it is both queued on its preferred LLC (pref_llc_queued) and runnable (!sched_delayed). Define that membership once in task_pref_llc_runnable(), and adjust the counter only through pref_llc_running_inc()/pref_llc_running_dec() from the four sites that change either input: account_llc_enqueue(), account_llc_dequeue(), set_delayed() and clear_delayed(). Gating every update on the same predicate keeps the delay, wake and dequeue paths from double-counting or underflowing; see the comments at those sites for the ordering. nr_llc_running and sd->llc_counts are not touched and stay on queued semantics. Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC") Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/ Reported-by: Zhan Xusheng Suggested-by: Chen Yu Signed-off-by: Tim Chen Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Kayra Cizmeci Cc: # v7.2.x Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com --- include/net/dst_cache.h | 109 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 109 insertions(+) create mode 100644 include/net/dst_cache.h (limited to 'include/net/dst_cache.h') diff --git a/include/net/dst_cache.h b/include/net/dst_cache.h new file mode 100644 index 000000000..196169959 --- /dev/null +++ b/include/net/dst_cache.h @@ -0,0 +1,109 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _NET_DST_CACHE_H +#define _NET_DST_CACHE_H + +#include +#include +#if IS_ENABLED(CONFIG_IPV6) +#include +#endif + +struct dst_cache { + struct dst_cache_pcpu __percpu *cache; + unsigned long reset_ts; +}; + +/** + * dst_cache_get - perform cache lookup + * @dst_cache: the cache + * + * The caller should use dst_cache_get_ip4() if it need to retrieve the + * source address to be used when xmitting to the cached dst. + * local BH must be disabled. + */ +struct dst_entry *dst_cache_get(struct dst_cache *dst_cache); + +/** + * dst_cache_get_ip4 - perform cache lookup and fetch ipv4 source address + * @dst_cache: the cache + * @saddr: return value for the retrieved source address + * + * local BH must be disabled. + */ +struct rtable *dst_cache_get_ip4(struct dst_cache *dst_cache, __be32 *saddr); + +/** + * dst_cache_set_ip4 - store the ipv4 dst into the cache + * @dst_cache: the cache + * @dst: the entry to be cached + * @saddr: the source address to be stored inside the cache + * + * local BH must be disabled. + */ +void dst_cache_set_ip4(struct dst_cache *dst_cache, struct dst_entry *dst, + __be32 saddr); + +#if IS_ENABLED(CONFIG_IPV6) + +/** + * dst_cache_set_ip6 - store the ipv6 dst into the cache + * @dst_cache: the cache + * @dst: the entry to be cached + * @saddr: the source address to be stored inside the cache + * + * local BH must be disabled. + */ +void dst_cache_set_ip6(struct dst_cache *dst_cache, struct dst_entry *dst, + const struct in6_addr *saddr); + +/** + * dst_cache_get_ip6 - perform cache lookup and fetch ipv6 source address + * @dst_cache: the cache + * @saddr: return value for the retrieved source address + * + * local BH must be disabled. + */ +struct dst_entry *dst_cache_get_ip6(struct dst_cache *dst_cache, + struct in6_addr *saddr); +#endif + +/** + * dst_cache_reset - invalidate the cache contents + * @dst_cache: the cache + * + * This does not free the cached dst to avoid races and contentions. + * the dst will be freed on later cache lookup. + */ +static inline void dst_cache_reset(struct dst_cache *dst_cache) +{ + WRITE_ONCE(dst_cache->reset_ts, jiffies); +} + +/** + * dst_cache_reset_now - invalidate the cache contents immediately + * @dst_cache: the cache + * + * The caller must be sure there are no concurrent users, as this frees + * all dst_cache users immediately, rather than waiting for the next + * per-cpu usage like dst_cache_reset does. Most callers should use the + * higher speed lazily-freed dst_cache_reset function instead. + */ +void dst_cache_reset_now(struct dst_cache *dst_cache); + +/** + * dst_cache_init - initialize the cache, allocating the required storage + * @dst_cache: the cache + * @gfp: allocation flags + */ +int dst_cache_init(struct dst_cache *dst_cache, gfp_t gfp); + +/** + * dst_cache_destroy - empty the cache and free the allocated storage + * @dst_cache: the cache + * + * No synchronization is enforced: it must be called only when the cache + * is unused. + */ +void dst_cache_destroy(struct dst_cache *dst_cache); + +#endif -- cgit v1.3.1