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 --- net/nsh/Kconfig | 10 ++++ net/nsh/Makefile | 2 + net/nsh/nsh.c | 153 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 165 insertions(+) create mode 100644 net/nsh/Kconfig create mode 100644 net/nsh/Makefile create mode 100644 net/nsh/nsh.c (limited to 'net/nsh') diff --git a/net/nsh/Kconfig b/net/nsh/Kconfig new file mode 100644 index 000000000..84f2a2823 --- /dev/null +++ b/net/nsh/Kconfig @@ -0,0 +1,10 @@ +# SPDX-License-Identifier: GPL-2.0-only +menuconfig NET_NSH + tristate "Network Service Header (NSH) protocol" + default n + help + Network Service Header is an implementation of Service Function + Chaining (RFC 7665). The current implementation in Linux supports + only MD type 1 and only with the openvswitch module. + + If unsure, say N. diff --git a/net/nsh/Makefile b/net/nsh/Makefile new file mode 100644 index 000000000..de34a615a --- /dev/null +++ b/net/nsh/Makefile @@ -0,0 +1,2 @@ +# SPDX-License-Identifier: GPL-2.0-only +obj-$(CONFIG_NET_NSH) += nsh.o diff --git a/net/nsh/nsh.c b/net/nsh/nsh.c new file mode 100644 index 000000000..bfb775806 --- /dev/null +++ b/net/nsh/nsh.c @@ -0,0 +1,153 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Network Service Header + * + * Copyright (c) 2017 Red Hat, Inc. -- Jiri Benc + */ + +#include +#include +#include +#include +#include +#include + +int nsh_push(struct sk_buff *skb, const struct nshhdr *pushed_nh) +{ + struct nshhdr *nh; + size_t length = nsh_hdr_len(pushed_nh); + u8 next_proto; + + if (skb->mac_len) { + next_proto = TUN_P_ETHERNET; + } else { + next_proto = tun_p_from_eth_p(skb->protocol); + if (!next_proto) + return -EAFNOSUPPORT; + } + + /* Add the NSH header */ + if (skb_cow_head(skb, length) < 0) + return -ENOMEM; + + skb_push(skb, length); + nh = (struct nshhdr *)(skb->data); + memcpy(nh, pushed_nh, length); + nh->np = next_proto; + skb_postpush_rcsum(skb, nh, length); + + skb->protocol = htons(ETH_P_NSH); + skb_reset_mac_header(skb); + skb_reset_network_header(skb); + skb_reset_mac_len(skb); + + return 0; +} +EXPORT_SYMBOL_GPL(nsh_push); + +int nsh_pop(struct sk_buff *skb) +{ + struct nshhdr *nh; + size_t length; + __be16 inner_proto; + + if (!pskb_may_pull(skb, NSH_BASE_HDR_LEN)) + return -ENOMEM; + nh = (struct nshhdr *)(skb->data); + length = nsh_hdr_len(nh); + if (length < NSH_BASE_HDR_LEN) + return -EINVAL; + inner_proto = tun_p_to_eth_p(nh->np); + if (!pskb_may_pull(skb, length)) + return -ENOMEM; + + if (!inner_proto) + return -EAFNOSUPPORT; + + skb_pull_rcsum(skb, length); + skb_reset_mac_header(skb); + skb_reset_network_header(skb); + skb_reset_mac_len(skb); + skb->protocol = inner_proto; + + return 0; +} +EXPORT_SYMBOL_GPL(nsh_pop); + +static struct sk_buff *nsh_gso_segment(struct sk_buff *skb, + netdev_features_t features) +{ + unsigned int outer_hlen, mac_len, nsh_len; + struct sk_buff *segs = ERR_PTR(-EINVAL); + u16 mac_offset = skb->mac_header; + __be16 outer_proto, proto; + + skb_reset_network_header(skb); + + outer_proto = skb->protocol; + outer_hlen = skb_mac_header_len(skb); + mac_len = skb->mac_len; + + if (unlikely(!pskb_may_pull(skb, NSH_BASE_HDR_LEN))) + goto out; + nsh_len = nsh_hdr_len(nsh_hdr(skb)); + if (nsh_len < NSH_BASE_HDR_LEN) + goto out; + if (unlikely(!pskb_may_pull(skb, nsh_len))) + goto out; + + proto = tun_p_to_eth_p(nsh_hdr(skb)->np); + if (!proto) + goto out; + + __skb_pull(skb, nsh_len); + + skb_reset_mac_header(skb); + skb->mac_len = proto == htons(ETH_P_TEB) ? ETH_HLEN : 0; + skb->protocol = proto; + + features &= NETIF_F_SG; + segs = skb_mac_gso_segment(skb, features); + if (IS_ERR_OR_NULL(segs)) { + skb_gso_error_unwind(skb, htons(ETH_P_NSH), nsh_len, + mac_offset, mac_len); + goto out; + } + + for (skb = segs; skb; skb = skb->next) { + skb->protocol = outer_proto; + __skb_push(skb, nsh_len + outer_hlen); + skb_reset_mac_header(skb); + skb_set_network_header(skb, outer_hlen); + skb->mac_len = mac_len; + } + +out: + return segs; +} + +static struct packet_offload nsh_packet_offload __read_mostly = { + .type = htons(ETH_P_NSH), + .priority = 15, + .callbacks = { + .gso_segment = nsh_gso_segment, + }, +}; + +static int __init nsh_init_module(void) +{ + dev_add_offload(&nsh_packet_offload); + return 0; +} + +static void __exit nsh_cleanup_module(void) +{ + dev_remove_offload(&nsh_packet_offload); +} + +module_init(nsh_init_module); +module_exit(nsh_cleanup_module); + +MODULE_AUTHOR("Jiri Benc "); +MODULE_DESCRIPTION("NSH protocol"); +MODULE_LICENSE("GPL v2"); -- cgit v1.3.1