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/ethtool/linkstate.c | 226 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 226 insertions(+) create mode 100644 net/ethtool/linkstate.c (limited to 'net/ethtool/linkstate.c') diff --git a/net/ethtool/linkstate.c b/net/ethtool/linkstate.c new file mode 100644 index 000000000..24569e929 --- /dev/null +++ b/net/ethtool/linkstate.c @@ -0,0 +1,226 @@ +// SPDX-License-Identifier: GPL-2.0-only + +#include +#include + +#include "common.h" +#include "netlink.h" + +struct linkstate_req_info { + struct ethnl_req_info base; +}; + +struct linkstate_reply_data { + struct ethnl_reply_data base; + int link; + int sqi; + int sqi_max; + struct ethtool_link_ext_stats link_stats; + bool link_ext_state_provided; + struct ethtool_link_ext_state_info ethtool_link_ext_state_info; +}; + +#define LINKSTATE_REPDATA(__reply_base) \ + container_of(__reply_base, struct linkstate_reply_data, base) + +const struct nla_policy ethnl_linkstate_get_policy[] = { + [ETHTOOL_A_LINKSTATE_HEADER] = + NLA_POLICY_NESTED(ethnl_header_policy_stats), +}; + +static int linkstate_get_sqi(struct phy_device *phydev) +{ + int ret; + + if (!phydev) + return -EOPNOTSUPP; + + mutex_lock(&phydev->lock); + if (!phydev->drv || !phydev->drv->get_sqi) + ret = -EOPNOTSUPP; + else if (!phydev->link) + ret = -ENETDOWN; + else + ret = phydev->drv->get_sqi(phydev); + mutex_unlock(&phydev->lock); + + return ret; +} + +static int linkstate_get_sqi_max(struct phy_device *phydev) +{ + int ret; + + if (!phydev) + return -EOPNOTSUPP; + + mutex_lock(&phydev->lock); + if (!phydev->drv || !phydev->drv->get_sqi_max) + ret = -EOPNOTSUPP; + else if (!phydev->link) + ret = -ENETDOWN; + else + ret = phydev->drv->get_sqi_max(phydev); + mutex_unlock(&phydev->lock); + + return ret; +}; + +static bool linkstate_sqi_critical_error(int sqi) +{ + return sqi < 0 && sqi != -EOPNOTSUPP && sqi != -ENETDOWN; +} + +static bool linkstate_sqi_valid(struct linkstate_reply_data *data) +{ + return data->sqi >= 0 && data->sqi_max >= 0 && + data->sqi <= data->sqi_max; +} + +static int linkstate_get_link_ext_state(struct net_device *dev, + struct linkstate_reply_data *data) +{ + int err; + + if (!dev->ethtool_ops->get_link_ext_state) + return -EOPNOTSUPP; + + err = dev->ethtool_ops->get_link_ext_state(dev, &data->ethtool_link_ext_state_info); + if (err) + return err; + + data->link_ext_state_provided = true; + + return 0; +} + +static int linkstate_prepare_data(const struct ethnl_req_info *req_base, + struct ethnl_reply_data *reply_base, + const struct genl_info *info) +{ + struct linkstate_reply_data *data = LINKSTATE_REPDATA(reply_base); + struct net_device *dev = reply_base->dev; + struct nlattr **tb = info->attrs; + struct phy_device *phydev; + int ret; + + phydev = ethnl_req_get_phydev(req_base, tb, ETHTOOL_A_LINKSTATE_HEADER, + info->extack); + if (IS_ERR(phydev)) + return PTR_ERR(phydev); + + ret = ethnl_ops_begin(dev); + if (ret < 0) + return ret; + data->link = __ethtool_get_link(dev); + + ret = linkstate_get_sqi(phydev); + if (linkstate_sqi_critical_error(ret)) + goto out; + data->sqi = ret; + + ret = linkstate_get_sqi_max(phydev); + if (linkstate_sqi_critical_error(ret)) + goto out; + data->sqi_max = ret; + + if (dev->flags & IFF_UP) { + ret = linkstate_get_link_ext_state(dev, data); + if (ret < 0 && ret != -EOPNOTSUPP && ret != -ENODATA) + goto out; + } + + ethtool_stats_init((u64 *)&data->link_stats, + sizeof(data->link_stats) / 8); + + if (req_base->flags & ETHTOOL_FLAG_STATS) { + if (phydev) + phy_ethtool_get_link_ext_stats(phydev, + &data->link_stats); + + if (dev->ethtool_ops->get_link_ext_stats) + dev->ethtool_ops->get_link_ext_stats(dev, + &data->link_stats); + } + + ret = 0; +out: + ethnl_ops_complete(dev); + return ret; +} + +static int linkstate_reply_size(const struct ethnl_req_info *req_base, + const struct ethnl_reply_data *reply_base) +{ + struct linkstate_reply_data *data = LINKSTATE_REPDATA(reply_base); + int len; + + len = nla_total_size(sizeof(u8)) /* LINKSTATE_LINK */ + + 0; + + if (linkstate_sqi_valid(data)) { + len += nla_total_size(sizeof(u32)); /* LINKSTATE_SQI */ + len += nla_total_size(sizeof(u32)); /* LINKSTATE_SQI_MAX */ + } + + if (data->link_ext_state_provided) + len += nla_total_size(sizeof(u8)); /* LINKSTATE_EXT_STATE */ + + if (data->ethtool_link_ext_state_info.__link_ext_substate) + len += nla_total_size(sizeof(u8)); /* LINKSTATE_EXT_SUBSTATE */ + + if (data->link_stats.link_down_events != ETHTOOL_STAT_NOT_SET) + len += nla_total_size(sizeof(u32)); + + return len; +} + +static int linkstate_fill_reply(struct sk_buff *skb, + const struct ethnl_req_info *req_base, + const struct ethnl_reply_data *reply_base) +{ + struct linkstate_reply_data *data = LINKSTATE_REPDATA(reply_base); + + if (data->link >= 0 && + nla_put_u8(skb, ETHTOOL_A_LINKSTATE_LINK, !!data->link)) + return -EMSGSIZE; + + if (linkstate_sqi_valid(data)) { + if (nla_put_u32(skb, ETHTOOL_A_LINKSTATE_SQI, data->sqi)) + return -EMSGSIZE; + + if (nla_put_u32(skb, ETHTOOL_A_LINKSTATE_SQI_MAX, + data->sqi_max)) + return -EMSGSIZE; + } + + if (data->link_ext_state_provided) { + if (nla_put_u8(skb, ETHTOOL_A_LINKSTATE_EXT_STATE, + data->ethtool_link_ext_state_info.link_ext_state)) + return -EMSGSIZE; + + if (data->ethtool_link_ext_state_info.__link_ext_substate && + nla_put_u8(skb, ETHTOOL_A_LINKSTATE_EXT_SUBSTATE, + data->ethtool_link_ext_state_info.__link_ext_substate)) + return -EMSGSIZE; + } + + if (data->link_stats.link_down_events != ETHTOOL_STAT_NOT_SET) + if (nla_put_u32(skb, ETHTOOL_A_LINKSTATE_EXT_DOWN_CNT, + data->link_stats.link_down_events)) + return -EMSGSIZE; + + return 0; +} + +const struct ethnl_request_ops ethnl_linkstate_request_ops = { + .request_cmd = ETHTOOL_MSG_LINKSTATE_GET, + .reply_cmd = ETHTOOL_MSG_LINKSTATE_GET_REPLY, + .hdr_attr = ETHTOOL_A_LINKSTATE_HEADER, + .req_info_size = sizeof(struct linkstate_req_info), + .reply_data_size = sizeof(struct linkstate_reply_data), + + .prepare_data = linkstate_prepare_data, + .reply_size = linkstate_reply_size, + .fill_reply = linkstate_fill_reply, +}; -- cgit v1.3.1