[PATCH OLK-6.6 v5 00/14] Cgroup-Based Interference Statistics
Pu Lehui (6): interference: Add CGROUP_IFS config interference: Add CGROUP_IFS_DEFAULT_ENABLED config cgroup: Export cgroup_init_cftypes as internal interface cgroup: Export cgroup_addrm_files as internal interface interference: Add cgroup v1 support for CGROUP_IFS openeuler_defconfig: Disable CGROUP_IFS config by default Tengda Wu (5): interference: Add sched rundelay interference track support interference: Add sched sleep interference track support interference: Add sched throttle interference track support interference: Exclude throttle time from rundelay interference: Add sched wakelat interference track support Xu Kuohai (3): interference: Add cgroup-based inteference statistics framework interference: Add smt interference track support interference: Add irq interference track support arch/arm64/configs/openeuler_defconfig | 1 + arch/x86/configs/openeuler_defconfig | 1 + arch/x86/kernel/smpboot.c | 5 +- drivers/base/arch_topology.c | 4 + include/linux/cgroup-defs.h | 4 + include/linux/cgroup.h | 189 ++++++++++++++ include/linux/sched.h | 9 + init/Kconfig | 23 ++ kernel/cgroup/Makefile | 1 + kernel/cgroup/cgroup-internal.h | 20 ++ kernel/cgroup/cgroup.c | 27 +- kernel/cgroup/ifs.c | 339 +++++++++++++++++++++++++ kernel/sched/core.c | 23 +- kernel/sched/cpuacct.c | 15 ++ kernel/sched/cputime.c | 8 +- kernel/sched/deadline.c | 4 +- kernel/sched/fair.c | 14 +- kernel/sched/rt.c | 4 +- kernel/sched/sched.h | 11 +- kernel/sched/stats.c | 2 + kernel/sched/stats.h | 54 +++- 21 files changed, 729 insertions(+), 29 deletions(-) create mode 100644 kernel/cgroup/ifs.c -- 2.34.1
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Add CGROUP_IFS config which will provide online low-overhead cgroup-based interference statistics. Signed-off-by: Pu Lehui <pulehui@huawei.com> --- init/Kconfig | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/init/Kconfig b/init/Kconfig index 38d41dd127c0..838cd2398276 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1341,6 +1341,18 @@ config CGROUP_FILES This supports catching misbehaving processes and return EMFILE instead of ENOMEM for kernel memory limits. +config CGROUP_IFS + bool "Cgroup-based Interference Statistics" + default n + select KERNFS + select IRQ_TIME_ACCOUTING + help + This option will provide online low-overhead interference + statistics based on cgroup, which will help in troubleshooting + performance issues. + + Say N if unsure. + endif # CGROUPS config UCOUNTS_PERCPU_COUNTER -- 2.34.1
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Add CGROUP_IFS_DEFAULT_ENABLED config which will enable cgroup-based interference statistics by default. Signed-off-by: Pu Lehui <pulehui@huawei.com> --- init/Kconfig | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/init/Kconfig b/init/Kconfig index 838cd2398276..8c305798bbc9 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1353,6 +1353,17 @@ config CGROUP_IFS Say N if unsure. +config CGROUP_IFS_DEFAULT_ENABLED + bool "Enable Cgroup-based Interference Statistics by default" + default y + depends on CGROUP_IFS + help + This option will enable cgroup-based interference statistics + by default. If this option is N, you need to enable cgroup-based + interference statistics by configuring the cgroup_ifs startup param. + + Say Y if unsure. + endif # CGROUPS config UCOUNTS_PERCPU_COUNTER -- 2.34.1
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Export cgroup_init_cftypes as internal interface for upcomming cgroup ifs framework. Signed-off-by: Pu Lehui <pulehui@huawei.com> --- kernel/cgroup/cgroup-internal.h | 1 + kernel/cgroup/cgroup.c | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/kernel/cgroup/cgroup-internal.h b/kernel/cgroup/cgroup-internal.h index f5fb12890645..831449ec81e5 100644 --- a/kernel/cgroup/cgroup-internal.h +++ b/kernel/cgroup/cgroup-internal.h @@ -297,5 +297,6 @@ void cgroup1_check_for_release(struct cgroup *cgrp); int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param); int cgroup1_get_tree(struct fs_context *fc); int cgroup1_reconfigure(struct fs_context *ctx); +int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); #endif /* __CGROUP_INTERNAL_H */ diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index 34f1a810f7b0..fac7113c122d 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -4414,7 +4414,7 @@ static void cgroup_exit_cftypes(struct cftype *cfts) } } -static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) +int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) { struct cftype *cft; int ret = 0; -- 2.34.1
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Export cgroup_addrm_files as internal interface for upcomming cgroup ifs framework. Signed-off-by: Pu Lehui <pulehui@huawei.com> --- kernel/cgroup/cgroup-internal.h | 3 +++ kernel/cgroup/cgroup.c | 9 +++------ 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/kernel/cgroup/cgroup-internal.h b/kernel/cgroup/cgroup-internal.h index 831449ec81e5..a379e523ae05 100644 --- a/kernel/cgroup/cgroup-internal.h +++ b/kernel/cgroup/cgroup-internal.h @@ -298,5 +298,8 @@ int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param); int cgroup1_get_tree(struct fs_context *fc); int cgroup1_reconfigure(struct fs_context *ctx); int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); +int cgroup_addrm_files(struct cgroup_subsys_state *css, + struct cgroup *cgrp, struct cftype cfts[], + bool is_add); #endif /* __CGROUP_INTERNAL_H */ diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index fac7113c122d..5dcfd236fd33 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -247,9 +247,6 @@ static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, struct cgroup_subsys *ss); static void css_release(struct percpu_ref *ref); static void kill_css(struct cgroup_subsys_state *css); -static int cgroup_addrm_files(struct cgroup_subsys_state *css, - struct cgroup *cgrp, struct cftype cfts[], - bool is_add); #ifdef CONFIG_DEBUG_CGROUP_REF #define CGROUP_REF_FN_ATTRS noinline @@ -4331,9 +4328,9 @@ static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, * Depending on @is_add, add or remove files defined by @cfts on @cgrp. * For removals, this function never fails. */ -static int cgroup_addrm_files(struct cgroup_subsys_state *css, - struct cgroup *cgrp, struct cftype cfts[], - bool is_add) +int cgroup_addrm_files(struct cgroup_subsys_state *css, + struct cgroup *cgrp, struct cftype cfts[], + bool is_add) { struct cftype *cft, *cft_end = NULL; int ret = 0; -- 2.34.1
From: Xu Kuohai <xukuohai@huawei.com> hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Add cgroup-based interference statistics framework. Signed-off-by: Xu Kuohai <xukuohai@huawei.com> Signed-off-by: Pu Lehui <pulehui@huawei.com> --- include/linux/cgroup-defs.h | 4 + include/linux/cgroup.h | 55 +++++++++++ kernel/cgroup/Makefile | 1 + kernel/cgroup/cgroup-internal.h | 16 ++++ kernel/cgroup/cgroup.c | 16 +++- kernel/cgroup/ifs.c | 163 ++++++++++++++++++++++++++++++++ 6 files changed, 254 insertions(+), 1 deletion(-) create mode 100644 kernel/cgroup/ifs.c diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h index f3fd0407d346..47372df6abbe 100644 --- a/include/linux/cgroup-defs.h +++ b/include/linux/cgroup-defs.h @@ -552,7 +552,11 @@ struct cgroup { struct bpf_local_storage __rcu *bpf_cgrp_storage; #endif +#ifdef CONFIG_CGROUP_IFS + KABI_USE(1, struct cgroup_ifs *ifs) +#else KABI_RESERVE(1) +#endif KABI_RESERVE(2) KABI_RESERVE(3) KABI_RESERVE(4) diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index 62cea15eb6df..fd5617e7f0de 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -860,4 +860,59 @@ static inline void cgroup_bpf_put(struct cgroup *cgrp) {} void cgroup_move_task_to_root(struct task_struct *tsk); #endif +#ifdef CONFIG_CGROUP_IFS + +enum ifs_types { + IFS_STUB, + NR_IFS_TYPES, +}; + +struct cgroup_ifs_cpu { + /* total time for each interference, in ns */ + u64 time[NR_IFS_TYPES]; +}; + +/* + * cgroup interference statistics + */ +struct cgroup_ifs { + /* per-cpu interference statistics tracking */ + struct cgroup_ifs_cpu __percpu *pcpu; +}; + +extern struct cgroup_ifs cgroup_root_ifs; + +DECLARE_STATIC_KEY_FALSE(cgrp_ifs_enabled); +static inline bool cgroup_ifs_enabled(void) +{ + return static_branch_unlikely(&cgrp_ifs_enabled); +} + +static inline struct cgroup_ifs *cgroup_ifs(struct cgroup *cgrp) +{ + return cgroup_ino(cgrp) == 1 ? &cgroup_root_ifs : cgrp->ifs; +} + +static inline struct cgroup_ifs *task_ifs(struct task_struct *task) +{ + return cgroup_ifs(task_dfl_cgroup(task)); +} + +static inline struct cgroup_ifs *current_ifs(void) +{ + return task_ifs(current); +} + +static inline void cgroup_ifs_account_delta(struct cgroup_ifs_cpu *ifsc, + int type, u64 delta) +{ + if (!cgroup_ifs_enabled()) + return; + + if (delta > 0) + ifsc->time[type] += delta; +} + +#endif /* CONFIG_CGROUP_IFS */ + #endif /* _LINUX_CGROUP_H */ diff --git a/kernel/cgroup/Makefile b/kernel/cgroup/Makefile index 12f8457ad1f9..8d0cc493fb23 100644 --- a/kernel/cgroup/Makefile +++ b/kernel/cgroup/Makefile @@ -7,3 +7,4 @@ obj-$(CONFIG_CGROUP_RDMA) += rdma.o obj-$(CONFIG_CPUSETS) += cpuset.o obj-$(CONFIG_CGROUP_MISC) += misc.o obj-$(CONFIG_CGROUP_DEBUG) += debug.o +obj-$(CONFIG_CGROUP_IFS) += ifs.o diff --git a/kernel/cgroup/cgroup-internal.h b/kernel/cgroup/cgroup-internal.h index a379e523ae05..0e24ff3c661a 100644 --- a/kernel/cgroup/cgroup-internal.h +++ b/kernel/cgroup/cgroup-internal.h @@ -302,4 +302,20 @@ int cgroup_addrm_files(struct cgroup_subsys_state *css, struct cgroup *cgrp, struct cftype cfts[], bool is_add); +#ifdef CONFIG_CGROUP_IFS +int cgroup_ifs_alloc(struct cgroup *cgrp); +void cgroup_ifs_free(struct cgroup *cgrp); +void cgroup_ifs_init(void); +int cgroup_ifs_add_files(struct cgroup_subsys_state *css, struct cgroup *cgrp); +void cgroup_ifs_rm_files(struct cgroup_subsys_state *css, struct cgroup *cgrp); +#else /* !CONFIG_CGROUP_IFS */ +static inline int cgroup_ifs_alloc(struct cgroup *cgrp) { return 0; } +static inline void cgroup_ifs_free(struct cgroup *cgrp) {} +static inline void cgroup_ifs_init(void) {} +static inline int cgroup_ifs_add_files(struct cgroup_subsys_state *css, + struct cgroup *cgrp) { return 0; } +static inline void cgroup_ifs_rm_files(struct cgroup_subsys_state *css, + struct cgroup *cgrp) {} +#endif /* CONFIG_CGROUP_IFS */ + #endif /* __CGROUP_INTERNAL_H */ diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index 5dcfd236fd33..01553d04c366 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -1700,6 +1700,7 @@ static void css_clear_dir(struct cgroup_subsys_state *css) if (cgroup_psi_enabled()) cgroup_addrm_files(css, cgrp, cgroup_psi_files, false); + cgroup_ifs_rm_files(css, cgrp); } else { cgroup_addrm_files(css, cgrp, cgroup1_base_files, false); @@ -1738,6 +1739,10 @@ static int css_populate_dir(struct cgroup_subsys_state *css) if (ret < 0) return ret; } + + ret = cgroup_ifs_add_files(css, cgrp); + if (ret < 0) + return ret; } else { ret = cgroup_addrm_files(css, cgrp, cgroup1_base_files, true); @@ -5477,6 +5482,7 @@ static void css_free_rwork_fn(struct work_struct *work) cgroup_put(cgroup_parent(cgrp)); kernfs_put(cgrp->kn); psi_cgroup_free(cgrp); + cgroup_ifs_free(cgrp); cgroup_rstat_exit(cgrp); kfree(cgrp); } else { @@ -5724,10 +5730,14 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name, if (ret) goto out_kernfs_remove; + ret = cgroup_ifs_alloc(cgrp); + if (ret) + goto out_psi_free; + if (cgrp->root == &cgrp_dfl_root) { ret = cgroup_bpf_inherit(cgrp); if (ret) - goto out_psi_free; + goto out_ifs_free; } /* @@ -5788,6 +5798,8 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name, return cgrp; +out_ifs_free: + cgroup_ifs_free(cgrp); out_psi_free: psi_cgroup_free(cgrp); out_kernfs_remove: @@ -6195,6 +6207,8 @@ int __init cgroup_init(void) BUG_ON(cgroup_init_cftypes(NULL, cgroup_psi_files)); BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files)); + cgroup_ifs_init(); + cgroup_rstat_boot(); get_user_ns(init_cgroup_ns.user_ns); diff --git a/kernel/cgroup/ifs.c b/kernel/cgroup/ifs.c new file mode 100644 index 000000000000..fc761cb02a6d --- /dev/null +++ b/kernel/cgroup/ifs.c @@ -0,0 +1,163 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Interference statistics for cgroup + * + * Copyright (C) 2025-2025 Huawei Technologies Co., Ltd + */ + +#include "cgroup-internal.h" + +static DEFINE_PER_CPU(struct cgroup_ifs_cpu, cgrp_root_ifs_cpu); +struct cgroup_ifs cgroup_root_ifs = { + .pcpu = &cgrp_root_ifs_cpu, +}; + +DEFINE_STATIC_KEY_FALSE(cgrp_ifs_enabled); + +#ifdef CONFIG_CGROUP_IFS_DEFAULT_ENABLED +static bool ifs_enable = true; +#else +static bool ifs_enable; +#endif + +static int __init setup_ifs(char *str) +{ + return kstrtobool(str, &ifs_enable) == 0; +} +__setup("cgroup_ifs=", setup_ifs); + +int cgroup_ifs_alloc(struct cgroup *cgrp) +{ + cgrp->ifs = kzalloc(sizeof(struct cgroup_ifs), GFP_KERNEL); + if (!cgrp->ifs) + return -ENOMEM; + + cgrp->ifs->pcpu = alloc_percpu(struct cgroup_ifs_cpu); + if (!cgrp->ifs->pcpu) { + kfree(cgrp->ifs); + return -ENOMEM; + } + + return 0; +} + +void cgroup_ifs_free(struct cgroup *cgrp) +{ + free_percpu(cgrp->ifs->pcpu); + kfree(cgrp->ifs); +} + +static const char *ifs_type_name(int type) +{ + char *name = NULL; + + switch (type) { + default: + break; + } + + return name; +} + +static int print_sum_time(struct cgroup_ifs *ifs, struct seq_file *seq) +{ + u64 time[NR_IFS_TYPES] = { 0 }; + int cpu; + int i; + + for_each_possible_cpu(cpu) { + for (i = 0; i < NR_IFS_TYPES; i++) + time[i] += per_cpu_ptr(ifs->pcpu, cpu)->time[i]; + } + + seq_printf(seq, "%-18s%s\n", "Interference", "Total Time (ns)"); + + for (i = 0; i < NR_IFS_TYPES; i++) + seq_printf(seq, "%-18s%llu\n", ifs_type_name(i), time[i]); + + return 0; +} + +static int cgroup_ifs_show(struct seq_file *seq, void *v) +{ + struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup; + struct cgroup_ifs __maybe_unused *ifs = cgroup_ifs(cgrp); + int ret; + + if (!ifs) { + pr_info("cgroup_ino(cgrp) = %ld\n", cgroup_ino(cgrp)); + return -EINVAL; + } + + ret = print_sum_time(ifs, seq); + if (ret) + return ret; + + return 0; +} + +static ssize_t cgroup_ifs_write(struct kernfs_open_file *of, char *buf, + size_t nbytes, loff_t off) +{ + struct cgroup *cgrp = seq_css(of->seq_file)->cgroup; + struct cgroup_ifs *ifs = cgroup_ifs(cgrp); + struct cgroup_ifs_cpu *ifsc; + bool clear; + int cpu; + + if (!ifs) + return -EOPNOTSUPP; + + if (kstrtobool(strstrip(buf), &clear) < 0) + return -EINVAL; + + if (!clear) { + for_each_possible_cpu(cpu) { + ifsc = per_cpu_ptr(ifs->pcpu, cpu); + memset(ifsc->time, 0, sizeof(ifsc->time)); + } + } + + return nbytes; +} + +static struct cftype cgroup_ifs_files[] = { + { + .name = "interference.stat", + .seq_show = cgroup_ifs_show, + .write = cgroup_ifs_write, + }, + { } /* terminate */ +}; + +void cgroup_ifs_init(void) +{ + if (!ifs_enable) + return; + + BUG_ON(cgroup_init_cftypes(NULL, cgroup_ifs_files)); + + static_branch_enable(&cgrp_ifs_enabled); +} + +int cgroup_ifs_add_files(struct cgroup_subsys_state *css, struct cgroup *cgrp) +{ + int ret = 0; + + if (!cgroup_ifs_enabled()) + return 0; + + ret = cgroup_addrm_files(css, cgrp, cgroup_ifs_files, true); + if (ret < 0) { + cgroup_addrm_files(css, cgrp, cgroup_ifs_files, false); + return ret; + } + + return 0; +} + +void cgroup_ifs_rm_files(struct cgroup_subsys_state *css, struct cgroup *cgrp) +{ + if (cgroup_ifs_enabled()) + cgroup_addrm_files(css, cgrp, cgroup_ifs_files, false); +} -- 2.34.1
From: Xu Kuohai <xukuohai@huawei.com> hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Add smt interference track support. Signed-off-by: Xu Kuohai <xukuohai@huawei.com> Signed-off-by: Pu Lehui <pulehui@huawei.com> --- arch/x86/kernel/smpboot.c | 5 +- drivers/base/arch_topology.c | 4 ++ include/linux/cgroup.h | 12 +++- kernel/cgroup/ifs.c | 103 +++++++++++++++++++++++++++++++++++ kernel/sched/core.c | 1 + 5 files changed, 123 insertions(+), 2 deletions(-) diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c index a5e84700b319..2ce5ddb7a043 100644 --- a/arch/x86/kernel/smpboot.c +++ b/arch/x86/kernel/smpboot.c @@ -61,6 +61,7 @@ #include <linux/cpuhotplug.h> #include <linux/mc146818rtc.h> #include <linux/acpi.h> +#include <linux/cgroup.h> #include <linux/cpuset.h> #include <asm/acpi.h> @@ -681,8 +682,10 @@ void set_cpu_sibling_map(int cpu) if (match_pkg(c, o) && !topology_same_node(c, o)) x86_has_numa_in_package = true; - if ((i == cpu) || (has_smt && match_smt(c, o))) + if ((i == cpu) || (has_smt && match_smt(c, o))) { link_mask(topology_sibling_cpumask, cpu, i); + cgroup_ifs_set_smt(topology_sibling_cpumask(cpu)); + } if ((i == cpu) || (has_mp && match_llc(c, o))) link_mask(cpu_llc_shared_mask, cpu, i); diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c index 1525d3e2f9d5..a26169094712 100644 --- a/drivers/base/arch_topology.c +++ b/drivers/base/arch_topology.c @@ -807,6 +807,8 @@ void update_siblings_masks(unsigned int cpuid) cpumask_set_cpu(cpuid, &cpu_topo->thread_sibling); cpumask_set_cpu(cpu, &cpuid_topo->thread_sibling); } + + cgroup_ifs_set_smt(&cpuid_topo->thread_sibling); } static void clear_cpu_topology(int cpu) @@ -823,6 +825,8 @@ static void clear_cpu_topology(int cpu) cpumask_set_cpu(cpu, &cpu_topo->core_sibling); cpumask_clear(&cpu_topo->thread_sibling); cpumask_set_cpu(cpu, &cpu_topo->thread_sibling); + + cgroup_ifs_set_smt(&cpu_topo->thread_sibling); } void __init reset_cpu_topology(void) diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index fd5617e7f0de..256d04d28ea2 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -863,7 +863,7 @@ void cgroup_move_task_to_root(struct task_struct *tsk); #ifdef CONFIG_CGROUP_IFS enum ifs_types { - IFS_STUB, + IFS_SMT, NR_IFS_TYPES, }; @@ -913,6 +913,16 @@ static inline void cgroup_ifs_account_delta(struct cgroup_ifs_cpu *ifsc, ifsc->time[type] += delta; } +void cgroup_ifs_account_smttime(struct task_struct *prev, + struct task_struct *next, + struct task_struct *idle); +void cgroup_ifs_set_smt(cpumask_t *sibling); + +#else /* !CONFIG_CGROUP_IFS */ +static inline void cgroup_ifs_account_smttime(struct task_struct *prev, + struct task_struct *next, + struct task_struct *idle) {} +static inline void cgroup_ifs_set_smt(cpumask_t *sibling) {} #endif /* CONFIG_CGROUP_IFS */ #endif /* _LINUX_CGROUP_H */ diff --git a/kernel/cgroup/ifs.c b/kernel/cgroup/ifs.c index fc761cb02a6d..634ce18e0773 100644 --- a/kernel/cgroup/ifs.c +++ b/kernel/cgroup/ifs.c @@ -5,8 +5,20 @@ * Copyright (C) 2025-2025 Huawei Technologies Co., Ltd */ +#include <linux/sched/clock.h> #include "cgroup-internal.h" +/* smt information */ +struct smt_info { + u64 total_time; + u64 prev_read_time; + u64 noidle_enter_time; + bool is_noidle; +}; + +static DEFINE_PER_CPU(struct smt_info, smt_info); +static DEFINE_PER_CPU_READ_MOSTLY(int, smt_sibling) = -1; + static DEFINE_PER_CPU(struct cgroup_ifs_cpu, cgrp_root_ifs_cpu); struct cgroup_ifs cgroup_root_ifs = { .pcpu = &cgrp_root_ifs_cpu, @@ -26,6 +38,94 @@ static int __init setup_ifs(char *str) } __setup("cgroup_ifs=", setup_ifs); +void cgroup_ifs_set_smt(cpumask_t *sibling) +{ + int cpu; + int cpuid1 = -1; + int cpuid2 = -1; + bool off = false; + + for_each_cpu(cpu, sibling) { + if (cpuid1 == -1) { + cpuid1 = cpu; + } else if (cpuid2 == -1) { + cpuid2 = cpu; + } else { + *per_cpu_ptr(&smt_sibling, cpu) = -1; + off = true; + } + } + + if (cpuid1 != -1) + *per_cpu_ptr(&smt_sibling, cpuid1) = off ? -1 : cpuid2; + + if (cpuid2 != -1) + *per_cpu_ptr(&smt_sibling, cpuid2) = off ? -1 : cpuid1; +} + +static void account_smttime(struct task_struct *task) +{ + u64 delta; + struct cgroup_ifs *ifs; + struct smt_info *info; + + ifs = task_ifs(task); + if (!ifs) + return; + + info = this_cpu_ptr(&smt_info); + + delta = info->total_time - info->prev_read_time; + info->prev_read_time = info->total_time; + + cgroup_ifs_account_delta(this_cpu_ptr(ifs->pcpu), IFS_SMT, delta); +} + +void cgroup_ifs_account_smttime(struct task_struct *prev, + struct task_struct *next, + struct task_struct *idle) +{ + struct smt_info *ci, *si; + u64 now, delta; + int sibling; + + sibling = this_cpu_read(smt_sibling); + if (sibling == -1 || prev == next) + return; + + ci = this_cpu_ptr(&smt_info); + si = per_cpu_ptr(&smt_info, sibling); + + now = sched_clock_cpu(smp_processor_id()); + + /* leave noidle */ + if (prev != idle && next == idle) { + ci->is_noidle = false; + /* account interference time */ + if (ci->noidle_enter_time && si->is_noidle) { + delta = now - ci->noidle_enter_time; + + ci->total_time += delta; + si->total_time += delta; + + si->noidle_enter_time = 0; + ci->noidle_enter_time = 0; + + account_smttime(prev); + } + /* enter noidle */ + } else if (prev == idle && next != idle) { + /* if the sibling is also nonidle, there is smt interference */ + if (si->is_noidle) { + ci->noidle_enter_time = now; + si->noidle_enter_time = now; + } + ci->is_noidle = true; + /* cgroup changed */ + } else if (task_ifs(prev) != task_ifs(next)) + account_smttime(prev); +} + int cgroup_ifs_alloc(struct cgroup *cgrp) { cgrp->ifs = kzalloc(sizeof(struct cgroup_ifs), GFP_KERNEL); @@ -52,6 +152,9 @@ static const char *ifs_type_name(int type) char *name = NULL; switch (type) { + case IFS_SMT: + name = "smt"; + break; default: break; } diff --git a/kernel/sched/core.c b/kernel/sched/core.c index b41f3f30ef57..17ca1d72b03e 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -6750,6 +6750,7 @@ static void __sched notrace __schedule(unsigned int sched_mode) migrate_disable_switch(rq, prev); psi_sched_switch(prev, next, !task_on_rq_queued(prev)); + cgroup_ifs_account_smttime(prev, next, rq->idle); trace_sched_switch(sched_mode & SM_MASK_PREEMPT, prev, next, prev_state); -- 2.34.1
From: Tengda Wu <wutengda2@huawei.com> hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Reuse the 'run_delay' in /proc/<pid>/schedstat to accounting 'rundelay' interference, which indicates how much time the cgroup tasks are in runnable state and wait for running. Signed-off-by: Tengda Wu <wutengda2@huawei.com> Signed-off-by: Pu Lehui <pulehui@huawei.com> --- include/linux/cgroup.h | 17 +++++++++++++++++ kernel/cgroup/ifs.c | 3 +++ kernel/sched/stats.h | 2 ++ 3 files changed, 22 insertions(+) diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index 256d04d28ea2..522f775e5139 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -864,6 +864,7 @@ void cgroup_move_task_to_root(struct task_struct *tsk); enum ifs_types { IFS_SMT, + IFS_RUNDELAY, NR_IFS_TYPES, }; @@ -918,11 +919,27 @@ void cgroup_ifs_account_smttime(struct task_struct *prev, struct task_struct *idle); void cgroup_ifs_set_smt(cpumask_t *sibling); +static inline void cgroup_ifs_account_rundelay(struct task_struct *task, + u64 delta) +{ + struct cgroup_ifs *ifs; + + if (!cgroup_ifs_enabled()) + return; + + ifs = task_ifs(task); + if (!ifs) + return; + + cgroup_ifs_account_delta(this_cpu_ptr(ifs->pcpu), IFS_RUNDELAY, delta); +} + #else /* !CONFIG_CGROUP_IFS */ static inline void cgroup_ifs_account_smttime(struct task_struct *prev, struct task_struct *next, struct task_struct *idle) {} static inline void cgroup_ifs_set_smt(cpumask_t *sibling) {} +static inline void cgroup_ifs_account_rundelay(struct task_struct *task, u64 delta) {} #endif /* CONFIG_CGROUP_IFS */ #endif /* _LINUX_CGROUP_H */ diff --git a/kernel/cgroup/ifs.c b/kernel/cgroup/ifs.c index 634ce18e0773..7bc11c82fce4 100644 --- a/kernel/cgroup/ifs.c +++ b/kernel/cgroup/ifs.c @@ -155,6 +155,9 @@ static const char *ifs_type_name(int type) case IFS_SMT: name = "smt"; break; + case IFS_RUNDELAY: + name = "rundelay"; + break; default: break; } diff --git a/kernel/sched/stats.h b/kernel/sched/stats.h index a6d7206d969d..6d40d4d55b27 100644 --- a/kernel/sched/stats.h +++ b/kernel/sched/stats.h @@ -241,6 +241,7 @@ static inline void sched_info_dequeue(struct rq *rq, struct task_struct *t) t->sched_info.last_queued = 0; t->sched_info.run_delay += delta; + cgroup_ifs_account_rundelay(t, delta); rq_sched_info_dequeue(rq, delta); } @@ -263,6 +264,7 @@ static void sched_info_arrive(struct rq *rq, struct task_struct *t) t->sched_info.last_arrival = now; t->sched_info.pcount++; + cgroup_ifs_account_rundelay(t, delta); rq_sched_info_arrive(rq, delta); } -- 2.34.1
From: Tengda Wu <wutengda2@huawei.com> hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Reuse the 'sum_sleep_runtime' in /proc/schedstats to accounting task 'sleep' interference, which indicates how much time the cgroup tasks are dequeued and wait for enqueuing. Signed-off-by: Tengda Wu <wutengda2@huawei.com> Signed-off-by: Pu Lehui <pulehui@huawei.com> --- include/linux/cgroup.h | 26 ++++++++++++++++++++++++++ kernel/cgroup/ifs.c | 35 +++++++++++++++++++++++++++++++++-- kernel/sched/core.c | 1 + kernel/sched/stats.c | 2 ++ 4 files changed, 62 insertions(+), 2 deletions(-) diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index 522f775e5139..84273989b573 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -865,6 +865,9 @@ void cgroup_move_task_to_root(struct task_struct *tsk); enum ifs_types { IFS_SMT, IFS_RUNDELAY, +#ifdef CONFIG_SCHEDSTATS + IFS_SLEEP, +#endif NR_IFS_TYPES, }; @@ -934,12 +937,35 @@ static inline void cgroup_ifs_account_rundelay(struct task_struct *task, cgroup_ifs_account_delta(this_cpu_ptr(ifs->pcpu), IFS_RUNDELAY, delta); } +#ifdef CONFIG_SCHEDSTATS +static inline void cgroup_ifs_account_sleep(struct task_struct *task, + u64 delta) +{ + struct cgroup_ifs *ifs; + + if (!cgroup_ifs_enabled()) + return; + + ifs = task_ifs(task); + if (!ifs) + return; + + cgroup_ifs_account_delta(this_cpu_ptr(ifs->pcpu), IFS_SLEEP, delta); +} + +void cgroup_ifs_enable_sleep_account(void); +#endif + #else /* !CONFIG_CGROUP_IFS */ static inline void cgroup_ifs_account_smttime(struct task_struct *prev, struct task_struct *next, struct task_struct *idle) {} static inline void cgroup_ifs_set_smt(cpumask_t *sibling) {} static inline void cgroup_ifs_account_rundelay(struct task_struct *task, u64 delta) {} +#ifdef CONFIG_SCHEDSTATS +static inline void cgroup_ifs_account_sleep(struct task_struct *task, u64 delta) {} +static inline void cgroup_ifs_enable_sleep_account(void) {} +#endif #endif /* CONFIG_CGROUP_IFS */ #endif /* _LINUX_CGROUP_H */ diff --git a/kernel/cgroup/ifs.c b/kernel/cgroup/ifs.c index 7bc11c82fce4..c2de386315c2 100644 --- a/kernel/cgroup/ifs.c +++ b/kernel/cgroup/ifs.c @@ -32,6 +32,10 @@ static bool ifs_enable = true; static bool ifs_enable; #endif +#ifdef CONFIG_SCHEDSTATS +static bool ifs_sleep_enable; +#endif + static int __init setup_ifs(char *str) { return kstrtobool(str, &ifs_enable) == 0; @@ -158,6 +162,11 @@ static const char *ifs_type_name(int type) case IFS_RUNDELAY: name = "rundelay"; break; +#ifdef CONFIG_SCHEDSTATS + case IFS_SLEEP: + name = "sleep"; + break; +#endif default: break; } @@ -165,6 +174,15 @@ static const char *ifs_type_name(int type) return name; } +static bool should_print(int type) +{ +#ifdef CONFIG_SCHEDSTATS + if (type == IFS_SLEEP) + return ifs_sleep_enable; +#endif + return true; +} + static int print_sum_time(struct cgroup_ifs *ifs, struct seq_file *seq) { u64 time[NR_IFS_TYPES] = { 0 }; @@ -172,14 +190,20 @@ static int print_sum_time(struct cgroup_ifs *ifs, struct seq_file *seq) int i; for_each_possible_cpu(cpu) { - for (i = 0; i < NR_IFS_TYPES; i++) + for (i = 0; i < NR_IFS_TYPES; i++) { + if (!should_print(i)) + continue; time[i] += per_cpu_ptr(ifs->pcpu, cpu)->time[i]; + } } seq_printf(seq, "%-18s%s\n", "Interference", "Total Time (ns)"); - for (i = 0; i < NR_IFS_TYPES; i++) + for (i = 0; i < NR_IFS_TYPES; i++) { + if (!should_print(i)) + continue; seq_printf(seq, "%-18s%llu\n", ifs_type_name(i), time[i]); + } return 0; } @@ -267,3 +291,10 @@ void cgroup_ifs_rm_files(struct cgroup_subsys_state *css, struct cgroup *cgrp) if (cgroup_ifs_enabled()) cgroup_addrm_files(css, cgrp, cgroup_ifs_files, false); } + +#ifdef CONFIG_SCHEDSTATS +void cgroup_ifs_enable_sleep_account(void) +{ + ifs_sleep_enable = true; +} +#endif diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 17ca1d72b03e..7b77fa528c3c 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4667,6 +4667,7 @@ static void set_schedstats(bool enabled) { if (enabled) { compute_skid(); + cgroup_ifs_enable_sleep_account(); static_branch_enable(&sched_schedstats); } else { static_branch_disable(&sched_schedstats); diff --git a/kernel/sched/stats.c b/kernel/sched/stats.c index 306f26fde69a..e3f1026ba5e8 100644 --- a/kernel/sched/stats.c +++ b/kernel/sched/stats.c @@ -67,6 +67,7 @@ void __update_stats_enqueue_sleeper(struct rq *rq, struct task_struct *p, if (p) { account_scheduler_latency(p, delta >> 10, 1); trace_sched_stat_sleep(p, delta); + cgroup_ifs_account_sleep(p, delta); } } @@ -91,6 +92,7 @@ void __update_stats_enqueue_sleeper(struct rq *rq, struct task_struct *p, } trace_sched_stat_blocked(p, delta); + cgroup_ifs_account_sleep(p, delta); /* * Blocking time is in units of nanosecs, so shift by -- 2.34.1
From: Tengda Wu <wutengda2@huawei.com> hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Collect the duration of bandwidth throttling when the cgroup is throttled. Signed-off-by: Tengda Wu <wutengda2@huawei.com> Signed-off-by: Pu Lehui <pulehui@huawei.com> --- include/linux/cgroup.h | 19 +++++++++++++++++++ kernel/cgroup/ifs.c | 3 +++ kernel/sched/fair.c | 6 +++++- 3 files changed, 27 insertions(+), 1 deletion(-) diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index 84273989b573..8fc762b3841c 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -865,6 +865,7 @@ void cgroup_move_task_to_root(struct task_struct *tsk); enum ifs_types { IFS_SMT, IFS_RUNDELAY, + IFS_THROTTLE, #ifdef CONFIG_SCHEDSTATS IFS_SLEEP, #endif @@ -937,6 +938,23 @@ static inline void cgroup_ifs_account_rundelay(struct task_struct *task, cgroup_ifs_account_delta(this_cpu_ptr(ifs->pcpu), IFS_RUNDELAY, delta); } +static inline void cgroup_ifs_account_throttle(struct cgroup *cgrp, int cpu, + u64 delta) +{ + struct cgroup_ifs *ifs; + struct cgroup_ifs_cpu *ifsc; + + if (!cgroup_ifs_enabled()) + return; + + ifs = cgroup_ifs(cgrp); + if (!ifs) + return; + + ifsc = per_cpu_ptr(ifs->pcpu, cpu); /* XXX: set another cpu data ? */ + cgroup_ifs_account_delta(ifsc, IFS_THROTTLE, delta); +} + #ifdef CONFIG_SCHEDSTATS static inline void cgroup_ifs_account_sleep(struct task_struct *task, u64 delta) @@ -962,6 +980,7 @@ static inline void cgroup_ifs_account_smttime(struct task_struct *prev, struct task_struct *idle) {} static inline void cgroup_ifs_set_smt(cpumask_t *sibling) {} static inline void cgroup_ifs_account_rundelay(struct task_struct *task, u64 delta) {} +static inline void cgroup_ifs_account_throttle(struct cgroup *cgrp, int cpu, u64 delta) {} #ifdef CONFIG_SCHEDSTATS static inline void cgroup_ifs_account_sleep(struct task_struct *task, u64 delta) {} static inline void cgroup_ifs_enable_sleep_account(void) {} diff --git a/kernel/cgroup/ifs.c b/kernel/cgroup/ifs.c index c2de386315c2..dad42b38d16e 100644 --- a/kernel/cgroup/ifs.c +++ b/kernel/cgroup/ifs.c @@ -162,6 +162,9 @@ static const char *ifs_type_name(int type) case IFS_RUNDELAY: name = "rundelay"; break; + case IFS_THROTTLE: + name = "throttle"; + break; #ifdef CONFIG_SCHEDSTATS case IFS_SLEEP: name = "sleep"; diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index ddaa8dd71c3e..be95fd6394bc 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -6053,8 +6053,12 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq) raw_spin_lock(&cfs_b->lock); if (cfs_rq->throttled_clock) { - cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock; + u64 delta = rq_clock(rq) - cfs_rq->throttled_clock; + + cfs_b->throttled_time += delta; cfs_rq->throttled_clock = 0; + cgroup_ifs_account_throttle(cfs_rq->tg->css.cgroup, + cpu_of(rq), delta); } list_del_rcu(&cfs_rq->throttled_list); raw_spin_unlock(&cfs_b->lock); -- 2.34.1
From: Tengda Wu <wutengda2@huawei.com> hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Now that the throttle metric is in place, rundelay no longer needs to track throttle time. Signed-off-by: Tengda Wu <wutengda2@huawei.com> Signed-off-by: Pu Lehui <pulehui@huawei.com> --- include/linux/sched.h | 4 ++++ kernel/sched/core.c | 18 ++++++++++-------- kernel/sched/deadline.c | 4 ++-- kernel/sched/fair.c | 8 +++++--- kernel/sched/rt.c | 4 ++-- kernel/sched/sched.h | 11 ++++++++++- kernel/sched/stats.h | 27 +++++++++++++++++++++++++-- 7 files changed, 58 insertions(+), 18 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index a694cc11dea5..b31db870e23b 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -994,6 +994,10 @@ struct task_struct { /* delay due to memory thrashing */ unsigned in_thrashing:1; #endif +#ifdef CONFIG_CGROUP_IFS + /* Run delayed due to bandwidth throttling */ + KABI_FILL_HOLE(unsigned in_throttle:1) +#endif unsigned long atomic_flags; /* Flags requiring atomic access. */ diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 7b77fa528c3c..082056ba8bdc 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -154,6 +154,16 @@ const_debug unsigned int sysctl_sched_nr_migrate = SCHED_NR_MIGRATE_BREAK; __read_mostly int scheduler_running; +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) +int sched_task_is_throttled(struct task_struct *p, int cpu) +{ + if (p->sched_class->task_is_throttled) + return p->sched_class->task_is_throttled(p, cpu); + + return 0; +} +#endif + #ifdef CONFIG_SCHED_CORE DEFINE_STATIC_KEY_FALSE(__sched_core_enabled); @@ -268,14 +278,6 @@ void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags) resched_curr(rq); } -static int sched_task_is_throttled(struct task_struct *p, int cpu) -{ - if (p->sched_class->task_is_throttled) - return p->sched_class->task_is_throttled(p, cpu); - - return 0; -} - static struct task_struct *sched_core_next(struct task_struct *p, unsigned long cookie) { struct rb_node *node = &p->core_node; diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c index dd13bbe8c0e4..3aa12a73c9ce 100644 --- a/kernel/sched/deadline.c +++ b/kernel/sched/deadline.c @@ -2715,7 +2715,7 @@ static void prio_changed_dl(struct rq *rq, struct task_struct *p, #endif } -#ifdef CONFIG_SCHED_CORE +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) static int task_is_throttled_dl(struct task_struct *p, int cpu) { return p->dl.dl_throttled; @@ -2754,7 +2754,7 @@ DEFINE_SCHED_CLASS(dl) = { .switched_to = switched_to_dl, .update_curr = update_curr_dl, -#ifdef CONFIG_SCHED_CORE +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) .task_is_throttled = task_is_throttled_dl, #endif }; diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index be95fd6394bc..cf3f147cba4d 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -14348,7 +14348,11 @@ bool cfs_prio_less(const struct task_struct *a, const struct task_struct *b, return delta > 0; } +#else +static inline void task_tick_core(struct rq *rq, struct task_struct *curr) {} +#endif +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) static int task_is_throttled_fair(struct task_struct *p, int cpu) { struct cfs_rq *cfs_rq; @@ -14360,8 +14364,6 @@ static int task_is_throttled_fair(struct task_struct *p, int cpu) #endif return throttled_hierarchy(cfs_rq); } -#else -static inline void task_tick_core(struct rq *rq, struct task_struct *curr) {} #endif #ifdef CONFIG_SCHED_STEAL @@ -15096,7 +15098,7 @@ DEFINE_SCHED_CLASS(fair) = { .task_change_group = task_change_group_fair, #endif -#ifdef CONFIG_SCHED_CORE +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) .task_is_throttled = task_is_throttled_fair, #endif diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c index 77bb7ee8cce0..e2eaa8ffd009 100644 --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -2692,7 +2692,7 @@ static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) return 0; } -#ifdef CONFIG_SCHED_CORE +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) static int task_is_throttled_rt(struct task_struct *p, int cpu) { struct rt_rq *rt_rq; @@ -2740,7 +2740,7 @@ DEFINE_SCHED_CLASS(rt) = { .update_curr = update_curr_rt, -#ifdef CONFIG_SCHED_CORE +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) .task_is_throttled = task_is_throttled_rt, #endif diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 761870540a21..41ba8ad1a753 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -546,6 +546,15 @@ extern void sched_release_group(struct task_group *tg); extern void sched_move_task(struct task_struct *tsk); +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) +extern int sched_task_is_throttled(struct task_struct *p, int cpu); +#else +static inline int sched_task_is_throttled(struct task_struct *p, int cpu) +{ + return 0; +} +#endif + #ifdef CONFIG_QOS_SCHED_SMART_GRID extern void start_auto_affinity(struct auto_affinity *auto_affi); extern void stop_auto_affinity(struct auto_affinity *auto_affi); @@ -2451,7 +2460,7 @@ struct sched_class { void (*task_change_group)(struct task_struct *p); #endif -#ifdef CONFIG_SCHED_CORE +#if defined(CONFIG_SCHED_CORE) || defined(CONFIG_CGROUP_IFS) int (*task_is_throttled)(struct task_struct *p, int cpu); #endif KABI_RESERVE(1) diff --git a/kernel/sched/stats.h b/kernel/sched/stats.h index 6d40d4d55b27..12966561d5d4 100644 --- a/kernel/sched/stats.h +++ b/kernel/sched/stats.h @@ -134,6 +134,24 @@ __schedstats_from_se(struct sched_entity *se) #define QOS_THROTTLED 2 #endif +#ifdef CONFIG_CGROUP_IFS +static inline void ifs_account_rundelay(struct task_struct *task, u64 delta) +{ + /* + * No need to include bandwidth throttling time in rundelay, + * leave it to the throttle metric. + */ + if (unlikely(task->in_throttle)) { + task->in_throttle = 0; + return; + } + + cgroup_ifs_account_rundelay(task, delta); +} +#else +static inline void ifs_account_rundelay(struct task_struct *task, u64 delta) {} +#endif + #ifdef CONFIG_PSI void psi_task_change(struct task_struct *task, int clear, int set); void psi_task_switch(struct task_struct *prev, struct task_struct *next, @@ -241,7 +259,7 @@ static inline void sched_info_dequeue(struct rq *rq, struct task_struct *t) t->sched_info.last_queued = 0; t->sched_info.run_delay += delta; - cgroup_ifs_account_rundelay(t, delta); + ifs_account_rundelay(t, delta); rq_sched_info_dequeue(rq, delta); } @@ -264,7 +282,7 @@ static void sched_info_arrive(struct rq *rq, struct task_struct *t) t->sched_info.last_arrival = now; t->sched_info.pcount++; - cgroup_ifs_account_rundelay(t, delta); + ifs_account_rundelay(t, delta); rq_sched_info_arrive(rq, delta); } @@ -277,6 +295,11 @@ static inline void sched_info_enqueue(struct rq *rq, struct task_struct *t) { if (!t->sched_info.last_queued) t->sched_info.last_queued = rq_clock(rq); + +#ifdef CONFIG_CGROUP_IFS + if (!t->in_throttle && sched_task_is_throttled(t, task_cpu(t))) + t->in_throttle = 1; +#endif } /* -- 2.34.1
From: Tengda Wu <wutengda2@huawei.com> hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Measure the wakeup latency of tasks from sched_waking/sched_wakeup_new to running as a form of scheduling interference. Signed-off-by: Tengda Wu <wutengda2@huawei.com> Signed-off-by: Pu Lehui <pulehui@huawei.com> --- include/linux/cgroup.h | 17 +++++++++++++++++ include/linux/sched.h | 5 +++++ kernel/cgroup/ifs.c | 3 +++ kernel/sched/core.c | 3 +++ kernel/sched/stats.h | 29 ++++++++++++++++++++++++++++- 5 files changed, 56 insertions(+), 1 deletion(-) diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index 8fc762b3841c..5ac1b53c25e4 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -865,6 +865,7 @@ void cgroup_move_task_to_root(struct task_struct *tsk); enum ifs_types { IFS_SMT, IFS_RUNDELAY, + IFS_WAKELAT, IFS_THROTTLE, #ifdef CONFIG_SCHEDSTATS IFS_SLEEP, @@ -938,6 +939,21 @@ static inline void cgroup_ifs_account_rundelay(struct task_struct *task, cgroup_ifs_account_delta(this_cpu_ptr(ifs->pcpu), IFS_RUNDELAY, delta); } +static inline void cgroup_ifs_account_wakelat(struct task_struct *task, + u64 delta) +{ + struct cgroup_ifs *ifs; + + if (!cgroup_ifs_enabled()) + return; + + ifs = task_ifs(task); + if (!ifs) + return; + + cgroup_ifs_account_delta(this_cpu_ptr(ifs->pcpu), IFS_WAKELAT, delta); +} + static inline void cgroup_ifs_account_throttle(struct cgroup *cgrp, int cpu, u64 delta) { @@ -980,6 +996,7 @@ static inline void cgroup_ifs_account_smttime(struct task_struct *prev, struct task_struct *idle) {} static inline void cgroup_ifs_set_smt(cpumask_t *sibling) {} static inline void cgroup_ifs_account_rundelay(struct task_struct *task, u64 delta) {} +static inline void cgroup_ifs_account_wakelat(struct task_struct *task, u64 delta) {} static inline void cgroup_ifs_account_throttle(struct cgroup *cgrp, int cpu, u64 delta) {} #ifdef CONFIG_SCHEDSTATS static inline void cgroup_ifs_account_sleep(struct task_struct *task, u64 delta) {} diff --git a/include/linux/sched.h b/include/linux/sched.h index b31db870e23b..3979c34e9b83 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -390,7 +390,12 @@ struct sched_info { /* When were we last queued to run? */ unsigned long long last_queued; +#ifdef CONFIG_CGROUP_IFS + /* When were we last waking to run? */ + KABI_USE(1, unsigned long long last_waking) +#else KABI_RESERVE(1) +#endif KABI_RESERVE(2) #endif /* CONFIG_SCHED_INFO */ }; diff --git a/kernel/cgroup/ifs.c b/kernel/cgroup/ifs.c index dad42b38d16e..edebb5f00602 100644 --- a/kernel/cgroup/ifs.c +++ b/kernel/cgroup/ifs.c @@ -162,6 +162,9 @@ static const char *ifs_type_name(int type) case IFS_RUNDELAY: name = "rundelay"; break; + case IFS_WAKELAT: + name = "wakelat"; + break; case IFS_THROTTLE: name = "throttle"; break; diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 082056ba8bdc..6c9da25a3d45 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4246,6 +4246,7 @@ int try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) trace_sched_waking(p); ttwu_do_wakeup(p); + ifs_task_waking(p); goto out; } @@ -4261,6 +4262,7 @@ int try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) break; trace_sched_waking(p); + ifs_task_waking(p); /* * Ensure we load p->on_rq _after_ p->state, otherwise it would @@ -4935,6 +4937,7 @@ void wake_up_new_task(struct task_struct *p) activate_task(rq, p, ENQUEUE_NOCLOCK); trace_sched_wakeup_new(p); + ifs_task_waking(p); check_preempt_curr(rq, p, WF_FORK); #ifdef CONFIG_SMP if (p->sched_class->task_woken) { diff --git a/kernel/sched/stats.h b/kernel/sched/stats.h index 12966561d5d4..1b736b931247 100644 --- a/kernel/sched/stats.h +++ b/kernel/sched/stats.h @@ -2,6 +2,8 @@ #ifndef _KERNEL_STATS_H #define _KERNEL_STATS_H +#include <linux/sched/clock.h> + #ifdef CONFIG_SCHEDSTATS extern struct static_key_false sched_schedstats; @@ -148,8 +150,31 @@ static inline void ifs_account_rundelay(struct task_struct *task, u64 delta) cgroup_ifs_account_rundelay(task, delta); } + +static inline void ifs_task_waking(struct task_struct *t) +{ + if (t->sched_info.last_waking) + return; + + t->sched_info.last_waking = sched_clock(); +} + +static inline void ifs_task_arrive(struct task_struct *t) +{ + unsigned long long now, delta = 0; + + if (!t->sched_info.last_waking) + return; + + now = sched_clock(); + delta = now - t->sched_info.last_waking; + t->sched_info.last_waking = 0; + cgroup_ifs_account_wakelat(t, delta); +} #else static inline void ifs_account_rundelay(struct task_struct *task, u64 delta) {} +static inline void ifs_task_waking(struct task_struct *t) {} +static inline void ifs_task_arrive(struct task_struct *t) {} #endif #ifdef CONFIG_PSI @@ -336,8 +361,10 @@ sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *n if (prev != rq->idle) sched_info_depart(rq, prev); - if (next != rq->idle) + if (next != rq->idle) { sched_info_arrive(rq, next); + ifs_task_arrive(next); + } } #else /* !CONFIG_SCHED_INFO: */ -- 2.34.1
From: Xu Kuohai <xukuohai@huawei.com> hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Add softirq and hardirq interference track support. Signed-off-by: Xu Kuohai <xukuohai@huawei.com> Signed-off-by: Pu Lehui <pulehui@huawei.com> --- include/linux/cgroup.h | 45 ++++++++++++++++++++++++++++++++++++++++++ kernel/cgroup/ifs.c | 23 +++++++++++++++++++++ kernel/sched/cputime.c | 8 ++++++-- 3 files changed, 74 insertions(+), 2 deletions(-) diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index 5ac1b53c25e4..d6f43bca5ecb 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -867,6 +867,10 @@ enum ifs_types { IFS_RUNDELAY, IFS_WAKELAT, IFS_THROTTLE, +#ifdef CONFIG_IRQ_TIME_ACCOUNTING + IFS_SOFTIRQ, + IFS_HARDIRQ, +#endif #ifdef CONFIG_SCHEDSTATS IFS_SLEEP, #endif @@ -971,6 +975,42 @@ static inline void cgroup_ifs_account_throttle(struct cgroup *cgrp, int cpu, cgroup_ifs_account_delta(ifsc, IFS_THROTTLE, delta); } +#ifdef CONFIG_IRQ_TIME_ACCOUNTING +static inline void cgroup_ifs_account_softirq(u64 delta) +{ + struct cgroup_ifs *ifs; + struct cgroup_ifs_cpu *ifsc; + + if (!cgroup_ifs_enabled()) + return; + + ifs = current_ifs(); + if (!ifs) + return; + + ifsc = this_cpu_ptr(ifs->pcpu); + cgroup_ifs_account_delta(ifsc, IFS_SOFTIRQ, delta); +} + +static inline void cgroup_ifs_account_hardirq(u64 delta) +{ + struct cgroup_ifs *ifs; + struct cgroup_ifs_cpu *ifsc; + + if (!cgroup_ifs_enabled()) + return; + + ifs = current_ifs(); + if (!ifs) + return; + + ifsc = this_cpu_ptr(ifs->pcpu); + cgroup_ifs_account_delta(ifsc, IFS_HARDIRQ, delta); +} + +void cgroup_ifs_enable_irq_account(void); +#endif + #ifdef CONFIG_SCHEDSTATS static inline void cgroup_ifs_account_sleep(struct task_struct *task, u64 delta) @@ -998,6 +1038,11 @@ static inline void cgroup_ifs_set_smt(cpumask_t *sibling) {} static inline void cgroup_ifs_account_rundelay(struct task_struct *task, u64 delta) {} static inline void cgroup_ifs_account_wakelat(struct task_struct *task, u64 delta) {} static inline void cgroup_ifs_account_throttle(struct cgroup *cgrp, int cpu, u64 delta) {} +#ifdef CONFIG_IRQ_TIME_ACCOUNTING +static inline void cgroup_ifs_account_softirq(u64 delta) {} +static inline void cgroup_ifs_account_hardirq(u64 delta) {} +static inline void cgroup_ifs_enable_irq_account(void) {} +#endif #ifdef CONFIG_SCHEDSTATS static inline void cgroup_ifs_account_sleep(struct task_struct *task, u64 delta) {} static inline void cgroup_ifs_enable_sleep_account(void) {} diff --git a/kernel/cgroup/ifs.c b/kernel/cgroup/ifs.c index edebb5f00602..7ac051913920 100644 --- a/kernel/cgroup/ifs.c +++ b/kernel/cgroup/ifs.c @@ -36,6 +36,10 @@ static bool ifs_enable; static bool ifs_sleep_enable; #endif +#ifdef CONFIG_IRQ_TIME_ACCOUNTING +static bool ifs_irq_enable; +#endif + static int __init setup_ifs(char *str) { return kstrtobool(str, &ifs_enable) == 0; @@ -168,6 +172,14 @@ static const char *ifs_type_name(int type) case IFS_THROTTLE: name = "throttle"; break; +#ifdef CONFIG_IRQ_TIME_ACCOUNTING + case IFS_SOFTIRQ: + name = "softirq"; + break; + case IFS_HARDIRQ: + name = "hardirq"; + break; +#endif #ifdef CONFIG_SCHEDSTATS case IFS_SLEEP: name = "sleep"; @@ -185,6 +197,10 @@ static bool should_print(int type) #ifdef CONFIG_SCHEDSTATS if (type == IFS_SLEEP) return ifs_sleep_enable; +#endif +#ifdef CONFIG_IRQ_TIME_ACCOUNTING + if (type == IFS_SOFTIRQ || type == IFS_HARDIRQ) + return ifs_irq_enable; #endif return true; } @@ -304,3 +320,10 @@ void cgroup_ifs_enable_sleep_account(void) ifs_sleep_enable = true; } #endif + +#ifdef CONFIG_IRQ_TIME_ACCOUNTING +void cgroup_ifs_enable_irq_account(void) +{ + ifs_irq_enable = true; +} +#endif diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c index b453f8a6a7c7..499812509f7d 100644 --- a/kernel/sched/cputime.c +++ b/kernel/sched/cputime.c @@ -27,6 +27,7 @@ static int sched_clock_irqtime; void enable_sched_clock_irqtime(void) { sched_clock_irqtime = 1; + cgroup_ifs_enable_irq_account(); } void disable_sched_clock_irqtime(void) @@ -71,10 +72,13 @@ void irqtime_account_irq(struct task_struct *curr, unsigned int offset) * in that case, so as not to confuse scheduler with a special task * that do not consume any time, but still wants to run. */ - if (pc & HARDIRQ_MASK) + if (pc & HARDIRQ_MASK) { irqtime_account_delta(irqtime, delta, CPUTIME_IRQ); - else if ((pc & SOFTIRQ_OFFSET) && curr != this_cpu_ksoftirqd()) + cgroup_ifs_account_hardirq(delta); + } else if ((pc & SOFTIRQ_OFFSET) && curr != this_cpu_ksoftirqd()) { irqtime_account_delta(irqtime, delta, CPUTIME_SOFTIRQ); + cgroup_ifs_account_softirq(delta); + } } static u64 irqtime_tick_accounted(u64 maxtime) -- 2.34.1
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Add cgroup v1 support for CGROUP_IFS. Signed-off-by: Pu Lehui <pulehui@huawei.com> --- kernel/cgroup/ifs.c | 10 ++++++++++ kernel/sched/cpuacct.c | 15 +++++++++++++++ 2 files changed, 25 insertions(+) diff --git a/kernel/cgroup/ifs.c b/kernel/cgroup/ifs.c index 7ac051913920..6306b8fa5076 100644 --- a/kernel/cgroup/ifs.c +++ b/kernel/cgroup/ifs.c @@ -282,6 +282,16 @@ static struct cftype cgroup_ifs_files[] = { { } /* terminate */ }; +struct cftype cgroup_v1_ifs_files[] = { + { + .name = "interference.stat", + .flags = CFTYPE_NO_PREFIX, + .seq_show = cgroup_ifs_show, + .write = cgroup_ifs_write, + }, + { } /* terminate */ +}; + void cgroup_ifs_init(void) { if (!ifs_enable) diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c index 2a1e44822bc1..6fee560173c6 100644 --- a/kernel/sched/cpuacct.c +++ b/kernel/sched/cpuacct.c @@ -386,3 +386,18 @@ static int __init cgroup_v1_psi_init(void) late_initcall_sync(cgroup_v1_psi_init); #endif + +#ifdef CONFIG_CGROUP_IFS +extern struct cftype cgroup_v1_ifs_files[]; + +static int __init cgroup_v1_ifs_init(void) +{ + if (!cgroup_ifs_enabled()) + return 0; + + cgroup_add_legacy_cftypes(&cpuacct_cgrp_subsys, cgroup_v1_ifs_files); + + return 0; +} +late_initcall_sync(cgroup_v1_ifs_init); +#endif -- 2.34.1
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICAOAT -------------------------------- Disable CGROUP_IFS config by default in openeuler_defconfig. Signed-off-by: Pu Lehui <pulehui@huawei.com> --- arch/arm64/configs/openeuler_defconfig | 1 + arch/x86/configs/openeuler_defconfig | 1 + 2 files changed, 2 insertions(+) diff --git a/arch/arm64/configs/openeuler_defconfig b/arch/arm64/configs/openeuler_defconfig index 8906984a3399..b170a3f4ad7d 100644 --- a/arch/arm64/configs/openeuler_defconfig +++ b/arch/arm64/configs/openeuler_defconfig @@ -196,6 +196,7 @@ CONFIG_SOCK_CGROUP_DATA=y CONFIG_CGROUP_V1_KILL=y CONFIG_CGROUP_V1_STAT=y CONFIG_CGROUP_FILES=y +# CONFIG_CGROUP_IFS is not set CONFIG_UCOUNTS_PERCPU_COUNTER=y CONFIG_NAMESPACES=y CONFIG_UTS_NS=y diff --git a/arch/x86/configs/openeuler_defconfig b/arch/x86/configs/openeuler_defconfig index 0513530b18b6..8d5434651880 100644 --- a/arch/x86/configs/openeuler_defconfig +++ b/arch/x86/configs/openeuler_defconfig @@ -214,6 +214,7 @@ CONFIG_SOCK_CGROUP_DATA=y CONFIG_CGROUP_V1_KILL=y CONFIG_CGROUP_V1_STAT=y CONFIG_CGROUP_FILES=y +# CONFIG_CGROUP_IFS is not set CONFIG_UCOUNTS_PERCPU_COUNTER=y CONFIG_NAMESPACES=y CONFIG_UTS_NS=y -- 2.34.1
反馈： 您发送到kernel@openeuler.org的补丁/补丁集，已成功转换为PR！ PR链接地址： https://gitee.com/openeuler/kernel/pulls/16489 邮件列表地址：https://mailweb.openeuler.org/archives/list/kernel@openeuler.org/message/H6M... FeedBack: The patch(es) which you have sent to kernel@openeuler.org mailing list has been converted to a pull request successfully! Pull request link: https://gitee.com/openeuler/kernel/pulls/16489 Mailing list address: https://mailweb.openeuler.org/archives/list/kernel@openeuler.org/message/H6M...
participants (2)
-
patchwork bot -
Pu Lehui