Unverified Commit 518c4345 authored by openeuler-ci-bot's avatar openeuler-ci-bot Committed by Gitee
Browse files

!7916 v5 Introduce NUMA isolation and consolidation 

Merge Pull Request from: @ci-robot 
 
PR sync from: Hui Tang <tanghui20@huawei.com>
https://mailweb.openeuler.org/hyperkitty/list/kernel@openeuler.org/message/Q3Z5CC6SXAGYZMBWKVRFUSZBV3W7HLML/ 
Introduce NUMA isolation and consolidation.

Guan Jing (1):
  sched: Add can_migrate_task hook

Hui Tang (10):
  sched: Introduce CONFIG_TASK_PLACEMENT_BY_CPU_RANGE
  sched: Some fixes for select_rq hook
  bpf:programmable: Add nodemask operation collection
  sched: Introduce task relationship by net and memory
  bpf:programmable: Add helper to get memory and net relationship
  sched: Add ioctl to get relationship
  sched: Update numa group preferred node periodically
  bpf:programmable: Add helper to set preferred node
  sched: Introduce CONFIG_QOS_SCHED_NUMA_ICON
  config: Enable NUMA isolation and consolidation by default


-- 
2.34.1
 
https://gitee.com/openeuler/kernel/issues/I9GZAQ 
 
Link:https://gitee.com/openeuler/kernel/pulls/7916

 

Reviewed-by: default avatarZhang Jianhua <chris.zjh@huawei.com>
Reviewed-by: default avatarZucheng Zheng <zhengzucheng@huawei.com>
Reviewed-by: default avatarzhangyi (F) <yi.zhang@huawei.com>
Signed-off-by: default avatarJialin Zhang <zhangjialin11@huawei.com>
parents d07e4530 aab27d37

arch/arm64/configs/openeuler_defconfig

+3 −1
Original line number Diff line number Diff line
@@ -162,6 +162,8 @@ CONFIG_FAIR_GROUP_SCHED=y 
CONFIG_CFS_BANDWIDTH=y

CONFIG_RT_GROUP_SCHED=y

CONFIG_QOS_SCHED_DYNAMIC_AFFINITY=y

CONFIG_SCHED_TASK_RELATIONSHIP=y

CONFIG_QOS_SCHED_NUMA_ICON=y

CONFIG_QOS_SCHED_SMART_GRID=y

CONFIG_CGROUP_PIDS=y

CONFIG_CGROUP_RDMA=y
@@ -234,7 +236,7 @@ CONFIG_KALLSYMS=y 
CONFIG_KALLSYMS_ALL=y

CONFIG_KALLSYMS_BASE_RELATIVE=y

# CONFIG_BPF_LSM is not set

# CONFIG_BPF_SCHED is not set

CONFIG_BPF_SCHED=y

CONFIG_BPF_SYSCALL=y

CONFIG_ARCH_WANT_DEFAULT_BPF_JIT=y

CONFIG_BPF_JIT_ALWAYS_ON=y

arch/x86/configs/openeuler_defconfig

+2 −0
Original line number Diff line number Diff line
@@ -167,6 +167,8 @@ CONFIG_FAIR_GROUP_SCHED=y 
CONFIG_CFS_BANDWIDTH=y

CONFIG_RT_GROUP_SCHED=y

CONFIG_QOS_SCHED_DYNAMIC_AFFINITY=y

# CONFIG_SCHED_TASK_RELATIONSHIP is not set

# CONFIG_QOS_SCHED_NUMA_ICON is not set

# CONFIG_QOS_SCHED_SMART_GRID is not set

CONFIG_CGROUP_PIDS=y

CONFIG_CGROUP_RDMA=y

fs/exec.c

+2 −0
Original line number Diff line number Diff line
@@ -38,6 +38,7 @@ 
#include <linux/sched/coredump.h>

#include <linux/sched/signal.h>

#include <linux/sched/numa_balancing.h>

#include <linux/sched/relationship.h>

#include <linux/sched/task.h>

#include <linux/pagemap.h>

#include <linux/perf_event.h>
@@ -1822,6 +1823,7 @@ static int bprm_execve(struct linux_binprm *bprm, 
	rseq_execve(current);

	acct_update_integrals(current);

	task_numa_free(current, false);

	task_relationship_free(current, true);

	return retval;



out:

include/linux/sched.h

+57 −1
Original line number Diff line number Diff line
@@ -26,6 +26,7 @@ 
#include <linux/resource.h>

#include <linux/latencytop.h>

#include <linux/sched/prio.h>

#include <linux/sched/relationship.h>

#include <linux/sched/types.h>

#include <linux/signal_types.h>

#include <linux/mm_types_task.h>
@@ -1437,11 +1438,15 @@ struct task_struct { 
	KABI_USE(7, void *pf_io_worker)

#if defined(CONFIG_QOS_SCHED_DYNAMIC_AFFINITY) && !defined(__GENKSYMS__)

	KABI_USE(8, cpumask_t *prefer_cpus)

	KABI_USE(9, const cpumask_t *select_cpus)

#else

	KABI_RESERVE(8)

#endif

#if defined(CONFIG_TASK_PLACEMENT_BY_CPU_RANGE) && !defined(__GENKSYMS__)

	KABI_USE(9, const cpumask_t *select_cpus)

#else

	KABI_RESERVE(9)

#endif



#if (defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)) && defined(CONFIG_X86)

	KABI_USE(10, unsigned int sequential_io)

	KABI_USE(11, unsigned int sequential_io_avg)
@@ -1464,7 +1469,11 @@ struct task_struct { 
#else

	KABI_RESERVE(13)

#endif

#if defined(CONFIG_SCHED_TASK_RELATIONSHIP) && !defined(__GENKSYMS__)

	KABI_USE(14, struct task_relationship *rship)

#else

	KABI_RESERVE(14)

#endif

	KABI_RESERVE(15)

	KABI_RESERVE(16)

	KABI_AUX_PTR(task_struct)
@@ -2351,6 +2360,21 @@ struct bpf_sched_cpu_stats { 
	KABI_RESERVE(4)

};



struct bpf_node_stats {

	unsigned long util;

	unsigned long compute_capacity;

	unsigned int weight;



	KABI_RESERVE(1)

	KABI_RESERVE(2)

	KABI_RESERVE(3)

	KABI_RESERVE(4)

	KABI_RESERVE(5)

	KABI_RESERVE(6)

	KABI_RESERVE(7)

	KABI_RESERVE(8)

};



struct cpumask_op_args {

	unsigned int op_type;

	void *arg1;
@@ -2374,6 +2398,28 @@ enum cpumask_op_type { 
	CPUMASK_CPULIST_PARSE

};



enum nodemask_op_type {

	NODEMASK_EMPTY,

	NODEMASK_NODE_ISSET,

	NODEMASK_NODES_CLEAR,

	NODEMASK_NODE_SET,

	NODEMASK_NODE_CLEAR,

	NODEMASK_NODELIST_PARSE,

	NODEMASK_TO_CPUMASK,

	NODEMASK_NODES_ANDNOT,

	NODEMASK_NODES_AND,

	NODEMASK_NODES_OR,

	NODEMASK_WEIGHT,

	NODEMASK_ONLINE

};



struct nodemask_op_args {

	enum nodemask_op_type op_type;

	void *arg1;

	void *arg2;

	void *arg3;

};



struct sched_migrate_ctx {

	struct task_struct *task;

	struct cpumask *select_idle_mask;
@@ -2402,5 +2448,15 @@ struct sched_affine_ctx { 
	KABI_RESERVE(3)

	KABI_RESERVE(4)

};



struct sched_migrate_node {

	int src_cpu;

	int dst_cpu;



	KABI_RESERVE(1)

	KABI_RESERVE(2)

	KABI_RESERVE(3)

	KABI_RESERVE(4)

};

#endif

#endif

include/linux/sched/relationship.h

0 → 100644
+202 −0
Original line number Diff line number Diff line 
/* SPDX-License-Identifier: GPL-2.0 */

#ifndef _LINUX_SCHED_RELATIONSHIP_H

#define _LINUX_SCHED_RELATIONSHIP_H



#include <linux/nodemask.h>

#include <linux/jump_label.h>

#include <linux/refcount.h>

#include <uapi/linux/sched_ctrl.h>



#define FAULT_NODES_MAX 4



struct task_struct;

struct rq;



#ifdef CONFIG_SCHED_DEBUG

struct seq_file;

#endif



struct fault_array_info {

	int nid;

	unsigned long val;

};



struct relationship_comm {

	int nr_tasks;

	int gid;

	nodemask_t preferred_node;

};



struct bpf_net_relationship {

	struct relationship_comm comm;

	unsigned long grp_rxtx_bytes;

	unsigned long grp_remote_rxtx_bytes;

};



struct bpf_mm_relationship {

	struct relationship_comm comm;

	unsigned long grp_total_faults;

	struct fault_array_info grp_faults_ordered[FAULT_NODES_MAX];

	struct fault_array_info grp_score_ordered[FAULT_NODES_MAX];

};



struct bpf_relationship_get_args {

	struct bpf_mm_relationship mm;

	struct bpf_net_relationship net;

};



struct bpf_relationship_set_args {

	nodemask_t preferred_node;

};



struct relationship_hdr {

	refcount_t refcount;

	spinlock_t lock;

	int nr_tasks;

	int gid;

	nodemask_t preferred_nid;

};



enum net_req_type {

	NET_RS_TYPE_INVALID = 0,

	NET_RS_TYPE_LOCAL,

	NET_RS_TYPE_RX,

	NET_RS_TYPE_TX,

	NET_RS_TYPE_MAX

};



struct net_relationship_req {

	enum net_req_type net_rship_type;

	pid_t rx_pid;

	pid_t tx_pid;

	int nic_nid;

	int rx_dev_idx;

	int rx_dev_queue_idx;

	u64 rx_dev_netns_cookie;

	unsigned long rxtx_bytes;



	/* reserved */

	unsigned long rxtx_cnt;

};



struct net_relationship_callback {

	struct callback_head twork;

	atomic_t active;

	pid_t src_pid;

	struct net_relationship_req req;

};



struct net_group {

	struct rcu_head rcu;

	struct relationship_hdr hdr;

	unsigned long rxtx_bytes;



	/* reserved */

	unsigned long rxtx_cnt;

};



struct numa_fault_ext {

	struct fault_array_info faults_ordered[FAULT_NODES_MAX];

};



struct task_relationship {

	/* network relationship */

	struct net_group __rcu *net_group;

	spinlock_t net_lock;

	int nic_nid;

	int rx_dev_idx;

	int rx_dev_queue_idx;

	unsigned long rx_dev_netns_cookie;

	unsigned long rxtx_remote_bytes;

	unsigned long rxtx_remote_update_next;

	unsigned long rxtx_remote_buffer;

	unsigned long rxtx_bytes;

	unsigned long rxtx_buffer;

	unsigned long rxtx_update_next;

	struct net_relationship_callback cb;



	/* extras numa fault data */

	struct numa_fault_ext faults;



#ifdef CONFIG_NUMA_BALANCING

	/* preferred nodes adjust */

	u64 node_stamp;

	struct callback_head node_work;

#endif

};



#ifdef CONFIG_BPF_SCHED

struct sched_preferred_node_ctx {

	struct task_struct *tsk;

	nodemask_t preferred_node;



	KABI_RESERVE(1)

	KABI_RESERVE(2)

	KABI_RESERVE(3)

	KABI_RESERVE(4)

};

#endif



extern void task_relationship_enable(void);

extern void task_relationship_disable(void);



#ifdef CONFIG_SCHED_DEBUG

extern void sched_show_relationship(struct task_struct *p, struct seq_file *m);

#endif



#ifdef CONFIG_SCHED_TASK_RELATIONSHIP

extern int sched_relationship_fork(struct task_struct *p);

extern void sched_relationship_free(struct task_struct *p);

void task_relationship_free(struct task_struct *tsk, bool reset);

extern bool task_relationship_supported(struct task_struct *tsk);

extern int sched_net_relationship_submit(struct net_relationship_req *req);

extern void

sctl_sched_get_net_relationship(struct task_struct *tsk,

				struct sctl_net_relationship_info *info);

extern void

sctl_sched_get_mem_relationship(struct task_struct *tsk,

				struct sctl_mem_relationship_info *info);

extern void sched_get_mm_relationship(struct task_struct *tsk,

			       struct bpf_relationship_get_args *args);

extern void sched_get_relationship(struct task_struct *tsk,

				   struct bpf_relationship_get_args *args);

extern void numa_faults_update_and_sort(int nid, int new,

					  struct fault_array_info *stats);

extern void task_tick_relationship(struct rq *rq, struct task_struct *curr);



extern void task_preferred_node_work(struct callback_head *work);

extern void

sched_set_curr_preferred_node(struct bpf_relationship_set_args *args);



DECLARE_STATIC_KEY_FALSE(__relationship_switch);

static inline bool task_relationship_used(void)

{

	return static_branch_unlikely(&__relationship_switch);

}

#else

static inline bool task_relationship_used(void)

{

	return false;

}



static inline int sched_relationship_fork(struct task_struct *p)

{

	return 0;

}



static inline void sched_relationship_free(struct task_struct *p) {}



static inline void

task_relationship_free(struct task_struct *tsk, bool reset) {}



static inline int

sched_net_relationship_submit(struct net_relationship_req *req)

{

	return 0;

}



static inline void

task_tick_relationship(struct rq *rq, struct task_struct *curr) {}

#endif



#endif