Commit 9de44ea0 authored by Hui Tang's avatar Hui Tang
Browse files

bpf:programmable: Add nodemask operation collection 

hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I9GZAQ


CVE: NA

--------------------------------

Introduce the helper to process a series nodemask operations,
such as 'nodes_empty', 'node_isset' etc.

Signed-off-by: default avatarHui Tang <tanghui20@huawei.com>
parent 7743b2f7

include/linux/sched.h

+22 −0
Original line number Diff line number Diff line
@@ -2378,6 +2378,28 @@ 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;

include/uapi/linux/bpf.h

+29 −0
Original line number Diff line number Diff line
@@ -3874,6 +3874,34 @@ union bpf_attr { 
 *		check src_cpu whether share cache with dst_cpu.

 *	Return

 *		yes 1, no 0.

 *

 * int bpf_nodemask_op(struct nodemask_op_args *op, int len)

 *	Description

 *		A series of nodemask-related operations. Perform different

 *		operations base on *op*->type. User also need fill other

 *		*op* field base on *op*->type. *op*->type is one of them

 *

 *		**NODEMASK_EMPTY**

 *			nodes_empty(op->arg1) returned.

 *		**NODEMASK_NODE_ISSET**

 *			node_isset(op->arg1, op->arg2) returned

 *		**NODEMASK_NODES_CLEAR**

 *			0 returned

 *		**NODEMASK_NODE_CLEAR**

 *			unset op->arg1 from op->arg2, 0 returned

 *		**NODEMASK_NODE_SET**

 *			set op->arg1 to op->arg2, 0 returned

 *		**NODEMASK_WEIGHT**

 *			nodes_weight(op->arg1) returned

 *		**NODEMASK_NODELIST_PARSE**

 *			str *op->arg1* to nodemask_t *op->arg2*,

 *			0 on success, or a negative error in case of failure.

 *		**NODEMASK_TO_CPUMASK**

 *			nodemask_t *arg1* to cpumask_t *op->arg2*, 0 returned.

 *		**NODEMASK_ONLINE**

 *			set online nodes to nodemask_t *op->arg1*, 0 returned.

 *	Return

 *		View above.

 */

#define __BPF_FUNC_MAPPER(FN)		\

	FN(unspec),			\
@@ -4046,6 +4074,7 @@ union bpf_attr { 
	FN(sched_entity_to_tg),		\

	FN(cpumask_op),			\

	FN(cpus_share_cache),		\

	FN(nodemask_op),		\

	/* */



/* integer value in 'imm' field of BPF_CALL instruction selects which helper

kernel/sched/bpf_sched.c

+71 −0
Original line number Diff line number Diff line
@@ -260,6 +260,75 @@ static const struct bpf_func_proto bpf_cpumask_op_proto = { 
	.arg2_type	= ARG_CONST_SIZE,

};



BPF_CALL_2(bpf_nodemask_op, struct nodemask_op_args *, op, int, len)

{

	struct cpumask *cpumask;

	nodemask_t mask;

	int nid;



	if (len != sizeof(*op) || !op->arg1)

		return -EINVAL;



	switch (op->op_type) {

	case NODEMASK_EMPTY:

		mask = *(nodemask_t *)op->arg1;

		return nodes_empty(mask);

	case NODEMASK_NODE_ISSET:

		mask = *(nodemask_t *)op->arg2;

		return node_isset(*(int *)op->arg1, mask);

	case NODEMASK_NODES_CLEAR:

		__nodes_clear((nodemask_t *)op->arg1, MAX_NUMNODES);

		break;

	case NODEMASK_NODE_CLEAR:

		__node_clear(*(int *)op->arg1, (nodemask_t *)op->arg2);

		break;

	case NODEMASK_NODE_SET:

		__node_set(*(int *)op->arg1, (nodemask_t *)op->arg2);

		break;

	case NODEMASK_NODES_AND:

		__nodes_and((nodemask_t *)op->arg1, (nodemask_t *)op->arg2,

			    (nodemask_t *)op->arg3, MAX_NUMNODES);

		break;

	case NODEMASK_NODES_ANDNOT:

		__nodes_andnot((nodemask_t *)op->arg1, (nodemask_t *)op->arg2,

			       (nodemask_t *)op->arg3, MAX_NUMNODES);

		break;

	case NODEMASK_NODES_OR:

		__nodes_or((nodemask_t *)op->arg1, (nodemask_t *)op->arg2,

			   (nodemask_t *)op->arg3, MAX_NUMNODES);

		break;

	case NODEMASK_WEIGHT:

		mask = *(nodemask_t *)op->arg1;

		return nodes_weight(mask);

	case NODEMASK_NODELIST_PARSE:

		return __nodelist_parse((const char *)op->arg1,

					(nodemask_t *)op->arg2, MAX_NUMNODES);

	case NODEMASK_TO_CPUMASK:

		mask = *(nodemask_t *)op->arg1;

		cpumask = (struct cpumask *)op->arg2;

		cpumask_clear(cpumask);

		for_each_node_mask(nid, mask) {

			cpumask_or(cpumask, cpumask, cpumask_of_node(nid));

		}

		break;

	case NODEMASK_ONLINE:

		*(nodemask_t *)op->arg1 = node_online_map;

		break;

	default:

		return -EINVAL;

	}



	return 0;

}



static const struct bpf_func_proto bpf_nodemask_op_proto = {

	.func		= bpf_nodemask_op,

	.gpl_only	= false,

	.ret_type	= RET_INTEGER,

	.arg1_type	= ARG_PTR_TO_MEM,

	.arg2_type	= ARG_CONST_SIZE,

};



BPF_CALL_2(bpf_cpus_share_cache, int, src_cpu, int, dst_cpu)

{

	if ((unsigned int)src_cpu >= nr_cpu_ids ||
@@ -299,6 +368,8 @@ bpf_sched_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 
		return &bpf_cpumask_op_proto;

	case BPF_FUNC_cpus_share_cache:

		return &bpf_cpus_share_cache_proto;

	case BPF_FUNC_nodemask_op:

		return &bpf_nodemask_op_proto;

	default:

		return bpf_base_func_proto(func_id);

	}

scripts/bpf_helpers_doc.py

+2 −0
Original line number Diff line number Diff line
@@ -445,6 +445,7 @@ class PrinterHelpers(Printer): 
            'struct sched_migrate_ctx',

            'struct sched_affine_ctx',

            'struct sched_migrate_node',

            'struct nodemask_op_args',

    ]

    known_types = {

            '...',
@@ -498,6 +499,7 @@ class PrinterHelpers(Printer): 
            'struct sched_migrate_ctx',

            'struct sched_affine_ctx',

            'struct sched_migrate_node',

            'struct nodemask_op_args',

    }

    mapped_types = {

            'u8': '__u8',

tools/include/uapi/linux/bpf.h

+29 −0
Original line number Diff line number Diff line
@@ -3874,6 +3874,34 @@ union bpf_attr { 
 *		check src_cpu whether share cache with dst_cpu.

 *	Return

 *		true yes, false no.

 *

 * int bpf_nodemask_op(struct nodemask_op_args *op, int len)

 *	Description

 *		A series of nodemask-related operations. Perform different

 *		operations base on *op*->type. User also need fill other

 *		*op* field base on *op*->type. *op*->type is one of them

 *

 *		**NODEMASK_EMPTY**

 *			nodes_empty(op->arg1) returned.

 *		**NODEMASK_NODE_ISSET**

 *			node_isset(op->arg1, op->arg2) returned

 *		**NODEMASK_NODES_CLEAR**

 *			0 returned

 *		**NODEMASK_NODE_CLEAR**

 *			unset op->arg1 from op->arg2, 0 returned

 *		**NODEMASK_NODE_SET**

 *			set op->arg1 to op->arg2, 0 returned

 *		**NODEMASK_WEIGHT**

 *			nodes_weight(op->arg1) returned

 *		**NODEMASK_NODELIST_PARSE**

 *			str *op->arg1* to nodemask_t *op->arg2*,

 *			0 on success, or a negative error in case of failure.

 *		**NODEMASK_TO_CPUMASK**

 *			nodemask_t *arg1* to cpumask_t *op->arg2*, 0 returned.

 *		**NODEMASK_ONLINE**

 *			set online nodes to nodemask_t *op->arg1*, 0 returned.

 *	Return

 *		View above.

 */

#define __BPF_FUNC_MAPPER(FN)		\

	FN(unspec),			\
@@ -4046,6 +4074,7 @@ union bpf_attr { 
	FN(sched_entity_to_tg),		\

	FN(cpumask_op),			\

	FN(cpus_share_cache),		\

	FN(nodemask_op),		\

	/* */



/* integer value in 'imm' field of BPF_CALL instruction selects which helper