xref: /linux/kernel/time/timer_migration.h (revision a23e1966932464e1c5226cb9ac4ce1d5fc10ba22)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 #ifndef _KERNEL_TIME_MIGRATION_H
3 #define _KERNEL_TIME_MIGRATION_H
4 
5 /* Per group capacity. Must be a power of 2! */
6 #define TMIGR_CHILDREN_PER_GROUP 8
7 
8 /**
9  * struct tmigr_event - a timer event associated to a CPU
10  * @nextevt:	The node to enqueue an event in the parent group queue
11  * @cpu:	The CPU to which this event belongs
12  * @ignore:	Hint whether the event could be ignored; it is set when
13  *		CPU or group is active;
14  */
15 struct tmigr_event {
16 	struct timerqueue_node	nextevt;
17 	unsigned int		cpu;
18 	bool			ignore;
19 };
20 
21 /**
22  * struct tmigr_group - timer migration hierarchy group
23  * @lock:		Lock protecting the event information and group hierarchy
24  *			information during setup
25  * @parent:		Pointer to the parent group
26  * @groupevt:		Next event of the group which is only used when the
27  *			group is !active. The group event is then queued into
28  *			the parent timer queue.
29  *			Ignore bit of @groupevt is set when the group is active.
30  * @next_expiry:	Base monotonic expiry time of the next event of the
31  *			group; It is used for the racy lockless check whether a
32  *			remote expiry is required; it is always reliable
33  * @events:		Timer queue for child events queued in the group
34  * @migr_state:		State of the group (see union tmigr_state)
35  * @level:		Hierarchy level of the group; Required during setup
36  * @numa_node:		Required for setup only to make sure CPU and low level
37  *			group information is NUMA local. It is set to NUMA node
38  *			as long as the group level is per NUMA node (level <
39  *			tmigr_crossnode_level); otherwise it is set to
40  *			NUMA_NO_NODE
41  * @num_children:	Counter of group children to make sure the group is only
42  *			filled with TMIGR_CHILDREN_PER_GROUP; Required for setup
43  *			only
44  * @childmask:		childmask of the group in the parent group; is set
45  *			during setup and will never change; can be read
46  *			lockless
47  * @list:		List head that is added to the per level
48  *			tmigr_level_list; is required during setup when a
49  *			new group needs to be connected to the existing
50  *			hierarchy groups
51  */
52 struct tmigr_group {
53 	raw_spinlock_t		lock;
54 	struct tmigr_group	*parent;
55 	struct tmigr_event	groupevt;
56 	u64			next_expiry;
57 	struct timerqueue_head	events;
58 	atomic_t		migr_state;
59 	unsigned int		level;
60 	int			numa_node;
61 	unsigned int		num_children;
62 	u8			childmask;
63 	struct list_head	list;
64 };
65 
66 /**
67  * struct tmigr_cpu - timer migration per CPU group
68  * @lock:		Lock protecting the tmigr_cpu group information
69  * @online:		Indicates whether the CPU is online; In deactivate path
70  *			it is required to know whether the migrator in the top
71  *			level group is to be set offline, while a timer is
72  *			pending. Then another online CPU needs to be notified to
73  *			take over the migrator role. Furthermore the information
74  *			is required in CPU hotplug path as the CPU is able to go
75  *			idle before the timer migration hierarchy hotplug AP is
76  *			reached. During this phase, the CPU has to handle the
77  *			global timers on its own and must not act as a migrator.
78  * @idle:		Indicates whether the CPU is idle in the timer migration
79  *			hierarchy
80  * @remote:		Is set when timers of the CPU are expired remotely
81  * @tmgroup:		Pointer to the parent group
82  * @childmask:		childmask of tmigr_cpu in the parent group
83  * @wakeup:		Stores the first timer when the timer migration
84  *			hierarchy is completely idle and remote expiry was done;
85  *			is returned to timer code in the idle path and is only
86  *			used in idle path.
87  * @cpuevt:		CPU event which could be enqueued into the parent group
88  */
89 struct tmigr_cpu {
90 	raw_spinlock_t		lock;
91 	bool			online;
92 	bool			idle;
93 	bool			remote;
94 	struct tmigr_group	*tmgroup;
95 	u8			childmask;
96 	u64			wakeup;
97 	struct tmigr_event	cpuevt;
98 };
99 
100 /**
101  * union tmigr_state - state of tmigr_group
102  * @state:	Combined version of the state - only used for atomic
103  *		read/cmpxchg function
104  * @struct:	Split version of the state - only use the struct members to
105  *		update information to stay independent of endianness
106  */
107 union tmigr_state {
108 	u32 state;
109 	/**
110 	 * struct - split state of tmigr_group
111 	 * @active:	Contains each childmask bit of the active children
112 	 * @migrator:	Contains childmask of the child which is migrator
113 	 * @seq:	Sequence counter needs to be increased when an update
114 	 *		to the tmigr_state is done. It prevents a race when
115 	 *		updates in the child groups are propagated in changed
116 	 *		order. Detailed information about the scenario is
117 	 *		given in the documentation at the begin of
118 	 *		timer_migration.c.
119 	 */
120 	struct {
121 		u8	active;
122 		u8	migrator;
123 		u16	seq;
124 	} __packed;
125 };
126 
127 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
128 extern void tmigr_handle_remote(void);
129 extern bool tmigr_requires_handle_remote(void);
130 extern void tmigr_cpu_activate(void);
131 extern u64 tmigr_cpu_deactivate(u64 nextevt);
132 extern u64 tmigr_cpu_new_timer(u64 nextevt);
133 extern u64 tmigr_quick_check(u64 nextevt);
134 #else
135 static inline void tmigr_handle_remote(void) { }
136 static inline bool tmigr_requires_handle_remote(void) { return false; }
137 static inline void tmigr_cpu_activate(void) { }
138 #endif
139 
140 #endif
141