Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_PSI_TYPES_H
3#define _LINUX_PSI_TYPES_H
4
5#include <linux/kthread.h>
6#include <linux/seqlock.h>
7#include <linux/types.h>
8#include <linux/kref.h>
9#include <linux/wait.h>
10
11#ifdef CONFIG_PSI
12
13/* Tracked task states */
14enum psi_task_count {
15 NR_IOWAIT,
16 NR_MEMSTALL,
17 NR_RUNNING,
18 /*
19 * For IO and CPU stalls the presence of running/oncpu tasks
20 * in the domain means a partial rather than a full stall.
21 * For memory it's not so simple because of page reclaimers:
22 * they are running/oncpu while representing a stall. To tell
23 * whether a domain has productivity left or not, we need to
24 * distinguish between regular running (i.e. productive)
25 * threads and memstall ones.
26 */
27 NR_MEMSTALL_RUNNING,
28 NR_PSI_TASK_COUNTS = 4,
29};
30
31/* Task state bitmasks */
32#define TSK_IOWAIT (1 << NR_IOWAIT)
33#define TSK_MEMSTALL (1 << NR_MEMSTALL)
34#define TSK_RUNNING (1 << NR_RUNNING)
35#define TSK_MEMSTALL_RUNNING (1 << NR_MEMSTALL_RUNNING)
36
37/* Only one task can be scheduled, no corresponding task count */
38#define TSK_ONCPU (1 << NR_PSI_TASK_COUNTS)
39
40/* Resources that workloads could be stalled on */
41enum psi_res {
42 PSI_IO,
43 PSI_MEM,
44 PSI_CPU,
45#ifdef CONFIG_IRQ_TIME_ACCOUNTING
46 PSI_IRQ,
47#endif
48 NR_PSI_RESOURCES,
49};
50
51/*
52 * Pressure states for each resource:
53 *
54 * SOME: Stalled tasks & working tasks
55 * FULL: Stalled tasks & no working tasks
56 */
57enum psi_states {
58 PSI_IO_SOME,
59 PSI_IO_FULL,
60 PSI_MEM_SOME,
61 PSI_MEM_FULL,
62 PSI_CPU_SOME,
63 PSI_CPU_FULL,
64#ifdef CONFIG_IRQ_TIME_ACCOUNTING
65 PSI_IRQ_FULL,
66#endif
67 /* Only per-CPU, to weigh the CPU in the global average: */
68 PSI_NONIDLE,
69 NR_PSI_STATES,
70};
71
72/* Use one bit in the state mask to track TSK_ONCPU */
73#define PSI_ONCPU (1 << NR_PSI_STATES)
74
75enum psi_aggregators {
76 PSI_AVGS = 0,
77 PSI_POLL,
78 NR_PSI_AGGREGATORS,
79};
80
81struct psi_group_cpu {
82 /* 1st cacheline updated by the scheduler */
83
84 /* Aggregator needs to know of concurrent changes */
85 seqcount_t seq ____cacheline_aligned_in_smp;
86
87 /* States of the tasks belonging to this group */
88 unsigned int tasks[NR_PSI_TASK_COUNTS];
89
90 /* Aggregate pressure state derived from the tasks */
91 u32 state_mask;
92
93 /* Period time sampling buckets for each state of interest (ns) */
94 u32 times[NR_PSI_STATES];
95
96 /* Time of last task change in this group (rq_clock) */
97 u64 state_start;
98
99 /* 2nd cacheline updated by the aggregator */
100
101 /* Delta detection against the sampling buckets */
102 u32 times_prev[NR_PSI_AGGREGATORS][NR_PSI_STATES]
103 ____cacheline_aligned_in_smp;
104};
105
106/* PSI growth tracking window */
107struct psi_window {
108 /* Window size in ns */
109 u64 size;
110
111 /* Start time of the current window in ns */
112 u64 start_time;
113
114 /* Value at the start of the window */
115 u64 start_value;
116
117 /* Value growth in the previous window */
118 u64 prev_growth;
119};
120
121struct psi_trigger {
122 /* PSI state being monitored by the trigger */
123 enum psi_states state;
124
125 /* User-spacified threshold in ns */
126 u64 threshold;
127
128 /* List node inside triggers list */
129 struct list_head node;
130
131 /* Backpointer needed during trigger destruction */
132 struct psi_group *group;
133
134 /* Wait queue for polling */
135 wait_queue_head_t event_wait;
136
137 /* Pending event flag */
138 int event;
139
140 /* Tracking window */
141 struct psi_window win;
142
143 /*
144 * Time last event was generated. Used for rate-limiting
145 * events to one per window
146 */
147 u64 last_event_time;
148
149 /* Deferred event(s) from previous ratelimit window */
150 bool pending_event;
151};
152
153struct psi_group {
154 struct psi_group *parent;
155 bool enabled;
156
157 /* Protects data used by the aggregator */
158 struct mutex avgs_lock;
159
160 /* Per-cpu task state & time tracking */
161 struct psi_group_cpu __percpu *pcpu;
162
163 /* Running pressure averages */
164 u64 avg_total[NR_PSI_STATES - 1];
165 u64 avg_last_update;
166 u64 avg_next_update;
167
168 /* Aggregator work control */
169 struct delayed_work avgs_work;
170
171 /* Total stall times and sampled pressure averages */
172 u64 total[NR_PSI_AGGREGATORS][NR_PSI_STATES - 1];
173 unsigned long avg[NR_PSI_STATES - 1][3];
174
175 /* Monitor work control */
176 struct task_struct __rcu *poll_task;
177 struct timer_list poll_timer;
178 wait_queue_head_t poll_wait;
179 atomic_t poll_wakeup;
180
181 /* Protects data used by the monitor */
182 struct mutex trigger_lock;
183
184 /* Configured polling triggers */
185 struct list_head triggers;
186 u32 nr_triggers[NR_PSI_STATES - 1];
187 u32 poll_states;
188 u64 poll_min_period;
189
190 /* Total stall times at the start of monitor activation */
191 u64 polling_total[NR_PSI_STATES - 1];
192 u64 polling_next_update;
193 u64 polling_until;
194};
195
196#else /* CONFIG_PSI */
197
198#define NR_PSI_RESOURCES 0
199
200struct psi_group { };
201
202#endif /* CONFIG_PSI */
203
204#endif /* _LINUX_PSI_TYPES_H */