xref: /xnu-12377.61.12/bsd/sys/work_interval.h (revision 4d495c6e23c53686cf65f45067f79024cf5dcee8)

/*
 * Copyright (c) 2015-2017 Apple Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 *
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 *
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 *
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */

#ifndef _SYS_WORK_INTERVAL_H
#define _SYS_WORK_INTERVAL_H

#include <stdint.h>
#include <sys/cdefs.h>
#include <sys/_types/_size_t.h>

#include <mach/port.h>

__BEGIN_DECLS

/*
 * A work interval is a repeatable unit of work characterized by a
 * start, finish, and deadline.
 *
 * Trusted clients with deadline-sensitive work may report information
 * about the execution of their work using the work interval facility.
 * This is intended to be a higher-level semantic than realtime scheduling,
 * which operates at the level of thread block/unblock. A high level
 * operation may have many blocking points, including IPC to other tasks,
 * and this this metric will capture the overall time to complete a unit of
 * work.
 *
 * A work interval is defined by several timestamps, namely (S)tart,
 * (F)inish, (D)eadline, and (N)ext start.
 *
 *   ... ----+==================+--------+--+==== ...
 *           |                  |        |  |
 *           S                  F        D  N
 *
 *           \__________________/
 *                  Active
 *           \___________________________/
 *                   Work Interval
 *
 *                               \_________/
 *                                    |
 *   report information here ---------+
 *
 * Definitions:
 *
 *   Start: Absolute time when the current deadline-oriented work began. Due
 *          to scheduling latency, preemption, and blocking points, the
 *          thread controlling the work interval may actually begin
 *          executing after this ideal time (which may be the previous work
 *          interval's "next start")
 *   Finish: Absolute time when the current deadline-oriented work finished.
 *          This will typically be a timestamp taken before reporting using
 *          the work interval interface.
 *   Deadline: Absolute time by which the current work was expected to finish.
 *          In cases where the amount of computation (or preemption, or time
 *          spent blocked) causes the active period to take longer than
 *          expected, F may be greater than D.
 *   Next start: Absolute time when the next deadline-oriented work is
 *          expected to begin. This is typically the same as Deadline.
 *   Active: The fraction of the work interval spent completing the work. In
 *          cases where the Finish time exceeded the Deadline, this fraction
 *          will be >1.0.
 *
 * Basic Use:
 *
 *   Clients should report information for a work interval after finishing
 *   work for the current interval but before the next work interval begins.
 *
 *   If Finish far exceeds the previously expected Deadline, the
 *   caller may adjust Next Start to align to a multiple of the period
 *   (and skip over several work intervals that could not be
 *   executed).
 *
 * Caution (!):
 *
 *   Because the information supplied via this facility directly influences power
 *   management decisions, clients should strive to be as accurate as possible.
 *   Failure to do so will adversely impact system power and performance.
 *
 * Work Interval Auto Join Support:
 *
 * Work intervals support an optional flag WORK_INTERVAL_FLAG_ENABLE_AUTO_JOIN
 * which allows RT threads from the same home thread group to join work
 * intervals via wakeup relationship tracking. Based on the join policy,
 * RT threads can temporarily join the work interval of other RT threads
 * which make them runnable. The auto joined thread remains in the work
 * interval until it blocks or terminates. The mechanism works through
 * make runnable heuristic and it should be used with extreme caution.
 * If a client specifies this flag, it gives up explicit control over its
 * thread group membership and threads unrelated to the work interval
 * could become part of the thread group. This could lead to serious power
 * and performance issues. If the make runnable heuristic does not work
 * for a client use case, it should adopt work_interval_join_port() or
 * work_interval_join() to explicitly declare its intent.
 *
 * Work Interval Deferred Finish Support:
 *
 * Another advanced feature for work intervals is the ability to defer the finish
 * calls for the work interval until all auto-joined threads for the work interval
 * have blocked or terminated. This feature is enabled via an optional flag
 * WORK_INTERVAL_FLAG_ENABLE_DEFERRED_FINISH and is valid only if the work interval
 * is configured with the WORK_INTERVAL_FLAG_ENABLE_AUTO_JOIN flag as well. The
 * deferred finish mechanism allows the work interval to defer the finish call
 * for the work interval until all auto-join threads have blocked/terminated
 * (and have therefore un-joined the work interval) or one of the work interval
 * threads calls start for the next frame. The deferred finish works only for
 * workloads that have no temporal overlap across frames i.e. previous frame has to
 * finish before next frame can start. This feature should be used with caution
 * since auto-joined threads would delay finish calls to the performance controller
 * which could lead to poor performance and battery life.
 */

/* Flags to be passed with work_interval_create() */

/* If interval is joinable, create no longer implicitly joins, you must use work_interval_join */
#define WORK_INTERVAL_FLAG_JOINABLE                     (0x1)
/* Only threads that join the group are measured together, otherwise the group is the creator's home group */
#define WORK_INTERVAL_FLAG_GROUP                        (0x2)
/* Specifies that the work interval is being created by a client who doesn't
 * necessarily have the PRIV_WORK_INTERVAL entitlement. Skip privilege checks.
 * This can only be masked in for work intervals of types COREAUDIO, CA_CLIENT
 * and DEFAULT */
#define WORK_INTERVAL_FLAG_UNRESTRICTED                 (0x4)

/* [Advanced Flag] Read section on "Work Interval Auto Join Support" above for details */
#define WORK_INTERVAL_FLAG_ENABLE_AUTO_JOIN             (0x8)
/* [Advanced Flag] Read section on "Work Interval Deferred Finish Support" above for details */
#define WORK_INTERVAL_FLAG_ENABLE_DEFERRED_FINISH       (0x10)

/* Kernel-supplied flag: Work interval has been ignored by the kernel */
#define WORK_INTERVAL_FLAG_IGNORED                      (0x20)

/* Specifies that the work interval requests the system to provide just enough performance
 * to be able to finish at the provided deadline and no sooner. */
#define WORK_INTERVAL_FLAG_FINISH_AT_DEADLINE           (0x40)

/* Internal-only flag: workinterval will have the SET_WORKLOAD_ID operation called on it immediately
 * after creation. */
#define WORK_INTERVAL_FLAG_HAS_WORKLOAD_ID              (0x80)

/* Internal-only flag: Telemetry data will be aggregated from threads while they are joined to
 * the work interval. */
#define WORK_INTERVAL_FLAG_ENABLE_TELEMETRY_DATA        (0x100)

/* Flags to describe the interval flavor to the performance controller */
#define WORK_INTERVAL_TYPE_MASK                 (0xF0000000)
#define WORK_INTERVAL_TYPE_DEFAULT              (0x0 << 28)
#define WORK_INTERVAL_TYPE_COREAUDIO            (0x1 << 28)
#define WORK_INTERVAL_TYPE_COREANIMATION        (0x2 << 28)
#define WORK_INTERVAL_TYPE_CA_RENDER_SERVER     (0x2 << 28)
#define WORK_INTERVAL_TYPE_CA_CLIENT            (0x3 << 28)
#define WORK_INTERVAL_TYPE_HID_DELIVERY         (0x4 << 28)
#define WORK_INTERVAL_TYPE_COREMEDIA            (0x5 << 28)
#define WORK_INTERVAL_TYPE_ARKIT                (0x6 << 28)
#define WORK_INTERVAL_TYPE_FRAME_COMPOSITOR     (0x7 << 28)
#define WORK_INTERVAL_TYPE_LAST                 (0xF << 28)

#ifndef KERNEL

typedef struct work_interval *work_interval_t;
typedef struct work_interval_instance *work_interval_instance_t;
typedef struct work_interval_data *work_interval_data_t;

/*
 * Create a new work interval handle.
 *
 * May fail with EALREADY if the current group already has a work interval.
 *
 * With no flags:
 *      Auto-joins the work interval to the creating thread
 *      May only use interval_handle from creating thread
 *      Data provided affects native thread group
 *
 * With the JOINABLE flag
 *      interval_handle is usable by the process
 *      creating thread does not auto-join
 *      notifying thread must have joined when notifying
 *
 * With the GROUP flag
 *      creates a new thread group to isolate the joined threads from
 *      the rest of the process for performance controller analysis
 *      Threads which join the work interval become members of this new group
 *
 * TODO: Add a name parameter so that clients can name the work interval
 * Can also take the thread name from the notifying thread
 *
 * Requires the 'com.apple.private.kernel.work-interval' entitlement (PRIV_WORK_INTERVAL)
 *
 * Note that joining a work interval supersedes automatic thread group management via vouchers
 */
int     work_interval_create(work_interval_t *interval_handle, uint32_t flags);

/* Returns the flags used for the work interval when it was created.
 *
 * May fail with EINVAL if the port isn't from a prior call to
 * work_interval_copy_port.
 */
int work_interval_get_flags_from_port(mach_port_t port, uint32_t *flags);


/*
 * Notify the power management subsystem that the work for a current interval has completed
 *
 * Only the process which created the work interval may notify
 */
int     work_interval_notify(work_interval_t    interval_handle,
    uint64_t start, uint64_t finish,
    uint64_t deadline, uint64_t next_start,
    uint32_t flags);

/*
 * Notify, with "finish" implicitly set to the current time
 *
 * Only the process which created the work interval may notify
 */
int     work_interval_notify_simple(work_interval_t interval_handle,
    uint64_t start, uint64_t deadline,
    uint64_t next_start);

/*
 * Deallocate work interval handle
 * For non-JOINABLE, also removes thread from work interval
 * For JOINABLE, does not remove thread (needs a leave as well)
 */
int     work_interval_destroy(work_interval_t interval_handle);

/*
 * Join work interval via work interval handle
 * Only allowed if interval is using the joinable and group flags
 *
 * Supersedes automatic thread group management via vouchers
 */
int     work_interval_join(work_interval_t interval_handle);

/*
 * extract Mach send right representing work interval thread group
 * Returns a +1 send right ref, which must be deallocated via mach_port_deallocate
 * Only allowed if interval is joinable, otherwise returns ENOTSUP
 *
 * Supersedes automatic thread group management via vouchers
 */
int     work_interval_copy_port(work_interval_t interval_handle, mach_port_t *port);

/*
 * Join work interval via Mach send right
 *
 * Does NOT consume Mach send right, must deallocate with mach_port_deallocate after using
 * It's safe to deallocate the right after joining, the thread will stay joined
 *
 * Can be sent to clients via xpc_dictionary_copy_mach_send, and similar
 *
 * Supersedes automatic thread group management via vouchers
 *
 * If the underlying work interval object is terminated then this may return ENOENT
 * <rdar://problem/31819320>
 */
int     work_interval_join_port(mach_port_t port);

/*
 * Leave the current thread's work interval
 */
int     work_interval_leave(void);

#endif /* !KERNEL */


#if PRIVATE

/* Private interface between Libsyscall and xnu */
#define WORK_INTERVAL_OPERATION_CREATE  0x00000001      /* deprecated */
#define WORK_INTERVAL_OPERATION_DESTROY 0x00000002      /* arg is NULL */
#define WORK_INTERVAL_OPERATION_NOTIFY  0x00000003      /* arg is a work_interval_notification_t */
#define WORK_INTERVAL_OPERATION_CREATE2 0x00000004      /* arg is a work_interval_create_params */
#define WORK_INTERVAL_OPERATION_JOIN    0x00000005      /* arg is a port_name */
#define WORK_INTERVAL_OPERATION_GET_FLAGS 0x00000009    /* arg is a port name */
#define WORK_INTERVAL_OPERATION_SET_NAME  0x0000000a    /* arg is name string (char[WORK_INTERVAL_NAME_MAX])*/
#define WORK_INTERVAL_OPERATION_SET_WORKLOAD_ID  0x0000000b    /* arg is a work_interval_workload_id_params */
#define WORK_INTERVAL_NAME_MAX  32
#define WORK_INTERVAL_WORKLOAD_ID_NAME_MAX  64

/* Flags passed in work_interval_workload_id_params wlidp_flags field */
#define WORK_INTERVAL_WORKLOAD_ID_HAS_ID                (1u << 0)
#define WORK_INTERVAL_WORKLOAD_ID_RT_ALLOWED            (1u << 1)
#define WORK_INTERVAL_WORKLOAD_ID_RT_CRITICAL           (1u << 2)
/* Work interval is allowed to provide complexity values per frame as part of {start, update, finish} calls */
#define WORK_INTERVAL_WORKLOAD_ID_COMPLEXITY_ALLOWED    (1u << 3)

/* Flags allowed to be passed in from userspace as part of kern_work_interval_set_workload_id() */
#define WORK_INTERVAL_SET_WORKLOAD_ID_FLAGS_MASK        (WORK_INTERVAL_WORKLOAD_ID_RT_CRITICAL | WORK_INTERVAL_WORKLOAD_ID_RT_ALLOWED | WORK_INTERVAL_WORKLOAD_ID_COMPLEXITY_ALLOWED)

#ifdef XNU_KERNEL_PRIVATE
/* Marker that workinterval was joined before workload ID was set */
#define WORK_INTERVAL_WORKLOAD_ID_ALREADY_JOINED        (1u << 31)

#endif /* XNU_KERNEL_PRIVATE */

struct work_interval_notification {
	uint64_t        start;
	uint64_t        finish;
	uint64_t        deadline;
	uint64_t        next_start;
	uint32_t        notify_flags;
	uint32_t        create_flags;
};
typedef struct work_interval_notification *work_interval_notification_t;

struct work_interval_create_params {
	uint64_t        wicp_id;          /* in/out param */
	mach_port_name_t wicp_port;       /* in/out param */
	uint32_t        wicp_create_flags;
};

struct work_interval_workload_id_params {
	uint32_t        wlidp_flags;            /* in/out param */
	uint32_t        wlidp_wicreate_flags;   /* in/out param */
	uint64_t        wlidp_name;             /* in param (pointer to char[WORK_INTERVAL_WORKLOAD_ID_NAME_MAX])*/
	uint64_t        wlidp_syscall_mask[2];  /* out param (needs to fit MACH_TRAP_TABLE_COUNT + nsysent bits) */
};


int     __work_interval_ctl(uint32_t operation, uint64_t work_interval_id, void *arg, size_t len);

#endif /* PRIVATE */

__END_DECLS

#endif /* _SYS_WORK_INTERVAL_H */