C interface of the Scheduler kernel object, see L4::Scheduler for the C++ interface. More...

Collaboration diagram for Scheduler:

Data Structures
struct	l4_sched_cpu_set_t
	CPU sets. More...

struct	l4_sched_param_t
	Scheduler parameter set. More...

Typedefs
typedef struct l4_sched_cpu_set_t	l4_sched_cpu_set_t
	CPU sets.

typedef struct l4_sched_param_t	l4_sched_param_t
	Scheduler parameter set.

Enumerations
enum	L4_scheduler_classes { L4_SCHEDULER_CLASS_FIXED_PRIO = 1UL << 1 , L4_SCHEDULER_CLASS_WFQ = 1UL << 2 }
	Supported scheduler classes. More...

enum	L4_scheduler_ops { L4_SCHEDULER_INFO_OP = 0UL , L4_SCHEDULER_RUN_THREAD_OP = 1UL , L4_SCHEDULER_IDLE_TIME_OP = 2UL }
	Operations on the Scheduler object. More...

Functions
l4_sched_cpu_set_t	l4_sched_cpu_set (l4_umword_t offset, unsigned char granularity, l4_umword_t map=1) L4_NOTHROW

l4_msgtag_t	l4_scheduler_info (l4_cap_idx_t scheduler, l4_umword_t cpu_max, l4_sched_cpu_set_t cpus) L4_NOTHROW
	Get scheduler information.

l4_msgtag_t	l4_scheduler_info_with_classes (l4_cap_idx_t scheduler, l4_umword_t cpu_max, l4_sched_cpu_set_t cpus, l4_umword_t *sched_classes) L4_NOTHROW
	Get scheduler information.

l4_sched_param_t	l4_sched_param (unsigned prio, l4_umword_t quantum=0) L4_NOTHROW
	Construct scheduler parameter.

l4_msgtag_t	l4_scheduler_run_thread (l4_cap_idx_t scheduler, l4_cap_idx_t thread, l4_sched_param_t const *sp) L4_NOTHROW
	Run a thread on a Scheduler.

l4_msgtag_t	l4_scheduler_idle_time (l4_cap_idx_t scheduler, l4_sched_cpu_set_t const cpus, l4_kernel_clock_t us) L4_NOTHROW
	Query the idle time (in µs) of a CPU.

int	l4_scheduler_is_online (l4_cap_idx_t scheduler, l4_umword_t cpu) L4_NOTHROW
	Query if a CPU is online.

Detailed Description

C interface of the Scheduler kernel object, see L4::Scheduler for the C++ interface.

The Scheduler interface allows a client to manage CPU resources. The API provides functions to query scheduler information, check the online state of CPUs, query CPU idle time and to start threads on defined CPU sets.

The scheduler offers a virtual device IRQ which triggers when the number of online cores changes, e.g. due to hotplug events. In contrast to hardware IRQs, this IRQ implements a limited functionality:

Only IRQ line 0 is supported, no MSI vectors.
The IRQ is edge-triggered and the IRQ mode cannot be changed.
As the IRQ is edge-triggered, it does not have to be explicitly unmasked.

It depends on the platform, which hotplug events actually trigger the IRQ. Many platforms only support triggering the IRQ when a CPU core different from the boot CPU goes online.

Include File: #include <l4/sys/scheduler.h>

scheduler.h
Scheduler object functions.

Enumeration Type Documentation

◆ L4_scheduler_classes

enum L4_scheduler_classes

Supported scheduler classes.

Enumerator
L4_SCHEDULER_CLASS_FIXED_PRIO	Fixed-priority scheduler.
L4_SCHEDULER_CLASS_WFQ	Weighted fair queuing scheduler.

Definition at line 57 of file scheduler.h.

◆ L4_scheduler_ops

enum L4_scheduler_ops

Operations on the Scheduler object.

Enumerator
L4_SCHEDULER_INFO_OP	Query infos about the scheduler.
L4_SCHEDULER_RUN_THREAD_OP	Run a thread on this scheduler.
L4_SCHEDULER_IDLE_TIME_OP	Query idle time for the scheduler.

Definition at line 262 of file scheduler.h.

Function Documentation

◆ l4_sched_cpu_set()

l4_sched_cpu_set_t l4_sched_cpu_set	(	l4_umword_t	offset,
		unsigned char	granularity,
		l4_umword_t	map = `1`
	)

inline

Parameters

offset	Offset. Must be a multiple of 2^granularity.
granularity	Granularity in log2 notation.
map	Bitmap of CPUs, defaults to 1 in C++.

Returns: CPU set.

Examples: examples/sys/migrate/thread_migrate.cc.

Definition at line 272 of file scheduler.h.

References l4_sched_cpu_set_t::gran_offset, and l4_sched_cpu_set_t::map.

Referenced by l4_sched_param().

Here is the caller graph for this function:

◆ l4_sched_param()

l4_sched_param_t l4_sched_param	(	unsigned	prio,
		l4_umword_t	quantum = `0`
	)

inline

Construct scheduler parameter.

The l4_sched_param_t::affinity of the returned value contains all CPUs.

Examples: examples/sys/aliens/main.c, examples/sys/migrate/thread_migrate.cc, examples/sys/singlestep/main.c, examples/sys/start-with-exc/main.c, and examples/sys/utcb-ipc/main.c.

Definition at line 282 of file scheduler.h.

References l4_sched_param_t::affinity, l4_sched_cpu_set(), l4_sched_param_t::prio, and l4_sched_param_t::quantum.

Here is the call graph for this function:

◆ l4_scheduler_idle_time()

l4_msgtag_t l4_scheduler_idle_time	(	l4_cap_idx_t	scheduler,
		l4_sched_cpu_set_t const *	cpus,
		l4_kernel_clock_t *	us
	)

inline

Query the idle time (in µs) of a CPU.

Parameters

	scheduler	Scheduler object.
	cpus	Set of CPUs to query. Only the idle time of the first selected CPU in `cpus.map` is queried.
[out]	us	Idle time of queried CPU in µs.

Return values

0	Success.
-L4_EINVAL	Invalid CPU requested in cpu set.

This function retrieves the idle time in µs of the first selected CPU in cpus.map. The idle time is the accumulated time a CPU has spent in the idle thread since its last reset. To calculate a load estimate l one has to retrieve the idle time at the beginning (i1) and the end (i2) of a known time interval t. The load is then calculated as l = 1 - (i2 - i1)/t.

The idle time is only defined for online CPUs. Reading the idle time from offline CPUs is undefined and may result in either getting -L4_EINVAL or calculating an estimated (incorrect) load of 1.

Note: The idle time statistics of remote CPUs is updated on context switch events only, hence may not be up-to-date when requested cross-CPU. To get up-to-date idle time you should use a thread running on the same CPU of which the idle time is requested.

Definition at line 396 of file scheduler.h.

References l4_utcb().

Here is the call graph for this function:

◆ l4_scheduler_info()

l4_msgtag_t l4_scheduler_info	(	l4_cap_idx_t	scheduler,
		l4_umword_t *	cpu_max,
		l4_sched_cpu_set_t *	cpus
	)

inline

Get scheduler information.

Parameters

	scheduler	Scheduler object.
[out]	cpu_max	Maximum number of CPUs ever available. Optional, can be NULL.
[in,out]	cpus	cpus.offset is first CPU of interest. cpus.granularity (see l4_sched_cpu_set_t). cpus.map Bitmap of online CPUs. Pass NULL if this information is not required.

Return values

0	Success.
-L4_ERANGE	The given CPU offset is larger than the maximum number of CPUs.

Definition at line 374 of file scheduler.h.

References l4_utcb().

Here is the call graph for this function:

◆ l4_scheduler_info_with_classes()

l4_msgtag_t l4_scheduler_info_with_classes	(	l4_cap_idx_t	scheduler,
		l4_umword_t *	cpu_max,
		l4_sched_cpu_set_t *	cpus,
		l4_umword_t *	sched_classes
	)

inline

Get scheduler information.

Parameters

	scheduler	Scheduler object.
[out]	cpu_max	Maximum number of CPUs ever available. Optional, can be NULL.
[in,out]	cpus	cpus.offset is first CPU of interest. cpus.granularity (see l4_sched_cpu_set_t). cpus.map Bitmap of online CPUs. Pass NULL if this information is not required.
[out]	sched_classes	A bitmap of available scheduling classes (see L4_scheduler_classes). Optional, can be NULL.

Return values

0	Success.
-L4_ERANGE	The given CPU offset is larger than the maximum number of CPUs.

This function delivers the same information as l4_scheduler_info plus the available scheduler classes (see L4_scheduler_classes).

Definition at line 381 of file scheduler.h.

References l4_utcb().

Here is the call graph for this function:

◆ l4_scheduler_is_online()

int l4_scheduler_is_online	(	l4_cap_idx_t	scheduler,
		l4_umword_t	cpu
	)

inline

Query if a CPU is online.

Parameters

scheduler	Scheduler object.
cpu	CPU number whose online status should be queried.

Return values

true	The CPU is online.
false	The CPU is offline

Definition at line 403 of file scheduler.h.

References l4_utcb().

Here is the call graph for this function:

◆ l4_scheduler_run_thread()

l4_msgtag_t l4_scheduler_run_thread	(	l4_cap_idx_t	scheduler,
		l4_cap_idx_t	thread,
		l4_sched_param_t const *	sp
	)

inline

Run a thread on a Scheduler.

Parameters

scheduler	Scheduler object.
thread	Capability of the thread to run.
sp	Scheduling parameters.

Return values

0	Success.
-L4_EINVAL	Invalid size of the scheduling parameter.

This function launches a thread on a CPU determined by the scheduling parameter sp.affinity. A thread can be intentionally stopped by migrating it on an offline or an invalid CPU. The thread is only guaranteed to run if the CPU it is migrated to is currently online.

Note

If the target CPU is currently not online, there is no guarantee that the thread will ever run, even if the CPU comes online later on.

A scheduler may impose a policy with regard to selecting CPUs. However the scheduler is required to ensure the following two properties:

Two threads with disjoint CPU sets must be scheduled to different CPUs.
Two threads with identical CPU sets selecting only a single CPU must be scheduled to the same CPU.

Examples: examples/sys/aliens/main.c, examples/sys/singlestep/main.c, examples/sys/start-with-exc/main.c, and examples/sys/utcb-ipc/main.c.

Definition at line 389 of file scheduler.h.

References l4_utcb().

Here is the call graph for this function:

Data Structures

Typedefs

Enumerations

Functions

Detailed Description

Enumeration Type Documentation

◆ L4_scheduler_classes

◆ L4_scheduler_ops

Function Documentation

◆ l4_sched_cpu_set()

◆ l4_sched_param()

◆ l4_scheduler_idle_time()

◆ l4_scheduler_info()

◆ l4_scheduler_info_with_classes()

◆ l4_scheduler_is_online()

◆ l4_scheduler_run_thread()