Scheduler selection and interface dispatch¶
The common interface¶
sys/sys/sched.h describes the interface used by kernel callers. Its
SCHED_ITF_* X macros enumerate return types, names, and argument pairs
for operations such as sched_add(), sched_switch(), and
sched_wakeup(). These lists generate declarations, instance slots,
and dispatch machinery from a common description.
The interface has several categories:
SCHED_ITF_DISPATCHdescribes calls dispatched directly to the selected instance through indirect functions.SCHED_ITF_WRAPdescribes calls that have a common wrapper around an instance operation; initialization is an example.SCHED_ITF_INTERNALdescribes implementation hooks used by the framework.Common functions are declared separately in the header and can build on the selected implementation.
Read the comments on the interface lists before inferring a caller contract from one implementation’s function body. Flags, lock transitions, and thread-state requirements are part of that contract.
Registration and boot selection¶
An implementation defines SCHED_IMPLEMENTATION and uses
DECLARE_SCHEDULER to register an instance and its selection name in a
linker set. The descriptor contains the implementation’s function pointers.
The implementation namespace headers let the source use ordinary interface
names while the linked implementation symbols have distinct prefixes.
sched_instance_select() in sched_shim.c reads the
kern.sched.name boot setting and searches the registered selections.
The shim’s initial requested name is ULE. If no matching selection is
found, the code can fall back to the first available registration. The
later initialization wrapper rejects a missing active instance.
Thus the implementations available in a particular kernel and the name
requested during boot both matter. The kern.sched.name sysctl is not
a runtime scheduler hot-swap interface. kern.sched.available reports
the registered selection names.
Dispatch and initialization¶
The shim’s DEFINE_SHIM expansion uses DEFINE_IFUNC for dispatched
calls. The resolver obtains the corresponding function pointer from
active_sched. Other initialization hooks explicitly call through
that descriptor after preparing shared state such as CPU topology.
Keep three names separate when navigating the code:
The interface spelling, such as
sched_switch.The implementation’s linked symbol, with an implementation-specific prefix.
A documentation scope used to distinguish implementation source listings.
The ULE implementation and 4BSD implementation references use documentation scopes
scheduler.ule and scheduler.bsd4. Those scopes keep similarly named
implementation functions distinct inside Sphinx; they do not add another
kernel dispatch layer.
Reading a scheduling call¶
Start at the caller’s thread state and lock state. Find the operation in
sched.h, identify its category, and then follow the selected instance’s
implementation. On a system using ULE, interpreting a 4BSD queue or priority
calculation does not explain the same call’s ULE behavior.
The instance framework also means a source-level function name alone may be insufficient for a debugger or symbol search. Check the namespace headers and the linked implementation prefix when locating a concrete body.