File:  [NetBSD Developer Wiki] / wikisrc / projects / project / kernel_continuations.mdwn
Revision 1.4: download - view: text, annotated - select for diffs
Tue Jan 3 20:51:24 2017 UTC (4 years, 2 months ago) by riastradh
Branches: MAIN
CVS tags: HEAD
Remove preapproved funding designations.

Per discussion between board and core, these stale designations will
be replaced by something we hope to be more lively and maintained.

    1: [[!template id=project
    3: title="Kernel continuations"
    5: contact="""
    6: [tech-kern](,
    7: [board](,
    8: [core](
    9: """
   11: category="kernel"
   12: difficulty="hard"
   14: description="""
   15: This project proposal is a subtask of [[smp_networking]].
   17: The goal of this project is to implement continuations at the kernel level.
   18: Most of the pieces are already available in the kernel, so this can be
   19: reworded as: combine *callouts*, *softints*, and *workqueues* into a single
   20: framework.  Continuations are meant to be cheap; very cheap.
   22: These continuations are a dispatching system for making callbacks at
   23: scheduled times or in different thread/interrupt contexts.
   24: They aren't "continuations" in the usual sense such as you might find
   25: in Scheme code.
   27: Please note that the main goal of this project is to simplify the
   28: implementation of [[SMP networking|smp_networking]], so care must be taken
   29: in the design of the interface to support all the features required for
   30: this other project.
   32: The proposed interface looks like the following.  This interface is mostly
   33: derived from the `callout(9)` API and is a superset of the softint(9) API.
   34: The most significant change is that workqueue items are not tied to a
   35: specific kernel thread.
   37: * `kcont_t *kcont_create(kcont_wq_t *wq, kmutex_t *lock, void
   38:   (*func)(void *, kcont_t *), void *arg, int flags);`
   40:   A `wq` must be supplied.  It may be one returned by
   41:   `kcont_workqueue_acquire` or a predefined workqueue such as (sorted from
   42:   highest priority to lowest):
   44:   * `wq_softserial`, `wq_softnet`, `wq_softbio`, `wq_softclock`
   45:   * `wq_prihigh`, `wq_primedhigh`, `wq_primedlow`, `wq_prilow`
   47:   `lock`, if non-NULL, should be locked before calling `func(arg)` and
   48:   released afterwards.  However, if the lock is released and/or destroyed
   49:   before the called function returns, then, before returning,
   50:   `kcont_set_mutex` must be called with either a new mutex to be released
   51:   or `NULL`.  If acquiring lock would block, other pending kernel
   52:   continuations which depend on other locks may be dispatched in the
   53:   meantime.  However, all continuations sharing the same set of `{ wq, lock,
   54:   [ci] }` need to be processed in the order they were scheduled.
   56:   `flags` must be 0.  This field is just provided for extensibility.
   58: * `int kcont_schedule(kcont_t *kc, struct cpu_info *ci, int nticks);`
   60:   If the continuation is marked as *INVOKING*, an error of `EBUSY` should
   61:   be returned.  If `nticks` is 0, the continuation is marked as *INVOKING*
   62:   while *EXPIRED* and *PENDING* are cleared, and the continuation is
   63:   scheduled to be invoked without delay.  Otherwise, the continuation is
   64:   marked as *PENDING* while *EXPIRED* status is cleared, and the timer
   65:   reset to `nticks`.  Once the timer expires, the continuation is marked as
   66:   *EXPIRED* and *INVOKING*, and the *PENDING* status is cleared.  If `ci`
   67:   is non-NULL, the continuation is invoked on the specified CPU if the
   68:   continuations's workqueue has per-cpu queues.  If that workqueue does not
   69:   provide per-cpu queues, an error of `ENOENT` is returned.  Otherwise when
   70:   `ci` is `NULL`, the continuation is invoked on either the current CPU or
   71:   the next available CPU depending on whether the continuation's workqueue
   72:   has per-cpu queues or not, respectively.
   74: * `void kcont_destroy(kcont_t *kc);`
   76: * `kmutex_t *kcont_getmutex(kcont_t *kc);`
   78:   Returns the lock currently associated with the continuation `kc`.
   80: * `void kcont_setarg(kcont_t *kc, void *arg);`
   82:   Updates `arg` in the continuation `kc`.  If no lock is associated with
   83:   the continuation, then `arg` may be changed at any time; however, if the
   84:   continuation is being invoked, it may not pick up the change.  Otherwise,
   85:   `kcont_setarg` must only be called when the associated lock is locked.
   87: * `kmutex_t *kcont_setmutex(kcont_t *kc, kmutex_t *lock);`
   89:   Updates the lock associated with the continuation `kc` and returns the
   90:   previous lock.  If no lock is currently associated with the continuation,
   91:   then calling this function with a lock other than NULL will trigger an
   92:   assertion failure.  Otherwise, `kcont_setmutex` must be called only when
   93:   the existing lock (which will be replaced) is locked.  If
   94:   `kcont_setmutex` is called as a result of the invokation of func, then
   95:   after kcont_setmutex has been called but before func returns, the
   96:   replaced lock must have been released, and the replacement lock, if
   97:   non-NULL, must be locked upon return.
   99: * `void kcont_setfunc(kcont_t *kc, void (*func)(void *), void *arg);`
  101:   Updates `func` and `arg` in the continuation `kc`.  If no lock is
  102:   associated with the continuation, then only arg may be changed.
  103:   Otherwise, `kcont_setfunc` must be called only when the associated lock
  104:   is locked.
  106: * `bool kcont_stop(kcont_t *kc);`
  108:   The `kcont_stop function` stops the timer associated the continuation
  109:   handle kc.  The *PENDING* and *EXPIRED* status for the continuation
  110:   handle is cleared.  It is safe to call `kcont_stop` on a continuation
  111:   handle that is not pending, so long as it is initialized.  `kcont_stop`
  112:   will return a non-zero value if the continuation was *EXPIRED*.
  114: * `bool kcont_pending(kcont_t *kc);`
  116:   The `kcont_pending` function tests the *PENDING* status of the
  117:   continuation handle `kc`.  A *PENDING* continuation is one who's timer
  118:   has been started and has not expired.  Note that it is possible for a
  119:   continuation's timer to have expired without being invoked if the
  120:   continuation's lock could not be acquired or there are higher priority
  121:   threads preventing its invokation.  Note that it is only safe to test
  122:   *PENDING* status when holding the continuation's lock.
  124: * `bool kcont_expired(kcont_t *kc);`
  126:   Tests to see if the continuation's function has been invoked since the
  127:   last `kcont_schedule`.
  129: * `bool kcont_active(kcont_t *kc);`
  131: * `bool kcont_invoking(kcont_t *kc);`
  133:   Tests the *INVOKING* status of the handle `kc`.  This flag is set just
  134:   before a continuation's function is being called.  Since the scheduling
  135:   of the worker threads may induce delays, other pending higher-priority
  136:   code may run before the continuation function is allowed to run.  This
  137:   may create a race condition if this higher-priority code deallocates
  138:   storage containing one or more continuation structures whose continuation
  139:   functions are about to be run.  In such cases, one technique to prevent
  140:   references to deallocated storage would be to test whether any
  141:   continuation functions are in the *INVOKING* state using
  142:   `kcont_invoking`, and if so, to mark the data structure and defer storage
  143:   deallocation until the continuation function is allowed to run.  For this
  144:   handshake protocol to work, the continuation function will have to use
  145:   the `kcont_ack` function to clear this flag.
  147: * `bool kcont_ack(kcont_t *kc);`
  149:   Clears the *INVOKING* state in the continuation handle `kc`.  This is
  150:   used in situations where it is necessary to protect against the race
  151:   condition described under `kcont_invoking`.
  153: * `kcont_wq_t *kcont_workqueue_acquire(pri_t pri, int flags);`
  155:   Returns a workqueue that matches the specified criteria.  Thus if
  156:   multiple requesters ask for the same criteria, they are all returned the
  157:   same workqueue.  `pri` specifies the priority at which the kernel thread
  158:   which empties the workqueue should run.
  160:   If `flags` is 0 then the standard operation is required.  However, the
  161:   following flag(s) may be bitwise ORed together:
  163:   * `WQ_PERCPU` specifies that the workqueue should have a separate queue
  164:     for each CPU, thus allowing continuations to be invoked on specific CPUs.
  166: * `int kcont_workqueue_release(kcont_wq_t *wq);`
  168:   Releases an acquired workqueue.  On the last release, the workqueue's
  169:   resources are freed and the workqueue is destroyed.
  170: """
  171: ]]
  173: [[!tag smp_networking]]

CVSweb for NetBSD wikisrc <> software: FreeBSD-CVSweb