* Copyright (c) 2003 Apple Computer, 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
* 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,
* 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@
*/
-/*-
+/*
* Copyright (c) 1999,2000 Jonathan Lemon <jlemon@FreeBSD.org>
* All rights reserved.
*
#if defined(XNU_KERNEL_PRIVATE)
-typedef int (*kevent_callback_t)(struct kqueue *, struct kevent_internal_s *, void *);
-typedef void (*kqueue_continue_t)(struct kqueue *, void *, int);
+typedef int (*kevent_callback_t)(struct kevent_qos_s *, struct kevent_ctx_s *);
#include <stdint.h>
#include <kern/locks.h>
-#include <sys/pthread_shims.h>
#include <mach/thread_policy.h>
+#include <pthread/workqueue_internal.h>
/*
* Lock ordering:
* proc fd lock -> kq lock -> kq-waitq-set lock -> thread lock
*
* WorkQ/WorkLoop kqueues (from above):
- * proc fd lock -> kq lock -> kq-request lock -> pthread kext locks -> thread lock
+ * proc fd lock -> kq lock -> workq lock -> thread lock
*
* Whenever kqueues interact with source locks, it drops all of its own
* locks in exchange for a use-reference on the knote used to synchronize
*
* Standard file-based kqueues (from below):
* XXX lock -> kq lock -> kq-waitq-set lock -> thread lock
- * Standard file-based kqueues with non-kq-aware sources (from below):
- * XXX lock -> kq-waitq-set lock -> thread lock
*
* WorkQ/WorkLoop kqueues (from below):
- * XXX lock -> kq lock -> kq-request lock -> pthread kext locks -> thread lock
- * WorkQ/WorkLoop kqueues with non-kq-aware sources (from below):
- * XXX -> kq-waitq-set lock -> kq-request lock -> pthread kext locks -> thread lock
+ * XXX lock -> kq lock -> workq lock -> thread lock
*/
-#define KQEXTENT 256 /* linear growth by this amount */
+#define KQEXTENT 256 /* linear growth by this amount */
+
+struct knote_lock_ctx {
+ struct knote *knlc_knote;
+ thread_t knlc_thread;
+ uintptr_t knlc_waiters;
+ LIST_ENTRY(knote_lock_ctx) knlc_link;
+#if DEBUG || DEVELOPMENT
+#define KNOTE_LOCK_CTX_UNLOCKED 0
+#define KNOTE_LOCK_CTX_LOCKED 1
+#define KNOTE_LOCK_CTX_WAITING 2
+ int knlc_state;
+#endif
+};
+LIST_HEAD(knote_locks, knote_lock_ctx);
+
+#if DEBUG || DEVELOPMENT
+/*
+ * KNOTE_LOCK_CTX(name) is a convenience macro to define a knote lock context on
+ * the stack named `name`. In development kernels, it uses tricks to make sure
+ * not locks was still held when exiting the C-scope that contains this context.
+ */
+static inline void
+knote_lock_ctx_chk(struct knote_lock_ctx *knlc)
+{
+ /* evil hackery to make sure no one forgets to unlock */
+ assert(knlc->knlc_state == KNOTE_LOCK_CTX_UNLOCKED);
+}
+#define KNOTE_LOCK_CTX(n) \
+ struct knote_lock_ctx n __attribute__((cleanup(knote_lock_ctx_chk))); \
+ n.knlc_state = KNOTE_LOCK_CTX_UNLOCKED
+#else
+#define KNOTE_LOCK_CTX(n) \
+ struct knote_lock_ctx n
+#endif
+
+
+__options_decl(kq_state_t, uint16_t, {
+ KQ_SEL = 0x0001, /* select was recorded for kq */
+ KQ_SLEEP = 0x0002, /* thread is waiting for events */
+ KQ_PROCWAIT = 0x0004, /* thread waiting for processing */
+ KQ_KEV32 = 0x0008, /* kq is used with 32-bit events */
+ KQ_KEV64 = 0x0010, /* kq is used with 64-bit events */
+ KQ_KEV_QOS = 0x0020, /* kq events carry QoS info */
+ KQ_WORKQ = 0x0040, /* KQ is bound to process workq */
+ KQ_WORKLOOP = 0x0080, /* KQ is part of a workloop */
+ KQ_PROCESSING = 0x0100, /* KQ is being processed */
+ KQ_DRAIN = 0x0200, /* kq is draining */
+ KQ_WAKEUP = 0x0400, /* kq awakened while processing */
+ KQ_DYNAMIC = 0x0800, /* kqueue is dynamically managed */
+ KQ_R2K_ARMED = 0x1000, /* ast notification armed */
+ KQ_HAS_TURNSTILE = 0x2000, /* this kqueue has a turnstile */
+});
/*
* kqueue - common core definition of a kqueue
* derived from this definition.
*/
struct kqueue {
- struct waitq_set kq_wqs; /* private waitq set */
- lck_spin_t kq_lock; /* kqueue lock */
- uint16_t kq_state; /* state of the kq */
- uint16_t kq_level; /* nesting level of the kq */
- uint32_t kq_count; /* number of queued events */
- struct proc *kq_p; /* process containing kqueue */
- struct kqtailq kq_queue[1]; /* variable array of kqtailq structs */
+ struct {
+ struct waitq_set kq_wqs; /* private waitq set */
+ lck_spin_t kq_lock; /* kqueue lock */
+ kq_state_t kq_state; /* state of the kq */
+ union {
+ uint16_t kq_waitq_hook;/* prepost hook (kqwl/kqwq) */
+ uint16_t kq_level; /* nesting level of the kq */
+ };
+ uint32_t kq_count; /* number of queued events */
+ struct proc *kq_p; /* process containing kqueue */
+ struct knote_locks kq_knlocks; /* list of knote locks held */
+ }; /* make sure struct padding is put before kq_queue */
+ struct kqtailq kq_queue[0]; /* variable array of queues */
};
-#define KQ_SEL 0x001 /* select was recorded for kq */
-#define KQ_SLEEP 0x002 /* thread is waiting for events */
-#define KQ_PROCWAIT 0x004 /* thread waiting for processing */
-#define KQ_KEV32 0x008 /* kq is used with 32-bit events */
-#define KQ_KEV64 0x010 /* kq is used with 64-bit events */
-#define KQ_KEV_QOS 0x020 /* kq events carry QoS info */
-#define KQ_WORKQ 0x040 /* KQ is bound to process workq */
-#define KQ_WORKLOOP 0x080 /* KQ is part of a workloop */
-#define KQ_PROCESSING 0x100 /* KQ is being processed */
-#define KQ_DRAIN 0x200 /* kq is draining */
-#define KQ_WAKEUP 0x400 /* kq awakened while processing */
-#define KQ_DYNAMIC 0x800 /* kqueue is dynamically managed */
-#define KQ_NO_WQ_THREAD 0x1000 /* kq will not have workqueue threads dynamically created */
/*
* kqfile - definition of a typical kqueue opened as a file descriptor
* via the kqueue() system call.
*/
struct kqfile {
struct kqueue kqf_kqueue; /* common kqueue core */
+ struct kqtailq kqf_queue; /* queue of woken up knotes */
struct kqtailq kqf_suppressed; /* suppression queue */
struct selinfo kqf_sel; /* parent select/kqueue info */
-};
-
#define kqf_wqs kqf_kqueue.kq_wqs
#define kqf_lock kqf_kqueue.kq_lock
#define kqf_state kqf_kqueue.kq_state
#define kqf_level kqf_kqueue.kq_level
#define kqf_count kqf_kqueue.kq_count
#define kqf_p kqf_kqueue.kq_p
-#define kqf_queue kqf_kqueue.kq_queue
-
-#define QOS_INDEX_KQFILE 0 /* number of qos levels in a file kq */
-
-struct kqr_bound {
- struct kqtailq kqrb_suppressed; /* Per-QoS suppression queues */
- thread_t kqrb_thread; /* thread to satisfy request */
-};
-
-/*
- * kqrequest - per-QoS thread request status
- */
-struct kqrequest {
-#if 0
- union {
- struct kqr_bound kqru_bound; /* used when thread is bound */
- struct workq_threadreq_s kqru_req; /* used when request oustanding */
- } kqr_u;
-#define kqr_suppressed kqr_u.kqru_bound.kqrb_suppressed
-#define kqr_thread kqr_u.kqru_bound.kqrb_thread
-#define kqr_req kqr_u.kqru_req
-#else
- struct kqr_bound kqr_bound; /* used when thread is bound */
- struct workq_threadreq_s kqr_req; /* used when request oustanding */
-#define kqr_suppressed kqr_bound.kqrb_suppressed
-#define kqr_thread kqr_bound.kqrb_thread
-#endif
- uint8_t kqr_state; /* KQ/workq interaction state */
- uint8_t kqr_wakeup_indexes; /* QoS/override levels that woke */
- uint16_t kqr_dsync_waiters:13, /* number of dispatch sync waiters */
- kqr_dsync_owner_qos:3; /* Qos override on dispatch sync owner */
- uint16_t kqr_sync_suppress_count; /* number of suppressed sync ipc knotes */
- kq_index_t kqr_stayactive_qos:3, /* max QoS of statyactive knotes */
- kqr_owner_override_is_sync:1, /* sync owner has sync ipc override */
- kqr_override_index:3, /* highest wakeup override index */
- kqr_has_sync_override:1; /* Qos/override at UI is sync ipc override */
-
- /* set under both the kqlock and the filt_wllock */
- kq_index_t :0; /* prevent bitfields coalescing <rdar://problem/31854115> */
- kq_index_t kqr_qos_index:4, /* QoS for the thread request */
- kqr_dsync_waiters_qos:4; /* override from dispatch sync waiters */
};
+#define QOS_INDEX_KQFILE 0 /* number of qos levels in a file kq */
-#define KQR_PROCESSING 0x01 /* requested thread is running the q */
-#define KQR_THREQUESTED 0x02 /* thread has been requested from workq */
-#define KQR_WAKEUP 0x04 /* wakeup called during processing */
-#define KQR_BOUND 0x08 /* servicing thread is bound */
-#define KQR_THOVERCOMMIT 0x20 /* overcommit needed for thread requests */
-#define KQR_DRAIN 0x40 /* cancel initiated - drain fulfill */
-#define KQR_R2K_NOTIF_ARMED 0x80 /* ast notifications armed */
/*
* WorkQ kqueues need to request threads to service the triggered
* knotes in the queue. These threads are brought up on a
#define KQWQ_QOS_MANAGER (THREAD_QOS_LAST)
#endif
-#if !defined(KQWQ_NQOS)
-#define KQWQ_NQOS (KQWQ_QOS_MANAGER + 1)
-#endif
-
-/*
- * Workq thread start out a particular effective-requested-QoS, but
- * additional events processed by the filters may represent
- * backlogged events that may themselves have a higher requested-QoS.
- * To represent this, the filter may apply an override to a knote's
- * requested QoS.
- *
- * We further segregate these overridden knotes into different buckets
- * by <requested, override> grouping. This allows easy matching of
- * knotes to process vs. the highest workq thread override applied.
- *
- * Only certain override patterns need to be supported. A knote
- * cannot have an effective-requested-QoS of UNSPECIFIED - because
- * the kevent->qos (when canonicalized) will always be above that
- * or indicate manager. And we don't allow an override to specify
- * manager. This results in the following buckets being needed:
- *
- * Effective-Requested QoS
- * MAINT BG UTIL DEFAULT UINIT UINTER MANAGER
- * override:
- * MAINT 0
- * BG 1 6
- * UTILITY 2 7 11
- * DEFAULT 3 8 12 15
- * UINIT 4 9 13 16 18
- * UINTER 5 10 14 17 19 20
- * 21
- */
#if !defined(KQWQ_NBUCKETS)
-#define KQWQ_NBUCKETS 22
+#define KQWQ_NBUCKETS (KQWQ_QOS_MANAGER + 1)
#endif
/*
* values.
*/
struct kqworkq {
- struct kqueue kqwq_kqueue;
- struct kqtailq kqwq_queuecont[KQWQ_NBUCKETS-1]; /* continue array of queues */
- struct kqrequest kqwq_request[KQWQ_NQOS]; /* per-QoS request states */
- lck_spin_t kqwq_reqlock; /* kqueue request lock */
+ struct kqueue kqwq_kqueue;
+ struct kqtailq kqwq_queue[KQWQ_NBUCKETS]; /* array of queues */
+ struct kqtailq kqwq_suppressed[KQWQ_NBUCKETS]; /* Per-QoS suppression queues */
+ workq_threadreq_s kqwq_request[KQWQ_NBUCKETS]; /* per-QoS request states */
};
-#define kqwq_wqs kqwq_kqueue.kq_wqs
-#define kqwq_lock kqwq_kqueue.kq_lock
-#define kqwq_state kqwq_kqueue.kq_state
-#define kqwq_level kqwq_kqueue.kq_level
-#define kqwq_count kqwq_kqueue.kq_count
-#define kqwq_p kqwq_kqueue.kq_p
-#define kqwq_queue kqwq_kqueue.kq_queue
-
-#define kqwq_req_lock(kqwq) lck_spin_lock(&kqwq->kqwq_reqlock)
-#define kqwq_req_unlock(kqwq) lck_spin_unlock(&kqwq->kqwq_reqlock)
-#define kqwq_req_held(kqwq) LCK_SPIN_ASSERT(&kqwq->kqwq_reqlock, LCK_ASSERT_OWNED)
-
-#define KQWQ_THMANAGER 0x10 /* expect manager thread to run the queue */
+#define kqwq_wqs kqwq_kqueue.kq_wqs
+#define kqwq_lock kqwq_kqueue.kq_lock
+#define kqwq_state kqwq_kqueue.kq_state
+#define kqwq_waitq_hook kqwq_kqueue.kq_waitq_hook
+#define kqwq_count kqwq_kqueue.kq_count
+#define kqwq_p kqwq_kqueue.kq_p
/*
* WorkLoop kqueues need to request a thread to service the triggered
* NOTE: "lane" support is TBD.
*/
struct kqworkloop {
- struct kqueue kqwl_kqueue; /* queue of events */
- struct kqtailq kqwl_queuecont[KQWL_NBUCKETS-1]; /* continue array of queues */
- struct kqrequest kqwl_request; /* thread request state */
- lck_spin_t kqwl_reqlock; /* kqueue request lock */
- lck_mtx_t kqwl_statelock; /* state/debounce lock */
- thread_t kqwl_owner; /* current [sync] owner thread */
- uint32_t kqwl_retains; /* retain references */
- kqueue_id_t kqwl_dynamicid; /* dynamic identity */
- SLIST_ENTRY(kqworkloop) kqwl_hashlink; /* linkage for search list */
+ struct kqueue kqwl_kqueue; /* queue of events */
+ struct kqtailq kqwl_queue[KQWL_NBUCKETS]; /* array of queues */
+ struct kqtailq kqwl_suppressed; /* Per-QoS suppression queues */
+ workq_threadreq_s kqwl_request; /* thread request state */
+ lck_spin_t kqwl_statelock; /* state/debounce lock */
+ thread_t kqwl_owner; /* current [sync] owner thread */
+ uint32_t kqwl_retains; /* retain references */
+#define KQWL_STAYACTIVE_FIRED_BIT (1 << 0)
+ uint8_t kqwl_wakeup_indexes; /* QoS/override levels that woke */
+ kq_index_t kqwl_stayactive_qos; /* max QoS of statyactive knotes */
+ struct turnstile *kqwl_turnstile; /* turnstile for sync IPC/waiters */
+ kqueue_id_t kqwl_dynamicid; /* dynamic identity */
+ uint64_t kqwl_params; /* additional parameters */
+ LIST_ENTRY(kqworkloop) kqwl_hashlink; /* linkage for search list */
+#if CONFIG_WORKLOOP_DEBUG
+#define KQWL_HISTORY_COUNT 32
+#define KQWL_HISTORY_WRITE_ENTRY(kqwl, ...) ({ \
+ struct kqworkloop *__kqwl = (kqwl); \
+ unsigned int __index = os_atomic_inc_orig(&__kqwl->kqwl_index, relaxed); \
+ __kqwl->kqwl_history[__index % KQWL_HISTORY_COUNT] = \
+ (struct kqwl_history)__VA_ARGS__; \
+ })
+ struct kqwl_history {
+ thread_t updater; /* Note: updates can be reordered */
+ thread_t servicer;
+ thread_t old_owner;
+ thread_t new_owner;
+
+ uint64_t kev_ident;
+ int16_t error;
+ uint16_t kev_flags;
+ uint32_t kev_fflags;
+
+ uint64_t kev_mask;
+ uint64_t kev_value;
+ uint64_t in_value;
+ } kqwl_history[KQWL_HISTORY_COUNT];
+ unsigned int kqwl_index;
+#endif // CONFIG_WORKLOOP_DEBUG
};
+LIST_HEAD(kqwllist, kqworkloop);
-SLIST_HEAD(kqlist, kqworkloop);
+typedef union {
+ struct kqueue *kq;
+ struct kqworkq *kqwq;
+ struct kqfile *kqf;
+ struct kqworkloop *kqwl;
+} __attribute__((transparent_union)) kqueue_t;
-#define kqwl_wqs kqwl_kqueue.kq_wqs
-#define kqwl_lock kqwl_kqueue.kq_lock
-#define kqwl_state kqwl_kqueue.kq_state
-#define kqwl_level kqwl_kqueue.kq_level
-#define kqwl_count kqwl_kqueue.kq_count
-#define kqwl_p kqwl_kqueue.kq_p
-#define kqwl_queue kqwl_kqueue.kq_queue
-#define kqwl_req_lock(kqwl) lck_spin_lock(&kqwl->kqwl_reqlock)
-#define kqwl_req_unlock(kqwl) lck_spin_unlock(&kqwl->kqwl_reqlock)
-#define kqwl_req_held(kqwl) LCK_SPIN_ASSERT(&kqwl->kqwl_reqlock, LCK_ASSERT_OWNED)
+#define kqwl_wqs kqwl_kqueue.kq_wqs
+#define kqwl_lock kqwl_kqueue.kq_lock
+#define kqwl_state kqwl_kqueue.kq_state
+#define kqwl_waitq_hook kqwl_kqueue.kq_waitq_hook
+#define kqwl_count kqwl_kqueue.kq_count
+#define kqwl_p kqwl_kqueue.kq_p
#define KQ_WORKLOOP_RETAINS_MAX UINT32_MAX
-extern int workloop_fulfill_threadreq(struct proc *p, workq_threadreq_t req, thread_t thread, int flags);
+extern void kqueue_threadreq_unbind(struct proc *p, workq_threadreq_t);
+
+// called with the kq req held
+#define KQUEUE_THREADERQ_BIND_NO_INHERITOR_UPDATE 0x1
+extern void kqueue_threadreq_bind(struct proc *p, workq_threadreq_t req,
+ thread_t thread, unsigned int flags);
+
+struct turnstile *kqueue_threadreq_get_turnstile(workq_threadreq_t kqr);
+
+// called with the wq lock held
+extern void kqueue_threadreq_bind_prepost(struct proc *p, workq_threadreq_t req,
+ struct uthread *uth);
+
+// called with no lock held
+extern void kqueue_threadreq_bind_commit(struct proc *p, thread_t thread);
+
+extern void kqueue_threadreq_cancel(struct proc *p, workq_threadreq_t req);
-extern struct kqueue *kqueue_alloc(struct proc *, unsigned int);
+// lock not held as kqwl_params is immutable after creation
+extern workq_threadreq_param_t kqueue_threadreq_workloop_param(workq_threadreq_t req);
+
+extern struct kqueue *kqueue_alloc(struct proc *);
extern void kqueue_dealloc(struct kqueue *);
+extern void kqworkq_dealloc(struct kqworkq *kqwq);
extern void knotes_dealloc(struct proc *);
+extern void kqworkloops_dealloc(struct proc *);
-extern void kevent_register(struct kqueue *, struct kevent_internal_s *, struct proc *);
-extern int kqueue_scan(struct kqueue *, kevent_callback_t, kqueue_continue_t,
- void *, struct filt_process_s *, struct timeval *, struct proc *);
+extern int kevent_register(struct kqueue *, struct kevent_qos_s *,
+ struct knote **);
+extern int kqueue_scan(struct kqueue *, int flags,
+ struct kevent_ctx_s *, kevent_callback_t);
extern int kqueue_stat(struct kqueue *, void *, int, proc_t);
#endif /* XNU_KERNEL_PRIVATE */
#endif /* !_SYS_EVENTVAR_H_ */
-
-
-
-