[dragonfly.git] / sys / kern / kern_upcall.c

/*
 * Copyright (c) 2003,2004,2006 The DragonFly Project.  All rights reserved.
 * 
 * This code is derived from software contributed to The DragonFly Project
 * by Matthew Dillon <dillon@backplane.com>
 * 
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in
 *    the documentation and/or other materials provided with the
 *    distribution.
 * 3. Neither the name of The DragonFly Project nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific, prior written permission.
 * 
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 * 
 * $DragonFly: src/sys/kern/kern_upcall.c,v 1.11 2006/09/10 21:35:10 dillon Exp $
 */

/*
 * Implement upcall registration and dispatch.
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/proc.h>
#include <sys/upcall.h>
#include <sys/thread2.h>
#include <sys/upcall.h>
#include <sys/malloc.h>
#include <sys/sysproto.h>
#include <sys/lock.h>
#include <sys/signalvar.h>

#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/vm_kern.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>

#include <machine/cpu.h>

MALLOC_DEFINE(M_UPCALL, "upcalls", "upcall registration structures");

#ifdef SMP

static void
sigupcall_remote(void *arg)
{
 	struct lwp *lp = arg;
	if (lp == lwkt_preempted_proc())
		sigupcall();
}

#endif

/*
 * upc_register:
 *
 *	Register an upcall context wrapper and procedure.  Note that the
 *	upcall context is set globally for the process, not for each upcall.
 *
 * ARGS(struct upcall *upc, upcall_func_t ctx, upcall_func_t func, void *data)
 */
int
sys_upc_register(struct upc_register_args *uap)
{
    struct lwp *lp = curthread->td_lwp;
    struct vmspace *vm = curproc->p_vmspace;
    struct vmupcall *vu;

    if (vm->vm_upccount >= UPCALL_MAXCOUNT)
	return(EFBIG);

    vu = kmalloc(sizeof(struct vmupcall), M_UPCALL, M_WAITOK|M_ZERO);
    vu->vu_ctx = uap->ctxfunc;
    vu->vu_func = uap->func;
    vu->vu_data = uap->data;
    vu->vu_lwp = lp;
    lp->lwp_upcall = uap->upc;

    if (vm->vm_upcalls != NULL)
	vu->vu_id = vm->vm_upcalls->vu_id + 1;
    else
	vu->vu_id = UPC_RESERVED;
    vu->vu_next = vm->vm_upcalls;
    vm->vm_upcalls = vu;
    ++vm->vm_upccount;
    uap->sysmsg_result = vu->vu_id;
    return(0);
}

/*
 * upc_control:
 *
 * ARGS(int cmd, int upcid, void *data)
 */
int
sys_upc_control(struct upc_control_args *uap)
{
    struct lwp *lp = curthread->td_lwp;
    struct lwp *targlp;
    struct vmspace *vms = curproc->p_vmspace;
    struct vmupcall *vu;
    struct vmupcall *vu_send;
    struct vmupcall **vupp;
    int error;

    switch(uap->cmd) {
    case UPC_CONTROL_DISPATCH:
	/*
	 * Dispatch the specified upcall id or the next pending id if -1.
	 * the upcall will be marked pending but an actual upcall will only
	 * occur if userland is not in a critical section and the userland
	 * pending bit is not set.
	 *
	 * You can dispatch an upcall associated with your process or another
	 * process sharing the same VM space.
	 */
	error = (uap->upcid == -1) ? 0 : ENOENT;
	for (vu = vms->vm_upcalls; vu; vu = vu->vu_next) {
	    if (vu->vu_id == uap->upcid || 
		(uap->upcid == -1 &&
		vu->vu_pending >= (int)(intptr_t)uap->data && vu->vu_lwp == lp)
	    ) {
		if (vu->vu_pending < (int)(intptr_t)uap->data)
		    vu->vu_pending = (int)(intptr_t)uap->data;
		error = 0;
		targlp = vu->vu_lwp;
		targlp->lwp_proc->p_flag |= P_UPCALLPEND;	/* XXX lwp flags */
		if (targlp->lwp_proc->p_flag & P_UPCALLWAIT)
		    wakeup(&targlp->lwp_upcall);
#ifdef SMP
		if (targlp->lwp_thread->td_gd != mycpu)
		    lwkt_send_ipiq(targlp->lwp_thread->td_gd, sigupcall_remote, targlp);
		else
		    sigupcall();
#else
		sigupcall();
#endif
		break;
	    }
	}
	break;
    case UPC_CONTROL_NEXT:
	/*
	 * This is used by the context code to fetch the next pending upcall.
	 * The context code has two choices:  (A) it can drop
	 * upcall->crit_count and set upcall->pending then make this call
	 * unconditionally or * (B) it can drop upcall->crit_count and then
	 * test upcall->pending and only make this call if upcall->pending
	 * is set.  If upcall->pending is clear the context code can pop
	 * the upcall stack itself and return without entering into the kernel
	 * again.  (B) is more efficient but leaves a small window of
	 * opportunity where multiple upcalls can pushdown the stack.
	 *
	 * If another upcall is pending the crit_count will be bumped and
	 * the function, data, and context pointers will be returned in
	 * registers (C cannot call this routine).  If no more upcalls are
	 * pending the pending bit will be cleared and the 'data' argument
	 * is expected to be pointing at the upcall context which we will
	 * then pop, returning to the original code that was interrupted
	 * (NOT the context code).
	 */
	vu_send = NULL;
	for (vu = vms->vm_upcalls; vu; vu = vu->vu_next) {
	    if (vu->vu_lwp == lp && vu->vu_pending) {
		if (vu_send)
		    break;
		vu_send = vu;
	    }
	}
	/*
	 * vu_send may be NULL, indicating that no more upcalls are pending
	 * for this cpu.  We set the userland pending bit based on whether
	 * additional upcalls are pending or not.
	 */
	error = fetchupcall(vu_send, vu != NULL, uap->data);
	break;
    case UPC_CONTROL_DELETE:
	/*
	 * Delete the specified upcall id.  If the upcall id is -1, delete
	 * all upcall id's associated with the current process.
	 */
	error = (uap->upcid == -1) ? 0 : ENOENT;
	vupp = &vms->vm_upcalls;
	while ((vu = *vupp) != NULL) {
	    if (vu->vu_id == uap->upcid || 
		(uap->upcid == -1 && vu->vu_lwp == lp)
	    ) {
		*vupp = vu->vu_next;
		error = 0;
		kfree(vu, M_UPCALL);
	    } else {
		vupp = &vu->vu_next;
	    }
	}
	break;
    case UPC_CONTROL_POLL:
    case UPC_CONTROL_POLLANDCLEAR:
    case UPC_CONTROL_WAIT:
	/*
	 * If upcid is -1 poll for the first pending upcall and return the
	 * id or 0 if no upcalls are pending.
	 *
	 * If upcid is a particular upcall then poll that upcall and return
	 * its pending status (0 or 1).  For POLLANDCLEAR, also clear the
	 * pending status.  The userland pending bit is not modified by
	 * this call (maybe we should modify it for poll-and-clear).
	 */
	error = (uap->upcid == -1) ? 0 : ENOENT;
	for (vu = vms->vm_upcalls; vu; vu = vu->vu_next) {
	    if (vu->vu_id == uap->upcid || 
		(uap->upcid == -1 &&
		 vu->vu_pending >= (int)(intptr_t)uap->data && vu->vu_lwp == lp)
	    ) {
		error = 0;
		if (uap->upcid == -1)
		    uap->sysmsg_result = vu->vu_id;
		else
		    uap->sysmsg_result = vu->vu_pending;
		if (uap->cmd == UPC_CONTROL_POLLANDCLEAR)
		    vu->vu_pending = 0;
		break;
	    }
	}
	if (uap->cmd == UPC_CONTROL_WAIT && vu == NULL) {
	    lp->lwp_proc->p_flag |= P_UPCALLWAIT;	/* XXX lwp flags */
	    tsleep(&lp->lwp_upcall, PCATCH, "wupcall", 0);
	    lp->lwp_proc->p_flag &= ~P_UPCALLWAIT;	/* XXX lwp flags */
	}
	break;
    default:
	error = EINVAL;
	break;
    }
    return(error);
}

void
upc_release(struct vmspace *vm, struct lwp *lp)
{
    struct vmupcall **vupp;
    struct vmupcall *vu;

    vupp = &vm->vm_upcalls;
    while ((vu = *vupp) != NULL) {
	if (vu->vu_lwp == lp) {
	    *vupp = vu->vu_next;
	    kfree(vu, M_UPCALL);
	    --vm->vm_upccount;
	} else {
	    vupp = &vu->vu_next;
	}
    }
}

/*
 * XXX eventually we should sort by vu_pending priority and dispatch
 * the highest priority upcall first.
 */
void
postupcall(struct lwp *lp)
{
    struct vmspace *vm = lp->lwp_proc->p_vmspace;
    struct vmupcall *vu;
    struct vmupcall *vu_send = NULL;

    for (vu = vm->vm_upcalls; vu; vu = vu->vu_next) {
	if (vu->vu_lwp == lp && vu->vu_pending) {
	    if (vu_send) {
		sendupcall(vu, 1);
		return;
	    }
	    vu_send = vu;
	}
    }
    if (vu_send)
	sendupcall(vu_send, 0);
}
Commit	Line	Data
a722be49	1	/*
1f33f833	2	* Copyright (c) 2003,2004,2006 The DragonFly Project. All rights reserved.
8c10bfcf MD	3	*
	4	* This code is derived from software contributed to The DragonFly Project
	5	* by Matthew Dillon <dillon@backplane.com>
	6	*
a722be49 MD	7	* Redistribution and use in source and binary forms, with or without
	8	* modification, are permitted provided that the following conditions
	9	* are met:
8c10bfcf	10	*
a722be49 MD	11	* 1. Redistributions of source code must retain the above copyright
	12	* notice, this list of conditions and the following disclaimer.
	13	* 2. Redistributions in binary form must reproduce the above copyright
8c10bfcf MD	14	* notice, this list of conditions and the following disclaimer in
	15	* the documentation and/or other materials provided with the
	16	* distribution.
	17	* 3. Neither the name of The DragonFly Project nor the names of its
	18	* contributors may be used to endorse or promote products derived
	19	* from this software without specific, prior written permission.
	20	*
	21	* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
	22	* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
	23	* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
	24	* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
	25	* COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
	26	* INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
	27	* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
	28	* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
	29	* AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
	30	* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
	31	* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
a722be49	32	* SUCH DAMAGE.
8c10bfcf	33	*
1f33f833	34	* $DragonFly: src/sys/kern/kern_upcall.c,v 1.11 2006/09/10 21:35:10 dillon Exp $
a722be49 MD	35	*/
	36
	37	/*
	38	* Implement upcall registration and dispatch.
	39	*/
	40
	41	#include <sys/param.h>
	42	#include <sys/systm.h>
	43	#include <sys/kernel.h>
	44	#include <sys/proc.h>
	45	#include <sys/upcall.h>
	46	#include <sys/thread2.h>
	47	#include <sys/upcall.h>
	48	#include <sys/malloc.h>
	49	#include <sys/sysproto.h>
	50	#include <sys/lock.h>
	51	#include <sys/signalvar.h>
	52
	53	#include <vm/vm.h>
	54	#include <vm/vm_param.h>
	55	#include <vm/vm_kern.h>
	56	#include <vm/pmap.h>
	57	#include <vm/vm_map.h>
	58
	59	#include <machine/cpu.h>
	60
	61	MALLOC_DEFINE(M_UPCALL, "upcalls", "upcall registration structures");
	62
	63	#ifdef SMP
	64
	65	static void
	66	sigupcall_remote(void *arg)
	67	{
4170cc8b	68	struct lwp *lp = arg;
553ea3c8	69	if (lp == lwkt_preempted_proc())
a722be49 MD	70	sigupcall();
	71	}
	72
	73	#endif
	74
	75	/*
	76	* upc_register:
	77	*
	78	* Register an upcall context wrapper and procedure. Note that the
	79	* upcall context is set globally for the process, not for each upcall.
	80	*
	81	* ARGS(struct upcall upc, upcall_func_t ctx, upcall_func_t func, void data)
	82	*/
	83	int
753fd850	84	sys_upc_register(struct upc_register_args *uap)
a722be49	85	{
4170cc8b SS	86	struct lwp *lp = curthread->td_lwp;
4170cc8b SS	87	struct vmspace *vm = curproc->p_vmspace;
a722be49 MD	88	struct vmupcall *vu;
	89
	90	if (vm->vm_upccount >= UPCALL_MAXCOUNT)
	91	return(EFBIG);
	92
efda3bd0	93	vu = kmalloc(sizeof(struct vmupcall), M_UPCALL, M_WAITOK\|M_ZERO);
a722be49 MD	94	vu->vu_ctx = uap->ctxfunc;
	95	vu->vu_func = uap->func;
	96	vu->vu_data = uap->data;
4170cc8b SS	97	vu->vu_lwp = lp;
4170cc8b SS	98	lp->lwp_upcall = uap->upc;
a722be49 MD	99
	100	if (vm->vm_upcalls != NULL)
	101	vu->vu_id = vm->vm_upcalls->vu_id + 1;
	102	else
1f33f833	103	vu->vu_id = UPC_RESERVED;
a722be49 MD	104	vu->vu_next = vm->vm_upcalls;
	105	vm->vm_upcalls = vu;
	106	++vm->vm_upccount;
	107	uap->sysmsg_result = vu->vu_id;
	108	return(0);
	109	}
	110
	111	/*
	112	* upc_control:
	113	*
	114	* ARGS(int cmd, int upcid, void *data)
	115	*/
	116	int
753fd850	117	sys_upc_control(struct upc_control_args *uap)
a722be49	118	{
4170cc8b SS	119	struct lwp *lp = curthread->td_lwp;
	120	struct lwp *targlp;
	121	struct vmspace *vms = curproc->p_vmspace;
a722be49 MD	122	struct vmupcall *vu;
	123	struct vmupcall *vu_send;
	124	struct vmupcall **vupp;
	125	int error;
	126
	127	switch(uap->cmd) {
	128	case UPC_CONTROL_DISPATCH:
	129	/*
	130	* Dispatch the specified upcall id or the next pending id if -1.
	131	* the upcall will be marked pending but an actual upcall will only
	132	* occur if userland is not in a critical section and the userland
	133	* pending bit is not set.
	134	*
	135	* You can dispatch an upcall associated with your process or another
	136	* process sharing the same VM space.
	137	*/
	138	error = (uap->upcid == -1) ? 0 : ENOENT;
	139	for (vu = vms->vm_upcalls; vu; vu = vu->vu_next) {
	140	if (vu->vu_id == uap->upcid \|\|
973c11b9 MD	141	(uap->upcid == -1 &&
973c11b9 MD	142	vu->vu_pending >= (int)(intptr_t)uap->data && vu->vu_lwp == lp)
a722be49	143	) {
973c11b9 MD	144	if (vu->vu_pending < (int)(intptr_t)uap->data)
973c11b9 MD	145	vu->vu_pending = (int)(intptr_t)uap->data;
a722be49	146	error = 0;
4170cc8b SS	147	targlp = vu->vu_lwp;
	148	targlp->lwp_proc->p_flag \|= P_UPCALLPEND; /* XXX lwp flags */
	149	if (targlp->lwp_proc->p_flag & P_UPCALLWAIT)
	150	wakeup(&targlp->lwp_upcall);
a722be49	151	#ifdef SMP
4170cc8b SS	152	if (targlp->lwp_thread->td_gd != mycpu)
4170cc8b SS	153	lwkt_send_ipiq(targlp->lwp_thread->td_gd, sigupcall_remote, targlp);
a722be49 MD	154	else
	155	sigupcall();
	156	#else
	157	sigupcall();
	158	#endif
	159	break;
	160	}
	161	}
	162	break;
	163	case UPC_CONTROL_NEXT:
	164	/*
	165	* This is used by the context code to fetch the next pending upcall.
	166	* The context code has two choices: (A) it can drop
	167	* upcall->crit_count and set upcall->pending then make this call
	168	* unconditionally or * (B) it can drop upcall->crit_count and then
	169	* test upcall->pending and only make this call if upcall->pending
	170	* is set. If upcall->pending is clear the context code can pop
	171	* the upcall stack itself and return without entering into the kernel
	172	* again. (B) is more efficient but leaves a small window of
	173	* opportunity where multiple upcalls can pushdown the stack.
	174	*
	175	* If another upcall is pending the crit_count will be bumped and
	176	* the function, data, and context pointers will be returned in
	177	* registers (C cannot call this routine). If no more upcalls are
	178	* pending the pending bit will be cleared and the 'data' argument
	179	* is expected to be pointing at the upcall context which we will
	180	* then pop, returning to the original code that was interrupted
	181	* (NOT the context code).
	182	*/
	183	vu_send = NULL;
	184	for (vu = vms->vm_upcalls; vu; vu = vu->vu_next) {
4170cc8b	185	if (vu->vu_lwp == lp && vu->vu_pending) {
a722be49 MD	186	if (vu_send)
	187	break;
	188	vu_send = vu;
	189	}
	190	}
	191	/*
	192	* vu_send may be NULL, indicating that no more upcalls are pending
	193	* for this cpu. We set the userland pending bit based on whether
	194	* additional upcalls are pending or not.
	195	*/
0a455ac5	196	error = fetchupcall(vu_send, vu != NULL, uap->data);
a722be49 MD	197	break;
	198	case UPC_CONTROL_DELETE:
	199	/*
	200	* Delete the specified upcall id. If the upcall id is -1, delete
	201	* all upcall id's associated with the current process.
	202	*/
	203	error = (uap->upcid == -1) ? 0 : ENOENT;
	204	vupp = &vms->vm_upcalls;
	205	while ((vu = *vupp) != NULL) {
	206	if (vu->vu_id == uap->upcid \|\|
4170cc8b	207	(uap->upcid == -1 && vu->vu_lwp == lp)
a722be49 MD	208	) {
	209	*vupp = vu->vu_next;
	210	error = 0;
efda3bd0	211	kfree(vu, M_UPCALL);
a722be49 MD	212	} else {
	213	vupp = &vu->vu_next;
	214	}
	215	}
	216	break;
	217	case UPC_CONTROL_POLL:
	218	case UPC_CONTROL_POLLANDCLEAR:
fe8c5e17	219	case UPC_CONTROL_WAIT:
a722be49 MD	220	/*
	221	* If upcid is -1 poll for the first pending upcall and return the
	222	* id or 0 if no upcalls are pending.
	223	*
	224	* If upcid is a particular upcall then poll that upcall and return
	225	* its pending status (0 or 1). For POLLANDCLEAR, also clear the
	226	* pending status. The userland pending bit is not modified by
	227	* this call (maybe we should modify it for poll-and-clear).
	228	*/
	229	error = (uap->upcid == -1) ? 0 : ENOENT;
	230	for (vu = vms->vm_upcalls; vu; vu = vu->vu_next) {
	231	if (vu->vu_id == uap->upcid \|\|
973c11b9 MD	232	(uap->upcid == -1 &&
973c11b9 MD	233	vu->vu_pending >= (int)(intptr_t)uap->data && vu->vu_lwp == lp)
a722be49 MD	234	) {
	235	error = 0;
	236	if (uap->upcid == -1)
	237	uap->sysmsg_result = vu->vu_id;
	238	else
	239	uap->sysmsg_result = vu->vu_pending;
	240	if (uap->cmd == UPC_CONTROL_POLLANDCLEAR)
	241	vu->vu_pending = 0;
	242	break;
	243	}
	244	}
fe8c5e17	245	if (uap->cmd == UPC_CONTROL_WAIT && vu == NULL) {
4170cc8b SS	246	lp->lwp_proc->p_flag \|= P_UPCALLWAIT; /* XXX lwp flags */
	247	tsleep(&lp->lwp_upcall, PCATCH, "wupcall", 0);
	248	lp->lwp_proc->p_flag &= ~P_UPCALLWAIT; /* XXX lwp flags */
fe8c5e17	249	}
a722be49 MD	250	break;
	251	default:
	252	error = EINVAL;
	253	break;
	254	}
	255	return(error);
	256	}
	257
	258	void
4170cc8b	259	upc_release(struct vmspace vm, struct lwp lp)
a722be49 MD	260	{
	261	struct vmupcall **vupp;
	262	struct vmupcall *vu;
	263
	264	vupp = &vm->vm_upcalls;
	265	while ((vu = *vupp) != NULL) {
4170cc8b	266	if (vu->vu_lwp == lp) {
a722be49	267	*vupp = vu->vu_next;
efda3bd0	268	kfree(vu, M_UPCALL);
a722be49 MD	269	--vm->vm_upccount;
	270	} else {
	271	vupp = &vu->vu_next;
	272	}
	273	}
	274	}
	275
093dd88e MD	276	/*
	277	* XXX eventually we should sort by vu_pending priority and dispatch
	278	* the highest priority upcall first.
	279	*/
a722be49	280	void
4170cc8b	281	postupcall(struct lwp *lp)
a722be49	282	{
4170cc8b	283	struct vmspace *vm = lp->lwp_proc->p_vmspace;
a722be49 MD	284	struct vmupcall *vu;
	285	struct vmupcall *vu_send = NULL;
	286
	287	for (vu = vm->vm_upcalls; vu; vu = vu->vu_next) {
4170cc8b	288	if (vu->vu_lwp == lp && vu->vu_pending) {
a722be49 MD	289	if (vu_send) {
	290	sendupcall(vu, 1);
	291	return;
	292	}
	293	vu_send = vu;
	294	}
	295	}
	296	if (vu_send)
	297	sendupcall(vu_send, 0);
	298	}
	299