4 * Copyright (c) 2006 The DragonFly Project. All rights reserved.
6 * This code is derived from software contributed to The DragonFly Project
7 * by Matthew Dillon <dillon@backplane.com>
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in
17 * the documentation and/or other materials provided with the
19 * 3. Neither the name of The DragonFly Project nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific, prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 #include <sys/param.h>
38 #include <sys/kernel.h>
39 #include <sys/systm.h>
40 #include <sys/sysproto.h>
41 #include <sys/kern_syscall.h>
43 #include <sys/thread.h>
45 #include <sys/malloc.h>
46 #include <sys/sysctl.h>
47 #include <sys/vkernel.h>
48 #include <sys/vmspace.h>
50 #include <vm/vm_extern.h>
53 #include <machine/vmparam.h>
55 #include <sys/sysref2.h>
56 #include <sys/mplock2.h>
58 static struct vmspace_entry *vkernel_find_vmspace(struct vkernel_proc *vkp,
60 static void vmspace_entry_delete(struct vmspace_entry *ve,
61 struct vkernel_proc *vkp);
63 static MALLOC_DEFINE(M_VKERNEL, "vkernel", "VKernel structures");
66 * vmspace_create (void *id, int type, void *data)
68 * Create a VMSPACE under the control of the caller with the specified id.
69 * An id of NULL cannot be used. The type and data fields must currently
72 * The vmspace starts out completely empty. Memory may be mapped into the
73 * VMSPACE with vmspace_mmap() and MAP_VPAGETABLE section(s) controlled
74 * with vmspace_mcontrol().
79 sys_vmspace_create(struct vmspace_create_args *uap)
81 struct vmspace_entry *ve;
82 struct vkernel_proc *vkp;
83 struct proc *p = curproc;
86 if (vkernel_enable == 0)
90 * Create a virtual kernel side-structure for the process if one
93 * Implement a simple resolution for SMP races.
95 if ((vkp = p->p_vkernel) == NULL) {
96 vkp = kmalloc(sizeof(*vkp), M_VKERNEL, M_WAITOK|M_ZERO);
97 lwkt_gettoken(&proc_token);
98 if (p->p_vkernel == NULL) {
100 lwkt_token_init(&vkp->token, "vkernel");
104 kfree(vkp, M_VKERNEL);
107 lwkt_reltoken(&proc_token);
113 * Create a new VMSPACE, disallow conflicting ids
115 ve = kmalloc(sizeof(struct vmspace_entry), M_VKERNEL, M_WAITOK|M_ZERO);
116 ve->vmspace = vmspace_alloc(VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
118 pmap_pinit2(vmspace_pmap(ve->vmspace));
120 lwkt_gettoken(&vkp->token);
121 if (RB_INSERT(vmspace_rb_tree, &vkp->root, ve)) {
122 sysref_put(&ve->vmspace->vm_sysref);
123 kfree(ve, M_VKERNEL);
128 lwkt_reltoken(&vkp->token);
134 * Destroy a VMSPACE given its identifier.
139 sys_vmspace_destroy(struct vmspace_destroy_args *uap)
141 struct vkernel_proc *vkp;
142 struct vmspace_entry *ve;
146 if ((vkp = curproc->p_vkernel) == NULL) {
150 lwkt_gettoken(&vkp->token);
151 if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
159 vmspace_entry_delete(ve, vkp);
162 lwkt_reltoken(&vkp->token);
169 * vmspace_ctl (void *id, int cmd, struct trapframe *tframe,
170 * struct vextframe *vframe);
172 * Transfer control to a VMSPACE. Control is returned after the specified
173 * number of microseconds or if a page fault, signal, trap, or system call
174 * occurs. The context is updated as appropriate.
179 sys_vmspace_ctl(struct vmspace_ctl_args *uap)
181 struct vkernel_proc *vkp;
182 struct vkernel_lwp *vklp;
183 struct vmspace_entry *ve;
189 lp = curthread->td_lwp;
192 if ((vkp = p->p_vkernel) == NULL)
196 lwkt_gettoken(&vkp->token);
197 if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
203 * Signal mailbox interlock
205 if (p->p_flag & P_MAILBOX) {
206 lwkt_gettoken(&p->p_token);
207 p->p_flag &= ~P_MAILBOX;
208 lwkt_reltoken(&p->p_token);
214 case VMSPACE_CTL_RUN:
216 * Save the caller's register context, swap VM spaces, and
217 * install the passed register context. Return with
218 * EJUSTRETURN so the syscall code doesn't adjust the context.
220 atomic_add_int(&ve->refs, 1);
221 framesz = sizeof(struct trapframe);
222 if ((vklp = lp->lwp_vkernel) == NULL) {
223 vklp = kmalloc(sizeof(*vklp), M_VKERNEL,
225 lp->lwp_vkernel = vklp;
227 vklp->user_trapframe = uap->tframe;
228 vklp->user_vextframe = uap->vframe;
229 bcopy(uap->sysmsg_frame, &vklp->save_trapframe, framesz);
230 bcopy(&curthread->td_tls, &vklp->save_vextframe.vx_tls,
231 sizeof(vklp->save_vextframe.vx_tls));
232 error = copyin(uap->tframe, uap->sysmsg_frame, framesz);
234 error = copyin(&uap->vframe->vx_tls,
236 sizeof(struct savetls));
239 error = cpu_sanitize_frame(uap->sysmsg_frame);
241 error = cpu_sanitize_tls(&curthread->td_tls);
243 bcopy(&vklp->save_trapframe, uap->sysmsg_frame,
245 bcopy(&vklp->save_vextframe.vx_tls, &curthread->td_tls,
246 sizeof(vklp->save_vextframe.vx_tls));
248 atomic_subtract_int(&ve->refs, 1);
251 pmap_setlwpvm(lp, ve->vmspace);
253 set_vkernel_fp(uap->sysmsg_frame);
262 lwkt_reltoken(&vkp->token);
268 * vmspace_mmap(id, addr, len, prot, flags, fd, offset)
270 * map memory within a VMSPACE. This function is just like a normal mmap()
271 * but operates on the vmspace's memory map. Most callers use this to create
272 * a MAP_VPAGETABLE mapping.
277 sys_vmspace_mmap(struct vmspace_mmap_args *uap)
279 struct vkernel_proc *vkp;
280 struct vmspace_entry *ve;
284 * We hold the vmspace token to serialize calls to vkernel_find_vmspace.
286 lwkt_gettoken(&vmspace_token);
287 if ((vkp = curproc->p_vkernel) == NULL) {
293 * NOTE: kern_mmap() can block so we need to temporarily ref ve->refs.
295 lwkt_gettoken(&vkp->token);
296 if ((ve = vkernel_find_vmspace(vkp, uap->id)) != NULL) {
297 atomic_add_int(&ve->refs, 1);
298 error = kern_mmap(ve->vmspace, uap->addr, uap->len,
299 uap->prot, uap->flags,
300 uap->fd, uap->offset, &uap->sysmsg_resultp);
301 atomic_subtract_int(&ve->refs, 1);
305 lwkt_reltoken(&vkp->token);
307 lwkt_reltoken(&vmspace_token);
312 * vmspace_munmap(id, addr, len)
314 * unmap memory within a VMSPACE.
319 sys_vmspace_munmap(struct vmspace_munmap_args *uap)
321 struct vkernel_proc *vkp;
322 struct vmspace_entry *ve;
325 vm_size_t size, pageoff;
330 if ((vkp = curproc->p_vkernel) == NULL) {
334 lwkt_gettoken(&vkp->token);
335 if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
341 * NOTE: kern_munmap() can block so we need to temporarily
344 atomic_add_int(&ve->refs, 1);
347 * Copied from sys_munmap()
349 addr = (vm_offset_t)uap->addr;
352 pageoff = (addr & PAGE_MASK);
355 size = (vm_size_t)round_page(size);
356 if (size < uap->len) { /* wrap */
360 tmpaddr = addr + size; /* workaround gcc4 opt */
361 if (tmpaddr < addr) { /* wrap */
370 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) {
374 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS) {
378 map = &ve->vmspace->vm_map;
379 if (!vm_map_check_protection(map, addr, tmpaddr, VM_PROT_NONE, FALSE)) {
383 vm_map_remove(map, addr, addr + size);
386 atomic_subtract_int(&ve->refs, 1);
388 lwkt_reltoken(&vkp->token);
395 * vmspace_pread(id, buf, nbyte, flags, offset)
397 * Read data from a vmspace. The number of bytes read is returned or
398 * -1 if an unrecoverable error occured. If the number of bytes read is
399 * less then the request size, a page fault occured in the VMSPACE which
400 * the caller must resolve in order to proceed.
402 * (not implemented yet)
406 sys_vmspace_pread(struct vmspace_pread_args *uap)
408 struct vkernel_proc *vkp;
409 struct vmspace_entry *ve;
413 if ((vkp = curproc->p_vkernel) == NULL) {
417 lwkt_gettoken(&vkp->token);
418 if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
424 lwkt_reltoken(&vkp->token);
431 * vmspace_pwrite(id, buf, nbyte, flags, offset)
433 * Write data to a vmspace. The number of bytes written is returned or
434 * -1 if an unrecoverable error occured. If the number of bytes written is
435 * less then the request size, a page fault occured in the VMSPACE which
436 * the caller must resolve in order to proceed.
438 * (not implemented yet)
442 sys_vmspace_pwrite(struct vmspace_pwrite_args *uap)
444 struct vkernel_proc *vkp;
445 struct vmspace_entry *ve;
449 if ((vkp = curproc->p_vkernel) == NULL) {
453 lwkt_gettoken(&vkp->token);
454 if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
460 lwkt_reltoken(&vkp->token);
467 * vmspace_mcontrol(id, addr, len, behav, value)
469 * madvise/mcontrol support for a vmspace.
474 sys_vmspace_mcontrol(struct vmspace_mcontrol_args *uap)
476 struct vkernel_proc *vkp;
477 struct vmspace_entry *ve;
478 vm_offset_t start, end;
479 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
483 if ((vkp = curproc->p_vkernel) == NULL) {
487 lwkt_gettoken(&vkp->token);
488 if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
494 * NOTE: kern_madvise() can block so we need to temporarily
497 atomic_add_int(&ve->refs, 1);
500 * This code is basically copied from sys_mcontrol()
502 if (uap->behav < 0 || uap->behav > MADV_CONTROL_END) {
507 if (tmpaddr < (vm_offset_t)uap->addr) {
511 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) {
515 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS) {
520 start = trunc_page((vm_offset_t) uap->addr);
521 end = round_page(tmpaddr);
523 error = vm_map_madvise(&ve->vmspace->vm_map, start, end,
524 uap->behav, uap->value);
526 atomic_subtract_int(&ve->refs, 1);
528 lwkt_reltoken(&vkp->token);
535 * Red black tree functions
537 static int rb_vmspace_compare(struct vmspace_entry *, struct vmspace_entry *);
538 RB_GENERATE(vmspace_rb_tree, vmspace_entry, rb_entry, rb_vmspace_compare);
541 * a->start is address, and the only field has to be initialized.
542 * The caller must hold vkp->token.
544 * The caller must hold vkp->token.
547 rb_vmspace_compare(struct vmspace_entry *a, struct vmspace_entry *b)
549 if ((char *)a->id < (char *)b->id)
551 else if ((char *)a->id > (char *)b->id)
557 * The caller must hold vkp->token.
561 rb_vmspace_delete(struct vmspace_entry *ve, void *data)
563 struct vkernel_proc *vkp = data;
565 KKASSERT(ve->refs == 0);
566 vmspace_entry_delete(ve, vkp);
571 * Remove a vmspace_entry from the RB tree and destroy it. We have to clean
572 * up the pmap, the vm_map, then destroy the vmspace.
574 * This function must remove the ve immediately before it might potentially
577 * The caller must hold vkp->token.
581 vmspace_entry_delete(struct vmspace_entry *ve, struct vkernel_proc *vkp)
583 RB_REMOVE(vmspace_rb_tree, &vkp->root, ve);
585 pmap_remove_pages(vmspace_pmap(ve->vmspace),
586 VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
587 vm_map_remove(&ve->vmspace->vm_map,
588 VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
589 sysref_put(&ve->vmspace->vm_sysref);
590 kfree(ve, M_VKERNEL);
594 * Locate the ve for (id), return the ve or NULL. If found this function
595 * will bump ve->refs which prevents the ve from being immediately destroyed
596 * (but it can still be removed).
598 * The caller must hold vkp->token.
601 struct vmspace_entry *
602 vkernel_find_vmspace(struct vkernel_proc *vkp, void *id)
604 struct vmspace_entry *ve;
605 struct vmspace_entry key;
608 ve = RB_FIND(vmspace_rb_tree, &vkp->root, &key);
613 * Manage vkernel refs, used by the kernel when fork()ing or exit()ing
619 vkernel_inherit(struct proc *p1, struct proc *p2)
621 struct vkernel_proc *vkp;
624 KKASSERT(vkp->refs > 0);
625 atomic_add_int(&vkp->refs, 1);
633 vkernel_exit(struct proc *p)
635 struct vkernel_proc *vkp;
641 * Restore the original VM context if we are killed while running
644 * This isn't supposed to happen. What is supposed to happen is
645 * that the process should enter vkernel_trap() before the handling
648 RB_FOREACH(lp, lwp_rb_tree, &p->p_lwp_tree) {
649 vkernel_lwp_exit(lp);
653 * Dereference the common area
656 KKASSERT(vkp->refs > 0);
658 if (atomic_fetchadd_int(&vkp->refs, -1) == 1) {
659 lwkt_gettoken(&vkp->token);
660 RB_SCAN(vmspace_rb_tree, &vkp->root, NULL,
661 rb_vmspace_delete, vkp);
662 lwkt_reltoken(&vkp->token);
663 kfree(vkp, M_VKERNEL);
671 vkernel_lwp_exit(struct lwp *lp)
673 struct vkernel_lwp *vklp;
674 struct vmspace_entry *ve;
676 if ((vklp = lp->lwp_vkernel) != NULL) {
677 if ((ve = vklp->ve) != NULL) {
678 kprintf("Warning, pid %d killed with "
679 "active VC!\n", lp->lwp_proc->p_pid);
680 pmap_setlwpvm(lp, lp->lwp_proc->p_vmspace);
682 KKASSERT(ve->refs > 0);
683 atomic_subtract_int(&ve->refs, 1);
685 lp->lwp_vkernel = NULL;
686 kfree(vklp, M_VKERNEL);
691 * A VM space under virtual kernel control trapped out or made a system call
692 * or otherwise needs to return control to the virtual kernel context.
697 vkernel_trap(struct lwp *lp, struct trapframe *frame)
699 struct proc *p = lp->lwp_proc;
700 struct vmspace_entry *ve;
701 struct vkernel_lwp *vklp;
705 * Which vmspace entry was running?
707 vklp = lp->lwp_vkernel;
710 KKASSERT(ve != NULL);
713 * Switch the LWP vmspace back to the virtual kernel's VM space.
716 pmap_setlwpvm(lp, p->p_vmspace);
717 KKASSERT(ve->refs > 0);
718 atomic_subtract_int(&ve->refs, 1);
719 /* ve is invalid once we kill our ref */
722 * Copy the emulated process frame to the virtual kernel process.
723 * The emulated process cannot change TLS descriptors so don't
724 * bother saving them, we already have a copy.
726 * Restore the virtual kernel's saved context so the virtual kernel
727 * process can resume.
729 error = copyout(frame, vklp->user_trapframe, sizeof(*frame));
730 bcopy(&vklp->save_trapframe, frame, sizeof(*frame));
731 bcopy(&vklp->save_vextframe.vx_tls, &curthread->td_tls,
732 sizeof(vklp->save_vextframe.vx_tls));
734 cpu_vkernel_trap(frame, error);