2 * Copyright (c) 1982, 1986 The Regents of the University of California.
3 * Copyright (c) 1989, 1990 William Jolitz
4 * Copyright (c) 1994 John Dyson
7 * This code is derived from software contributed to Berkeley by
8 * the Systems Programming Group of the University of Utah Computer
9 * Science Department, and William Jolitz.
11 * Redistribution and use in source and binary :forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91
40 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
52 #include <sys/socketvar.h>
53 #include <sys/sf_buf.h>
54 #include <sys/syscall.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysent.h>
57 #include <sys/unistd.h>
58 #include <machine/cpu.h>
59 #include <machine/frame.h>
60 #include <machine/pcb.h>
61 #include <machine/sysarch.h>
63 #include <sys/mutex.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_kern.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_param.h>
72 #include <vm/vm_pageout.h>
74 #include <vm/uma_int.h>
76 #include <machine/md_var.h>
79 * struct switchframe and trapframe must both be a multiple of 8
80 * for correct stack alignment.
82 CTASSERT(sizeof(struct switchframe) == 24);
83 CTASSERT(sizeof(struct trapframe) == 80);
86 #define NSFBUFS (512 + maxusers * 16)
90 static int nsfbufspeak;
91 static int nsfbufsused;
93 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufs, CTLFLAG_RDTUN, &nsfbufs, 0,
94 "Maximum number of sendfile(2) sf_bufs available");
95 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufspeak, CTLFLAG_RD, &nsfbufspeak, 0,
96 "Number of sendfile(2) sf_bufs at peak usage");
97 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufsused, CTLFLAG_RD, &nsfbufsused, 0,
98 "Number of sendfile(2) sf_bufs in use");
100 static void sf_buf_init(void *arg);
101 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
103 LIST_HEAD(sf_head, sf_buf);
106 * A hash table of active sendfile(2) buffers
108 static struct sf_head *sf_buf_active;
109 static u_long sf_buf_hashmask;
111 #define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask)
113 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
114 static u_int sf_buf_alloc_want;
117 * A lock used to synchronize access to the hash table and free list
119 static struct mtx sf_buf_lock;
122 * Finish a fork operation, with process p2 nearly set up.
123 * Copy and update the pcb, set up the stack so that the child
124 * ready to run and return to user mode.
127 cpu_fork(register struct thread *td1, register struct proc *p2,
128 struct thread *td2, int flags)
131 struct trapframe *tf;
132 struct switchframe *sf;
135 if ((flags & RFPROC) == 0)
137 pcb2 = (struct pcb *)(td2->td_kstack + td2->td_kstack_pages * PAGE_SIZE) - 1;
139 #ifndef CPU_XSCALE_CORE3
140 pmap_use_minicache(td2->td_kstack, td2->td_kstack_pages * PAGE_SIZE);
144 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
146 bcopy(&td1->td_proc->p_md, mdp2, sizeof(*mdp2));
147 pcb2->un_32.pcb32_sp = td2->td_kstack +
148 USPACE_SVC_STACK_TOP - sizeof(*pcb2);
149 pcb2->pcb_vfpcpu = -1;
151 td2->td_frame = tf = (struct trapframe *)STACKALIGN(
152 pcb2->un_32.pcb32_sp - sizeof(struct trapframe));
153 *tf = *td1->td_frame;
154 sf = (struct switchframe *)tf - 1;
155 sf->sf_r4 = (u_int)fork_return;
156 sf->sf_r5 = (u_int)td2;
157 sf->sf_pc = (u_int)fork_trampoline;
158 tf->tf_spsr &= ~PSR_C_bit;
161 pcb2->un_32.pcb32_sp = (u_int)sf;
162 KASSERT((pcb2->un_32.pcb32_sp & 7) == 0,
163 ("cpu_fork: Incorrect stack alignment"));
165 /* Setup to release spin count in fork_exit(). */
166 td2->td_md.md_spinlock_count = 1;
167 td2->td_md.md_saved_cspr = 0;
168 #ifdef ARM_TP_ADDRESS
169 td2->td_md.md_tp = *(register_t *)ARM_TP_ADDRESS;
171 td2->td_md.md_tp = (register_t) get_tls();
176 cpu_thread_swapin(struct thread *td)
181 cpu_thread_swapout(struct thread *td)
186 * Detatch mapped page and release resources back to the system.
189 sf_buf_free(struct sf_buf *sf)
192 mtx_lock(&sf_buf_lock);
194 if (sf->ref_count == 0) {
195 TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
197 pmap_kremove(sf->kva);
199 LIST_REMOVE(sf, list_entry);
200 if (sf_buf_alloc_want > 0)
201 wakeup(&sf_buf_freelist);
203 mtx_unlock(&sf_buf_lock);
207 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
210 sf_buf_init(void *arg)
212 struct sf_buf *sf_bufs;
217 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
219 sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
220 TAILQ_INIT(&sf_buf_freelist);
221 sf_base = kva_alloc(nsfbufs * PAGE_SIZE);
222 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
224 for (i = 0; i < nsfbufs; i++) {
225 sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
226 TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
228 sf_buf_alloc_want = 0;
229 mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
233 * Get an sf_buf from the freelist. Will block if none are available.
236 sf_buf_alloc(struct vm_page *m, int flags)
238 struct sf_head *hash_list;
242 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
243 mtx_lock(&sf_buf_lock);
244 LIST_FOREACH(sf, hash_list, list_entry) {
247 if (sf->ref_count == 1) {
248 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
250 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
255 while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
256 if (flags & SFB_NOWAIT)
259 SFSTAT_INC(sf_allocwait);
260 error = msleep(&sf_buf_freelist, &sf_buf_lock,
261 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
266 * If we got a signal, don't risk going back to sleep.
271 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
273 LIST_REMOVE(sf, list_entry);
274 LIST_INSERT_HEAD(hash_list, sf, list_entry);
278 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
279 pmap_kenter(sf->kva, VM_PAGE_TO_PHYS(sf->m));
281 mtx_unlock(&sf_buf_lock);
286 cpu_set_syscall_retval(struct thread *td, int error)
288 struct trapframe *frame;
294 frame = td->td_frame;
299 * __syscall returns an off_t while most other syscalls return an
300 * int. As an off_t is 64-bits and an int is 32-bits we need to
301 * place the returned data into r1. As the lseek and frerebsd6_lseek
302 * syscalls also return an off_t they do not need this fixup.
307 call = *(u_int32_t *)(frame->tf_pc - INSN_SIZE) & 0x000fffff;
309 if (call == SYS___syscall) {
310 register_t *ap = &frame->tf_r0;
311 register_t code = ap[_QUAD_LOWWORD];
312 if (td->td_proc->p_sysent->sv_mask)
313 code &= td->td_proc->p_sysent->sv_mask;
314 fixup = (code != SYS_freebsd6_lseek && code != SYS_lseek)
323 frame->tf_r1 = td->td_retval[0];
325 frame->tf_r0 = td->td_retval[0];
326 frame->tf_r1 = td->td_retval[1];
328 frame->tf_spsr &= ~PSR_C_bit; /* carry bit */
332 * Reconstruct the pc to point at the swi.
334 frame->tf_pc -= INSN_SIZE;
340 frame->tf_r0 = error;
341 frame->tf_spsr |= PSR_C_bit; /* carry bit */
347 * Initialize machine state (pcb and trap frame) for a new thread about to
348 * upcall. Put enough state in the new thread's PCB to get it to go back
349 * userret(), where we can intercept it again to set the return (upcall)
350 * Address and stack, along with those from upcals that are from other sources
351 * such as those generated in thread_userret() itself.
354 cpu_set_upcall(struct thread *td, struct thread *td0)
356 struct trapframe *tf;
357 struct switchframe *sf;
359 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
360 bcopy(td0->td_pcb, td->td_pcb, sizeof(struct pcb));
362 sf = (struct switchframe *)tf - 1;
363 sf->sf_r4 = (u_int)fork_return;
364 sf->sf_r5 = (u_int)td;
365 sf->sf_pc = (u_int)fork_trampoline;
366 tf->tf_spsr &= ~PSR_C_bit;
368 td->td_pcb->un_32.pcb32_sp = (u_int)sf;
369 KASSERT((td->td_pcb->un_32.pcb32_sp & 7) == 0,
370 ("cpu_set_upcall: Incorrect stack alignment"));
372 /* Setup to release spin count in fork_exit(). */
373 td->td_md.md_spinlock_count = 1;
374 td->td_md.md_saved_cspr = 0;
378 * Set that machine state for performing an upcall that has to
379 * be done in thread_userret() so that those upcalls generated
380 * in thread_userret() itself can be done as well.
383 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
386 struct trapframe *tf = td->td_frame;
388 tf->tf_usr_sp = STACKALIGN((int)stack->ss_sp + stack->ss_size
389 - sizeof(struct trapframe));
390 tf->tf_pc = (int)entry;
391 tf->tf_r0 = (int)arg;
392 tf->tf_spsr = PSR_USR32_MODE;
396 cpu_set_user_tls(struct thread *td, void *tls_base)
399 td->td_md.md_tp = (register_t)tls_base;
400 if (td == curthread) {
402 #ifdef ARM_TP_ADDRESS
403 *(register_t *)ARM_TP_ADDRESS = (register_t)tls_base;
405 set_tls((void *)tls_base);
413 cpu_thread_exit(struct thread *td)
418 cpu_thread_alloc(struct thread *td)
420 td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_pages *
423 * Ensure td_frame is aligned to an 8 byte boundary as it will be
424 * placed into the stack pointer which must be 8 byte aligned in
427 td->td_frame = (struct trapframe *)STACKALIGN((u_int)td->td_kstack +
428 USPACE_SVC_STACK_TOP - sizeof(struct pcb) -
429 sizeof(struct trapframe));
431 #ifndef CPU_XSCALE_CORE3
432 pmap_use_minicache(td->td_kstack, td->td_kstack_pages * PAGE_SIZE);
438 cpu_thread_free(struct thread *td)
443 cpu_thread_clean(struct thread *td)
448 * Intercept the return address from a freshly forked process that has NOT
449 * been scheduled yet.
451 * This is needed to make kernel threads stay in kernel mode.
454 cpu_set_fork_handler(struct thread *td, void (*func)(void *), void *arg)
456 struct switchframe *sf;
457 struct trapframe *tf;
460 sf = (struct switchframe *)tf - 1;
461 sf->sf_r4 = (u_int)func;
462 sf->sf_r5 = (u_int)arg;
463 td->td_pcb->un_32.pcb32_sp = (u_int)sf;
464 KASSERT((td->td_pcb->un_32.pcb32_sp & 7) == 0,
465 ("cpu_set_fork_handler: Incorrect stack alignment"));
469 * Software interrupt handler for queued VM system processing.
475 if (busdma_swi_pending)
480 cpu_exit(struct thread *td)