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>
77 #include <machine/vfp.h>
80 * struct switchframe and trapframe must both be a multiple of 8
81 * for correct stack alignment.
83 CTASSERT(sizeof(struct switchframe) == 24);
84 CTASSERT(sizeof(struct trapframe) == 80);
87 #define NSFBUFS (512 + maxusers * 16)
91 static int nsfbufspeak;
92 static int nsfbufsused;
94 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufs, CTLFLAG_RDTUN, &nsfbufs, 0,
95 "Maximum number of sendfile(2) sf_bufs available");
96 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufspeak, CTLFLAG_RD, &nsfbufspeak, 0,
97 "Number of sendfile(2) sf_bufs at peak usage");
98 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufsused, CTLFLAG_RD, &nsfbufsused, 0,
99 "Number of sendfile(2) sf_bufs in use");
101 static void sf_buf_init(void *arg);
102 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
104 LIST_HEAD(sf_head, sf_buf);
107 * A hash table of active sendfile(2) buffers
109 static struct sf_head *sf_buf_active;
110 static u_long sf_buf_hashmask;
112 #define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask)
114 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
115 static u_int sf_buf_alloc_want;
118 * A lock used to synchronize access to the hash table and free list
120 static struct mtx sf_buf_lock;
123 * Finish a fork operation, with process p2 nearly set up.
124 * Copy and update the pcb, set up the stack so that the child
125 * ready to run and return to user mode.
128 cpu_fork(register struct thread *td1, register struct proc *p2,
129 struct thread *td2, int flags)
132 struct trapframe *tf;
133 struct switchframe *sf;
136 if ((flags & RFPROC) == 0)
138 pcb2 = (struct pcb *)(td2->td_kstack + td2->td_kstack_pages * PAGE_SIZE) - 1;
140 #ifndef CPU_XSCALE_CORE3
141 pmap_use_minicache(td2->td_kstack, td2->td_kstack_pages * PAGE_SIZE);
145 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
147 bcopy(&td1->td_proc->p_md, mdp2, sizeof(*mdp2));
148 pcb2->un_32.pcb32_sp = td2->td_kstack +
149 USPACE_SVC_STACK_TOP - sizeof(*pcb2);
150 pcb2->pcb_vfpcpu = -1;
151 pcb2->pcb_vfpstate.fpscr = VFPSCR_DN | VFPSCR_FZ;
153 td2->td_frame = tf = (struct trapframe *)STACKALIGN(
154 pcb2->un_32.pcb32_sp - sizeof(struct trapframe));
155 *tf = *td1->td_frame;
156 sf = (struct switchframe *)tf - 1;
157 sf->sf_r4 = (u_int)fork_return;
158 sf->sf_r5 = (u_int)td2;
159 sf->sf_pc = (u_int)fork_trampoline;
160 tf->tf_spsr &= ~PSR_C;
163 pcb2->un_32.pcb32_sp = (u_int)sf;
164 KASSERT((pcb2->un_32.pcb32_sp & 7) == 0,
165 ("cpu_fork: Incorrect stack alignment"));
167 /* Setup to release spin count in fork_exit(). */
168 td2->td_md.md_spinlock_count = 1;
169 td2->td_md.md_saved_cspr = 0;
170 #ifdef ARM_TP_ADDRESS
171 td2->td_md.md_tp = *(register_t *)ARM_TP_ADDRESS;
173 td2->td_md.md_tp = (register_t) get_tls();
178 cpu_thread_swapin(struct thread *td)
183 cpu_thread_swapout(struct thread *td)
188 * Detatch mapped page and release resources back to the system.
191 sf_buf_free(struct sf_buf *sf)
194 mtx_lock(&sf_buf_lock);
196 if (sf->ref_count == 0) {
197 TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
199 pmap_kremove(sf->kva);
201 LIST_REMOVE(sf, list_entry);
202 if (sf_buf_alloc_want > 0)
203 wakeup(&sf_buf_freelist);
205 mtx_unlock(&sf_buf_lock);
209 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
212 sf_buf_init(void *arg)
214 struct sf_buf *sf_bufs;
219 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
221 sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
222 TAILQ_INIT(&sf_buf_freelist);
223 sf_base = kva_alloc(nsfbufs * PAGE_SIZE);
224 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
226 for (i = 0; i < nsfbufs; i++) {
227 sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
228 TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
230 sf_buf_alloc_want = 0;
231 mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
235 * Get an sf_buf from the freelist. Will block if none are available.
238 sf_buf_alloc(struct vm_page *m, int flags)
240 struct sf_head *hash_list;
244 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
245 mtx_lock(&sf_buf_lock);
246 LIST_FOREACH(sf, hash_list, list_entry) {
249 if (sf->ref_count == 1) {
250 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
252 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
257 while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
258 if (flags & SFB_NOWAIT)
261 SFSTAT_INC(sf_allocwait);
262 error = msleep(&sf_buf_freelist, &sf_buf_lock,
263 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
268 * If we got a signal, don't risk going back to sleep.
273 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
275 LIST_REMOVE(sf, list_entry);
276 LIST_INSERT_HEAD(hash_list, sf, list_entry);
280 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
281 pmap_kenter(sf->kva, VM_PAGE_TO_PHYS(sf->m));
283 mtx_unlock(&sf_buf_lock);
288 cpu_set_syscall_retval(struct thread *td, int error)
290 struct trapframe *frame;
296 frame = td->td_frame;
301 * __syscall returns an off_t while most other syscalls return an
302 * int. As an off_t is 64-bits and an int is 32-bits we need to
303 * place the returned data into r1. As the lseek and frerebsd6_lseek
304 * syscalls also return an off_t they do not need this fixup.
309 call = *(u_int32_t *)(frame->tf_pc - INSN_SIZE) & 0x000fffff;
311 if (call == SYS___syscall) {
312 register_t *ap = &frame->tf_r0;
313 register_t code = ap[_QUAD_LOWWORD];
314 if (td->td_proc->p_sysent->sv_mask)
315 code &= td->td_proc->p_sysent->sv_mask;
316 fixup = (code != SYS_freebsd6_lseek && code != SYS_lseek)
325 frame->tf_r1 = td->td_retval[0];
327 frame->tf_r0 = td->td_retval[0];
328 frame->tf_r1 = td->td_retval[1];
330 frame->tf_spsr &= ~PSR_C; /* carry bit */
334 * Reconstruct the pc to point at the swi.
336 frame->tf_pc -= INSN_SIZE;
342 frame->tf_r0 = error;
343 frame->tf_spsr |= PSR_C; /* carry bit */
349 * Initialize machine state (pcb and trap frame) for a new thread about to
350 * upcall. Put enough state in the new thread's PCB to get it to go back
351 * userret(), where we can intercept it again to set the return (upcall)
352 * Address and stack, along with those from upcals that are from other sources
353 * such as those generated in thread_userret() itself.
356 cpu_set_upcall(struct thread *td, struct thread *td0)
358 struct trapframe *tf;
359 struct switchframe *sf;
361 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
362 bcopy(td0->td_pcb, td->td_pcb, sizeof(struct pcb));
364 sf = (struct switchframe *)tf - 1;
365 sf->sf_r4 = (u_int)fork_return;
366 sf->sf_r5 = (u_int)td;
367 sf->sf_pc = (u_int)fork_trampoline;
368 tf->tf_spsr &= ~PSR_C;
370 td->td_pcb->un_32.pcb32_sp = (u_int)sf;
371 KASSERT((td->td_pcb->un_32.pcb32_sp & 7) == 0,
372 ("cpu_set_upcall: Incorrect stack alignment"));
374 /* Setup to release spin count in fork_exit(). */
375 td->td_md.md_spinlock_count = 1;
376 td->td_md.md_saved_cspr = 0;
380 * Set that machine state for performing an upcall that has to
381 * be done in thread_userret() so that those upcalls generated
382 * in thread_userret() itself can be done as well.
385 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
388 struct trapframe *tf = td->td_frame;
390 tf->tf_usr_sp = STACKALIGN((int)stack->ss_sp + stack->ss_size
391 - sizeof(struct trapframe));
392 tf->tf_pc = (int)entry;
393 tf->tf_r0 = (int)arg;
394 tf->tf_spsr = PSR_USR32_MODE;
398 cpu_set_user_tls(struct thread *td, void *tls_base)
401 td->td_md.md_tp = (register_t)tls_base;
402 if (td == curthread) {
404 #ifdef ARM_TP_ADDRESS
405 *(register_t *)ARM_TP_ADDRESS = (register_t)tls_base;
407 set_tls((void *)tls_base);
415 cpu_thread_exit(struct thread *td)
420 cpu_thread_alloc(struct thread *td)
422 td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_pages *
425 * Ensure td_frame is aligned to an 8 byte boundary as it will be
426 * placed into the stack pointer which must be 8 byte aligned in
429 td->td_frame = (struct trapframe *)STACKALIGN((u_int)td->td_kstack +
430 USPACE_SVC_STACK_TOP - sizeof(struct pcb) -
431 sizeof(struct trapframe));
433 #ifndef CPU_XSCALE_CORE3
434 pmap_use_minicache(td->td_kstack, td->td_kstack_pages * PAGE_SIZE);
440 cpu_thread_free(struct thread *td)
445 cpu_thread_clean(struct thread *td)
450 * Intercept the return address from a freshly forked process that has NOT
451 * been scheduled yet.
453 * This is needed to make kernel threads stay in kernel mode.
456 cpu_set_fork_handler(struct thread *td, void (*func)(void *), void *arg)
458 struct switchframe *sf;
459 struct trapframe *tf;
462 sf = (struct switchframe *)tf - 1;
463 sf->sf_r4 = (u_int)func;
464 sf->sf_r5 = (u_int)arg;
465 td->td_pcb->un_32.pcb32_sp = (u_int)sf;
466 KASSERT((td->td_pcb->un_32.pcb32_sp & 7) == 0,
467 ("cpu_set_fork_handler: Incorrect stack alignment"));
471 * Software interrupt handler for queued VM system processing.
477 if (busdma_swi_pending)
482 cpu_exit(struct thread *td)