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
5 * Copyright (c) 2001 Jake Burkholder.
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
10 * Science Department, and William Jolitz.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91
37 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
38 * from: FreeBSD: src/sys/i386/i386/vm_machdep.c,v 1.167 2001/07/12
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/linker_set.h>
52 #include <sys/malloc.h>
54 #include <sys/mutex.h>
56 #include <sys/sysent.h>
57 #include <sys/sf_buf.h>
58 #include <sys/sched.h>
59 #include <sys/sysctl.h>
60 #include <sys/unistd.h>
61 #include <sys/vmmeter.h>
63 #include <dev/ofw/openfirm.h>
66 #include <vm/vm_extern.h>
68 #include <vm/vm_kern.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_pageout.h>
72 #include <vm/vm_param.h>
74 #include <vm/uma_int.h>
76 #include <machine/cache.h>
77 #include <machine/cpu.h>
78 #include <machine/fp.h>
79 #include <machine/frame.h>
80 #include <machine/fsr.h>
81 #include <machine/md_var.h>
82 #include <machine/ofw_machdep.h>
83 #include <machine/ofw_mem.h>
84 #include <machine/pcb.h>
85 #include <machine/tlb.h>
86 #include <machine/tstate.h>
89 #define NSFBUFS (512 + maxusers * 16)
92 static void sf_buf_init(void *arg);
93 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
96 * Expanded sf_freelist head. Really an SLIST_HEAD() in disguise, with the
97 * sf_freelist head with the sf_lock mutex.
100 SLIST_HEAD(, sf_buf) sf_head;
104 static u_int sf_buf_alloc_want;
106 PMAP_STATS_VAR(uma_nsmall_alloc);
107 PMAP_STATS_VAR(uma_nsmall_alloc_oc);
108 PMAP_STATS_VAR(uma_nsmall_free);
111 cpu_exit(struct thread *td)
116 p->p_md.md_sigtramp = NULL;
117 if (p->p_md.md_utrap != NULL) {
118 utrap_free(p->p_md.md_utrap);
119 p->p_md.md_utrap = NULL;
124 cpu_thread_exit(struct thread *td)
130 cpu_thread_clean(struct thread *td)
136 cpu_thread_alloc(struct thread *td)
140 pcb = (struct pcb *)((td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
141 sizeof(struct pcb)) & ~0x3fUL);
143 td->td_frame = (struct trapframe *)pcb - 1;
148 cpu_thread_free(struct thread *td)
154 cpu_thread_swapin(struct thread *td)
160 cpu_thread_swapout(struct thread *td)
166 cpu_set_syscall_retval(struct thread *td, int error)
171 td->td_frame->tf_out[0] = td->td_retval[0];
172 td->td_frame->tf_out[1] = td->td_retval[1];
173 td->td_frame->tf_tstate &= ~TSTATE_XCC_C;
178 * Undo the tpc advancement we have done on syscall
179 * enter, we want to reexecute the system call.
181 td->td_frame->tf_tpc = td->td_pcb->pcb_tpc;
182 td->td_frame->tf_tnpc -= 4;
189 if (td->td_proc->p_sysent->sv_errsize) {
190 if (error >= td->td_proc->p_sysent->sv_errsize)
191 error = -1; /* XXX */
193 error = td->td_proc->p_sysent->sv_errtbl[error];
195 td->td_frame->tf_out[0] = error;
196 td->td_frame->tf_tstate |= TSTATE_XCC_C;
202 cpu_set_upcall(struct thread *td, struct thread *td0)
204 struct trapframe *tf;
208 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
212 fr = (struct frame *)tf - 1;
213 fr->fr_local[0] = (u_long)fork_return;
214 fr->fr_local[1] = (u_long)td;
215 fr->fr_local[2] = (u_long)tf;
216 pcb->pcb_pc = (u_long)fork_trampoline - 8;
217 pcb->pcb_sp = (u_long)fr - SPOFF;
219 /* Setup to release the spin count in fork_exit(). */
220 td->td_md.md_spinlock_count = 1;
221 td->td_md.md_saved_pil = 0;
225 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
228 struct trapframe *tf;
234 sp = (uint64_t)stack->ss_sp + stack->ss_size;
235 tf->tf_out[0] = (uint64_t)arg;
236 tf->tf_out[6] = sp - SPOFF - sizeof(struct frame);
237 tf->tf_tpc = (uint64_t)entry;
238 tf->tf_tnpc = tf->tf_tpc + 4;
240 td->td_retval[0] = tf->tf_out[0];
241 td->td_retval[1] = tf->tf_out[1];
245 cpu_set_user_tls(struct thread *td, void *tls_base)
250 td->td_frame->tf_global[7] = (uint64_t)tls_base;
255 * Finish a fork operation, with process p2 nearly set up.
256 * Copy and update the pcb, set up the stack so that the child
257 * ready to run and return to user mode.
260 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
262 struct trapframe *tf;
270 KASSERT(td1 == curthread || td1 == &thread0,
271 ("cpu_fork: p1 not curproc and not proc0"));
273 if ((flags & RFPROC) == 0)
276 p2->p_md.md_sigtramp = td1->td_proc->p_md.md_sigtramp;
277 p2->p_md.md_utrap = utrap_hold(td1->td_proc->p_md.md_utrap);
279 /* The pcb must be aligned on a 64-byte boundary. */
281 pcb2 = (struct pcb *)((td2->td_kstack + td2->td_kstack_pages *
282 PAGE_SIZE - sizeof(struct pcb)) & ~0x3fUL);
286 * Ensure that p1's pcb is up to date.
289 if ((td1->td_frame->tf_fprs & FPRS_FEF) != 0)
290 savefpctx(pcb1->pcb_ufp);
292 /* Make sure the copied windows are spilled. */
294 /* Copy the pcb (this will copy the windows saved in the pcb, too). */
295 bcopy(pcb1, pcb2, sizeof(*pcb1));
298 * If we're creating a new user process and we're sharing the address
299 * space, the parent's top most frame must be saved in the pcb. The
300 * child will pop the frame when it returns to user mode, and may
301 * overwrite it with its own data causing much suffering for the
302 * parent. We check if its already in the pcb, and if not copy it
303 * in. Its unlikely that the copyin will fail, but if so there's not
304 * much we can do. The parent will likely crash soon anyway in that
307 if ((flags & RFMEM) != 0 && td1 != &thread0) {
308 sp = td1->td_frame->tf_sp;
309 for (i = 0; i < pcb1->pcb_nsaved; i++) {
310 if (pcb1->pcb_rwsp[i] == sp)
313 if (i == pcb1->pcb_nsaved) {
314 error = copyin((caddr_t)sp + SPOFF, &pcb1->pcb_rw[i],
315 sizeof(struct rwindow));
317 pcb1->pcb_rwsp[i] = sp;
324 * Create a new fresh stack for the new process.
325 * Copy the trap frame for the return to user mode as if from a
326 * syscall. This copies most of the user mode register values.
328 tf = (struct trapframe *)pcb2 - 1;
329 bcopy(td1->td_frame, tf, sizeof(*tf));
331 tf->tf_out[0] = 0; /* Child returns zero */
333 tf->tf_tstate &= ~TSTATE_XCC_C; /* success */
337 fp = (struct frame *)tf - 1;
338 fp->fr_local[0] = (u_long)fork_return;
339 fp->fr_local[1] = (u_long)td2;
340 fp->fr_local[2] = (u_long)tf;
341 /* Terminate stack traces at this frame. */
342 fp->fr_pc = fp->fr_fp = 0;
343 pcb2->pcb_sp = (u_long)fp - SPOFF;
344 pcb2->pcb_pc = (u_long)fork_trampoline - 8;
346 /* Setup to release the spin count in fork_exit(). */
347 td2->td_md.md_spinlock_count = 1;
348 td2->td_md.md_saved_pil = 0;
351 * Now, cpu_switch() can schedule the new process.
358 static char bspec[64] = "";
372 if ((chosen = OF_finddevice("/chosen")) != 0) {
373 if (OF_getprop(chosen, "bootpath", bspec, sizeof(bspec)) == -1)
375 bspec[sizeof(bspec) - 1] = '\0';
382 * Intercept the return address from a freshly forked process that has NOT
383 * been scheduled yet.
385 * This is needed to make kernel threads stay in kernel mode.
388 cpu_set_fork_handler(struct thread *td, void (*func)(void *), void *arg)
394 fp = (struct frame *)(pcb->pcb_sp + SPOFF);
395 fp->fr_local[0] = (u_long)func;
396 fp->fr_local[1] = (u_long)arg;
400 is_physical_memory(vm_paddr_t addr)
402 struct ofw_mem_region *mr;
404 for (mr = sparc64_memreg; mr < sparc64_memreg + sparc64_nmemreg; mr++)
405 if (addr >= mr->mr_start && addr < mr->mr_start + mr->mr_size)
411 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
414 sf_buf_init(void *arg)
416 struct sf_buf *sf_bufs;
421 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
423 mtx_init(&sf_freelist.sf_lock, "sf_bufs list lock", NULL, MTX_DEF);
424 SLIST_INIT(&sf_freelist.sf_head);
425 sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE);
426 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
428 for (i = 0; i < nsfbufs; i++) {
429 sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
430 SLIST_INSERT_HEAD(&sf_freelist.sf_head, &sf_bufs[i], free_list);
432 sf_buf_alloc_want = 0;
436 * Get an sf_buf from the freelist. Will block if none are available.
439 sf_buf_alloc(struct vm_page *m, int flags)
444 mtx_lock(&sf_freelist.sf_lock);
445 while ((sf = SLIST_FIRST(&sf_freelist.sf_head)) == NULL) {
446 if (flags & SFB_NOWAIT)
449 mbstat.sf_allocwait++;
450 error = msleep(&sf_freelist, &sf_freelist.sf_lock,
451 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
455 * If we got a signal, don't risk going back to sleep.
461 SLIST_REMOVE_HEAD(&sf_freelist.sf_head, free_list);
464 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
465 pmap_qenter(sf->kva, &sf->m, 1);
467 mtx_unlock(&sf_freelist.sf_lock);
472 * Release resources back to the system.
475 sf_buf_free(struct sf_buf *sf)
478 pmap_qremove(sf->kva, 1);
479 mtx_lock(&sf_freelist.sf_lock);
480 SLIST_INSERT_HEAD(&sf_freelist.sf_head, sf, free_list);
482 if (sf_buf_alloc_want > 0)
483 wakeup(&sf_freelist);
484 mtx_unlock(&sf_freelist.sf_lock);
491 /* Nothing to do here - busdma bounce buffers are not implemented. */
495 uma_small_alloc(uma_zone_t zone, int bytes, u_int8_t *flags, int wait)
497 static vm_pindex_t color;
503 PMAP_STATS_INC(uma_nsmall_alloc);
505 *flags = UMA_SLAB_PRIV;
507 if ((wait & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
508 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
510 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
513 pflags |= VM_ALLOC_ZERO;
516 m = vm_page_alloc(NULL, color++, pflags | VM_ALLOC_NOOBJ);
526 pa = VM_PAGE_TO_PHYS(m);
527 if (dcache_color_ignore == 0 && m->md.color != DCACHE_COLOR(pa)) {
528 KASSERT(m->md.colors[0] == 0 && m->md.colors[1] == 0,
529 ("uma_small_alloc: free page still has mappings!"));
530 PMAP_STATS_INC(uma_nsmall_alloc_oc);
531 m->md.color = DCACHE_COLOR(pa);
532 dcache_page_inval(pa);
534 va = (void *)TLB_PHYS_TO_DIRECT(pa);
535 if ((wait & M_ZERO) && (m->flags & PG_ZERO) == 0)
536 cpu_block_zero(va, PAGE_SIZE);
541 uma_small_free(void *mem, int size, u_int8_t flags)
545 PMAP_STATS_INC(uma_nsmall_free);
546 m = PHYS_TO_VM_PAGE(TLB_DIRECT_TO_PHYS((vm_offset_t)mem));
549 atomic_subtract_int(&cnt.v_wire_count, 1);