1 /* $NetBSD: vm_machdep.c,v 1.63 2002/09/25 22:21:21 thorpej Exp $ */ 2 3 /* 4 * Copyright (c) 1996 5 * The President and Fellows of Harvard College. All rights reserved. 6 * Copyright (c) 1992, 1993 7 * The Regents of the University of California. All rights reserved. 8 * 9 * This software was developed by the Computer Systems Engineering group 10 * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and 11 * contributed to Berkeley. 12 * 13 * All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Lawrence Berkeley Laboratory. 17 * This product includes software developed by Harvard University. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions 21 * are met: 22 * 1. Redistributions of source code must retain the above copyright 23 * notice, this list of conditions and the following disclaimer. 24 * 2. Redistributions in binary form must reproduce the above copyright 25 * notice, this list of conditions and the following disclaimer in the 26 * documentation and/or other materials provided with the distribution. 27 * 3. All advertising materials mentioning features or use of this software 28 * must display the following acknowledgement: 29 * This product includes software developed by Harvard University. 30 * This product includes software developed by the University of 31 * California, Berkeley and its contributors. 32 * 4. Neither the name of the University nor the names of its contributors 33 * may be used to endorse or promote products derived from this software 34 * without specific prior written permission. 35 * 36 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 37 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 38 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 39 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 40 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 41 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 42 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 43 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 44 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 45 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 46 * SUCH DAMAGE. 47 * 48 * @(#)vm_machdep.c 8.2 (Berkeley) 9/23/93 49 */ 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/proc.h> 54 #include <sys/user.h> 55 #include <sys/core.h> 56 #include <sys/malloc.h> 57 #include <sys/buf.h> 58 #include <sys/exec.h> 59 #include <sys/vnode.h> 60 61 #include <uvm/uvm_extern.h> 62 63 #include <machine/cpu.h> 64 #include <machine/frame.h> 65 #include <machine/trap.h> 66 67 #include <sparc/sparc/cpuvar.h> 68 69 /* 70 * Move pages from one kernel virtual address to another. 71 */ 72 void 73 pagemove(from, to, size) 74 caddr_t from, to; 75 size_t size; 76 { 77 paddr_t pa; 78 79 if (size & PGOFSET || (int)from & PGOFSET || (int)to & PGOFSET) 80 panic("pagemove 1"); 81 while (size > 0) { 82 if (pmap_extract(pmap_kernel(), (vaddr_t)from, &pa) == FALSE) 83 panic("pagemove 2"); 84 pmap_kremove((vaddr_t)from, PAGE_SIZE); 85 pmap_kenter_pa((vaddr_t)to, pa, VM_PROT_READ | VM_PROT_WRITE); 86 from += PAGE_SIZE; 87 to += PAGE_SIZE; 88 size -= PAGE_SIZE; 89 } 90 pmap_update(pmap_kernel()); 91 } 92 93 94 /* 95 * Map a user I/O request into kernel virtual address space. 96 * Note: the pages are already locked by uvm_vslock(), so we 97 * do not need to pass an access_type to pmap_enter(). 98 */ 99 void 100 vmapbuf(bp, len) 101 struct buf *bp; 102 vsize_t len; 103 { 104 struct pmap *upmap, *kpmap; 105 vaddr_t uva; /* User VA (map from) */ 106 vaddr_t kva; /* Kernel VA (new to) */ 107 paddr_t pa; /* physical address */ 108 vsize_t off; 109 110 if ((bp->b_flags & B_PHYS) == 0) 111 panic("vmapbuf"); 112 113 /* 114 * XXX: It might be better to round/trunc to a 115 * segment boundary to avoid VAC problems! 116 */ 117 bp->b_saveaddr = bp->b_data; 118 uva = trunc_page((vaddr_t)bp->b_data); 119 off = (vaddr_t)bp->b_data - uva; 120 len = round_page(off + len); 121 kva = uvm_km_valloc_wait(kernel_map, len); 122 bp->b_data = (caddr_t)(kva + off); 123 124 /* 125 * We have to flush any write-back cache on the 126 * user-space mappings so our new mappings will 127 * have the correct contents. 128 */ 129 if (CACHEINFO.c_vactype != VAC_NONE) 130 cpuinfo.cache_flush((caddr_t)uva, len); 131 132 upmap = vm_map_pmap(&bp->b_proc->p_vmspace->vm_map); 133 kpmap = vm_map_pmap(kernel_map); 134 do { 135 if (pmap_extract(upmap, uva, &pa) == FALSE) 136 panic("vmapbuf: null page frame"); 137 /* Now map the page into kernel space. */ 138 pmap_enter(kpmap, kva, pa, 139 VM_PROT_READ|VM_PROT_WRITE, PMAP_WIRED); 140 uva += PAGE_SIZE; 141 kva += PAGE_SIZE; 142 len -= PAGE_SIZE; 143 } while (len); 144 pmap_update(kpmap); 145 } 146 147 /* 148 * Unmap a previously-mapped user I/O request. 149 */ 150 void 151 vunmapbuf(bp, len) 152 struct buf *bp; 153 vsize_t len; 154 { 155 vaddr_t kva; 156 vsize_t off; 157 158 if ((bp->b_flags & B_PHYS) == 0) 159 panic("vunmapbuf"); 160 161 kva = trunc_page((vaddr_t)bp->b_data); 162 off = (vaddr_t)bp->b_data - kva; 163 len = round_page(off + len); 164 pmap_remove(vm_map_pmap(kernel_map), kva, kva + len); 165 pmap_update(vm_map_pmap(kernel_map)); 166 uvm_km_free_wakeup(kernel_map, kva, len); 167 bp->b_data = bp->b_saveaddr; 168 bp->b_saveaddr = NULL; 169 170 #if 0 /* XXX: The flush above is sufficient, right? */ 171 if (CACHEINFO.c_vactype != VAC_NONE) 172 cpuinfo.cache_flush(bp->b_data, len); 173 #endif 174 } 175 176 177 /* 178 * The offset of the topmost frame in the kernel stack. 179 */ 180 #define TOPFRAMEOFF (USPACE-sizeof(struct trapframe)-sizeof(struct frame)) 181 182 /* 183 * Finish a fork operation, with process p2 nearly set up. 184 * Copy and update the pcb and trap frame, making the child ready to run. 185 * 186 * Rig the child's kernel stack so that it will start out in 187 * proc_trampoline() and call child_return() with p2 as an 188 * argument. This causes the newly-created child process to go 189 * directly to user level with an apparent return value of 0 from 190 * fork(), while the parent process returns normally. 191 * 192 * p1 is the process being forked; if p1 == &proc0, we are creating 193 * a kernel thread, and the return path and argument are specified with 194 * `func' and `arg'. 195 * 196 * If an alternate user-level stack is requested (with non-zero values 197 * in both the stack and stacksize args), set up the user stack pointer 198 * accordingly. 199 */ 200 void 201 cpu_fork(p1, p2, stack, stacksize, func, arg) 202 struct proc *p1, *p2; 203 void *stack; 204 size_t stacksize; 205 void (*func) __P((void *)); 206 void *arg; 207 { 208 struct pcb *opcb = &p1->p_addr->u_pcb; 209 struct pcb *npcb = &p2->p_addr->u_pcb; 210 struct trapframe *tf2; 211 struct rwindow *rp; 212 213 /* 214 * Save all user registers to p1's stack or, in the case of 215 * user registers and invalid stack pointers, to opcb. 216 * We then copy the whole pcb to p2; when switch() selects p2 217 * to run, it will run at the `proc_trampoline' stub, rather 218 * than returning at the copying code below. 219 * 220 * If process p1 has an FPU state, we must copy it. If it is 221 * the FPU user, we must save the FPU state first. 222 */ 223 224 if (p1 == curproc) { 225 write_user_windows(); 226 opcb->pcb_psr = getpsr(); 227 } 228 #ifdef DIAGNOSTIC 229 else if (p1 != &proc0) 230 panic("cpu_fork: curproc"); 231 #endif 232 233 bcopy((caddr_t)opcb, (caddr_t)npcb, sizeof(struct pcb)); 234 if (p1->p_md.md_fpstate) { 235 if (p1 == cpuinfo.fpproc) 236 savefpstate(p1->p_md.md_fpstate); 237 else if (p1->p_md.md_fpumid != -1) 238 panic("FPU on module %d; fix this", p1->p_md.md_fpumid); 239 p2->p_md.md_fpstate = malloc(sizeof(struct fpstate), 240 M_SUBPROC, M_WAITOK); 241 bcopy(p1->p_md.md_fpstate, p2->p_md.md_fpstate, 242 sizeof(struct fpstate)); 243 } else 244 p2->p_md.md_fpstate = NULL; 245 246 p2->p_md.md_fpumid = -1; 247 248 /* 249 * Setup (kernel) stack frame that will by-pass the child 250 * out of the kernel. (The trap frame invariably resides at 251 * the tippity-top of the u. area.) 252 */ 253 tf2 = p2->p_md.md_tf = (struct trapframe *) 254 ((int)npcb + USPACE - sizeof(*tf2)); 255 256 /* Copy parent's trapframe */ 257 *tf2 = *(struct trapframe *)((int)opcb + USPACE - sizeof(*tf2)); 258 259 /* 260 * If specified, give the child a different stack. 261 */ 262 if (stack != NULL) 263 tf2->tf_out[6] = (u_int)stack + stacksize; 264 265 /* 266 * The fork system call always uses the old system call 267 * convention; clear carry and skip trap instruction as 268 * in syscall(). 269 * note: proc_trampoline() sets a fresh psr when returning 270 * to user mode. 271 */ 272 /*tf2->tf_psr &= ~PSR_C; -* success */ 273 tf2->tf_pc = tf2->tf_npc; 274 tf2->tf_npc = tf2->tf_pc + 4; 275 276 /* Set return values in child mode */ 277 tf2->tf_out[0] = 0; 278 tf2->tf_out[1] = 1; 279 280 /* Construct kernel frame to return to in cpu_switch() */ 281 rp = (struct rwindow *)((u_int)npcb + TOPFRAMEOFF); 282 rp->rw_local[0] = (int)func; /* Function to call */ 283 rp->rw_local[1] = (int)arg; /* and its argument */ 284 285 npcb->pcb_pc = (int)proc_trampoline - 8; 286 npcb->pcb_sp = (int)rp; 287 npcb->pcb_psr &= ~PSR_CWP; /* Run in window #0 */ 288 npcb->pcb_wim = 1; /* Fence at window #1 */ 289 } 290 291 /* 292 * cpu_exit is called as the last action during exit. 293 * 294 * We clean up a little and then call switchexit() with the old proc 295 * as an argument. switchexit() switches to the idle context, schedules 296 * the old vmspace and stack to be freed, then selects a new process to 297 * run. 298 */ 299 void 300 cpu_exit(p) 301 struct proc *p; 302 { 303 struct fpstate *fs; 304 305 if ((fs = p->p_md.md_fpstate) != NULL) { 306 if (p == cpuinfo.fpproc) { 307 savefpstate(fs); 308 cpuinfo.fpproc = NULL; 309 } 310 free((void *)fs, M_SUBPROC); 311 } 312 switchexit(p); 313 /* NOTREACHED */ 314 } 315 316 /* 317 * cpu_coredump is called to write a core dump header. 318 * (should this be defined elsewhere? machdep.c?) 319 */ 320 int 321 cpu_coredump(p, vp, cred, chdr) 322 struct proc *p; 323 struct vnode *vp; 324 struct ucred *cred; 325 struct core *chdr; 326 { 327 int error; 328 struct md_coredump md_core; 329 struct coreseg cseg; 330 331 CORE_SETMAGIC(*chdr, COREMAGIC, MID_MACHINE, 0); 332 chdr->c_hdrsize = ALIGN(sizeof(*chdr)); 333 chdr->c_seghdrsize = ALIGN(sizeof(cseg)); 334 chdr->c_cpusize = sizeof(md_core); 335 336 md_core.md_tf = *p->p_md.md_tf; 337 if (p->p_md.md_fpstate) { 338 if (p == cpuinfo.fpproc) 339 savefpstate(p->p_md.md_fpstate); 340 md_core.md_fpstate = *p->p_md.md_fpstate; 341 } else 342 bzero((caddr_t)&md_core.md_fpstate, sizeof(struct fpstate)); 343 344 CORE_SETMAGIC(cseg, CORESEGMAGIC, MID_MACHINE, CORE_CPU); 345 cseg.c_addr = 0; 346 cseg.c_size = chdr->c_cpusize; 347 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&cseg, chdr->c_seghdrsize, 348 (off_t)chdr->c_hdrsize, UIO_SYSSPACE, 349 IO_NODELOCKED|IO_UNIT, cred, NULL, p); 350 if (error) 351 return error; 352 353 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&md_core, sizeof(md_core), 354 (off_t)(chdr->c_hdrsize + chdr->c_seghdrsize), UIO_SYSSPACE, 355 IO_NODELOCKED|IO_UNIT, cred, NULL, p); 356 if (!error) 357 chdr->c_nseg++; 358 359 return error; 360 } 361