1 /* $NetBSD: vm_machdep.c,v 1.58 2001/07/29 23:56:33 chs 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 #include <sys/map.h> 61 62 #include <uvm/uvm_extern.h> 63 64 #include <machine/cpu.h> 65 #include <machine/frame.h> 66 #include <machine/trap.h> 67 68 #include <sparc/sparc/cpuvar.h> 69 70 /* 71 * Move pages from one kernel virtual address to another. 72 */ 73 void 74 pagemove(from, to, size) 75 caddr_t from, to; 76 size_t size; 77 { 78 paddr_t pa; 79 80 if (size & PGOFSET || (int)from & PGOFSET || (int)to & PGOFSET) 81 panic("pagemove 1"); 82 while (size > 0) { 83 if (pmap_extract(pmap_kernel(), (vaddr_t)from, &pa) == FALSE) 84 panic("pagemove 2"); 85 pmap_kremove((vaddr_t)from, PAGE_SIZE); 86 pmap_kenter_pa((vaddr_t)to, pa, VM_PROT_READ | VM_PROT_WRITE); 87 from += PAGE_SIZE; 88 to += PAGE_SIZE; 89 size -= PAGE_SIZE; 90 } 91 pmap_update(); 92 } 93 94 95 /* 96 * Map a user I/O request into kernel virtual address space. 97 * Note: the pages are already locked by uvm_vslock(), so we 98 * do not need to pass an access_type to pmap_enter(). 99 */ 100 void 101 vmapbuf(bp, len) 102 struct buf *bp; 103 vsize_t len; 104 { 105 struct pmap *upmap, *kpmap; 106 vaddr_t uva; /* User VA (map from) */ 107 vaddr_t kva; /* Kernel VA (new to) */ 108 paddr_t pa; /* physical address */ 109 vsize_t off; 110 111 if ((bp->b_flags & B_PHYS) == 0) 112 panic("vmapbuf"); 113 114 /* 115 * XXX: It might be better to round/trunc to a 116 * segment boundary to avoid VAC problems! 117 */ 118 bp->b_saveaddr = bp->b_data; 119 uva = trunc_page((vaddr_t)bp->b_data); 120 off = (vaddr_t)bp->b_data - uva; 121 len = round_page(off + len); 122 kva = uvm_km_valloc_wait(kernel_map, len); 123 bp->b_data = (caddr_t)(kva + off); 124 125 /* 126 * We have to flush any write-back cache on the 127 * user-space mappings so our new mappings will 128 * have the correct contents. 129 */ 130 if (CACHEINFO.c_vactype != VAC_NONE) 131 cpuinfo.cache_flush((caddr_t)uva, len); 132 133 upmap = vm_map_pmap(&bp->b_proc->p_vmspace->vm_map); 134 kpmap = vm_map_pmap(kernel_map); 135 do { 136 if (pmap_extract(upmap, uva, &pa) == FALSE) 137 panic("vmapbuf: null page frame"); 138 /* Now map the page into kernel space. */ 139 pmap_enter(kpmap, kva, pa, 140 VM_PROT_READ|VM_PROT_WRITE, PMAP_WIRED); 141 uva += PAGE_SIZE; 142 kva += PAGE_SIZE; 143 len -= PAGE_SIZE; 144 } while (len); 145 pmap_update(); 146 } 147 148 /* 149 * Unmap a previously-mapped user I/O request. 150 */ 151 void 152 vunmapbuf(bp, len) 153 struct buf *bp; 154 vsize_t len; 155 { 156 vaddr_t kva; 157 vsize_t off; 158 159 if ((bp->b_flags & B_PHYS) == 0) 160 panic("vunmapbuf"); 161 162 kva = trunc_page((vaddr_t)bp->b_data); 163 off = (vaddr_t)bp->b_data - kva; 164 len = round_page(off + len); 165 166 /* This will call pmap_remove() for us. */ 167 uvm_km_free_wakeup(kernel_map, kva, len); 168 bp->b_data = bp->b_saveaddr; 169 bp->b_saveaddr = NULL; 170 171 #if 0 /* XXX: The flush above is sufficient, right? */ 172 if (CACHEINFO.c_vactype != VAC_NONE) 173 cpuinfo.cache_flush(bp->b_data, len); 174 #endif 175 } 176 177 178 /* 179 * The offset of the topmost frame in the kernel stack. 180 */ 181 #define TOPFRAMEOFF (USPACE-sizeof(struct trapframe)-sizeof(struct frame)) 182 183 /* 184 * Finish a fork operation, with process p2 nearly set up. 185 * Copy and update the pcb and trap frame, making the child ready to run. 186 * 187 * Rig the child's kernel stack so that it will start out in 188 * proc_trampoline() and call child_return() with p2 as an 189 * argument. This causes the newly-created child process to go 190 * directly to user level with an apparent return value of 0 from 191 * fork(), while the parent process returns normally. 192 * 193 * p1 is the process being forked; if p1 == &proc0, we are creating 194 * a kernel thread, and the return path and argument are specified with 195 * `func' and `arg'. 196 * 197 * If an alternate user-level stack is requested (with non-zero values 198 * in both the stack and stacksize args), set up the user stack pointer 199 * accordingly. 200 */ 201 void 202 cpu_fork(p1, p2, stack, stacksize, func, arg) 203 struct proc *p1, *p2; 204 void *stack; 205 size_t stacksize; 206 void (*func) __P((void *)); 207 void *arg; 208 { 209 struct pcb *opcb = &p1->p_addr->u_pcb; 210 struct pcb *npcb = &p2->p_addr->u_pcb; 211 struct trapframe *tf2; 212 struct rwindow *rp; 213 214 /* 215 * Save all user registers to p1's stack or, in the case of 216 * user registers and invalid stack pointers, to opcb. 217 * We then copy the whole pcb to p2; when switch() selects p2 218 * to run, it will run at the `proc_trampoline' stub, rather 219 * than returning at the copying code below. 220 * 221 * If process p1 has an FPU state, we must copy it. If it is 222 * the FPU user, we must save the FPU state first. 223 */ 224 225 if (p1 == curproc) { 226 write_user_windows(); 227 opcb->pcb_psr = getpsr(); 228 } 229 #ifdef DIAGNOSTIC 230 else if (p1 != &proc0) 231 panic("cpu_fork: curproc"); 232 #endif 233 234 bcopy((caddr_t)opcb, (caddr_t)npcb, sizeof(struct pcb)); 235 if (p1->p_md.md_fpstate) { 236 if (p1 == cpuinfo.fpproc) 237 savefpstate(p1->p_md.md_fpstate); 238 else if (p1->p_md.md_fpumid != -1) 239 panic("FPU on module %d; fix this", p1->p_md.md_fpumid); 240 p2->p_md.md_fpstate = malloc(sizeof(struct fpstate), 241 M_SUBPROC, M_WAITOK); 242 bcopy(p1->p_md.md_fpstate, p2->p_md.md_fpstate, 243 sizeof(struct fpstate)); 244 } else 245 p2->p_md.md_fpstate = NULL; 246 247 p2->p_md.md_fpumid = -1; 248 249 /* 250 * Setup (kernel) stack frame that will by-pass the child 251 * out of the kernel. (The trap frame invariably resides at 252 * the tippity-top of the u. area.) 253 */ 254 tf2 = p2->p_md.md_tf = (struct trapframe *) 255 ((int)npcb + USPACE - sizeof(*tf2)); 256 257 /* Copy parent's trapframe */ 258 *tf2 = *(struct trapframe *)((int)opcb + USPACE - sizeof(*tf2)); 259 260 /* 261 * If specified, give the child a different stack. 262 */ 263 if (stack != NULL) 264 tf2->tf_out[6] = (u_int)stack + stacksize; 265 266 /* Duplicate efforts of syscall(), but slightly differently */ 267 if (tf2->tf_global[1] & SYSCALL_G2RFLAG) { 268 /* jmp %g2 (or %g7, deprecated) on success */ 269 tf2->tf_npc = tf2->tf_global[2]; 270 } else { 271 /* 272 * old system call convention: clear C on success 273 * note: proc_trampoline() sets a fresh psr when 274 * returning to user mode. 275 */ 276 /*tf2->tf_psr &= ~PSR_C; -* success */ 277 } 278 279 /* Set return values in child mode */ 280 tf2->tf_out[0] = 0; 281 tf2->tf_out[1] = 1; 282 283 /* Construct kernel frame to return to in cpu_switch() */ 284 rp = (struct rwindow *)((u_int)npcb + TOPFRAMEOFF); 285 rp->rw_local[0] = (int)func; /* Function to call */ 286 rp->rw_local[1] = (int)arg; /* and its argument */ 287 288 npcb->pcb_pc = (int)proc_trampoline - 8; 289 npcb->pcb_sp = (int)rp; 290 npcb->pcb_psr &= ~PSR_CWP; /* Run in window #0 */ 291 npcb->pcb_wim = 1; /* Fence at window #1 */ 292 293 } 294 295 /* 296 * cpu_exit is called as the last action during exit. 297 * 298 * We clean up a little and then call switchexit() with the old proc 299 * as an argument. switchexit() switches to the idle context, schedules 300 * the old vmspace and stack to be freed, then selects a new process to 301 * run. 302 */ 303 void 304 cpu_exit(p) 305 struct proc *p; 306 { 307 struct fpstate *fs; 308 309 if ((fs = p->p_md.md_fpstate) != NULL) { 310 if (p == cpuinfo.fpproc) { 311 savefpstate(fs); 312 cpuinfo.fpproc = NULL; 313 } 314 free((void *)fs, M_SUBPROC); 315 } 316 switchexit(p); 317 /* NOTREACHED */ 318 } 319 320 /* 321 * cpu_coredump is called to write a core dump header. 322 * (should this be defined elsewhere? machdep.c?) 323 */ 324 int 325 cpu_coredump(p, vp, cred, chdr) 326 struct proc *p; 327 struct vnode *vp; 328 struct ucred *cred; 329 struct core *chdr; 330 { 331 int error; 332 struct md_coredump md_core; 333 struct coreseg cseg; 334 335 CORE_SETMAGIC(*chdr, COREMAGIC, MID_MACHINE, 0); 336 chdr->c_hdrsize = ALIGN(sizeof(*chdr)); 337 chdr->c_seghdrsize = ALIGN(sizeof(cseg)); 338 chdr->c_cpusize = sizeof(md_core); 339 340 md_core.md_tf = *p->p_md.md_tf; 341 if (p->p_md.md_fpstate) { 342 if (p == cpuinfo.fpproc) 343 savefpstate(p->p_md.md_fpstate); 344 md_core.md_fpstate = *p->p_md.md_fpstate; 345 } else 346 bzero((caddr_t)&md_core.md_fpstate, sizeof(struct fpstate)); 347 348 CORE_SETMAGIC(cseg, CORESEGMAGIC, MID_MACHINE, CORE_CPU); 349 cseg.c_addr = 0; 350 cseg.c_size = chdr->c_cpusize; 351 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&cseg, chdr->c_seghdrsize, 352 (off_t)chdr->c_hdrsize, UIO_SYSSPACE, 353 IO_NODELOCKED|IO_UNIT, cred, NULL, p); 354 if (error) 355 return error; 356 357 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&md_core, sizeof(md_core), 358 (off_t)(chdr->c_hdrsize + chdr->c_seghdrsize), UIO_SYSSPACE, 359 IO_NODELOCKED|IO_UNIT, cred, NULL, p); 360 if (!error) 361 chdr->c_nseg++; 362 363 return error; 364 } 365