1 /* $NetBSD: vm_machdep.c,v 1.54 2000/05/28 05:49:04 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 #include <sys/map.h> 61 62 #include <vm/vm.h> 63 #include <vm/vm_kern.h> 64 65 #include <machine/cpu.h> 66 #include <machine/frame.h> 67 #include <machine/trap.h> 68 69 #include <sparc/sparc/cpuvar.h> 70 71 /* 72 * Move pages from one kernel virtual address to another. 73 */ 74 void 75 pagemove(from, to, size) 76 caddr_t from, to; 77 size_t size; 78 { 79 paddr_t pa; 80 81 if (size & PGOFSET || (int)from & PGOFSET || (int)to & PGOFSET) 82 panic("pagemove 1"); 83 while (size > 0) { 84 if (pmap_extract(pmap_kernel(), (vaddr_t)from, &pa) == FALSE) 85 panic("pagemove 2"); 86 pmap_remove(pmap_kernel(), 87 (vaddr_t)from, (vaddr_t)from + PAGE_SIZE); 88 pmap_enter(pmap_kernel(), 89 (vaddr_t)to, pa, VM_PROT_READ|VM_PROT_WRITE, 90 VM_PROT_READ|VM_PROT_WRITE|PMAP_WIRED); 91 from += PAGE_SIZE; 92 to += PAGE_SIZE; 93 size -= PAGE_SIZE; 94 } 95 } 96 97 98 /* 99 * Map a user I/O request into kernel virtual address space. 100 * Note: the pages are already locked by uvm_vslock(), so we 101 * do not need to pass an access_type to pmap_enter(). 102 */ 103 void 104 vmapbuf(bp, len) 105 struct buf *bp; 106 vsize_t len; 107 { 108 struct pmap *upmap, *kpmap; 109 vaddr_t uva; /* User VA (map from) */ 110 vaddr_t kva; /* Kernel VA (new to) */ 111 paddr_t pa; /* physical address */ 112 vsize_t off; 113 114 if ((bp->b_flags & B_PHYS) == 0) 115 panic("vmapbuf"); 116 117 /* 118 * XXX: It might be better to round/trunc to a 119 * segment boundary to avoid VAC problems! 120 */ 121 bp->b_saveaddr = bp->b_data; 122 uva = trunc_page((vaddr_t)bp->b_data); 123 off = (vaddr_t)bp->b_data - uva; 124 len = round_page(off + len); 125 kva = uvm_km_valloc_wait(kernel_map, len); 126 bp->b_data = (caddr_t)(kva + off); 127 128 /* 129 * We have to flush any write-back cache on the 130 * user-space mappings so our new mappings will 131 * have the correct contents. 132 */ 133 if (CACHEINFO.c_vactype != VAC_NONE) 134 cpuinfo.cache_flush((caddr_t)uva, len); 135 136 upmap = vm_map_pmap(&bp->b_proc->p_vmspace->vm_map); 137 kpmap = vm_map_pmap(kernel_map); 138 do { 139 if (pmap_extract(upmap, uva, &pa) == FALSE) 140 panic("vmapbuf: null page frame"); 141 /* Now map the page into kernel space. */ 142 pmap_enter(kpmap, kva, pa, 143 VM_PROT_READ|VM_PROT_WRITE, PMAP_WIRED); 144 uva += PAGE_SIZE; 145 kva += PAGE_SIZE; 146 len -= PAGE_SIZE; 147 } while (len); 148 } 149 150 /* 151 * Unmap a previously-mapped user I/O request. 152 */ 153 void 154 vunmapbuf(bp, len) 155 struct buf *bp; 156 vsize_t len; 157 { 158 vaddr_t kva; 159 vsize_t off; 160 161 if ((bp->b_flags & B_PHYS) == 0) 162 panic("vunmapbuf"); 163 164 kva = trunc_page((vaddr_t)bp->b_data); 165 off = (vaddr_t)bp->b_data - kva; 166 len = round_page(off + len); 167 168 /* This will call pmap_remove() for us. */ 169 uvm_km_free_wakeup(kernel_map, kva, len); 170 bp->b_data = bp->b_saveaddr; 171 bp->b_saveaddr = NULL; 172 173 #if 0 /* XXX: The flush above is sufficient, right? */ 174 if (CACHEINFO.c_vactype != VAC_NONE) 175 cpuinfo.cache_flush(bp->b_data, len); 176 #endif 177 } 178 179 180 /* 181 * The offset of the topmost frame in the kernel stack. 182 */ 183 #define TOPFRAMEOFF (USPACE-sizeof(struct trapframe)-sizeof(struct frame)) 184 185 /* 186 * Finish a fork operation, with process p2 nearly set up. 187 * Copy and update the pcb and trap frame, making the child ready to run. 188 * 189 * Rig the child's kernel stack so that it will start out in 190 * proc_trampoline() and call child_return() with p2 as an 191 * argument. This causes the newly-created child process to go 192 * directly to user level with an apparent return value of 0 from 193 * fork(), while the parent process returns normally. 194 * 195 * p1 is the process being forked; if p1 == &proc0, we are creating 196 * a kernel thread, and the return path and argument are specified with 197 * `func' and `arg'. 198 * 199 * If an alternate user-level stack is requested (with non-zero values 200 * in both the stack and stacksize args), set up the user stack pointer 201 * accordingly. 202 */ 203 void 204 cpu_fork(p1, p2, stack, stacksize, func, arg) 205 struct proc *p1, *p2; 206 void *stack; 207 size_t stacksize; 208 void (*func) __P((void *)); 209 void *arg; 210 { 211 struct pcb *opcb = &p1->p_addr->u_pcb; 212 struct pcb *npcb = &p2->p_addr->u_pcb; 213 struct trapframe *tf2; 214 struct rwindow *rp; 215 216 /* 217 * Save all user registers to p1's stack or, in the case of 218 * user registers and invalid stack pointers, to opcb. 219 * We then copy the whole pcb to p2; when switch() selects p2 220 * to run, it will run at the `proc_trampoline' stub, rather 221 * than returning at the copying code below. 222 * 223 * If process p1 has an FPU state, we must copy it. If it is 224 * the FPU user, we must save the FPU state first. 225 */ 226 227 if (p1 == curproc) { 228 write_user_windows(); 229 opcb->pcb_psr = getpsr(); 230 } 231 #ifdef DIAGNOSTIC 232 else if (p1 != &proc0) 233 panic("cpu_fork: curproc"); 234 #endif 235 236 bcopy((caddr_t)opcb, (caddr_t)npcb, sizeof(struct pcb)); 237 if (p1->p_md.md_fpstate) { 238 if (p1 == cpuinfo.fpproc) 239 savefpstate(p1->p_md.md_fpstate); 240 else if (p1->p_md.md_fpumid != -1) 241 panic("FPU on module %d; fix this", p1->p_md.md_fpumid); 242 p2->p_md.md_fpstate = malloc(sizeof(struct fpstate), 243 M_SUBPROC, M_WAITOK); 244 bcopy(p1->p_md.md_fpstate, p2->p_md.md_fpstate, 245 sizeof(struct fpstate)); 246 } else 247 p2->p_md.md_fpstate = NULL; 248 249 p2->p_md.md_fpumid = -1; 250 251 /* 252 * Setup (kernel) stack frame that will by-pass the child 253 * out of the kernel. (The trap frame invariably resides at 254 * the tippity-top of the u. area.) 255 */ 256 tf2 = p2->p_md.md_tf = (struct trapframe *) 257 ((int)npcb + USPACE - sizeof(*tf2)); 258 259 /* Copy parent's trapframe */ 260 *tf2 = *(struct trapframe *)((int)opcb + USPACE - sizeof(*tf2)); 261 262 /* 263 * If specified, give the child a different stack. 264 */ 265 if (stack != NULL) 266 tf2->tf_out[6] = (u_int)stack + stacksize; 267 268 /* Duplicate efforts of syscall(), but slightly differently */ 269 if (tf2->tf_global[1] & SYSCALL_G2RFLAG) { 270 /* jmp %g2 (or %g7, deprecated) on success */ 271 tf2->tf_npc = tf2->tf_global[2]; 272 } else { 273 /* 274 * old system call convention: clear C on success 275 * note: proc_trampoline() sets a fresh psr when 276 * returning to user mode. 277 */ 278 /*tf2->tf_psr &= ~PSR_C; -* success */ 279 } 280 281 /* Set return values in child mode */ 282 tf2->tf_out[0] = 0; 283 tf2->tf_out[1] = 1; 284 285 /* Construct kernel frame to return to in cpu_switch() */ 286 rp = (struct rwindow *)((u_int)npcb + TOPFRAMEOFF); 287 rp->rw_local[0] = (int)func; /* Function to call */ 288 rp->rw_local[1] = (int)arg; /* and its argument */ 289 290 npcb->pcb_pc = (int)proc_trampoline - 8; 291 npcb->pcb_sp = (int)rp; 292 npcb->pcb_psr &= ~PSR_CWP; /* Run in window #0 */ 293 npcb->pcb_wim = 1; /* Fence at window #1 */ 294 295 } 296 297 /* 298 * cpu_exit is called as the last action during exit. 299 * 300 * We clean up a little and then call switchexit() with the old proc 301 * as an argument. switchexit() switches to the idle context, schedules 302 * the old vmspace and stack to be freed, then selects a new process to 303 * run. 304 */ 305 void 306 cpu_exit(p) 307 struct proc *p; 308 { 309 struct fpstate *fs; 310 311 if ((fs = p->p_md.md_fpstate) != NULL) { 312 if (p == cpuinfo.fpproc) { 313 savefpstate(fs); 314 cpuinfo.fpproc = NULL; 315 } 316 free((void *)fs, M_SUBPROC); 317 } 318 switchexit(p); 319 /* NOTREACHED */ 320 } 321 322 /* 323 * cpu_coredump is called to write a core dump header. 324 * (should this be defined elsewhere? machdep.c?) 325 */ 326 int 327 cpu_coredump(p, vp, cred, chdr) 328 struct proc *p; 329 struct vnode *vp; 330 struct ucred *cred; 331 struct core *chdr; 332 { 333 int error; 334 struct md_coredump md_core; 335 struct coreseg cseg; 336 337 CORE_SETMAGIC(*chdr, COREMAGIC, MID_MACHINE, 0); 338 chdr->c_hdrsize = ALIGN(sizeof(*chdr)); 339 chdr->c_seghdrsize = ALIGN(sizeof(cseg)); 340 chdr->c_cpusize = sizeof(md_core); 341 342 md_core.md_tf = *p->p_md.md_tf; 343 if (p->p_md.md_fpstate) { 344 if (p == cpuinfo.fpproc) 345 savefpstate(p->p_md.md_fpstate); 346 md_core.md_fpstate = *p->p_md.md_fpstate; 347 } else 348 bzero((caddr_t)&md_core.md_fpstate, sizeof(struct fpstate)); 349 350 CORE_SETMAGIC(cseg, CORESEGMAGIC, MID_MACHINE, CORE_CPU); 351 cseg.c_addr = 0; 352 cseg.c_size = chdr->c_cpusize; 353 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&cseg, chdr->c_seghdrsize, 354 (off_t)chdr->c_hdrsize, UIO_SYSSPACE, 355 IO_NODELOCKED|IO_UNIT, cred, NULL, p); 356 if (error) 357 return error; 358 359 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&md_core, sizeof(md_core), 360 (off_t)(chdr->c_hdrsize + chdr->c_seghdrsize), UIO_SYSSPACE, 361 IO_NODELOCKED|IO_UNIT, cred, NULL, p); 362 if (!error) 363 chdr->c_nseg++; 364 365 return error; 366 } 367