1 /*- 2 * Copyright (c) 1982, 1986, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * the Systems Programming Group of the University of Utah Computer 7 * Science Department, and William Jolitz. 8 * 9 * %sccs.include.redist.c% 10 * 11 * @(#)vm_machdep.c 8.3 (Berkeley) 01/21/94 12 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 13 */ 14 15 #include <sys/param.h> 16 #include <sys/systm.h> 17 #include <sys/proc.h> 18 #include <sys/malloc.h> 19 #include <sys/buf.h> 20 #include <sys/vnode.h> 21 #include <sys/user.h> 22 23 #include <machine/cpu.h> 24 25 #include <vm/vm.h> 26 #include <vm/vm_kern.h> 27 28 /* 29 * Finish a fork operation, with process p2 nearly set up. 30 * Copy and update the kernel stack and pcb, making the child 31 * ready to run, and marking it so that it can return differently 32 * than the parent. Returns 1 in the child process, 0 in the parent. 33 * We currently double-map the user area so that the stack is at the same 34 * address in each process; in the future we will probably relocate 35 * the frame pointers on the stack after copying. 36 */ 37 cpu_fork(p1, p2) 38 register struct proc *p1, *p2; 39 { 40 register struct user *up = p2->p_addr; 41 int foo, offset, addr, i; 42 extern char kstack[]; 43 extern int mvesp(); 44 45 /* 46 * Copy pcb and stack from proc p1 to p2. 47 * We do this as cheaply as possible, copying only the active 48 * part of the stack. The stack and pcb need to agree; 49 * this is tricky, as the final pcb is constructed by savectx, 50 * but its frame isn't yet on the stack when the stack is copied. 51 * mi_switch compensates for this when the child eventually runs. 52 * This should be done differently, with a single call 53 * that copies and updates the pcb+stack, 54 * replacing the bcopy and savectx. 55 */ 56 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 57 offset = mvesp() - (int)kstack; 58 bcopy((caddr_t)kstack + offset, (caddr_t)p2->p_addr + offset, 59 (unsigned) ctob(UPAGES) - offset); 60 p2->p_md.md_regs = p1->p_md.md_regs; 61 62 /* 63 * Wire top of address space of child to it's u. 64 * First, fault in a page of pte's to map it. 65 */ 66 addr = trunc_page((u_int)vtopte(kstack)); 67 (void)vm_map_pageable(&p2->p_vmspace->vm_map, addr, addr+NBPG, FALSE); 68 for (i=0; i < UPAGES; i++) 69 pmap_enter(&p2->p_vmspace->vm_pmap, (vm_offset_t)kstack+i*NBPG, 70 pmap_extract(kernel_pmap, ((int)p2->p_addr)+i*NBPG), 71 VM_PROT_READ, 1); 72 pmap_activate(&p2->p_vmspace->vm_pmap, &up->u_pcb); 73 74 /* 75 * 76 * Arrange for a non-local goto when the new process 77 * is started, to resume here, returning nonzero from setjmp. 78 */ 79 if (savectx(up, 1)) { 80 /* 81 * Return 1 in child. 82 */ 83 return (1); 84 } 85 return (0); 86 } 87 88 #include "npx.h" 89 #if NNPX > 0 90 extern struct proc *npxproc; 91 #endif 92 93 #ifdef notyet 94 /* 95 * cpu_exit is called as the last action during exit. 96 * 97 * We change to an inactive address space and a "safe" stack, 98 * passing thru an argument to the new stack. Now, safely isolated 99 * from the resources we're shedding, we release the address space 100 * and any remaining machine-dependent resources, including the 101 * memory for the user structure and kernel stack. 102 * 103 * Next, we assign a dummy context to be written over by mi_switch, 104 * calling it to send this process off to oblivion. 105 * [The nullpcb allows us to minimize cost in mi_switch() by not having 106 * a special case]. 107 */ 108 struct proc *switch_to_inactive(); 109 cpu_exit(p) 110 register struct proc *p; 111 { 112 static struct pcb nullpcb; /* pcb to overwrite on last switch */ 113 114 #if NNPX > 0 115 /* free cporcessor (if we have it) */ 116 if( p == npxproc) npxproc =0; 117 #endif 118 119 /* move to inactive space and stack, passing arg accross */ 120 p = switch_to_inactive(p); 121 122 /* drop per-process resources */ 123 vmspace_free(p->p_vmspace); 124 kmem_free(kernel_map, (vm_offset_t)p->p_addr, ctob(UPAGES)); 125 126 p->p_addr = (struct user *) &nullpcb; 127 mi_switch(); 128 /* NOTREACHED */ 129 } 130 #else 131 cpu_exit(p) 132 register struct proc *p; 133 { 134 135 /* free coprocessor (if we have it) */ 136 #if NNPX > 0 137 if( p == npxproc) npxproc =0; 138 #endif 139 140 curproc = p; 141 mi_switch(); 142 } 143 144 cpu_wait(p) struct proc *p; { 145 146 /* drop per-process resources */ 147 vmspace_free(p->p_vmspace); 148 kmem_free(kernel_map, (vm_offset_t)p->p_addr, ctob(UPAGES)); 149 } 150 #endif 151 152 /* 153 * Dump the machine specific header information at the start of a core dump. 154 */ 155 cpu_coredump(p, vp, cred) 156 struct proc *p; 157 struct vnode *vp; 158 struct ucred *cred; 159 { 160 161 return (vn_rdwr(UIO_WRITE, vp, (caddr_t) p->p_addr, ctob(UPAGES), 162 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, 163 p)); 164 } 165 166 /* 167 * Set a red zone in the kernel stack after the u. area. 168 */ 169 setredzone(pte, vaddr) 170 u_short *pte; 171 caddr_t vaddr; 172 { 173 /* eventually do this by setting up an expand-down stack segment 174 for ss0: selector, allowing stack access down to top of u. 175 this means though that protection violations need to be handled 176 thru a double fault exception that must do an integral task 177 switch to a known good context, within which a dump can be 178 taken. a sensible scheme might be to save the initial context 179 used by sched (that has physical memory mapped 1:1 at bottom) 180 and take the dump while still in mapped mode */ 181 } 182 183 /* 184 * Move pages from one kernel virtual address to another. 185 * Both addresses are assumed to reside in the Sysmap, 186 * and size must be a multiple of CLSIZE. 187 */ 188 pagemove(from, to, size) 189 register caddr_t from, to; 190 int size; 191 { 192 register struct pte *fpte, *tpte; 193 194 if (size % CLBYTES) 195 panic("pagemove"); 196 fpte = kvtopte(from); 197 tpte = kvtopte(to); 198 while (size > 0) { 199 *tpte++ = *fpte; 200 *(int *)fpte++ = 0; 201 from += NBPG; 202 to += NBPG; 203 size -= NBPG; 204 } 205 tlbflush(); 206 } 207 208 /* 209 * Convert kernel VA to physical address 210 */ 211 kvtop(addr) 212 register caddr_t addr; 213 { 214 vm_offset_t va; 215 216 va = pmap_extract(kernel_pmap, (vm_offset_t)addr); 217 if (va == 0) 218 panic("kvtop: zero page frame"); 219 return((int)va); 220 } 221 222 #ifdef notdef 223 /* 224 * The probe[rw] routines should probably be redone in assembler 225 * for efficiency. 226 */ 227 prober(addr) 228 register u_int addr; 229 { 230 register int page; 231 register struct proc *p; 232 233 if (addr >= USRSTACK) 234 return(0); 235 p = u.u_procp; 236 page = btop(addr); 237 if (page < dptov(p, p->p_dsize) || page > sptov(p, p->p_ssize)) 238 return(1); 239 return(0); 240 } 241 242 probew(addr) 243 register u_int addr; 244 { 245 register int page; 246 register struct proc *p; 247 248 if (addr >= USRSTACK) 249 return(0); 250 p = u.u_procp; 251 page = btop(addr); 252 if (page < dptov(p, p->p_dsize) || page > sptov(p, p->p_ssize)) 253 return((*(int *)vtopte(p, page) & PG_PROT) == PG_UW); 254 return(0); 255 } 256 257 /* 258 * NB: assumes a physically contiguous kernel page table 259 * (makes life a LOT simpler). 260 */ 261 kernacc(addr, count, rw) 262 register u_int addr; 263 int count, rw; 264 { 265 register struct pde *pde; 266 register struct pte *pte; 267 register int ix, cnt; 268 extern long Syssize; 269 270 if (count <= 0) 271 return(0); 272 pde = (struct pde *)((u_int)u.u_procp->p_p0br + u.u_procp->p_szpt * NBPG); 273 ix = (addr & PD_MASK) >> PD_SHIFT; 274 cnt = ((addr + count + (1 << PD_SHIFT) - 1) & PD_MASK) >> PD_SHIFT; 275 cnt -= ix; 276 for (pde += ix; cnt; cnt--, pde++) 277 if (pde->pd_v == 0) 278 return(0); 279 ix = btop(addr-0xfe000000); 280 cnt = btop(addr-0xfe000000+count+NBPG-1); 281 if (cnt > (int)&Syssize) 282 return(0); 283 cnt -= ix; 284 for (pte = &Sysmap[ix]; cnt; cnt--, pte++) 285 if (pte->pg_v == 0 /*|| (rw == B_WRITE && pte->pg_prot == 1)*/) 286 return(0); 287 return(1); 288 } 289 290 useracc(addr, count, rw) 291 register u_int addr; 292 int count, rw; 293 { 294 register int (*func)(); 295 register u_int addr2; 296 extern int prober(), probew(); 297 298 if (count <= 0) 299 return(0); 300 addr2 = addr; 301 addr += count; 302 func = (rw == B_READ) ? prober : probew; 303 do { 304 if ((*func)(addr2) == 0) 305 return(0); 306 addr2 = (addr2 + NBPG) & ~PGOFSET; 307 } while (addr2 < addr); 308 return(1); 309 } 310 #endif 311 312 extern vm_map_t phys_map; 313 314 /* 315 * Map an IO request into kernel virtual address space. Requests fall into 316 * one of five catagories: 317 * 318 * B_PHYS|B_UAREA: User u-area swap. 319 * Address is relative to start of u-area (p_addr). 320 * B_PHYS|B_PAGET: User page table swap. 321 * Address is a kernel VA in usrpt (Usrptmap). 322 * B_PHYS|B_DIRTY: Dirty page push. 323 * Address is a VA in proc2's address space. 324 * B_PHYS|B_PGIN: Kernel pagein of user pages. 325 * Address is VA in user's address space. 326 * B_PHYS: User "raw" IO request. 327 * Address is VA in user's address space. 328 * 329 * All requests are (re)mapped into kernel VA space via the useriomap 330 * (a name with only slightly more meaning than "kernelmap") 331 */ 332 vmapbuf(bp) 333 register struct buf *bp; 334 { 335 register int npf; 336 register caddr_t addr; 337 register long flags = bp->b_flags; 338 struct proc *p; 339 int off; 340 vm_offset_t kva; 341 register vm_offset_t pa; 342 343 if ((flags & B_PHYS) == 0) 344 panic("vmapbuf"); 345 addr = bp->b_saveaddr = bp->b_un.b_addr; 346 off = (int)addr & PGOFSET; 347 p = bp->b_proc; 348 npf = btoc(round_page(bp->b_bcount + off)); 349 kva = kmem_alloc_wait(phys_map, ctob(npf)); 350 bp->b_un.b_addr = (caddr_t) (kva + off); 351 while (npf--) { 352 pa = pmap_extract(&p->p_vmspace->vm_pmap, (vm_offset_t)addr); 353 if (pa == 0) 354 panic("vmapbuf: null page frame"); 355 pmap_enter(vm_map_pmap(phys_map), kva, trunc_page(pa), 356 VM_PROT_READ|VM_PROT_WRITE, TRUE); 357 addr += PAGE_SIZE; 358 kva += PAGE_SIZE; 359 } 360 } 361 362 /* 363 * Free the io map PTEs associated with this IO operation. 364 * We also invalidate the TLB entries and restore the original b_addr. 365 */ 366 vunmapbuf(bp) 367 register struct buf *bp; 368 { 369 register int npf; 370 register caddr_t addr = bp->b_un.b_addr; 371 vm_offset_t kva; 372 373 if ((bp->b_flags & B_PHYS) == 0) 374 panic("vunmapbuf"); 375 npf = btoc(round_page(bp->b_bcount + ((int)addr & PGOFSET))); 376 kva = (vm_offset_t)((int)addr & ~PGOFSET); 377 kmem_free_wakeup(phys_map, kva, ctob(npf)); 378 bp->b_un.b_addr = bp->b_saveaddr; 379 bp->b_saveaddr = NULL; 380 } 381