1 /* $NetBSD: vm_machdep.c,v 1.94 2008/01/05 22:51:34 martin 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/cdefs.h> 52 __KERNEL_RCSID(0, "$NetBSD: vm_machdep.c,v 1.94 2008/01/05 22:51:34 martin Exp $"); 53 54 #include "opt_multiprocessor.h" 55 #include "opt_coredump.h" 56 57 #include <sys/param.h> 58 #include <sys/systm.h> 59 #include <sys/proc.h> 60 #include <sys/user.h> 61 #include <sys/core.h> 62 #include <sys/malloc.h> 63 #include <sys/buf.h> 64 #include <sys/exec.h> 65 #include <sys/vnode.h> 66 #include <sys/simplelock.h> 67 68 #include <uvm/uvm_extern.h> 69 70 #include <machine/cpu.h> 71 #include <machine/frame.h> 72 #include <machine/trap.h> 73 74 #include <sparc/sparc/cpuvar.h> 75 76 /* 77 * Map a user I/O request into kernel virtual address space. 78 * Note: the pages are already locked by uvm_vslock(), so we 79 * do not need to pass an access_type to pmap_enter(). 80 */ 81 void 82 vmapbuf(struct buf *bp, vsize_t len) 83 { 84 struct pmap *upmap, *kpmap; 85 vaddr_t uva; /* User VA (map from) */ 86 vaddr_t kva; /* Kernel VA (new to) */ 87 paddr_t pa; /* physical address */ 88 vsize_t off; 89 90 if ((bp->b_flags & B_PHYS) == 0) 91 panic("vmapbuf"); 92 93 /* 94 * XXX: It might be better to round/trunc to a 95 * segment boundary to avoid VAC problems! 96 */ 97 bp->b_saveaddr = bp->b_data; 98 uva = trunc_page((vaddr_t)bp->b_data); 99 off = (vaddr_t)bp->b_data - uva; 100 len = round_page(off + len); 101 kva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA); 102 bp->b_data = (void *)(kva + off); 103 104 /* 105 * We have to flush any write-back cache on the 106 * user-space mappings so our new mappings will 107 * have the correct contents. 108 */ 109 if (CACHEINFO.c_vactype != VAC_NONE) 110 cache_flush((void *)uva, len); 111 112 upmap = vm_map_pmap(&bp->b_proc->p_vmspace->vm_map); 113 kpmap = vm_map_pmap(kernel_map); 114 do { 115 if (pmap_extract(upmap, uva, &pa) == false) 116 panic("vmapbuf: null page frame"); 117 /* Now map the page into kernel space. */ 118 pmap_enter(kpmap, kva, pa, 119 VM_PROT_READ|VM_PROT_WRITE, PMAP_WIRED); 120 uva += PAGE_SIZE; 121 kva += PAGE_SIZE; 122 len -= PAGE_SIZE; 123 } while (len); 124 pmap_update(kpmap); 125 } 126 127 /* 128 * Unmap a previously-mapped user I/O request. 129 */ 130 void 131 vunmapbuf(struct buf *bp, vsize_t len) 132 { 133 vaddr_t kva; 134 vsize_t off; 135 136 if ((bp->b_flags & B_PHYS) == 0) 137 panic("vunmapbuf"); 138 139 kva = trunc_page((vaddr_t)bp->b_data); 140 off = (vaddr_t)bp->b_data - kva; 141 len = round_page(off + len); 142 pmap_remove(vm_map_pmap(kernel_map), kva, kva + len); 143 pmap_update(vm_map_pmap(kernel_map)); 144 uvm_km_free(kernel_map, kva, len, UVM_KMF_VAONLY); 145 bp->b_data = bp->b_saveaddr; 146 bp->b_saveaddr = NULL; 147 148 #if 0 /* XXX: The flush above is sufficient, right? */ 149 if (CACHEINFO.c_vactype != VAC_NONE) 150 cpuinfo.cache_flush(bp->b_data, len); 151 #endif 152 } 153 154 155 void 156 cpu_proc_fork(struct proc *p1, struct proc *p2) 157 { 158 159 p2->p_md.md_flags = p1->p_md.md_flags; 160 } 161 162 163 /* 164 * The offset of the topmost frame in the kernel stack. 165 */ 166 #define TOPFRAMEOFF (USPACE-sizeof(struct trapframe)-sizeof(struct frame)) 167 168 /* 169 * Finish a fork operation, with process l2 nearly set up. 170 * Copy and update the pcb and trap frame, making the child ready to run. 171 * 172 * Rig the child's kernel stack so that it will start out in 173 * lwp_trampoline() and call child_return() with l2 as an 174 * argument. This causes the newly-created child process to go 175 * directly to user level with an apparent return value of 0 from 176 * fork(), while the parent process returns normally. 177 * 178 * l1 is the process being forked; if l1 == &lwp0, we are creating 179 * a kernel thread, and the return path and argument are specified with 180 * `func' and `arg'. 181 * 182 * If an alternate user-level stack is requested (with non-zero values 183 * in both the stack and stacksize args), set up the user stack pointer 184 * accordingly. 185 */ 186 void 187 cpu_lwp_fork(struct lwp *l1, struct lwp *l2, 188 void *stack, size_t stacksize, 189 void (*func)(void *), void *arg) 190 { 191 struct pcb *opcb = &l1->l_addr->u_pcb; 192 struct pcb *npcb = &l2->l_addr->u_pcb; 193 struct trapframe *tf2; 194 struct rwindow *rp; 195 196 /* 197 * Save all user registers to l1's stack or, in the case of 198 * user registers and invalid stack pointers, to opcb. 199 * We then copy the whole pcb to l2; when switch() selects l2 200 * to run, it will run at the `lwp_trampoline' stub, rather 201 * than returning at the copying code below. 202 * 203 * If process l1 has an FPU state, we must copy it. If it is 204 * the FPU user, we must save the FPU state first. 205 */ 206 207 if (l1 == curlwp) { 208 write_user_windows(); 209 opcb->pcb_psr = getpsr(); 210 } 211 212 bcopy((void *)opcb, (void *)npcb, sizeof(struct pcb)); 213 if (l1->l_md.md_fpstate != NULL) { 214 struct cpu_info *cpi; 215 int s; 216 217 l2->l_md.md_fpstate = malloc(sizeof(struct fpstate), 218 M_SUBPROC, M_WAITOK); 219 220 FPU_LOCK(s); 221 if ((cpi = l1->l_md.md_fpu) != NULL) { 222 if (cpi->fplwp != l1) 223 panic("FPU(%d): fplwp %p", 224 cpi->ci_cpuid, cpi->fplwp); 225 if (l1 == cpuinfo.fplwp) 226 savefpstate(l1->l_md.md_fpstate); 227 #if defined(MULTIPROCESSOR) 228 else 229 XCALL1(savefpstate, l1->l_md.md_fpstate, 230 1 << cpi->ci_cpuid); 231 #endif 232 } 233 bcopy(l1->l_md.md_fpstate, l2->l_md.md_fpstate, 234 sizeof(struct fpstate)); 235 FPU_UNLOCK(s); 236 } else 237 l2->l_md.md_fpstate = NULL; 238 239 l2->l_md.md_fpu = NULL; 240 241 /* 242 * Setup (kernel) stack frame that will by-pass the child 243 * out of the kernel. (The trap frame invariably resides at 244 * the tippity-top of the u. area.) 245 */ 246 tf2 = l2->l_md.md_tf = (struct trapframe *) 247 ((int)npcb + USPACE - sizeof(*tf2)); 248 249 /* Copy parent's trapframe */ 250 *tf2 = *(struct trapframe *)((int)opcb + USPACE - sizeof(*tf2)); 251 252 /* 253 * If specified, give the child a different stack. 254 */ 255 if (stack != NULL) 256 tf2->tf_out[6] = (u_int)stack + stacksize; 257 258 /* 259 * The fork system call always uses the old system call 260 * convention; clear carry and skip trap instruction as 261 * in syscall(). 262 * note: lwp_trampoline() sets a fresh psr when returning 263 * to user mode. 264 */ 265 /*tf2->tf_psr &= ~PSR_C; -* success */ 266 tf2->tf_pc = tf2->tf_npc; 267 tf2->tf_npc = tf2->tf_pc + 4; 268 269 /* Set return values in child mode */ 270 tf2->tf_out[0] = 0; 271 tf2->tf_out[1] = 1; 272 273 /* Construct kernel frame to return to in cpu_switch() */ 274 rp = (struct rwindow *)((u_int)npcb + TOPFRAMEOFF); 275 rp->rw_local[0] = (int)func; /* Function to call */ 276 rp->rw_local[1] = (int)arg; /* and its argument */ 277 rp->rw_local[2] = (int)l2; /* the new LWP */ 278 279 npcb->pcb_pc = (int)lwp_trampoline - 8; 280 npcb->pcb_sp = (int)rp; 281 npcb->pcb_psr &= ~PSR_CWP; /* Run in window #0 */ 282 npcb->pcb_wim = 1; /* Fence at window #1 */ 283 } 284 285 /* 286 * Cleanup FPU state. 287 */ 288 void 289 cpu_lwp_free(struct lwp *l, int proc) 290 { 291 struct fpstate *fs; 292 293 if ((fs = l->l_md.md_fpstate) != NULL) { 294 struct cpu_info *cpi; 295 int s; 296 297 FPU_LOCK(s); 298 if ((cpi = l->l_md.md_fpu) != NULL) { 299 if (cpi->fplwp != l) 300 panic("FPU(%d): fplwp %p", 301 cpi->ci_cpuid, cpi->fplwp); 302 if (l == cpuinfo.fplwp) 303 savefpstate(fs); 304 #if defined(MULTIPROCESSOR) 305 else 306 XCALL1(savefpstate, fs, 1 << cpi->ci_cpuid); 307 #endif 308 cpi->fplwp = NULL; 309 } 310 l->l_md.md_fpu = NULL; 311 FPU_UNLOCK(s); 312 } 313 } 314 315 void 316 cpu_lwp_free2(struct lwp *l) 317 { 318 struct fpstate *fs; 319 320 if ((fs = l->l_md.md_fpstate) != NULL) 321 free((void *)fs, M_SUBPROC); 322 } 323 324 #ifdef COREDUMP 325 /* 326 * cpu_coredump is called to write a core dump header. 327 * (should this be defined elsewhere? machdep.c?) 328 */ 329 int 330 cpu_coredump(struct lwp *l, void *iocookie, struct core *chdr) 331 { 332 int error; 333 struct md_coredump md_core; 334 struct coreseg cseg; 335 336 if (iocookie == NULL) { 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 chdr->c_nseg++; 342 return 0; 343 } 344 345 md_core.md_tf = *l->l_md.md_tf; 346 if (l->l_md.md_fpstate) { 347 if (l == cpuinfo.fplwp) 348 savefpstate(l->l_md.md_fpstate); 349 md_core.md_fpstate = *l->l_md.md_fpstate; 350 } else 351 bzero((void *)&md_core.md_fpstate, sizeof(struct fpstate)); 352 353 CORE_SETMAGIC(cseg, CORESEGMAGIC, MID_MACHINE, CORE_CPU); 354 cseg.c_addr = 0; 355 cseg.c_size = chdr->c_cpusize; 356 357 error = coredump_write(iocookie, UIO_SYSSPACE, &cseg, 358 chdr->c_seghdrsize); 359 if (error) 360 return error; 361 362 return coredump_write(iocookie, UIO_SYSSPACE, &md_core, 363 sizeof(md_core)); 364 } 365 #endif 366