sys/platform/vkernel/i386/vm_machdep.c

   1 /*-
   2  * Copyright (c) 1982, 1986 The Regents of the University of California.
   3  * Copyright (c) 1989, 1990 William Jolitz
   4  * Copyright (c) 1994 John Dyson
   5  * All rights reserved.
   6  *
   7  * This code is derived from software contributed to Berkeley by
   8  * the Systems Programming Group of the University of Utah Computer
   9  * Science Department, and William Jolitz.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 3. All advertising materials mentioning features or use of this software
  20  *    must display the following acknowledgement:
  21  *      This product includes software developed by the University of
  22  *      California, Berkeley and its contributors.
  23  * 4. Neither the name of the University nor the names of its contributors
  24  *    may be used to endorse or promote products derived from this software
  25  *    without specific prior written permission.
  26  *
  27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  37  * SUCH DAMAGE.
  38  *
  39  *      from: @(#)vm_machdep.c  7.3 (Berkeley) 5/13/91
  40  *      Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
  41  * $FreeBSD: src/sys/i386/i386/vm_machdep.c,v 1.132.2.9 2003/01/25 19:02:23 dillon Exp $
  42  * $DragonFly: src/sys/platform/vkernel/i386/vm_machdep.c,v 1.10 2008/08/02 05:22:21 dillon Exp $
  43  */
  44
  45 #include "use_npx.h"
  46 #include "use_isa.h"
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/malloc.h>
  51 #include <sys/proc.h>
  52 #include <sys/buf.h>
  53 #include <sys/interrupt.h>
  54 #include <sys/vnode.h>
  55 #include <sys/vmmeter.h>
  56 #include <sys/kernel.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/unistd.h>
  59
  60 #include <machine/clock.h>
  61 #include <machine/cpu.h>
  62 #include <machine/md_var.h>
  63 #include <machine/smp.h>
  64 #include <machine/pcb.h>
  65 #include <machine/pcb_ext.h>
  66 #include <machine/vm86.h>
  67 #include <machine/segments.h>
  68 #include <machine/globaldata.h> /* npxthread */
  69
  70 #include <vm/vm.h>
  71 #include <vm/vm_param.h>
  72 #include <sys/lock.h>
  73 #include <vm/vm_kern.h>
  74 #include <vm/vm_page.h>
  75 #include <vm/vm_map.h>
  76 #include <vm/vm_extern.h>
  77
  78 #include <sys/user.h>
  79
  80 #include <sys/thread2.h>
  81 #include <sys/mplock2.h>
  82
  83 #include <bus/isa/isa.h>
  84
  85 #include <stdio.h>
  86 #include <stdlib.h>
  87
  88 char machine[] = MACHINE;
  89 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD,
  90               machine, 0, "Machine class");
  91
  92 char cpu_vendor[] = "DragonFly";        /* XXX */
  93 u_int cpu_id = 0x80000000;              /* XXX */
  94
  95 /*
  96  * Finish a fork operation, with lwp lp2 nearly set up.
  97  * Copy and update the pcb, set up the stack so that the child
  98  * ready to run and return to user mode.
  99  */
 100 void
 101 cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
 102 {
 103         struct pcb *pcb2;
 104
 105         if ((flags & RFPROC) == 0) {
 106                 if ((flags & RFMEM) == 0) {
 107                         /* unshare user LDT */
 108                         struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
 109                         struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
 110                         if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
 111                                 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
 112                                 user_ldt_free(pcb1);
 113                                 pcb1->pcb_ldt = pcb_ldt;
 114                                 set_user_ldt(pcb1);
 115                         }
 116                 }
 117                 return;
 118         }
 119
 120 #if NNPX > 0
 121         /* Ensure that lp1's pcb is up to date. */
 122         if (mdcpu->gd_npxthread == lp1->lwp_thread)
 123                 npxsave(lp1->lwp_thread->td_savefpu);
 124 #endif
 125
 126         /*
 127          * Copy lp1's PCB.  This really only applies to the
 128          * debug registers and FP state, but its faster to just copy the
 129          * whole thing.  Because we only save the PCB at switchout time,
 130          * the register state may not be current.
 131          */
 132         pcb2 = lp2->lwp_thread->td_pcb;
 133         *pcb2 = *lp1->lwp_thread->td_pcb;
 134
 135         /*
 136          * Create a new fresh stack for the new process.
 137          * Copy the trap frame for the return to user mode as if from a
 138          * syscall.  This copies the user mode register values.  The
 139          * 16 byte offset saves space for vm86, and must match
 140          * common_tss.esp0 (kernel stack pointer on entry from user mode)
 141          *
 142          * pcb_esp must allocate an additional call-return pointer below
 143          * the trap frame which will be restored by cpu_restore from
 144          * PCB_EIP, and the thread's td_sp pointer must allocate an
 145          * additonal two worsd below the pcb_esp call-return pointer to
 146          * hold the LWKT restore function pointer and eflags.
 147          *
 148          * The LWKT restore function pointer must be set to cpu_restore,
 149          * which is our standard heavy weight process switch-in function.
 150          * YYY eventually we should shortcut fork_return and fork_trampoline
 151          * to use the LWKT restore function directly so we can get rid of
 152          * all the extra crap we are setting up.
 153          */
 154         lp2->lwp_md.md_regs = (struct trapframe *)((char *)pcb2 - 16) - 1;
 155         bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));
 156
 157         /*
 158          * Set registers for trampoline to user mode.  Leave space for the
 159          * return address on stack.  These are the kernel mode register values.
 160          */
 161         pcb2->pcb_unused01 = 0;
 162         pcb2->pcb_edi = 0;
 163         pcb2->pcb_esi = (int)fork_return;       /* fork_trampoline argument */
 164         pcb2->pcb_ebp = 0;
 165         pcb2->pcb_esp = (int)lp2->lwp_md.md_regs - sizeof(void *);
 166         pcb2->pcb_ebx = (int)lp2;               /* fork_trampoline argument */
 167         pcb2->pcb_eip = (int)fork_trampoline;
 168         lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
 169         *(u_int32_t *)lp2->lwp_thread->td_sp = PSL_USER;
 170         lp2->lwp_thread->td_sp -= sizeof(void *);
 171         *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;
 172
 173         /*
 174          * pcb2->pcb_ldt:       duplicated below, if necessary.
 175          * pcb2->pcb_savefpu:   cloned above.
 176          * pcb2->pcb_flags:     cloned above (always 0 here?).
 177          * pcb2->pcb_onfault:   cloned above (always NULL here?).
 178          */
 179
 180         /*
 181          * XXX don't copy the i/o pages.  this should probably be fixed.
 182          */
 183         pcb2->pcb_ext = 0;
 184
 185         /* Copy the LDT, if necessary. */
 186         if (pcb2->pcb_ldt != 0) {
 187                 if (flags & RFMEM) {
 188                         pcb2->pcb_ldt->ldt_refcnt++;
 189                 } else {
 190                         pcb2->pcb_ldt = user_ldt_alloc(pcb2,
 191                                 pcb2->pcb_ldt->ldt_len);
 192                 }
 193         }
 194         bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
 195               sizeof(lp2->lwp_thread->td_tls));
 196         /*
 197          * Now, cpu_switch() can schedule the new process.
 198          * pcb_esp is loaded pointing to the cpu_switch() stack frame
 199          * containing the return address when exiting cpu_switch.
 200          * This will normally be to fork_trampoline(), which will have
 201          * %ebx loaded with the new proc's pointer.  fork_trampoline()
 202          * will set up a stack to call fork_return(p, frame); to complete
 203          * the return to user-mode.
 204          */
 205 }
 206
 207 /*
 208  * Prepare new lwp to return to the address specified in params.
 209  */
 210 int
 211 cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
 212 {
 213         struct trapframe *regs = lp->lwp_md.md_regs;
 214         void *bad_return = NULL;
 215         int error;
 216
 217         regs->tf_eip = (int)params->func;
 218         regs->tf_esp = (int)params->stack;
 219         /* Set up argument for function call */
 220         regs->tf_esp -= sizeof(params->arg);
 221         error = copyout(&params->arg, (void *)regs->tf_esp,
 222                         sizeof(params->arg));
 223         if (error)
 224                 return (error);
 225         /*
 226          * Set up fake return address.  As the lwp function may never return,
 227          * we simply copy out a NULL pointer and force the lwp to receive
 228          * a SIGSEGV if it returns anyways.
 229          */
 230         regs->tf_esp -= sizeof(void *);
 231         error = copyout(&bad_return, (void *)regs->tf_esp,
 232                         sizeof(bad_return));
 233         if (error)
 234                 return (error);
 235
 236         cpu_set_fork_handler(lp,
 237             (void (*)(void *, struct trapframe *))generic_lwp_return, lp);
 238         return (0);
 239 }
 240
 241 /*
 242  * Intercept the return address from a freshly forked process that has NOT
 243  * been scheduled yet.
 244  *
 245  * This is needed to make kernel threads stay in kernel mode.
 246  */
 247 void
 248 cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *),
 249                      void *arg)
 250 {
 251         /*
 252          * Note that the trap frame follows the args, so the function
 253          * is really called like this:  func(arg, frame);
 254          */
 255         lp->lwp_thread->td_pcb->pcb_esi = (int) func;   /* function */
 256         lp->lwp_thread->td_pcb->pcb_ebx = (int) arg;    /* first arg */
 257 }
 258
 259 void
 260 cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
 261 {
 262         td->td_pcb->pcb_esi = (int)func;
 263         td->td_pcb->pcb_ebx = (int) arg;
 264         td->td_switch = cpu_lwkt_switch;
 265         td->td_sp -= sizeof(void *);
 266         *(void **)td->td_sp = rfunc;    /* exit function on return */
 267         td->td_sp -= sizeof(void *);
 268         *(void **)td->td_sp = cpu_kthread_restore;
 269 }
 270
 271 void
 272 cpu_lwp_exit(void)
 273 {
 274         struct thread *td = curthread;
 275         struct pcb *pcb;
 276         struct pcb_ext *ext;
 277
 278 #if NNPX > 0
 279         npxexit();
 280 #endif  /* NNPX */
 281
 282         /*
 283          * If we were using a private TSS do a forced-switch to ourselves
 284          * to switch back to the common TSS before freeing it.
 285          */
 286         pcb = td->td_pcb;
 287         if ((ext = pcb->pcb_ext) != NULL) {
 288                 crit_enter();
 289                 pcb->pcb_ext = NULL;
 290                 td->td_switch(td);
 291                 crit_exit();
 292                 kmem_free(&kernel_map, (vm_offset_t)ext, ctob(IOPAGES + 1));
 293         }
 294         user_ldt_free(pcb);
 295         if (pcb->pcb_flags & PCB_DBREGS) {
 296                 /*
 297                  * disable all hardware breakpoints
 298                  */
 299                 reset_dbregs();
 300                 pcb->pcb_flags &= ~PCB_DBREGS;
 301         }
 302         td->td_gd->gd_cnt.v_swtch++;
 303
 304         crit_enter_quick(td);
 305         if (td->td_flags & TDF_TSLEEPQ)
 306                 tsleep_remove(td);
 307         lwkt_deschedule_self(td);
 308         lwkt_remove_tdallq(td);
 309         cpu_thread_exit();
 310 }
 311
 312 /*
 313  * Terminate the current thread.  The caller must have already acquired
 314  * the thread's rwlock and placed it on a reap list or otherwise notified
 315  * a reaper of its existance.  We set a special assembly switch function which
 316  * releases td_rwlock after it has cleaned up the MMU state and switched
 317  * out the stack.
 318  *
 319  * Must be caller from a critical section and with the thread descheduled.
 320  */
 321 void
 322 cpu_thread_exit(void)
 323 {
 324         curthread->td_switch = cpu_exit_switch;
 325         curthread->td_flags |= TDF_EXITING;
 326         lwkt_switch();
 327         panic("cpu_exit");
 328 }
 329
 330 /*
 331  * Process Reaper.  Called after the caller has acquired the thread's
 332  * rwlock and removed it from the reap list.
 333  */
 334 void
 335 cpu_proc_wait(struct proc *p)
 336 {
 337         /* drop per-process resources */
 338         pmap_dispose_proc(p);
 339 }
 340
 341 #ifdef notyet
 342 static void
 343 setredzone(u_short *pte, caddr_t vaddr)
 344 {
 345 /* eventually do this by setting up an expand-down stack segment
 346    for ss0: selector, allowing stack access down to top of u.
 347    this means though that protection violations need to be handled
 348    thru a double fault exception that must do an integral task
 349    switch to a known good context, within which a dump can be
 350    taken. a sensible scheme might be to save the initial context
 351    used by sched (that has physical memory mapped 1:1 at bottom)
 352    and take the dump while still in mapped mode */
 353 }
 354 #endif
 355
 356 /*
 357  * Convert kernel VA to physical address
 358  */
 359 vm_paddr_t
 360 kvtop(void *addr)
 361 {
 362         vm_paddr_t pa;
 363
 364         pa = pmap_kextract((vm_offset_t)addr);
 365         if (pa == 0)
 366                 panic("kvtop: zero page frame");
 367         return (pa);
 368 }
 369
 370 SYSCTL_DECL(_vm_stats_misc);
 371
 372 /*
 373  * Tell whether this address is in some physical memory region.
 374  * Currently used by the kernel coredump code in order to avoid
 375  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
 376  * or other unpredictable behaviour.
 377  */
 378
 379 int
 380 is_physical_memory(vm_offset_t addr)
 381 {
 382         return 1;
 383 }
 384
 385 /*
 386  * Used by /dev/kmem to determine if we can safely read or write
 387  * the requested KVA range.  Some portions of kernel memory are
 388  * not governed by our virtual page table.
 389  */
 390 extern int32_t _end;
 391 extern void _start(void);
 392
 393 int
 394 kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
 395 {
 396         vm_offset_t addr;
 397
 398         if (saddr >= trunc_page((vm_offset_t)&_start) && eaddr <= round_page((vm_offset_t)&_end))
 399                 return 0;
 400         if (saddr < KvaStart)
 401                 return EFAULT;
 402         if (eaddr >= KvaEnd)
 403                 return EFAULT;
 404         for (addr = saddr; addr < eaddr; addr += PAGE_SIZE)  {
 405                 if (pmap_extract(&kernel_pmap, addr) == 0)
 406                         return EFAULT;
 407         }
 408         if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
 409                 return EFAULT;
 410         return 0;
 411 }
 412