sys/platform/vkernel64/x86_64/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  * Copyright (c) 2008 The DragonFly Project.
   6  * All rights reserved.
   7  *
   8  * This code is derived from software contributed to Berkeley by
   9  * the Systems Programming Group of the University of Utah Computer
  10  * Science Department, and William Jolitz.
  11  *
  12  * Redistribution and use in source and binary forms, with or without
  13  * modification, are permitted provided that the following conditions
  14  * are met:
  15  * 1. Redistributions of source code must retain the above copyright
  16  *    notice, this list of conditions and the following disclaimer.
  17  * 2. Redistributions in binary form must reproduce the above copyright
  18  *    notice, this list of conditions and the following disclaimer in the
  19  *    documentation and/or other materials provided with the distribution.
  20  * 3. All advertising materials mentioning features or use of this software
  21  *    must display the following acknowledgement:
  22  *      This product includes software developed by the University of
  23  *      California, Berkeley and its contributors.
  24  * 4. Neither the name of the University nor the names of its contributors
  25  *    may be used to endorse or promote products derived from this software
  26  *    without specific prior written permission.
  27  *
  28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  38  * SUCH DAMAGE.
  39  *
  40  *      from: @(#)vm_machdep.c  7.3 (Berkeley) 5/13/91
  41  *      Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
  42  * $FreeBSD: src/sys/i386/i386/vm_machdep.c,v 1.132.2.9 2003/01/25 19:02:23 dillon Exp $
  43  */
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/malloc.h>
  48 #include <sys/proc.h>
  49 #include <sys/buf.h>
  50 #include <sys/interrupt.h>
  51 #include <sys/vnode.h>
  52 #include <sys/vmmeter.h>
  53 #include <sys/kernel.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/unistd.h>
  56 #include <sys/lwp.h>
  57
  58 #include <machine/clock.h>
  59 #include <machine/cpu.h>
  60 #include <machine/md_var.h>
  61 #include <machine/smp.h>
  62 #include <machine/pcb.h>
  63 #include <machine/pcb_ext.h>
  64 #include <machine/segments.h>
  65 #include <machine/globaldata.h> /* npxthread */
  66
  67 #include <vm/vm.h>
  68 #include <vm/vm_param.h>
  69 #include <sys/lock.h>
  70 #include <vm/vm_kern.h>
  71 #include <vm/vm_page.h>
  72 #include <vm/vm_map.h>
  73 #include <vm/vm_extern.h>
  74
  75 #include <sys/thread2.h>
  76
  77 #include <bus/isa/isa.h>
  78
  79 char machine[] = MACHINE;
  80 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD,
  81               machine, 0, "Machine class");
  82
  83 u_int cpu_vendor_id = 0;                /* XXX */
  84
  85 /*
  86  * Finish a fork operation, with lwp lp2 nearly set up.
  87  * Copy and update the pcb, set up the stack so that the child
  88  * ready to run and return to user mode.
  89  */
  90 void
  91 cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
  92 {
  93         struct pcb *pcb2;
  94
  95         if ((flags & RFPROC) == 0) {
  96                 if ((flags & RFMEM) == 0) {
  97                         /* unshare user LDT */
  98                         struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
  99                         struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
 100                         if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
 101                                 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
 102                                 user_ldt_free(pcb1);
 103                                 pcb1->pcb_ldt = pcb_ldt;
 104                                 set_user_ldt(pcb1);
 105                         }
 106                 }
 107                 return;
 108         }
 109
 110         /* Ensure that lp1's pcb is up to date. */
 111         if (mdcpu->gd_npxthread == lp1->lwp_thread)
 112                 npxsave(lp1->lwp_thread->td_savefpu);
 113
 114         /*
 115          * Copy lp1's PCB.  This really only applies to the
 116          * debug registers and FP state, but its faster to just copy the
 117          * whole thing.  Because we only save the PCB at switchout time,
 118          * the register state may not be current.
 119          */
 120         pcb2 = lp2->lwp_thread->td_pcb;
 121         *pcb2 = *lp1->lwp_thread->td_pcb;
 122
 123         /*
 124          * Create a new fresh stack for the new process.
 125          * Copy the trap frame for the return to user mode as if from a
 126          * syscall.  This copies the user mode register values.
 127          *
 128          * pcb_rsp must allocate an additional call-return pointer below
 129          * the trap frame which will be restored by cpu_heavy_restore from
 130          * PCB_RIP, and the thread's td_sp pointer must allocate an
 131          * additonal two quadwords below the pcb_rsp call-return pointer to
 132          * hold the LWKT restore function pointer and rflags.
 133          *
 134          * The LWKT restore function pointer must be set to cpu_heavy_restore,
 135          * which is our standard heavy-weight process switch-in function.
 136          * YYY eventually we should shortcut fork_return and fork_trampoline
 137          * to use the LWKT restore function directly so we can get rid of
 138          * all the extra crap we are setting up.
 139          */
 140         lp2->lwp_md.md_regs = (struct trapframe *)pcb2 - 1;
 141         bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));
 142
 143         /*
 144          * Set registers for trampoline to user mode.  Leave space for the
 145          * return address on stack.  These are the kernel mode register values.
 146          */
 147         pcb2->pcb_unused01 = 0;
 148         pcb2->pcb_rbx = (unsigned long)fork_return;     /* fork_trampoline argument */
 149         pcb2->pcb_rbp = 0;
 150         pcb2->pcb_rsp = (unsigned long)lp2->lwp_md.md_regs - sizeof(void *);
 151         pcb2->pcb_r12 = (unsigned long)lp2;             /* fork_trampoline argument */
 152         pcb2->pcb_r13 = 0;
 153         pcb2->pcb_r14 = 0;
 154         pcb2->pcb_r15 = 0;
 155         pcb2->pcb_rip = (unsigned long)fork_trampoline;
 156         lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_rsp - sizeof(void *));
 157         *(u_int64_t *)lp2->lwp_thread->td_sp = PSL_USER;
 158         lp2->lwp_thread->td_sp -= sizeof(void *);
 159         *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;
 160
 161         /*
 162          * pcb2->pcb_ldt:       duplicated below, if necessary.
 163          * pcb2->pcb_savefpu:   cloned above.
 164          * pcb2->pcb_flags:     cloned above (always 0 here?).
 165          * pcb2->pcb_onfault:   cloned above (always NULL here?).
 166          */
 167
 168         /*
 169          * XXX don't copy the i/o pages.  this should probably be fixed.
 170          */
 171         pcb2->pcb_ext = NULL;
 172
 173         /* Copy the LDT, if necessary. */
 174         if (pcb2->pcb_ldt != NULL) {
 175                 if (flags & RFMEM) {
 176                         pcb2->pcb_ldt->ldt_refcnt++;
 177                 } else {
 178                         pcb2->pcb_ldt = user_ldt_alloc(pcb2,
 179                                 pcb2->pcb_ldt->ldt_len);
 180                 }
 181         }
 182         bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
 183               sizeof(lp2->lwp_thread->td_tls));
 184         /*
 185          * Now, cpu_switch() can schedule the new lwp.
 186          * pcb_rsp is loaded pointing to the cpu_switch() stack frame
 187          * containing the return address when exiting cpu_switch.
 188          * This will normally be to fork_trampoline(), which will have
 189          * %rbx loaded with the new lwp's pointer.  fork_trampoline()
 190          * will set up a stack to call fork_return(lp, frame); to complete
 191          * the return to user-mode.
 192          */
 193 }
 194
 195 /*
 196  * Prepare new lwp to return to the address specified in params.
 197  */
 198 int
 199 cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
 200 {
 201         struct trapframe *regs = lp->lwp_md.md_regs;
 202         void *bad_return = NULL;
 203         int error;
 204
 205         regs->tf_rip = (long)params->lwp_func;
 206         regs->tf_rsp = (long)params->lwp_stack;
 207         /* Set up argument for function call */
 208         regs->tf_rdi = (long)params->lwp_arg; /* JG Can this be in userspace addresses? */
 209         /*
 210          * Set up fake return address.  As the lwp function may never return,
 211          * we simply copy out a NULL pointer and force the lwp to receive
 212          * a SIGSEGV if it returns anyways.
 213          */
 214         regs->tf_rsp -= sizeof(void *);
 215         error = copyout(&bad_return, (void *)regs->tf_rsp, sizeof(bad_return));
 216         if (error)
 217                 return (error);
 218
 219         cpu_set_fork_handler(lp,
 220             (void (*)(void *, struct trapframe *))generic_lwp_return, lp);
 221         return (0);
 222 }
 223
 224 /*
 225  * Intercept the return address from a freshly forked process that has NOT
 226  * been scheduled yet.
 227  *
 228  * This is needed to make kernel threads stay in kernel mode.
 229  */
 230 void
 231 cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *),
 232                      void *arg)
 233 {
 234         /*
 235          * Note that the trap frame follows the args, so the function
 236          * is really called like this:  func(arg, frame);
 237          */
 238         lp->lwp_thread->td_pcb->pcb_rbx = (long)func;   /* function */
 239         lp->lwp_thread->td_pcb->pcb_r12 = (long)arg;    /* first arg */
 240 }
 241
 242 void
 243 cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
 244 {
 245         td->td_pcb->pcb_rbx = (long)func;
 246         td->td_pcb->pcb_r12 = (long)arg;
 247         td->td_switch = cpu_lwkt_switch;
 248         td->td_sp -= sizeof(void *);
 249         *(void **)td->td_sp = rfunc;    /* exit function on return */
 250         td->td_sp -= sizeof(void *);
 251         *(void **)td->td_sp = cpu_kthread_restore;
 252 }
 253
 254 void
 255 cpu_lwp_exit(void)
 256 {
 257         struct thread *td = curthread;
 258         struct pcb *pcb;
 259
 260         pcb = td->td_pcb;
 261
 262         /* Some i386 functionality was dropped */
 263         KKASSERT(pcb->pcb_ext == NULL);
 264
 265         /*
 266          * disable all hardware breakpoints
 267          */
 268         if (pcb->pcb_flags & PCB_DBREGS) {
 269                 reset_dbregs();
 270                 pcb->pcb_flags &= ~PCB_DBREGS;
 271         }
 272         td->td_gd->gd_cnt.v_swtch++;
 273
 274         crit_enter_quick(td);
 275         if (td->td_flags & TDF_TSLEEPQ)
 276                 tsleep_remove(td);
 277         lwkt_deschedule_self(td);
 278         lwkt_remove_tdallq(td);
 279         cpu_thread_exit();
 280 }
 281
 282 /*
 283  * Terminate the current thread.  The caller must have already acquired
 284  * the thread's rwlock and placed it on a reap list or otherwise notified
 285  * a reaper of its existance.  We set a special assembly switch function which
 286  * releases td_rwlock after it has cleaned up the MMU state and switched
 287  * out the stack.
 288  *
 289  * Must be caller from a critical section and with the thread descheduled.
 290  */
 291 void
 292 cpu_thread_exit(void)
 293 {
 294         npxexit();
 295         curthread->td_switch = cpu_exit_switch;
 296         curthread->td_flags |= TDF_EXITING;
 297         lwkt_switch();
 298         panic("cpu_thread_exit: lwkt_switch() unexpectedly returned");
 299 }
 300
 301 /*
 302  * Used by /dev/kmem to determine if we can safely read or write
 303  * the requested KVA range.  Some portions of kernel memory are
 304  * not governed by our virtual page table.
 305  */
 306 extern int64_t _end;
 307 extern void _start(void);
 308
 309 int
 310 kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
 311 {
 312         vm_offset_t addr;
 313
 314         if (saddr >= trunc_page((vm_offset_t)&_start) &&
 315             eaddr <= round_page((vm_offset_t)&_end)) {
 316                 return 0;
 317         }
 318         if (saddr < KvaStart)
 319                 return EFAULT;
 320         if (eaddr >= KvaEnd)
 321                 return EFAULT;
 322         for (addr = saddr; addr < eaddr; addr += PAGE_SIZE)  {
 323                 if (pmap_kextract(addr) == 0)
 324                         return EFAULT;
 325         }
 326         if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
 327                 return EFAULT;
 328         return 0;
 329 }