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  * $DragonFly: src/sys/platform/pc64/amd64/vm_machdep.c,v 1.3 2008/08/29 17:07:10 dillon Exp $
  44  */
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/malloc.h>
  49 #include <sys/proc.h>
  50 #include <sys/buf.h>
  51 #include <sys/interrupt.h>
  52 #include <sys/vnode.h>
  53 #include <sys/vmmeter.h>
  54 #include <sys/kernel.h>
  55 #include <sys/sysctl.h>
  56 #include <sys/unistd.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 /*
  84  * Finish a fork operation, with lwp lp2 nearly set up.
  85  * Copy and update the pcb, set up the stack so that the child
  86  * ready to run and return to user mode.
  87  */
  88 void
  89 cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
  90 {
  91         struct pcb *pcb2;
  92
  93         if ((flags & RFPROC) == 0) {
  94                 if ((flags & RFMEM) == 0) {
  95                         /* unshare user LDT */
  96                         struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
  97                         struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
  98                         if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
  99                                 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
 100                                 user_ldt_free(pcb1);
 101                                 pcb1->pcb_ldt = pcb_ldt;
 102                                 set_user_ldt(pcb1);
 103                         }
 104                 }
 105                 return;
 106         }
 107
 108         /* Ensure that lp1's pcb is up to date. */
 109         if (mdcpu->gd_npxthread == lp1->lwp_thread)
 110                 npxsave(lp1->lwp_thread->td_savefpu);
 111
 112         /*
 113          * Copy lp1's PCB.  This really only applies to the
 114          * debug registers and FP state, but its faster to just copy the
 115          * whole thing.  Because we only save the PCB at switchout time,
 116          * the register state may not be current.
 117          */
 118         pcb2 = lp2->lwp_thread->td_pcb;
 119         *pcb2 = *lp1->lwp_thread->td_pcb;
 120
 121         /*
 122          * Create a new fresh stack for the new process.
 123          * Copy the trap frame for the return to user mode as if from a
 124          * syscall.  This copies the user mode register values.
 125          *
 126          * pcb_rsp must allocate an additional call-return pointer below
 127          * the trap frame which will be restored by cpu_heavy_restore from
 128          * PCB_RIP, and the thread's td_sp pointer must allocate an
 129          * additonal two quadwords below the pcb_rsp call-return pointer to
 130          * hold the LWKT restore function pointer and rflags.
 131          *
 132          * The LWKT restore function pointer must be set to cpu_heavy_restore,
 133          * which is our standard heavy-weight process switch-in function.
 134          * YYY eventually we should shortcut fork_return and fork_trampoline
 135          * to use the LWKT restore function directly so we can get rid of
 136          * all the extra crap we are setting up.
 137          */
 138         lp2->lwp_md.md_regs = (struct trapframe *)pcb2 - 1;
 139         bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));
 140
 141         /*
 142          * Set registers for trampoline to user mode.  Leave space for the
 143          * return address on stack.  These are the kernel mode register values.
 144          */
 145         pcb2->pcb_unused01 = 0;
 146         pcb2->pcb_rbx = (unsigned long)fork_return;     /* fork_trampoline argument */
 147         pcb2->pcb_rbp = 0;
 148         pcb2->pcb_rsp = (unsigned long)lp2->lwp_md.md_regs - sizeof(void *);
 149         pcb2->pcb_r12 = (unsigned long)lp2;             /* fork_trampoline argument */
 150         pcb2->pcb_r13 = 0;
 151         pcb2->pcb_r14 = 0;
 152         pcb2->pcb_r15 = 0;
 153         pcb2->pcb_rip = (unsigned long)fork_trampoline;
 154         lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_rsp - sizeof(void *));
 155         *(u_int64_t *)lp2->lwp_thread->td_sp = PSL_USER;
 156         lp2->lwp_thread->td_sp -= sizeof(void *);
 157         *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;
 158
 159         /*
 160          * pcb2->pcb_ldt:       duplicated below, if necessary.
 161          * pcb2->pcb_savefpu:   cloned above.
 162          * pcb2->pcb_flags:     cloned above (always 0 here?).
 163          * pcb2->pcb_onfault:   cloned above (always NULL here?).
 164          */
 165
 166         /*
 167          * XXX don't copy the i/o pages.  this should probably be fixed.
 168          */
 169         pcb2->pcb_ext = 0;
 170
 171         /* Copy the LDT, if necessary. */
 172         if (pcb2->pcb_ldt != 0) {
 173                 if (flags & RFMEM) {
 174                         pcb2->pcb_ldt->ldt_refcnt++;
 175                 } else {
 176                         pcb2->pcb_ldt = user_ldt_alloc(pcb2,
 177                                 pcb2->pcb_ldt->ldt_len);
 178                 }
 179         }
 180         bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
 181               sizeof(lp2->lwp_thread->td_tls));
 182         /*
 183          * Now, cpu_switch() can schedule the new lwp.
 184          * pcb_rsp is loaded pointing to the cpu_switch() stack frame
 185          * containing the return address when exiting cpu_switch.
 186          * This will normally be to fork_trampoline(), which will have
 187          * %rbx loaded with the new lwp's pointer.  fork_trampoline()
 188          * will set up a stack to call fork_return(lp, frame); to complete
 189          * the return to user-mode.
 190          */
 191 }
 192
 193 /*
 194  * Prepare new lwp to return to the address specified in params.
 195  */
 196 int
 197 cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
 198 {
 199         struct trapframe *regs = lp->lwp_md.md_regs;
 200         void *bad_return = NULL;
 201         int error;
 202
 203         regs->tf_rip = (long)params->func;
 204         regs->tf_rsp = (long)params->stack;
 205         /* Set up argument for function call */
 206         regs->tf_rdi = (long)params->arg; /* JG Can this be in userspace addresses? */
 207         /*
 208          * Set up fake return address.  As the lwp function may never return,
 209          * we simply copy out a NULL pointer and force the lwp to receive
 210          * a SIGSEGV if it returns anyways.
 211          */
 212         regs->tf_rsp -= sizeof(void *);
 213         error = copyout(&bad_return, (void *)regs->tf_rsp, sizeof(bad_return));
 214         if (error)
 215                 return (error);
 216
 217         cpu_set_fork_handler(lp,
 218             (void (*)(void *, struct trapframe *))generic_lwp_return, lp);
 219         return (0);
 220 }
 221
 222 /*
 223  * Intercept the return address from a freshly forked process that has NOT
 224  * been scheduled yet.
 225  *
 226  * This is needed to make kernel threads stay in kernel mode.
 227  */
 228 void
 229 cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *),
 230                      void *arg)
 231 {
 232         /*
 233          * Note that the trap frame follows the args, so the function
 234          * is really called like this:  func(arg, frame);
 235          */
 236         lp->lwp_thread->td_pcb->pcb_rbx = (long)func;   /* function */
 237         lp->lwp_thread->td_pcb->pcb_r12 = (long)arg;    /* first arg */
 238 }
 239
 240 void
 241 cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
 242 {
 243         td->td_pcb->pcb_rbx = (long)func;
 244         td->td_pcb->pcb_r12 = (long)arg;
 245         td->td_switch = cpu_lwkt_switch;
 246         td->td_sp -= sizeof(void *);
 247         *(void **)td->td_sp = rfunc;    /* exit function on return */
 248         td->td_sp -= sizeof(void *);
 249         *(void **)td->td_sp = cpu_kthread_restore;
 250 }
 251
 252 void
 253 cpu_lwp_exit(void)
 254 {
 255         struct thread *td = curthread;
 256         struct pcb *pcb;
 257         npxexit();
 258         pcb = td->td_pcb;
 259
 260         /* Some i386 functionality was dropped */
 261         KKASSERT(pcb->pcb_ext == NULL);
 262
 263         /*
 264          * disable all hardware breakpoints
 265          */
 266         if (pcb->pcb_flags & PCB_DBREGS) {
 267                 reset_dbregs();
 268                 pcb->pcb_flags &= ~PCB_DBREGS;
 269         }
 270         td->td_gd->gd_cnt.v_swtch++;
 271
 272         crit_enter_quick(td);
 273         lwkt_deschedule_self(td);
 274         lwkt_remove_tdallq(td);
 275         cpu_thread_exit();
 276 }
 277
 278 /*
 279  * Terminate the current thread.  The caller must have already acquired
 280  * the thread's rwlock and placed it on a reap list or otherwise notified
 281  * a reaper of its existance.  We set a special assembly switch function which
 282  * releases td_rwlock after it has cleaned up the MMU state and switched
 283  * out the stack.
 284  *
 285  * Must be caller from a critical section and with the thread descheduled.
 286  */
 287 void
 288 cpu_thread_exit(void)
 289 {
 290         curthread->td_switch = cpu_exit_switch;
 291         curthread->td_flags |= TDF_EXITING;
 292         lwkt_switch();
 293         panic("cpu_thread_exit: lwkt_switch() unexpectedly returned");
 294 }
 295
 296 /*
 297  * Process Reaper.  Called after the caller has acquired the thread's
 298  * rwlock and removed it from the reap list.
 299  */
 300 void
 301 cpu_proc_wait(struct proc *p)
 302 {
 303         /* drop per-process resources */
 304         pmap_dispose_proc(p);
 305 }
 306
 307 int
 308 grow_stack(struct proc *p, u_long sp)
 309 {
 310         int rv;
 311
 312         rv = vm_map_growstack (p, sp);
 313         if (rv != KERN_SUCCESS)
 314                 return (0);
 315
 316         return (1);
 317 }
 318
 319 /*
 320  * Tell whether this address is in some physical memory region.
 321  * Currently used by the kernel coredump code in order to avoid
 322  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
 323  * or other unpredictable behaviour.
 324  */
 325
 326 int
 327 is_physical_memory(vm_offset_t addr)
 328 {
 329         return 1;
 330 }
 331
 332 /*
 333  * Used by /dev/kmem to determine if we can safely read or write
 334  * the requested KVA range.  Some portions of kernel memory are
 335  * not governed by our virtual page table.
 336  */
 337 extern int64_t _end;
 338 extern void _start(void);
 339
 340 int
 341 kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
 342 {
 343         vm_offset_t addr;
 344
 345         if (saddr >= trunc_page((vm_offset_t)&_start) && eaddr <= round_page((vm_offset_t)&_end))
 346                 return 0;
 347         if (saddr < KvaStart)
 348                 return EFAULT;
 349         if (eaddr >= KvaEnd)
 350                 return EFAULT;
 351         for (addr = saddr; addr < eaddr; addr += PAGE_SIZE)  {
 352                 if (pmap_extract(&kernel_pmap, addr) == 0)
 353                         return EFAULT;
 354         }
 355         if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
 356                 return EFAULT;
 357         return 0;
 358 }