The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/i386/i386/vm_machdep.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    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  */
   42 
   43 #include <sys/cdefs.h>
   44 __FBSDID("$FreeBSD: releng/11.2/sys/i386/i386/vm_machdep.c 332760 2018-04-19 07:15:40Z avg $");
   45 
   46 #include "opt_isa.h"
   47 #include "opt_npx.h"
   48 #include "opt_reset.h"
   49 #include "opt_cpu.h"
   50 #include "opt_xbox.h"
   51 
   52 #include <sys/param.h>
   53 #include <sys/systm.h>
   54 #include <sys/bio.h>
   55 #include <sys/buf.h>
   56 #include <sys/kernel.h>
   57 #include <sys/ktr.h>
   58 #include <sys/lock.h>
   59 #include <sys/malloc.h>
   60 #include <sys/mbuf.h>
   61 #include <sys/mutex.h>
   62 #include <sys/pioctl.h>
   63 #include <sys/proc.h>
   64 #include <sys/sysent.h>
   65 #include <sys/sf_buf.h>
   66 #include <sys/smp.h>
   67 #include <sys/sched.h>
   68 #include <sys/sysctl.h>
   69 #include <sys/unistd.h>
   70 #include <sys/vnode.h>
   71 #include <sys/vmmeter.h>
   72 
   73 #include <machine/cpu.h>
   74 #include <machine/cputypes.h>
   75 #include <machine/md_var.h>
   76 #include <machine/pcb.h>
   77 #include <machine/pcb_ext.h>
   78 #include <machine/smp.h>
   79 #include <machine/vm86.h>
   80 
   81 #include <vm/vm.h>
   82 #include <vm/vm_extern.h>
   83 #include <vm/vm_kern.h>
   84 #include <vm/vm_page.h>
   85 #include <vm/vm_map.h>
   86 #include <vm/vm_param.h>
   87 
   88 #ifdef XBOX
   89 #include <machine/xbox.h>
   90 #endif
   91 
   92 #ifndef NSFBUFS
   93 #define NSFBUFS         (512 + maxusers * 16)
   94 #endif
   95 
   96 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
   97     "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
   98 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
   99     "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
  100 _Static_assert(__OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
  101     "__OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
  102 
  103 union savefpu *
  104 get_pcb_user_save_td(struct thread *td)
  105 {
  106         vm_offset_t p;
  107 
  108         p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
  109             roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
  110         KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
  111         return ((union savefpu *)p);
  112 }
  113 
  114 union savefpu *
  115 get_pcb_user_save_pcb(struct pcb *pcb)
  116 {
  117         vm_offset_t p;
  118 
  119         p = (vm_offset_t)(pcb + 1);
  120         return ((union savefpu *)p);
  121 }
  122 
  123 struct pcb *
  124 get_pcb_td(struct thread *td)
  125 {
  126         vm_offset_t p;
  127 
  128         p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
  129             roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
  130             sizeof(struct pcb);
  131         return ((struct pcb *)p);
  132 }
  133 
  134 void *
  135 alloc_fpusave(int flags)
  136 {
  137         void *res;
  138         struct savefpu_ymm *sf;
  139 
  140         res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
  141         if (use_xsave) {
  142                 sf = (struct savefpu_ymm *)res;
  143                 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
  144                 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
  145         }
  146         return (res);
  147 }
  148 /*
  149  * Finish a fork operation, with process p2 nearly set up.
  150  * Copy and update the pcb, set up the stack so that the child
  151  * ready to run and return to user mode.
  152  */
  153 void
  154 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
  155 {
  156         struct proc *p1;
  157         struct pcb *pcb2;
  158         struct mdproc *mdp2;
  159 
  160         p1 = td1->td_proc;
  161         if ((flags & RFPROC) == 0) {
  162                 if ((flags & RFMEM) == 0) {
  163                         /* unshare user LDT */
  164                         struct mdproc *mdp1 = &p1->p_md;
  165                         struct proc_ldt *pldt, *pldt1;
  166 
  167                         mtx_lock_spin(&dt_lock);
  168                         if ((pldt1 = mdp1->md_ldt) != NULL &&
  169                             pldt1->ldt_refcnt > 1) {
  170                                 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
  171                                 if (pldt == NULL)
  172                                         panic("could not copy LDT");
  173                                 mdp1->md_ldt = pldt;
  174                                 set_user_ldt(mdp1);
  175                                 user_ldt_deref(pldt1);
  176                         } else
  177                                 mtx_unlock_spin(&dt_lock);
  178                 }
  179                 return;
  180         }
  181 
  182         /* Ensure that td1's pcb is up to date. */
  183         if (td1 == curthread)
  184                 td1->td_pcb->pcb_gs = rgs();
  185         critical_enter();
  186         if (PCPU_GET(fpcurthread) == td1)
  187                 npxsave(td1->td_pcb->pcb_save);
  188         critical_exit();
  189 
  190         /* Point the pcb to the top of the stack */
  191         pcb2 = get_pcb_td(td2);
  192         td2->td_pcb = pcb2;
  193 
  194         /* Copy td1's pcb */
  195         bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
  196 
  197         /* Properly initialize pcb_save */
  198         pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
  199         bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
  200             cpu_max_ext_state_size);
  201 
  202         /* Point mdproc and then copy over td1's contents */
  203         mdp2 = &p2->p_md;
  204         bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
  205 
  206         /*
  207          * Create a new fresh stack for the new process.
  208          * Copy the trap frame for the return to user mode as if from a
  209          * syscall.  This copies most of the user mode register values.
  210          * The -16 is so we can expand the trapframe if we go to vm86.
  211          */
  212         td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
  213         bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
  214 
  215         td2->td_frame->tf_eax = 0;              /* Child returns zero */
  216         td2->td_frame->tf_eflags &= ~PSL_C;     /* success */
  217         td2->td_frame->tf_edx = 1;
  218 
  219         /*
  220          * If the parent process has the trap bit set (i.e. a debugger had
  221          * single stepped the process to the system call), we need to clear
  222          * the trap flag from the new frame unless the debugger had set PF_FORK
  223          * on the parent.  Otherwise, the child will receive a (likely
  224          * unexpected) SIGTRAP when it executes the first instruction after
  225          * returning  to userland.
  226          */
  227         if ((p1->p_pfsflags & PF_FORK) == 0)
  228                 td2->td_frame->tf_eflags &= ~PSL_T;
  229 
  230         /*
  231          * Set registers for trampoline to user mode.  Leave space for the
  232          * return address on stack.  These are the kernel mode register values.
  233          */
  234 #if defined(PAE) || defined(PAE_TABLES)
  235         pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
  236 #else
  237         pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
  238 #endif
  239         pcb2->pcb_edi = 0;
  240         pcb2->pcb_esi = (int)fork_return;       /* fork_trampoline argument */
  241         pcb2->pcb_ebp = 0;
  242         pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
  243         pcb2->pcb_ebx = (int)td2;               /* fork_trampoline argument */
  244         pcb2->pcb_eip = (int)fork_trampoline;
  245         pcb2->pcb_psl = PSL_KERNEL;             /* ints disabled */
  246         /*-
  247          * pcb2->pcb_dr*:       cloned above.
  248          * pcb2->pcb_savefpu:   cloned above.
  249          * pcb2->pcb_flags:     cloned above.
  250          * pcb2->pcb_onfault:   cloned above (always NULL here?).
  251          * pcb2->pcb_gs:        cloned above.
  252          * pcb2->pcb_ext:       cleared below.
  253          */
  254 
  255         /*
  256          * XXX don't copy the i/o pages.  this should probably be fixed.
  257          */
  258         pcb2->pcb_ext = 0;
  259 
  260         /* Copy the LDT, if necessary. */
  261         mtx_lock_spin(&dt_lock);
  262         if (mdp2->md_ldt != NULL) {
  263                 if (flags & RFMEM) {
  264                         mdp2->md_ldt->ldt_refcnt++;
  265                 } else {
  266                         mdp2->md_ldt = user_ldt_alloc(mdp2,
  267                             mdp2->md_ldt->ldt_len);
  268                         if (mdp2->md_ldt == NULL)
  269                                 panic("could not copy LDT");
  270                 }
  271         }
  272         mtx_unlock_spin(&dt_lock);
  273 
  274         /* Setup to release spin count in fork_exit(). */
  275         td2->td_md.md_spinlock_count = 1;
  276         td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
  277 
  278         /*
  279          * Now, cpu_switch() can schedule the new process.
  280          * pcb_esp is loaded pointing to the cpu_switch() stack frame
  281          * containing the return address when exiting cpu_switch.
  282          * This will normally be to fork_trampoline(), which will have
  283          * %ebx loaded with the new proc's pointer.  fork_trampoline()
  284          * will set up a stack to call fork_return(p, frame); to complete
  285          * the return to user-mode.
  286          */
  287 }
  288 
  289 /*
  290  * Intercept the return address from a freshly forked process that has NOT
  291  * been scheduled yet.
  292  *
  293  * This is needed to make kernel threads stay in kernel mode.
  294  */
  295 void
  296 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
  297 {
  298         /*
  299          * Note that the trap frame follows the args, so the function
  300          * is really called like this:  func(arg, frame);
  301          */
  302         td->td_pcb->pcb_esi = (int) func;       /* function */
  303         td->td_pcb->pcb_ebx = (int) arg;        /* first arg */
  304 }
  305 
  306 void
  307 cpu_exit(struct thread *td)
  308 {
  309 
  310         /*
  311          * If this process has a custom LDT, release it.  Reset pc->pcb_gs
  312          * and %gs before we free it in case they refer to an LDT entry.
  313          */
  314         mtx_lock_spin(&dt_lock);
  315         if (td->td_proc->p_md.md_ldt) {
  316                 td->td_pcb->pcb_gs = _udatasel;
  317                 load_gs(_udatasel);
  318                 user_ldt_free(td);
  319         } else
  320                 mtx_unlock_spin(&dt_lock);
  321 }
  322 
  323 void
  324 cpu_thread_exit(struct thread *td)
  325 {
  326 
  327         critical_enter();
  328         if (td == PCPU_GET(fpcurthread))
  329                 npxdrop();
  330         critical_exit();
  331 
  332         /* Disable any hardware breakpoints. */
  333         if (td->td_pcb->pcb_flags & PCB_DBREGS) {
  334                 reset_dbregs();
  335                 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
  336         }
  337 }
  338 
  339 void
  340 cpu_thread_clean(struct thread *td)
  341 {
  342         struct pcb *pcb;
  343 
  344         pcb = td->td_pcb; 
  345         if (pcb->pcb_ext != NULL) {
  346                 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
  347                 /*
  348                  * XXX do we need to move the TSS off the allocated pages
  349                  * before freeing them?  (not done here)
  350                  */
  351                 kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext,
  352                     ctob(IOPAGES + 1));
  353                 pcb->pcb_ext = NULL;
  354         }
  355 }
  356 
  357 void
  358 cpu_thread_swapin(struct thread *td)
  359 {
  360 }
  361 
  362 void
  363 cpu_thread_swapout(struct thread *td)
  364 {
  365 }
  366 
  367 void
  368 cpu_thread_alloc(struct thread *td)
  369 {
  370         struct pcb *pcb;
  371         struct xstate_hdr *xhdr;
  372 
  373         td->td_pcb = pcb = get_pcb_td(td);
  374         td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1;
  375         pcb->pcb_ext = NULL; 
  376         pcb->pcb_save = get_pcb_user_save_pcb(pcb);
  377         if (use_xsave) {
  378                 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
  379                 bzero(xhdr, sizeof(*xhdr));
  380                 xhdr->xstate_bv = xsave_mask;
  381         }
  382 }
  383 
  384 void
  385 cpu_thread_free(struct thread *td)
  386 {
  387 
  388         cpu_thread_clean(td);
  389 }
  390 
  391 void
  392 cpu_set_syscall_retval(struct thread *td, int error)
  393 {
  394 
  395         switch (error) {
  396         case 0:
  397                 td->td_frame->tf_eax = td->td_retval[0];
  398                 td->td_frame->tf_edx = td->td_retval[1];
  399                 td->td_frame->tf_eflags &= ~PSL_C;
  400                 break;
  401 
  402         case ERESTART:
  403                 /*
  404                  * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
  405                  * 0x80 is 2 bytes. We saved this in tf_err.
  406                  */
  407                 td->td_frame->tf_eip -= td->td_frame->tf_err;
  408                 break;
  409 
  410         case EJUSTRETURN:
  411                 break;
  412 
  413         default:
  414                 td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error);
  415                 td->td_frame->tf_eflags |= PSL_C;
  416                 break;
  417         }
  418 }
  419 
  420 /*
  421  * Initialize machine state, mostly pcb and trap frame for a new
  422  * thread, about to return to userspace.  Put enough state in the new
  423  * thread's PCB to get it to go back to the fork_return(), which
  424  * finalizes the thread state and handles peculiarities of the first
  425  * return to userspace for the new thread.
  426  */
  427 void
  428 cpu_copy_thread(struct thread *td, struct thread *td0)
  429 {
  430         struct pcb *pcb2;
  431 
  432         /* Point the pcb to the top of the stack. */
  433         pcb2 = td->td_pcb;
  434 
  435         /*
  436          * Copy the upcall pcb.  This loads kernel regs.
  437          * Those not loaded individually below get their default
  438          * values here.
  439          */
  440         bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
  441         pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
  442             PCB_KERNNPX);
  443         pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
  444         bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
  445             cpu_max_ext_state_size);
  446 
  447         /*
  448          * Create a new fresh stack for the new thread.
  449          */
  450         bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
  451 
  452         /* If the current thread has the trap bit set (i.e. a debugger had
  453          * single stepped the process to the system call), we need to clear
  454          * the trap flag from the new frame. Otherwise, the new thread will
  455          * receive a (likely unexpected) SIGTRAP when it executes the first
  456          * instruction after returning to userland.
  457          */
  458         td->td_frame->tf_eflags &= ~PSL_T;
  459 
  460         /*
  461          * Set registers for trampoline to user mode.  Leave space for the
  462          * return address on stack.  These are the kernel mode register values.
  463          */
  464         pcb2->pcb_edi = 0;
  465         pcb2->pcb_esi = (int)fork_return;                   /* trampoline arg */
  466         pcb2->pcb_ebp = 0;
  467         pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
  468         pcb2->pcb_ebx = (int)td;                            /* trampoline arg */
  469         pcb2->pcb_eip = (int)fork_trampoline;
  470         pcb2->pcb_psl &= ~(PSL_I);      /* interrupts must be disabled */
  471         pcb2->pcb_gs = rgs();
  472         /*
  473          * If we didn't copy the pcb, we'd need to do the following registers:
  474          * pcb2->pcb_cr3:       cloned above.
  475          * pcb2->pcb_dr*:       cloned above.
  476          * pcb2->pcb_savefpu:   cloned above.
  477          * pcb2->pcb_flags:     cloned above.
  478          * pcb2->pcb_onfault:   cloned above (always NULL here?).
  479          * pcb2->pcb_gs:        cloned above.
  480          * pcb2->pcb_ext:       cleared below.
  481          */
  482         pcb2->pcb_ext = NULL;
  483 
  484         /* Setup to release spin count in fork_exit(). */
  485         td->td_md.md_spinlock_count = 1;
  486         td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
  487 }
  488 
  489 /*
  490  * Set that machine state for performing an upcall that starts
  491  * the entry function with the given argument.
  492  */
  493 void
  494 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
  495     stack_t *stack)
  496 {
  497 
  498         /* 
  499          * Do any extra cleaning that needs to be done.
  500          * The thread may have optional components
  501          * that are not present in a fresh thread.
  502          * This may be a recycled thread so make it look
  503          * as though it's newly allocated.
  504          */
  505         cpu_thread_clean(td);
  506 
  507         /*
  508          * Set the trap frame to point at the beginning of the entry
  509          * function.
  510          */
  511         td->td_frame->tf_ebp = 0; 
  512         td->td_frame->tf_esp =
  513             (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
  514         td->td_frame->tf_eip = (int)entry;
  515 
  516         /* Return address sentinel value to stop stack unwinding. */
  517         suword((void *)td->td_frame->tf_esp, 0);
  518 
  519         /* Pass the argument to the entry point. */
  520         suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
  521             (int)arg);
  522 }
  523 
  524 int
  525 cpu_set_user_tls(struct thread *td, void *tls_base)
  526 {
  527         struct segment_descriptor sd;
  528         uint32_t base;
  529 
  530         /*
  531          * Construct a descriptor and store it in the pcb for
  532          * the next context switch.  Also store it in the gdt
  533          * so that the load of tf_fs into %fs will activate it
  534          * at return to userland.
  535          */
  536         base = (uint32_t)tls_base;
  537         sd.sd_lobase = base & 0xffffff;
  538         sd.sd_hibase = (base >> 24) & 0xff;
  539         sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
  540         sd.sd_hilimit = 0xf;
  541         sd.sd_type  = SDT_MEMRWA;
  542         sd.sd_dpl   = SEL_UPL;
  543         sd.sd_p     = 1;
  544         sd.sd_xx    = 0;
  545         sd.sd_def32 = 1;
  546         sd.sd_gran  = 1;
  547         critical_enter();
  548         /* set %gs */
  549         td->td_pcb->pcb_gsd = sd;
  550         if (td == curthread) {
  551                 PCPU_GET(fsgs_gdt)[1] = sd;
  552                 load_gs(GSEL(GUGS_SEL, SEL_UPL));
  553         }
  554         critical_exit();
  555         return (0);
  556 }
  557 
  558 /*
  559  * Convert kernel VA to physical address
  560  */
  561 vm_paddr_t
  562 kvtop(void *addr)
  563 {
  564         vm_paddr_t pa;
  565 
  566         pa = pmap_kextract((vm_offset_t)addr);
  567         if (pa == 0)
  568                 panic("kvtop: zero page frame");
  569         return (pa);
  570 }
  571 
  572 /*
  573  * Get an sf_buf from the freelist.  May block if none are available.
  574  */
  575 void
  576 sf_buf_map(struct sf_buf *sf, int flags)
  577 {
  578         pt_entry_t opte, *ptep;
  579 
  580         /*
  581          * Update the sf_buf's virtual-to-physical mapping, flushing the
  582          * virtual address from the TLB.  Since the reference count for 
  583          * the sf_buf's old mapping was zero, that mapping is not 
  584          * currently in use.  Consequently, there is no need to exchange 
  585          * the old and new PTEs atomically, even under PAE.
  586          */
  587         ptep = vtopte(sf->kva);
  588         opte = *ptep;
  589         *ptep = VM_PAGE_TO_PHYS(sf->m) | pgeflag | PG_RW | PG_V |
  590             pmap_cache_bits(sf->m->md.pat_mode, 0);
  591 
  592         /*
  593          * Avoid unnecessary TLB invalidations: If the sf_buf's old
  594          * virtual-to-physical mapping was not used, then any processor
  595          * that has invalidated the sf_buf's virtual address from its TLB
  596          * since the last used mapping need not invalidate again.
  597          */
  598 #ifdef SMP
  599         if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
  600                 CPU_ZERO(&sf->cpumask);
  601 
  602         sf_buf_shootdown(sf, flags);
  603 #else
  604         if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
  605                 pmap_invalidate_page(kernel_pmap, sf->kva);
  606 #endif
  607 }
  608 
  609 #ifdef SMP
  610 void
  611 sf_buf_shootdown(struct sf_buf *sf, int flags)
  612 {
  613         cpuset_t other_cpus;
  614         u_int cpuid;
  615 
  616         sched_pin();
  617         cpuid = PCPU_GET(cpuid);
  618         if (!CPU_ISSET(cpuid, &sf->cpumask)) {
  619                 CPU_SET(cpuid, &sf->cpumask);
  620                 invlpg(sf->kva);
  621         }
  622         if ((flags & SFB_CPUPRIVATE) == 0) {
  623                 other_cpus = all_cpus;
  624                 CPU_CLR(cpuid, &other_cpus);
  625                 CPU_NAND(&other_cpus, &sf->cpumask);
  626                 if (!CPU_EMPTY(&other_cpus)) {
  627                         CPU_OR(&sf->cpumask, &other_cpus);
  628                         smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap);
  629                 }
  630         }
  631         sched_unpin();
  632 }
  633 #endif
  634 
  635 /*
  636  * MD part of sf_buf_free().
  637  */
  638 int
  639 sf_buf_unmap(struct sf_buf *sf)
  640 {
  641 
  642         return (0);
  643 }
  644 
  645 static void
  646 sf_buf_invalidate(struct sf_buf *sf)
  647 {
  648         vm_page_t m = sf->m;
  649 
  650         /*
  651          * Use pmap_qenter to update the pte for
  652          * existing mapping, in particular, the PAT
  653          * settings are recalculated.
  654          */
  655         pmap_qenter(sf->kva, &m, 1);
  656         pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE);
  657 }
  658 
  659 /*
  660  * Invalidate the cache lines that may belong to the page, if
  661  * (possibly old) mapping of the page by sf buffer exists.  Returns
  662  * TRUE when mapping was found and cache invalidated.
  663  */
  664 boolean_t
  665 sf_buf_invalidate_cache(vm_page_t m)
  666 {
  667 
  668         return (sf_buf_process_page(m, sf_buf_invalidate));
  669 }
  670 
  671 /*
  672  * Software interrupt handler for queued VM system processing.
  673  */   
  674 void  
  675 swi_vm(void *dummy) 
  676 {     
  677         if (busdma_swi_pending != 0)
  678                 busdma_swi();
  679 }
  680 
  681 /*
  682  * Tell whether this address is in some physical memory region.
  683  * Currently used by the kernel coredump code in order to avoid
  684  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
  685  * or other unpredictable behaviour.
  686  */
  687 
  688 int
  689 is_physical_memory(vm_paddr_t addr)
  690 {
  691 
  692 #ifdef DEV_ISA
  693         /* The ISA ``memory hole''. */
  694         if (addr >= 0xa0000 && addr < 0x100000)
  695                 return 0;
  696 #endif
  697 
  698         /*
  699          * stuff other tests for known memory-mapped devices (PCI?)
  700          * here
  701          */
  702 
  703         return 1;
  704 }

Cache object: 908374ea18dcc32f27ad18a591d28f63


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.