amd64/amd64/vm_machdep.c

/*-
 * Copyright (c) 1982, 1986 The Regents of the University of California.
 * Copyright (c) 1989, 1990 William Jolitz
 * All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * the Systems Programming Group of the University of Utah Computer
 * Science Department, and William Jolitz.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the University of
 *	California, Berkeley and its contributors.
 * 4. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
 *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
 *	$Id: vm_machdep.c,v 1.10 1994/01/21 17:11:38 davidg Exp $
 */

#include "npx.h"
#include "param.h"
#include "systm.h"
#include "proc.h"
#include "malloc.h"
#include "buf.h"
#include "user.h"

#include "../include/cpu.h"

#include "vm/vm.h"
#include "vm/vm_kern.h"

/*
 * Finish a fork operation, with process p2 nearly set up.
 * Copy and update the kernel stack and pcb, making the child
 * ready to run, and marking it so that it can return differently
 * than the parent.  Returns 1 in the child process, 0 in the parent.
 * We currently double-map the user area so that the stack is at the same
 * address in each process; in the future we will probably relocate
 * the frame pointers on the stack after copying.
 */
int
cpu_fork(p1, p2)
	register struct proc *p1, *p2;
{
	register struct user *up = p2->p_addr;
	int foo, offset, addr, i;
	extern char kstack[];
	extern int mvesp();

	/*
	 * Copy pcb and stack from proc p1 to p2.
	 * We do this as cheaply as possible, copying only the active
	 * part of the stack.  The stack and pcb need to agree;
	 * this is tricky, as the final pcb is constructed by savectx,
	 * but its frame isn't yet on the stack when the stack is copied.
	 * swtch compensates for this when the child eventually runs.
	 * This should be done differently, with a single call
	 * that copies and updates the pcb+stack,
	 * replacing the bcopy and savectx.
	 */
	p2->p_addr->u_pcb = p1->p_addr->u_pcb;
	offset = mvesp() - (int)kstack;
	bcopy((caddr_t)kstack + offset, (caddr_t)p2->p_addr + offset,
	    (unsigned) ctob(UPAGES) - offset);
	p2->p_regs = p1->p_regs;

	/*
	 * Wire top of address space of child to it's kstack.
	 * First, fault in a page of pte's to map it.
	 */
#if 0
        addr = trunc_page((u_int)vtopte(kstack));
	vm_map_pageable(&p2->p_vmspace->vm_map, addr, addr+NBPG, FALSE);
	for (i=0; i < UPAGES; i++)
		pmap_enter(&p2->p_vmspace->vm_pmap, kstack+i*NBPG,
			   pmap_extract(kernel_pmap, ((int)p2->p_addr)+i*NBPG),
			   /*
			    * The user area has to be mapped writable because
			    * it contains the kernel stack (when CR0_WP is on
			    * on a 486 there is no user-read/kernel-write
			    * mode).  It is protected from user mode access
			    * by the segment limits.
			    */
			   VM_PROT_READ|VM_PROT_WRITE, TRUE);
#endif
	pmap_activate(&p2->p_vmspace->vm_pmap, &up->u_pcb);

	/*
	 *
	 * Arrange for a non-local goto when the new process
	 * is started, to resume here, returning nonzero from setjmp.
	 */
	if (savectx(up, 1)) {
		/*
		 * Return 1 in child.
		 */
		return (1);
	}
	return (0);
}

#ifdef notyet
/*
 * cpu_exit is called as the last action during exit.
 *
 * We change to an inactive address space and a "safe" stack,
 * passing thru an argument to the new stack. Now, safely isolated
 * from the resources we're shedding, we release the address space
 * and any remaining machine-dependent resources, including the
 * memory for the user structure and kernel stack.
 *
 * Next, we assign a dummy context to be written over by swtch,
 * calling it to send this process off to oblivion.
 * [The nullpcb allows us to minimize cost in swtch() by not having
 * a special case].
 */
struct proc *swtch_to_inactive();
volatile void
cpu_exit(p)
	register struct proc *p;
{
	static struct pcb nullpcb;	/* pcb to overwrite on last swtch */

#if NNPX > 0
	npxexit(p);
#endif	/* NNPX */

	/* move to inactive space and stack, passing arg accross */
	p = swtch_to_inactive(p);

	/* drop per-process resources */
	vmspace_free(p->p_vmspace);
	kmem_free(kernel_map, (vm_offset_t)p->p_addr, ctob(UPAGES));

	p->p_addr = (struct user *) &nullpcb;
	splclock();
	swtch();
	/* NOTREACHED */
}
#else
void
cpu_exit(p)
	register struct proc *p;
{

#if NNPX > 0
	npxexit(p);
#endif	/* NNPX */
	splclock();
	curproc = 0;
	swtch();
	/*
	 * This is to shutup the compiler, and if swtch() failed I suppose
	 * this would be a good thing.  This keeps gcc happy because panic
	 * is a volatile void function as well.
	 */
	panic("cpu_exit");
}

void
cpu_wait(p) struct proc *p; {
/*	extern vm_map_t upages_map; */
	extern char kstack[];

	/* drop per-process resources */
 	pmap_remove(vm_map_pmap(kernel_map), (vm_offset_t) p->p_addr,
		((vm_offset_t) p->p_addr) + ctob(UPAGES));
	kmem_free(kernel_map, (vm_offset_t)p->p_addr, ctob(UPAGES));
	vmspace_free(p->p_vmspace);
}
#endif

/*
 * Set a red zone in the kernel stack after the u. area.
 */
void
setredzone(pte, vaddr)
	u_short *pte;
	caddr_t vaddr;
{
/* eventually do this by setting up an expand-down stack segment
   for ss0: selector, allowing stack access down to top of u.
   this means though that protection violations need to be handled
   thru a double fault exception that must do an integral task
   switch to a known good context, within which a dump can be
   taken. a sensible scheme might be to save the initial context
   used by sched (that has physical memory mapped 1:1 at bottom)
   and take the dump while still in mapped mode */
}

/*
 * Move pages from one kernel virtual address to another.
 * Both addresses are assumed to reside in the Sysmap,
 * and size must be a multiple of CLSIZE.
 */
void
pagemove(from, to, size)
	register caddr_t from, to;
	int size;
{
	register struct pte *fpte, *tpte;

	if (size % CLBYTES)
		panic("pagemove");
	fpte = kvtopte(from);
	tpte = kvtopte(to);
	while (size > 0) {
		*tpte++ = *fpte;
		*(int *)fpte++ = 0;
		from += NBPG;
		to += NBPG;
		size -= NBPG;
	}
	tlbflush();
}

/*
 * Convert kernel VA to physical address
 */
u_long
kvtop(void *addr)
{
	vm_offset_t va;

	va = pmap_extract(kernel_pmap, (vm_offset_t)addr);
	if (va == 0)
		panic("kvtop: zero page frame");
	return((int)va);
}

extern vm_map_t phys_map;

/*
 * Map an IO request into kernel virtual address space.  Requests fall into
 * one of five catagories:
 *
 *	B_PHYS|B_UAREA:	User u-area swap.
 *			Address is relative to start of u-area (p_addr).
 *	B_PHYS|B_PAGET:	User page table swap.
 *			Address is a kernel VA in usrpt (Usrptmap).
 *	B_PHYS|B_DIRTY:	Dirty page push.
 *			Address is a VA in proc2's address space.
 *	B_PHYS|B_PGIN:	Kernel pagein of user pages.
 *			Address is VA in user's address space.
 *	B_PHYS:		User "raw" IO request.
 *			Address is VA in user's address space.
 *
 * All requests are (re)mapped into kernel VA space via the useriomap
 * (a name with only slightly more meaning than "kernelmap")
 */
void
vmapbuf(bp)
	register struct buf *bp;
{
	register int npf;
	register caddr_t addr;
	register long flags = bp->b_flags;
	struct proc *p;
	int off;
	vm_offset_t kva;
	register vm_offset_t pa;

	if ((flags & B_PHYS) == 0)
		panic("vmapbuf");
	addr = bp->b_saveaddr = bp->b_un.b_addr;
	off = (int)addr & PGOFSET;
	p = bp->b_proc;
	npf = btoc(round_page(bp->b_bcount + off));
	kva = kmem_alloc_wait(phys_map, ctob(npf));
	bp->b_un.b_addr = (caddr_t) (kva + off);
	while (npf--) {
		pa = pmap_extract(&p->p_vmspace->vm_pmap, (vm_offset_t)addr);
		if (pa == 0)
			panic("vmapbuf: null page frame");
		pmap_enter(vm_map_pmap(phys_map), kva, trunc_page(pa),
			   VM_PROT_READ|VM_PROT_WRITE, TRUE);
		addr += PAGE_SIZE;
		kva += PAGE_SIZE;
	}
}

/*
 * Free the io map PTEs associated with this IO operation.
 * We also invalidate the TLB entries and restore the original b_addr.
 */
void
vunmapbuf(bp)
	register struct buf *bp;
{
	register int npf;
	register caddr_t addr = bp->b_un.b_addr;
	vm_offset_t kva;

	if ((bp->b_flags & B_PHYS) == 0)
		panic("vunmapbuf");
	npf = btoc(round_page(bp->b_bcount + ((int)addr & PGOFSET)));
	kva = (vm_offset_t)((int)addr & ~PGOFSET);
	kmem_free_wakeup(phys_map, kva, ctob(npf));
	bp->b_un.b_addr = bp->b_saveaddr;
	bp->b_saveaddr = NULL;
}

/*
 * Force reset the processor by invalidating the entire address space!
 */
void
cpu_reset() {

	/* force a shutdown by unmapping entire address space ! */
	bzero((caddr_t) PTD, NBPG);

	/* "good night, sweet prince .... <THUNK!>" */
	tlbflush();
	/* NOTREACHED */
	while(1);
}

/*
 * Grow the user stack to allow for 'sp'. This version grows the stack in
 *	chunks of DFLSSIZ. It is expected (required) that there is an
 *	integer number of DFLSSIZ chunks in MAXSSIZ.
 */
int
grow(p, sp)
	struct proc *p;
	int sp;
{
	unsigned int nss;
	caddr_t v;
	struct vmspace *vm = p->p_vmspace;

	if ((caddr_t)sp <= vm->vm_maxsaddr || (unsigned)sp >= (unsigned)USRSTACK)
	    return (1);

	nss = roundup(USRSTACK - (unsigned)sp, PAGE_SIZE);

	if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur)
		return (0);

	if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT,
	    DFLSSIZ) < nss) {
		int grow_amount;
		/*
		 * If necessary, grow the VM that the stack occupies
		 * to allow for the rlimit. This allows us to not have
		 * to allocate all of the VM up-front in execve (which
		 * is expensive).
		 * Grow the VM by the amount requested rounded up to
		 * the nearest DFLSSIZ to provide for some hysteresis.
		 */
		grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), DFLSSIZ);
		v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT,
		    DFLSSIZ) - grow_amount;
		/*
		 * If there isn't enough room to extend by DFLSSIZ, then
		 * just extend to the maximum size
		 */
		if (v < vm->vm_maxsaddr) {
			v = vm->vm_maxsaddr;
			grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT);
		}
		if (vm_allocate(&vm->vm_map, (vm_offset_t *)&v,
		    grow_amount, FALSE) != KERN_SUCCESS) {
			return (0);
		}
		vm->vm_ssize += grow_amount >> PAGE_SHIFT;
	}

	return (1);
}