pmap.c

早期freebsd实现

/* 
 * Copyright (c) 1992 OMRON Corporation.
 * Copyright (c) 1991, 1993
 *	The Regents of the University of California.  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.
 *
 * 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: hp300/hp300/pmap.c	8.2 (Berkeley) 11/14/93
 *
 *	@(#)pmap.c	8.2 (Berkeley) 12/6/93
 */

/*
 * LUNA physical map management code taken from:
 * HP9000/300 series physical map management code.
 *
 * Supports:
 *	68030 with on-chip MMU	(LUNA-I)
 *	68040 with on-chip MMU	(LUNA-II)
 *
 * Notes:
 *	Don't even pay lip service to multiprocessor support.
 *
 *	We assume TLB entries don't have process tags (except for the
 *	supervisor/user distinction) so we only invalidate TLB entries
 *	when changing mappings for the current (or kernel) pmap.  This is
 *	technically not true for the 68551 but we flush the TLB on every
 *	context switch, so it effectively winds up that way.
 *
 *	Bitwise and/or operations are significantly faster than bitfield
 *	references so we use them when accessing STE/PTEs in the pmap_pte_*
 *	macros.  Note also that the two are not always equivalent; e.g.:
 *		(*(int *)pte & PG_PROT) [4] != pte->pg_prot [1]
 *	and a couple of routines that deal with protection and wiring take
 *	some shortcuts that assume the and/or definitions.
 *
 *	This implementation will only work for PAGE_SIZE == NBPG
 *	(i.e. 4096 bytes).
 */

/*
 *	Manages physical address maps.
 *
 *	In addition to hardware address maps, this
 *	module is called upon to provide software-use-only
 *	maps which may or may not be stored in the same
 *	form as hardware maps.  These pseudo-maps are
 *	used to store intermediate results from copy
 *	operations to and from address spaces.
 *
 *	Since the information managed by this module is
 *	also stored by the logical address mapping module,
 *	this module may throw away valid virtual-to-physical
 *	mappings at almost any time.  However, invalidations
 *	of virtual-to-physical mappings must be done as
 *	requested.
 *
 *	In order to cope with hardware architectures which
 *	make virtual-to-physical map invalidates expensive,
 *	this module may delay invalidate or reduced protection
 *	operations until such time as they are actually
 *	necessary.  This module is given full information as
 *	to which processors are currently using which maps,
 *	and to when physical maps must be made correct.
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/malloc.h>
#include <sys/user.h>

#include <luna68k/luna68k/pte.h>

#include <vm/vm.h>
#include <vm/vm_kern.h>
#include <vm/vm_page.h>

#include <machine/cpu.h>

#ifdef PMAPSTATS
struct {
	int collectscans;
	int collectpages;
	int kpttotal;
	int kptinuse;
	int kptmaxuse;
} kpt_stats;
struct {
	int kernel;	/* entering kernel mapping */
	int user;	/* entering user mapping */
	int ptpneeded;	/* needed to allocate a PT page */
	int nochange;	/* no change at all */
	int pwchange;	/* no mapping change, just wiring or protection */
	int wchange;	/* no mapping change, just wiring */
	int pchange;	/* no mapping change, just protection */
	int mchange;	/* was mapped but mapping to different page */
	int managed;	/* a managed page */
	int firstpv;	/* first mapping for this PA */
	int secondpv;	/* second mapping for this PA */
	int ci;		/* cache inhibited */
	int unmanaged;	/* not a managed page */
	int flushes;	/* cache flushes */
} enter_stats;
struct {
	int calls;
	int removes;
	int pvfirst;
	int pvsearch;
	int ptinvalid;
	int uflushes;
	int sflushes;
} remove_stats;
struct {
	int calls;
	int changed;
	int alreadyro;
	int alreadyrw;
} protect_stats;
struct chgstats {
	int setcalls;
	int sethits;
	int setmiss;
	int clrcalls;
	int clrhits;
	int clrmiss;
} changebit_stats[16];
#endif

#ifdef DEBUG
int debugmap = 0;
int pmapdebug = 0x2000;
#define PDB_FOLLOW	0x0001
#define PDB_INIT	0x0002
#define PDB_ENTER	0x0004
#define PDB_REMOVE	0x0008
#define PDB_CREATE	0x0010
#define PDB_PTPAGE	0x0020
#define PDB_CACHE	0x0040
#define PDB_BITS	0x0080
#define PDB_COLLECT	0x0100
#define PDB_PROTECT	0x0200
#define PDB_SEGTAB	0x0400
#define PDB_MULTIMAP	0x0800
#define PDB_PARANOIA	0x2000
#define PDB_WIRING	0x4000
#define PDB_PVDUMP	0x8000

#ifdef HAVEVAC
int pmapvacflush = 0;
#define	PVF_ENTER	0x01
#define	PVF_REMOVE	0x02
#define	PVF_PROTECT	0x04
#define	PVF_TOTAL	0x80
#endif

#if defined(LUNA2)
int dowriteback = 1;	/* 68040: enable writeback caching */
int dokwriteback = 1;	/* 68040: enable writeback caching of kernel AS */
#endif

extern vm_offset_t pager_sva, pager_eva;
#endif

/*
 * Get STEs and PTEs for user/kernel address space
 */
#if defined(LUNA2)
#define	pmap_ste1(m, v)	\
	(&((m)->pm_stab[(vm_offset_t)(v) >> SG4_SHIFT1]))
/* XXX assumes physically contiguous ST pages (if more than one) */
#define pmap_ste2(m, v) \
	(&((m)->pm_stab[(st_entry_t *)(*(u_int *)pmap_ste1(m, v) & SG4_ADDR1) \
			- (m)->pm_stpa + (((v) & SG4_MASK2) >> SG4_SHIFT2)]))
#define	pmap_ste(m, v)	\
	(&((m)->pm_stab[(vm_offset_t)(v) \
			>> (mmutype == MMU_68040 ? SG4_SHIFT1 : SG_ISHIFT)]))
#define pmap_ste_v(m, v) \
	(mmutype == MMU_68040 \
	 ? ((*(int *)pmap_ste1(m, v) & SG_V) && \
	    (*(int *)pmap_ste2(m, v) & SG_V)) \
	 : (*(int *)pmap_ste(m, v) & SG_V))
#else
#define	pmap_ste(m, v)	 (&((m)->pm_stab[(vm_offset_t)(v) >> SG_ISHIFT]))
#define pmap_ste_v(m, v) (*(int *)pmap_ste(m, v) & SG_V)
#endif

#define pmap_pte(m, v)	(&((m)->pm_ptab[(vm_offset_t)(v) >> PG_SHIFT]))
#define pmap_pte_pa(pte)	(*(int *)(pte) & PG_FRAME)
#define pmap_pte_w(pte)		(*(int *)(pte) & PG_W)
#define pmap_pte_ci(pte)	(*(int *)(pte) & PG_CI)
#define pmap_pte_m(pte)		(*(int *)(pte) & PG_M)
#define pmap_pte_u(pte)		(*(int *)(pte) & PG_U)
#define pmap_pte_prot(pte)	(*(int *)(pte) & PG_PROT)
#define pmap_pte_v(pte)		(*(int *)(pte) & PG_V)

#define pmap_pte_set_w(pte, v) \
	if (v) *(int *)(pte) |= PG_W; else *(int *)(pte) &= ~PG_W
#define pmap_pte_set_prot(pte, v) \
	if (v) *(int *)(pte) |= PG_PROT; else *(int *)(pte) &= ~PG_PROT
#define pmap_pte_w_chg(pte, nw)		((nw) ^ pmap_pte_w(pte))
#define pmap_pte_prot_chg(pte, np)	((np) ^ pmap_pte_prot(pte))

/*
 * Given a map and a machine independent protection code,
 * convert to an luna protection code.
 */
#define pte_prot(m, p)	(protection_codes[p])
int	protection_codes[8];

/*
 * Kernel page table page management.
 */
struct kpt_page {
	struct kpt_page *kpt_next;	/* link on either used or free list */
	vm_offset_t	kpt_va;		/* always valid kernel VA */
	vm_offset_t	kpt_pa;		/* PA of this page (for speed) */
};
struct kpt_page *kpt_free_list, *kpt_used_list;
struct kpt_page *kpt_pages;

/*
 * Kernel segment/page table and page table map.
 * The page table map gives us a level of indirection we need to dynamically
 * expand the page table.  It is essentially a copy of the segment table
 * with PTEs instead of STEs.  All are initialized in locore at boot time.
 * Sysmap will initially contain VM_KERNEL_PT_PAGES pages of PTEs.
 * Segtabzero is an empty segment table which all processes share til they
 * reference something.
 */
st_entry_t	*Sysseg;
pt_entry_t	*Sysmap, *Sysptmap;
st_entry_t	*Segtabzero, *Segtabzeropa;
vm_size_t	Sysptsize = VM_KERNEL_PT_PAGES;

struct pmap	kernel_pmap_store;
vm_map_t	pt_map;

vm_offset_t    	avail_start;	/* PA of first available physical page */
vm_offset_t	avail_end;	/* PA of last available physical page */
vm_size_t	mem_size;	/* memory size in bytes */
vm_offset_t	virtual_avail;  /* VA of first avail page (after kernel bss)*/
vm_offset_t	virtual_end;	/* VA of last avail page (end of kernel AS) */
vm_offset_t	vm_first_phys;	/* PA of first managed page */
vm_offset_t	vm_last_phys;	/* PA just past last managed page */
boolean_t	pmap_initialized = FALSE;	/* Has pmap_init completed? */
char		*pmap_attributes;	/* reference and modify bits */
#ifdef HAVEVAC
int		pmap_aliasmask;	/* seperation at which VA aliasing ok */
#endif
#if defined(LUNA2)
int		protostfree;	/* prototype (default) free ST map */
#endif

/*
 * Internal routines
 */
void pmap_remove_mapping __P((pmap_t, vm_offset_t, pt_entry_t *, int));
boolean_t pmap_testbit	__P((vm_offset_t, int));
void pmap_changebit	__P((vm_offset_t, int, boolean_t));
void pmap_enter_ptpage	__P((pmap_t, vm_offset_t));
#ifdef DEBUG
void pmap_pvdump	__P((vm_offset_t));
void pmap_check_wiring	__P((char *, vm_offset_t));
#endif

/* pmap_remove_mapping flags */
#define	PRM_TFLUSH	1
#define	PRM_CFLUSH	2

/*
 * Bootstrap memory allocator. This function allows for early dynamic
 * memory allocation until the virtual memory system has been bootstrapped.
 * After that point, either kmem_alloc or malloc should be used. This
 * function works by stealing pages from the (to be) managed page pool,
 * stealing virtual address space, then mapping the pages and zeroing them.
 *
 * It should be used from pmap_bootstrap till vm_page_startup, afterwards
 * it cannot be used, and will generate a panic if tried. Note that this
 * memory will never be freed, and in essence it is wired down.
 */
void *
pmap_bootstrap_alloc(size) {
	vm_offset_t val;
	int i;
	extern boolean_t vm_page_startup_initialized;
	
	if (vm_page_startup_initialized)
		panic("pmap_bootstrap_alloc: called after startup initialized");
	size = round_page(size);
	val = virtual_avail;

	virtual_avail = pmap_map(virtual_avail, avail_start,
		avail_start + size, VM_PROT_READ|VM_PROT_WRITE);
	avail_start += size;

	blkclr ((caddr_t) val, size);
	return ((void *) val);
}

/*
 *	Initialize the pmap module.
 *	Called by vm_init, to initialize any structures that the pmap
 *	system needs to map virtual memory.
 */
void
pmap_init(phys_start, phys_end)
	vm_offset_t	phys_start, phys_end;
{
	vm_offset_t	addr, addr2;
	vm_size_t	npg, s;
	int		rv;
	extern char kstack[];

#ifdef DEBUG
	if (pmapdebug & PDB_FOLLOW)
		printf("pmap_init(%x, %x)\n", phys_start, phys_end);
#endif
	/*
	 * Now that kernel map has been allocated, we can mark as
	 * unavailable regions which we have mapped in locore.
	 */
	addr = (vm_offset_t) Sysmap;
	vm_object_reference(kernel_object);
	(void) vm_map_find(kernel_map, kernel_object, addr,
			   &addr, LUNA_MAX_PTSIZE, FALSE);
	/*
	 * If this fails it is probably because the static portion of
	 * the kernel page table isn't big enough and we overran the
	 * page table map.   Need to adjust pmap_size() in luna_init.c.
	 */
	if (addr != (vm_offset_t)Sysmap)
		goto bogons;

	addr = (vm_offset_t) kstack;
	vm_object_reference(kernel_object);
	(void) vm_map_find(kernel_map, kernel_object, addr,
			   &addr, luna_ptob(UPAGES), FALSE);
	if (addr != (vm_offset_t)kstack)
bogons:
		panic("pmap_init: bogons in the VM system!\n");

#ifdef DEBUG
	if (pmapdebug & PDB_INIT) {
		printf("pmap_init: Sysseg %x, Sysmap %x, Sysptmap %x\n",
		       Sysseg, Sysmap, Sysptmap);
		printf("  pstart %x, pend %x, vstart %x, vend %x\n",
		       avail_start, avail_end, virtual_avail, virtual_end);
	}
#endif

	/*
	 * Allocate memory for random pmap data structures.  Includes the
	 * initial segment table, pv_head_table and pmap_attributes.
	 */
	npg = atop(phys_end - phys_start);
	s = (vm_size_t) (LUNA_STSIZE + sizeof(struct pv_entry) * npg + npg);
	s = round_page(s);
	addr = (vm_offset_t) kmem_alloc(kernel_map, s);
	Segtabzero = (st_entry_t *) addr;
	Segtabzeropa = (st_entry_t *) pmap_extract(kernel_pmap, addr);
	addr += LUNA_STSIZE;
	pv_table = (pv_entry_t) addr;
	addr += sizeof(struct pv_entry) * npg;
	pmap_attributes = (char *) addr;
#ifdef DEBUG
	if (pmapdebug & PDB_INIT)
		printf("pmap_init: %x bytes: npg %x s0 %x(%x) tbl %x atr %x\n",
		       s, npg, Segtabzero, Segtabzeropa,
		       pv_table, pmap_attributes);
#endif

	/*
	 * Allocate physical memory for kernel PT pages and their management.
	 * We need 1 PT page per possible task plus some slop.
	 */
	npg = min(atop(LUNA_MAX_KPTSIZE), maxproc+16);
	s = ptoa(npg) + round_page(npg * sizeof(struct kpt_page));

	/*
	 * Verify that space will be allocated in region for which
	 * we already have kernel PT pages.
	 */
	addr = 0;
	rv = vm_map_find(kernel_map, NULL, 0, &addr, s, TRUE);
	if (rv != KERN_SUCCESS || addr + s >= (vm_offset_t)Sysmap)
		panic("pmap_init: kernel PT too small");
	vm_map_remove(kernel_map, addr, addr + s);

	/*
	 * Now allocate the space and link the pages together to
	 * form the KPT free list.
	 */
	addr = (vm_offset_t) kmem_alloc(kernel_map, s);
	s = ptoa(npg);
	addr2 = addr + s;
	kpt_pages = &((struct kpt_page *)addr2)[npg];
	kpt_free_list = (struct kpt_page *) 0;
	do {
		addr2 -= LUNA_PAGE_SIZE;
		(--kpt_pages)->kpt_next = kpt_free_list;
		kpt_free_list = kpt_pages;
		kpt_pages->kpt_va = addr2;
		kpt_pages->kpt_pa = pmap_extract(kernel_pmap, addr2);
	} while (addr != addr2);
#ifdef PMAPSTATS
	kpt_stats.kpttotal = atop(s);
#endif
#ifdef DEBUG
	if (pmapdebug & PDB_INIT)
		printf("pmap_init: KPT: %d pages from %x to %x\n",
		       atop(s), addr, addr + s);
#endif

	/*
	 * Slightly modified version of kmem_suballoc() to get page table
	 * map where we want it.
	 */
	addr = LUNA_PTBASE;
	s = min(LUNA_PTMAXSIZE, maxproc*LUNA_MAX_PTSIZE);
	addr2 = addr + s;
	rv = vm_map_find(kernel_map, NULL, 0, &addr, s, TRUE);
	if (rv != KERN_SUCCESS)
		panic("pmap_init: cannot allocate space for PT map");
	pmap_reference(vm_map_pmap(kernel_map));
	pt_map = vm_map_create(vm_map_pmap(kernel_map), addr, addr2, TRUE);
	if (pt_map == NULL)
		panic("pmap_init: cannot create pt_map");
	rv = vm_map_submap(kernel_map, addr, addr2, pt_map);
	if (rv != KERN_SUCCESS)
		panic("pmap_init: cannot map range to pt_map");
#ifdef DEBUG
	if (pmapdebug & PDB_INIT)
		printf("pmap_init: pt_map [%x - %x)\n", addr, addr2);
#endif

#if defined(LUNA2)
	if (mmutype == MMU_68040) {
		protostfree = ~l2tobm(0);
		for (rv = MAXUL2SIZE; rv < sizeof(protostfree)*NBBY; rv++)
			protostfree &= ~l2tobm(rv);
	}
#endif

	/*
	 * Now it is safe to enable pv_table recording.
	 */
	vm_first_phys = phys_start;
	vm_last_phys = phys_end;
	pmap_initialized = TRUE;
}

/*
 *	Used to map a range of physical addresses into kernel
 *	virtual address space.
 *
 *	For now, VM is already on, we only need to map the
 *	specified memory.
 */
vm_offset_t
pmap_map(virt, start, end, prot)
	vm_offset_t	virt;