Home | History | Annotate | Line # | Download | only in sun3x
pmap.c revision 1.88.2.3
      1  1.88.2.3     yamt /*	$NetBSD: pmap.c,v 1.88.2.3 2007/02/26 09:08:37 yamt Exp $	*/
      2       1.1      gwr 
      3       1.1      gwr /*-
      4      1.10   jeremy  * Copyright (c) 1996, 1997 The NetBSD Foundation, Inc.
      5       1.1      gwr  * All rights reserved.
      6       1.1      gwr  *
      7       1.1      gwr  * This code is derived from software contributed to The NetBSD Foundation
      8       1.1      gwr  * by Jeremy Cooper.
      9       1.1      gwr  *
     10       1.1      gwr  * Redistribution and use in source and binary forms, with or without
     11       1.1      gwr  * modification, are permitted provided that the following conditions
     12       1.1      gwr  * are met:
     13       1.1      gwr  * 1. Redistributions of source code must retain the above copyright
     14       1.1      gwr  *    notice, this list of conditions and the following disclaimer.
     15       1.1      gwr  * 2. Redistributions in binary form must reproduce the above copyright
     16       1.1      gwr  *    notice, this list of conditions and the following disclaimer in the
     17       1.1      gwr  *    documentation and/or other materials provided with the distribution.
     18       1.1      gwr  * 3. All advertising materials mentioning features or use of this software
     19       1.1      gwr  *    must display the following acknowledgement:
     20       1.1      gwr  *        This product includes software developed by the NetBSD
     21       1.1      gwr  *        Foundation, Inc. and its contributors.
     22       1.1      gwr  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23       1.1      gwr  *    contributors may be used to endorse or promote products derived
     24       1.1      gwr  *    from this software without specific prior written permission.
     25       1.1      gwr  *
     26       1.1      gwr  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27       1.1      gwr  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28       1.1      gwr  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29       1.1      gwr  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30       1.1      gwr  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31       1.1      gwr  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32       1.1      gwr  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33       1.1      gwr  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34       1.1      gwr  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35       1.1      gwr  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36       1.1      gwr  * POSSIBILITY OF SUCH DAMAGE.
     37       1.1      gwr  */
     38       1.1      gwr 
     39       1.1      gwr /*
     40       1.1      gwr  * XXX These comments aren't quite accurate.  Need to change.
     41       1.1      gwr  * The sun3x uses the MC68851 Memory Management Unit, which is built
     42       1.1      gwr  * into the CPU.  The 68851 maps virtual to physical addresses using
     43       1.1      gwr  * a multi-level table lookup, which is stored in the very memory that
     44       1.1      gwr  * it maps.  The number of levels of lookup is configurable from one
     45       1.1      gwr  * to four.  In this implementation, we use three, named 'A' through 'C'.
     46       1.1      gwr  *
     47       1.1      gwr  * The MMU translates virtual addresses into physical addresses by
     48      1.84      wiz  * traversing these tables in a process called a 'table walk'.  The most
     49       1.1      gwr  * significant 7 bits of the Virtual Address ('VA') being translated are
     50       1.1      gwr  * used as an index into the level A table, whose base in physical memory
     51       1.1      gwr  * is stored in a special MMU register, the 'CPU Root Pointer' or CRP.  The
     52       1.1      gwr  * address found at that index in the A table is used as the base
     53       1.1      gwr  * address for the next table, the B table.  The next six bits of the VA are
     54       1.1      gwr  * used as an index into the B table, which in turn gives the base address
     55       1.1      gwr  * of the third and final C table.
     56       1.1      gwr  *
     57       1.1      gwr  * The next six bits of the VA are used as an index into the C table to
     58       1.1      gwr  * locate a Page Table Entry (PTE).  The PTE is a physical address in memory
     59       1.1      gwr  * to which the remaining 13 bits of the VA are added, producing the
     60       1.1      gwr  * mapped physical address.
     61       1.1      gwr  *
     62       1.1      gwr  * To map the entire memory space in this manner would require 2114296 bytes
     63       1.1      gwr  * of page tables per process - quite expensive.  Instead we will
     64       1.1      gwr  * allocate a fixed but considerably smaller space for the page tables at
     65       1.1      gwr  * the time the VM system is initialized.  When the pmap code is asked by
     66       1.1      gwr  * the kernel to map a VA to a PA, it allocates tables as needed from this
     67       1.1      gwr  * pool.  When there are no more tables in the pool, tables are stolen
     68       1.1      gwr  * from the oldest mapped entries in the tree.  This is only possible
     69       1.1      gwr  * because all memory mappings are stored in the kernel memory map
     70       1.1      gwr  * structures, independent of the pmap structures.  A VA which references
     71       1.1      gwr  * one of these invalidated maps will cause a page fault.  The kernel
     72       1.1      gwr  * will determine that the page fault was caused by a task using a valid
     73       1.1      gwr  * VA, but for some reason (which does not concern it), that address was
     74       1.1      gwr  * not mapped.  It will ask the pmap code to re-map the entry and then
     75       1.1      gwr  * it will resume executing the faulting task.
     76       1.1      gwr  *
     77       1.1      gwr  * In this manner the most efficient use of the page table space is
     78       1.1      gwr  * achieved.  Tasks which do not execute often will have their tables
     79       1.1      gwr  * stolen and reused by tasks which execute more frequently.  The best
     80       1.1      gwr  * size for the page table pool will probably be determined by
     81       1.1      gwr  * experimentation.
     82       1.1      gwr  *
     83       1.1      gwr  * You read all of the comments so far.  Good for you.
     84       1.1      gwr  * Now go play!
     85       1.1      gwr  */
     86       1.1      gwr 
     87       1.1      gwr /*** A Note About the 68851 Address Translation Cache
     88       1.1      gwr  * The MC68851 has a 64 entry cache, called the Address Translation Cache
     89       1.1      gwr  * or 'ATC'.  This cache stores the most recently used page descriptors
     90       1.1      gwr  * accessed by the MMU when it does translations.  Using a marker called a
     91       1.1      gwr  * 'task alias' the MMU can store the descriptors from 8 different table
     92       1.1      gwr  * spaces concurrently.  The task alias is associated with the base
     93       1.1      gwr  * address of the level A table of that address space.  When an address
     94       1.1      gwr  * space is currently active (the CRP currently points to its A table)
     95       1.1      gwr  * the only cached descriptors that will be obeyed are ones which have a
     96       1.1      gwr  * matching task alias of the current space associated with them.
     97       1.1      gwr  *
     98       1.1      gwr  * Since the cache is always consulted before any table lookups are done,
     99       1.1      gwr  * it is important that it accurately reflect the state of the MMU tables.
    100       1.1      gwr  * Whenever a change has been made to a table that has been loaded into
    101       1.1      gwr  * the MMU, the code must be sure to flush any cached entries that are
    102       1.1      gwr  * affected by the change.  These instances are documented in the code at
    103       1.1      gwr  * various points.
    104       1.1      gwr  */
    105       1.1      gwr /*** A Note About the Note About the 68851 Address Translation Cache
    106       1.1      gwr  * 4 months into this code I discovered that the sun3x does not have
    107       1.1      gwr  * a MC68851 chip. Instead, it has a version of this MMU that is part of the
    108       1.1      gwr  * the 68030 CPU.
    109       1.1      gwr  * All though it behaves very similarly to the 68851, it only has 1 task
    110       1.8      gwr  * alias and a 22 entry cache.  So sadly (or happily), the first paragraph
    111       1.8      gwr  * of the previous note does not apply to the sun3x pmap.
    112       1.1      gwr  */
    113      1.83    lukem 
    114      1.83    lukem #include <sys/cdefs.h>
    115  1.88.2.3     yamt __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.88.2.3 2007/02/26 09:08:37 yamt Exp $");
    116      1.45      gwr 
    117      1.45      gwr #include "opt_ddb.h"
    118      1.82   martin #include "opt_pmap_debug.h"
    119       1.1      gwr 
    120       1.1      gwr #include <sys/param.h>
    121       1.1      gwr #include <sys/systm.h>
    122       1.1      gwr #include <sys/proc.h>
    123       1.1      gwr #include <sys/malloc.h>
    124      1.56  tsutsui #include <sys/pool.h>
    125       1.1      gwr #include <sys/user.h>
    126       1.1      gwr #include <sys/queue.h>
    127      1.20  thorpej #include <sys/kcore.h>
    128      1.38      gwr 
    129      1.38      gwr #include <uvm/uvm.h>
    130      1.43      mrg 
    131       1.1      gwr #include <machine/cpu.h>
    132      1.17      gwr #include <machine/kcore.h>
    133      1.33      gwr #include <machine/mon.h>
    134       1.1      gwr #include <machine/pmap.h>
    135       1.1      gwr #include <machine/pte.h>
    136      1.37      gwr #include <machine/vmparam.h>
    137      1.75      chs #include <m68k/cacheops.h>
    138      1.33      gwr 
    139      1.33      gwr #include <sun3/sun3/cache.h>
    140      1.33      gwr #include <sun3/sun3/machdep.h>
    141       1.1      gwr 
    142       1.1      gwr #include "pmap_pvt.h"
    143       1.1      gwr 
    144       1.1      gwr /* XXX - What headers declare these? */
    145       1.1      gwr extern struct pcb *curpcb;
    146       1.1      gwr extern int physmem;
    147       1.7      gwr 
    148       1.1      gwr /* Defined in locore.s */
    149       1.1      gwr extern char kernel_text[];
    150       1.1      gwr 
    151       1.1      gwr /* Defined by the linker */
    152       1.1      gwr extern char etext[], edata[], end[];
    153       1.1      gwr extern char *esym;	/* DDB */
    154       1.1      gwr 
    155       1.7      gwr /*************************** DEBUGGING DEFINITIONS ***********************
    156       1.7      gwr  * Macros, preprocessor defines and variables used in debugging can make *
    157       1.7      gwr  * code hard to read.  Anything used exclusively for debugging purposes  *
    158       1.7      gwr  * is defined here to avoid having such mess scattered around the file.  *
    159       1.7      gwr  *************************************************************************/
    160       1.8      gwr #ifdef	PMAP_DEBUG
    161       1.7      gwr /*
    162       1.7      gwr  * To aid the debugging process, macros should be expanded into smaller steps
    163       1.7      gwr  * that accomplish the same goal, yet provide convenient places for placing
    164       1.8      gwr  * breakpoints.  When this code is compiled with PMAP_DEBUG mode defined, the
    165       1.7      gwr  * 'INLINE' keyword is defined to an empty string.  This way, any function
    166       1.7      gwr  * defined to be a 'static INLINE' will become 'outlined' and compiled as
    167       1.7      gwr  * a separate function, which is much easier to debug.
    168       1.7      gwr  */
    169       1.7      gwr #define	INLINE	/* nothing */
    170       1.7      gwr 
    171       1.1      gwr /*
    172       1.7      gwr  * It is sometimes convenient to watch the activity of a particular table
    173       1.7      gwr  * in the system.  The following variables are used for that purpose.
    174       1.1      gwr  */
    175       1.7      gwr a_tmgr_t *pmap_watch_atbl = 0;
    176       1.7      gwr b_tmgr_t *pmap_watch_btbl = 0;
    177       1.7      gwr c_tmgr_t *pmap_watch_ctbl = 0;
    178       1.1      gwr 
    179       1.7      gwr int pmap_debug = 0;
    180       1.7      gwr #define DPRINT(args) if (pmap_debug) printf args
    181       1.7      gwr 
    182       1.7      gwr #else	/********** Stuff below is defined if NOT debugging **************/
    183       1.7      gwr 
    184       1.7      gwr #define	INLINE	inline
    185      1.10   jeremy #define DPRINT(args)  /* nada */
    186       1.7      gwr 
    187      1.10   jeremy #endif	/* PMAP_DEBUG */
    188       1.7      gwr /*********************** END OF DEBUGGING DEFINITIONS ********************/
    189       1.1      gwr 
    190       1.1      gwr /*** Management Structure - Memory Layout
    191       1.1      gwr  * For every MMU table in the sun3x pmap system there must be a way to
    192       1.1      gwr  * manage it; we must know which process is using it, what other tables
    193       1.1      gwr  * depend on it, and whether or not it contains any locked pages.  This
    194       1.1      gwr  * is solved by the creation of 'table management'  or 'tmgr'
    195       1.1      gwr  * structures.  One for each MMU table in the system.
    196       1.1      gwr  *
    197       1.1      gwr  *                        MAP OF MEMORY USED BY THE PMAP SYSTEM
    198       1.1      gwr  *
    199       1.1      gwr  *      towards lower memory
    200       1.1      gwr  * kernAbase -> +-------------------------------------------------------+
    201       1.1      gwr  *              | Kernel     MMU A level table                          |
    202       1.1      gwr  * kernBbase -> +-------------------------------------------------------+
    203       1.1      gwr  *              | Kernel     MMU B level tables                         |
    204       1.1      gwr  * kernCbase -> +-------------------------------------------------------+
    205       1.1      gwr  *              |                                                       |
    206       1.1      gwr  *              | Kernel     MMU C level tables                         |
    207       1.1      gwr  *              |                                                       |
    208       1.7      gwr  * mmuCbase  -> +-------------------------------------------------------+
    209       1.7      gwr  *              | User       MMU C level tables                         |
    210       1.1      gwr  * mmuAbase  -> +-------------------------------------------------------+
    211       1.1      gwr  *              |                                                       |
    212       1.1      gwr  *              | User       MMU A level tables                         |
    213       1.1      gwr  *              |                                                       |
    214       1.1      gwr  * mmuBbase  -> +-------------------------------------------------------+
    215       1.1      gwr  *              | User       MMU B level tables                         |
    216       1.1      gwr  * tmgrAbase -> +-------------------------------------------------------+
    217       1.1      gwr  *              |  TMGR A level table structures                        |
    218       1.1      gwr  * tmgrBbase -> +-------------------------------------------------------+
    219       1.1      gwr  *              |  TMGR B level table structures                        |
    220       1.1      gwr  * tmgrCbase -> +-------------------------------------------------------+
    221       1.1      gwr  *              |  TMGR C level table structures                        |
    222       1.1      gwr  * pvbase    -> +-------------------------------------------------------+
    223       1.1      gwr  *              |  Physical to Virtual mapping table (list heads)       |
    224       1.1      gwr  * pvebase   -> +-------------------------------------------------------+
    225       1.1      gwr  *              |  Physical to Virtual mapping table (list elements)    |
    226       1.1      gwr  *              |                                                       |
    227       1.1      gwr  *              +-------------------------------------------------------+
    228       1.1      gwr  *      towards higher memory
    229       1.1      gwr  *
    230       1.1      gwr  * For every A table in the MMU A area, there will be a corresponding
    231       1.1      gwr  * a_tmgr structure in the TMGR A area.  The same will be true for
    232       1.1      gwr  * the B and C tables.  This arrangement will make it easy to find the
    233       1.1      gwr  * controling tmgr structure for any table in the system by use of
    234       1.1      gwr  * (relatively) simple macros.
    235       1.1      gwr  */
    236       1.7      gwr 
    237       1.7      gwr /*
    238       1.8      gwr  * Global variables for storing the base addresses for the areas
    239       1.1      gwr  * labeled above.
    240       1.1      gwr  */
    241      1.69      chs static vaddr_t  	kernAphys;
    242       1.1      gwr static mmu_long_dte_t	*kernAbase;
    243       1.1      gwr static mmu_short_dte_t	*kernBbase;
    244       1.1      gwr static mmu_short_pte_t	*kernCbase;
    245      1.15      gwr static mmu_short_pte_t	*mmuCbase;
    246      1.15      gwr static mmu_short_dte_t	*mmuBbase;
    247       1.1      gwr static mmu_long_dte_t	*mmuAbase;
    248       1.1      gwr static a_tmgr_t		*Atmgrbase;
    249       1.1      gwr static b_tmgr_t		*Btmgrbase;
    250       1.1      gwr static c_tmgr_t		*Ctmgrbase;
    251      1.15      gwr static pv_t 		*pvbase;
    252       1.1      gwr static pv_elem_t	*pvebase;
    253      1.15      gwr struct pmap 		kernel_pmap;
    254       1.1      gwr 
    255       1.8      gwr /*
    256       1.8      gwr  * This holds the CRP currently loaded into the MMU.
    257       1.8      gwr  */
    258       1.8      gwr struct mmu_rootptr kernel_crp;
    259       1.8      gwr 
    260       1.8      gwr /*
    261       1.8      gwr  * Just all around global variables.
    262       1.1      gwr  */
    263       1.1      gwr static TAILQ_HEAD(a_pool_head_struct, a_tmgr_struct) a_pool;
    264       1.1      gwr static TAILQ_HEAD(b_pool_head_struct, b_tmgr_struct) b_pool;
    265       1.1      gwr static TAILQ_HEAD(c_pool_head_struct, c_tmgr_struct) c_pool;
    266       1.7      gwr 
    267       1.7      gwr 
    268       1.7      gwr /*
    269       1.7      gwr  * Flags used to mark the safety/availability of certain operations or
    270       1.7      gwr  * resources.
    271       1.7      gwr  */
    272  1.88.2.2     yamt /* Safe to use pmap_bootstrap_alloc(). */
    273  1.88.2.3     yamt static bool bootstrap_alloc_enabled = false;
    274  1.88.2.2     yamt /* Temporary virtual pages are in use */
    275  1.88.2.2     yamt int tmp_vpages_inuse;
    276       1.1      gwr 
    277       1.1      gwr /*
    278       1.1      gwr  * XXX:  For now, retain the traditional variables that were
    279       1.1      gwr  * used in the old pmap/vm interface (without NONCONTIG).
    280       1.1      gwr  */
    281      1.81  thorpej /* Kernel virtual address space available: */
    282      1.81  thorpej vaddr_t	virtual_avail, virtual_end;
    283       1.1      gwr /* Physical address space available: */
    284      1.69      chs paddr_t	avail_start, avail_end;
    285       1.1      gwr 
    286       1.7      gwr /* This keep track of the end of the contiguously mapped range. */
    287      1.69      chs vaddr_t virtual_contig_end;
    288       1.7      gwr 
    289       1.7      gwr /* Physical address used by pmap_next_page() */
    290      1.69      chs paddr_t avail_next;
    291       1.7      gwr 
    292       1.7      gwr /* These are used by pmap_copy_page(), etc. */
    293      1.69      chs vaddr_t tmp_vpages[2];
    294       1.1      gwr 
    295      1.56  tsutsui /* memory pool for pmap structures */
    296      1.56  tsutsui struct pool	pmap_pmap_pool;
    297      1.56  tsutsui 
    298       1.7      gwr /*
    299       1.7      gwr  * The 3/80 is the only member of the sun3x family that has non-contiguous
    300       1.1      gwr  * physical memory.  Memory is divided into 4 banks which are physically
    301       1.1      gwr  * locatable on the system board.  Although the size of these banks varies
    302       1.1      gwr  * with the size of memory they contain, their base addresses are
    303       1.1      gwr  * permenently fixed.  The following structure, which describes these
    304       1.1      gwr  * banks, is initialized by pmap_bootstrap() after it reads from a similar
    305       1.1      gwr  * structure provided by the ROM Monitor.
    306       1.1      gwr  *
    307       1.1      gwr  * For the other machines in the sun3x architecture which do have contiguous
    308       1.1      gwr  * RAM, this list will have only one entry, which will describe the entire
    309       1.1      gwr  * range of available memory.
    310       1.1      gwr  */
    311      1.20  thorpej struct pmap_physmem_struct avail_mem[SUN3X_NPHYS_RAM_SEGS];
    312       1.1      gwr u_int total_phys_mem;
    313       1.1      gwr 
    314       1.7      gwr /*************************************************************************/
    315       1.7      gwr 
    316       1.7      gwr /*
    317       1.7      gwr  * XXX - Should "tune" these based on statistics.
    318       1.7      gwr  *
    319       1.7      gwr  * My first guess about the relative numbers of these needed is
    320       1.7      gwr  * based on the fact that a "typical" process will have several
    321       1.7      gwr  * pages mapped at low virtual addresses (text, data, bss), then
    322       1.7      gwr  * some mapped shared libraries, and then some stack pages mapped
    323       1.7      gwr  * near the high end of the VA space.  Each process can use only
    324       1.7      gwr  * one A table, and most will use only two B tables (maybe three)
    325       1.7      gwr  * and probably about four C tables.  Therefore, the first guess
    326       1.7      gwr  * at the relative numbers of these needed is 1:2:4 -gwr
    327       1.7      gwr  *
    328       1.7      gwr  * The number of C tables needed is closely related to the amount
    329       1.7      gwr  * of physical memory available plus a certain amount attributable
    330       1.7      gwr  * to the use of double mappings.  With a few simulation statistics
    331       1.7      gwr  * we can find a reasonably good estimation of this unknown value.
    332       1.7      gwr  * Armed with that and the above ratios, we have a good idea of what
    333       1.7      gwr  * is needed at each level. -j
    334       1.7      gwr  *
    335       1.7      gwr  * Note: It is not physical memory memory size, but the total mapped
    336       1.7      gwr  * virtual space required by the combined working sets of all the
    337       1.7      gwr  * currently _runnable_ processes.  (Sleeping ones don't count.)
    338       1.7      gwr  * The amount of physical memory should be irrelevant. -gwr
    339       1.7      gwr  */
    340      1.22   jeremy #ifdef	FIXED_NTABLES
    341       1.7      gwr #define NUM_A_TABLES	16
    342       1.7      gwr #define NUM_B_TABLES	32
    343       1.7      gwr #define NUM_C_TABLES	64
    344      1.22   jeremy #else
    345      1.22   jeremy unsigned int	NUM_A_TABLES, NUM_B_TABLES, NUM_C_TABLES;
    346      1.22   jeremy #endif	/* FIXED_NTABLES */
    347       1.7      gwr 
    348       1.7      gwr /*
    349       1.7      gwr  * This determines our total virtual mapping capacity.
    350       1.7      gwr  * Yes, it is a FIXED value so we can pre-allocate.
    351       1.7      gwr  */
    352       1.7      gwr #define NUM_USER_PTES	(NUM_C_TABLES * MMU_C_TBL_SIZE)
    353      1.15      gwr 
    354      1.15      gwr /*
    355      1.15      gwr  * The size of the Kernel Virtual Address Space (KVAS)
    356      1.15      gwr  * for purposes of MMU table allocation is -KERNBASE
    357      1.15      gwr  * (length from KERNBASE to 0xFFFFffff)
    358      1.15      gwr  */
    359      1.15      gwr #define	KVAS_SIZE		(-KERNBASE)
    360      1.15      gwr 
    361      1.15      gwr /* Numbers of kernel MMU tables to support KVAS_SIZE. */
    362      1.15      gwr #define KERN_B_TABLES	(KVAS_SIZE >> MMU_TIA_SHIFT)
    363      1.15      gwr #define KERN_C_TABLES	(KVAS_SIZE >> MMU_TIB_SHIFT)
    364      1.15      gwr #define	NUM_KERN_PTES	(KVAS_SIZE >> MMU_TIC_SHIFT)
    365       1.7      gwr 
    366       1.7      gwr /*************************** MISCELANEOUS MACROS *************************/
    367      1.55  tsutsui #define pmap_lock(pmap) simple_lock(&pmap->pm_lock)
    368      1.55  tsutsui #define pmap_unlock(pmap) simple_unlock(&pmap->pm_lock)
    369      1.55  tsutsui #define pmap_add_ref(pmap) ++pmap->pm_refcount
    370      1.55  tsutsui #define pmap_del_ref(pmap) --pmap->pm_refcount
    371      1.55  tsutsui #define pmap_refcount(pmap) pmap->pm_refcount
    372      1.64  thorpej 
    373      1.64  thorpej void *pmap_bootstrap_alloc(int);
    374       1.7      gwr 
    375      1.86      chs static INLINE void *mmu_ptov(paddr_t);
    376      1.86      chs static INLINE paddr_t mmu_vtop(void *);
    377       1.7      gwr 
    378       1.7      gwr #if	0
    379  1.88.2.2     yamt static INLINE a_tmgr_t *mmuA2tmgr(mmu_long_dte_t *);
    380      1.26   jeremy #endif /* 0 */
    381  1.88.2.2     yamt static INLINE b_tmgr_t *mmuB2tmgr(mmu_short_dte_t *);
    382  1.88.2.2     yamt static INLINE c_tmgr_t *mmuC2tmgr(mmu_short_pte_t *);
    383       1.7      gwr 
    384      1.86      chs static INLINE pv_t *pa2pv(paddr_t);
    385      1.86      chs static INLINE int   pteidx(mmu_short_pte_t *);
    386      1.86      chs static INLINE pmap_t current_pmap(void);
    387       1.7      gwr 
    388       1.7      gwr /*
    389       1.7      gwr  * We can always convert between virtual and physical addresses
    390       1.7      gwr  * for anything in the range [KERNBASE ... avail_start] because
    391       1.7      gwr  * that range is GUARANTEED to be mapped linearly.
    392       1.7      gwr  * We rely heavily upon this feature!
    393       1.7      gwr  */
    394       1.7      gwr static INLINE void *
    395      1.86      chs mmu_ptov(paddr_t pa)
    396       1.7      gwr {
    397      1.69      chs 	vaddr_t va;
    398       1.7      gwr 
    399       1.7      gwr 	va = (pa + KERNBASE);
    400       1.8      gwr #ifdef	PMAP_DEBUG
    401       1.7      gwr 	if ((va < KERNBASE) || (va >= virtual_contig_end))
    402       1.7      gwr 		panic("mmu_ptov");
    403       1.7      gwr #endif
    404  1.88.2.2     yamt 	return (void *)va;
    405       1.7      gwr }
    406      1.69      chs 
    407      1.86      chs static INLINE paddr_t
    408      1.86      chs mmu_vtop(void *vva)
    409       1.7      gwr {
    410      1.69      chs 	vaddr_t va;
    411       1.7      gwr 
    412      1.69      chs 	va = (vaddr_t)vva;
    413       1.8      gwr #ifdef	PMAP_DEBUG
    414       1.7      gwr 	if ((va < KERNBASE) || (va >= virtual_contig_end))
    415      1.72  tsutsui 		panic("mmu_vtop");
    416       1.7      gwr #endif
    417  1.88.2.2     yamt 	return va - KERNBASE;
    418       1.7      gwr }
    419       1.7      gwr 
    420       1.7      gwr /*
    421       1.7      gwr  * These macros map MMU tables to their corresponding manager structures.
    422       1.1      gwr  * They are needed quite often because many of the pointers in the pmap
    423       1.1      gwr  * system reference MMU tables and not the structures that control them.
    424       1.1      gwr  * There needs to be a way to find one when given the other and these
    425       1.1      gwr  * macros do so by taking advantage of the memory layout described above.
    426       1.1      gwr  * Here's a quick step through the first macro, mmuA2tmgr():
    427       1.1      gwr  *
    428       1.1      gwr  * 1) find the offset of the given MMU A table from the base of its table
    429       1.1      gwr  *    pool (table - mmuAbase).
    430       1.1      gwr  * 2) convert this offset into a table index by dividing it by the
    431       1.1      gwr  *    size of one MMU 'A' table. (sizeof(mmu_long_dte_t) * MMU_A_TBL_SIZE)
    432       1.1      gwr  * 3) use this index to select the corresponding 'A' table manager
    433       1.1      gwr  *    structure from the 'A' table manager pool (Atmgrbase[index]).
    434       1.1      gwr  */
    435       1.7      gwr /*  This function is not currently used. */
    436       1.7      gwr #if	0
    437       1.7      gwr static INLINE a_tmgr_t *
    438      1.86      chs mmuA2tmgr(mmu_long_dte_t *mmuAtbl)
    439       1.7      gwr {
    440      1.69      chs 	int idx;
    441       1.7      gwr 
    442       1.7      gwr 	/* Which table is this in? */
    443       1.7      gwr 	idx = (mmuAtbl - mmuAbase) / MMU_A_TBL_SIZE;
    444       1.8      gwr #ifdef	PMAP_DEBUG
    445       1.7      gwr 	if ((idx < 0) || (idx >= NUM_A_TABLES))
    446       1.7      gwr 		panic("mmuA2tmgr");
    447       1.7      gwr #endif
    448  1.88.2.2     yamt 	return &Atmgrbase[idx];
    449       1.7      gwr }
    450       1.7      gwr #endif	/* 0 */
    451       1.7      gwr 
    452       1.7      gwr static INLINE b_tmgr_t *
    453      1.86      chs mmuB2tmgr(mmu_short_dte_t *mmuBtbl)
    454       1.7      gwr {
    455      1.69      chs 	int idx;
    456       1.7      gwr 
    457       1.7      gwr 	/* Which table is this in? */
    458       1.7      gwr 	idx = (mmuBtbl - mmuBbase) / MMU_B_TBL_SIZE;
    459       1.8      gwr #ifdef	PMAP_DEBUG
    460       1.7      gwr 	if ((idx < 0) || (idx >= NUM_B_TABLES))
    461       1.7      gwr 		panic("mmuB2tmgr");
    462       1.7      gwr #endif
    463  1.88.2.2     yamt 	return &Btmgrbase[idx];
    464       1.7      gwr }
    465       1.7      gwr 
    466       1.7      gwr /* mmuC2tmgr			INTERNAL
    467       1.7      gwr  **
    468       1.7      gwr  * Given a pte known to belong to a C table, return the address of
    469       1.7      gwr  * that table's management structure.
    470       1.7      gwr  */
    471       1.7      gwr static INLINE c_tmgr_t *
    472      1.86      chs mmuC2tmgr(mmu_short_pte_t *mmuCtbl)
    473       1.7      gwr {
    474      1.69      chs 	int idx;
    475       1.7      gwr 
    476       1.7      gwr 	/* Which table is this in? */
    477       1.7      gwr 	idx = (mmuCtbl - mmuCbase) / MMU_C_TBL_SIZE;
    478       1.8      gwr #ifdef	PMAP_DEBUG
    479       1.7      gwr 	if ((idx < 0) || (idx >= NUM_C_TABLES))
    480       1.7      gwr 		panic("mmuC2tmgr");
    481       1.7      gwr #endif
    482  1.88.2.2     yamt 	return &Ctmgrbase[idx];
    483       1.7      gwr }
    484       1.7      gwr 
    485       1.8      gwr /* This is now a function call below.
    486       1.1      gwr  * #define pa2pv(pa) \
    487       1.1      gwr  *	(&pvbase[(unsigned long)\
    488      1.25    veego  *		m68k_btop(pa)\
    489       1.1      gwr  *	])
    490       1.1      gwr  */
    491       1.1      gwr 
    492       1.7      gwr /* pa2pv			INTERNAL
    493       1.7      gwr  **
    494       1.7      gwr  * Return the pv_list_head element which manages the given physical
    495       1.7      gwr  * address.
    496       1.7      gwr  */
    497       1.7      gwr static INLINE pv_t *
    498      1.86      chs pa2pv(paddr_t pa)
    499       1.7      gwr {
    500      1.69      chs 	struct pmap_physmem_struct *bank;
    501      1.69      chs 	int idx;
    502       1.7      gwr 
    503       1.7      gwr 	bank = &avail_mem[0];
    504       1.7      gwr 	while (pa >= bank->pmem_end)
    505       1.7      gwr 		bank = bank->pmem_next;
    506       1.7      gwr 
    507       1.7      gwr 	pa -= bank->pmem_start;
    508      1.25    veego 	idx = bank->pmem_pvbase + m68k_btop(pa);
    509       1.8      gwr #ifdef	PMAP_DEBUG
    510       1.7      gwr 	if ((idx < 0) || (idx >= physmem))
    511       1.7      gwr 		panic("pa2pv");
    512       1.7      gwr #endif
    513       1.7      gwr 	return &pvbase[idx];
    514       1.7      gwr }
    515       1.7      gwr 
    516       1.7      gwr /* pteidx			INTERNAL
    517       1.7      gwr  **
    518       1.7      gwr  * Return the index of the given PTE within the entire fixed table of
    519       1.7      gwr  * PTEs.
    520       1.7      gwr  */
    521       1.7      gwr static INLINE int
    522      1.86      chs pteidx(mmu_short_pte_t *pte)
    523       1.7      gwr {
    524  1.88.2.2     yamt 
    525  1.88.2.2     yamt 	return pte - kernCbase;
    526       1.7      gwr }
    527       1.7      gwr 
    528       1.7      gwr /*
    529       1.8      gwr  * This just offers a place to put some debugging checks,
    530      1.76  thorpej  * and reduces the number of places "curlwp" appears...
    531       1.7      gwr  */
    532      1.86      chs static INLINE pmap_t
    533      1.86      chs current_pmap(void)
    534       1.7      gwr {
    535       1.7      gwr 	struct vmspace *vm;
    536      1.67      chs 	struct vm_map *map;
    537       1.7      gwr 	pmap_t	pmap;
    538       1.7      gwr 
    539      1.76  thorpej 	if (curlwp == NULL)
    540       1.9      gwr 		pmap = &kernel_pmap;
    541       1.9      gwr 	else {
    542      1.76  thorpej 		vm = curproc->p_vmspace;
    543       1.9      gwr 		map = &vm->vm_map;
    544       1.9      gwr 		pmap = vm_map_pmap(map);
    545       1.9      gwr 	}
    546       1.7      gwr 
    547  1.88.2.2     yamt 	return pmap;
    548       1.7      gwr }
    549       1.7      gwr 
    550       1.7      gwr 
    551       1.1      gwr /*************************** FUNCTION DEFINITIONS ************************
    552       1.1      gwr  * These appear here merely for the compiler to enforce type checking on *
    553       1.1      gwr  * all function calls.                                                   *
    554       1.7      gwr  *************************************************************************/
    555       1.1      gwr 
    556  1.88.2.2     yamt /*
    557  1.88.2.2     yamt  * Internal functions
    558  1.88.2.2     yamt  */
    559  1.88.2.2     yamt a_tmgr_t *get_a_table(void);
    560  1.88.2.2     yamt b_tmgr_t *get_b_table(void);
    561  1.88.2.2     yamt c_tmgr_t *get_c_table(void);
    562  1.88.2.3     yamt int free_a_table(a_tmgr_t *, bool);
    563  1.88.2.3     yamt int free_b_table(b_tmgr_t *, bool);
    564  1.88.2.3     yamt int free_c_table(c_tmgr_t *, bool);
    565  1.88.2.2     yamt 
    566  1.88.2.2     yamt void pmap_bootstrap_aalign(int);
    567  1.88.2.2     yamt void pmap_alloc_usermmu(void);
    568  1.88.2.2     yamt void pmap_alloc_usertmgr(void);
    569  1.88.2.2     yamt void pmap_alloc_pv(void);
    570  1.88.2.2     yamt void pmap_init_a_tables(void);
    571  1.88.2.2     yamt void pmap_init_b_tables(void);
    572  1.88.2.2     yamt void pmap_init_c_tables(void);
    573  1.88.2.2     yamt void pmap_init_pv(void);
    574  1.88.2.2     yamt void pmap_clear_pv(paddr_t, int);
    575  1.88.2.3     yamt static INLINE bool is_managed(paddr_t);
    576  1.88.2.2     yamt 
    577  1.88.2.3     yamt bool pmap_remove_a(a_tmgr_t *, vaddr_t, vaddr_t);
    578  1.88.2.3     yamt bool pmap_remove_b(b_tmgr_t *, vaddr_t, vaddr_t);
    579  1.88.2.3     yamt bool pmap_remove_c(c_tmgr_t *, vaddr_t, vaddr_t);
    580  1.88.2.2     yamt void pmap_remove_pte(mmu_short_pte_t *);
    581  1.88.2.2     yamt 
    582  1.88.2.2     yamt void pmap_enter_kernel(vaddr_t, paddr_t, vm_prot_t);
    583  1.88.2.2     yamt static INLINE void pmap_remove_kernel(vaddr_t, vaddr_t);
    584  1.88.2.2     yamt static INLINE void pmap_protect_kernel(vaddr_t, vaddr_t, vm_prot_t);
    585  1.88.2.3     yamt static INLINE bool pmap_extract_kernel(vaddr_t, paddr_t *);
    586  1.88.2.2     yamt vaddr_t pmap_get_pteinfo(u_int, pmap_t *, c_tmgr_t **);
    587  1.88.2.2     yamt static INLINE int pmap_dereference(pmap_t);
    588  1.88.2.2     yamt 
    589  1.88.2.3     yamt bool pmap_stroll(pmap_t, vaddr_t, a_tmgr_t **, b_tmgr_t **, c_tmgr_t **,
    590  1.88.2.2     yamt     mmu_short_pte_t **, int *, int *, int *);
    591  1.88.2.2     yamt void pmap_bootstrap_copyprom(void);
    592  1.88.2.2     yamt void pmap_takeover_mmu(void);
    593  1.88.2.2     yamt void pmap_bootstrap_setprom(void);
    594      1.86      chs static void pmap_page_upload(void);
    595       1.1      gwr 
    596  1.88.2.2     yamt #ifdef PMAP_DEBUG
    597  1.88.2.2     yamt /* Debugging function definitions */
    598  1.88.2.2     yamt void  pv_list(paddr_t, int);
    599  1.88.2.2     yamt #endif /* PMAP_DEBUG */
    600  1.88.2.2     yamt 
    601       1.1      gwr /** Interface functions
    602       1.1      gwr  ** - functions required by the Mach VM Pmap interface, with MACHINE_CONTIG
    603       1.1      gwr  **   defined.
    604  1.88.2.2     yamt  **   The new UVM doesn't require them so now INTERNAL.
    605       1.1      gwr  **/
    606  1.88.2.2     yamt static INLINE void pmap_pinit(pmap_t);
    607  1.88.2.2     yamt static INLINE void pmap_release(pmap_t);
    608       1.1      gwr 
    609       1.1      gwr /********************************** CODE ********************************
    610       1.1      gwr  * Functions that are called from other parts of the kernel are labeled *
    611       1.1      gwr  * as 'INTERFACE' functions.  Functions that are only called from       *
    612       1.1      gwr  * within the pmap module are labeled as 'INTERNAL' functions.          *
    613       1.1      gwr  * Functions that are internal, but are not (currently) used at all are *
    614       1.1      gwr  * labeled 'INTERNAL_X'.                                                *
    615       1.1      gwr  ************************************************************************/
    616       1.1      gwr 
    617       1.1      gwr /* pmap_bootstrap			INTERNAL
    618       1.1      gwr  **
    619      1.33      gwr  * Initializes the pmap system.  Called at boot time from
    620      1.33      gwr  * locore2.c:_vm_init()
    621       1.1      gwr  *
    622       1.1      gwr  * Reminder: having a pmap_bootstrap_alloc() and also having the VM
    623       1.1      gwr  *           system implement pmap_steal_memory() is redundant.
    624       1.1      gwr  *           Don't release this code without removing one or the other!
    625       1.1      gwr  */
    626      1.86      chs void
    627      1.86      chs pmap_bootstrap(vaddr_t nextva)
    628       1.1      gwr {
    629       1.1      gwr 	struct physmemory *membank;
    630       1.1      gwr 	struct pmap_physmem_struct *pmap_membank;
    631      1.69      chs 	vaddr_t va, eva;
    632      1.69      chs 	paddr_t pa;
    633       1.1      gwr 	int b, c, i, j;	/* running table counts */
    634      1.40      gwr 	int size, resvmem;
    635       1.1      gwr 
    636       1.1      gwr 	/*
    637       1.1      gwr 	 * This function is called by __bootstrap after it has
    638       1.1      gwr 	 * determined the type of machine and made the appropriate
    639       1.1      gwr 	 * patches to the ROM vectors (XXX- I don't quite know what I meant
    640       1.1      gwr 	 * by that.)  It allocates and sets up enough of the pmap system
    641       1.1      gwr 	 * to manage the kernel's address space.
    642       1.1      gwr 	 */
    643       1.1      gwr 
    644       1.1      gwr 	/*
    645       1.7      gwr 	 * Determine the range of kernel virtual and physical
    646       1.7      gwr 	 * space available. Note that we ABSOLUTELY DEPEND on
    647       1.7      gwr 	 * the fact that the first bank of memory (4MB) is
    648       1.7      gwr 	 * mapped linearly to KERNBASE (which we guaranteed in
    649       1.7      gwr 	 * the first instructions of locore.s).
    650       1.7      gwr 	 * That is plenty for our bootstrap work.
    651       1.1      gwr 	 */
    652      1.25    veego 	virtual_avail = m68k_round_page(nextva);
    653       1.7      gwr 	virtual_contig_end = KERNBASE + 0x400000; /* +4MB */
    654       1.1      gwr 	virtual_end = VM_MAX_KERNEL_ADDRESS;
    655       1.7      gwr 	/* Don't need avail_start til later. */
    656       1.1      gwr 
    657       1.7      gwr 	/* We may now call pmap_bootstrap_alloc(). */
    658  1.88.2.3     yamt 	bootstrap_alloc_enabled = true;
    659       1.1      gwr 
    660       1.1      gwr 	/*
    661       1.1      gwr 	 * This is a somewhat unwrapped loop to deal with
    662       1.1      gwr 	 * copying the PROM's 'phsymem' banks into the pmap's
    663       1.1      gwr 	 * banks.  The following is always assumed:
    664       1.1      gwr 	 * 1. There is always at least one bank of memory.
    665       1.1      gwr 	 * 2. There is always a last bank of memory, and its
    666       1.1      gwr 	 *    pmem_next member must be set to NULL.
    667       1.1      gwr 	 */
    668       1.1      gwr 	membank = romVectorPtr->v_physmemory;
    669       1.1      gwr 	pmap_membank = avail_mem;
    670       1.1      gwr 	total_phys_mem = 0;
    671       1.1      gwr 
    672      1.40      gwr 	for (;;) { /* break on !membank */
    673       1.1      gwr 		pmap_membank->pmem_start = membank->address;
    674       1.1      gwr 		pmap_membank->pmem_end = membank->address + membank->size;
    675       1.1      gwr 		total_phys_mem += membank->size;
    676      1.40      gwr 		membank = membank->next;
    677      1.40      gwr 		if (!membank)
    678      1.40      gwr 			break;
    679       1.1      gwr 		/* This silly syntax arises because pmap_membank
    680       1.1      gwr 		 * is really a pre-allocated array, but it is put into
    681       1.1      gwr 		 * use as a linked list.
    682       1.1      gwr 		 */
    683       1.1      gwr 		pmap_membank->pmem_next = pmap_membank + 1;
    684       1.1      gwr 		pmap_membank = pmap_membank->pmem_next;
    685       1.1      gwr 	}
    686      1.40      gwr 	/* This is the last element. */
    687      1.40      gwr 	pmap_membank->pmem_next = NULL;
    688       1.1      gwr 
    689       1.1      gwr 	/*
    690      1.40      gwr 	 * Note: total_phys_mem, physmem represent
    691      1.40      gwr 	 * actual physical memory, including that
    692      1.40      gwr 	 * reserved for the PROM monitor.
    693       1.1      gwr 	 */
    694      1.40      gwr 	physmem = btoc(total_phys_mem);
    695       1.1      gwr 
    696       1.1      gwr 	/*
    697      1.60  tsutsui 	 * Avail_end is set to the first byte of physical memory
    698      1.60  tsutsui 	 * after the end of the last bank.  We use this only to
    699      1.60  tsutsui 	 * determine if a physical address is "managed" memory.
    700      1.60  tsutsui 	 * This address range should be reduced to prevent the
    701      1.40      gwr 	 * physical pages needed by the PROM monitor from being used
    702      1.40      gwr 	 * in the VM system.
    703       1.1      gwr 	 */
    704      1.40      gwr 	resvmem = total_phys_mem - *(romVectorPtr->memoryAvail);
    705      1.40      gwr 	resvmem = m68k_round_page(resvmem);
    706      1.60  tsutsui 	avail_end = pmap_membank->pmem_end - resvmem;
    707       1.1      gwr 
    708       1.1      gwr 	/*
    709      1.15      gwr 	 * First allocate enough kernel MMU tables to map all
    710      1.15      gwr 	 * of kernel virtual space from KERNBASE to 0xFFFFFFFF.
    711       1.1      gwr 	 * Note: All must be aligned on 256 byte boundaries.
    712      1.15      gwr 	 * Start with the level-A table (one of those).
    713       1.1      gwr 	 */
    714      1.69      chs 	size = sizeof(mmu_long_dte_t) * MMU_A_TBL_SIZE;
    715       1.7      gwr 	kernAbase = pmap_bootstrap_alloc(size);
    716      1.71  tsutsui 	memset(kernAbase, 0, size);
    717       1.1      gwr 
    718      1.15      gwr 	/* Now the level-B kernel tables... */
    719      1.15      gwr 	size = sizeof(mmu_short_dte_t) * MMU_B_TBL_SIZE * KERN_B_TABLES;
    720       1.7      gwr 	kernBbase = pmap_bootstrap_alloc(size);
    721      1.71  tsutsui 	memset(kernBbase, 0, size);
    722       1.1      gwr 
    723      1.15      gwr 	/* Now the level-C kernel tables... */
    724      1.15      gwr 	size = sizeof(mmu_short_pte_t) * MMU_C_TBL_SIZE * KERN_C_TABLES;
    725      1.15      gwr 	kernCbase = pmap_bootstrap_alloc(size);
    726      1.71  tsutsui 	memset(kernCbase, 0, size);
    727       1.7      gwr 	/*
    728       1.7      gwr 	 * Note: In order for the PV system to work correctly, the kernel
    729       1.7      gwr 	 * and user-level C tables must be allocated contiguously.
    730       1.7      gwr 	 * Nothing should be allocated between here and the allocation of
    731       1.7      gwr 	 * mmuCbase below.  XXX: Should do this as one allocation, and
    732       1.7      gwr 	 * then compute a pointer for mmuCbase instead of this...
    733      1.15      gwr 	 *
    734      1.15      gwr 	 * Allocate user MMU tables.
    735      1.70      wiz 	 * These must be contiguous with the preceding.
    736       1.7      gwr 	 */
    737      1.22   jeremy 
    738      1.22   jeremy #ifndef	FIXED_NTABLES
    739      1.22   jeremy 	/*
    740      1.22   jeremy 	 * The number of user-level C tables that should be allocated is
    741      1.22   jeremy 	 * related to the size of physical memory.  In general, there should
    742      1.22   jeremy 	 * be enough tables to map four times the amount of available RAM.
    743      1.22   jeremy 	 * The extra amount is needed because some table space is wasted by
    744      1.22   jeremy 	 * fragmentation.
    745      1.22   jeremy 	 */
    746      1.22   jeremy 	NUM_C_TABLES = (total_phys_mem * 4) / (MMU_C_TBL_SIZE * MMU_PAGE_SIZE);
    747      1.22   jeremy 	NUM_B_TABLES = NUM_C_TABLES / 2;
    748      1.22   jeremy 	NUM_A_TABLES = NUM_B_TABLES / 2;
    749      1.22   jeremy #endif	/* !FIXED_NTABLES */
    750      1.22   jeremy 
    751      1.15      gwr 	size = sizeof(mmu_short_pte_t) * MMU_C_TBL_SIZE	* NUM_C_TABLES;
    752      1.15      gwr 	mmuCbase = pmap_bootstrap_alloc(size);
    753      1.15      gwr 
    754      1.15      gwr 	size = sizeof(mmu_short_dte_t) * MMU_B_TBL_SIZE	* NUM_B_TABLES;
    755      1.15      gwr 	mmuBbase = pmap_bootstrap_alloc(size);
    756       1.1      gwr 
    757      1.69      chs 	size = sizeof(mmu_long_dte_t) * MMU_A_TBL_SIZE * NUM_A_TABLES;
    758      1.15      gwr 	mmuAbase = pmap_bootstrap_alloc(size);
    759       1.7      gwr 
    760       1.7      gwr 	/*
    761       1.7      gwr 	 * Fill in the never-changing part of the kernel tables.
    762       1.7      gwr 	 * For simplicity, the kernel's mappings will be editable as a
    763       1.1      gwr 	 * flat array of page table entries at kernCbase.  The
    764       1.1      gwr 	 * higher level 'A' and 'B' tables must be initialized to point
    765       1.1      gwr 	 * to this lower one.
    766       1.1      gwr 	 */
    767       1.1      gwr 	b = c = 0;
    768       1.1      gwr 
    769       1.7      gwr 	/*
    770       1.7      gwr 	 * Invalidate all mappings below KERNBASE in the A table.
    771       1.1      gwr 	 * This area has already been zeroed out, but it is good
    772       1.1      gwr 	 * practice to explicitly show that we are interpreting
    773       1.1      gwr 	 * it as a list of A table descriptors.
    774       1.1      gwr 	 */
    775       1.1      gwr 	for (i = 0; i < MMU_TIA(KERNBASE); i++) {
    776       1.1      gwr 		kernAbase[i].addr.raw = 0;
    777       1.1      gwr 	}
    778       1.1      gwr 
    779       1.7      gwr 	/*
    780       1.7      gwr 	 * Set up the kernel A and B tables so that they will reference the
    781       1.1      gwr 	 * correct spots in the contiguous table of PTEs allocated for the
    782       1.1      gwr 	 * kernel's virtual memory space.
    783       1.1      gwr 	 */
    784       1.1      gwr 	for (i = MMU_TIA(KERNBASE); i < MMU_A_TBL_SIZE; i++) {
    785       1.1      gwr 		kernAbase[i].attr.raw =
    786  1.88.2.2     yamt 		    MMU_LONG_DTE_LU | MMU_LONG_DTE_SUPV | MMU_DT_SHORT;
    787       1.7      gwr 		kernAbase[i].addr.raw = mmu_vtop(&kernBbase[b]);
    788       1.1      gwr 
    789  1.88.2.2     yamt 		for (j = 0; j < MMU_B_TBL_SIZE; j++) {
    790  1.88.2.2     yamt 			kernBbase[b + j].attr.raw =
    791  1.88.2.2     yamt 			    mmu_vtop(&kernCbase[c]) | MMU_DT_SHORT;
    792       1.1      gwr 			c += MMU_C_TBL_SIZE;
    793       1.1      gwr 		}
    794       1.1      gwr 		b += MMU_B_TBL_SIZE;
    795       1.1      gwr 	}
    796       1.1      gwr 
    797       1.7      gwr 	pmap_alloc_usermmu();	/* Allocate user MMU tables.        */
    798       1.7      gwr 	pmap_alloc_usertmgr();	/* Allocate user MMU table managers.*/
    799       1.7      gwr 	pmap_alloc_pv();	/* Allocate physical->virtual map.  */
    800       1.7      gwr 
    801       1.7      gwr 	/*
    802       1.7      gwr 	 * We are now done with pmap_bootstrap_alloc().  Round up
    803       1.7      gwr 	 * `virtual_avail' to the nearest page, and set the flag
    804       1.7      gwr 	 * to prevent use of pmap_bootstrap_alloc() hereafter.
    805       1.7      gwr 	 */
    806      1.79  thorpej 	pmap_bootstrap_aalign(PAGE_SIZE);
    807  1.88.2.3     yamt 	bootstrap_alloc_enabled = false;
    808       1.7      gwr 
    809       1.7      gwr 	/*
    810       1.7      gwr 	 * Now that we are done with pmap_bootstrap_alloc(), we
    811       1.7      gwr 	 * must save the virtual and physical addresses of the
    812       1.7      gwr 	 * end of the linearly mapped range, which are stored in
    813       1.7      gwr 	 * virtual_contig_end and avail_start, respectively.
    814       1.7      gwr 	 * These variables will never change after this point.
    815       1.7      gwr 	 */
    816       1.7      gwr 	virtual_contig_end = virtual_avail;
    817       1.7      gwr 	avail_start = virtual_avail - KERNBASE;
    818       1.7      gwr 
    819       1.7      gwr 	/*
    820       1.7      gwr 	 * `avail_next' is a running pointer used by pmap_next_page() to
    821       1.7      gwr 	 * keep track of the next available physical page to be handed
    822       1.7      gwr 	 * to the VM system during its initialization, in which it
    823       1.7      gwr 	 * asks for physical pages, one at a time.
    824       1.7      gwr 	 */
    825       1.7      gwr 	avail_next = avail_start;
    826       1.7      gwr 
    827       1.7      gwr 	/*
    828       1.7      gwr 	 * Now allocate some virtual addresses, but not the physical pages
    829       1.7      gwr 	 * behind them.  Note that virtual_avail is already page-aligned.
    830       1.7      gwr 	 *
    831       1.7      gwr 	 * tmp_vpages[] is an array of two virtual pages used for temporary
    832       1.7      gwr 	 * kernel mappings in the pmap module to facilitate various physical
    833       1.7      gwr 	 * address-oritented operations.
    834       1.7      gwr 	 */
    835       1.7      gwr 	tmp_vpages[0] = virtual_avail;
    836      1.79  thorpej 	virtual_avail += PAGE_SIZE;
    837       1.7      gwr 	tmp_vpages[1] = virtual_avail;
    838      1.79  thorpej 	virtual_avail += PAGE_SIZE;
    839       1.7      gwr 
    840       1.7      gwr 	/** Initialize the PV system **/
    841       1.7      gwr 	pmap_init_pv();
    842       1.7      gwr 
    843       1.7      gwr 	/*
    844       1.7      gwr 	 * Fill in the kernel_pmap structure and kernel_crp.
    845       1.7      gwr 	 */
    846       1.7      gwr 	kernAphys = mmu_vtop(kernAbase);
    847       1.7      gwr 	kernel_pmap.pm_a_tmgr = NULL;
    848       1.7      gwr 	kernel_pmap.pm_a_phys = kernAphys;
    849       1.7      gwr 	kernel_pmap.pm_refcount = 1; /* always in use */
    850      1.55  tsutsui 	simple_lock_init(&kernel_pmap.pm_lock);
    851       1.7      gwr 
    852       1.7      gwr 	kernel_crp.rp_attr = MMU_LONG_DTE_LU | MMU_DT_LONG;
    853       1.7      gwr 	kernel_crp.rp_addr = kernAphys;
    854       1.7      gwr 
    855       1.1      gwr 	/*
    856       1.1      gwr 	 * Now pmap_enter_kernel() may be used safely and will be
    857       1.7      gwr 	 * the main interface used hereafter to modify the kernel's
    858       1.7      gwr 	 * virtual address space.  Note that since we are still running
    859       1.7      gwr 	 * under the PROM's address table, none of these table modifications
    860       1.7      gwr 	 * actually take effect until pmap_takeover_mmu() is called.
    861       1.1      gwr 	 *
    862       1.7      gwr 	 * Note: Our tables do NOT have the PROM linear mappings!
    863       1.7      gwr 	 * Only the mappings created here exist in our tables, so
    864       1.7      gwr 	 * remember to map anything we expect to use.
    865       1.1      gwr 	 */
    866      1.69      chs 	va = (vaddr_t)KERNBASE;
    867       1.7      gwr 	pa = 0;
    868       1.1      gwr 
    869       1.1      gwr 	/*
    870       1.7      gwr 	 * The first page of the kernel virtual address space is the msgbuf
    871       1.7      gwr 	 * page.  The page attributes (data, non-cached) are set here, while
    872       1.7      gwr 	 * the address is assigned to this global pointer in cpu_startup().
    873      1.29      gwr 	 * It is non-cached, mostly due to paranoia.
    874       1.1      gwr 	 */
    875      1.29      gwr 	pmap_enter_kernel(va, pa|PMAP_NC, VM_PROT_ALL);
    876  1.88.2.2     yamt 	va += PAGE_SIZE;
    877  1.88.2.2     yamt 	pa += PAGE_SIZE;
    878       1.1      gwr 
    879       1.7      gwr 	/* Next page is used as the temporary stack. */
    880       1.1      gwr 	pmap_enter_kernel(va, pa, VM_PROT_ALL);
    881  1.88.2.2     yamt 	va += PAGE_SIZE;
    882  1.88.2.2     yamt 	pa += PAGE_SIZE;
    883       1.1      gwr 
    884       1.1      gwr 	/*
    885       1.1      gwr 	 * Map all of the kernel's text segment as read-only and cacheable.
    886       1.1      gwr 	 * (Cacheable is implied by default).  Unfortunately, the last bytes
    887       1.1      gwr 	 * of kernel text and the first bytes of kernel data will often be
    888       1.1      gwr 	 * sharing the same page.  Therefore, the last page of kernel text
    889  1.88.2.2     yamt 	 * has to be mapped as read/write, to accommodate the data.
    890       1.1      gwr 	 */
    891      1.69      chs 	eva = m68k_trunc_page((vaddr_t)etext);
    892      1.79  thorpej 	for (; va < eva; va += PAGE_SIZE, pa += PAGE_SIZE)
    893       1.1      gwr 		pmap_enter_kernel(va, pa, VM_PROT_READ|VM_PROT_EXECUTE);
    894       1.1      gwr 
    895       1.7      gwr 	/*
    896       1.7      gwr 	 * Map all of the kernel's data as read/write and cacheable.
    897       1.7      gwr 	 * This includes: data, BSS, symbols, and everything in the
    898       1.7      gwr 	 * contiguous memory used by pmap_bootstrap_alloc()
    899       1.1      gwr 	 */
    900      1.79  thorpej 	for (; pa < avail_start; va += PAGE_SIZE, pa += PAGE_SIZE)
    901       1.1      gwr 		pmap_enter_kernel(va, pa, VM_PROT_READ|VM_PROT_WRITE);
    902       1.1      gwr 
    903       1.7      gwr 	/*
    904       1.7      gwr 	 * At this point we are almost ready to take over the MMU.  But first
    905       1.7      gwr 	 * we must save the PROM's address space in our map, as we call its
    906       1.7      gwr 	 * routines and make references to its data later in the kernel.
    907       1.1      gwr 	 */
    908       1.7      gwr 	pmap_bootstrap_copyprom();
    909       1.7      gwr 	pmap_takeover_mmu();
    910      1.13      gwr 	pmap_bootstrap_setprom();
    911       1.1      gwr 
    912       1.1      gwr 	/* Notify the VM system of our page size. */
    913      1.79  thorpej 	uvmexp.pagesize = PAGE_SIZE;
    914      1.43      mrg 	uvm_setpagesize();
    915      1.37      gwr 
    916      1.37      gwr 	pmap_page_upload();
    917       1.1      gwr }
    918       1.1      gwr 
    919       1.1      gwr 
    920       1.1      gwr /* pmap_alloc_usermmu			INTERNAL
    921       1.1      gwr  **
    922       1.1      gwr  * Called from pmap_bootstrap() to allocate MMU tables that will
    923       1.1      gwr  * eventually be used for user mappings.
    924       1.1      gwr  */
    925      1.86      chs void
    926      1.86      chs pmap_alloc_usermmu(void)
    927       1.1      gwr {
    928  1.88.2.2     yamt 
    929       1.7      gwr 	/* XXX: Moved into caller. */
    930       1.1      gwr }
    931       1.1      gwr 
    932       1.1      gwr /* pmap_alloc_pv			INTERNAL
    933       1.1      gwr  **
    934       1.1      gwr  * Called from pmap_bootstrap() to allocate the physical
    935       1.1      gwr  * to virtual mapping list.  Each physical page of memory
    936       1.1      gwr  * in the system has a corresponding element in this list.
    937       1.1      gwr  */
    938      1.86      chs void
    939      1.86      chs pmap_alloc_pv(void)
    940       1.1      gwr {
    941       1.1      gwr 	int	i;
    942       1.1      gwr 	unsigned int	total_mem;
    943       1.1      gwr 
    944       1.7      gwr 	/*
    945       1.7      gwr 	 * Allocate a pv_head structure for every page of physical
    946       1.1      gwr 	 * memory that will be managed by the system.  Since memory on
    947       1.1      gwr 	 * the 3/80 is non-contiguous, we cannot arrive at a total page
    948       1.1      gwr 	 * count by subtraction of the lowest available address from the
    949       1.1      gwr 	 * highest, but rather we have to step through each memory
    950       1.1      gwr 	 * bank and add the number of pages in each to the total.
    951       1.1      gwr 	 *
    952       1.1      gwr 	 * At this time we also initialize the offset of each bank's
    953       1.1      gwr 	 * starting pv_head within the pv_head list so that the physical
    954       1.1      gwr 	 * memory state routines (pmap_is_referenced(),
    955       1.1      gwr 	 * pmap_is_modified(), et al.) can quickly find coresponding
    956       1.1      gwr 	 * pv_heads in spite of the non-contiguity.
    957       1.1      gwr 	 */
    958       1.1      gwr 	total_mem = 0;
    959      1.20  thorpej 	for (i = 0; i < SUN3X_NPHYS_RAM_SEGS; i++) {
    960      1.25    veego 		avail_mem[i].pmem_pvbase = m68k_btop(total_mem);
    961  1.88.2.2     yamt 		total_mem += avail_mem[i].pmem_end - avail_mem[i].pmem_start;
    962       1.1      gwr 		if (avail_mem[i].pmem_next == NULL)
    963       1.1      gwr 			break;
    964       1.1      gwr 	}
    965  1.88.2.2     yamt 	pvbase = (pv_t *)pmap_bootstrap_alloc(sizeof(pv_t) *
    966  1.88.2.2     yamt 	    m68k_btop(total_phys_mem));
    967       1.1      gwr }
    968       1.1      gwr 
    969       1.1      gwr /* pmap_alloc_usertmgr			INTERNAL
    970       1.1      gwr  **
    971       1.1      gwr  * Called from pmap_bootstrap() to allocate the structures which
    972       1.1      gwr  * facilitate management of user MMU tables.  Each user MMU table
    973       1.1      gwr  * in the system has one such structure associated with it.
    974       1.1      gwr  */
    975      1.86      chs void
    976      1.86      chs pmap_alloc_usertmgr(void)
    977       1.1      gwr {
    978       1.1      gwr 	/* Allocate user MMU table managers */
    979       1.7      gwr 	/* It would be a lot simpler to just make these BSS, but */
    980       1.7      gwr 	/* we may want to change their size at boot time... -j */
    981  1.88.2.2     yamt 	Atmgrbase =
    982  1.88.2.2     yamt 	    (a_tmgr_t *)pmap_bootstrap_alloc(sizeof(a_tmgr_t) * NUM_A_TABLES);
    983  1.88.2.2     yamt 	Btmgrbase =
    984  1.88.2.2     yamt 	    (b_tmgr_t *)pmap_bootstrap_alloc(sizeof(b_tmgr_t) * NUM_B_TABLES);
    985  1.88.2.2     yamt 	Ctmgrbase =
    986  1.88.2.2     yamt 	    (c_tmgr_t *)pmap_bootstrap_alloc(sizeof(c_tmgr_t) * NUM_C_TABLES);
    987       1.1      gwr 
    988       1.7      gwr 	/*
    989       1.7      gwr 	 * Allocate PV list elements for the physical to virtual
    990       1.1      gwr 	 * mapping system.
    991       1.1      gwr 	 */
    992  1.88.2.2     yamt 	pvebase = (pv_elem_t *)pmap_bootstrap_alloc(sizeof(pv_elem_t) *
    993  1.88.2.2     yamt 	    (NUM_USER_PTES + NUM_KERN_PTES));
    994       1.1      gwr }
    995       1.1      gwr 
    996       1.1      gwr /* pmap_bootstrap_copyprom()			INTERNAL
    997       1.1      gwr  **
    998       1.1      gwr  * Copy the PROM mappings into our own tables.  Note, we
    999       1.1      gwr  * can use physical addresses until __bootstrap returns.
   1000       1.1      gwr  */
   1001      1.86      chs void
   1002      1.86      chs pmap_bootstrap_copyprom(void)
   1003       1.1      gwr {
   1004      1.33      gwr 	struct sunromvec *romp;
   1005       1.1      gwr 	int *mon_ctbl;
   1006       1.1      gwr 	mmu_short_pte_t *kpte;
   1007       1.1      gwr 	int i, len;
   1008       1.1      gwr 
   1009       1.1      gwr 	romp = romVectorPtr;
   1010       1.1      gwr 
   1011       1.1      gwr 	/*
   1012      1.33      gwr 	 * Copy the mappings in SUN3X_MON_KDB_BASE...SUN3X_MONEND
   1013      1.33      gwr 	 * Note: mon_ctbl[0] maps SUN3X_MON_KDB_BASE
   1014       1.1      gwr 	 */
   1015       1.1      gwr 	mon_ctbl = *romp->monptaddr;
   1016      1.33      gwr 	i = m68k_btop(SUN3X_MON_KDB_BASE - KERNBASE);
   1017       1.1      gwr 	kpte = &kernCbase[i];
   1018      1.33      gwr 	len = m68k_btop(SUN3X_MONEND - SUN3X_MON_KDB_BASE);
   1019       1.1      gwr 
   1020       1.1      gwr 	for (i = 0; i < len; i++) {
   1021       1.1      gwr 		kpte[i].attr.raw = mon_ctbl[i];
   1022       1.1      gwr 	}
   1023       1.1      gwr 
   1024       1.1      gwr 	/*
   1025       1.1      gwr 	 * Copy the mappings at MON_DVMA_BASE (to the end).
   1026       1.1      gwr 	 * Note, in here, mon_ctbl[0] maps MON_DVMA_BASE.
   1027      1.32      gwr 	 * Actually, we only want the last page, which the
   1028      1.32      gwr 	 * PROM has set up for use by the "ie" driver.
   1029      1.32      gwr 	 * (The i82686 needs its SCP there.)
   1030      1.32      gwr 	 * If we copy all the mappings, pmap_enter_kernel
   1031      1.32      gwr 	 * may complain about finding valid PTEs that are
   1032      1.32      gwr 	 * not recorded in our PV lists...
   1033       1.1      gwr 	 */
   1034       1.1      gwr 	mon_ctbl = *romp->shadowpteaddr;
   1035      1.33      gwr 	i = m68k_btop(SUN3X_MON_DVMA_BASE - KERNBASE);
   1036       1.1      gwr 	kpte = &kernCbase[i];
   1037      1.33      gwr 	len = m68k_btop(SUN3X_MON_DVMA_SIZE);
   1038  1.88.2.2     yamt 	for (i = (len - 1); i < len; i++) {
   1039       1.1      gwr 		kpte[i].attr.raw = mon_ctbl[i];
   1040       1.1      gwr 	}
   1041       1.1      gwr }
   1042       1.1      gwr 
   1043       1.1      gwr /* pmap_takeover_mmu			INTERNAL
   1044       1.1      gwr  **
   1045       1.1      gwr  * Called from pmap_bootstrap() after it has copied enough of the
   1046       1.1      gwr  * PROM mappings into the kernel map so that we can use our own
   1047       1.1      gwr  * MMU table.
   1048       1.1      gwr  */
   1049      1.86      chs void
   1050      1.86      chs pmap_takeover_mmu(void)
   1051       1.1      gwr {
   1052       1.1      gwr 
   1053      1.13      gwr 	loadcrp(&kernel_crp);
   1054       1.1      gwr }
   1055       1.1      gwr 
   1056      1.13      gwr /* pmap_bootstrap_setprom()			INTERNAL
   1057      1.13      gwr  **
   1058      1.13      gwr  * Set the PROM mappings so it can see kernel space.
   1059      1.13      gwr  * Note that physical addresses are used here, which
   1060      1.13      gwr  * we can get away with because this runs with the
   1061      1.13      gwr  * low 1GB set for transparent translation.
   1062      1.13      gwr  */
   1063      1.86      chs void
   1064      1.86      chs pmap_bootstrap_setprom(void)
   1065      1.13      gwr {
   1066      1.13      gwr 	mmu_long_dte_t *mon_dte;
   1067      1.13      gwr 	extern struct mmu_rootptr mon_crp;
   1068      1.13      gwr 	int i;
   1069      1.13      gwr 
   1070  1.88.2.2     yamt 	mon_dte = (mmu_long_dte_t *)mon_crp.rp_addr;
   1071      1.13      gwr 	for (i = MMU_TIA(KERNBASE); i < MMU_TIA(KERN_END); i++) {
   1072      1.13      gwr 		mon_dte[i].attr.raw = kernAbase[i].attr.raw;
   1073      1.13      gwr 		mon_dte[i].addr.raw = kernAbase[i].addr.raw;
   1074      1.13      gwr 	}
   1075      1.13      gwr }
   1076      1.13      gwr 
   1077      1.13      gwr 
   1078       1.1      gwr /* pmap_init			INTERFACE
   1079       1.1      gwr  **
   1080       1.1      gwr  * Called at the end of vm_init() to set up the pmap system to go
   1081       1.7      gwr  * into full time operation.  All initialization of kernel_pmap
   1082       1.7      gwr  * should be already done by now, so this should just do things
   1083       1.7      gwr  * needed for user-level pmaps to work.
   1084       1.1      gwr  */
   1085      1.86      chs void
   1086      1.86      chs pmap_init(void)
   1087       1.1      gwr {
   1088  1.88.2.2     yamt 
   1089       1.1      gwr 	/** Initialize the manager pools **/
   1090       1.1      gwr 	TAILQ_INIT(&a_pool);
   1091       1.1      gwr 	TAILQ_INIT(&b_pool);
   1092       1.1      gwr 	TAILQ_INIT(&c_pool);
   1093       1.1      gwr 
   1094       1.1      gwr 	/**************************************************************
   1095       1.1      gwr 	 * Initialize all tmgr structures and MMU tables they manage. *
   1096       1.1      gwr 	 **************************************************************/
   1097       1.1      gwr 	/** Initialize A tables **/
   1098       1.1      gwr 	pmap_init_a_tables();
   1099       1.1      gwr 	/** Initialize B tables **/
   1100       1.1      gwr 	pmap_init_b_tables();
   1101       1.1      gwr 	/** Initialize C tables **/
   1102       1.1      gwr 	pmap_init_c_tables();
   1103      1.56  tsutsui 
   1104      1.56  tsutsui 	/** Initialize the pmap pools **/
   1105      1.56  tsutsui 	pool_init(&pmap_pmap_pool, sizeof(struct pmap), 0, 0, 0, "pmappl",
   1106      1.74  thorpej 	    &pool_allocator_nointr);
   1107       1.1      gwr }
   1108       1.1      gwr 
   1109       1.1      gwr /* pmap_init_a_tables()			INTERNAL
   1110       1.1      gwr  **
   1111       1.1      gwr  * Initializes all A managers, their MMU A tables, and inserts
   1112       1.1      gwr  * them into the A manager pool for use by the system.
   1113       1.1      gwr  */
   1114      1.86      chs void
   1115      1.86      chs pmap_init_a_tables(void)
   1116       1.1      gwr {
   1117       1.1      gwr 	int i;
   1118       1.1      gwr 	a_tmgr_t *a_tbl;
   1119       1.1      gwr 
   1120      1.86      chs 	for (i = 0; i < NUM_A_TABLES; i++) {
   1121       1.1      gwr 		/* Select the next available A manager from the pool */
   1122       1.1      gwr 		a_tbl = &Atmgrbase[i];
   1123       1.1      gwr 
   1124       1.7      gwr 		/*
   1125       1.7      gwr 		 * Clear its parent entry.  Set its wired and valid
   1126       1.1      gwr 		 * entry count to zero.
   1127       1.1      gwr 		 */
   1128       1.1      gwr 		a_tbl->at_parent = NULL;
   1129       1.1      gwr 		a_tbl->at_wcnt = a_tbl->at_ecnt = 0;
   1130       1.1      gwr 
   1131       1.1      gwr 		/* Assign it the next available MMU A table from the pool */
   1132       1.1      gwr 		a_tbl->at_dtbl = &mmuAbase[i * MMU_A_TBL_SIZE];
   1133       1.1      gwr 
   1134       1.7      gwr 		/*
   1135       1.7      gwr 		 * Initialize the MMU A table with the table in the `proc0',
   1136       1.1      gwr 		 * or kernel, mapping.  This ensures that every process has
   1137       1.1      gwr 		 * the kernel mapped in the top part of its address space.
   1138       1.1      gwr 		 */
   1139  1.88.2.2     yamt 		memcpy(a_tbl->at_dtbl, kernAbase,
   1140  1.88.2.2     yamt 		    MMU_A_TBL_SIZE * sizeof(mmu_long_dte_t));
   1141       1.1      gwr 
   1142       1.7      gwr 		/*
   1143       1.7      gwr 		 * Finally, insert the manager into the A pool,
   1144       1.1      gwr 		 * making it ready to be used by the system.
   1145       1.1      gwr 		 */
   1146       1.1      gwr 		TAILQ_INSERT_TAIL(&a_pool, a_tbl, at_link);
   1147       1.1      gwr     }
   1148       1.1      gwr }
   1149       1.1      gwr 
   1150       1.1      gwr /* pmap_init_b_tables()			INTERNAL
   1151       1.1      gwr  **
   1152       1.1      gwr  * Initializes all B table managers, their MMU B tables, and
   1153       1.1      gwr  * inserts them into the B manager pool for use by the system.
   1154       1.1      gwr  */
   1155      1.86      chs void
   1156      1.86      chs pmap_init_b_tables(void)
   1157       1.1      gwr {
   1158      1.86      chs 	int i, j;
   1159       1.1      gwr 	b_tmgr_t *b_tbl;
   1160       1.1      gwr 
   1161      1.86      chs 	for (i = 0; i < NUM_B_TABLES; i++) {
   1162       1.1      gwr 		/* Select the next available B manager from the pool */
   1163       1.1      gwr 		b_tbl = &Btmgrbase[i];
   1164       1.1      gwr 
   1165       1.1      gwr 		b_tbl->bt_parent = NULL;	/* clear its parent,  */
   1166       1.1      gwr 		b_tbl->bt_pidx = 0;		/* parent index,      */
   1167       1.1      gwr 		b_tbl->bt_wcnt = 0;		/* wired entry count, */
   1168       1.1      gwr 		b_tbl->bt_ecnt = 0;		/* valid entry count. */
   1169       1.1      gwr 
   1170       1.1      gwr 		/* Assign it the next available MMU B table from the pool */
   1171       1.1      gwr 		b_tbl->bt_dtbl = &mmuBbase[i * MMU_B_TBL_SIZE];
   1172       1.1      gwr 
   1173       1.1      gwr 		/* Invalidate every descriptor in the table */
   1174  1.88.2.2     yamt 		for (j = 0; j < MMU_B_TBL_SIZE; j++)
   1175       1.1      gwr 			b_tbl->bt_dtbl[j].attr.raw = MMU_DT_INVALID;
   1176       1.1      gwr 
   1177       1.1      gwr 		/* Insert the manager into the B pool */
   1178       1.1      gwr 		TAILQ_INSERT_TAIL(&b_pool, b_tbl, bt_link);
   1179       1.1      gwr 	}
   1180       1.1      gwr }
   1181       1.1      gwr 
   1182       1.1      gwr /* pmap_init_c_tables()			INTERNAL
   1183       1.1      gwr  **
   1184       1.1      gwr  * Initializes all C table managers, their MMU C tables, and
   1185       1.1      gwr  * inserts them into the C manager pool for use by the system.
   1186       1.1      gwr  */
   1187      1.86      chs void
   1188      1.86      chs pmap_init_c_tables(void)
   1189       1.1      gwr {
   1190      1.86      chs 	int i, j;
   1191       1.1      gwr 	c_tmgr_t *c_tbl;
   1192       1.1      gwr 
   1193      1.86      chs 	for (i = 0; i < NUM_C_TABLES; i++) {
   1194       1.1      gwr 		/* Select the next available C manager from the pool */
   1195       1.1      gwr 		c_tbl = &Ctmgrbase[i];
   1196       1.1      gwr 
   1197       1.1      gwr 		c_tbl->ct_parent = NULL;	/* clear its parent,  */
   1198       1.1      gwr 		c_tbl->ct_pidx = 0;		/* parent index,      */
   1199       1.1      gwr 		c_tbl->ct_wcnt = 0;		/* wired entry count, */
   1200      1.26   jeremy 		c_tbl->ct_ecnt = 0;		/* valid entry count, */
   1201      1.26   jeremy 		c_tbl->ct_pmap = NULL;		/* parent pmap,       */
   1202      1.26   jeremy 		c_tbl->ct_va = 0;		/* base of managed range */
   1203       1.1      gwr 
   1204       1.1      gwr 		/* Assign it the next available MMU C table from the pool */
   1205       1.1      gwr 		c_tbl->ct_dtbl = &mmuCbase[i * MMU_C_TBL_SIZE];
   1206       1.1      gwr 
   1207  1.88.2.2     yamt 		for (j = 0; j < MMU_C_TBL_SIZE; j++)
   1208       1.1      gwr 			c_tbl->ct_dtbl[j].attr.raw = MMU_DT_INVALID;
   1209       1.1      gwr 
   1210       1.1      gwr 		TAILQ_INSERT_TAIL(&c_pool, c_tbl, ct_link);
   1211       1.1      gwr 	}
   1212       1.1      gwr }
   1213       1.1      gwr 
   1214       1.1      gwr /* pmap_init_pv()			INTERNAL
   1215       1.1      gwr  **
   1216       1.1      gwr  * Initializes the Physical to Virtual mapping system.
   1217       1.1      gwr  */
   1218      1.86      chs void
   1219      1.86      chs pmap_init_pv(void)
   1220       1.1      gwr {
   1221      1.86      chs 	int i;
   1222       1.7      gwr 
   1223       1.7      gwr 	/* Initialize every PV head. */
   1224      1.25    veego 	for (i = 0; i < m68k_btop(total_phys_mem); i++) {
   1225       1.7      gwr 		pvbase[i].pv_idx = PVE_EOL;	/* Indicate no mappings */
   1226       1.7      gwr 		pvbase[i].pv_flags = 0;		/* Zero out page flags  */
   1227       1.7      gwr 	}
   1228       1.1      gwr }
   1229       1.1      gwr 
   1230  1.88.2.2     yamt /* is_managed				INTERNAL
   1231  1.88.2.2     yamt  **
   1232  1.88.2.2     yamt  * Determine if the given physical address is managed by the PV system.
   1233  1.88.2.2     yamt  * Note that this logic assumes that no one will ask for the status of
   1234  1.88.2.2     yamt  * addresses which lie in-between the memory banks on the 3/80.  If they
   1235  1.88.2.2     yamt  * do so, it will falsely report that it is managed.
   1236  1.88.2.2     yamt  *
   1237  1.88.2.2     yamt  * Note: A "managed" address is one that was reported to the VM system as
   1238  1.88.2.2     yamt  * a "usable page" during system startup.  As such, the VM system expects the
   1239  1.88.2.2     yamt  * pmap module to keep an accurate track of the useage of those pages.
   1240  1.88.2.2     yamt  * Any page not given to the VM system at startup does not exist (as far as
   1241  1.88.2.2     yamt  * the VM system is concerned) and is therefore "unmanaged."  Examples are
   1242  1.88.2.2     yamt  * those pages which belong to the ROM monitor and the memory allocated before
   1243  1.88.2.2     yamt  * the VM system was started.
   1244  1.88.2.2     yamt  */
   1245  1.88.2.3     yamt static INLINE bool
   1246  1.88.2.2     yamt is_managed(paddr_t pa)
   1247  1.88.2.2     yamt {
   1248  1.88.2.2     yamt 	if (pa >= avail_start && pa < avail_end)
   1249  1.88.2.3     yamt 		return true;
   1250  1.88.2.2     yamt 	else
   1251  1.88.2.3     yamt 		return false;
   1252  1.88.2.2     yamt }
   1253  1.88.2.2     yamt 
   1254       1.1      gwr /* get_a_table			INTERNAL
   1255       1.1      gwr  **
   1256       1.1      gwr  * Retrieve and return a level A table for use in a user map.
   1257       1.1      gwr  */
   1258       1.1      gwr a_tmgr_t *
   1259      1.86      chs get_a_table(void)
   1260       1.1      gwr {
   1261       1.1      gwr 	a_tmgr_t *tbl;
   1262       1.7      gwr 	pmap_t pmap;
   1263       1.1      gwr 
   1264       1.1      gwr 	/* Get the top A table in the pool */
   1265      1.86      chs 	tbl = TAILQ_FIRST(&a_pool);
   1266       1.7      gwr 	if (tbl == NULL) {
   1267       1.7      gwr 		/*
   1268      1.85      wiz 		 * XXX - Instead of panicking here and in other get_x_table
   1269       1.7      gwr 		 * functions, we do have the option of sleeping on the head of
   1270       1.7      gwr 		 * the table pool.  Any function which updates the table pool
   1271       1.7      gwr 		 * would then issue a wakeup() on the head, thus waking up any
   1272       1.7      gwr 		 * processes waiting for a table.
   1273       1.7      gwr 		 *
   1274       1.7      gwr 		 * Actually, the place to sleep would be when some process
   1275       1.7      gwr 		 * asks for a "wired" mapping that would run us short of
   1276       1.7      gwr 		 * mapping resources.  This design DEPENDS on always having
   1277       1.7      gwr 		 * some mapping resources in the pool for stealing, so we
   1278       1.7      gwr 		 * must make sure we NEVER let the pool become empty. -gwr
   1279       1.7      gwr 		 */
   1280       1.1      gwr 		panic("get_a_table: out of A tables.");
   1281       1.7      gwr 	}
   1282       1.7      gwr 
   1283       1.1      gwr 	TAILQ_REMOVE(&a_pool, tbl, at_link);
   1284       1.7      gwr 	/*
   1285       1.7      gwr 	 * If the table has a non-null parent pointer then it is in use.
   1286       1.1      gwr 	 * Forcibly abduct it from its parent and clear its entries.
   1287       1.1      gwr 	 * No re-entrancy worries here.  This table would not be in the
   1288       1.1      gwr 	 * table pool unless it was available for use.
   1289       1.7      gwr 	 *
   1290  1.88.2.3     yamt 	 * Note that the second argument to free_a_table() is false.  This
   1291       1.7      gwr 	 * indicates that the table should not be relinked into the A table
   1292       1.7      gwr 	 * pool.  That is a job for the function that called us.
   1293       1.1      gwr 	 */
   1294       1.1      gwr 	if (tbl->at_parent) {
   1295  1.88.2.2     yamt 		KASSERT(tbl->at_wcnt == 0);
   1296       1.7      gwr 		pmap = tbl->at_parent;
   1297  1.88.2.3     yamt 		free_a_table(tbl, false);
   1298       1.7      gwr 		pmap->pm_a_tmgr = NULL;
   1299       1.7      gwr 		pmap->pm_a_phys = kernAphys;
   1300       1.1      gwr 	}
   1301       1.1      gwr 	return tbl;
   1302       1.1      gwr }
   1303       1.1      gwr 
   1304       1.1      gwr /* get_b_table			INTERNAL
   1305       1.1      gwr  **
   1306       1.1      gwr  * Return a level B table for use.
   1307       1.1      gwr  */
   1308       1.1      gwr b_tmgr_t *
   1309      1.86      chs get_b_table(void)
   1310       1.1      gwr {
   1311       1.1      gwr 	b_tmgr_t *tbl;
   1312       1.1      gwr 
   1313       1.1      gwr 	/* See 'get_a_table' for comments. */
   1314      1.86      chs 	tbl = TAILQ_FIRST(&b_pool);
   1315       1.1      gwr 	if (tbl == NULL)
   1316       1.1      gwr 		panic("get_b_table: out of B tables.");
   1317       1.1      gwr 	TAILQ_REMOVE(&b_pool, tbl, bt_link);
   1318       1.1      gwr 	if (tbl->bt_parent) {
   1319  1.88.2.2     yamt 		KASSERT(tbl->bt_wcnt == 0);
   1320       1.1      gwr 		tbl->bt_parent->at_dtbl[tbl->bt_pidx].attr.raw = MMU_DT_INVALID;
   1321       1.1      gwr 		tbl->bt_parent->at_ecnt--;
   1322  1.88.2.3     yamt 		free_b_table(tbl, false);
   1323       1.1      gwr 	}
   1324       1.1      gwr 	return tbl;
   1325       1.1      gwr }
   1326       1.1      gwr 
   1327       1.1      gwr /* get_c_table			INTERNAL
   1328       1.1      gwr  **
   1329       1.1      gwr  * Return a level C table for use.
   1330       1.1      gwr  */
   1331       1.1      gwr c_tmgr_t *
   1332      1.86      chs get_c_table(void)
   1333       1.1      gwr {
   1334       1.1      gwr 	c_tmgr_t *tbl;
   1335       1.1      gwr 
   1336       1.1      gwr 	/* See 'get_a_table' for comments */
   1337      1.86      chs 	tbl = TAILQ_FIRST(&c_pool);
   1338       1.1      gwr 	if (tbl == NULL)
   1339       1.1      gwr 		panic("get_c_table: out of C tables.");
   1340       1.1      gwr 	TAILQ_REMOVE(&c_pool, tbl, ct_link);
   1341       1.1      gwr 	if (tbl->ct_parent) {
   1342  1.88.2.2     yamt 		KASSERT(tbl->ct_wcnt == 0);
   1343       1.1      gwr 		tbl->ct_parent->bt_dtbl[tbl->ct_pidx].attr.raw = MMU_DT_INVALID;
   1344       1.1      gwr 		tbl->ct_parent->bt_ecnt--;
   1345  1.88.2.3     yamt 		free_c_table(tbl, false);
   1346       1.1      gwr 	}
   1347       1.1      gwr 	return tbl;
   1348       1.1      gwr }
   1349       1.1      gwr 
   1350       1.7      gwr /*
   1351       1.7      gwr  * The following 'free_table' and 'steal_table' functions are called to
   1352       1.1      gwr  * detach tables from their current obligations (parents and children) and
   1353       1.1      gwr  * prepare them for reuse in another mapping.
   1354       1.1      gwr  *
   1355       1.1      gwr  * Free_table is used when the calling function will handle the fate
   1356       1.1      gwr  * of the parent table, such as returning it to the free pool when it has
   1357       1.1      gwr  * no valid entries.  Functions that do not want to handle this should
   1358       1.1      gwr  * call steal_table, in which the parent table's descriptors and entry
   1359       1.1      gwr  * count are automatically modified when this table is removed.
   1360       1.1      gwr  */
   1361       1.1      gwr 
   1362       1.1      gwr /* free_a_table			INTERNAL
   1363       1.1      gwr  **
   1364       1.1      gwr  * Unmaps the given A table and all child tables from their current
   1365       1.1      gwr  * mappings.  Returns the number of pages that were invalidated.
   1366       1.7      gwr  * If 'relink' is true, the function will return the table to the head
   1367       1.7      gwr  * of the available table pool.
   1368       1.1      gwr  *
   1369       1.1      gwr  * Cache note: The MC68851 will automatically flush all
   1370       1.1      gwr  * descriptors derived from a given A table from its
   1371       1.1      gwr  * Automatic Translation Cache (ATC) if we issue a
   1372       1.1      gwr  * 'PFLUSHR' instruction with the base address of the
   1373       1.1      gwr  * table.  This function should do, and does so.
   1374       1.1      gwr  * Note note: We are using an MC68030 - there is no
   1375       1.1      gwr  * PFLUSHR.
   1376       1.1      gwr  */
   1377      1.86      chs int
   1378  1.88.2.3     yamt free_a_table(a_tmgr_t *a_tbl, bool relink)
   1379       1.1      gwr {
   1380       1.1      gwr 	int i, removed_cnt;
   1381       1.1      gwr 	mmu_long_dte_t	*dte;
   1382       1.1      gwr 	mmu_short_dte_t *dtbl;
   1383  1.88.2.2     yamt 	b_tmgr_t	*b_tbl;
   1384  1.88.2.2     yamt 	uint8_t at_wired, bt_wired;
   1385       1.1      gwr 
   1386       1.7      gwr 	/*
   1387       1.7      gwr 	 * Flush the ATC cache of all cached descriptors derived
   1388       1.1      gwr 	 * from this table.
   1389      1.22   jeremy 	 * Sun3x does not use 68851's cached table feature
   1390       1.1      gwr 	 * flush_atc_crp(mmu_vtop(a_tbl->dte));
   1391       1.1      gwr 	 */
   1392       1.1      gwr 
   1393       1.7      gwr 	/*
   1394       1.7      gwr 	 * Remove any pending cache flushes that were designated
   1395       1.1      gwr 	 * for the pmap this A table belongs to.
   1396       1.1      gwr 	 * a_tbl->parent->atc_flushq[0] = 0;
   1397      1.22   jeremy 	 * Not implemented in sun3x.
   1398       1.1      gwr 	 */
   1399       1.1      gwr 
   1400       1.7      gwr 	/*
   1401       1.7      gwr 	 * All A tables in the system should retain a map for the
   1402       1.1      gwr 	 * kernel. If the table contains any valid descriptors
   1403       1.1      gwr 	 * (other than those for the kernel area), invalidate them all,
   1404       1.1      gwr 	 * stopping short of the kernel's entries.
   1405       1.1      gwr 	 */
   1406       1.1      gwr 	removed_cnt = 0;
   1407  1.88.2.2     yamt 	at_wired = a_tbl->at_wcnt;
   1408       1.1      gwr 	if (a_tbl->at_ecnt) {
   1409       1.1      gwr 		dte = a_tbl->at_dtbl;
   1410  1.88.2.2     yamt 		for (i = 0; i < MMU_TIA(KERNBASE); i++) {
   1411       1.7      gwr 			/*
   1412       1.7      gwr 			 * If a table entry points to a valid B table, free
   1413       1.1      gwr 			 * it and its children.
   1414       1.1      gwr 			 */
   1415       1.1      gwr 			if (MMU_VALID_DT(dte[i])) {
   1416       1.7      gwr 				/*
   1417       1.7      gwr 				 * The following block does several things,
   1418       1.1      gwr 				 * from innermost expression to the
   1419       1.1      gwr 				 * outermost:
   1420       1.1      gwr 				 * 1) It extracts the base (cc 1996)
   1421       1.1      gwr 				 *    address of the B table pointed
   1422       1.1      gwr 				 *    to in the A table entry dte[i].
   1423       1.1      gwr 				 * 2) It converts this base address into
   1424       1.1      gwr 				 *    the virtual address it can be
   1425       1.1      gwr 				 *    accessed with. (all MMU tables point
   1426       1.1      gwr 				 *    to physical addresses.)
   1427       1.1      gwr 				 * 3) It finds the corresponding manager
   1428       1.1      gwr 				 *    structure which manages this MMU table.
   1429       1.1      gwr 				 * 4) It frees the manager structure.
   1430       1.1      gwr 				 *    (This frees the MMU table and all
   1431       1.1      gwr 				 *    child tables. See 'free_b_table' for
   1432       1.1      gwr 				 *    details.)
   1433       1.1      gwr 				 */
   1434       1.7      gwr 				dtbl = mmu_ptov(dte[i].addr.raw);
   1435  1.88.2.2     yamt 				b_tbl = mmuB2tmgr(dtbl);
   1436  1.88.2.2     yamt 				bt_wired = b_tbl->bt_wcnt;
   1437  1.88.2.3     yamt 				removed_cnt += free_b_table(b_tbl, true);
   1438  1.88.2.2     yamt 				if (bt_wired)
   1439  1.88.2.2     yamt 					a_tbl->at_wcnt--;
   1440       1.8      gwr 				dte[i].attr.raw = MMU_DT_INVALID;
   1441       1.1      gwr 			}
   1442       1.8      gwr 		}
   1443       1.8      gwr 		a_tbl->at_ecnt = 0;
   1444       1.1      gwr 	}
   1445  1.88.2.2     yamt 	KASSERT(a_tbl->at_wcnt == 0);
   1446  1.88.2.2     yamt 
   1447       1.7      gwr 	if (relink) {
   1448       1.7      gwr 		a_tbl->at_parent = NULL;
   1449  1.88.2.2     yamt 		if (!at_wired)
   1450  1.88.2.2     yamt 			TAILQ_REMOVE(&a_pool, a_tbl, at_link);
   1451       1.7      gwr 		TAILQ_INSERT_HEAD(&a_pool, a_tbl, at_link);
   1452       1.7      gwr 	}
   1453       1.1      gwr 	return removed_cnt;
   1454       1.1      gwr }
   1455       1.1      gwr 
   1456       1.1      gwr /* free_b_table			INTERNAL
   1457       1.1      gwr  **
   1458       1.1      gwr  * Unmaps the given B table and all its children from their current
   1459       1.1      gwr  * mappings.  Returns the number of pages that were invalidated.
   1460       1.1      gwr  * (For comments, see 'free_a_table()').
   1461       1.1      gwr  */
   1462      1.86      chs int
   1463  1.88.2.3     yamt free_b_table(b_tmgr_t *b_tbl, bool relink)
   1464       1.1      gwr {
   1465       1.1      gwr 	int i, removed_cnt;
   1466       1.1      gwr 	mmu_short_dte_t *dte;
   1467       1.1      gwr 	mmu_short_pte_t	*dtbl;
   1468  1.88.2.2     yamt 	c_tmgr_t	*c_tbl;
   1469  1.88.2.2     yamt 	uint8_t bt_wired, ct_wired;
   1470       1.1      gwr 
   1471       1.1      gwr 	removed_cnt = 0;
   1472  1.88.2.2     yamt 	bt_wired = b_tbl->bt_wcnt;
   1473       1.1      gwr 	if (b_tbl->bt_ecnt) {
   1474       1.1      gwr 		dte = b_tbl->bt_dtbl;
   1475  1.88.2.2     yamt 		for (i = 0; i < MMU_B_TBL_SIZE; i++) {
   1476       1.1      gwr 			if (MMU_VALID_DT(dte[i])) {
   1477       1.7      gwr 				dtbl = mmu_ptov(MMU_DTE_PA(dte[i]));
   1478  1.88.2.2     yamt 				c_tbl = mmuC2tmgr(dtbl);
   1479  1.88.2.2     yamt 				ct_wired = c_tbl->ct_wcnt;
   1480  1.88.2.3     yamt 				removed_cnt += free_c_table(c_tbl, true);
   1481  1.88.2.2     yamt 				if (ct_wired)
   1482  1.88.2.2     yamt 					b_tbl->bt_wcnt--;
   1483       1.8      gwr 				dte[i].attr.raw = MMU_DT_INVALID;
   1484       1.1      gwr 			}
   1485       1.8      gwr 		}
   1486       1.8      gwr 		b_tbl->bt_ecnt = 0;
   1487       1.1      gwr 	}
   1488  1.88.2.2     yamt 	KASSERT(b_tbl->bt_wcnt == 0);
   1489       1.1      gwr 
   1490       1.7      gwr 	if (relink) {
   1491       1.7      gwr 		b_tbl->bt_parent = NULL;
   1492  1.88.2.2     yamt 		if (!bt_wired)
   1493  1.88.2.2     yamt 			TAILQ_REMOVE(&b_pool, b_tbl, bt_link);
   1494       1.7      gwr 		TAILQ_INSERT_HEAD(&b_pool, b_tbl, bt_link);
   1495       1.7      gwr 	}
   1496       1.1      gwr 	return removed_cnt;
   1497       1.1      gwr }
   1498       1.1      gwr 
   1499       1.1      gwr /* free_c_table			INTERNAL
   1500       1.1      gwr  **
   1501       1.1      gwr  * Unmaps the given C table from use and returns it to the pool for
   1502       1.1      gwr  * re-use.  Returns the number of pages that were invalidated.
   1503       1.1      gwr  *
   1504       1.1      gwr  * This function preserves any physical page modification information
   1505       1.1      gwr  * contained in the page descriptors within the C table by calling
   1506       1.1      gwr  * 'pmap_remove_pte().'
   1507       1.1      gwr  */
   1508      1.86      chs int
   1509  1.88.2.3     yamt free_c_table(c_tmgr_t *c_tbl, bool relink)
   1510       1.1      gwr {
   1511  1.88.2.2     yamt 	mmu_short_pte_t *c_pte;
   1512       1.1      gwr 	int i, removed_cnt;
   1513  1.88.2.2     yamt 	uint8_t ct_wired;
   1514       1.1      gwr 
   1515       1.1      gwr 	removed_cnt = 0;
   1516  1.88.2.2     yamt 	ct_wired = c_tbl->ct_wcnt;
   1517       1.8      gwr 	if (c_tbl->ct_ecnt) {
   1518  1.88.2.2     yamt 		for (i = 0; i < MMU_C_TBL_SIZE; i++) {
   1519  1.88.2.2     yamt 			c_pte = &c_tbl->ct_dtbl[i];
   1520  1.88.2.2     yamt 			if (MMU_VALID_DT(*c_pte)) {
   1521  1.88.2.2     yamt 				if (c_pte->attr.raw & MMU_SHORT_PTE_WIRED)
   1522  1.88.2.2     yamt 					c_tbl->ct_wcnt--;
   1523  1.88.2.2     yamt 				pmap_remove_pte(c_pte);
   1524       1.1      gwr 				removed_cnt++;
   1525       1.1      gwr 			}
   1526       1.8      gwr 		}
   1527       1.8      gwr 		c_tbl->ct_ecnt = 0;
   1528       1.8      gwr 	}
   1529  1.88.2.2     yamt 	KASSERT(c_tbl->ct_wcnt == 0);
   1530       1.8      gwr 
   1531       1.7      gwr 	if (relink) {
   1532       1.7      gwr 		c_tbl->ct_parent = NULL;
   1533  1.88.2.2     yamt 		if (!ct_wired)
   1534  1.88.2.2     yamt 			TAILQ_REMOVE(&c_pool, c_tbl, ct_link);
   1535       1.7      gwr 		TAILQ_INSERT_HEAD(&c_pool, c_tbl, ct_link);
   1536       1.7      gwr 	}
   1537       1.1      gwr 	return removed_cnt;
   1538       1.1      gwr }
   1539       1.1      gwr 
   1540       1.1      gwr 
   1541       1.1      gwr /* pmap_remove_pte			INTERNAL
   1542       1.1      gwr  **
   1543       1.1      gwr  * Unmap the given pte and preserve any page modification
   1544       1.1      gwr  * information by transfering it to the pv head of the
   1545       1.1      gwr  * physical page it maps to.  This function does not update
   1546       1.1      gwr  * any reference counts because it is assumed that the calling
   1547       1.8      gwr  * function will do so.
   1548       1.1      gwr  */
   1549       1.1      gwr void
   1550      1.86      chs pmap_remove_pte(mmu_short_pte_t *pte)
   1551       1.1      gwr {
   1552       1.7      gwr 	u_short     pv_idx, targ_idx;
   1553      1.69      chs 	paddr_t     pa;
   1554       1.1      gwr 	pv_t       *pv;
   1555       1.1      gwr 
   1556       1.1      gwr 	pa = MMU_PTE_PA(*pte);
   1557       1.1      gwr 	if (is_managed(pa)) {
   1558       1.1      gwr 		pv = pa2pv(pa);
   1559       1.7      gwr 		targ_idx = pteidx(pte);	/* Index of PTE being removed    */
   1560       1.7      gwr 
   1561       1.7      gwr 		/*
   1562       1.7      gwr 		 * If the PTE being removed is the first (or only) PTE in
   1563       1.7      gwr 		 * the list of PTEs currently mapped to this page, remove the
   1564       1.7      gwr 		 * PTE by changing the index found on the PV head.  Otherwise
   1565       1.7      gwr 		 * a linear search through the list will have to be executed
   1566       1.7      gwr 		 * in order to find the PVE which points to the PTE being
   1567       1.7      gwr 		 * removed, so that it may be modified to point to its new
   1568       1.7      gwr 		 * neighbor.
   1569       1.7      gwr 		 */
   1570      1.69      chs 
   1571       1.7      gwr 		pv_idx = pv->pv_idx;	/* Index of first PTE in PV list */
   1572       1.7      gwr 		if (pv_idx == targ_idx) {
   1573       1.7      gwr 			pv->pv_idx = pvebase[targ_idx].pve_next;
   1574       1.7      gwr 		} else {
   1575      1.69      chs 
   1576       1.7      gwr 			/*
   1577      1.32      gwr 			 * Find the PV element pointing to the target
   1578      1.32      gwr 			 * element.  Note: may have pv_idx==PVE_EOL
   1579       1.7      gwr 			 */
   1580      1.69      chs 
   1581      1.32      gwr 			for (;;) {
   1582      1.32      gwr 				if (pv_idx == PVE_EOL) {
   1583      1.32      gwr 					goto pv_not_found;
   1584      1.32      gwr 				}
   1585      1.32      gwr 				if (pvebase[pv_idx].pve_next == targ_idx)
   1586      1.32      gwr 					break;
   1587       1.7      gwr 				pv_idx = pvebase[pv_idx].pve_next;
   1588       1.7      gwr 			}
   1589      1.69      chs 
   1590       1.7      gwr 			/*
   1591       1.7      gwr 			 * At this point, pv_idx is the index of the PV
   1592       1.7      gwr 			 * element just before the target element in the list.
   1593       1.7      gwr 			 * Unlink the target.
   1594       1.7      gwr 			 */
   1595      1.69      chs 
   1596       1.7      gwr 			pvebase[pv_idx].pve_next = pvebase[targ_idx].pve_next;
   1597       1.7      gwr 		}
   1598      1.69      chs 
   1599       1.7      gwr 		/*
   1600       1.7      gwr 		 * Save the mod/ref bits of the pte by simply
   1601       1.1      gwr 		 * ORing the entire pte onto the pv_flags member
   1602       1.1      gwr 		 * of the pv structure.
   1603       1.1      gwr 		 * There is no need to use a separate bit pattern
   1604       1.1      gwr 		 * for usage information on the pv head than that
   1605       1.1      gwr 		 * which is used on the MMU ptes.
   1606       1.1      gwr 		 */
   1607      1.69      chs 
   1608  1.88.2.2     yamt  pv_not_found:
   1609       1.7      gwr 		pv->pv_flags |= (u_short) pte->attr.raw;
   1610       1.1      gwr 	}
   1611       1.1      gwr 	pte->attr.raw = MMU_DT_INVALID;
   1612       1.1      gwr }
   1613       1.1      gwr 
   1614       1.1      gwr /* pmap_stroll			INTERNAL
   1615       1.1      gwr  **
   1616       1.1      gwr  * Retrieve the addresses of all table managers involved in the mapping of
   1617      1.77      wiz  * the given virtual address.  If the table walk completed successfully,
   1618  1.88.2.3     yamt  * return true.  If it was only partially successful, return false.
   1619       1.1      gwr  * The table walk performed by this function is important to many other
   1620       1.1      gwr  * functions in this module.
   1621       1.7      gwr  *
   1622       1.7      gwr  * Note: This function ought to be easier to read.
   1623       1.1      gwr  */
   1624  1.88.2.3     yamt bool
   1625      1.86      chs pmap_stroll(pmap_t pmap, vaddr_t va, a_tmgr_t **a_tbl, b_tmgr_t **b_tbl,
   1626      1.86      chs     c_tmgr_t **c_tbl, mmu_short_pte_t **pte, int *a_idx, int *b_idx,
   1627      1.86      chs     int *pte_idx)
   1628       1.1      gwr {
   1629       1.1      gwr 	mmu_long_dte_t *a_dte;   /* A: long descriptor table          */
   1630       1.1      gwr 	mmu_short_dte_t *b_dte;  /* B: short descriptor table         */
   1631       1.1      gwr 
   1632       1.1      gwr 	if (pmap == pmap_kernel())
   1633  1.88.2.3     yamt 		return false;
   1634       1.1      gwr 
   1635       1.7      gwr 	/* Does the given pmap have its own A table? */
   1636       1.7      gwr 	*a_tbl = pmap->pm_a_tmgr;
   1637       1.1      gwr 	if (*a_tbl == NULL)
   1638  1.88.2.3     yamt 		return false; /* No.  Return unknown. */
   1639       1.1      gwr 	/* Does the A table have a valid B table
   1640       1.1      gwr 	 * under the corresponding table entry?
   1641       1.1      gwr 	 */
   1642       1.1      gwr 	*a_idx = MMU_TIA(va);
   1643       1.1      gwr 	a_dte = &((*a_tbl)->at_dtbl[*a_idx]);
   1644       1.1      gwr 	if (!MMU_VALID_DT(*a_dte))
   1645  1.88.2.3     yamt 		return false; /* No. Return unknown. */
   1646       1.1      gwr 	/* Yes. Extract B table from the A table. */
   1647       1.7      gwr 	*b_tbl = mmuB2tmgr(mmu_ptov(a_dte->addr.raw));
   1648  1.88.2.2     yamt 	/*
   1649  1.88.2.2     yamt 	 * Does the B table have a valid C table
   1650       1.1      gwr 	 * under the corresponding table entry?
   1651       1.1      gwr 	 */
   1652       1.1      gwr 	*b_idx = MMU_TIB(va);
   1653       1.1      gwr 	b_dte = &((*b_tbl)->bt_dtbl[*b_idx]);
   1654       1.1      gwr 	if (!MMU_VALID_DT(*b_dte))
   1655  1.88.2.3     yamt 		return false; /* No. Return unknown. */
   1656       1.1      gwr 	/* Yes. Extract C table from the B table. */
   1657       1.7      gwr 	*c_tbl = mmuC2tmgr(mmu_ptov(MMU_DTE_PA(*b_dte)));
   1658       1.1      gwr 	*pte_idx = MMU_TIC(va);
   1659       1.1      gwr 	*pte = &((*c_tbl)->ct_dtbl[*pte_idx]);
   1660       1.1      gwr 
   1661  1.88.2.3     yamt 	return true;
   1662       1.1      gwr }
   1663       1.1      gwr 
   1664       1.1      gwr /* pmap_enter			INTERFACE
   1665       1.1      gwr  **
   1666       1.1      gwr  * Called by the kernel to map a virtual address
   1667       1.1      gwr  * to a physical address in the given process map.
   1668       1.1      gwr  *
   1669       1.1      gwr  * Note: this function should apply an exclusive lock
   1670       1.1      gwr  * on the pmap system for its duration.  (it certainly
   1671       1.1      gwr  * would save my hair!!)
   1672       1.7      gwr  * This function ought to be easier to read.
   1673       1.1      gwr  */
   1674      1.86      chs int
   1675      1.86      chs pmap_enter(pmap_t pmap, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
   1676       1.1      gwr {
   1677  1.88.2.3     yamt 	bool insert, managed; /* Marks the need for PV insertion.*/
   1678       1.7      gwr 	u_short nidx;            /* PV list index                     */
   1679      1.52   jeremy 	int mapflags;            /* Flags for the mapping (see NOTE1) */
   1680       1.8      gwr 	u_int a_idx, b_idx, pte_idx; /* table indices                 */
   1681       1.1      gwr 	a_tmgr_t *a_tbl;         /* A: long descriptor table manager  */
   1682       1.1      gwr 	b_tmgr_t *b_tbl;         /* B: short descriptor table manager */
   1683       1.1      gwr 	c_tmgr_t *c_tbl;         /* C: short page table manager       */
   1684       1.1      gwr 	mmu_long_dte_t *a_dte;   /* A: long descriptor table          */
   1685       1.1      gwr 	mmu_short_dte_t *b_dte;  /* B: short descriptor table         */
   1686       1.1      gwr 	mmu_short_pte_t *c_pte;  /* C: short page descriptor table    */
   1687       1.1      gwr 	pv_t      *pv;           /* pv list head                      */
   1688  1.88.2.3     yamt 	bool wired;         /* is the mapping to be wired?       */
   1689       1.1      gwr 	enum {NONE, NEWA, NEWB, NEWC} llevel; /* used at end   */
   1690       1.1      gwr 
   1691       1.1      gwr 	if (pmap == pmap_kernel()) {
   1692       1.1      gwr 		pmap_enter_kernel(va, pa, prot);
   1693      1.61      chs 		return 0;
   1694       1.1      gwr 	}
   1695       1.7      gwr 
   1696      1.52   jeremy 	/*
   1697      1.52   jeremy 	 * Determine if the mapping should be wired.
   1698      1.52   jeremy 	 */
   1699      1.52   jeremy 	wired = ((flags & PMAP_WIRED) != 0);
   1700      1.52   jeremy 
   1701      1.52   jeremy 	/*
   1702      1.52   jeremy 	 * NOTE1:
   1703      1.52   jeremy 	 *
   1704      1.52   jeremy 	 * On November 13, 1999, someone changed the pmap_enter() API such
   1705      1.52   jeremy 	 * that it now accepts a 'flags' argument.  This new argument
   1706      1.52   jeremy 	 * contains bit-flags for the architecture-independent (UVM) system to
   1707      1.52   jeremy 	 * use in signalling certain mapping requirements to the architecture-
   1708      1.52   jeremy 	 * dependent (pmap) system.  The argument it replaces, 'wired', is now
   1709      1.52   jeremy 	 * one of the flags within it.
   1710      1.52   jeremy 	 *
   1711      1.52   jeremy 	 * In addition to flags signaled by the architecture-independent
   1712      1.52   jeremy 	 * system, parts of the architecture-dependent section of the sun3x
   1713      1.52   jeremy 	 * kernel pass their own flags in the lower, unused bits of the
   1714      1.52   jeremy 	 * physical address supplied to this function.  These flags are
   1715      1.52   jeremy 	 * extracted and stored in the temporary variable 'mapflags'.
   1716      1.52   jeremy 	 *
   1717      1.52   jeremy 	 * Extract sun3x specific flags from the physical address.
   1718      1.52   jeremy 	 */
   1719  1.88.2.2     yamt 	mapflags = (pa & ~MMU_PAGE_MASK);
   1720  1.88.2.2     yamt 	pa &= MMU_PAGE_MASK;
   1721       1.7      gwr 
   1722       1.7      gwr 	/*
   1723      1.22   jeremy 	 * Determine if the physical address being mapped is on-board RAM.
   1724      1.22   jeremy 	 * Any other area of the address space is likely to belong to a
   1725      1.22   jeremy 	 * device and hence it would be disasterous to cache its contents.
   1726       1.7      gwr 	 */
   1727  1.88.2.3     yamt 	if ((managed = is_managed(pa)) == false)
   1728      1.52   jeremy 		mapflags |= PMAP_NC;
   1729       1.7      gwr 
   1730       1.7      gwr 	/*
   1731       1.7      gwr 	 * For user mappings we walk along the MMU tables of the given
   1732       1.1      gwr 	 * pmap, reaching a PTE which describes the virtual page being
   1733       1.1      gwr 	 * mapped or changed.  If any level of the walk ends in an invalid
   1734       1.1      gwr 	 * entry, a table must be allocated and the entry must be updated
   1735       1.1      gwr 	 * to point to it.
   1736       1.1      gwr 	 * There is a bit of confusion as to whether this code must be
   1737       1.1      gwr 	 * re-entrant.  For now we will assume it is.  To support
   1738       1.1      gwr 	 * re-entrancy we must unlink tables from the table pool before
   1739       1.1      gwr 	 * we assume we may use them.  Tables are re-linked into the pool
   1740       1.1      gwr 	 * when we are finished with them at the end of the function.
   1741       1.1      gwr 	 * But I don't feel like doing that until we have proof that this
   1742       1.1      gwr 	 * needs to be re-entrant.
   1743       1.1      gwr 	 * 'llevel' records which tables need to be relinked.
   1744       1.1      gwr 	 */
   1745       1.1      gwr 	llevel = NONE;
   1746       1.1      gwr 
   1747       1.7      gwr 	/*
   1748       1.7      gwr 	 * Step 1 - Retrieve the A table from the pmap.  If it has no
   1749       1.7      gwr 	 * A table, allocate a new one from the available pool.
   1750       1.1      gwr 	 */
   1751       1.1      gwr 
   1752       1.7      gwr 	a_tbl = pmap->pm_a_tmgr;
   1753       1.7      gwr 	if (a_tbl == NULL) {
   1754       1.7      gwr 		/*
   1755       1.7      gwr 		 * This pmap does not currently have an A table.  Allocate
   1756       1.7      gwr 		 * a new one.
   1757       1.7      gwr 		 */
   1758       1.7      gwr 		a_tbl = get_a_table();
   1759       1.7      gwr 		a_tbl->at_parent = pmap;
   1760       1.7      gwr 
   1761       1.7      gwr 		/*
   1762       1.7      gwr 		 * Assign this new A table to the pmap, and calculate its
   1763       1.7      gwr 		 * physical address so that loadcrp() can be used to make
   1764       1.7      gwr 		 * the table active.
   1765       1.7      gwr 		 */
   1766       1.7      gwr 		pmap->pm_a_tmgr = a_tbl;
   1767       1.7      gwr 		pmap->pm_a_phys = mmu_vtop(a_tbl->at_dtbl);
   1768       1.7      gwr 
   1769       1.7      gwr 		/*
   1770       1.7      gwr 		 * If the process receiving a new A table is the current
   1771       1.7      gwr 		 * process, we are responsible for setting the MMU so that
   1772       1.9      gwr 		 * it becomes the current address space.  This only adds
   1773       1.9      gwr 		 * new mappings, so no need to flush anything.
   1774       1.7      gwr 		 */
   1775       1.9      gwr 		if (pmap == current_pmap()) {
   1776       1.9      gwr 			kernel_crp.rp_addr = pmap->pm_a_phys;
   1777       1.9      gwr 			loadcrp(&kernel_crp);
   1778       1.9      gwr 		}
   1779       1.7      gwr 
   1780       1.1      gwr 		if (!wired)
   1781       1.1      gwr 			llevel = NEWA;
   1782       1.1      gwr 	} else {
   1783       1.7      gwr 		/*
   1784       1.7      gwr 		 * Use the A table already allocated for this pmap.
   1785       1.1      gwr 		 * Unlink it from the A table pool if necessary.
   1786       1.1      gwr 		 */
   1787       1.1      gwr 		if (wired && !a_tbl->at_wcnt)
   1788       1.1      gwr 			TAILQ_REMOVE(&a_pool, a_tbl, at_link);
   1789       1.1      gwr 	}
   1790       1.1      gwr 
   1791       1.7      gwr 	/*
   1792       1.7      gwr 	 * Step 2 - Walk into the B table.  If there is no valid B table,
   1793       1.1      gwr 	 * allocate one.
   1794       1.1      gwr 	 */
   1795       1.1      gwr 
   1796       1.1      gwr 	a_idx = MMU_TIA(va);            /* Calculate the TIA of the VA. */
   1797       1.1      gwr 	a_dte = &a_tbl->at_dtbl[a_idx]; /* Retrieve descriptor from table */
   1798       1.1      gwr 	if (MMU_VALID_DT(*a_dte)) {     /* Is the descriptor valid? */
   1799       1.7      gwr 		/* The descriptor is valid.  Use the B table it points to. */
   1800       1.1      gwr 		/*************************************
   1801       1.1      gwr 		 *               a_idx               *
   1802       1.1      gwr 		 *                 v                 *
   1803       1.1      gwr 		 * a_tbl -> +-+-+-+-+-+-+-+-+-+-+-+- *
   1804       1.1      gwr 		 *          | | | | | | | | | | | |  *
   1805       1.1      gwr 		 *          +-+-+-+-+-+-+-+-+-+-+-+- *
   1806       1.1      gwr 		 *                 |                 *
   1807       1.1      gwr 		 *                 \- b_tbl -> +-+-  *
   1808       1.1      gwr 		 *                             | |   *
   1809       1.1      gwr 		 *                             +-+-  *
   1810       1.1      gwr 		 *************************************/
   1811       1.7      gwr 		b_dte = mmu_ptov(a_dte->addr.raw);
   1812       1.1      gwr 		b_tbl = mmuB2tmgr(b_dte);
   1813       1.7      gwr 
   1814       1.7      gwr 		/*
   1815       1.7      gwr 		 * If the requested mapping must be wired, but this table
   1816       1.7      gwr 		 * being used to map it is not, the table must be removed
   1817       1.7      gwr 		 * from the available pool and its wired entry count
   1818       1.7      gwr 		 * incremented.
   1819       1.7      gwr 		 */
   1820       1.1      gwr 		if (wired && !b_tbl->bt_wcnt) {
   1821       1.1      gwr 			TAILQ_REMOVE(&b_pool, b_tbl, bt_link);
   1822       1.7      gwr 			a_tbl->at_wcnt++;
   1823       1.1      gwr 		}
   1824       1.1      gwr 	} else {
   1825       1.7      gwr 		/* The descriptor is invalid.  Allocate a new B table. */
   1826       1.7      gwr 		b_tbl = get_b_table();
   1827       1.7      gwr 
   1828       1.1      gwr 		/* Point the parent A table descriptor to this new B table. */
   1829       1.7      gwr 		a_dte->addr.raw = mmu_vtop(b_tbl->bt_dtbl);
   1830       1.7      gwr 		a_dte->attr.raw = MMU_LONG_DTE_LU | MMU_DT_SHORT;
   1831       1.7      gwr 		a_tbl->at_ecnt++; /* Update parent's valid entry count */
   1832       1.7      gwr 
   1833       1.1      gwr 		/* Create the necessary back references to the parent table */
   1834       1.1      gwr 		b_tbl->bt_parent = a_tbl;
   1835       1.1      gwr 		b_tbl->bt_pidx = a_idx;
   1836       1.7      gwr 
   1837       1.7      gwr 		/*
   1838       1.7      gwr 		 * If this table is to be wired, make sure the parent A table
   1839       1.1      gwr 		 * wired count is updated to reflect that it has another wired
   1840       1.1      gwr 		 * entry.
   1841       1.1      gwr 		 */
   1842       1.1      gwr 		if (wired)
   1843       1.1      gwr 			a_tbl->at_wcnt++;
   1844       1.1      gwr 		else if (llevel == NONE)
   1845       1.1      gwr 			llevel = NEWB;
   1846       1.1      gwr 	}
   1847       1.1      gwr 
   1848       1.7      gwr 	/*
   1849       1.7      gwr 	 * Step 3 - Walk into the C table, if there is no valid C table,
   1850       1.1      gwr 	 * allocate one.
   1851       1.1      gwr 	 */
   1852       1.1      gwr 
   1853       1.1      gwr 	b_idx = MMU_TIB(va);            /* Calculate the TIB of the VA */
   1854       1.1      gwr 	b_dte = &b_tbl->bt_dtbl[b_idx]; /* Retrieve descriptor from table */
   1855       1.1      gwr 	if (MMU_VALID_DT(*b_dte)) {     /* Is the descriptor valid? */
   1856       1.7      gwr 		/* The descriptor is valid.  Use the C table it points to. */
   1857       1.1      gwr 		/**************************************
   1858       1.1      gwr 		 *               c_idx                *
   1859       1.1      gwr 		 * |                v                 *
   1860       1.1      gwr 		 * \- b_tbl -> +-+-+-+-+-+-+-+-+-+-+- *
   1861       1.1      gwr 		 *             | | | | | | | | | | |  *
   1862       1.1      gwr 		 *             +-+-+-+-+-+-+-+-+-+-+- *
   1863       1.1      gwr 		 *                  |                 *
   1864       1.1      gwr 		 *                  \- c_tbl -> +-+-- *
   1865       1.1      gwr 		 *                              | | | *
   1866       1.1      gwr 		 *                              +-+-- *
   1867       1.1      gwr 		 **************************************/
   1868       1.7      gwr 		c_pte = mmu_ptov(MMU_PTE_PA(*b_dte));
   1869       1.1      gwr 		c_tbl = mmuC2tmgr(c_pte);
   1870       1.7      gwr 
   1871       1.7      gwr 		/* If mapping is wired and table is not */
   1872       1.1      gwr 		if (wired && !c_tbl->ct_wcnt) {
   1873       1.1      gwr 			TAILQ_REMOVE(&c_pool, c_tbl, ct_link);
   1874       1.1      gwr 			b_tbl->bt_wcnt++;
   1875       1.1      gwr 		}
   1876       1.1      gwr 	} else {
   1877       1.7      gwr 		/* The descriptor is invalid.  Allocate a new C table. */
   1878       1.7      gwr 		c_tbl = get_c_table();
   1879       1.7      gwr 
   1880       1.1      gwr 		/* Point the parent B table descriptor to this new C table. */
   1881       1.7      gwr 		b_dte->attr.raw = mmu_vtop(c_tbl->ct_dtbl);
   1882       1.7      gwr 		b_dte->attr.raw |= MMU_DT_SHORT;
   1883       1.7      gwr 		b_tbl->bt_ecnt++; /* Update parent's valid entry count */
   1884       1.7      gwr 
   1885       1.1      gwr 		/* Create the necessary back references to the parent table */
   1886       1.1      gwr 		c_tbl->ct_parent = b_tbl;
   1887       1.1      gwr 		c_tbl->ct_pidx = b_idx;
   1888      1.26   jeremy 		/*
   1889      1.26   jeremy 		 * Store the pmap and base virtual managed address for faster
   1890      1.26   jeremy 		 * retrieval in the PV functions.
   1891      1.26   jeremy 		 */
   1892      1.26   jeremy 		c_tbl->ct_pmap = pmap;
   1893      1.26   jeremy 		c_tbl->ct_va = (va & (MMU_TIA_MASK|MMU_TIB_MASK));
   1894       1.7      gwr 
   1895       1.7      gwr 		/*
   1896       1.7      gwr 		 * If this table is to be wired, make sure the parent B table
   1897       1.1      gwr 		 * wired count is updated to reflect that it has another wired
   1898       1.1      gwr 		 * entry.
   1899       1.1      gwr 		 */
   1900       1.1      gwr 		if (wired)
   1901       1.1      gwr 			b_tbl->bt_wcnt++;
   1902       1.1      gwr 		else if (llevel == NONE)
   1903       1.1      gwr 			llevel = NEWC;
   1904       1.1      gwr 	}
   1905       1.1      gwr 
   1906       1.7      gwr 	/*
   1907       1.7      gwr 	 * Step 4 - Deposit a page descriptor (PTE) into the appropriate
   1908       1.1      gwr 	 * slot of the C table, describing the PA to which the VA is mapped.
   1909       1.1      gwr 	 */
   1910       1.1      gwr 
   1911       1.1      gwr 	pte_idx = MMU_TIC(va);
   1912       1.1      gwr 	c_pte = &c_tbl->ct_dtbl[pte_idx];
   1913       1.1      gwr 	if (MMU_VALID_DT(*c_pte)) { /* Is the entry currently valid? */
   1914       1.7      gwr 		/*
   1915       1.7      gwr 		 * The PTE is currently valid.  This particular call
   1916       1.1      gwr 		 * is just a synonym for one (or more) of the following
   1917       1.1      gwr 		 * operations:
   1918       1.7      gwr 		 *     change protection of a page
   1919       1.1      gwr 		 *     change wiring status of a page
   1920       1.1      gwr 		 *     remove the mapping of a page
   1921       1.1      gwr 		 */
   1922       1.7      gwr 
   1923       1.7      gwr 		/* First check if this is a wiring operation. */
   1924  1.88.2.2     yamt 		if (c_pte->attr.raw & MMU_SHORT_PTE_WIRED) {
   1925       1.7      gwr 			/*
   1926  1.88.2.2     yamt 			 * The existing mapping is wired, so adjust wired
   1927  1.88.2.2     yamt 			 * entry count here. If new mapping is still wired,
   1928  1.88.2.2     yamt 			 * wired entry count will be incremented again later.
   1929       1.7      gwr 			 */
   1930  1.88.2.2     yamt 			c_tbl->ct_wcnt--;
   1931  1.88.2.2     yamt 			if (!wired) {
   1932  1.88.2.2     yamt 				/*
   1933  1.88.2.2     yamt 				 * The mapping of this PTE is being changed
   1934  1.88.2.2     yamt 				 * from wired to unwired.
   1935  1.88.2.2     yamt 				 * Adjust wired entry counts in each table and
   1936  1.88.2.2     yamt 				 * set llevel flag to put unwired tables back
   1937  1.88.2.2     yamt 				 * into the active pool.
   1938  1.88.2.2     yamt 				 */
   1939  1.88.2.2     yamt 				if (c_tbl->ct_wcnt == 0) {
   1940  1.88.2.2     yamt 					llevel = NEWC;
   1941  1.88.2.2     yamt 					if (--b_tbl->bt_wcnt == 0) {
   1942  1.88.2.2     yamt 						llevel = NEWB;
   1943  1.88.2.2     yamt 						if (--a_tbl->at_wcnt == 0) {
   1944  1.88.2.2     yamt 							llevel = NEWA;
   1945  1.88.2.2     yamt 						}
   1946  1.88.2.2     yamt 					}
   1947  1.88.2.2     yamt 				}
   1948  1.88.2.2     yamt 			}
   1949       1.7      gwr 		}
   1950       1.7      gwr 
   1951       1.1      gwr 		/* Is the new address the same as the old? */
   1952       1.1      gwr 		if (MMU_PTE_PA(*c_pte) == pa) {
   1953       1.7      gwr 			/*
   1954       1.7      gwr 			 * Yes, mark that it does not need to be reinserted
   1955       1.7      gwr 			 * into the PV list.
   1956       1.7      gwr 			 */
   1957  1.88.2.3     yamt 			insert = false;
   1958       1.7      gwr 
   1959       1.7      gwr 			/*
   1960       1.7      gwr 			 * Clear all but the modified, referenced and wired
   1961       1.7      gwr 			 * bits on the PTE.
   1962       1.7      gwr 			 */
   1963       1.7      gwr 			c_pte->attr.raw &= (MMU_SHORT_PTE_M
   1964  1.88.2.2     yamt 			    | MMU_SHORT_PTE_USED | MMU_SHORT_PTE_WIRED);
   1965       1.1      gwr 		} else {
   1966       1.1      gwr 			/* No, remove the old entry */
   1967       1.1      gwr 			pmap_remove_pte(c_pte);
   1968  1.88.2.3     yamt 			insert = true;
   1969       1.1      gwr 		}
   1970       1.8      gwr 
   1971       1.8      gwr 		/*
   1972       1.8      gwr 		 * TLB flush is only necessary if modifying current map.
   1973       1.8      gwr 		 * However, in pmap_enter(), the pmap almost always IS
   1974       1.8      gwr 		 * the current pmap, so don't even bother to check.
   1975       1.8      gwr 		 */
   1976       1.8      gwr 		TBIS(va);
   1977       1.1      gwr 	} else {
   1978       1.7      gwr 		/*
   1979       1.7      gwr 		 * The PTE is invalid.  Increment the valid entry count in
   1980       1.8      gwr 		 * the C table manager to reflect the addition of a new entry.
   1981       1.7      gwr 		 */
   1982       1.1      gwr 		c_tbl->ct_ecnt++;
   1983       1.8      gwr 
   1984       1.8      gwr 		/* XXX - temporarily make sure the PTE is cleared. */
   1985       1.8      gwr 		c_pte->attr.raw = 0;
   1986       1.1      gwr 
   1987       1.7      gwr 		/* It will also need to be inserted into the PV list. */
   1988  1.88.2.3     yamt 		insert = true;
   1989       1.7      gwr 	}
   1990       1.7      gwr 
   1991       1.7      gwr 	/*
   1992       1.7      gwr 	 * If page is changing from unwired to wired status, set an unused bit
   1993       1.7      gwr 	 * within the PTE to indicate that it is wired.  Also increment the
   1994       1.7      gwr 	 * wired entry count in the C table manager.
   1995       1.7      gwr 	 */
   1996       1.7      gwr 	if (wired) {
   1997       1.1      gwr 		c_pte->attr.raw |= MMU_SHORT_PTE_WIRED;
   1998       1.7      gwr 		c_tbl->ct_wcnt++;
   1999       1.1      gwr 	}
   2000       1.1      gwr 
   2001       1.7      gwr 	/*
   2002       1.7      gwr 	 * Map the page, being careful to preserve modify/reference/wired
   2003       1.7      gwr 	 * bits.  At this point it is assumed that the PTE either has no bits
   2004       1.7      gwr 	 * set, or if there are set bits, they are only modified, reference or
   2005       1.7      gwr 	 * wired bits.  If not, the following statement will cause erratic
   2006       1.7      gwr 	 * behavior.
   2007       1.7      gwr 	 */
   2008       1.8      gwr #ifdef	PMAP_DEBUG
   2009       1.7      gwr 	if (c_pte->attr.raw & ~(MMU_SHORT_PTE_M |
   2010       1.7      gwr 		MMU_SHORT_PTE_USED | MMU_SHORT_PTE_WIRED)) {
   2011       1.7      gwr 		printf("pmap_enter: junk left in PTE at %p\n", c_pte);
   2012       1.7      gwr 		Debugger();
   2013       1.7      gwr 	}
   2014       1.7      gwr #endif
   2015       1.7      gwr 	c_pte->attr.raw |= ((u_long) pa | MMU_DT_PAGE);
   2016       1.7      gwr 
   2017       1.7      gwr 	/*
   2018       1.7      gwr 	 * If the mapping should be read-only, set the write protect
   2019       1.7      gwr 	 * bit in the PTE.
   2020       1.7      gwr 	 */
   2021       1.7      gwr 	if (!(prot & VM_PROT_WRITE))
   2022       1.7      gwr 		c_pte->attr.raw |= MMU_SHORT_PTE_WP;
   2023       1.7      gwr 
   2024       1.7      gwr 	/*
   2025      1.87      chs 	 * Mark the PTE as used and/or modified as specified by the flags arg.
   2026      1.87      chs 	 */
   2027      1.87      chs 	if (flags & VM_PROT_ALL) {
   2028      1.87      chs 		c_pte->attr.raw |= MMU_SHORT_PTE_USED;
   2029      1.87      chs 		if (flags & VM_PROT_WRITE) {
   2030      1.87      chs 			c_pte->attr.raw |= MMU_SHORT_PTE_M;
   2031      1.87      chs 		}
   2032      1.87      chs 	}
   2033      1.87      chs 
   2034      1.87      chs 	/*
   2035       1.7      gwr 	 * If the mapping should be cache inhibited (indicated by the flag
   2036       1.7      gwr 	 * bits found on the lower order of the physical address.)
   2037       1.7      gwr 	 * mark the PTE as a cache inhibited page.
   2038       1.7      gwr 	 */
   2039      1.52   jeremy 	if (mapflags & PMAP_NC)
   2040       1.7      gwr 		c_pte->attr.raw |= MMU_SHORT_PTE_CI;
   2041       1.7      gwr 
   2042       1.7      gwr 	/*
   2043       1.7      gwr 	 * If the physical address being mapped is managed by the PV
   2044       1.7      gwr 	 * system then link the pte into the list of pages mapped to that
   2045       1.7      gwr 	 * address.
   2046       1.7      gwr 	 */
   2047       1.7      gwr 	if (insert && managed) {
   2048       1.7      gwr 		pv = pa2pv(pa);
   2049       1.7      gwr 		nidx = pteidx(c_pte);
   2050       1.7      gwr 
   2051       1.7      gwr 		pvebase[nidx].pve_next = pv->pv_idx;
   2052       1.7      gwr 		pv->pv_idx = nidx;
   2053       1.7      gwr 	}
   2054       1.1      gwr 
   2055  1.88.2.2     yamt 	/* Move any allocated or unwired tables back into the active pool. */
   2056       1.1      gwr 
   2057       1.1      gwr 	switch (llevel) {
   2058       1.1      gwr 		case NEWA:
   2059       1.1      gwr 			TAILQ_INSERT_TAIL(&a_pool, a_tbl, at_link);
   2060       1.1      gwr 			/* FALLTHROUGH */
   2061       1.1      gwr 		case NEWB:
   2062       1.1      gwr 			TAILQ_INSERT_TAIL(&b_pool, b_tbl, bt_link);
   2063       1.1      gwr 			/* FALLTHROUGH */
   2064       1.1      gwr 		case NEWC:
   2065       1.1      gwr 			TAILQ_INSERT_TAIL(&c_pool, c_tbl, ct_link);
   2066       1.1      gwr 			/* FALLTHROUGH */
   2067       1.1      gwr 		default:
   2068       1.1      gwr 			break;
   2069       1.1      gwr 	}
   2070      1.51  thorpej 
   2071      1.61      chs 	return 0;
   2072       1.1      gwr }
   2073       1.1      gwr 
   2074       1.1      gwr /* pmap_enter_kernel			INTERNAL
   2075       1.1      gwr  **
   2076       1.1      gwr  * Map the given virtual address to the given physical address within the
   2077       1.1      gwr  * kernel address space.  This function exists because the kernel map does
   2078       1.1      gwr  * not do dynamic table allocation.  It consists of a contiguous array of ptes
   2079       1.1      gwr  * and can be edited directly without the need to walk through any tables.
   2080       1.1      gwr  *
   2081       1.1      gwr  * XXX: "Danger, Will Robinson!"
   2082       1.1      gwr  * Note that the kernel should never take a fault on any page
   2083       1.1      gwr  * between [ KERNBASE .. virtual_avail ] and this is checked in
   2084       1.1      gwr  * trap.c for kernel-mode MMU faults.  This means that mappings
   2085       1.1      gwr  * created in that range must be implicily wired. -gwr
   2086       1.1      gwr  */
   2087      1.86      chs void
   2088      1.86      chs pmap_enter_kernel(vaddr_t va, paddr_t pa, vm_prot_t prot)
   2089       1.1      gwr {
   2090  1.88.2.3     yamt 	bool       was_valid, insert;
   2091      1.32      gwr 	u_short         pte_idx;
   2092      1.69      chs 	int             flags;
   2093       1.1      gwr 	mmu_short_pte_t *pte;
   2094       1.7      gwr 	pv_t            *pv;
   2095      1.69      chs 	paddr_t     old_pa;
   2096       1.7      gwr 
   2097      1.32      gwr 	flags = (pa & ~MMU_PAGE_MASK);
   2098      1.32      gwr 	pa &= MMU_PAGE_MASK;
   2099      1.32      gwr 
   2100      1.32      gwr 	if (is_managed(pa))
   2101  1.88.2.3     yamt 		insert = true;
   2102      1.32      gwr 	else
   2103  1.88.2.3     yamt 		insert = false;
   2104       1.7      gwr 
   2105       1.7      gwr 	/*
   2106       1.7      gwr 	 * Calculate the index of the PTE being modified.
   2107       1.7      gwr 	 */
   2108  1.88.2.2     yamt 	pte_idx = (u_long)m68k_btop(va - KERNBASE);
   2109       1.1      gwr 
   2110      1.22   jeremy 	/* This array is traditionally named "Sysmap" */
   2111       1.7      gwr 	pte = &kernCbase[pte_idx];
   2112       1.7      gwr 
   2113       1.7      gwr 	if (MMU_VALID_DT(*pte)) {
   2114  1.88.2.3     yamt 		was_valid = true;
   2115       1.7      gwr 		/*
   2116      1.32      gwr 		 * If the PTE already maps a different
   2117      1.32      gwr 		 * physical address, umap and pv_unlink.
   2118      1.24   jeremy 		 */
   2119      1.24   jeremy 		old_pa = MMU_PTE_PA(*pte);
   2120      1.32      gwr 		if (pa != old_pa)
   2121      1.32      gwr 			pmap_remove_pte(pte);
   2122      1.32      gwr 		else {
   2123      1.24   jeremy 		    /*
   2124      1.32      gwr 		     * Old PA and new PA are the same.  No need to
   2125      1.32      gwr 		     * relink the mapping within the PV list.
   2126      1.24   jeremy 		     */
   2127  1.88.2.3     yamt 		     insert = false;
   2128       1.8      gwr 
   2129       1.7      gwr 		    /*
   2130      1.24   jeremy 		     * Save any mod/ref bits on the PTE.
   2131       1.7      gwr 		     */
   2132      1.24   jeremy 		    pte->attr.raw &= (MMU_SHORT_PTE_USED|MMU_SHORT_PTE_M);
   2133       1.7      gwr 		}
   2134       1.7      gwr 	} else {
   2135       1.8      gwr 		pte->attr.raw = MMU_DT_INVALID;
   2136  1.88.2.3     yamt 		was_valid = false;
   2137       1.7      gwr 	}
   2138       1.7      gwr 
   2139       1.7      gwr 	/*
   2140       1.8      gwr 	 * Map the page.  Being careful to preserve modified/referenced bits
   2141       1.8      gwr 	 * on the PTE.
   2142       1.7      gwr 	 */
   2143       1.7      gwr 	pte->attr.raw |= (pa | MMU_DT_PAGE);
   2144       1.1      gwr 
   2145       1.1      gwr 	if (!(prot & VM_PROT_WRITE)) /* If access should be read-only */
   2146       1.1      gwr 		pte->attr.raw |= MMU_SHORT_PTE_WP;
   2147       1.7      gwr 	if (flags & PMAP_NC)
   2148       1.1      gwr 		pte->attr.raw |= MMU_SHORT_PTE_CI;
   2149       1.8      gwr 	if (was_valid)
   2150       1.7      gwr 		TBIS(va);
   2151       1.1      gwr 
   2152       1.7      gwr 	/*
   2153       1.7      gwr 	 * Insert the PTE into the PV system, if need be.
   2154       1.7      gwr 	 */
   2155       1.7      gwr 	if (insert) {
   2156       1.7      gwr 		pv = pa2pv(pa);
   2157       1.7      gwr 		pvebase[pte_idx].pve_next = pv->pv_idx;
   2158       1.7      gwr 		pv->pv_idx = pte_idx;
   2159       1.7      gwr 	}
   2160      1.34      gwr }
   2161      1.34      gwr 
   2162      1.86      chs void
   2163      1.86      chs pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
   2164      1.49      chs {
   2165      1.69      chs 	mmu_short_pte_t	*pte;
   2166      1.69      chs 
   2167      1.69      chs 	/* This array is traditionally named "Sysmap" */
   2168      1.69      chs 	pte = &kernCbase[(u_long)m68k_btop(va - KERNBASE)];
   2169      1.69      chs 
   2170      1.69      chs 	KASSERT(!MMU_VALID_DT(*pte));
   2171      1.69      chs 	pte->attr.raw = MMU_DT_INVALID | MMU_DT_PAGE | (pa & MMU_PAGE_MASK);
   2172      1.69      chs 	if (!(prot & VM_PROT_WRITE))
   2173      1.69      chs 		pte->attr.raw |= MMU_SHORT_PTE_WP;
   2174      1.49      chs }
   2175      1.49      chs 
   2176      1.86      chs void
   2177      1.86      chs pmap_kremove(vaddr_t va, vsize_t len)
   2178      1.49      chs {
   2179      1.69      chs 	int idx, eidx;
   2180      1.69      chs 
   2181      1.69      chs #ifdef	PMAP_DEBUG
   2182  1.88.2.2     yamt 	if ((va & PGOFSET) || (len & PGOFSET))
   2183      1.72  tsutsui 		panic("pmap_kremove: alignment");
   2184      1.69      chs #endif
   2185      1.69      chs 
   2186      1.69      chs 	idx  = m68k_btop(va - KERNBASE);
   2187      1.69      chs 	eidx = m68k_btop(va + len - KERNBASE);
   2188      1.69      chs 
   2189      1.69      chs 	while (idx < eidx) {
   2190      1.69      chs 		kernCbase[idx++].attr.raw = MMU_DT_INVALID;
   2191      1.69      chs 		TBIS(va);
   2192      1.79  thorpej 		va += PAGE_SIZE;
   2193      1.49      chs 	}
   2194      1.49      chs }
   2195      1.49      chs 
   2196      1.35   jeremy /* pmap_map			INTERNAL
   2197      1.35   jeremy  **
   2198      1.35   jeremy  * Map a contiguous range of physical memory into a contiguous range of
   2199      1.35   jeremy  * the kernel virtual address space.
   2200      1.35   jeremy  *
   2201      1.35   jeremy  * Used for device mappings and early mapping of the kernel text/data/bss.
   2202      1.35   jeremy  * Returns the first virtual address beyond the end of the range.
   2203      1.34      gwr  */
   2204      1.86      chs vaddr_t
   2205      1.86      chs pmap_map(vaddr_t va, paddr_t pa, paddr_t endpa, int prot)
   2206      1.34      gwr {
   2207      1.34      gwr 	int sz;
   2208      1.34      gwr 
   2209      1.34      gwr 	sz = endpa - pa;
   2210      1.34      gwr 	do {
   2211      1.34      gwr 		pmap_enter_kernel(va, pa, prot);
   2212      1.79  thorpej 		va += PAGE_SIZE;
   2213      1.79  thorpej 		pa += PAGE_SIZE;
   2214      1.79  thorpej 		sz -= PAGE_SIZE;
   2215      1.34      gwr 	} while (sz > 0);
   2216      1.73    chris 	pmap_update(pmap_kernel());
   2217  1.88.2.2     yamt 	return va;
   2218  1.88.2.2     yamt }
   2219  1.88.2.2     yamt 
   2220  1.88.2.2     yamt /* pmap_protect_kernel			INTERNAL
   2221  1.88.2.2     yamt  **
   2222  1.88.2.2     yamt  * Apply the given protection code to a kernel address range.
   2223  1.88.2.2     yamt  */
   2224  1.88.2.2     yamt static INLINE void
   2225  1.88.2.2     yamt pmap_protect_kernel(vaddr_t startva, vaddr_t endva, vm_prot_t prot)
   2226  1.88.2.2     yamt {
   2227  1.88.2.2     yamt 	vaddr_t va;
   2228  1.88.2.2     yamt 	mmu_short_pte_t *pte;
   2229  1.88.2.2     yamt 
   2230  1.88.2.2     yamt 	pte = &kernCbase[(unsigned long) m68k_btop(startva - KERNBASE)];
   2231  1.88.2.2     yamt 	for (va = startva; va < endva; va += PAGE_SIZE, pte++) {
   2232  1.88.2.2     yamt 		if (MMU_VALID_DT(*pte)) {
   2233  1.88.2.2     yamt 		    switch (prot) {
   2234  1.88.2.2     yamt 		        case VM_PROT_ALL:
   2235  1.88.2.2     yamt 		            break;
   2236  1.88.2.2     yamt 		        case VM_PROT_EXECUTE:
   2237  1.88.2.2     yamt 		        case VM_PROT_READ:
   2238  1.88.2.2     yamt 		        case VM_PROT_READ|VM_PROT_EXECUTE:
   2239  1.88.2.2     yamt 		            pte->attr.raw |= MMU_SHORT_PTE_WP;
   2240  1.88.2.2     yamt 		            break;
   2241  1.88.2.2     yamt 		        case VM_PROT_NONE:
   2242  1.88.2.2     yamt 		            /* this is an alias for 'pmap_remove_kernel' */
   2243  1.88.2.2     yamt 		            pmap_remove_pte(pte);
   2244  1.88.2.2     yamt 		            break;
   2245  1.88.2.2     yamt 		        default:
   2246  1.88.2.2     yamt 		            break;
   2247  1.88.2.2     yamt 		    }
   2248  1.88.2.2     yamt 		    /*
   2249  1.88.2.2     yamt 		     * since this is the kernel, immediately flush any cached
   2250  1.88.2.2     yamt 		     * descriptors for this address.
   2251  1.88.2.2     yamt 		     */
   2252  1.88.2.2     yamt 		    TBIS(va);
   2253  1.88.2.2     yamt 		}
   2254  1.88.2.2     yamt 	}
   2255       1.1      gwr }
   2256       1.1      gwr 
   2257       1.1      gwr /* pmap_protect			INTERFACE
   2258       1.1      gwr  **
   2259       1.7      gwr  * Apply the given protection to the given virtual address range within
   2260       1.1      gwr  * the given map.
   2261       1.1      gwr  *
   2262       1.1      gwr  * It is ok for the protection applied to be stronger than what is
   2263       1.1      gwr  * specified.  We use this to our advantage when the given map has no
   2264       1.7      gwr  * mapping for the virtual address.  By skipping a page when this
   2265       1.1      gwr  * is discovered, we are effectively applying a protection of VM_PROT_NONE,
   2266       1.1      gwr  * and therefore do not need to map the page just to apply a protection
   2267       1.1      gwr  * code.  Only pmap_enter() needs to create new mappings if they do not exist.
   2268       1.7      gwr  *
   2269       1.7      gwr  * XXX - This function could be speeded up by using pmap_stroll() for inital
   2270       1.7      gwr  *       setup, and then manual scrolling in the for() loop.
   2271       1.1      gwr  */
   2272      1.86      chs void
   2273      1.86      chs pmap_protect(pmap_t pmap, vaddr_t startva, vaddr_t endva, vm_prot_t prot)
   2274       1.1      gwr {
   2275  1.88.2.3     yamt 	bool iscurpmap;
   2276       1.1      gwr 	int a_idx, b_idx, c_idx;
   2277       1.1      gwr 	a_tmgr_t *a_tbl;
   2278       1.1      gwr 	b_tmgr_t *b_tbl;
   2279       1.1      gwr 	c_tmgr_t *c_tbl;
   2280       1.1      gwr 	mmu_short_pte_t *pte;
   2281       1.1      gwr 
   2282       1.1      gwr 	if (pmap == pmap_kernel()) {
   2283       1.7      gwr 		pmap_protect_kernel(startva, endva, prot);
   2284       1.1      gwr 		return;
   2285       1.1      gwr 	}
   2286       1.1      gwr 
   2287      1.11   jeremy 	/*
   2288      1.12   jeremy 	 * In this particular pmap implementation, there are only three
   2289      1.12   jeremy 	 * types of memory protection: 'all' (read/write/execute),
   2290      1.12   jeremy 	 * 'read-only' (read/execute) and 'none' (no mapping.)
   2291      1.12   jeremy 	 * It is not possible for us to treat 'executable' as a separate
   2292      1.12   jeremy 	 * protection type.  Therefore, protection requests that seek to
   2293      1.12   jeremy 	 * remove execute permission while retaining read or write, and those
   2294      1.12   jeremy 	 * that make little sense (write-only for example) are ignored.
   2295      1.11   jeremy 	 */
   2296      1.12   jeremy 	switch (prot) {
   2297      1.12   jeremy 		case VM_PROT_NONE:
   2298      1.12   jeremy 			/*
   2299      1.12   jeremy 			 * A request to apply the protection code of
   2300      1.12   jeremy 			 * 'VM_PROT_NONE' is a synonym for pmap_remove().
   2301      1.12   jeremy 			 */
   2302      1.12   jeremy 			pmap_remove(pmap, startva, endva);
   2303      1.12   jeremy 			return;
   2304      1.12   jeremy 		case	VM_PROT_EXECUTE:
   2305      1.12   jeremy 		case	VM_PROT_READ:
   2306      1.12   jeremy 		case	VM_PROT_READ|VM_PROT_EXECUTE:
   2307      1.12   jeremy 			/* continue */
   2308      1.12   jeremy 			break;
   2309      1.12   jeremy 		case	VM_PROT_WRITE:
   2310      1.12   jeremy 		case	VM_PROT_WRITE|VM_PROT_READ:
   2311      1.12   jeremy 		case	VM_PROT_WRITE|VM_PROT_EXECUTE:
   2312      1.12   jeremy 		case	VM_PROT_ALL:
   2313      1.12   jeremy 			/* None of these should happen in a sane system. */
   2314      1.12   jeremy 			return;
   2315      1.11   jeremy 	}
   2316      1.11   jeremy 
   2317      1.11   jeremy 	/*
   2318      1.11   jeremy 	 * If the pmap has no A table, it has no mappings and therefore
   2319      1.11   jeremy 	 * there is nothing to protect.
   2320      1.11   jeremy 	 */
   2321      1.11   jeremy 	if ((a_tbl = pmap->pm_a_tmgr) == NULL)
   2322      1.11   jeremy 		return;
   2323      1.11   jeremy 
   2324      1.11   jeremy 	a_idx = MMU_TIA(startva);
   2325      1.11   jeremy 	b_idx = MMU_TIB(startva);
   2326      1.11   jeremy 	c_idx = MMU_TIC(startva);
   2327  1.88.2.1     yamt 	b_tbl = NULL;
   2328  1.88.2.1     yamt 	c_tbl = NULL;
   2329      1.11   jeremy 
   2330       1.7      gwr 	iscurpmap = (pmap == current_pmap());
   2331      1.11   jeremy 	while (startva < endva) {
   2332      1.11   jeremy 		if (b_tbl || MMU_VALID_DT(a_tbl->at_dtbl[a_idx])) {
   2333      1.11   jeremy 		  if (b_tbl == NULL) {
   2334      1.11   jeremy 		    b_tbl = (b_tmgr_t *) a_tbl->at_dtbl[a_idx].addr.raw;
   2335      1.69      chs 		    b_tbl = mmu_ptov((vaddr_t)b_tbl);
   2336      1.69      chs 		    b_tbl = mmuB2tmgr((mmu_short_dte_t *)b_tbl);
   2337      1.11   jeremy 		  }
   2338      1.11   jeremy 		  if (c_tbl || MMU_VALID_DT(b_tbl->bt_dtbl[b_idx])) {
   2339      1.11   jeremy 		    if (c_tbl == NULL) {
   2340      1.11   jeremy 		      c_tbl = (c_tmgr_t *) MMU_DTE_PA(b_tbl->bt_dtbl[b_idx]);
   2341      1.69      chs 		      c_tbl = mmu_ptov((vaddr_t)c_tbl);
   2342      1.69      chs 		      c_tbl = mmuC2tmgr((mmu_short_pte_t *)c_tbl);
   2343      1.11   jeremy 		    }
   2344      1.11   jeremy 		    if (MMU_VALID_DT(c_tbl->ct_dtbl[c_idx])) {
   2345      1.11   jeremy 		      pte = &c_tbl->ct_dtbl[c_idx];
   2346      1.12   jeremy 		      /* make the mapping read-only */
   2347      1.12   jeremy 		      pte->attr.raw |= MMU_SHORT_PTE_WP;
   2348      1.11   jeremy 		      /*
   2349      1.11   jeremy 		       * If we just modified the current address space,
   2350      1.11   jeremy 		       * flush any translations for the modified page from
   2351      1.11   jeremy 		       * the translation cache and any data from it in the
   2352      1.11   jeremy 		       * data cache.
   2353      1.11   jeremy 		       */
   2354      1.11   jeremy 		      if (iscurpmap)
   2355      1.11   jeremy 		          TBIS(startva);
   2356      1.11   jeremy 		    }
   2357      1.79  thorpej 		    startva += PAGE_SIZE;
   2358       1.1      gwr 
   2359      1.11   jeremy 		    if (++c_idx >= MMU_C_TBL_SIZE) { /* exceeded C table? */
   2360      1.11   jeremy 		      c_tbl = NULL;
   2361      1.11   jeremy 		      c_idx = 0;
   2362      1.11   jeremy 		      if (++b_idx >= MMU_B_TBL_SIZE) { /* exceeded B table? */
   2363      1.11   jeremy 		        b_tbl = NULL;
   2364      1.11   jeremy 		        b_idx = 0;
   2365      1.11   jeremy 		      }
   2366      1.11   jeremy 		    }
   2367      1.11   jeremy 		  } else { /* C table wasn't valid */
   2368      1.11   jeremy 		    c_tbl = NULL;
   2369      1.11   jeremy 		    c_idx = 0;
   2370      1.11   jeremy 		    startva += MMU_TIB_RANGE;
   2371      1.11   jeremy 		    if (++b_idx >= MMU_B_TBL_SIZE) { /* exceeded B table? */
   2372      1.11   jeremy 		      b_tbl = NULL;
   2373      1.11   jeremy 		      b_idx = 0;
   2374      1.11   jeremy 		    }
   2375      1.11   jeremy 		  } /* C table */
   2376      1.11   jeremy 		} else { /* B table wasn't valid */
   2377      1.11   jeremy 		  b_tbl = NULL;
   2378      1.11   jeremy 		  b_idx = 0;
   2379      1.11   jeremy 		  startva += MMU_TIA_RANGE;
   2380      1.11   jeremy 		  a_idx++;
   2381      1.11   jeremy 		} /* B table */
   2382       1.1      gwr 	}
   2383       1.1      gwr }
   2384       1.1      gwr 
   2385      1.47  thorpej /* pmap_unwire				INTERFACE
   2386       1.1      gwr  **
   2387      1.47  thorpej  * Clear the wired attribute of the specified page.
   2388       1.1      gwr  *
   2389       1.1      gwr  * This function is called from vm_fault.c to unwire
   2390      1.47  thorpej  * a mapping.
   2391       1.1      gwr  */
   2392      1.86      chs void
   2393      1.86      chs pmap_unwire(pmap_t pmap, vaddr_t va)
   2394       1.1      gwr {
   2395       1.1      gwr 	int a_idx, b_idx, c_idx;
   2396       1.1      gwr 	a_tmgr_t *a_tbl;
   2397       1.1      gwr 	b_tmgr_t *b_tbl;
   2398       1.1      gwr 	c_tmgr_t *c_tbl;
   2399       1.1      gwr 	mmu_short_pte_t *pte;
   2400       1.1      gwr 
   2401       1.1      gwr 	/* Kernel mappings always remain wired. */
   2402       1.1      gwr 	if (pmap == pmap_kernel())
   2403       1.1      gwr 		return;
   2404       1.1      gwr 
   2405       1.7      gwr 	/*
   2406       1.7      gwr 	 * Walk through the tables.  If the walk terminates without
   2407       1.1      gwr 	 * a valid PTE then the address wasn't wired in the first place.
   2408       1.1      gwr 	 * Return immediately.
   2409       1.1      gwr 	 */
   2410       1.1      gwr 	if (pmap_stroll(pmap, va, &a_tbl, &b_tbl, &c_tbl, &pte, &a_idx,
   2411  1.88.2.3     yamt 		&b_idx, &c_idx) == false)
   2412       1.1      gwr 		return;
   2413       1.1      gwr 
   2414       1.1      gwr 
   2415       1.1      gwr 	/* Is the PTE wired?  If not, return. */
   2416       1.1      gwr 	if (!(pte->attr.raw & MMU_SHORT_PTE_WIRED))
   2417       1.1      gwr 		return;
   2418       1.1      gwr 
   2419       1.1      gwr 	/* Remove the wiring bit. */
   2420       1.1      gwr 	pte->attr.raw &= ~(MMU_SHORT_PTE_WIRED);
   2421       1.1      gwr 
   2422       1.7      gwr 	/*
   2423       1.7      gwr 	 * Decrement the wired entry count in the C table.
   2424       1.1      gwr 	 * If it reaches zero the following things happen:
   2425       1.1      gwr 	 * 1. The table no longer has any wired entries and is considered
   2426       1.1      gwr 	 *    unwired.
   2427       1.1      gwr 	 * 2. It is placed on the available queue.
   2428       1.1      gwr 	 * 3. The parent table's wired entry count is decremented.
   2429       1.1      gwr 	 * 4. If it reaches zero, this process repeats at step 1 and
   2430       1.1      gwr 	 *    stops at after reaching the A table.
   2431       1.1      gwr 	 */
   2432       1.7      gwr 	if (--c_tbl->ct_wcnt == 0) {
   2433       1.1      gwr 		TAILQ_INSERT_TAIL(&c_pool, c_tbl, ct_link);
   2434       1.7      gwr 		if (--b_tbl->bt_wcnt == 0) {
   2435       1.1      gwr 			TAILQ_INSERT_TAIL(&b_pool, b_tbl, bt_link);
   2436       1.7      gwr 			if (--a_tbl->at_wcnt == 0) {
   2437       1.1      gwr 				TAILQ_INSERT_TAIL(&a_pool, a_tbl, at_link);
   2438       1.1      gwr 			}
   2439       1.1      gwr 		}
   2440       1.1      gwr 	}
   2441       1.1      gwr }
   2442       1.1      gwr 
   2443       1.1      gwr /* pmap_copy				INTERFACE
   2444       1.1      gwr  **
   2445       1.1      gwr  * Copy the mappings of a range of addresses in one pmap, into
   2446       1.1      gwr  * the destination address of another.
   2447       1.1      gwr  *
   2448       1.1      gwr  * This routine is advisory.  Should we one day decide that MMU tables
   2449       1.1      gwr  * may be shared by more than one pmap, this function should be used to
   2450       1.1      gwr  * link them together.  Until that day however, we do nothing.
   2451       1.1      gwr  */
   2452       1.1      gwr void
   2453      1.86      chs pmap_copy(pmap_t pmap_a, pmap_t pmap_b, vaddr_t dst, vsize_t len, vaddr_t src)
   2454       1.1      gwr {
   2455  1.88.2.2     yamt 
   2456       1.1      gwr 	/* not implemented. */
   2457       1.1      gwr }
   2458       1.1      gwr 
   2459       1.1      gwr /* pmap_copy_page			INTERFACE
   2460       1.1      gwr  **
   2461       1.1      gwr  * Copy the contents of one physical page into another.
   2462       1.1      gwr  *
   2463       1.7      gwr  * This function makes use of two virtual pages allocated in pmap_bootstrap()
   2464      1.24   jeremy  * to map the two specified physical pages into the kernel address space.
   2465       1.7      gwr  *
   2466       1.7      gwr  * Note: We could use the transparent translation registers to make the
   2467       1.7      gwr  * mappings.  If we do so, be sure to disable interrupts before using them.
   2468       1.1      gwr  */
   2469      1.86      chs void
   2470      1.86      chs pmap_copy_page(paddr_t srcpa, paddr_t dstpa)
   2471       1.1      gwr {
   2472      1.69      chs 	vaddr_t srcva, dstva;
   2473      1.23   jeremy 	int s;
   2474      1.24   jeremy 
   2475      1.24   jeremy 	srcva = tmp_vpages[0];
   2476      1.24   jeremy 	dstva = tmp_vpages[1];
   2477       1.1      gwr 
   2478      1.58  thorpej 	s = splvm();
   2479      1.69      chs #ifdef DIAGNOSTIC
   2480      1.24   jeremy 	if (tmp_vpages_inuse++)
   2481      1.24   jeremy 		panic("pmap_copy_page: temporary vpages are in use.");
   2482      1.69      chs #endif
   2483      1.23   jeremy 
   2484      1.23   jeremy 	/* Map pages as non-cacheable to avoid cache polution? */
   2485      1.69      chs 	pmap_kenter_pa(srcva, srcpa, VM_PROT_READ);
   2486  1.88.2.2     yamt 	pmap_kenter_pa(dstva, dstpa, VM_PROT_READ | VM_PROT_WRITE);
   2487       1.7      gwr 
   2488      1.79  thorpej 	/* Hand-optimized version of bcopy(src, dst, PAGE_SIZE) */
   2489  1.88.2.2     yamt 	copypage((char *)srcva, (char *)dstva);
   2490      1.24   jeremy 
   2491      1.79  thorpej 	pmap_kremove(srcva, PAGE_SIZE);
   2492      1.79  thorpej 	pmap_kremove(dstva, PAGE_SIZE);
   2493      1.24   jeremy 
   2494      1.69      chs #ifdef DIAGNOSTIC
   2495      1.24   jeremy 	--tmp_vpages_inuse;
   2496      1.69      chs #endif
   2497      1.23   jeremy 	splx(s);
   2498       1.1      gwr }
   2499       1.1      gwr 
   2500       1.1      gwr /* pmap_zero_page			INTERFACE
   2501       1.1      gwr  **
   2502       1.1      gwr  * Zero the contents of the specified physical page.
   2503       1.1      gwr  *
   2504       1.7      gwr  * Uses one of the virtual pages allocated in pmap_boostrap()
   2505      1.24   jeremy  * to map the specified page into the kernel address space.
   2506       1.1      gwr  */
   2507      1.86      chs void
   2508      1.86      chs pmap_zero_page(paddr_t dstpa)
   2509       1.1      gwr {
   2510      1.69      chs 	vaddr_t dstva;
   2511      1.23   jeremy 	int s;
   2512      1.23   jeremy 
   2513      1.24   jeremy 	dstva = tmp_vpages[1];
   2514      1.58  thorpej 	s = splvm();
   2515      1.69      chs #ifdef DIAGNOSTIC
   2516      1.26   jeremy 	if (tmp_vpages_inuse++)
   2517      1.24   jeremy 		panic("pmap_zero_page: temporary vpages are in use.");
   2518      1.69      chs #endif
   2519      1.24   jeremy 
   2520      1.24   jeremy 	/* The comments in pmap_copy_page() above apply here also. */
   2521  1.88.2.2     yamt 	pmap_kenter_pa(dstva, dstpa, VM_PROT_READ | VM_PROT_WRITE);
   2522      1.24   jeremy 
   2523      1.79  thorpej 	/* Hand-optimized version of bzero(ptr, PAGE_SIZE) */
   2524  1.88.2.2     yamt 	zeropage((char *)dstva);
   2525       1.1      gwr 
   2526      1.79  thorpej 	pmap_kremove(dstva, PAGE_SIZE);
   2527      1.69      chs #ifdef DIAGNOSTIC
   2528      1.24   jeremy 	--tmp_vpages_inuse;
   2529      1.69      chs #endif
   2530      1.23   jeremy 	splx(s);
   2531       1.1      gwr }
   2532       1.1      gwr 
   2533       1.1      gwr /* pmap_collect			INTERFACE
   2534       1.1      gwr  **
   2535       1.7      gwr  * Called from the VM system when we are about to swap out
   2536       1.7      gwr  * the process using this pmap.  This should give up any
   2537       1.7      gwr  * resources held here, including all its MMU tables.
   2538       1.1      gwr  */
   2539      1.86      chs void
   2540      1.86      chs pmap_collect(pmap_t pmap)
   2541       1.1      gwr {
   2542  1.88.2.2     yamt 
   2543       1.7      gwr 	/* XXX - todo... */
   2544       1.1      gwr }
   2545       1.1      gwr 
   2546  1.88.2.2     yamt /* pmap_pinit			INTERNAL
   2547  1.88.2.2     yamt  **
   2548  1.88.2.2     yamt  * Initialize a pmap structure.
   2549  1.88.2.2     yamt  */
   2550  1.88.2.2     yamt static INLINE void
   2551  1.88.2.2     yamt pmap_pinit(pmap_t pmap)
   2552  1.88.2.2     yamt {
   2553  1.88.2.2     yamt 
   2554  1.88.2.2     yamt 	memset(pmap, 0, sizeof(struct pmap));
   2555  1.88.2.2     yamt 	pmap->pm_a_tmgr = NULL;
   2556  1.88.2.2     yamt 	pmap->pm_a_phys = kernAphys;
   2557  1.88.2.2     yamt 	pmap->pm_refcount = 1;
   2558  1.88.2.2     yamt 	simple_lock_init(&pmap->pm_lock);
   2559  1.88.2.2     yamt }
   2560  1.88.2.2     yamt 
   2561       1.1      gwr /* pmap_create			INTERFACE
   2562       1.1      gwr  **
   2563       1.1      gwr  * Create and return a pmap structure.
   2564       1.1      gwr  */
   2565      1.86      chs pmap_t
   2566      1.86      chs pmap_create(void)
   2567       1.1      gwr {
   2568       1.1      gwr 	pmap_t	pmap;
   2569       1.1      gwr 
   2570      1.56  tsutsui 	pmap = pool_get(&pmap_pmap_pool, PR_WAITOK);
   2571       1.1      gwr 	pmap_pinit(pmap);
   2572       1.1      gwr 	return pmap;
   2573       1.1      gwr }
   2574       1.1      gwr 
   2575  1.88.2.2     yamt /* pmap_release				INTERNAL
   2576       1.1      gwr  **
   2577       1.1      gwr  * Release any resources held by the given pmap.
   2578       1.1      gwr  *
   2579       1.1      gwr  * This is the reverse analog to pmap_pinit.  It does not
   2580       1.1      gwr  * necessarily mean for the pmap structure to be deallocated,
   2581       1.1      gwr  * as in pmap_destroy.
   2582       1.1      gwr  */
   2583  1.88.2.2     yamt static INLINE void
   2584      1.86      chs pmap_release(pmap_t pmap)
   2585       1.1      gwr {
   2586  1.88.2.2     yamt 
   2587       1.7      gwr 	/*
   2588       1.7      gwr 	 * As long as the pmap contains no mappings,
   2589       1.1      gwr 	 * which always should be the case whenever
   2590       1.1      gwr 	 * this function is called, there really should
   2591       1.1      gwr 	 * be nothing to do.
   2592       1.1      gwr 	 */
   2593       1.1      gwr #ifdef	PMAP_DEBUG
   2594       1.1      gwr 	if (pmap == pmap_kernel())
   2595       1.9      gwr 		panic("pmap_release: kernel pmap");
   2596       1.1      gwr #endif
   2597       1.9      gwr 	/*
   2598       1.9      gwr 	 * XXX - If this pmap has an A table, give it back.
   2599       1.9      gwr 	 * The pmap SHOULD be empty by now, and pmap_remove
   2600       1.9      gwr 	 * should have already given back the A table...
   2601       1.9      gwr 	 * However, I see:  pmap->pm_a_tmgr->at_ecnt == 1
   2602       1.9      gwr 	 * at this point, which means some mapping was not
   2603       1.9      gwr 	 * removed when it should have been. -gwr
   2604       1.9      gwr 	 */
   2605       1.7      gwr 	if (pmap->pm_a_tmgr != NULL) {
   2606       1.9      gwr 		/* First make sure we are not using it! */
   2607       1.9      gwr 		if (kernel_crp.rp_addr == pmap->pm_a_phys) {
   2608       1.9      gwr 			kernel_crp.rp_addr = kernAphys;
   2609       1.9      gwr 			loadcrp(&kernel_crp);
   2610       1.9      gwr 		}
   2611      1.13      gwr #ifdef	PMAP_DEBUG /* XXX - todo! */
   2612      1.13      gwr 		/* XXX - Now complain... */
   2613      1.13      gwr 		printf("pmap_release: still have table\n");
   2614      1.13      gwr 		Debugger();
   2615      1.13      gwr #endif
   2616  1.88.2.3     yamt 		free_a_table(pmap->pm_a_tmgr, true);
   2617       1.7      gwr 		pmap->pm_a_tmgr = NULL;
   2618       1.7      gwr 		pmap->pm_a_phys = kernAphys;
   2619       1.7      gwr 	}
   2620       1.1      gwr }
   2621       1.1      gwr 
   2622       1.1      gwr /* pmap_reference			INTERFACE
   2623       1.1      gwr  **
   2624       1.1      gwr  * Increment the reference count of a pmap.
   2625       1.1      gwr  */
   2626      1.86      chs void
   2627      1.86      chs pmap_reference(pmap_t pmap)
   2628       1.1      gwr {
   2629      1.55  tsutsui 	pmap_lock(pmap);
   2630      1.55  tsutsui 	pmap_add_ref(pmap);
   2631      1.55  tsutsui 	pmap_unlock(pmap);
   2632       1.1      gwr }
   2633       1.1      gwr 
   2634       1.1      gwr /* pmap_dereference			INTERNAL
   2635       1.1      gwr  **
   2636       1.1      gwr  * Decrease the reference count on the given pmap
   2637       1.1      gwr  * by one and return the current count.
   2638       1.1      gwr  */
   2639  1.88.2.2     yamt static INLINE int
   2640      1.86      chs pmap_dereference(pmap_t pmap)
   2641       1.1      gwr {
   2642       1.1      gwr 	int rtn;
   2643       1.1      gwr 
   2644      1.55  tsutsui 	pmap_lock(pmap);
   2645      1.55  tsutsui 	rtn = pmap_del_ref(pmap);
   2646      1.55  tsutsui 	pmap_unlock(pmap);
   2647       1.1      gwr 
   2648       1.1      gwr 	return rtn;
   2649       1.1      gwr }
   2650       1.1      gwr 
   2651       1.1      gwr /* pmap_destroy			INTERFACE
   2652       1.1      gwr  **
   2653       1.1      gwr  * Decrement a pmap's reference count and delete
   2654       1.1      gwr  * the pmap if it becomes zero.  Will be called
   2655       1.1      gwr  * only after all mappings have been removed.
   2656       1.1      gwr  */
   2657      1.86      chs void
   2658      1.86      chs pmap_destroy(pmap_t pmap)
   2659       1.1      gwr {
   2660  1.88.2.2     yamt 
   2661       1.1      gwr 	if (pmap_dereference(pmap) == 0) {
   2662       1.1      gwr 		pmap_release(pmap);
   2663      1.56  tsutsui 		pool_put(&pmap_pmap_pool, pmap);
   2664       1.1      gwr 	}
   2665       1.1      gwr }
   2666       1.1      gwr 
   2667       1.1      gwr /* pmap_is_referenced			INTERFACE
   2668       1.1      gwr  **
   2669       1.1      gwr  * Determine if the given physical page has been
   2670       1.1      gwr  * referenced (read from [or written to.])
   2671       1.1      gwr  */
   2672  1.88.2.3     yamt bool
   2673      1.86      chs pmap_is_referenced(struct vm_page *pg)
   2674       1.1      gwr {
   2675      1.49      chs 	paddr_t   pa = VM_PAGE_TO_PHYS(pg);
   2676       1.1      gwr 	pv_t      *pv;
   2677      1.69      chs 	int       idx;
   2678       1.1      gwr 
   2679       1.7      gwr 	/*
   2680       1.7      gwr 	 * Check the flags on the pv head.  If they are set,
   2681       1.1      gwr 	 * return immediately.  Otherwise a search must be done.
   2682       1.7      gwr 	 */
   2683      1.69      chs 
   2684      1.69      chs 	pv = pa2pv(pa);
   2685       1.1      gwr 	if (pv->pv_flags & PV_FLAGS_USED)
   2686  1.88.2.3     yamt 		return true;
   2687      1.32      gwr 
   2688      1.32      gwr 	/*
   2689      1.32      gwr 	 * Search through all pv elements pointing
   2690      1.32      gwr 	 * to this page and query their reference bits
   2691      1.32      gwr 	 */
   2692      1.32      gwr 
   2693      1.69      chs 	for (idx = pv->pv_idx; idx != PVE_EOL; idx = pvebase[idx].pve_next) {
   2694      1.32      gwr 		if (MMU_PTE_USED(kernCbase[idx])) {
   2695  1.88.2.3     yamt 			return true;
   2696      1.32      gwr 		}
   2697       1.7      gwr 	}
   2698  1.88.2.3     yamt 	return false;
   2699       1.1      gwr }
   2700       1.1      gwr 
   2701       1.1      gwr /* pmap_is_modified			INTERFACE
   2702       1.1      gwr  **
   2703       1.1      gwr  * Determine if the given physical page has been
   2704       1.1      gwr  * modified (written to.)
   2705       1.1      gwr  */
   2706  1.88.2.3     yamt bool
   2707      1.86      chs pmap_is_modified(struct vm_page *pg)
   2708       1.1      gwr {
   2709      1.49      chs 	paddr_t   pa = VM_PAGE_TO_PHYS(pg);
   2710       1.1      gwr 	pv_t      *pv;
   2711      1.69      chs 	int       idx;
   2712       1.1      gwr 
   2713       1.1      gwr 	/* see comments in pmap_is_referenced() */
   2714       1.1      gwr 	pv = pa2pv(pa);
   2715      1.32      gwr 	if (pv->pv_flags & PV_FLAGS_MDFY)
   2716  1.88.2.3     yamt 		return true;
   2717      1.32      gwr 
   2718      1.32      gwr 	for (idx = pv->pv_idx;
   2719      1.32      gwr 		 idx != PVE_EOL;
   2720      1.32      gwr 		 idx = pvebase[idx].pve_next) {
   2721      1.32      gwr 
   2722      1.32      gwr 		if (MMU_PTE_MODIFIED(kernCbase[idx])) {
   2723  1.88.2.3     yamt 			return true;
   2724      1.32      gwr 		}
   2725       1.7      gwr 	}
   2726       1.7      gwr 
   2727  1.88.2.3     yamt 	return false;
   2728       1.1      gwr }
   2729       1.1      gwr 
   2730       1.1      gwr /* pmap_page_protect			INTERFACE
   2731       1.1      gwr  **
   2732       1.1      gwr  * Applies the given protection to all mappings to the given
   2733       1.1      gwr  * physical page.
   2734       1.1      gwr  */
   2735      1.86      chs void
   2736      1.86      chs pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   2737       1.1      gwr {
   2738      1.49      chs 	paddr_t   pa = VM_PAGE_TO_PHYS(pg);
   2739       1.1      gwr 	pv_t      *pv;
   2740      1.69      chs 	int       idx;
   2741      1.69      chs 	vaddr_t va;
   2742       1.1      gwr 	struct mmu_short_pte_struct *pte;
   2743       1.8      gwr 	c_tmgr_t  *c_tbl;
   2744       1.8      gwr 	pmap_t    pmap, curpmap;
   2745       1.1      gwr 
   2746       1.8      gwr 	curpmap = current_pmap();
   2747       1.1      gwr 	pv = pa2pv(pa);
   2748      1.32      gwr 
   2749      1.69      chs 	for (idx = pv->pv_idx; idx != PVE_EOL; idx = pvebase[idx].pve_next) {
   2750       1.7      gwr 		pte = &kernCbase[idx];
   2751       1.1      gwr 		switch (prot) {
   2752       1.1      gwr 			case VM_PROT_ALL:
   2753       1.1      gwr 				/* do nothing */
   2754       1.1      gwr 				break;
   2755       1.7      gwr 			case VM_PROT_EXECUTE:
   2756       1.1      gwr 			case VM_PROT_READ:
   2757       1.1      gwr 			case VM_PROT_READ|VM_PROT_EXECUTE:
   2758       1.8      gwr 				/*
   2759       1.8      gwr 				 * Determine the virtual address mapped by
   2760       1.8      gwr 				 * the PTE and flush ATC entries if necessary.
   2761       1.8      gwr 				 */
   2762       1.8      gwr 				va = pmap_get_pteinfo(idx, &pmap, &c_tbl);
   2763      1.69      chs 				pte->attr.raw |= MMU_SHORT_PTE_WP;
   2764       1.8      gwr 				if (pmap == curpmap || pmap == pmap_kernel())
   2765       1.8      gwr 					TBIS(va);
   2766       1.1      gwr 				break;
   2767       1.1      gwr 			case VM_PROT_NONE:
   2768       1.7      gwr 				/* Save the mod/ref bits. */
   2769       1.7      gwr 				pv->pv_flags |= pte->attr.raw;
   2770       1.7      gwr 				/* Invalidate the PTE. */
   2771       1.7      gwr 				pte->attr.raw = MMU_DT_INVALID;
   2772       1.8      gwr 
   2773       1.8      gwr 				/*
   2774       1.8      gwr 				 * Update table counts.  And flush ATC entries
   2775       1.8      gwr 				 * if necessary.
   2776       1.8      gwr 				 */
   2777       1.8      gwr 				va = pmap_get_pteinfo(idx, &pmap, &c_tbl);
   2778       1.8      gwr 
   2779       1.8      gwr 				/*
   2780       1.8      gwr 				 * If the PTE belongs to the kernel map,
   2781       1.8      gwr 				 * be sure to flush the page it maps.
   2782       1.8      gwr 				 */
   2783       1.8      gwr 				if (pmap == pmap_kernel()) {
   2784       1.8      gwr 					TBIS(va);
   2785       1.8      gwr 				} else {
   2786       1.8      gwr 					/*
   2787       1.8      gwr 					 * The PTE belongs to a user map.
   2788       1.8      gwr 					 * update the entry count in the C
   2789       1.8      gwr 					 * table to which it belongs and flush
   2790       1.8      gwr 					 * the ATC if the mapping belongs to
   2791       1.8      gwr 					 * the current pmap.
   2792       1.8      gwr 					 */
   2793       1.8      gwr 					c_tbl->ct_ecnt--;
   2794       1.8      gwr 					if (pmap == curpmap)
   2795       1.8      gwr 						TBIS(va);
   2796       1.8      gwr 				}
   2797       1.1      gwr 				break;
   2798       1.1      gwr 			default:
   2799       1.1      gwr 				break;
   2800       1.1      gwr 		}
   2801       1.1      gwr 	}
   2802       1.8      gwr 
   2803       1.8      gwr 	/*
   2804       1.8      gwr 	 * If the protection code indicates that all mappings to the page
   2805       1.8      gwr 	 * be removed, truncate the PV list to zero entries.
   2806       1.8      gwr 	 */
   2807       1.7      gwr 	if (prot == VM_PROT_NONE)
   2808       1.7      gwr 		pv->pv_idx = PVE_EOL;
   2809       1.1      gwr }
   2810       1.1      gwr 
   2811       1.7      gwr /* pmap_get_pteinfo		INTERNAL
   2812       1.1      gwr  **
   2813       1.7      gwr  * Called internally to find the pmap and virtual address within that
   2814       1.8      gwr  * map to which the pte at the given index maps.  Also includes the PTE's C
   2815       1.8      gwr  * table manager.
   2816       1.1      gwr  *
   2817       1.7      gwr  * Returns the pmap in the argument provided, and the virtual address
   2818       1.7      gwr  * by return value.
   2819       1.1      gwr  */
   2820      1.86      chs vaddr_t
   2821      1.86      chs pmap_get_pteinfo(u_int idx, pmap_t *pmap, c_tmgr_t **tbl)
   2822       1.1      gwr {
   2823      1.69      chs 	vaddr_t     va = 0;
   2824       1.1      gwr 
   2825       1.7      gwr 	/*
   2826       1.7      gwr 	 * Determine if the PTE is a kernel PTE or a user PTE.
   2827       1.1      gwr 	 */
   2828       1.8      gwr 	if (idx >= NUM_KERN_PTES) {
   2829       1.7      gwr 		/*
   2830       1.7      gwr 		 * The PTE belongs to a user mapping.
   2831       1.7      gwr 		 */
   2832       1.8      gwr 		/* XXX: Would like an inline for this to validate idx... */
   2833      1.26   jeremy 		*tbl = &Ctmgrbase[(idx - NUM_KERN_PTES) / MMU_C_TBL_SIZE];
   2834      1.26   jeremy 
   2835      1.26   jeremy 		*pmap = (*tbl)->ct_pmap;
   2836      1.26   jeremy 		/*
   2837      1.26   jeremy 		 * To find the va to which the PTE maps, we first take
   2838      1.26   jeremy 		 * the table's base virtual address mapping which is stored
   2839      1.26   jeremy 		 * in ct_va.  We then increment this address by a page for
   2840      1.26   jeremy 		 * every slot skipped until we reach the PTE.
   2841      1.26   jeremy 		 */
   2842  1.88.2.2     yamt 		va = (*tbl)->ct_va;
   2843      1.26   jeremy 		va += m68k_ptob(idx % MMU_C_TBL_SIZE);
   2844       1.7      gwr 	} else {
   2845       1.7      gwr 		/*
   2846       1.7      gwr 		 * The PTE belongs to the kernel map.
   2847       1.7      gwr 		 */
   2848       1.8      gwr 		*pmap = pmap_kernel();
   2849       1.8      gwr 
   2850      1.25    veego 		va = m68k_ptob(idx);
   2851       1.7      gwr 		va += KERNBASE;
   2852       1.7      gwr 	}
   2853       1.7      gwr 
   2854       1.1      gwr 	return va;
   2855       1.1      gwr }
   2856       1.1      gwr 
   2857       1.1      gwr /* pmap_clear_modify			INTERFACE
   2858       1.1      gwr  **
   2859       1.1      gwr  * Clear the modification bit on the page at the specified
   2860       1.1      gwr  * physical address.
   2861       1.1      gwr  *
   2862       1.1      gwr  */
   2863  1.88.2.3     yamt bool
   2864      1.86      chs pmap_clear_modify(struct vm_page *pg)
   2865       1.1      gwr {
   2866      1.49      chs 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   2867  1.88.2.3     yamt 	bool rv;
   2868      1.49      chs 
   2869      1.49      chs 	rv = pmap_is_modified(pg);
   2870       1.1      gwr 	pmap_clear_pv(pa, PV_FLAGS_MDFY);
   2871      1.49      chs 	return rv;
   2872       1.1      gwr }
   2873       1.1      gwr 
   2874       1.1      gwr /* pmap_clear_reference			INTERFACE
   2875       1.1      gwr  **
   2876       1.1      gwr  * Clear the referenced bit on the page at the specified
   2877       1.1      gwr  * physical address.
   2878       1.1      gwr  */
   2879  1.88.2.3     yamt bool
   2880      1.86      chs pmap_clear_reference(struct vm_page *pg)
   2881       1.1      gwr {
   2882      1.49      chs 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   2883  1.88.2.3     yamt 	bool rv;
   2884      1.49      chs 
   2885      1.49      chs 	rv = pmap_is_referenced(pg);
   2886       1.1      gwr 	pmap_clear_pv(pa, PV_FLAGS_USED);
   2887      1.49      chs 	return rv;
   2888       1.1      gwr }
   2889       1.1      gwr 
   2890       1.1      gwr /* pmap_clear_pv			INTERNAL
   2891       1.1      gwr  **
   2892       1.1      gwr  * Clears the specified flag from the specified physical address.
   2893       1.1      gwr  * (Used by pmap_clear_modify() and pmap_clear_reference().)
   2894       1.1      gwr  *
   2895       1.1      gwr  * Flag is one of:
   2896       1.1      gwr  *   PV_FLAGS_MDFY - Page modified bit.
   2897       1.1      gwr  *   PV_FLAGS_USED - Page used (referenced) bit.
   2898       1.1      gwr  *
   2899       1.1      gwr  * This routine must not only clear the flag on the pv list
   2900       1.1      gwr  * head.  It must also clear the bit on every pte in the pv
   2901       1.1      gwr  * list associated with the address.
   2902       1.1      gwr  */
   2903      1.86      chs void
   2904      1.86      chs pmap_clear_pv(paddr_t pa, int flag)
   2905       1.1      gwr {
   2906       1.1      gwr 	pv_t      *pv;
   2907      1.69      chs 	int       idx;
   2908      1.69      chs 	vaddr_t   va;
   2909       1.7      gwr 	pmap_t          pmap;
   2910       1.1      gwr 	mmu_short_pte_t *pte;
   2911       1.7      gwr 	c_tmgr_t        *c_tbl;
   2912       1.1      gwr 
   2913       1.1      gwr 	pv = pa2pv(pa);
   2914       1.1      gwr 	pv->pv_flags &= ~(flag);
   2915      1.69      chs 	for (idx = pv->pv_idx; idx != PVE_EOL; idx = pvebase[idx].pve_next) {
   2916       1.7      gwr 		pte = &kernCbase[idx];
   2917       1.1      gwr 		pte->attr.raw &= ~(flag);
   2918      1.69      chs 
   2919       1.7      gwr 		/*
   2920       1.7      gwr 		 * The MC68030 MMU will not set the modified or
   2921       1.7      gwr 		 * referenced bits on any MMU tables for which it has
   2922       1.7      gwr 		 * a cached descriptor with its modify bit set.  To insure
   2923       1.7      gwr 		 * that it will modify these bits on the PTE during the next
   2924       1.7      gwr 		 * time it is written to or read from, we must flush it from
   2925       1.7      gwr 		 * the ATC.
   2926       1.7      gwr 		 *
   2927       1.7      gwr 		 * Ordinarily it is only necessary to flush the descriptor
   2928       1.7      gwr 		 * if it is used in the current address space.  But since I
   2929       1.7      gwr 		 * am not sure that there will always be a notion of
   2930       1.7      gwr 		 * 'the current address space' when this function is called,
   2931       1.7      gwr 		 * I will skip the test and always flush the address.  It
   2932       1.7      gwr 		 * does no harm.
   2933       1.7      gwr 		 */
   2934      1.69      chs 
   2935       1.8      gwr 		va = pmap_get_pteinfo(idx, &pmap, &c_tbl);
   2936       1.7      gwr 		TBIS(va);
   2937       1.1      gwr 	}
   2938       1.1      gwr }
   2939       1.1      gwr 
   2940  1.88.2.2     yamt /* pmap_extract_kernel		INTERNAL
   2941  1.88.2.2     yamt  **
   2942  1.88.2.2     yamt  * Extract a translation from the kernel address space.
   2943  1.88.2.2     yamt  */
   2944  1.88.2.3     yamt static INLINE bool
   2945  1.88.2.2     yamt pmap_extract_kernel(vaddr_t va, paddr_t *pap)
   2946  1.88.2.2     yamt {
   2947  1.88.2.2     yamt 	mmu_short_pte_t *pte;
   2948  1.88.2.2     yamt 
   2949  1.88.2.2     yamt 	pte = &kernCbase[(u_int)m68k_btop(va - KERNBASE)];
   2950  1.88.2.2     yamt 	if (!MMU_VALID_DT(*pte))
   2951  1.88.2.3     yamt 		return false;
   2952  1.88.2.2     yamt 	if (pap != NULL)
   2953  1.88.2.2     yamt 		*pap = MMU_PTE_PA(*pte);
   2954  1.88.2.3     yamt 	return true;
   2955  1.88.2.2     yamt }
   2956  1.88.2.2     yamt 
   2957       1.1      gwr /* pmap_extract			INTERFACE
   2958       1.1      gwr  **
   2959       1.1      gwr  * Return the physical address mapped by the virtual address
   2960      1.48  thorpej  * in the specified pmap.
   2961       1.1      gwr  *
   2962       1.1      gwr  * Note: this function should also apply an exclusive lock
   2963       1.1      gwr  * on the pmap system during its duration.
   2964       1.1      gwr  */
   2965  1.88.2.3     yamt bool
   2966      1.86      chs pmap_extract(pmap_t pmap, vaddr_t va, paddr_t *pap)
   2967       1.1      gwr {
   2968       1.1      gwr 	int a_idx, b_idx, pte_idx;
   2969       1.1      gwr 	a_tmgr_t	*a_tbl;
   2970       1.1      gwr 	b_tmgr_t	*b_tbl;
   2971       1.1      gwr 	c_tmgr_t	*c_tbl;
   2972       1.1      gwr 	mmu_short_pte_t	*c_pte;
   2973       1.1      gwr 
   2974       1.1      gwr 	if (pmap == pmap_kernel())
   2975      1.48  thorpej 		return pmap_extract_kernel(va, pap);
   2976       1.1      gwr 
   2977       1.1      gwr 	if (pmap_stroll(pmap, va, &a_tbl, &b_tbl, &c_tbl,
   2978  1.88.2.3     yamt 		&c_pte, &a_idx, &b_idx, &pte_idx) == false)
   2979  1.88.2.3     yamt 		return false;
   2980       1.1      gwr 
   2981       1.7      gwr 	if (!MMU_VALID_DT(*c_pte))
   2982  1.88.2.3     yamt 		return false;
   2983       1.7      gwr 
   2984      1.48  thorpej 	if (pap != NULL)
   2985      1.48  thorpej 		*pap = MMU_PTE_PA(*c_pte);
   2986  1.88.2.3     yamt 	return true;
   2987       1.1      gwr }
   2988       1.1      gwr 
   2989       1.1      gwr /* pmap_remove_kernel		INTERNAL
   2990       1.1      gwr  **
   2991       1.1      gwr  * Remove the mapping of a range of virtual addresses from the kernel map.
   2992       1.9      gwr  * The arguments are already page-aligned.
   2993       1.1      gwr  */
   2994  1.88.2.2     yamt static INLINE void
   2995      1.86      chs pmap_remove_kernel(vaddr_t sva, vaddr_t eva)
   2996       1.1      gwr {
   2997       1.9      gwr 	int idx, eidx;
   2998       1.9      gwr 
   2999       1.9      gwr #ifdef	PMAP_DEBUG
   3000       1.9      gwr 	if ((sva & PGOFSET) || (eva & PGOFSET))
   3001       1.9      gwr 		panic("pmap_remove_kernel: alignment");
   3002       1.9      gwr #endif
   3003       1.1      gwr 
   3004      1.25    veego 	idx  = m68k_btop(sva - KERNBASE);
   3005      1.25    veego 	eidx = m68k_btop(eva - KERNBASE);
   3006       1.9      gwr 
   3007      1.24   jeremy 	while (idx < eidx) {
   3008       1.9      gwr 		pmap_remove_pte(&kernCbase[idx++]);
   3009      1.24   jeremy 		TBIS(sva);
   3010      1.79  thorpej 		sva += PAGE_SIZE;
   3011      1.24   jeremy 	}
   3012       1.1      gwr }
   3013       1.1      gwr 
   3014       1.1      gwr /* pmap_remove			INTERFACE
   3015       1.1      gwr  **
   3016       1.1      gwr  * Remove the mapping of a range of virtual addresses from the given pmap.
   3017       1.7      gwr  *
   3018       1.1      gwr  */
   3019      1.86      chs void
   3020      1.88  tsutsui pmap_remove(pmap_t pmap, vaddr_t sva, vaddr_t eva)
   3021       1.1      gwr {
   3022       1.7      gwr 
   3023       1.1      gwr 	if (pmap == pmap_kernel()) {
   3024      1.88  tsutsui 		pmap_remove_kernel(sva, eva);
   3025       1.1      gwr 		return;
   3026       1.1      gwr 	}
   3027       1.1      gwr 
   3028       1.7      gwr 	/*
   3029       1.7      gwr 	 * If the pmap doesn't have an A table of its own, it has no mappings
   3030       1.7      gwr 	 * that can be removed.
   3031       1.1      gwr 	 */
   3032       1.7      gwr 	if (pmap->pm_a_tmgr == NULL)
   3033       1.7      gwr 		return;
   3034       1.7      gwr 
   3035       1.7      gwr 	/*
   3036       1.7      gwr 	 * Remove the specified range from the pmap.  If the function
   3037       1.7      gwr 	 * returns true, the operation removed all the valid mappings
   3038       1.7      gwr 	 * in the pmap and freed its A table.  If this happened to the
   3039       1.7      gwr 	 * currently loaded pmap, the MMU root pointer must be reloaded
   3040       1.7      gwr 	 * with the default 'kernel' map.
   3041       1.7      gwr 	 */
   3042      1.88  tsutsui 	if (pmap_remove_a(pmap->pm_a_tmgr, sva, eva)) {
   3043       1.9      gwr 		if (kernel_crp.rp_addr == pmap->pm_a_phys) {
   3044       1.9      gwr 			kernel_crp.rp_addr = kernAphys;
   3045       1.9      gwr 			loadcrp(&kernel_crp);
   3046       1.9      gwr 			/* will do TLB flush below */
   3047       1.9      gwr 		}
   3048       1.7      gwr 		pmap->pm_a_tmgr = NULL;
   3049       1.7      gwr 		pmap->pm_a_phys = kernAphys;
   3050       1.1      gwr 	}
   3051       1.9      gwr 
   3052       1.9      gwr 	/*
   3053       1.9      gwr 	 * If we just modified the current address space,
   3054       1.9      gwr 	 * make sure to flush the MMU cache.
   3055       1.9      gwr 	 *
   3056       1.9      gwr 	 * XXX - this could be an unecessarily large flush.
   3057       1.9      gwr 	 * XXX - Could decide, based on the size of the VA range
   3058       1.9      gwr 	 * to be removed, whether to flush "by pages" or "all".
   3059       1.9      gwr 	 */
   3060       1.9      gwr 	if (pmap == current_pmap())
   3061       1.9      gwr 		TBIAU();
   3062       1.1      gwr }
   3063       1.1      gwr 
   3064       1.1      gwr /* pmap_remove_a			INTERNAL
   3065       1.1      gwr  **
   3066       1.1      gwr  * This is function number one in a set of three that removes a range
   3067       1.1      gwr  * of memory in the most efficient manner by removing the highest possible
   3068       1.1      gwr  * tables from the memory space.  This particular function attempts to remove
   3069       1.1      gwr  * as many B tables as it can, delegating the remaining fragmented ranges to
   3070       1.1      gwr  * pmap_remove_b().
   3071       1.1      gwr  *
   3072       1.7      gwr  * If the removal operation results in an empty A table, the function returns
   3073  1.88.2.3     yamt  * true.
   3074       1.7      gwr  *
   3075       1.1      gwr  * It's ugly but will do for now.
   3076       1.1      gwr  */
   3077  1.88.2.3     yamt bool
   3078      1.88  tsutsui pmap_remove_a(a_tmgr_t *a_tbl, vaddr_t sva, vaddr_t eva)
   3079       1.1      gwr {
   3080  1.88.2.3     yamt 	bool empty;
   3081       1.1      gwr 	int idx;
   3082      1.69      chs 	vaddr_t nstart, nend;
   3083       1.1      gwr 	b_tmgr_t *b_tbl;
   3084       1.1      gwr 	mmu_long_dte_t  *a_dte;
   3085       1.1      gwr 	mmu_short_dte_t *b_dte;
   3086  1.88.2.2     yamt 	uint8_t at_wired, bt_wired;
   3087       1.8      gwr 
   3088       1.7      gwr 	/*
   3089       1.7      gwr 	 * The following code works with what I call a 'granularity
   3090       1.7      gwr 	 * reduction algorithim'.  A range of addresses will always have
   3091       1.7      gwr 	 * the following properties, which are classified according to
   3092       1.7      gwr 	 * how the range relates to the size of the current granularity
   3093       1.7      gwr 	 * - an A table entry:
   3094       1.7      gwr 	 *
   3095       1.7      gwr 	 *            1 2       3 4
   3096       1.7      gwr 	 * -+---+---+---+---+---+---+---+-
   3097       1.7      gwr 	 * -+---+---+---+---+---+---+---+-
   3098       1.7      gwr 	 *
   3099       1.7      gwr 	 * A range will always start on a granularity boundary, illustrated
   3100       1.7      gwr 	 * by '+' signs in the table above, or it will start at some point
   3101       1.7      gwr 	 * inbetween a granularity boundary, as illustrated by point 1.
   3102       1.7      gwr 	 * The first step in removing a range of addresses is to remove the
   3103       1.7      gwr 	 * range between 1 and 2, the nearest granularity boundary.  This
   3104       1.7      gwr 	 * job is handled by the section of code governed by the
   3105       1.7      gwr 	 * 'if (start < nstart)' statement.
   3106       1.7      gwr 	 *
   3107       1.7      gwr 	 * A range will always encompass zero or more intergral granules,
   3108       1.7      gwr 	 * illustrated by points 2 and 3.  Integral granules are easy to
   3109       1.7      gwr 	 * remove.  The removal of these granules is the second step, and
   3110       1.7      gwr 	 * is handled by the code block 'if (nstart < nend)'.
   3111       1.7      gwr 	 *
   3112       1.7      gwr 	 * Lastly, a range will always end on a granularity boundary,
   3113       1.7      gwr 	 * ill. by point 3, or it will fall just beyond one, ill. by point
   3114       1.7      gwr 	 * 4.  The last step involves removing this range and is handled by
   3115       1.7      gwr 	 * the code block 'if (nend < end)'.
   3116       1.7      gwr 	 */
   3117      1.88  tsutsui 	nstart = MMU_ROUND_UP_A(sva);
   3118      1.88  tsutsui 	nend = MMU_ROUND_A(eva);
   3119       1.1      gwr 
   3120  1.88.2.2     yamt 	at_wired = a_tbl->at_wcnt;
   3121  1.88.2.2     yamt 
   3122      1.88  tsutsui 	if (sva < nstart) {
   3123       1.7      gwr 		/*
   3124       1.7      gwr 		 * This block is executed if the range starts between
   3125       1.7      gwr 		 * a granularity boundary.
   3126       1.7      gwr 		 *
   3127       1.7      gwr 		 * First find the DTE which is responsible for mapping
   3128       1.7      gwr 		 * the start of the range.
   3129       1.7      gwr 		 */
   3130      1.88  tsutsui 		idx = MMU_TIA(sva);
   3131       1.1      gwr 		a_dte = &a_tbl->at_dtbl[idx];
   3132       1.7      gwr 
   3133       1.7      gwr 		/*
   3134       1.7      gwr 		 * If the DTE is valid then delegate the removal of the sub
   3135       1.7      gwr 		 * range to pmap_remove_b(), which can remove addresses at
   3136       1.7      gwr 		 * a finer granularity.
   3137       1.7      gwr 		 */
   3138       1.1      gwr 		if (MMU_VALID_DT(*a_dte)) {
   3139       1.7      gwr 			b_dte = mmu_ptov(a_dte->addr.raw);
   3140       1.1      gwr 			b_tbl = mmuB2tmgr(b_dte);
   3141  1.88.2.2     yamt 			bt_wired = b_tbl->bt_wcnt;
   3142       1.7      gwr 
   3143       1.7      gwr 			/*
   3144       1.7      gwr 			 * The sub range to be removed starts at the start
   3145       1.7      gwr 			 * of the full range we were asked to remove, and ends
   3146       1.7      gwr 			 * at the greater of:
   3147       1.7      gwr 			 * 1. The end of the full range, -or-
   3148       1.7      gwr 			 * 2. The end of the full range, rounded down to the
   3149       1.7      gwr 			 *    nearest granularity boundary.
   3150       1.7      gwr 			 */
   3151      1.88  tsutsui 			if (eva < nstart)
   3152      1.88  tsutsui 				empty = pmap_remove_b(b_tbl, sva, eva);
   3153       1.7      gwr 			else
   3154      1.88  tsutsui 				empty = pmap_remove_b(b_tbl, sva, nstart);
   3155       1.7      gwr 
   3156       1.7      gwr 			/*
   3157  1.88.2.2     yamt 			 * If the child table no longer has wired entries,
   3158  1.88.2.2     yamt 			 * decrement wired entry count.
   3159  1.88.2.2     yamt 			 */
   3160  1.88.2.2     yamt 			if (bt_wired && b_tbl->bt_wcnt == 0)
   3161  1.88.2.2     yamt 				a_tbl->at_wcnt--;
   3162  1.88.2.2     yamt 
   3163  1.88.2.2     yamt 			/*
   3164       1.7      gwr 			 * If the removal resulted in an empty B table,
   3165       1.7      gwr 			 * invalidate the DTE that points to it and decrement
   3166       1.7      gwr 			 * the valid entry count of the A table.
   3167       1.7      gwr 			 */
   3168       1.7      gwr 			if (empty) {
   3169       1.7      gwr 				a_dte->attr.raw = MMU_DT_INVALID;
   3170       1.7      gwr 				a_tbl->at_ecnt--;
   3171       1.1      gwr 			}
   3172       1.1      gwr 		}
   3173       1.7      gwr 		/*
   3174       1.7      gwr 		 * If the DTE is invalid, the address range is already non-
   3175      1.68      wiz 		 * existent and can simply be skipped.
   3176       1.7      gwr 		 */
   3177       1.1      gwr 	}
   3178       1.1      gwr 	if (nstart < nend) {
   3179       1.7      gwr 		/*
   3180       1.8      gwr 		 * This block is executed if the range spans a whole number
   3181       1.7      gwr 		 * multiple of granules (A table entries.)
   3182       1.7      gwr 		 *
   3183       1.7      gwr 		 * First find the DTE which is responsible for mapping
   3184       1.7      gwr 		 * the start of the first granule involved.
   3185       1.7      gwr 		 */
   3186       1.1      gwr 		idx = MMU_TIA(nstart);
   3187       1.1      gwr 		a_dte = &a_tbl->at_dtbl[idx];
   3188       1.7      gwr 
   3189       1.7      gwr 		/*
   3190       1.7      gwr 		 * Remove entire sub-granules (B tables) one at a time,
   3191       1.7      gwr 		 * until reaching the end of the range.
   3192       1.7      gwr 		 */
   3193       1.7      gwr 		for (; nstart < nend; a_dte++, nstart += MMU_TIA_RANGE)
   3194       1.1      gwr 			if (MMU_VALID_DT(*a_dte)) {
   3195       1.7      gwr 				/*
   3196       1.7      gwr 				 * Find the B table manager for the
   3197       1.7      gwr 				 * entry and free it.
   3198       1.7      gwr 				 */
   3199       1.7      gwr 				b_dte = mmu_ptov(a_dte->addr.raw);
   3200       1.1      gwr 				b_tbl = mmuB2tmgr(b_dte);
   3201  1.88.2.2     yamt 				bt_wired = b_tbl->bt_wcnt;
   3202  1.88.2.2     yamt 
   3203  1.88.2.3     yamt 				free_b_table(b_tbl, true);
   3204       1.7      gwr 
   3205       1.7      gwr 				/*
   3206  1.88.2.2     yamt 				 * All child entries has been removed.
   3207  1.88.2.2     yamt 				 * If there were any wired entries in it,
   3208  1.88.2.2     yamt 				 * decrement wired entry count.
   3209  1.88.2.2     yamt 				 */
   3210  1.88.2.2     yamt 				if (bt_wired)
   3211  1.88.2.2     yamt 					a_tbl->at_wcnt--;
   3212  1.88.2.2     yamt 
   3213  1.88.2.2     yamt 				/*
   3214       1.7      gwr 				 * Invalidate the DTE that points to the
   3215       1.7      gwr 				 * B table and decrement the valid entry
   3216       1.7      gwr 				 * count of the A table.
   3217       1.7      gwr 				 */
   3218       1.1      gwr 				a_dte->attr.raw = MMU_DT_INVALID;
   3219       1.1      gwr 				a_tbl->at_ecnt--;
   3220       1.1      gwr 			}
   3221       1.1      gwr 	}
   3222      1.88  tsutsui 	if (nend < eva) {
   3223       1.7      gwr 		/*
   3224       1.7      gwr 		 * This block is executed if the range ends beyond a
   3225       1.7      gwr 		 * granularity boundary.
   3226       1.7      gwr 		 *
   3227       1.7      gwr 		 * First find the DTE which is responsible for mapping
   3228       1.7      gwr 		 * the start of the nearest (rounded down) granularity
   3229       1.7      gwr 		 * boundary.
   3230       1.7      gwr 		 */
   3231       1.1      gwr 		idx = MMU_TIA(nend);
   3232       1.1      gwr 		a_dte = &a_tbl->at_dtbl[idx];
   3233       1.7      gwr 
   3234       1.7      gwr 		/*
   3235       1.7      gwr 		 * If the DTE is valid then delegate the removal of the sub
   3236       1.7      gwr 		 * range to pmap_remove_b(), which can remove addresses at
   3237       1.7      gwr 		 * a finer granularity.
   3238       1.7      gwr 		 */
   3239       1.1      gwr 		if (MMU_VALID_DT(*a_dte)) {
   3240       1.7      gwr 			/*
   3241       1.7      gwr 			 * Find the B table manager for the entry
   3242       1.7      gwr 			 * and hand it to pmap_remove_b() along with
   3243       1.7      gwr 			 * the sub range.
   3244       1.7      gwr 			 */
   3245       1.7      gwr 			b_dte = mmu_ptov(a_dte->addr.raw);
   3246       1.1      gwr 			b_tbl = mmuB2tmgr(b_dte);
   3247  1.88.2.2     yamt 			bt_wired = b_tbl->bt_wcnt;
   3248       1.7      gwr 
   3249      1.88  tsutsui 			empty = pmap_remove_b(b_tbl, nend, eva);
   3250       1.7      gwr 
   3251       1.7      gwr 			/*
   3252  1.88.2.2     yamt 			 * If the child table no longer has wired entries,
   3253  1.88.2.2     yamt 			 * decrement wired entry count.
   3254  1.88.2.2     yamt 			 */
   3255  1.88.2.2     yamt 			if (bt_wired && b_tbl->bt_wcnt == 0)
   3256  1.88.2.2     yamt 				a_tbl->at_wcnt--;
   3257  1.88.2.2     yamt 			/*
   3258       1.7      gwr 			 * If the removal resulted in an empty B table,
   3259       1.7      gwr 			 * invalidate the DTE that points to it and decrement
   3260       1.7      gwr 			 * the valid entry count of the A table.
   3261       1.7      gwr 			 */
   3262       1.7      gwr 			if (empty) {
   3263       1.7      gwr 				a_dte->attr.raw = MMU_DT_INVALID;
   3264       1.7      gwr 				a_tbl->at_ecnt--;
   3265       1.7      gwr 			}
   3266       1.1      gwr 		}
   3267       1.1      gwr 	}
   3268       1.7      gwr 
   3269       1.7      gwr 	/*
   3270       1.7      gwr 	 * If there are no more entries in the A table, release it
   3271  1.88.2.3     yamt 	 * back to the available pool and return true.
   3272       1.7      gwr 	 */
   3273       1.7      gwr 	if (a_tbl->at_ecnt == 0) {
   3274  1.88.2.2     yamt 		KASSERT(a_tbl->at_wcnt == 0);
   3275       1.7      gwr 		a_tbl->at_parent = NULL;
   3276  1.88.2.2     yamt 		if (!at_wired)
   3277  1.88.2.2     yamt 			TAILQ_REMOVE(&a_pool, a_tbl, at_link);
   3278       1.7      gwr 		TAILQ_INSERT_HEAD(&a_pool, a_tbl, at_link);
   3279  1.88.2.3     yamt 		empty = true;
   3280       1.7      gwr 	} else {
   3281  1.88.2.2     yamt 		/*
   3282  1.88.2.2     yamt 		 * If the table doesn't have wired entries any longer
   3283  1.88.2.2     yamt 		 * but still has unwired entries, put it back into
   3284  1.88.2.2     yamt 		 * the available queue.
   3285  1.88.2.2     yamt 		 */
   3286  1.88.2.2     yamt 		if (at_wired && a_tbl->at_wcnt == 0)
   3287  1.88.2.2     yamt 			TAILQ_INSERT_TAIL(&a_pool, a_tbl, at_link);
   3288  1.88.2.3     yamt 		empty = false;
   3289       1.7      gwr 	}
   3290       1.7      gwr 
   3291       1.7      gwr 	return empty;
   3292       1.1      gwr }
   3293       1.1      gwr 
   3294       1.1      gwr /* pmap_remove_b			INTERNAL
   3295       1.1      gwr  **
   3296       1.1      gwr  * Remove a range of addresses from an address space, trying to remove entire
   3297       1.1      gwr  * C tables if possible.
   3298       1.7      gwr  *
   3299  1.88.2.3     yamt  * If the operation results in an empty B table, the function returns true.
   3300       1.1      gwr  */
   3301  1.88.2.3     yamt bool
   3302      1.88  tsutsui pmap_remove_b(b_tmgr_t *b_tbl, vaddr_t sva, vaddr_t eva)
   3303       1.1      gwr {
   3304  1.88.2.3     yamt 	bool empty;
   3305       1.1      gwr 	int idx;
   3306      1.69      chs 	vaddr_t nstart, nend, rstart;
   3307       1.1      gwr 	c_tmgr_t *c_tbl;
   3308       1.1      gwr 	mmu_short_dte_t  *b_dte;
   3309       1.1      gwr 	mmu_short_pte_t  *c_dte;
   3310  1.88.2.2     yamt 	uint8_t bt_wired, ct_wired;
   3311       1.1      gwr 
   3312      1.88  tsutsui 	nstart = MMU_ROUND_UP_B(sva);
   3313      1.88  tsutsui 	nend = MMU_ROUND_B(eva);
   3314       1.1      gwr 
   3315  1.88.2.2     yamt 	bt_wired = b_tbl->bt_wcnt;
   3316  1.88.2.2     yamt 
   3317      1.88  tsutsui 	if (sva < nstart) {
   3318      1.88  tsutsui 		idx = MMU_TIB(sva);
   3319       1.1      gwr 		b_dte = &b_tbl->bt_dtbl[idx];
   3320       1.1      gwr 		if (MMU_VALID_DT(*b_dte)) {
   3321       1.7      gwr 			c_dte = mmu_ptov(MMU_DTE_PA(*b_dte));
   3322       1.1      gwr 			c_tbl = mmuC2tmgr(c_dte);
   3323  1.88.2.2     yamt 			ct_wired = c_tbl->ct_wcnt;
   3324  1.88.2.2     yamt 
   3325      1.88  tsutsui 			if (eva < nstart)
   3326      1.88  tsutsui 				empty = pmap_remove_c(c_tbl, sva, eva);
   3327       1.7      gwr 			else
   3328      1.88  tsutsui 				empty = pmap_remove_c(c_tbl, sva, nstart);
   3329  1.88.2.2     yamt 
   3330  1.88.2.2     yamt 			/*
   3331  1.88.2.2     yamt 			 * If the child table no longer has wired entries,
   3332  1.88.2.2     yamt 			 * decrement wired entry count.
   3333  1.88.2.2     yamt 			 */
   3334  1.88.2.2     yamt 			if (ct_wired && c_tbl->ct_wcnt == 0)
   3335  1.88.2.2     yamt 				b_tbl->bt_wcnt--;
   3336  1.88.2.2     yamt 
   3337       1.7      gwr 			if (empty) {
   3338       1.7      gwr 				b_dte->attr.raw = MMU_DT_INVALID;
   3339       1.7      gwr 				b_tbl->bt_ecnt--;
   3340       1.1      gwr 			}
   3341       1.1      gwr 		}
   3342       1.1      gwr 	}
   3343       1.1      gwr 	if (nstart < nend) {
   3344       1.1      gwr 		idx = MMU_TIB(nstart);
   3345       1.1      gwr 		b_dte = &b_tbl->bt_dtbl[idx];
   3346       1.1      gwr 		rstart = nstart;
   3347       1.1      gwr 		while (rstart < nend) {
   3348       1.1      gwr 			if (MMU_VALID_DT(*b_dte)) {
   3349       1.7      gwr 				c_dte = mmu_ptov(MMU_DTE_PA(*b_dte));
   3350       1.1      gwr 				c_tbl = mmuC2tmgr(c_dte);
   3351  1.88.2.2     yamt 				ct_wired = c_tbl->ct_wcnt;
   3352  1.88.2.2     yamt 
   3353  1.88.2.3     yamt 				free_c_table(c_tbl, true);
   3354  1.88.2.2     yamt 
   3355  1.88.2.2     yamt 				/*
   3356  1.88.2.2     yamt 				 * All child entries has been removed.
   3357  1.88.2.2     yamt 				 * If there were any wired entries in it,
   3358  1.88.2.2     yamt 				 * decrement wired entry count.
   3359  1.88.2.2     yamt 				 */
   3360  1.88.2.2     yamt 				if (ct_wired)
   3361  1.88.2.2     yamt 					b_tbl->bt_wcnt--;
   3362  1.88.2.2     yamt 
   3363       1.1      gwr 				b_dte->attr.raw = MMU_DT_INVALID;
   3364       1.1      gwr 				b_tbl->bt_ecnt--;
   3365       1.1      gwr 			}
   3366       1.1      gwr 			b_dte++;
   3367       1.1      gwr 			rstart += MMU_TIB_RANGE;
   3368       1.1      gwr 		}
   3369       1.1      gwr 	}
   3370      1.88  tsutsui 	if (nend < eva) {
   3371       1.1      gwr 		idx = MMU_TIB(nend);
   3372       1.1      gwr 		b_dte = &b_tbl->bt_dtbl[idx];
   3373       1.1      gwr 		if (MMU_VALID_DT(*b_dte)) {
   3374       1.7      gwr 			c_dte = mmu_ptov(MMU_DTE_PA(*b_dte));
   3375       1.1      gwr 			c_tbl = mmuC2tmgr(c_dte);
   3376  1.88.2.2     yamt 			ct_wired = c_tbl->ct_wcnt;
   3377      1.88  tsutsui 			empty = pmap_remove_c(c_tbl, nend, eva);
   3378  1.88.2.2     yamt 
   3379  1.88.2.2     yamt 			/*
   3380  1.88.2.2     yamt 			 * If the child table no longer has wired entries,
   3381  1.88.2.2     yamt 			 * decrement wired entry count.
   3382  1.88.2.2     yamt 			 */
   3383  1.88.2.2     yamt 			if (ct_wired && c_tbl->ct_wcnt == 0)
   3384  1.88.2.2     yamt 				b_tbl->bt_wcnt--;
   3385  1.88.2.2     yamt 
   3386       1.7      gwr 			if (empty) {
   3387       1.7      gwr 				b_dte->attr.raw = MMU_DT_INVALID;
   3388       1.7      gwr 				b_tbl->bt_ecnt--;
   3389       1.7      gwr 			}
   3390       1.1      gwr 		}
   3391       1.1      gwr 	}
   3392       1.7      gwr 
   3393       1.7      gwr 	if (b_tbl->bt_ecnt == 0) {
   3394  1.88.2.2     yamt 		KASSERT(b_tbl->bt_wcnt == 0);
   3395       1.7      gwr 		b_tbl->bt_parent = NULL;
   3396  1.88.2.2     yamt 		if (!bt_wired)
   3397  1.88.2.2     yamt 			TAILQ_REMOVE(&b_pool, b_tbl, bt_link);
   3398       1.7      gwr 		TAILQ_INSERT_HEAD(&b_pool, b_tbl, bt_link);
   3399  1.88.2.3     yamt 		empty = true;
   3400       1.7      gwr 	} else {
   3401  1.88.2.2     yamt 		/*
   3402  1.88.2.2     yamt 		 * If the table doesn't have wired entries any longer
   3403  1.88.2.2     yamt 		 * but still has unwired entries, put it back into
   3404  1.88.2.2     yamt 		 * the available queue.
   3405  1.88.2.2     yamt 		 */
   3406  1.88.2.2     yamt 		if (bt_wired && b_tbl->bt_wcnt == 0)
   3407  1.88.2.2     yamt 			TAILQ_INSERT_TAIL(&b_pool, b_tbl, bt_link);
   3408  1.88.2.2     yamt 
   3409  1.88.2.3     yamt 		empty = false;
   3410       1.7      gwr 	}
   3411       1.7      gwr 
   3412       1.7      gwr 	return empty;
   3413       1.1      gwr }
   3414       1.1      gwr 
   3415       1.1      gwr /* pmap_remove_c			INTERNAL
   3416       1.1      gwr  **
   3417       1.1      gwr  * Remove a range of addresses from the given C table.
   3418       1.1      gwr  */
   3419  1.88.2.3     yamt bool
   3420      1.88  tsutsui pmap_remove_c(c_tmgr_t *c_tbl, vaddr_t sva, vaddr_t eva)
   3421       1.1      gwr {
   3422  1.88.2.3     yamt 	bool empty;
   3423       1.1      gwr 	int idx;
   3424       1.1      gwr 	mmu_short_pte_t *c_pte;
   3425  1.88.2.2     yamt 	uint8_t ct_wired;
   3426       1.1      gwr 
   3427  1.88.2.2     yamt 	ct_wired = c_tbl->ct_wcnt;
   3428  1.88.2.2     yamt 
   3429      1.88  tsutsui 	idx = MMU_TIC(sva);
   3430       1.1      gwr 	c_pte = &c_tbl->ct_dtbl[idx];
   3431  1.88.2.2     yamt 	for (; sva < eva; sva += MMU_PAGE_SIZE, c_pte++) {
   3432       1.7      gwr 		if (MMU_VALID_DT(*c_pte)) {
   3433  1.88.2.2     yamt 			if (c_pte->attr.raw & MMU_SHORT_PTE_WIRED)
   3434  1.88.2.2     yamt 				c_tbl->ct_wcnt--;
   3435       1.1      gwr 			pmap_remove_pte(c_pte);
   3436       1.7      gwr 			c_tbl->ct_ecnt--;
   3437       1.7      gwr 		}
   3438       1.1      gwr 	}
   3439       1.7      gwr 
   3440       1.7      gwr 	if (c_tbl->ct_ecnt == 0) {
   3441  1.88.2.2     yamt 		KASSERT(c_tbl->ct_wcnt == 0);
   3442       1.7      gwr 		c_tbl->ct_parent = NULL;
   3443  1.88.2.2     yamt 		if (!ct_wired)
   3444  1.88.2.2     yamt 			TAILQ_REMOVE(&c_pool, c_tbl, ct_link);
   3445       1.9      gwr 		TAILQ_INSERT_HEAD(&c_pool, c_tbl, ct_link);
   3446  1.88.2.3     yamt 		empty = true;
   3447       1.9      gwr 	} else {
   3448  1.88.2.2     yamt 		/*
   3449  1.88.2.2     yamt 		 * If the table doesn't have wired entries any longer
   3450  1.88.2.2     yamt 		 * but still has unwired entries, put it back into
   3451  1.88.2.2     yamt 		 * the available queue.
   3452  1.88.2.2     yamt 		 */
   3453  1.88.2.2     yamt 		if (ct_wired && c_tbl->ct_wcnt == 0)
   3454  1.88.2.2     yamt 			TAILQ_INSERT_TAIL(&c_pool, c_tbl, ct_link);
   3455  1.88.2.3     yamt 		empty = false;
   3456       1.9      gwr 	}
   3457       1.7      gwr 
   3458       1.9      gwr 	return empty;
   3459       1.1      gwr }
   3460       1.1      gwr 
   3461       1.1      gwr /* pmap_bootstrap_alloc			INTERNAL
   3462       1.1      gwr  **
   3463       1.1      gwr  * Used internally for memory allocation at startup when malloc is not
   3464       1.1      gwr  * available.  This code will fail once it crosses the first memory
   3465       1.1      gwr  * bank boundary on the 3/80.  Hopefully by then however, the VM system
   3466       1.1      gwr  * will be in charge of allocation.
   3467       1.1      gwr  */
   3468       1.1      gwr void *
   3469      1.86      chs pmap_bootstrap_alloc(int size)
   3470       1.1      gwr {
   3471       1.1      gwr 	void *rtn;
   3472       1.1      gwr 
   3473       1.8      gwr #ifdef	PMAP_DEBUG
   3474  1.88.2.3     yamt 	if (bootstrap_alloc_enabled == false) {
   3475       1.7      gwr 		mon_printf("pmap_bootstrap_alloc: disabled\n");
   3476       1.7      gwr 		sunmon_abort();
   3477       1.7      gwr 	}
   3478       1.7      gwr #endif
   3479       1.7      gwr 
   3480       1.1      gwr 	rtn = (void *) virtual_avail;
   3481       1.1      gwr 	virtual_avail += size;
   3482       1.1      gwr 
   3483       1.8      gwr #ifdef	PMAP_DEBUG
   3484       1.7      gwr 	if (virtual_avail > virtual_contig_end) {
   3485       1.7      gwr 		mon_printf("pmap_bootstrap_alloc: out of mem\n");
   3486       1.7      gwr 		sunmon_abort();
   3487       1.1      gwr 	}
   3488       1.7      gwr #endif
   3489       1.1      gwr 
   3490       1.1      gwr 	return rtn;
   3491       1.1      gwr }
   3492       1.1      gwr 
   3493       1.1      gwr /* pmap_bootstap_aalign			INTERNAL
   3494       1.1      gwr  **
   3495       1.7      gwr  * Used to insure that the next call to pmap_bootstrap_alloc() will
   3496       1.7      gwr  * return a chunk of memory aligned to the specified size.
   3497       1.8      gwr  *
   3498       1.8      gwr  * Note: This function will only support alignment sizes that are powers
   3499       1.8      gwr  * of two.
   3500       1.1      gwr  */
   3501      1.86      chs void
   3502      1.86      chs pmap_bootstrap_aalign(int size)
   3503       1.1      gwr {
   3504       1.7      gwr 	int off;
   3505       1.7      gwr 
   3506       1.7      gwr 	off = virtual_avail & (size - 1);
   3507       1.7      gwr 	if (off) {
   3508  1.88.2.2     yamt 		(void)pmap_bootstrap_alloc(size - off);
   3509       1.1      gwr 	}
   3510       1.1      gwr }
   3511       1.7      gwr 
   3512       1.8      gwr /* pmap_pa_exists
   3513       1.8      gwr  **
   3514       1.8      gwr  * Used by the /dev/mem driver to see if a given PA is memory
   3515       1.8      gwr  * that can be mapped.  (The PA is not in a hole.)
   3516       1.8      gwr  */
   3517      1.86      chs int
   3518      1.86      chs pmap_pa_exists(paddr_t pa)
   3519       1.8      gwr {
   3520      1.69      chs 	int i;
   3521      1.21      gwr 
   3522      1.21      gwr 	for (i = 0; i < SUN3X_NPHYS_RAM_SEGS; i++) {
   3523      1.21      gwr 		if ((pa >= avail_mem[i].pmem_start) &&
   3524      1.21      gwr 			(pa <  avail_mem[i].pmem_end))
   3525  1.88.2.2     yamt 			return 1;
   3526      1.21      gwr 		if (avail_mem[i].pmem_next == NULL)
   3527      1.21      gwr 			break;
   3528      1.21      gwr 	}
   3529  1.88.2.2     yamt 	return 0;
   3530       1.8      gwr }
   3531       1.8      gwr 
   3532      1.31      gwr /* Called only from locore.s and pmap.c */
   3533      1.86      chs void	_pmap_switch(pmap_t pmap);
   3534      1.31      gwr 
   3535      1.31      gwr /*
   3536      1.31      gwr  * _pmap_switch			INTERNAL
   3537      1.31      gwr  *
   3538      1.31      gwr  * This is called by locore.s:cpu_switch() when it is
   3539      1.31      gwr  * switching to a new process.  Load new translations.
   3540      1.31      gwr  * Note: done in-line by locore.s unless PMAP_DEBUG
   3541      1.24   jeremy  *
   3542      1.31      gwr  * Note that we do NOT allocate a context here, but
   3543      1.31      gwr  * share the "kernel only" context until we really
   3544      1.31      gwr  * need our own context for user-space mappings in
   3545      1.31      gwr  * pmap_enter_user().  [ s/context/mmu A table/ ]
   3546       1.1      gwr  */
   3547      1.86      chs void
   3548      1.86      chs _pmap_switch(pmap_t pmap)
   3549       1.1      gwr {
   3550       1.7      gwr 	u_long rootpa;
   3551       1.7      gwr 
   3552      1.31      gwr 	/*
   3553      1.31      gwr 	 * Only do reload/flush if we have to.
   3554      1.31      gwr 	 * Note that if the old and new process
   3555      1.31      gwr 	 * were BOTH using the "null" context,
   3556      1.31      gwr 	 * then this will NOT flush the TLB.
   3557      1.31      gwr 	 */
   3558       1.7      gwr 	rootpa = pmap->pm_a_phys;
   3559      1.31      gwr 	if (kernel_crp.rp_addr != rootpa) {
   3560      1.31      gwr 		DPRINT(("pmap_activate(%p)\n", pmap));
   3561       1.7      gwr 		kernel_crp.rp_addr = rootpa;
   3562       1.7      gwr 		loadcrp(&kernel_crp);
   3563       1.8      gwr 		TBIAU();
   3564      1.31      gwr 	}
   3565      1.31      gwr }
   3566      1.31      gwr 
   3567      1.31      gwr /*
   3568      1.31      gwr  * Exported version of pmap_activate().  This is called from the
   3569      1.31      gwr  * machine-independent VM code when a process is given a new pmap.
   3570      1.76  thorpej  * If (p == curlwp) do like cpu_switch would do; otherwise just
   3571      1.31      gwr  * take this as notification that the process has a new pmap.
   3572      1.31      gwr  */
   3573      1.86      chs void
   3574      1.86      chs pmap_activate(struct lwp *l)
   3575      1.31      gwr {
   3576  1.88.2.2     yamt 
   3577      1.76  thorpej 	if (l->l_proc == curproc) {
   3578      1.76  thorpej 		_pmap_switch(l->l_proc->p_vmspace->vm_map.pmap);
   3579       1.7      gwr 	}
   3580       1.1      gwr }
   3581       1.1      gwr 
   3582      1.30  thorpej /*
   3583      1.30  thorpej  * pmap_deactivate			INTERFACE
   3584      1.30  thorpej  **
   3585      1.30  thorpej  * This is called to deactivate the specified process's address space.
   3586      1.30  thorpej  */
   3587      1.86      chs void
   3588      1.86      chs pmap_deactivate(struct lwp *l)
   3589       1.1      gwr {
   3590  1.88.2.2     yamt 
   3591      1.69      chs 	/* Nothing to do. */
   3592       1.1      gwr }
   3593       1.1      gwr 
   3594      1.17      gwr /*
   3595      1.28      gwr  * Fill in the sun3x-specific part of the kernel core header
   3596      1.28      gwr  * for dumpsys().  (See machdep.c for the rest.)
   3597      1.17      gwr  */
   3598      1.86      chs void
   3599      1.86      chs pmap_kcore_hdr(struct sun3x_kcore_hdr *sh)
   3600      1.17      gwr {
   3601      1.17      gwr 	u_long spa, len;
   3602      1.17      gwr 	int i;
   3603      1.20  thorpej 
   3604      1.28      gwr 	sh->pg_frame = MMU_SHORT_PTE_BASEADDR;
   3605      1.28      gwr 	sh->pg_valid = MMU_DT_PAGE;
   3606      1.20  thorpej 	sh->contig_end = virtual_contig_end;
   3607      1.69      chs 	sh->kernCbase = (u_long)kernCbase;
   3608      1.20  thorpej 	for (i = 0; i < SUN3X_NPHYS_RAM_SEGS; i++) {
   3609      1.17      gwr 		spa = avail_mem[i].pmem_start;
   3610      1.25    veego 		spa = m68k_trunc_page(spa);
   3611      1.17      gwr 		len = avail_mem[i].pmem_end - spa;
   3612      1.25    veego 		len = m68k_round_page(len);
   3613      1.20  thorpej 		sh->ram_segs[i].start = spa;
   3614      1.20  thorpej 		sh->ram_segs[i].size  = len;
   3615      1.17      gwr 	}
   3616      1.17      gwr }
   3617      1.17      gwr 
   3618      1.81  thorpej 
   3619      1.81  thorpej /* pmap_virtual_space			INTERFACE
   3620      1.81  thorpej  **
   3621      1.81  thorpej  * Return the current available range of virtual addresses in the
   3622      1.81  thorpej  * arguuments provided.  Only really called once.
   3623      1.81  thorpej  */
   3624      1.86      chs void
   3625      1.86      chs pmap_virtual_space(vaddr_t *vstart, vaddr_t *vend)
   3626      1.81  thorpej {
   3627  1.88.2.2     yamt 
   3628      1.81  thorpej 	*vstart = virtual_avail;
   3629      1.81  thorpej 	*vend = virtual_end;
   3630      1.81  thorpej }
   3631       1.1      gwr 
   3632      1.37      gwr /*
   3633      1.37      gwr  * Provide memory to the VM system.
   3634      1.37      gwr  *
   3635      1.37      gwr  * Assume avail_start is always in the
   3636      1.37      gwr  * first segment as pmap_bootstrap does.
   3637      1.37      gwr  */
   3638      1.86      chs static void
   3639      1.86      chs pmap_page_upload(void)
   3640      1.37      gwr {
   3641      1.69      chs 	paddr_t	a, b;	/* memory range */
   3642      1.37      gwr 	int i;
   3643      1.37      gwr 
   3644      1.37      gwr 	/* Supply the memory in segments. */
   3645      1.37      gwr 	for (i = 0; i < SUN3X_NPHYS_RAM_SEGS; i++) {
   3646      1.37      gwr 		a = atop(avail_mem[i].pmem_start);
   3647      1.37      gwr 		b = atop(avail_mem[i].pmem_end);
   3648      1.37      gwr 		if (i == 0)
   3649      1.37      gwr 			a = atop(avail_start);
   3650      1.60  tsutsui 		if (avail_mem[i].pmem_end > avail_end)
   3651      1.60  tsutsui 			b = atop(avail_end);
   3652      1.37      gwr 
   3653      1.39  thorpej 		uvm_page_physload(a, b, a, b, VM_FREELIST_DEFAULT);
   3654      1.37      gwr 
   3655      1.37      gwr 		if (avail_mem[i].pmem_next == NULL)
   3656      1.37      gwr 			break;
   3657      1.37      gwr 	}
   3658       1.1      gwr }
   3659       1.8      gwr 
   3660       1.8      gwr /* pmap_count			INTERFACE
   3661       1.8      gwr  **
   3662       1.8      gwr  * Return the number of resident (valid) pages in the given pmap.
   3663       1.8      gwr  *
   3664       1.8      gwr  * Note:  If this function is handed the kernel map, it will report
   3665       1.8      gwr  * that it has no mappings.  Hopefully the VM system won't ask for kernel
   3666       1.8      gwr  * map statistics.
   3667       1.8      gwr  */
   3668      1.86      chs segsz_t
   3669      1.86      chs pmap_count(pmap_t pmap, int type)
   3670       1.8      gwr {
   3671       1.8      gwr 	u_int     count;
   3672       1.8      gwr 	int       a_idx, b_idx;
   3673       1.8      gwr 	a_tmgr_t *a_tbl;
   3674       1.8      gwr 	b_tmgr_t *b_tbl;
   3675       1.8      gwr 	c_tmgr_t *c_tbl;
   3676       1.8      gwr 
   3677       1.8      gwr 	/*
   3678       1.8      gwr 	 * If the pmap does not have its own A table manager, it has no
   3679       1.8      gwr 	 * valid entires.
   3680       1.8      gwr 	 */
   3681       1.8      gwr 	if (pmap->pm_a_tmgr == NULL)
   3682       1.8      gwr 		return 0;
   3683       1.8      gwr 
   3684       1.8      gwr 	a_tbl = pmap->pm_a_tmgr;
   3685       1.8      gwr 
   3686       1.8      gwr 	count = 0;
   3687       1.8      gwr 	for (a_idx = 0; a_idx < MMU_TIA(KERNBASE); a_idx++) {
   3688       1.8      gwr 	    if (MMU_VALID_DT(a_tbl->at_dtbl[a_idx])) {
   3689       1.8      gwr 	        b_tbl = mmuB2tmgr(mmu_ptov(a_tbl->at_dtbl[a_idx].addr.raw));
   3690       1.8      gwr 	        for (b_idx = 0; b_idx < MMU_B_TBL_SIZE; b_idx++) {
   3691       1.8      gwr 	            if (MMU_VALID_DT(b_tbl->bt_dtbl[b_idx])) {
   3692       1.8      gwr 	                c_tbl = mmuC2tmgr(
   3693       1.8      gwr 	                    mmu_ptov(MMU_DTE_PA(b_tbl->bt_dtbl[b_idx])));
   3694       1.8      gwr 	                if (type == 0)
   3695       1.8      gwr 	                    /*
   3696       1.8      gwr 	                     * A resident entry count has been requested.
   3697       1.8      gwr 	                     */
   3698       1.8      gwr 	                    count += c_tbl->ct_ecnt;
   3699       1.8      gwr 	                else
   3700       1.8      gwr 	                    /*
   3701       1.8      gwr 	                     * A wired entry count has been requested.
   3702       1.8      gwr 	                     */
   3703       1.8      gwr 	                    count += c_tbl->ct_wcnt;
   3704       1.8      gwr 	            }
   3705       1.8      gwr 	        }
   3706       1.8      gwr 	    }
   3707       1.8      gwr 	}
   3708       1.8      gwr 
   3709       1.8      gwr 	return count;
   3710       1.8      gwr }
   3711       1.8      gwr 
   3712       1.1      gwr /************************ SUN3 COMPATIBILITY ROUTINES ********************
   3713       1.1      gwr  * The following routines are only used by DDB for tricky kernel text    *
   3714       1.1      gwr  * text operations in db_memrw.c.  They are provided for sun3            *
   3715       1.1      gwr  * compatibility.                                                        *
   3716       1.1      gwr  *************************************************************************/
   3717       1.1      gwr /* get_pte			INTERNAL
   3718       1.1      gwr  **
   3719       1.1      gwr  * Return the page descriptor the describes the kernel mapping
   3720       1.1      gwr  * of the given virtual address.
   3721       1.1      gwr  */
   3722      1.86      chs extern u_long ptest_addr(u_long);	/* XXX: locore.s */
   3723      1.86      chs u_int
   3724      1.86      chs get_pte(vaddr_t va)
   3725      1.13      gwr {
   3726      1.13      gwr 	u_long pte_pa;
   3727      1.13      gwr 	mmu_short_pte_t *pte;
   3728      1.13      gwr 
   3729      1.13      gwr 	/* Get the physical address of the PTE */
   3730      1.13      gwr 	pte_pa = ptest_addr(va & ~PGOFSET);
   3731      1.13      gwr 
   3732      1.13      gwr 	/* Convert to a virtual address... */
   3733      1.13      gwr 	pte = (mmu_short_pte_t *) (KERNBASE + pte_pa);
   3734      1.13      gwr 
   3735      1.13      gwr 	/* Make sure it is in our level-C tables... */
   3736      1.13      gwr 	if ((pte < kernCbase) ||
   3737      1.13      gwr 		(pte >= &mmuCbase[NUM_USER_PTES]))
   3738      1.13      gwr 		return 0;
   3739      1.13      gwr 
   3740      1.13      gwr 	/* ... and just return its contents. */
   3741      1.13      gwr 	return (pte->attr.raw);
   3742      1.13      gwr }
   3743      1.13      gwr 
   3744       1.1      gwr 
   3745       1.1      gwr /* set_pte			INTERNAL
   3746       1.1      gwr  **
   3747       1.1      gwr  * Set the page descriptor that describes the kernel mapping
   3748       1.1      gwr  * of the given virtual address.
   3749       1.1      gwr  */
   3750      1.86      chs void
   3751      1.86      chs set_pte(vaddr_t va, u_int pte)
   3752       1.1      gwr {
   3753       1.1      gwr 	u_long idx;
   3754       1.1      gwr 
   3755       1.7      gwr 	if (va < KERNBASE)
   3756       1.7      gwr 		return;
   3757       1.7      gwr 
   3758      1.25    veego 	idx = (unsigned long) m68k_btop(va - KERNBASE);
   3759       1.1      gwr 	kernCbase[idx].attr.raw = pte;
   3760      1.33      gwr 	TBIS(va);
   3761       1.1      gwr }
   3762      1.42       is 
   3763      1.42       is /*
   3764      1.42       is  *	Routine:        pmap_procwr
   3765      1.42       is  *
   3766      1.42       is  *	Function:
   3767      1.42       is  *		Synchronize caches corresponding to [addr, addr+len) in p.
   3768      1.42       is  */
   3769      1.86      chs void
   3770      1.86      chs pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   3771      1.42       is {
   3772  1.88.2.2     yamt 
   3773      1.42       is 	(void)cachectl1(0x80000004, va, len, p);
   3774      1.42       is }
   3775      1.42       is 
   3776       1.7      gwr 
   3777       1.8      gwr #ifdef	PMAP_DEBUG
   3778       1.7      gwr /************************** DEBUGGING ROUTINES **************************
   3779       1.7      gwr  * The following routines are meant to be an aid to debugging the pmap  *
   3780       1.7      gwr  * system.  They are callable from the DDB command line and should be   *
   3781       1.7      gwr  * prepared to be handed unstable or incomplete states of the system.   *
   3782       1.7      gwr  ************************************************************************/
   3783       1.7      gwr 
   3784       1.7      gwr /* pv_list
   3785       1.7      gwr  **
   3786       1.7      gwr  * List all pages found on the pv list for the given physical page.
   3787       1.8      gwr  * To avoid endless loops, the listing will stop at the end of the list
   3788       1.7      gwr  * or after 'n' entries - whichever comes first.
   3789       1.7      gwr  */
   3790      1.86      chs void
   3791      1.86      chs pv_list(paddr_t pa, int n)
   3792       1.7      gwr {
   3793       1.7      gwr 	int  idx;
   3794      1.69      chs 	vaddr_t va;
   3795       1.7      gwr 	pv_t *pv;
   3796       1.7      gwr 	c_tmgr_t *c_tbl;
   3797       1.7      gwr 	pmap_t pmap;
   3798       1.7      gwr 
   3799       1.7      gwr 	pv = pa2pv(pa);
   3800       1.7      gwr 	idx = pv->pv_idx;
   3801      1.69      chs 	for (; idx != PVE_EOL && n > 0; idx = pvebase[idx].pve_next, n--) {
   3802       1.8      gwr 		va = pmap_get_pteinfo(idx, &pmap, &c_tbl);
   3803       1.7      gwr 		printf("idx %d, pmap 0x%x, va 0x%x, c_tbl %x\n",
   3804       1.7      gwr 			idx, (u_int) pmap, (u_int) va, (u_int) c_tbl);
   3805       1.7      gwr 	}
   3806       1.7      gwr }
   3807       1.8      gwr #endif	/* PMAP_DEBUG */
   3808       1.1      gwr 
   3809       1.1      gwr #ifdef NOT_YET
   3810       1.1      gwr /* and maybe not ever */
   3811       1.1      gwr /************************** LOW-LEVEL ROUTINES **************************
   3812      1.78      wiz  * These routines will eventually be re-written into assembly and placed*
   3813       1.1      gwr  * in locore.s.  They are here now as stubs so that the pmap module can *
   3814       1.1      gwr  * be linked as a standalone user program for testing.                  *
   3815       1.1      gwr  ************************************************************************/
   3816       1.1      gwr /* flush_atc_crp			INTERNAL
   3817       1.1      gwr  **
   3818       1.1      gwr  * Flush all page descriptors derived from the given CPU Root Pointer
   3819       1.1      gwr  * (CRP), or 'A' table as it is known here, from the 68851's automatic
   3820       1.1      gwr  * cache.
   3821       1.1      gwr  */
   3822      1.86      chs void
   3823      1.86      chs flush_atc_crp(int a_tbl)
   3824       1.1      gwr {
   3825       1.1      gwr 	mmu_long_rp_t rp;
   3826       1.1      gwr 
   3827       1.1      gwr 	/* Create a temporary root table pointer that points to the
   3828       1.1      gwr 	 * given A table.
   3829       1.1      gwr 	 */
   3830       1.1      gwr 	rp.attr.raw = ~MMU_LONG_RP_LU;
   3831       1.1      gwr 	rp.addr.raw = (unsigned int) a_tbl;
   3832       1.1      gwr 
   3833       1.1      gwr 	mmu_pflushr(&rp);
   3834       1.1      gwr 	/* mmu_pflushr:
   3835       1.1      gwr 	 * 	movel   sp(4)@,a0
   3836       1.1      gwr 	 * 	pflushr a0@
   3837       1.1      gwr 	 *	rts
   3838       1.1      gwr 	 */
   3839       1.1      gwr }
   3840       1.1      gwr #endif /* NOT_YET */
   3841