xref: /freebsd/sys/i386/include/pmap.h (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
1 /*-
2  * Copyright (c) 1991 Regents of the University of California.
3  * All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * the Systems Programming Group of the University of Utah Computer
7  * Science Department and William Jolitz of UUNET Technologies Inc.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  * Derived from hp300 version by Mike Hibler, this version by William
34  * Jolitz uses a recursive map [a pde points to the page directory] to
35  * map the page tables using the pagetables themselves. This is done to
36  * reduce the impact on kernel virtual memory for lots of sparse address
37  * space, and to reduce the cost of memory to each process.
38  *
39  *	from: hp300: @(#)pmap.h	7.2 (Berkeley) 12/16/90
40  *	from: @(#)pmap.h	7.4 (Berkeley) 5/12/91
41  * $FreeBSD$
42  */
43 
44 #ifndef _MACHINE_PMAP_H_
45 #define	_MACHINE_PMAP_H_
46 
47 /*
48  * Page-directory and page-table entries follow this format, with a few
49  * of the fields not present here and there, depending on a lot of things.
50  */
51 				/* ---- Intel Nomenclature ---- */
52 #define	PG_V		0x001	/* P	Valid			*/
53 #define PG_RW		0x002	/* R/W	Read/Write		*/
54 #define PG_U		0x004	/* U/S  User/Supervisor		*/
55 #define	PG_NC_PWT	0x008	/* PWT	Write through		*/
56 #define	PG_NC_PCD	0x010	/* PCD	Cache disable		*/
57 #define PG_A		0x020	/* A	Accessed		*/
58 #define	PG_M		0x040	/* D	Dirty			*/
59 #define	PG_PS		0x080	/* PS	Page size (0=4k,1=4M)	*/
60 #define	PG_PTE_PAT	0x080	/* PAT	PAT index		*/
61 #define	PG_G		0x100	/* G	Global			*/
62 #define	PG_AVAIL1	0x200	/*    /	Available for system	*/
63 #define	PG_AVAIL2	0x400	/*   <	programmers use		*/
64 #define	PG_AVAIL3	0x800	/*    \				*/
65 #define	PG_PDE_PAT	0x1000	/* PAT	PAT index		*/
66 #ifdef PAE
67 #define	PG_NX		(1ull<<63) /* No-execute */
68 #endif
69 
70 
71 /* Our various interpretations of the above */
72 #define PG_W		PG_AVAIL1	/* "Wired" pseudoflag */
73 #define	PG_MANAGED	PG_AVAIL2
74 #ifdef PAE
75 #define	PG_FRAME	(0x000ffffffffff000ull)
76 #define	PG_PS_FRAME	(0x000fffffffe00000ull)
77 #else
78 #define	PG_FRAME	(~PAGE_MASK)
79 #define	PG_PS_FRAME	(0xffc00000)
80 #endif
81 #define	PG_PROT		(PG_RW|PG_U)	/* all protection bits . */
82 #define PG_N		(PG_NC_PWT|PG_NC_PCD)	/* Non-cacheable */
83 
84 /* Page level cache control fields used to determine the PAT type */
85 #define PG_PDE_CACHE	(PG_PDE_PAT | PG_NC_PWT | PG_NC_PCD)
86 #define PG_PTE_CACHE	(PG_PTE_PAT | PG_NC_PWT | PG_NC_PCD)
87 
88 /*
89  * Promotion to a 2 or 4MB (PDE) page mapping requires that the corresponding
90  * 4KB (PTE) page mappings have identical settings for the following fields:
91  */
92 #define PG_PTE_PROMOTE	(PG_MANAGED | PG_W | PG_G | PG_PTE_PAT | \
93 	    PG_M | PG_A | PG_NC_PCD | PG_NC_PWT | PG_U | PG_RW | PG_V)
94 
95 /*
96  * Page Protection Exception bits
97  */
98 
99 #define PGEX_P		0x01	/* Protection violation vs. not present */
100 #define PGEX_W		0x02	/* during a Write cycle */
101 #define PGEX_U		0x04	/* access from User mode (UPL) */
102 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
103 #define PGEX_I		0x10	/* during an instruction fetch */
104 
105 /*
106  * Size of Kernel address space.  This is the number of page table pages
107  * (4MB each) to use for the kernel.  256 pages == 1 Gigabyte.
108  * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
109  * For PAE, the page table page unit size is 2MB.  This means that 512 pages
110  * is 1 Gigabyte.  Double everything.  It must be a multiple of 8 for PAE.
111  */
112 #ifndef KVA_PAGES
113 #ifdef PAE
114 #define KVA_PAGES	512
115 #else
116 #define KVA_PAGES	256
117 #endif
118 #endif
119 
120 /*
121  * Pte related macros
122  */
123 #define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDRSHIFT)|((pti)<<PAGE_SHIFT)))
124 
125 /* Initial number of kernel page tables. */
126 #ifndef NKPT
127 #ifdef PAE
128 /* 152 page tables needed to map 16G (76B "struct vm_page", 2M page tables). */
129 #define	NKPT		240
130 #else
131 /* 18 page tables needed to map 4G (72B "struct vm_page", 4M page tables). */
132 #define	NKPT		30
133 #endif
134 #endif
135 
136 #ifndef NKPDE
137 #define NKPDE	(KVA_PAGES)	/* number of page tables/pde's */
138 #endif
139 
140 /*
141  * The *PTDI values control the layout of virtual memory
142  *
143  * XXX This works for now, but I am not real happy with it, I'll fix it
144  * right after I fix locore.s and the magic 28K hole
145  */
146 #define	KPTDI		(NPDEPTD-NKPDE)	/* start of kernel virtual pde's */
147 #define	PTDPTDI		(KPTDI-NPGPTD)	/* ptd entry that points to ptd! */
148 
149 /*
150  * XXX doesn't really belong here I guess...
151  */
152 #define ISA_HOLE_START    0xa0000
153 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
154 
155 #ifndef LOCORE
156 
157 #include <sys/queue.h>
158 #include <sys/_cpuset.h>
159 #include <sys/_lock.h>
160 #include <sys/_mutex.h>
161 
162 #include <vm/_vm_radix.h>
163 
164 #ifdef PAE
165 
166 typedef uint64_t pdpt_entry_t;
167 typedef uint64_t pd_entry_t;
168 typedef uint64_t pt_entry_t;
169 
170 #define	PTESHIFT	(3)
171 #define	PDESHIFT	(3)
172 
173 #else
174 
175 typedef uint32_t pd_entry_t;
176 typedef uint32_t pt_entry_t;
177 
178 #define	PTESHIFT	(2)
179 #define	PDESHIFT	(2)
180 
181 #endif
182 
183 /*
184  * Address of current address space page table maps and directories.
185  */
186 #ifdef _KERNEL
187 extern pt_entry_t PTmap[];
188 extern pd_entry_t PTD[];
189 extern pd_entry_t PTDpde[];
190 
191 #ifdef PAE
192 extern pdpt_entry_t *IdlePDPT;
193 #endif
194 extern pd_entry_t *IdlePTD;	/* physical address of "Idle" state directory */
195 
196 /*
197  * Translate a virtual address to the kernel virtual address of its page table
198  * entry (PTE).  This can be used recursively.  If the address of a PTE as
199  * previously returned by this macro is itself given as the argument, then the
200  * address of the page directory entry (PDE) that maps the PTE will be
201  * returned.
202  *
203  * This macro may be used before pmap_bootstrap() is called.
204  */
205 #define	vtopte(va)	(PTmap + i386_btop(va))
206 
207 /*
208  * Translate a virtual address to its physical address.
209  *
210  * This macro may be used before pmap_bootstrap() is called.
211  */
212 #define	vtophys(va)	pmap_kextract((vm_offset_t)(va))
213 
214 #if defined(XEN)
215 #include <sys/param.h>
216 #include <machine/xen/xen-os.h>
217 #include <machine/xen/xenvar.h>
218 #include <machine/xen/xenpmap.h>
219 
220 extern pt_entry_t pg_nx;
221 
222 #define PG_KERNEL  (PG_V | PG_A | PG_RW | PG_M)
223 
224 #define MACH_TO_VM_PAGE(ma) PHYS_TO_VM_PAGE(xpmap_mtop((ma)))
225 #define VM_PAGE_TO_MACH(m) xpmap_ptom(VM_PAGE_TO_PHYS((m)))
226 
227 #define VTOM(va) xpmap_ptom(VTOP(va))
228 
229 static __inline vm_paddr_t
230 pmap_kextract_ma(vm_offset_t va)
231 {
232         vm_paddr_t ma;
233         if ((ma = PTD[va >> PDRSHIFT]) & PG_PS) {
234                 ma = (ma & ~(NBPDR - 1)) | (va & (NBPDR - 1));
235         } else {
236                 ma = (*vtopte(va) & PG_FRAME) | (va & PAGE_MASK);
237         }
238         return ma;
239 }
240 
241 static __inline vm_paddr_t
242 pmap_kextract(vm_offset_t va)
243 {
244         return xpmap_mtop(pmap_kextract_ma(va));
245 }
246 #define vtomach(va)     pmap_kextract_ma(((vm_offset_t) (va)))
247 
248 vm_paddr_t pmap_extract_ma(struct pmap *pmap, vm_offset_t va);
249 
250 void    pmap_kenter_ma(vm_offset_t va, vm_paddr_t pa);
251 void    pmap_map_readonly(struct pmap *pmap, vm_offset_t va, int len);
252 void    pmap_map_readwrite(struct pmap *pmap, vm_offset_t va, int len);
253 
254 static __inline pt_entry_t
255 pte_load_store(pt_entry_t *ptep, pt_entry_t v)
256 {
257 	pt_entry_t r;
258 
259 	r = *ptep;
260 	PT_SET_VA(ptep, v, TRUE);
261 	return (r);
262 }
263 
264 static __inline pt_entry_t
265 pte_load_store_ma(pt_entry_t *ptep, pt_entry_t v)
266 {
267 	pt_entry_t r;
268 
269 	r = *ptep;
270 	PT_SET_VA_MA(ptep, v, TRUE);
271 	return (r);
272 }
273 
274 #define	pte_load_clear(ptep)	pte_load_store((ptep), (pt_entry_t)0ULL)
275 
276 #define	pte_store(ptep, pte)	pte_load_store((ptep), (pt_entry_t)pte)
277 #define	pte_store_ma(ptep, pte)	pte_load_store_ma((ptep), (pt_entry_t)pte)
278 #define	pde_store_ma(ptep, pte)	pte_load_store_ma((ptep), (pt_entry_t)pte)
279 
280 #elif !defined(XEN)
281 
282 /*
283  * KPTmap is a linear mapping of the kernel page table.  It differs from the
284  * recursive mapping in two ways: (1) it only provides access to kernel page
285  * table pages, and not user page table pages, and (2) it provides access to
286  * a kernel page table page after the corresponding virtual addresses have
287  * been promoted to a 2/4MB page mapping.
288  *
289  * KPTmap is first initialized by locore to support just NPKT page table
290  * pages.  Later, it is reinitialized by pmap_bootstrap() to allow for
291  * expansion of the kernel page table.
292  */
293 extern pt_entry_t *KPTmap;
294 
295 /*
296  * Extract from the kernel page table the physical address that is mapped by
297  * the given virtual address "va".
298  *
299  * This function may be used before pmap_bootstrap() is called.
300  */
301 static __inline vm_paddr_t
302 pmap_kextract(vm_offset_t va)
303 {
304 	vm_paddr_t pa;
305 
306 	if ((pa = PTD[va >> PDRSHIFT]) & PG_PS) {
307 		pa = (pa & PG_PS_FRAME) | (va & PDRMASK);
308 	} else {
309 		/*
310 		 * Beware of a concurrent promotion that changes the PDE at
311 		 * this point!  For example, vtopte() must not be used to
312 		 * access the PTE because it would use the new PDE.  It is,
313 		 * however, safe to use the old PDE because the page table
314 		 * page is preserved by the promotion.
315 		 */
316 		pa = KPTmap[i386_btop(va)];
317 		pa = (pa & PG_FRAME) | (va & PAGE_MASK);
318 	}
319 	return (pa);
320 }
321 #endif
322 
323 #if !defined(XEN)
324 #define PT_UPDATES_FLUSH()
325 #endif
326 
327 #if defined(PAE) && !defined(XEN)
328 
329 #define	pde_cmpset(pdep, old, new) \
330 				atomic_cmpset_64((pdep), (old), (new))
331 
332 static __inline pt_entry_t
333 pte_load(pt_entry_t *ptep)
334 {
335 	pt_entry_t r;
336 
337 	__asm __volatile(
338 	    "lock; cmpxchg8b %1"
339 	    : "=A" (r)
340 	    : "m" (*ptep), "a" (0), "d" (0), "b" (0), "c" (0));
341 	return (r);
342 }
343 
344 static __inline pt_entry_t
345 pte_load_store(pt_entry_t *ptep, pt_entry_t v)
346 {
347 	pt_entry_t r;
348 
349 	r = *ptep;
350 	__asm __volatile(
351 	    "1:\n"
352 	    "\tlock; cmpxchg8b %1\n"
353 	    "\tjnz 1b"
354 	    : "+A" (r)
355 	    : "m" (*ptep), "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)));
356 	return (r);
357 }
358 
359 /* XXXRU move to atomic.h? */
360 static __inline int
361 atomic_cmpset_64(volatile uint64_t *dst, uint64_t exp, uint64_t src)
362 {
363 	int64_t res = exp;
364 
365 	__asm __volatile (
366 	"	lock ;			"
367 	"	cmpxchg8b %2 ;		"
368 	"	setz	%%al ;		"
369 	"	movzbl	%%al,%0 ;	"
370 	"# atomic_cmpset_64"
371 	: "+A" (res),			/* 0 (result) */
372 	  "=m" (*dst)			/* 1 */
373 	: "m" (*dst),			/* 2 */
374 	  "b" ((uint32_t)src),
375 	  "c" ((uint32_t)(src >> 32)));
376 
377 	return (res);
378 }
379 
380 #define	pte_load_clear(ptep)	pte_load_store((ptep), (pt_entry_t)0ULL)
381 
382 #define	pte_store(ptep, pte)	pte_load_store((ptep), (pt_entry_t)pte)
383 
384 extern pt_entry_t pg_nx;
385 
386 #elif !defined(PAE) && !defined (XEN)
387 
388 #define	pde_cmpset(pdep, old, new) \
389 				atomic_cmpset_int((pdep), (old), (new))
390 
391 static __inline pt_entry_t
392 pte_load(pt_entry_t *ptep)
393 {
394 	pt_entry_t r;
395 
396 	r = *ptep;
397 	return (r);
398 }
399 
400 static __inline pt_entry_t
401 pte_load_store(pt_entry_t *ptep, pt_entry_t pte)
402 {
403 	__asm volatile("xchgl %0, %1" : "+m" (*ptep), "+r" (pte));
404 	return (pte);
405 }
406 
407 #define	pte_load_clear(pte)	atomic_readandclear_int(pte)
408 
409 static __inline void
410 pte_store(pt_entry_t *ptep, pt_entry_t pte)
411 {
412 
413 	*ptep = pte;
414 }
415 
416 #endif /* PAE */
417 
418 #define	pte_clear(ptep)		pte_store((ptep), (pt_entry_t)0ULL)
419 
420 #define	pde_store(pdep, pde)	pte_store((pdep), (pde))
421 
422 #endif /* _KERNEL */
423 
424 /*
425  * Pmap stuff
426  */
427 struct	pv_entry;
428 struct	pv_chunk;
429 
430 struct md_page {
431 	TAILQ_HEAD(,pv_entry)	pv_list;
432 	int			pat_mode;
433 };
434 
435 struct pmap {
436 	struct mtx		pm_mtx;
437 	pd_entry_t		*pm_pdir;	/* KVA of page directory */
438 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
439 	cpuset_t		pm_active;	/* active on cpus */
440 	struct pmap_statistics	pm_stats;	/* pmap statistics */
441 	LIST_ENTRY(pmap) 	pm_list;	/* List of all pmaps */
442 #ifdef PAE
443 	pdpt_entry_t		*pm_pdpt;	/* KVA of page director pointer
444 						   table */
445 #endif
446 	struct vm_radix		pm_root;	/* spare page table pages */
447 };
448 
449 typedef struct pmap	*pmap_t;
450 
451 #ifdef _KERNEL
452 extern struct pmap	kernel_pmap_store;
453 #define kernel_pmap	(&kernel_pmap_store)
454 
455 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
456 #define	PMAP_LOCK_ASSERT(pmap, type) \
457 				mtx_assert(&(pmap)->pm_mtx, (type))
458 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
459 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
460 				    NULL, MTX_DEF | MTX_DUPOK)
461 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
462 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
463 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
464 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
465 #endif
466 
467 /*
468  * For each vm_page_t, there is a list of all currently valid virtual
469  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
470  */
471 typedef struct pv_entry {
472 	vm_offset_t	pv_va;		/* virtual address for mapping */
473 	TAILQ_ENTRY(pv_entry)	pv_next;
474 } *pv_entry_t;
475 
476 /*
477  * pv_entries are allocated in chunks per-process.  This avoids the
478  * need to track per-pmap assignments.
479  */
480 #define	_NPCM	11
481 #define	_NPCPV	336
482 struct pv_chunk {
483 	pmap_t			pc_pmap;
484 	TAILQ_ENTRY(pv_chunk)	pc_list;
485 	uint32_t		pc_map[_NPCM];	/* bitmap; 1 = free */
486 	TAILQ_ENTRY(pv_chunk)	pc_lru;
487 	struct pv_entry		pc_pventry[_NPCPV];
488 };
489 
490 #ifdef	_KERNEL
491 
492 extern caddr_t	CADDR1;
493 extern pt_entry_t *CMAP1;
494 extern vm_paddr_t phys_avail[];
495 extern vm_paddr_t dump_avail[];
496 extern int pseflag;
497 extern int pgeflag;
498 extern char *ptvmmap;		/* poor name! */
499 extern vm_offset_t virtual_avail;
500 extern vm_offset_t virtual_end;
501 
502 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
503 #define	pmap_page_is_write_mapped(m)	(((m)->aflags & PGA_WRITEABLE) != 0)
504 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
505 
506 /*
507  * Only the following functions or macros may be used before pmap_bootstrap()
508  * is called: pmap_kenter(), pmap_kextract(), pmap_kremove(), vtophys(), and
509  * vtopte().
510  */
511 void	pmap_bootstrap(vm_paddr_t);
512 int	pmap_cache_bits(int mode, boolean_t is_pde);
513 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
514 void	pmap_init_pat(void);
515 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
516 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
517 void	pmap_kremove(vm_offset_t);
518 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
519 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
520 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
521 boolean_t pmap_page_is_mapped(vm_page_t m);
522 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
523 void	pmap_unmapdev(vm_offset_t, vm_size_t);
524 pt_entry_t *pmap_pte(pmap_t, vm_offset_t) __pure2;
525 void	pmap_invalidate_page(pmap_t, vm_offset_t);
526 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
527 void	pmap_invalidate_all(pmap_t);
528 void	pmap_invalidate_cache(void);
529 void	pmap_invalidate_cache_pages(vm_page_t *pages, int count);
530 void	pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva);
531 
532 #endif /* _KERNEL */
533 
534 #endif /* !LOCORE */
535 
536 #endif /* !_MACHINE_PMAP_H_ */
537