xref: /freebsd/sys/i386/include/pmap.h (revision 830940567b49bb0c08dfaed40418999e76616909)
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 /*
85  * Promotion to a 2 or 4MB (PDE) page mapping requires that the corresponding
86  * 4KB (PTE) page mappings have identical settings for the following fields:
87  */
88 #define PG_PTE_PROMOTE	(PG_MANAGED | PG_W | PG_G | PG_PTE_PAT | \
89 	    PG_M | PG_A | PG_NC_PCD | PG_NC_PWT | PG_U | PG_RW | PG_V)
90 
91 /*
92  * Page Protection Exception bits
93  */
94 
95 #define PGEX_P		0x01	/* Protection violation vs. not present */
96 #define PGEX_W		0x02	/* during a Write cycle */
97 #define PGEX_U		0x04	/* access from User mode (UPL) */
98 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
99 #define PGEX_I		0x10	/* during an instruction fetch */
100 
101 /*
102  * Size of Kernel address space.  This is the number of page table pages
103  * (4MB each) to use for the kernel.  256 pages == 1 Gigabyte.
104  * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
105  * For PAE, the page table page unit size is 2MB.  This means that 512 pages
106  * is 1 Gigabyte.  Double everything.  It must be a multiple of 8 for PAE.
107  */
108 #ifndef KVA_PAGES
109 #ifdef PAE
110 #define KVA_PAGES	512
111 #else
112 #define KVA_PAGES	256
113 #endif
114 #endif
115 
116 /*
117  * Pte related macros
118  */
119 #define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDRSHIFT)|((pti)<<PAGE_SHIFT)))
120 
121 /* Initial number of kernel page tables. */
122 #ifndef NKPT
123 #ifdef PAE
124 /* 152 page tables needed to map 16G (76B "struct vm_page", 2M page tables). */
125 #define	NKPT		240
126 #else
127 /* 18 page tables needed to map 4G (72B "struct vm_page", 4M page tables). */
128 #define	NKPT		30
129 #endif
130 #endif
131 
132 #ifndef NKPDE
133 #define NKPDE	(KVA_PAGES)	/* number of page tables/pde's */
134 #endif
135 
136 /*
137  * The *PTDI values control the layout of virtual memory
138  *
139  * XXX This works for now, but I am not real happy with it, I'll fix it
140  * right after I fix locore.s and the magic 28K hole
141  */
142 #define	KPTDI		(NPDEPTD-NKPDE)	/* start of kernel virtual pde's */
143 #define	PTDPTDI		(KPTDI-NPGPTD)	/* ptd entry that points to ptd! */
144 
145 /*
146  * XXX doesn't really belong here I guess...
147  */
148 #define ISA_HOLE_START    0xa0000
149 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
150 
151 #ifndef LOCORE
152 
153 #include <sys/queue.h>
154 #include <sys/_lock.h>
155 #include <sys/_mutex.h>
156 
157 #ifdef PAE
158 
159 typedef uint64_t pdpt_entry_t;
160 typedef uint64_t pd_entry_t;
161 typedef uint64_t pt_entry_t;
162 
163 #define	PTESHIFT	(3)
164 #define	PDESHIFT	(3)
165 
166 #else
167 
168 typedef uint32_t pd_entry_t;
169 typedef uint32_t pt_entry_t;
170 
171 #define	PTESHIFT	(2)
172 #define	PDESHIFT	(2)
173 
174 #endif
175 
176 /*
177  * Address of current address space page table maps and directories.
178  */
179 #ifdef _KERNEL
180 extern pt_entry_t PTmap[];
181 extern pd_entry_t PTD[];
182 extern pd_entry_t PTDpde[];
183 
184 #ifdef PAE
185 extern pdpt_entry_t *IdlePDPT;
186 #endif
187 extern pd_entry_t *IdlePTD;	/* physical address of "Idle" state directory */
188 
189 /*
190  * virtual address to page table entry and
191  * to physical address.
192  * Note: these work recursively, thus vtopte of a pte will give
193  * the corresponding pde that in turn maps it.
194  */
195 #define	vtopte(va)	(PTmap + i386_btop(va))
196 #define	vtophys(va)	pmap_kextract((vm_offset_t)(va))
197 
198 #ifdef XEN
199 #include <sys/param.h>
200 #include <machine/xen/xen-os.h>
201 #include <machine/xen/xenvar.h>
202 #include <machine/xen/xenpmap.h>
203 
204 extern pt_entry_t pg_nx;
205 
206 #define PG_KERNEL  (PG_V | PG_A | PG_RW | PG_M)
207 
208 #define MACH_TO_VM_PAGE(ma) PHYS_TO_VM_PAGE(xpmap_mtop((ma)))
209 #define VM_PAGE_TO_MACH(m) xpmap_ptom(VM_PAGE_TO_PHYS((m)))
210 
211 static __inline vm_paddr_t
212 pmap_kextract_ma(vm_offset_t va)
213 {
214         vm_paddr_t ma;
215         if ((ma = PTD[va >> PDRSHIFT]) & PG_PS) {
216                 ma = (ma & ~(NBPDR - 1)) | (va & (NBPDR - 1));
217         } else {
218                 ma = (*vtopte(va) & PG_FRAME) | (va & PAGE_MASK);
219         }
220         return ma;
221 }
222 
223 static __inline vm_paddr_t
224 pmap_kextract(vm_offset_t va)
225 {
226         return xpmap_mtop(pmap_kextract_ma(va));
227 }
228 #define vtomach(va)     pmap_kextract_ma(((vm_offset_t) (va)))
229 
230 vm_paddr_t pmap_extract_ma(struct pmap *pmap, vm_offset_t va);
231 
232 void    pmap_kenter_ma(vm_offset_t va, vm_paddr_t pa);
233 void    pmap_map_readonly(struct pmap *pmap, vm_offset_t va, int len);
234 void    pmap_map_readwrite(struct pmap *pmap, vm_offset_t va, int len);
235 
236 static __inline pt_entry_t
237 pte_load_store(pt_entry_t *ptep, pt_entry_t v)
238 {
239 	pt_entry_t r;
240 
241 	v = xpmap_ptom(v);
242 	r = *ptep;
243 	PT_SET_VA(ptep, v, TRUE);
244 	return (r);
245 }
246 
247 static __inline pt_entry_t
248 pte_load_store_ma(pt_entry_t *ptep, pt_entry_t v)
249 {
250 	pt_entry_t r;
251 
252 	r = *ptep;
253 	PT_SET_VA_MA(ptep, v, TRUE);
254 	return (r);
255 }
256 
257 #define	pte_load_clear(ptep)	pte_load_store((ptep), (pt_entry_t)0ULL)
258 
259 #define	pte_store(ptep, pte)	pte_load_store((ptep), (pt_entry_t)pte)
260 #define	pte_store_ma(ptep, pte)	pte_load_store_ma((ptep), (pt_entry_t)pte)
261 #define	pde_store_ma(ptep, pte)	pte_load_store_ma((ptep), (pt_entry_t)pte)
262 
263 #elif !defined(XEN)
264 /*
265  *	Routine:	pmap_kextract
266  *	Function:
267  *		Extract the physical page address associated
268  *		kernel virtual address.
269  */
270 static __inline vm_paddr_t
271 pmap_kextract(vm_offset_t va)
272 {
273 	vm_paddr_t pa;
274 
275 	if ((pa = PTD[va >> PDRSHIFT]) & PG_PS) {
276 		pa = (pa & PG_PS_FRAME) | (va & PDRMASK);
277 	} else {
278 		pa = *vtopte(va);
279 		pa = (pa & PG_FRAME) | (va & PAGE_MASK);
280 	}
281 	return pa;
282 }
283 
284 #define PT_UPDATES_FLUSH()
285 #endif
286 
287 #if defined(PAE) && !defined(XEN)
288 
289 #define	pde_cmpset(pdep, old, new) \
290 				atomic_cmpset_64((pdep), (old), (new))
291 
292 static __inline pt_entry_t
293 pte_load(pt_entry_t *ptep)
294 {
295 	pt_entry_t r;
296 
297 	__asm __volatile(
298 	    "lock; cmpxchg8b %1"
299 	    : "=A" (r)
300 	    : "m" (*ptep), "a" (0), "d" (0), "b" (0), "c" (0));
301 	return (r);
302 }
303 
304 static __inline pt_entry_t
305 pte_load_store(pt_entry_t *ptep, pt_entry_t v)
306 {
307 	pt_entry_t r;
308 
309 	r = *ptep;
310 	__asm __volatile(
311 	    "1:\n"
312 	    "\tlock; cmpxchg8b %1\n"
313 	    "\tjnz 1b"
314 	    : "+A" (r)
315 	    : "m" (*ptep), "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)));
316 	return (r);
317 }
318 
319 /* XXXRU move to atomic.h? */
320 static __inline int
321 atomic_cmpset_64(volatile uint64_t *dst, uint64_t exp, uint64_t src)
322 {
323 	int64_t res = exp;
324 
325 	__asm __volatile (
326 	"	lock ;			"
327 	"	cmpxchg8b %2 ;		"
328 	"	setz	%%al ;		"
329 	"	movzbl	%%al,%0 ;	"
330 	"# atomic_cmpset_64"
331 	: "+A" (res),			/* 0 (result) */
332 	  "=m" (*dst)			/* 1 */
333 	: "m" (*dst),			/* 2 */
334 	  "b" ((uint32_t)src),
335 	  "c" ((uint32_t)(src >> 32)));
336 
337 	return (res);
338 }
339 
340 #define	pte_load_clear(ptep)	pte_load_store((ptep), (pt_entry_t)0ULL)
341 
342 #define	pte_store(ptep, pte)	pte_load_store((ptep), (pt_entry_t)pte)
343 
344 extern pt_entry_t pg_nx;
345 
346 #elif !defined(PAE) && !defined (XEN)
347 
348 #define	pde_cmpset(pdep, old, new) \
349 				atomic_cmpset_int((pdep), (old), (new))
350 
351 static __inline pt_entry_t
352 pte_load(pt_entry_t *ptep)
353 {
354 	pt_entry_t r;
355 
356 	r = *ptep;
357 	return (r);
358 }
359 
360 static __inline pt_entry_t
361 pte_load_store(pt_entry_t *ptep, pt_entry_t pte)
362 {
363 	__asm volatile("xchgl %0, %1" : "+m" (*ptep), "+r" (pte));
364 	return (pte);
365 }
366 
367 #define	pte_load_clear(pte)	atomic_readandclear_int(pte)
368 
369 static __inline void
370 pte_store(pt_entry_t *ptep, pt_entry_t pte)
371 {
372 
373 	*ptep = pte;
374 }
375 
376 #endif /* PAE */
377 
378 #define	pte_clear(ptep)		pte_store((ptep), (pt_entry_t)0ULL)
379 
380 #define	pde_store(pdep, pde)	pte_store((pdep), (pde))
381 
382 #endif /* _KERNEL */
383 
384 /*
385  * Pmap stuff
386  */
387 struct	pv_entry;
388 struct	pv_chunk;
389 
390 struct md_page {
391 	TAILQ_HEAD(,pv_entry)	pv_list;
392 	int			pat_mode;
393 };
394 
395 struct pmap {
396 	struct mtx		pm_mtx;
397 	pd_entry_t		*pm_pdir;	/* KVA of page directory */
398 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
399 	u_int			pm_active;	/* active on cpus */
400 	struct pmap_statistics	pm_stats;	/* pmap statistics */
401 	LIST_ENTRY(pmap) 	pm_list;	/* List of all pmaps */
402 #ifdef PAE
403 	pdpt_entry_t		*pm_pdpt;	/* KVA of page director pointer
404 						   table */
405 #endif
406 	vm_page_t		pm_root;	/* spare page table pages */
407 };
408 
409 typedef struct pmap	*pmap_t;
410 
411 #ifdef _KERNEL
412 extern struct pmap	kernel_pmap_store;
413 #define kernel_pmap	(&kernel_pmap_store)
414 
415 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
416 #define	PMAP_LOCK_ASSERT(pmap, type) \
417 				mtx_assert(&(pmap)->pm_mtx, (type))
418 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
419 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
420 				    NULL, MTX_DEF | MTX_DUPOK)
421 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
422 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
423 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
424 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
425 #endif
426 
427 /*
428  * For each vm_page_t, there is a list of all currently valid virtual
429  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
430  */
431 typedef struct pv_entry {
432 	vm_offset_t	pv_va;		/* virtual address for mapping */
433 	TAILQ_ENTRY(pv_entry)	pv_list;
434 } *pv_entry_t;
435 
436 /*
437  * pv_entries are allocated in chunks per-process.  This avoids the
438  * need to track per-pmap assignments.
439  */
440 #define	_NPCM	11
441 #define	_NPCPV	336
442 struct pv_chunk {
443 	pmap_t			pc_pmap;
444 	TAILQ_ENTRY(pv_chunk)	pc_list;
445 	uint32_t		pc_map[_NPCM];	/* bitmap; 1 = free */
446 	uint32_t		pc_spare[2];
447 	struct pv_entry		pc_pventry[_NPCPV];
448 };
449 
450 #ifdef	_KERNEL
451 
452 extern caddr_t	CADDR1;
453 extern pt_entry_t *CMAP1;
454 extern vm_paddr_t phys_avail[];
455 extern vm_paddr_t dump_avail[];
456 extern int pseflag;
457 extern int pgeflag;
458 extern char *ptvmmap;		/* poor name! */
459 extern vm_offset_t virtual_avail;
460 extern vm_offset_t virtual_end;
461 
462 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
463 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
464 
465 void	pmap_bootstrap(vm_paddr_t);
466 int	pmap_cache_bits(int mode, boolean_t is_pde);
467 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
468 void	pmap_init_pat(void);
469 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
470 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
471 void	pmap_kremove(vm_offset_t);
472 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
473 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
474 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
475 boolean_t pmap_page_is_mapped(vm_page_t m);
476 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
477 void	pmap_unmapdev(vm_offset_t, vm_size_t);
478 pt_entry_t *pmap_pte(pmap_t, vm_offset_t) __pure2;
479 void	pmap_set_pg(void);
480 void	pmap_invalidate_page(pmap_t, vm_offset_t);
481 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
482 void	pmap_invalidate_all(pmap_t);
483 void	pmap_invalidate_cache(void);
484 void	pmap_invalidate_cache_range(vm_offset_t, vm_offset_t);
485 
486 #endif /* _KERNEL */
487 
488 #endif /* !LOCORE */
489 
490 #endif /* !_MACHINE_PMAP_H_ */
491