xref: /netbsd-src/sys/arch/x86/include/pmap.h (revision a5847cc334d9a7029f6352b847e9e8d71a0f9e0c)
1 /*	$NetBSD: pmap.h,v 1.45 2011/11/08 17:16:52 cherry Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 /*
29  * Copyright (c) 2001 Wasabi Systems, Inc.
30  * All rights reserved.
31  *
32  * Written by Frank van der Linden for Wasabi Systems, Inc.
33  *
34  * Redistribution and use in source and binary forms, with or without
35  * modification, are permitted provided that the following conditions
36  * are met:
37  * 1. Redistributions of source code must retain the above copyright
38  *    notice, this list of conditions and the following disclaimer.
39  * 2. Redistributions in binary form must reproduce the above copyright
40  *    notice, this list of conditions and the following disclaimer in the
41  *    documentation and/or other materials provided with the distribution.
42  * 3. All advertising materials mentioning features or use of this software
43  *    must display the following acknowledgement:
44  *      This product includes software developed for the NetBSD Project by
45  *      Wasabi Systems, Inc.
46  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
47  *    or promote products derived from this software without specific prior
48  *    written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
52  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
53  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
54  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
55  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
56  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
57  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
58  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
59  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
60  * POSSIBILITY OF SUCH DAMAGE.
61  */
62 
63 /*
64  * pmap.h: see pmap.c for the history of this pmap module.
65  */
66 
67 #ifndef _X86_PMAP_H_
68 #define	_X86_PMAP_H_
69 
70 /*
71  * pl*_pi: index in the ptp page for a pde mapping a VA.
72  * (pl*_i below is the index in the virtual array of all pdes per level)
73  */
74 #define pl1_pi(VA)	(((VA_SIGN_POS(VA)) & L1_MASK) >> L1_SHIFT)
75 #define pl2_pi(VA)	(((VA_SIGN_POS(VA)) & L2_MASK) >> L2_SHIFT)
76 #define pl3_pi(VA)	(((VA_SIGN_POS(VA)) & L3_MASK) >> L3_SHIFT)
77 #define pl4_pi(VA)	(((VA_SIGN_POS(VA)) & L4_MASK) >> L4_SHIFT)
78 
79 /*
80  * pl*_i: generate index into pde/pte arrays in virtual space
81  *
82  * pl_i(va, X) == plX_i(va) <= pl_i_roundup(va, X)
83  */
84 #define pl1_i(VA)	(((VA_SIGN_POS(VA)) & L1_FRAME) >> L1_SHIFT)
85 #define pl2_i(VA)	(((VA_SIGN_POS(VA)) & L2_FRAME) >> L2_SHIFT)
86 #define pl3_i(VA)	(((VA_SIGN_POS(VA)) & L3_FRAME) >> L3_SHIFT)
87 #define pl4_i(VA)	(((VA_SIGN_POS(VA)) & L4_FRAME) >> L4_SHIFT)
88 #define pl_i(va, lvl) \
89         (((VA_SIGN_POS(va)) & ptp_masks[(lvl)-1]) >> ptp_shifts[(lvl)-1])
90 
91 #define	pl_i_roundup(va, lvl)	pl_i((va)+ ~ptp_masks[(lvl)-1], (lvl))
92 
93 /*
94  * PTP macros:
95  *   a PTP's index is the PD index of the PDE that points to it
96  *   a PTP's offset is the byte-offset in the PTE space that this PTP is at
97  *   a PTP's VA is the first VA mapped by that PTP
98  */
99 
100 #define ptp_va2o(va, lvl)	(pl_i(va, (lvl)+1) * PAGE_SIZE)
101 
102 /* size of a PDP: usually one page, except for PAE */
103 #ifdef PAE
104 #define PDP_SIZE 4
105 #else
106 #define PDP_SIZE 1
107 #endif
108 
109 
110 #if defined(_KERNEL)
111 /*
112  * pmap data structures: see pmap.c for details of locking.
113  */
114 
115 /*
116  * we maintain a list of all non-kernel pmaps
117  */
118 
119 LIST_HEAD(pmap_head, pmap); /* struct pmap_head: head of a pmap list */
120 
121 /*
122  * linked list of all non-kernel pmaps
123  */
124 extern struct pmap_head pmaps;
125 extern kmutex_t pmaps_lock;    /* protects pmaps */
126 
127 /*
128  * the pmap structure
129  *
130  * note that the pm_obj contains the lock pointer, the reference count,
131  * page list, and number of PTPs within the pmap.
132  *
133  * pm_lock is the same as the lock for vm object 0.  Changes to
134  * the other objects may only be made if that lock has been taken
135  * (the other object locks are only used when uvm_pagealloc is called)
136  *
137  * XXX If we ever support processor numbers higher than 31, we'll have
138  * XXX to rethink the CPU mask.
139  */
140 
141 struct pmap {
142 	struct uvm_object pm_obj[PTP_LEVELS-1]; /* objects for lvl >= 1) */
143 #define	pm_lock	pm_obj[0].vmobjlock
144 	kmutex_t pm_obj_lock[PTP_LEVELS-1];	/* locks for pm_objs */
145 	LIST_ENTRY(pmap) pm_list;	/* list (lck by pm_list lock) */
146 	pd_entry_t *pm_pdir;		/* VA of PD (lck by object lock) */
147 	paddr_t pm_pdirpa[PDP_SIZE];	/* PA of PDs (read-only after create) */
148 	struct vm_page *pm_ptphint[PTP_LEVELS-1];
149 					/* pointer to a PTP in our pmap */
150 	struct pmap_statistics pm_stats;  /* pmap stats (lck by object lock) */
151 
152 #if !defined(__x86_64__)
153 	vaddr_t pm_hiexec;		/* highest executable mapping */
154 #endif /* !defined(__x86_64__) */
155 	int pm_flags;			/* see below */
156 
157 	union descriptor *pm_ldt;	/* user-set LDT */
158 	size_t pm_ldt_len;		/* size of LDT in bytes */
159 	int pm_ldt_sel;			/* LDT selector */
160 	uint32_t pm_cpus;		/* mask of CPUs using pmap */
161 	uint32_t pm_kernel_cpus;	/* mask of CPUs using kernel part
162 					 of pmap */
163 	uint64_t pm_ncsw;		/* for assertions */
164 	struct vm_page *pm_gc_ptp;	/* pages from pmap g/c */
165 };
166 
167 /* macro to access pm_pdirpa slots */
168 #ifdef PAE
169 #define pmap_pdirpa(pmap, index) \
170 	((pmap)->pm_pdirpa[l2tol3(index)] + l2tol2(index) * sizeof(pd_entry_t))
171 #else
172 #define pmap_pdirpa(pmap, index) \
173 	((pmap)->pm_pdirpa[0] + (index) * sizeof(pd_entry_t))
174 #endif
175 
176 /*
177  * flag to be used for kernel mappings: PG_u on Xen/amd64,
178  * 0 otherwise.
179  */
180 #if defined(XEN) && defined(__x86_64__)
181 #define PG_k PG_u
182 #else
183 #define PG_k 0
184 #endif
185 
186 /*
187  * MD flags that we use for pmap_enter and pmap_kenter_pa:
188  */
189 
190 /*
191  * global kernel variables
192  */
193 
194 /*
195  * PDPpaddr is the physical address of the kernel's PDP.
196  * - i386 non-PAE and amd64: PDPpaddr corresponds directly to the %cr3
197  * value associated to the kernel process, proc0.
198  * - i386 PAE: it still represents the PA of the kernel's PDP (L2). Due to
199  * the L3 PD, it cannot be considered as the equivalent of a %cr3 any more.
200  * - Xen: it corresponds to the PFN of the kernel's PDP.
201  */
202 extern u_long PDPpaddr;
203 
204 extern int pmap_pg_g;			/* do we support PG_G? */
205 extern long nkptp[PTP_LEVELS];
206 
207 /*
208  * macros
209  */
210 
211 #define	pmap_resident_count(pmap)	((pmap)->pm_stats.resident_count)
212 #define	pmap_wired_count(pmap)		((pmap)->pm_stats.wired_count)
213 
214 #define pmap_clear_modify(pg)		pmap_clear_attrs(pg, PG_M)
215 #define pmap_clear_reference(pg)	pmap_clear_attrs(pg, PG_U)
216 #define pmap_copy(DP,SP,D,L,S)
217 #define pmap_is_modified(pg)		pmap_test_attrs(pg, PG_M)
218 #define pmap_is_referenced(pg)		pmap_test_attrs(pg, PG_U)
219 #define pmap_move(DP,SP,D,L,S)
220 #define pmap_phys_address(ppn)		(x86_ptob(ppn) & ~X86_MMAP_FLAG_MASK)
221 #define pmap_mmap_flags(ppn)		x86_mmap_flags(ppn)
222 #define pmap_valid_entry(E) 		((E) & PG_V) /* is PDE or PTE valid? */
223 
224 #if defined(__x86_64__) || defined(PAE)
225 #define X86_MMAP_FLAG_SHIFT	(64 - PGSHIFT)
226 #else
227 #define X86_MMAP_FLAG_SHIFT	(32 - PGSHIFT)
228 #endif
229 
230 #define X86_MMAP_FLAG_MASK	0xf
231 #define X86_MMAP_FLAG_PREFETCH	0x1
232 
233 /*
234  * prototypes
235  */
236 
237 void		pmap_activate(struct lwp *);
238 void		pmap_bootstrap(vaddr_t);
239 bool		pmap_clear_attrs(struct vm_page *, unsigned);
240 void		pmap_deactivate(struct lwp *);
241 void		pmap_page_remove (struct vm_page *);
242 void		pmap_remove(struct pmap *, vaddr_t, vaddr_t);
243 bool		pmap_test_attrs(struct vm_page *, unsigned);
244 void		pmap_write_protect(struct pmap *, vaddr_t, vaddr_t, vm_prot_t);
245 void		pmap_load(void);
246 paddr_t		pmap_init_tmp_pgtbl(paddr_t);
247 void		pmap_remove_all(struct pmap *);
248 void		pmap_ldt_sync(struct pmap *);
249 
250 void		pmap_emap_enter(vaddr_t, paddr_t, vm_prot_t);
251 void		pmap_emap_remove(vaddr_t, vsize_t);
252 void		pmap_emap_sync(bool);
253 
254 void		pmap_map_ptes(struct pmap *, struct pmap **, pd_entry_t **,
255 		    pd_entry_t * const **);
256 void		pmap_unmap_ptes(struct pmap *, struct pmap *);
257 
258 int		pmap_pdes_invalid(vaddr_t, pd_entry_t * const *, pd_entry_t *);
259 
260 u_int		x86_mmap_flags(paddr_t);
261 
262 bool		pmap_is_curpmap(struct pmap *);
263 
264 void		pmap_invalidate_pool_caches(void);
265 
266 vaddr_t reserve_dumppages(vaddr_t); /* XXX: not a pmap fn */
267 
268 typedef enum tlbwhy {
269 	TLBSHOOT_APTE,
270 	TLBSHOOT_KENTER,
271 	TLBSHOOT_KREMOVE,
272 	TLBSHOOT_FREE_PTP1,
273 	TLBSHOOT_FREE_PTP2,
274 	TLBSHOOT_REMOVE_PTE,
275 	TLBSHOOT_REMOVE_PTES,
276 	TLBSHOOT_SYNC_PV1,
277 	TLBSHOOT_SYNC_PV2,
278 	TLBSHOOT_WRITE_PROTECT,
279 	TLBSHOOT_ENTER,
280 	TLBSHOOT_UPDATE,
281 	TLBSHOOT_BUS_DMA,
282 	TLBSHOOT_BUS_SPACE,
283 	TLBSHOOT__MAX,
284 } tlbwhy_t;
285 
286 void		pmap_tlb_init(void);
287 void		pmap_tlb_shootdown(pmap_t, vaddr_t, pt_entry_t, tlbwhy_t);
288 void		pmap_tlb_shootnow(void);
289 void		pmap_tlb_intr(void);
290 
291 #define	__HAVE_PMAP_EMAP
292 
293 #define PMAP_GROWKERNEL		/* turn on pmap_growkernel interface */
294 #define PMAP_FORK		/* turn on pmap_fork interface */
295 
296 /*
297  * Do idle page zero'ing uncached to avoid polluting the cache.
298  */
299 bool	pmap_pageidlezero(paddr_t);
300 #define	PMAP_PAGEIDLEZERO(pa)	pmap_pageidlezero((pa))
301 
302 /*
303  * inline functions
304  */
305 
306 __inline static bool __unused
307 pmap_pdes_valid(vaddr_t va, pd_entry_t * const *pdes, pd_entry_t *lastpde)
308 {
309 	return pmap_pdes_invalid(va, pdes, lastpde) == 0;
310 }
311 
312 /*
313  * pmap_update_pg: flush one page from the TLB (or flush the whole thing
314  *	if hardware doesn't support one-page flushing)
315  */
316 
317 __inline static void __unused
318 pmap_update_pg(vaddr_t va)
319 {
320 	invlpg(va);
321 }
322 
323 /*
324  * pmap_update_2pg: flush two pages from the TLB
325  */
326 
327 __inline static void __unused
328 pmap_update_2pg(vaddr_t va, vaddr_t vb)
329 {
330 	invlpg(va);
331 	invlpg(vb);
332 }
333 
334 /*
335  * pmap_page_protect: change the protection of all recorded mappings
336  *	of a managed page
337  *
338  * => this function is a frontend for pmap_page_remove/pmap_clear_attrs
339  * => we only have to worry about making the page more protected.
340  *	unprotecting a page is done on-demand at fault time.
341  */
342 
343 __inline static void __unused
344 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
345 {
346 	if ((prot & VM_PROT_WRITE) == 0) {
347 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
348 			(void) pmap_clear_attrs(pg, PG_RW);
349 		} else {
350 			pmap_page_remove(pg);
351 		}
352 	}
353 }
354 
355 /*
356  * pmap_protect: change the protection of pages in a pmap
357  *
358  * => this function is a frontend for pmap_remove/pmap_write_protect
359  * => we only have to worry about making the page more protected.
360  *	unprotecting a page is done on-demand at fault time.
361  */
362 
363 __inline static void __unused
364 pmap_protect(struct pmap *pmap, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
365 {
366 	if ((prot & VM_PROT_WRITE) == 0) {
367 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
368 			pmap_write_protect(pmap, sva, eva, prot);
369 		} else {
370 			pmap_remove(pmap, sva, eva);
371 		}
372 	}
373 }
374 
375 /*
376  * various address inlines
377  *
378  *  vtopte: return a pointer to the PTE mapping a VA, works only for
379  *  user and PT addresses
380  *
381  *  kvtopte: return a pointer to the PTE mapping a kernel VA
382  */
383 
384 #include <lib/libkern/libkern.h>
385 
386 static __inline pt_entry_t * __unused
387 vtopte(vaddr_t va)
388 {
389 
390 	KASSERT(va < VM_MIN_KERNEL_ADDRESS);
391 
392 	return (PTE_BASE + pl1_i(va));
393 }
394 
395 static __inline pt_entry_t * __unused
396 kvtopte(vaddr_t va)
397 {
398 	pd_entry_t *pde;
399 
400 	KASSERT(va >= VM_MIN_KERNEL_ADDRESS);
401 
402 	pde = L2_BASE + pl2_i(va);
403 	if (*pde & PG_PS)
404 		return ((pt_entry_t *)pde);
405 
406 	return (PTE_BASE + pl1_i(va));
407 }
408 
409 paddr_t vtophys(vaddr_t);
410 vaddr_t	pmap_map(vaddr_t, paddr_t, paddr_t, vm_prot_t);
411 void	pmap_cpu_init_late(struct cpu_info *);
412 bool	sse2_idlezero_page(void *);
413 
414 
415 #ifdef XEN
416 
417 void	pmap_unmap_all_apdp_pdes(void);
418 #ifdef PAE
419 void	pmap_map_recursive_entries(void);
420 void	pmap_unmap_recursive_entries(void);
421 #endif /* PAE */
422 
423 #include <sys/bitops.h>
424 
425 #define XPTE_MASK	L1_FRAME
426 /* Selects the index of a PTE in (A)PTE_BASE */
427 #define XPTE_SHIFT	(L1_SHIFT - ilog2(sizeof(pt_entry_t)))
428 
429 /* PTE access inline fuctions */
430 
431 /*
432  * Get the machine address of the pointed pte
433  * We use hardware MMU to get value so works only for levels 1-3
434  */
435 
436 static __inline paddr_t
437 xpmap_ptetomach(pt_entry_t *pte)
438 {
439 	pt_entry_t *up_pte;
440 	vaddr_t va = (vaddr_t) pte;
441 
442 	va = ((va & XPTE_MASK) >> XPTE_SHIFT) | (vaddr_t) PTE_BASE;
443 	up_pte = (pt_entry_t *) va;
444 
445 	return (paddr_t) (((*up_pte) & PG_FRAME) + (((vaddr_t) pte) & (~PG_FRAME & ~VA_SIGN_MASK)));
446 }
447 
448 /*
449  * xpmap_update()
450  * Update an active pt entry with Xen
451  * Equivalent to *pte = npte
452  */
453 
454 static __inline void
455 xpmap_update (pt_entry_t *pte, pt_entry_t npte)
456 {
457         int s = splvm();
458 
459         xpq_queue_pte_update(xpmap_ptetomach(pte), npte);
460         xpq_flush_queue();
461         splx(s);
462 }
463 
464 
465 /* Xen helpers to change bits of a pte */
466 #define XPMAP_UPDATE_DIRECT	1	/* Update direct map entry flags too */
467 
468 paddr_t	vtomach(vaddr_t);
469 #define vtomfn(va) (vtomach(va) >> PAGE_SHIFT)
470 
471 void	pmap_apte_flush(struct pmap *);
472 void	pmap_unmap_apdp(void);
473 
474 #endif	/* XEN */
475 
476 /* pmap functions with machine addresses */
477 void	pmap_kenter_ma(vaddr_t, paddr_t, vm_prot_t, u_int);
478 int	pmap_enter_ma(struct pmap *, vaddr_t, paddr_t, paddr_t,
479 	    vm_prot_t, u_int, int);
480 bool	pmap_extract_ma(pmap_t, vaddr_t, paddr_t *);
481 
482 /*
483  * Hooks for the pool allocator.
484  */
485 #define	POOL_VTOPHYS(va)	vtophys((vaddr_t) (va))
486 
487 #endif /* _KERNEL */
488 
489 #endif /* _X86_PMAP_H_ */
490