xref: /netbsd-src/sys/arch/x86/include/pmap.h (revision e89934bbf778a6d6d6894877c4da59d0c7835b0f)
1 /*	$NetBSD: pmap.h,v 1.62 2017/02/11 14:11:24 maxv 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 #include <sys/kcpuset.h>
112 #include <uvm/pmap/pmap_pvt.h>
113 
114 /*
115  * pmap data structures: see pmap.c for details of locking.
116  */
117 
118 /*
119  * we maintain a list of all non-kernel pmaps
120  */
121 
122 LIST_HEAD(pmap_head, pmap); /* struct pmap_head: head of a pmap list */
123 
124 /*
125  * linked list of all non-kernel pmaps
126  */
127 extern struct pmap_head pmaps;
128 extern kmutex_t pmaps_lock;    /* protects pmaps */
129 
130 /*
131  * pool_cache(9) that PDPs are allocated from
132  */
133 extern struct pool_cache pmap_pdp_cache;
134 
135 /*
136  * the pmap structure
137  *
138  * note that the pm_obj contains the lock pointer, the reference count,
139  * page list, and number of PTPs within the pmap.
140  *
141  * pm_lock is the same as the lock for vm object 0.  Changes to
142  * the other objects may only be made if that lock has been taken
143  * (the other object locks are only used when uvm_pagealloc is called)
144  */
145 
146 struct pmap {
147 	struct uvm_object pm_obj[PTP_LEVELS-1]; /* objects for lvl >= 1) */
148 #define	pm_lock	pm_obj[0].vmobjlock
149 	kmutex_t pm_obj_lock[PTP_LEVELS-1];	/* locks for pm_objs */
150 	LIST_ENTRY(pmap) pm_list;	/* list (lck by pm_list lock) */
151 	pd_entry_t *pm_pdir;		/* VA of PD (lck by object lock) */
152 	paddr_t pm_pdirpa[PDP_SIZE];	/* PA of PDs (read-only after create) */
153 	struct vm_page *pm_ptphint[PTP_LEVELS-1];
154 					/* pointer to a PTP in our pmap */
155 	struct pmap_statistics pm_stats;  /* pmap stats (lck by object lock) */
156 
157 #if !defined(__x86_64__)
158 	vaddr_t pm_hiexec;		/* highest executable mapping */
159 #endif /* !defined(__x86_64__) */
160 	int pm_flags;			/* see below */
161 
162 	union descriptor *pm_ldt;	/* user-set LDT */
163 	size_t pm_ldt_len;		/* size of LDT in bytes */
164 	int pm_ldt_sel;			/* LDT selector */
165 	kcpuset_t *pm_cpus;		/* mask of CPUs using pmap */
166 	kcpuset_t *pm_kernel_cpus;	/* mask of CPUs using kernel part
167 					 of pmap */
168 	kcpuset_t *pm_xen_ptp_cpus;	/* mask of CPUs which have this pmap's
169 					 ptp mapped */
170 	uint64_t pm_ncsw;		/* for assertions */
171 	struct vm_page *pm_gc_ptp;	/* pages from pmap g/c */
172 };
173 
174 /* macro to access pm_pdirpa slots */
175 #ifdef PAE
176 #define pmap_pdirpa(pmap, index) \
177 	((pmap)->pm_pdirpa[l2tol3(index)] + l2tol2(index) * sizeof(pd_entry_t))
178 #else
179 #define pmap_pdirpa(pmap, index) \
180 	((pmap)->pm_pdirpa[0] + (index) * sizeof(pd_entry_t))
181 #endif
182 
183 /*
184  * flag to be used for kernel mappings: PG_u on Xen/amd64,
185  * 0 otherwise.
186  */
187 #if defined(XEN) && defined(__x86_64__)
188 #define PG_k PG_u
189 #else
190 #define PG_k 0
191 #endif
192 
193 /*
194  * MD flags that we use for pmap_enter and pmap_kenter_pa:
195  */
196 
197 /*
198  * global kernel variables
199  */
200 
201 /*
202  * PDPpaddr is the physical address of the kernel's PDP.
203  * - i386 non-PAE and amd64: PDPpaddr corresponds directly to the %cr3
204  * value associated to the kernel process, proc0.
205  * - i386 PAE: it still represents the PA of the kernel's PDP (L2). Due to
206  * the L3 PD, it cannot be considered as the equivalent of a %cr3 any more.
207  * - Xen: it corresponds to the PFN of the kernel's PDP.
208  */
209 extern u_long PDPpaddr;
210 
211 extern pd_entry_t pmap_pg_g;			/* do we support PG_G? */
212 extern pd_entry_t pmap_pg_nx;			/* do we support PG_NX? */
213 extern long nkptp[PTP_LEVELS];
214 
215 /*
216  * macros
217  */
218 
219 #define	pmap_resident_count(pmap)	((pmap)->pm_stats.resident_count)
220 #define	pmap_wired_count(pmap)		((pmap)->pm_stats.wired_count)
221 
222 #define pmap_clear_modify(pg)		pmap_clear_attrs(pg, PG_M)
223 #define pmap_clear_reference(pg)	pmap_clear_attrs(pg, PG_U)
224 #define pmap_copy(DP,SP,D,L,S)		__USE(L)
225 #define pmap_is_modified(pg)		pmap_test_attrs(pg, PG_M)
226 #define pmap_is_referenced(pg)		pmap_test_attrs(pg, PG_U)
227 #define pmap_move(DP,SP,D,L,S)
228 #define pmap_phys_address(ppn)		(x86_ptob(ppn) & ~X86_MMAP_FLAG_MASK)
229 #define pmap_mmap_flags(ppn)		x86_mmap_flags(ppn)
230 #define pmap_valid_entry(E) 		((E) & PG_V) /* is PDE or PTE valid? */
231 
232 #if defined(__x86_64__) || defined(PAE)
233 #define X86_MMAP_FLAG_SHIFT	(64 - PGSHIFT)
234 #else
235 #define X86_MMAP_FLAG_SHIFT	(32 - PGSHIFT)
236 #endif
237 
238 #define X86_MMAP_FLAG_MASK	0xf
239 #define X86_MMAP_FLAG_PREFETCH	0x1
240 
241 /*
242  * prototypes
243  */
244 
245 void		pmap_activate(struct lwp *);
246 void		pmap_bootstrap(vaddr_t);
247 bool		pmap_clear_attrs(struct vm_page *, unsigned);
248 bool		pmap_pv_clear_attrs(paddr_t, unsigned);
249 void		pmap_deactivate(struct lwp *);
250 void		pmap_page_remove(struct vm_page *);
251 void		pmap_pv_remove(paddr_t);
252 void		pmap_remove(struct pmap *, vaddr_t, vaddr_t);
253 bool		pmap_test_attrs(struct vm_page *, unsigned);
254 void		pmap_write_protect(struct pmap *, vaddr_t, vaddr_t, vm_prot_t);
255 void		pmap_load(void);
256 paddr_t		pmap_init_tmp_pgtbl(paddr_t);
257 void		pmap_remove_all(struct pmap *);
258 void		pmap_ldt_cleanup(struct lwp *);
259 void		pmap_ldt_sync(struct pmap *);
260 void		pmap_kremove_local(vaddr_t, vsize_t);
261 
262 void		pmap_emap_enter(vaddr_t, paddr_t, vm_prot_t);
263 void		pmap_emap_remove(vaddr_t, vsize_t);
264 void		pmap_emap_sync(bool);
265 
266 #define	__HAVE_PMAP_PV_TRACK	1
267 void		pmap_pv_init(void);
268 void		pmap_pv_track(paddr_t, psize_t);
269 void		pmap_pv_untrack(paddr_t, psize_t);
270 
271 void		pmap_map_ptes(struct pmap *, struct pmap **, pd_entry_t **,
272 		    pd_entry_t * const **);
273 void		pmap_unmap_ptes(struct pmap *, struct pmap *);
274 
275 int		pmap_pdes_invalid(vaddr_t, pd_entry_t * const *, pd_entry_t *);
276 
277 u_int		x86_mmap_flags(paddr_t);
278 
279 bool		pmap_is_curpmap(struct pmap *);
280 
281 #ifndef __HAVE_DIRECT_MAP
282 void		pmap_vpage_cpu_init(struct cpu_info *);
283 #endif
284 
285 vaddr_t reserve_dumppages(vaddr_t); /* XXX: not a pmap fn */
286 
287 typedef enum tlbwhy {
288 	TLBSHOOT_APTE,
289 	TLBSHOOT_KENTER,
290 	TLBSHOOT_KREMOVE,
291 	TLBSHOOT_FREE_PTP1,
292 	TLBSHOOT_FREE_PTP2,
293 	TLBSHOOT_REMOVE_PTE,
294 	TLBSHOOT_REMOVE_PTES,
295 	TLBSHOOT_SYNC_PV1,
296 	TLBSHOOT_SYNC_PV2,
297 	TLBSHOOT_WRITE_PROTECT,
298 	TLBSHOOT_ENTER,
299 	TLBSHOOT_UPDATE,
300 	TLBSHOOT_BUS_DMA,
301 	TLBSHOOT_BUS_SPACE,
302 	TLBSHOOT__MAX,
303 } tlbwhy_t;
304 
305 void		pmap_tlb_init(void);
306 void		pmap_tlb_cpu_init(struct cpu_info *);
307 void		pmap_tlb_shootdown(pmap_t, vaddr_t, pt_entry_t, tlbwhy_t);
308 void		pmap_tlb_shootnow(void);
309 void		pmap_tlb_intr(void);
310 
311 #define	__HAVE_PMAP_EMAP
312 
313 #define PMAP_GROWKERNEL		/* turn on pmap_growkernel interface */
314 #define PMAP_FORK		/* turn on pmap_fork interface */
315 
316 /*
317  * Do idle page zero'ing uncached to avoid polluting the cache.
318  */
319 bool	pmap_pageidlezero(paddr_t);
320 #define	PMAP_PAGEIDLEZERO(pa)	pmap_pageidlezero((pa))
321 
322 /*
323  * inline functions
324  */
325 
326 __inline static bool __unused
327 pmap_pdes_valid(vaddr_t va, pd_entry_t * const *pdes, pd_entry_t *lastpde)
328 {
329 	return pmap_pdes_invalid(va, pdes, lastpde) == 0;
330 }
331 
332 /*
333  * pmap_update_pg: flush one page from the TLB (or flush the whole thing
334  *	if hardware doesn't support one-page flushing)
335  */
336 
337 __inline static void __unused
338 pmap_update_pg(vaddr_t va)
339 {
340 	invlpg(va);
341 }
342 
343 /*
344  * pmap_update_2pg: flush two pages from the TLB
345  */
346 
347 __inline static void __unused
348 pmap_update_2pg(vaddr_t va, vaddr_t vb)
349 {
350 	invlpg(va);
351 	invlpg(vb);
352 }
353 
354 /*
355  * pmap_page_protect: change the protection of all recorded mappings
356  *	of a managed page
357  *
358  * => this function is a frontend for pmap_page_remove/pmap_clear_attrs
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_page_protect(struct vm_page *pg, vm_prot_t prot)
365 {
366 	if ((prot & VM_PROT_WRITE) == 0) {
367 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
368 			(void) pmap_clear_attrs(pg, PG_RW);
369 		} else {
370 			pmap_page_remove(pg);
371 		}
372 	}
373 }
374 
375 /*
376  * pmap_pv_protect: change the protection of all recorded mappings
377  *	of an unmanaged page
378  */
379 
380 __inline static void __unused
381 pmap_pv_protect(paddr_t pa, vm_prot_t prot)
382 {
383 	if ((prot & VM_PROT_WRITE) == 0) {
384 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
385 			(void) pmap_pv_clear_attrs(pa, PG_RW);
386 		} else {
387 			pmap_pv_remove(pa);
388 		}
389 	}
390 }
391 
392 /*
393  * pmap_protect: change the protection of pages in a pmap
394  *
395  * => this function is a frontend for pmap_remove/pmap_write_protect
396  * => we only have to worry about making the page more protected.
397  *	unprotecting a page is done on-demand at fault time.
398  */
399 
400 __inline static void __unused
401 pmap_protect(struct pmap *pmap, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
402 {
403 	if ((prot & VM_PROT_WRITE) == 0) {
404 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
405 			pmap_write_protect(pmap, sva, eva, prot);
406 		} else {
407 			pmap_remove(pmap, sva, eva);
408 		}
409 	}
410 }
411 
412 /*
413  * various address inlines
414  *
415  *  vtopte: return a pointer to the PTE mapping a VA, works only for
416  *  user and PT addresses
417  *
418  *  kvtopte: return a pointer to the PTE mapping a kernel VA
419  */
420 
421 #include <lib/libkern/libkern.h>
422 
423 static __inline pt_entry_t * __unused
424 vtopte(vaddr_t va)
425 {
426 
427 	KASSERT(va < VM_MIN_KERNEL_ADDRESS);
428 
429 	return (PTE_BASE + pl1_i(va));
430 }
431 
432 static __inline pt_entry_t * __unused
433 kvtopte(vaddr_t va)
434 {
435 	pd_entry_t *pde;
436 
437 	KASSERT(va >= VM_MIN_KERNEL_ADDRESS);
438 
439 	pde = L2_BASE + pl2_i(va);
440 	if (*pde & PG_PS)
441 		return ((pt_entry_t *)pde);
442 
443 	return (PTE_BASE + pl1_i(va));
444 }
445 
446 paddr_t vtophys(vaddr_t);
447 vaddr_t	pmap_map(vaddr_t, paddr_t, paddr_t, vm_prot_t);
448 void	pmap_cpu_init_late(struct cpu_info *);
449 bool	sse2_idlezero_page(void *);
450 
451 #ifdef XEN
452 #include <sys/bitops.h>
453 
454 #define XPTE_MASK	L1_FRAME
455 /* Selects the index of a PTE in (A)PTE_BASE */
456 #define XPTE_SHIFT	(L1_SHIFT - ilog2(sizeof(pt_entry_t)))
457 
458 /* PTE access inline fuctions */
459 
460 /*
461  * Get the machine address of the pointed pte
462  * We use hardware MMU to get value so works only for levels 1-3
463  */
464 
465 static __inline paddr_t
466 xpmap_ptetomach(pt_entry_t *pte)
467 {
468 	pt_entry_t *up_pte;
469 	vaddr_t va = (vaddr_t) pte;
470 
471 	va = ((va & XPTE_MASK) >> XPTE_SHIFT) | (vaddr_t) PTE_BASE;
472 	up_pte = (pt_entry_t *) va;
473 
474 	return (paddr_t) (((*up_pte) & PG_FRAME) + (((vaddr_t) pte) & (~PG_FRAME & ~VA_SIGN_MASK)));
475 }
476 
477 /* Xen helpers to change bits of a pte */
478 #define XPMAP_UPDATE_DIRECT	1	/* Update direct map entry flags too */
479 
480 paddr_t	vtomach(vaddr_t);
481 #define vtomfn(va) (vtomach(va) >> PAGE_SHIFT)
482 #endif	/* XEN */
483 
484 /* pmap functions with machine addresses */
485 void	pmap_kenter_ma(vaddr_t, paddr_t, vm_prot_t, u_int);
486 int	pmap_enter_ma(struct pmap *, vaddr_t, paddr_t, paddr_t,
487 	    vm_prot_t, u_int, int);
488 bool	pmap_extract_ma(pmap_t, vaddr_t, paddr_t *);
489 void	pmap_free_ptps(struct vm_page *);
490 
491 /*
492  * Hooks for the pool allocator.
493  */
494 #define	POOL_VTOPHYS(va)	vtophys((vaddr_t) (va))
495 
496 #ifdef __HAVE_DIRECT_MAP
497 
498 #define L4_SLOT_DIRECT		509
499 #define PDIR_SLOT_DIRECT	L4_SLOT_DIRECT
500 
501 #define PMAP_DIRECT_BASE	(VA_SIGN_NEG((L4_SLOT_DIRECT * NBPD_L4)))
502 #define PMAP_DIRECT_END		(VA_SIGN_NEG(((L4_SLOT_DIRECT + 1) * NBPD_L4)))
503 
504 #define PMAP_DIRECT_MAP(pa)	((vaddr_t)PMAP_DIRECT_BASE + (pa))
505 #define PMAP_DIRECT_UNMAP(va)	((paddr_t)(va) - PMAP_DIRECT_BASE)
506 
507 /*
508  * Alternate mapping hooks for pool pages.
509  */
510 #define PMAP_MAP_POOLPAGE(pa)	PMAP_DIRECT_MAP((pa))
511 #define PMAP_UNMAP_POOLPAGE(va)	PMAP_DIRECT_UNMAP((va))
512 
513 void	pagezero(vaddr_t);
514 
515 #endif /* __HAVE_DIRECT_MAP */
516 
517 #endif /* _KERNEL */
518 
519 #endif /* _X86_PMAP_H_ */
520