xref: /freebsd/sys/amd64/include/pmap.h (revision 2f6a179eb910129fb812c1ad1bdc300da1203dc0)
1 /*-
2  * Copyright (c) 2003 Peter Wemm.
3  * Copyright (c) 1991 Regents of the University of California.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department and William Jolitz of UUNET Technologies Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * Derived from hp300 version by Mike Hibler, this version by William
35  * Jolitz uses a recursive map [a pde points to the page directory] to
36  * map the page tables using the pagetables themselves. This is done to
37  * reduce the impact on kernel virtual memory for lots of sparse address
38  * space, and to reduce the cost of memory to each process.
39  *
40  *	from: hp300: @(#)pmap.h	7.2 (Berkeley) 12/16/90
41  *	from: @(#)pmap.h	7.4 (Berkeley) 5/12/91
42  * $FreeBSD$
43  */
44 
45 #ifndef _MACHINE_PMAP_H_
46 #define	_MACHINE_PMAP_H_
47 
48 /*
49  * Page-directory and page-table entries follow this format, with a few
50  * of the fields not present here and there, depending on a lot of things.
51  */
52 				/* ---- Intel Nomenclature ---- */
53 #define	PG_V		0x001	/* P	Valid			*/
54 #define PG_RW		0x002	/* R/W	Read/Write		*/
55 #define PG_U		0x004	/* U/S  User/Supervisor		*/
56 #define	PG_NC_PWT	0x008	/* PWT	Write through		*/
57 #define	PG_NC_PCD	0x010	/* PCD	Cache disable		*/
58 #define PG_A		0x020	/* A	Accessed		*/
59 #define	PG_M		0x040	/* D	Dirty			*/
60 #define	PG_PS		0x080	/* PS	Page size (0=4k,1=2M)	*/
61 #define	PG_PTE_PAT	0x080	/* PAT	PAT index		*/
62 #define	PG_G		0x100	/* G	Global			*/
63 #define	PG_AVAIL1	0x200	/*    /	Available for system	*/
64 #define	PG_AVAIL2	0x400	/*   <	programmers use		*/
65 #define	PG_AVAIL3	0x800	/*    \				*/
66 #define	PG_PDE_PAT	0x1000	/* PAT	PAT index		*/
67 #define	PG_NX		(1ul<<63) /* No-execute */
68 
69 
70 /* Our various interpretations of the above */
71 #define PG_W		PG_AVAIL1	/* "Wired" pseudoflag */
72 #define	PG_MANAGED	PG_AVAIL2
73 #define	PG_FRAME	(0x000ffffffffff000ul)
74 #define	PG_PS_FRAME	(0x000fffffffe00000ul)
75 #define	PG_PROT		(PG_RW|PG_U)	/* all protection bits . */
76 #define PG_N		(PG_NC_PWT|PG_NC_PCD)	/* Non-cacheable */
77 
78 /*
79  * Promotion to a 2MB (PDE) page mapping requires that the corresponding 4KB
80  * (PTE) page mappings have identical settings for the following fields:
81  */
82 #define	PG_PTE_PROMOTE	(PG_NX | PG_MANAGED | PG_W | PG_G | PG_PTE_PAT | \
83 	    PG_M | PG_A | PG_NC_PCD | PG_NC_PWT | PG_U | PG_RW | PG_V)
84 
85 /*
86  * Page Protection Exception bits
87  */
88 
89 #define PGEX_P		0x01	/* Protection violation vs. not present */
90 #define PGEX_W		0x02	/* during a Write cycle */
91 #define PGEX_U		0x04	/* access from User mode (UPL) */
92 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
93 #define PGEX_I		0x10	/* during an instruction fetch */
94 
95 /*
96  * Pte related macros.  This is complicated by having to deal with
97  * the sign extension of the 48th bit.
98  */
99 #define KVADDR(l4, l3, l2, l1) ( \
100 	((unsigned long)-1 << 47) | \
101 	((unsigned long)(l4) << PML4SHIFT) | \
102 	((unsigned long)(l3) << PDPSHIFT) | \
103 	((unsigned long)(l2) << PDRSHIFT) | \
104 	((unsigned long)(l1) << PAGE_SHIFT))
105 
106 #define UVADDR(l4, l3, l2, l1) ( \
107 	((unsigned long)(l4) << PML4SHIFT) | \
108 	((unsigned long)(l3) << PDPSHIFT) | \
109 	((unsigned long)(l2) << PDRSHIFT) | \
110 	((unsigned long)(l1) << PAGE_SHIFT))
111 
112 /* Initial number of kernel page tables. */
113 #ifndef NKPT
114 #define	NKPT		32
115 #endif
116 
117 #define NKPML4E		1		/* number of kernel PML4 slots */
118 #define NKPDPE		howmany(NKPT, NPDEPG)/* number of kernel PDP slots */
119 
120 #define	NUPML4E		(NPML4EPG/2)	/* number of userland PML4 pages */
121 #define	NUPDPE		(NUPML4E*NPDPEPG)/* number of userland PDP pages */
122 #define	NUPDE		(NUPDPE*NPDEPG)	/* number of userland PD entries */
123 
124 #define	NDMPML4E	1		/* number of dmap PML4 slots */
125 
126 /*
127  * The *PDI values control the layout of virtual memory
128  */
129 #define	PML4PML4I	(NPML4EPG/2)	/* Index of recursive pml4 mapping */
130 
131 #define	KPML4I		(NPML4EPG-1)	/* Top 512GB for KVM */
132 #define	DMPML4I		(KPML4I-1)	/* Next 512GB down for direct map */
133 
134 #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */
135 
136 /*
137  * XXX doesn't really belong here I guess...
138  */
139 #define ISA_HOLE_START    0xa0000
140 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
141 
142 #ifndef LOCORE
143 
144 #include <sys/queue.h>
145 #include <sys/_lock.h>
146 #include <sys/_mutex.h>
147 
148 typedef u_int64_t pd_entry_t;
149 typedef u_int64_t pt_entry_t;
150 typedef u_int64_t pdp_entry_t;
151 typedef u_int64_t pml4_entry_t;
152 
153 #define	PML4ESHIFT	(3)
154 #define	PDPESHIFT	(3)
155 #define	PTESHIFT	(3)
156 #define	PDESHIFT	(3)
157 
158 /*
159  * Address of current and alternate address space page table maps
160  * and directories.
161  * XXX it might be saner to just direct map all of physical memory
162  * into the kernel using 2MB pages.  We have enough space to do
163  * it (2^47 bits of KVM, while current max physical addressability
164  * is 2^40 physical bits).  Then we can get rid of the evil hole
165  * in the page tables and the evil overlapping.
166  */
167 #ifdef _KERNEL
168 #define	addr_PTmap	(KVADDR(PML4PML4I, 0, 0, 0))
169 #define	addr_PDmap	(KVADDR(PML4PML4I, PML4PML4I, 0, 0))
170 #define	addr_PDPmap	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
171 #define	addr_PML4map	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
172 #define	addr_PML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
173 #define	PTmap		((pt_entry_t *)(addr_PTmap))
174 #define	PDmap		((pd_entry_t *)(addr_PDmap))
175 #define	PDPmap		((pd_entry_t *)(addr_PDPmap))
176 #define	PML4map		((pd_entry_t *)(addr_PML4map))
177 #define	PML4pml4e	((pd_entry_t *)(addr_PML4pml4e))
178 
179 extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */
180 #endif
181 
182 #ifdef _KERNEL
183 /*
184  * virtual address to page table entry and
185  * to physical address.
186  * Note: these work recursively, thus vtopte of a pte will give
187  * the corresponding pde that in turn maps it.
188  */
189 pt_entry_t *vtopte(vm_offset_t);
190 #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va)))
191 
192 static __inline pt_entry_t
193 pte_load(pt_entry_t *ptep)
194 {
195 	pt_entry_t r;
196 
197 	r = *ptep;
198 	return (r);
199 }
200 
201 static __inline pt_entry_t
202 pte_load_store(pt_entry_t *ptep, pt_entry_t pte)
203 {
204 	pt_entry_t r;
205 
206 	__asm __volatile(
207 	    "xchgq %0,%1"
208 	    : "=m" (*ptep),
209 	      "=r" (r)
210 	    : "1" (pte),
211 	      "m" (*ptep));
212 	return (r);
213 }
214 
215 #define	pte_load_clear(pte)	atomic_readandclear_long(pte)
216 
217 static __inline void
218 pte_store(pt_entry_t *ptep, pt_entry_t pte)
219 {
220 
221 	*ptep = pte;
222 }
223 
224 #define	pte_clear(ptep)		pte_store((ptep), (pt_entry_t)0ULL)
225 
226 #define	pde_store(pdep, pde)	pte_store((pdep), (pde))
227 
228 extern pt_entry_t pg_nx;
229 
230 #endif /* _KERNEL */
231 
232 /*
233  * Pmap stuff
234  */
235 struct	pv_entry;
236 struct	pv_chunk;
237 
238 struct md_page {
239 	TAILQ_HEAD(,pv_entry)	pv_list;
240 };
241 
242 struct pmap {
243 	struct mtx		pm_mtx;
244 	pml4_entry_t		*pm_pml4;	/* KVA of level 4 page table */
245 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
246 	u_int			pm_active;	/* active on cpus */
247 	/* spare u_int here due to padding */
248 	struct pmap_statistics	pm_stats;	/* pmap statistics */
249 	vm_page_t		pm_root;	/* spare page table pages */
250 };
251 
252 typedef struct pmap	*pmap_t;
253 
254 #ifdef _KERNEL
255 extern struct pmap	kernel_pmap_store;
256 #define kernel_pmap	(&kernel_pmap_store)
257 
258 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
259 #define	PMAP_LOCK_ASSERT(pmap, type) \
260 				mtx_assert(&(pmap)->pm_mtx, (type))
261 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
262 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
263 				    NULL, MTX_DEF | MTX_DUPOK)
264 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
265 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
266 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
267 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
268 #endif
269 
270 /*
271  * For each vm_page_t, there is a list of all currently valid virtual
272  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
273  */
274 typedef struct pv_entry {
275 	vm_offset_t	pv_va;		/* virtual address for mapping */
276 	TAILQ_ENTRY(pv_entry)	pv_list;
277 } *pv_entry_t;
278 
279 /*
280  * pv_entries are allocated in chunks per-process.  This avoids the
281  * need to track per-pmap assignments.
282  */
283 #define	_NPCM	3
284 #define	_NPCPV	168
285 struct pv_chunk {
286 	pmap_t			pc_pmap;
287 	TAILQ_ENTRY(pv_chunk)	pc_list;
288 	uint64_t		pc_map[_NPCM];	/* bitmap; 1 = free */
289 	uint64_t		pc_spare[2];
290 	struct pv_entry		pc_pventry[_NPCPV];
291 };
292 
293 #ifdef	_KERNEL
294 
295 #define NPPROVMTRR		8
296 #define PPRO_VMTRRphysBase0	0x200
297 #define PPRO_VMTRRphysMask0	0x201
298 struct ppro_vmtrr {
299 	u_int64_t base, mask;
300 };
301 extern struct ppro_vmtrr PPro_vmtrr[NPPROVMTRR];
302 
303 extern caddr_t	CADDR1;
304 extern pt_entry_t *CMAP1;
305 extern vm_paddr_t phys_avail[];
306 extern vm_paddr_t dump_avail[];
307 extern vm_offset_t virtual_avail;
308 extern vm_offset_t virtual_end;
309 
310 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
311 
312 void	pmap_bootstrap(vm_paddr_t *);
313 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
314 void	pmap_init_pat(void);
315 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
316 void	pmap_kenter_attr(vm_offset_t va, vm_paddr_t pa, int mode);
317 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
318 vm_paddr_t pmap_kextract(vm_offset_t);
319 void	pmap_kremove(vm_offset_t);
320 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
321 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
322 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
323 boolean_t pmap_page_is_mapped(vm_page_t m);
324 void	pmap_unmapdev(vm_offset_t, vm_size_t);
325 void	pmap_invalidate_page(pmap_t, vm_offset_t);
326 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
327 void	pmap_invalidate_all(pmap_t);
328 void	pmap_invalidate_cache(void);
329 
330 #endif /* _KERNEL */
331 
332 #endif /* !LOCORE */
333 
334 #endif /* !_MACHINE_PMAP_H_ */
335