xref: /freebsd/sys/amd64/include/pmap.h (revision acd3428b7d3e94cef0e1881c868cb4b131d4ff41)
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=4M)	*/
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_PROT		(PG_RW|PG_U)	/* all protection bits . */
75 #define PG_N		(PG_NC_PWT|PG_NC_PCD)	/* Non-cacheable */
76 
77 /*
78  * Page Protection Exception bits
79  */
80 
81 #define PGEX_P		0x01	/* Protection violation vs. not present */
82 #define PGEX_W		0x02	/* during a Write cycle */
83 #define PGEX_U		0x04	/* access from User mode (UPL) */
84 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
85 #define PGEX_I		0x10	/* during an instruction fetch */
86 
87 /*
88  * Pte related macros.  This is complicated by having to deal with
89  * the sign extension of the 48th bit.
90  */
91 #define KVADDR(l4, l3, l2, l1) ( \
92 	((unsigned long)-1 << 47) | \
93 	((unsigned long)(l4) << PML4SHIFT) | \
94 	((unsigned long)(l3) << PDPSHIFT) | \
95 	((unsigned long)(l2) << PDRSHIFT) | \
96 	((unsigned long)(l1) << PAGE_SHIFT))
97 
98 #define UVADDR(l4, l3, l2, l1) ( \
99 	((unsigned long)(l4) << PML4SHIFT) | \
100 	((unsigned long)(l3) << PDPSHIFT) | \
101 	((unsigned long)(l2) << PDRSHIFT) | \
102 	((unsigned long)(l1) << PAGE_SHIFT))
103 
104 /* Initial number of kernel page tables */
105 #ifndef NKPT
106 #define	NKPT		240	/* Enough for 16GB (2MB page tables) */
107 #endif
108 
109 #define NKPML4E		1		/* number of kernel PML4 slots */
110 #define NKPDPE		1		/* number of kernel PDP slots */
111 #define	NKPDE		(NKPDPE*NPDEPG)	/* number of kernel PD slots */
112 
113 #define	NUPML4E		(NPML4EPG/2)	/* number of userland PML4 pages */
114 #define	NUPDPE		(NUPML4E*NPDPEPG)/* number of userland PDP pages */
115 #define	NUPDE		(NUPDPE*NPDEPG)	/* number of userland PD entries */
116 
117 #define	NDMPML4E	1		/* number of dmap PML4 slots */
118 
119 /*
120  * The *PDI values control the layout of virtual memory
121  */
122 #define	PML4PML4I	(NPML4EPG/2)	/* Index of recursive pml4 mapping */
123 
124 #define	KPML4I		(NPML4EPG-1)	/* Top 512GB for KVM */
125 #define	DMPML4I		(KPML4I-1)	/* Next 512GB down for direct map */
126 
127 #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */
128 
129 /*
130  * XXX doesn't really belong here I guess...
131  */
132 #define ISA_HOLE_START    0xa0000
133 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
134 
135 #ifndef LOCORE
136 
137 #include <sys/queue.h>
138 #include <sys/_lock.h>
139 #include <sys/_mutex.h>
140 
141 typedef u_int64_t pd_entry_t;
142 typedef u_int64_t pt_entry_t;
143 typedef u_int64_t pdp_entry_t;
144 typedef u_int64_t pml4_entry_t;
145 
146 #define	PML4ESHIFT	(3)
147 #define	PDPESHIFT	(3)
148 #define	PTESHIFT	(3)
149 #define	PDESHIFT	(3)
150 
151 /*
152  * Address of current and alternate address space page table maps
153  * and directories.
154  * XXX it might be saner to just direct map all of physical memory
155  * into the kernel using 2MB pages.  We have enough space to do
156  * it (2^47 bits of KVM, while current max physical addressability
157  * is 2^40 physical bits).  Then we can get rid of the evil hole
158  * in the page tables and the evil overlapping.
159  */
160 #ifdef _KERNEL
161 #define	addr_PTmap	(KVADDR(PML4PML4I, 0, 0, 0))
162 #define	addr_PDmap	(KVADDR(PML4PML4I, PML4PML4I, 0, 0))
163 #define	addr_PDPmap	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
164 #define	addr_PML4map	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
165 #define	addr_PML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
166 #define	PTmap		((pt_entry_t *)(addr_PTmap))
167 #define	PDmap		((pd_entry_t *)(addr_PDmap))
168 #define	PDPmap		((pd_entry_t *)(addr_PDPmap))
169 #define	PML4map		((pd_entry_t *)(addr_PML4map))
170 #define	PML4pml4e	((pd_entry_t *)(addr_PML4pml4e))
171 
172 extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */
173 #endif
174 
175 #ifdef _KERNEL
176 /*
177  * virtual address to page table entry and
178  * to physical address.
179  * Note: these work recursively, thus vtopte of a pte will give
180  * the corresponding pde that in turn maps it.
181  */
182 pt_entry_t *vtopte(vm_offset_t);
183 #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va)))
184 
185 static __inline pt_entry_t
186 pte_load(pt_entry_t *ptep)
187 {
188 	pt_entry_t r;
189 
190 	r = *ptep;
191 	return (r);
192 }
193 
194 static __inline pt_entry_t
195 pte_load_store(pt_entry_t *ptep, pt_entry_t pte)
196 {
197 	pt_entry_t r;
198 
199 	__asm __volatile(
200 	    "xchgq %0,%1"
201 	    : "=m" (*ptep),
202 	      "=r" (r)
203 	    : "1" (pte),
204 	      "m" (*ptep));
205 	return (r);
206 }
207 
208 #define	pte_load_clear(pte)	atomic_readandclear_long(pte)
209 
210 static __inline void
211 pte_store(pt_entry_t *ptep, pt_entry_t pte)
212 {
213 
214 	*ptep = pte;
215 }
216 
217 #define	pte_clear(ptep)		pte_store((ptep), (pt_entry_t)0ULL)
218 
219 #define	pde_store(pdep, pde)	pte_store((pdep), (pde))
220 
221 extern pt_entry_t pg_nx;
222 
223 #endif /* _KERNEL */
224 
225 /*
226  * Pmap stuff
227  */
228 struct	pv_entry;
229 struct	pv_chunk;
230 
231 struct md_page {
232 	int pv_list_count;
233 	TAILQ_HEAD(,pv_entry)	pv_list;
234 };
235 
236 struct pmap {
237 	struct mtx		pm_mtx;
238 	pml4_entry_t		*pm_pml4;	/* KVA of level 4 page table */
239 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
240 	u_int			pm_active;	/* active on cpus */
241 	/* spare u_int here due to padding */
242 	struct pmap_statistics	pm_stats;	/* pmap statistics */
243 };
244 
245 typedef struct pmap	*pmap_t;
246 
247 #ifdef _KERNEL
248 extern struct pmap	kernel_pmap_store;
249 #define kernel_pmap	(&kernel_pmap_store)
250 
251 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
252 #define	PMAP_LOCK_ASSERT(pmap, type) \
253 				mtx_assert(&(pmap)->pm_mtx, (type))
254 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
255 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
256 				    NULL, MTX_DEF | MTX_DUPOK)
257 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
258 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
259 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
260 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
261 #endif
262 
263 /*
264  * For each vm_page_t, there is a list of all currently valid virtual
265  * mappings of that page.  An entry is a pv_entry_t, the list is pv_table.
266  */
267 typedef struct pv_entry {
268 	vm_offset_t	pv_va;		/* virtual address for mapping */
269 	TAILQ_ENTRY(pv_entry)	pv_list;
270 } *pv_entry_t;
271 
272 /*
273  * pv_entries are allocated in chunks per-process.  This avoids the
274  * need to track per-pmap assignments.
275  */
276 #define	_NPCM	3
277 #define	_NPCPV	168
278 struct pv_chunk {
279 	pmap_t			pc_pmap;
280 	TAILQ_ENTRY(pv_chunk)	pc_list;
281 	uint64_t		pc_map[_NPCM];	/* bitmap; 1 = free */
282 	uint64_t		pc_spare[2];
283 	struct pv_entry		pc_pventry[_NPCPV];
284 };
285 
286 #ifdef	_KERNEL
287 
288 #define NPPROVMTRR		8
289 #define PPRO_VMTRRphysBase0	0x200
290 #define PPRO_VMTRRphysMask0	0x201
291 struct ppro_vmtrr {
292 	u_int64_t base, mask;
293 };
294 extern struct ppro_vmtrr PPro_vmtrr[NPPROVMTRR];
295 
296 extern caddr_t	CADDR1;
297 extern pt_entry_t *CMAP1;
298 extern vm_paddr_t avail_end;
299 extern vm_paddr_t phys_avail[];
300 extern vm_paddr_t dump_avail[];
301 extern vm_offset_t virtual_avail;
302 extern vm_offset_t virtual_end;
303 
304 #define	pmap_page_is_mapped(m)	(!TAILQ_EMPTY(&(m)->md.pv_list))
305 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
306 
307 void	pmap_bootstrap(vm_paddr_t *);
308 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
309 void	pmap_init_pat(void);
310 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
311 void	pmap_kenter_attr(vm_offset_t va, vm_paddr_t pa, int mode);
312 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
313 vm_paddr_t pmap_kextract(vm_offset_t);
314 void	pmap_kremove(vm_offset_t);
315 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
316 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
317 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
318 void	pmap_unmapdev(vm_offset_t, vm_size_t);
319 void	pmap_invalidate_page(pmap_t, vm_offset_t);
320 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
321 void	pmap_invalidate_all(pmap_t);
322 void	pmap_invalidate_cache(void);
323 
324 #endif /* _KERNEL */
325 
326 #endif /* !LOCORE */
327 
328 #endif /* !_MACHINE_PMAP_H_ */
329