xref: /freebsd/sys/amd64/include/pmap.h (revision 830940567b49bb0c08dfaed40418999e76616909)
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 /* Page level cache control fields used to determine the PAT type */
79 #define PG_PDE_CACHE	(PG_PDE_PAT | PG_NC_PWT | PG_NC_PCD)
80 #define PG_PTE_CACHE	(PG_PTE_PAT | PG_NC_PWT | PG_NC_PCD)
81 
82 /*
83  * Promotion to a 2MB (PDE) page mapping requires that the corresponding 4KB
84  * (PTE) page mappings have identical settings for the following fields:
85  */
86 #define	PG_PTE_PROMOTE	(PG_NX | PG_MANAGED | PG_W | PG_G | PG_PTE_PAT | \
87 	    PG_M | PG_A | PG_NC_PCD | PG_NC_PWT | PG_U | PG_RW | PG_V)
88 
89 /*
90  * Page Protection Exception bits
91  */
92 
93 #define PGEX_P		0x01	/* Protection violation vs. not present */
94 #define PGEX_W		0x02	/* during a Write cycle */
95 #define PGEX_U		0x04	/* access from User mode (UPL) */
96 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
97 #define PGEX_I		0x10	/* during an instruction fetch */
98 
99 /*
100  * Pte related macros.  This is complicated by having to deal with
101  * the sign extension of the 48th bit.
102  */
103 #define KVADDR(l4, l3, l2, l1) ( \
104 	((unsigned long)-1 << 47) | \
105 	((unsigned long)(l4) << PML4SHIFT) | \
106 	((unsigned long)(l3) << PDPSHIFT) | \
107 	((unsigned long)(l2) << PDRSHIFT) | \
108 	((unsigned long)(l1) << PAGE_SHIFT))
109 
110 #define UVADDR(l4, l3, l2, l1) ( \
111 	((unsigned long)(l4) << PML4SHIFT) | \
112 	((unsigned long)(l3) << PDPSHIFT) | \
113 	((unsigned long)(l2) << PDRSHIFT) | \
114 	((unsigned long)(l1) << PAGE_SHIFT))
115 
116 /* Initial number of kernel page tables. */
117 #ifndef NKPT
118 #define	NKPT		32
119 #endif
120 
121 #define NKPML4E		1		/* number of kernel PML4 slots */
122 #define NKPDPE		howmany(NKPT, NPDEPG)/* number of kernel PDP slots */
123 
124 #define	NUPML4E		(NPML4EPG/2)	/* number of userland PML4 pages */
125 #define	NUPDPE		(NUPML4E*NPDPEPG)/* number of userland PDP pages */
126 #define	NUPDE		(NUPDPE*NPDEPG)	/* number of userland PD entries */
127 
128 #define	NDMPML4E	1		/* number of dmap PML4 slots */
129 
130 /*
131  * The *PDI values control the layout of virtual memory
132  */
133 #define	PML4PML4I	(NPML4EPG/2)	/* Index of recursive pml4 mapping */
134 
135 #define	KPML4I		(NPML4EPG-1)	/* Top 512GB for KVM */
136 #define	DMPML4I		(KPML4I-1)	/* Next 512GB down for direct map */
137 
138 #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */
139 
140 /*
141  * XXX doesn't really belong here I guess...
142  */
143 #define ISA_HOLE_START    0xa0000
144 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
145 
146 #ifndef LOCORE
147 
148 #include <sys/queue.h>
149 #include <sys/_lock.h>
150 #include <sys/_mutex.h>
151 
152 typedef u_int64_t pd_entry_t;
153 typedef u_int64_t pt_entry_t;
154 typedef u_int64_t pdp_entry_t;
155 typedef u_int64_t pml4_entry_t;
156 
157 #define	PML4ESHIFT	(3)
158 #define	PDPESHIFT	(3)
159 #define	PTESHIFT	(3)
160 #define	PDESHIFT	(3)
161 
162 /*
163  * Address of current address space page table maps and directories.
164  */
165 #ifdef _KERNEL
166 #define	addr_PTmap	(KVADDR(PML4PML4I, 0, 0, 0))
167 #define	addr_PDmap	(KVADDR(PML4PML4I, PML4PML4I, 0, 0))
168 #define	addr_PDPmap	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
169 #define	addr_PML4map	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
170 #define	addr_PML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
171 #define	PTmap		((pt_entry_t *)(addr_PTmap))
172 #define	PDmap		((pd_entry_t *)(addr_PDmap))
173 #define	PDPmap		((pd_entry_t *)(addr_PDPmap))
174 #define	PML4map		((pd_entry_t *)(addr_PML4map))
175 #define	PML4pml4e	((pd_entry_t *)(addr_PML4pml4e))
176 
177 extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */
178 
179 /*
180  * virtual address to page table entry and
181  * to physical address.
182  * Note: these work recursively, thus vtopte of a pte will give
183  * the corresponding pde that in turn maps it.
184  */
185 pt_entry_t *vtopte(vm_offset_t);
186 #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va)))
187 
188 static __inline pt_entry_t
189 pte_load(pt_entry_t *ptep)
190 {
191 	pt_entry_t r;
192 
193 	r = *ptep;
194 	return (r);
195 }
196 
197 static __inline pt_entry_t
198 pte_load_store(pt_entry_t *ptep, pt_entry_t pte)
199 {
200 	pt_entry_t r;
201 
202 	__asm __volatile(
203 	    "xchgq %0,%1"
204 	    : "=m" (*ptep),
205 	      "=r" (r)
206 	    : "1" (pte),
207 	      "m" (*ptep));
208 	return (r);
209 }
210 
211 #define	pte_load_clear(pte)	atomic_readandclear_long(pte)
212 
213 static __inline void
214 pte_store(pt_entry_t *ptep, pt_entry_t pte)
215 {
216 
217 	*ptep = pte;
218 }
219 
220 #define	pte_clear(ptep)		pte_store((ptep), (pt_entry_t)0ULL)
221 
222 #define	pde_store(pdep, pde)	pte_store((pdep), (pde))
223 
224 extern pt_entry_t pg_nx;
225 
226 #endif /* _KERNEL */
227 
228 /*
229  * Pmap stuff
230  */
231 struct	pv_entry;
232 struct	pv_chunk;
233 
234 struct md_page {
235 	TAILQ_HEAD(,pv_entry)	pv_list;
236 	int			pat_mode;
237 };
238 
239 /*
240  * The kernel virtual address (KVA) of the level 4 page table page is always
241  * within the direct map (DMAP) region.
242  */
243 struct pmap {
244 	struct mtx		pm_mtx;
245 	pml4_entry_t		*pm_pml4;	/* KVA of level 4 page table */
246 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
247 	u_int			pm_active;	/* active on cpus */
248 	/* spare u_int here due to padding */
249 	struct pmap_statistics	pm_stats;	/* pmap statistics */
250 	vm_page_t		pm_root;	/* spare page table pages */
251 };
252 
253 typedef struct pmap	*pmap_t;
254 
255 #ifdef _KERNEL
256 extern struct pmap	kernel_pmap_store;
257 #define kernel_pmap	(&kernel_pmap_store)
258 
259 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
260 #define	PMAP_LOCK_ASSERT(pmap, type) \
261 				mtx_assert(&(pmap)->pm_mtx, (type))
262 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
263 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
264 				    NULL, MTX_DEF | MTX_DUPOK)
265 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
266 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
267 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
268 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
269 #endif
270 
271 /*
272  * For each vm_page_t, there is a list of all currently valid virtual
273  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
274  */
275 typedef struct pv_entry {
276 	vm_offset_t	pv_va;		/* virtual address for mapping */
277 	TAILQ_ENTRY(pv_entry)	pv_list;
278 } *pv_entry_t;
279 
280 /*
281  * pv_entries are allocated in chunks per-process.  This avoids the
282  * need to track per-pmap assignments.
283  */
284 #define	_NPCM	3
285 #define	_NPCPV	168
286 struct pv_chunk {
287 	pmap_t			pc_pmap;
288 	TAILQ_ENTRY(pv_chunk)	pc_list;
289 	uint64_t		pc_map[_NPCM];	/* bitmap; 1 = free */
290 	uint64_t		pc_spare[2];
291 	struct pv_entry		pc_pventry[_NPCPV];
292 };
293 
294 #ifdef	_KERNEL
295 
296 extern caddr_t	CADDR1;
297 extern pt_entry_t *CMAP1;
298 extern vm_paddr_t phys_avail[];
299 extern vm_paddr_t dump_avail[];
300 extern vm_offset_t virtual_avail;
301 extern vm_offset_t virtual_end;
302 
303 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
304 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
305 
306 void	pmap_bootstrap(vm_paddr_t *);
307 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
308 void	pmap_init_pat(void);
309 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
310 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
311 vm_paddr_t pmap_kextract(vm_offset_t);
312 void	pmap_kremove(vm_offset_t);
313 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
314 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
315 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
316 boolean_t pmap_page_is_mapped(vm_page_t m);
317 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
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