xref: /freebsd/sys/amd64/include/pmap.h (revision ce3adf4362fcca6a43e500b2531f0038adbfbd21)
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 /*
117  * Number of kernel PML4 slots.  Can be anywhere from 1 to 64 or so,
118  * but setting it larger than NDMPML4E makes no sense.
119  *
120  * Each slot provides .5 TB of kernel virtual space.
121  */
122 #define NKPML4E		4
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 /*
129  * NDMPML4E is the maximum number of PML4 entries that will be
130  * used to implement the direct map.  It must be a power of two,
131  * and should generally exceed NKPML4E.  The maximum possible
132  * value is 64; using 128 will make the direct map intrude into
133  * the recursive page table map.
134  */
135 #define	NDMPML4E	8
136 
137 /*
138  * These values control the layout of virtual memory.  The starting address
139  * of the direct map, which is controlled by DMPML4I, must be a multiple of
140  * its size.  (See the PHYS_TO_DMAP() and DMAP_TO_PHYS() macros.)
141  *
142  * Note: KPML4I is the index of the (single) level 4 page that maps
143  * the KVA that holds KERNBASE, while KPML4BASE is the index of the
144  * first level 4 page that maps VM_MIN_KERNEL_ADDRESS.  If NKPML4E
145  * is 1, these are the same, otherwise KPML4BASE < KPML4I and extra
146  * level 4 PDEs are needed to map from VM_MIN_KERNEL_ADDRESS up to
147  * KERNBASE.
148  *
149  * (KPML4I combines with KPDPI to choose where KERNBASE starts.
150  * Or, in other words, KPML4I provides bits 39..47 of KERNBASE,
151  * and KPDPI provides bits 30..38.)
152  */
153 #define	PML4PML4I	(NPML4EPG/2)	/* Index of recursive pml4 mapping */
154 
155 #define	KPML4BASE	(NPML4EPG-NKPML4E) /* KVM at highest addresses */
156 #define	DMPML4I		rounddown(KPML4BASE-NDMPML4E, NDMPML4E) /* Below KVM */
157 
158 #define	KPML4I		(NPML4EPG-1)
159 #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */
160 
161 /*
162  * XXX doesn't really belong here I guess...
163  */
164 #define ISA_HOLE_START    0xa0000
165 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
166 
167 #ifndef LOCORE
168 
169 #include <sys/queue.h>
170 #include <sys/_cpuset.h>
171 #include <sys/_lock.h>
172 #include <sys/_mutex.h>
173 
174 #include <vm/_vm_radix.h>
175 
176 typedef u_int64_t pd_entry_t;
177 typedef u_int64_t pt_entry_t;
178 typedef u_int64_t pdp_entry_t;
179 typedef u_int64_t pml4_entry_t;
180 
181 /*
182  * Address of current address space page table maps and directories.
183  */
184 #ifdef _KERNEL
185 #define	addr_PTmap	(KVADDR(PML4PML4I, 0, 0, 0))
186 #define	addr_PDmap	(KVADDR(PML4PML4I, PML4PML4I, 0, 0))
187 #define	addr_PDPmap	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
188 #define	addr_PML4map	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
189 #define	addr_PML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
190 #define	PTmap		((pt_entry_t *)(addr_PTmap))
191 #define	PDmap		((pd_entry_t *)(addr_PDmap))
192 #define	PDPmap		((pd_entry_t *)(addr_PDPmap))
193 #define	PML4map		((pd_entry_t *)(addr_PML4map))
194 #define	PML4pml4e	((pd_entry_t *)(addr_PML4pml4e))
195 
196 extern int nkpt;		/* Initial number of kernel page tables */
197 extern u_int64_t KPDPphys;	/* physical address of kernel level 3 */
198 extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */
199 
200 /*
201  * virtual address to page table entry and
202  * to physical address.
203  * Note: these work recursively, thus vtopte of a pte will give
204  * the corresponding pde that in turn maps it.
205  */
206 pt_entry_t *vtopte(vm_offset_t);
207 #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va)))
208 
209 #define	pte_load_store(ptep, pte)	atomic_swap_long(ptep, pte)
210 #define	pte_load_clear(ptep)		atomic_swap_long(ptep, 0)
211 #define	pte_store(ptep, pte) do { \
212 	*(u_long *)(ptep) = (u_long)(pte); \
213 } while (0)
214 #define	pte_clear(ptep)			pte_store(ptep, 0)
215 
216 #define	pde_store(pdep, pde)		pte_store(pdep, pde)
217 
218 extern pt_entry_t pg_nx;
219 
220 #endif /* _KERNEL */
221 
222 /*
223  * Pmap stuff
224  */
225 struct	pv_entry;
226 struct	pv_chunk;
227 
228 struct md_page {
229 	TAILQ_HEAD(,pv_entry)	pv_list;
230 	int			pv_gen;
231 	int			pat_mode;
232 };
233 
234 enum pmap_type {
235 	PT_X86,			/* regular x86 page tables */
236 	PT_EPT,			/* Intel's nested page tables */
237 	PT_RVI,			/* AMD's nested page tables */
238 };
239 
240 /*
241  * The kernel virtual address (KVA) of the level 4 page table page is always
242  * within the direct map (DMAP) region.
243  */
244 struct pmap {
245 	struct mtx		pm_mtx;
246 	pml4_entry_t		*pm_pml4;	/* KVA of level 4 page table */
247 	uint64_t		pm_cr3;
248 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
249 	cpuset_t		pm_active;	/* active on cpus */
250 	cpuset_t		pm_save;	/* Context valid on cpus mask */
251 	int			pm_pcid;	/* context id */
252 	enum pmap_type		pm_type;	/* regular or nested tables */
253 	struct pmap_statistics	pm_stats;	/* pmap statistics */
254 	struct vm_radix		pm_root;	/* spare page table pages */
255 	long			pm_eptgen;	/* EPT pmap generation id */
256 	int			pm_flags;
257 };
258 
259 typedef struct pmap	*pmap_t;
260 
261 #ifdef _KERNEL
262 extern struct pmap	kernel_pmap_store;
263 #define kernel_pmap	(&kernel_pmap_store)
264 
265 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
266 #define	PMAP_LOCK_ASSERT(pmap, type) \
267 				mtx_assert(&(pmap)->pm_mtx, (type))
268 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
269 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
270 				    NULL, MTX_DEF | MTX_DUPOK)
271 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
272 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
273 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
274 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
275 #endif
276 
277 /*
278  * For each vm_page_t, there is a list of all currently valid virtual
279  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
280  */
281 typedef struct pv_entry {
282 	vm_offset_t	pv_va;		/* virtual address for mapping */
283 	TAILQ_ENTRY(pv_entry)	pv_next;
284 } *pv_entry_t;
285 
286 /*
287  * pv_entries are allocated in chunks per-process.  This avoids the
288  * need to track per-pmap assignments.
289  */
290 #define	_NPCM	3
291 #define	_NPCPV	168
292 struct pv_chunk {
293 	pmap_t			pc_pmap;
294 	TAILQ_ENTRY(pv_chunk)	pc_list;
295 	uint64_t		pc_map[_NPCM];	/* bitmap; 1 = free */
296 	TAILQ_ENTRY(pv_chunk)	pc_lru;
297 	struct pv_entry		pc_pventry[_NPCPV];
298 };
299 
300 #ifdef	_KERNEL
301 
302 extern caddr_t	CADDR1;
303 extern pt_entry_t *CMAP1;
304 extern vm_paddr_t phys_avail[];
305 extern vm_paddr_t dump_avail[];
306 extern vm_offset_t virtual_avail;
307 extern vm_offset_t virtual_end;
308 
309 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
310 #define	pmap_page_is_write_mapped(m)	(((m)->aflags & PGA_WRITEABLE) != 0)
311 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
312 
313 void	pmap_bootstrap(vm_paddr_t *);
314 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
315 void	pmap_demote_DMAP(vm_paddr_t base, vm_size_t len, boolean_t invalidate);
316 void	pmap_init_pat(void);
317 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
318 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
319 vm_paddr_t pmap_kextract(vm_offset_t);
320 void	pmap_kremove(vm_offset_t);
321 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
322 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
323 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
324 boolean_t pmap_page_is_mapped(vm_page_t m);
325 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
326 void	pmap_unmapdev(vm_offset_t, vm_size_t);
327 void	pmap_invalidate_page(pmap_t, vm_offset_t);
328 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
329 void	pmap_invalidate_all(pmap_t);
330 void	pmap_invalidate_cache(void);
331 void	pmap_invalidate_cache_pages(vm_page_t *pages, int count);
332 void	pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva);
333 
334 #endif /* _KERNEL */
335 
336 #endif /* !LOCORE */
337 
338 #endif /* !_MACHINE_PMAP_H_ */
339