xref: /freebsd/sys/amd64/include/pmap.h (revision 5dae51da3da0cc94d17bd67b308fad304ebec7e0)
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	X86_PG_V	0x001	/* P	Valid			*/
54 #define	X86_PG_RW	0x002	/* R/W	Read/Write		*/
55 #define	X86_PG_U	0x004	/* U/S  User/Supervisor		*/
56 #define	X86_PG_NC_PWT	0x008	/* PWT	Write through		*/
57 #define	X86_PG_NC_PCD	0x010	/* PCD	Cache disable		*/
58 #define	X86_PG_A	0x020	/* A	Accessed		*/
59 #define	X86_PG_M	0x040	/* D	Dirty			*/
60 #define	X86_PG_PS	0x080	/* PS	Page size (0=4k,1=2M)	*/
61 #define	X86_PG_PTE_PAT	0x080	/* PAT	PAT index		*/
62 #define	X86_PG_G	0x100	/* G	Global			*/
63 #define	X86_PG_AVAIL1	0x200	/*    /	Available for system	*/
64 #define	X86_PG_AVAIL2	0x400	/*   <	programmers use		*/
65 #define	X86_PG_AVAIL3	0x800	/*    \				*/
66 #define	X86_PG_PDE_PAT	0x1000	/* PAT	PAT index		*/
67 #define	X86_PG_NX	(1ul<<63) /* No-execute */
68 #define	X86_PG_AVAIL(x)	(1ul << (x))
69 
70 /* Page level cache control fields used to determine the PAT type */
71 #define	X86_PG_PDE_CACHE (X86_PG_PDE_PAT | X86_PG_NC_PWT | X86_PG_NC_PCD)
72 #define	X86_PG_PTE_CACHE (X86_PG_PTE_PAT | X86_PG_NC_PWT | X86_PG_NC_PCD)
73 
74 /*
75  * Intel extended page table (EPT) bit definitions.
76  */
77 #define	EPT_PG_READ		0x001	/* R	Read		*/
78 #define	EPT_PG_WRITE		0x002	/* W	Write		*/
79 #define	EPT_PG_EXECUTE		0x004	/* X	Execute		*/
80 #define	EPT_PG_IGNORE_PAT	0x040	/* IPAT	Ignore PAT	*/
81 #define	EPT_PG_PS		0x080	/* PS	Page size	*/
82 #define	EPT_PG_A		0x100	/* A	Accessed	*/
83 #define	EPT_PG_M		0x200	/* D	Dirty		*/
84 #define	EPT_PG_MEMORY_TYPE(x)	((x) << 3) /* MT Memory Type	*/
85 
86 /*
87  * Define the PG_xx macros in terms of the bits on x86 PTEs.
88  */
89 #define	PG_V		X86_PG_V
90 #define	PG_RW		X86_PG_RW
91 #define	PG_U		X86_PG_U
92 #define	PG_NC_PWT	X86_PG_NC_PWT
93 #define	PG_NC_PCD	X86_PG_NC_PCD
94 #define	PG_A		X86_PG_A
95 #define	PG_M		X86_PG_M
96 #define	PG_PS		X86_PG_PS
97 #define	PG_PTE_PAT	X86_PG_PTE_PAT
98 #define	PG_G		X86_PG_G
99 #define	PG_AVAIL1	X86_PG_AVAIL1
100 #define	PG_AVAIL2	X86_PG_AVAIL2
101 #define	PG_AVAIL3	X86_PG_AVAIL3
102 #define	PG_PDE_PAT	X86_PG_PDE_PAT
103 #define	PG_NX		X86_PG_NX
104 #define	PG_PDE_CACHE	X86_PG_PDE_CACHE
105 #define	PG_PTE_CACHE	X86_PG_PTE_CACHE
106 
107 /* Our various interpretations of the above */
108 #define	PG_W		X86_PG_AVAIL3	/* "Wired" pseudoflag */
109 #define	PG_MANAGED	X86_PG_AVAIL2
110 #define	EPT_PG_EMUL_V	X86_PG_AVAIL(52)
111 #define	EPT_PG_EMUL_RW	X86_PG_AVAIL(53)
112 #define	PG_FRAME	(0x000ffffffffff000ul)
113 #define	PG_PS_FRAME	(0x000fffffffe00000ul)
114 
115 /*
116  * Promotion to a 2MB (PDE) page mapping requires that the corresponding 4KB
117  * (PTE) page mappings have identical settings for the following fields:
118  */
119 #define	PG_PTE_PROMOTE	(PG_NX | PG_MANAGED | PG_W | PG_G | PG_PTE_CACHE | \
120 	    PG_M | PG_A | PG_U | PG_RW | PG_V)
121 
122 /*
123  * Page Protection Exception bits
124  */
125 
126 #define PGEX_P		0x01	/* Protection violation vs. not present */
127 #define PGEX_W		0x02	/* during a Write cycle */
128 #define PGEX_U		0x04	/* access from User mode (UPL) */
129 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
130 #define PGEX_I		0x10	/* during an instruction fetch */
131 
132 /*
133  * undef the PG_xx macros that define bits in the regular x86 PTEs that
134  * have a different position in nested PTEs. This is done when compiling
135  * code that needs to be aware of the differences between regular x86 and
136  * nested PTEs.
137  *
138  * The appropriate bitmask will be calculated at runtime based on the pmap
139  * type.
140  */
141 #ifdef AMD64_NPT_AWARE
142 #undef PG_AVAIL1		/* X86_PG_AVAIL1 aliases with EPT_PG_M */
143 #undef PG_G
144 #undef PG_A
145 #undef PG_M
146 #undef PG_PDE_PAT
147 #undef PG_PDE_CACHE
148 #undef PG_PTE_PAT
149 #undef PG_PTE_CACHE
150 #undef PG_RW
151 #undef PG_V
152 #endif
153 
154 /*
155  * Pte related macros.  This is complicated by having to deal with
156  * the sign extension of the 48th bit.
157  */
158 #define KVADDR(l4, l3, l2, l1) ( \
159 	((unsigned long)-1 << 47) | \
160 	((unsigned long)(l4) << PML4SHIFT) | \
161 	((unsigned long)(l3) << PDPSHIFT) | \
162 	((unsigned long)(l2) << PDRSHIFT) | \
163 	((unsigned long)(l1) << PAGE_SHIFT))
164 
165 #define UVADDR(l4, l3, l2, l1) ( \
166 	((unsigned long)(l4) << PML4SHIFT) | \
167 	((unsigned long)(l3) << PDPSHIFT) | \
168 	((unsigned long)(l2) << PDRSHIFT) | \
169 	((unsigned long)(l1) << PAGE_SHIFT))
170 
171 /*
172  * Number of kernel PML4 slots.  Can be anywhere from 1 to 64 or so,
173  * but setting it larger than NDMPML4E makes no sense.
174  *
175  * Each slot provides .5 TB of kernel virtual space.
176  */
177 #define NKPML4E		4
178 
179 #define	NUPML4E		(NPML4EPG/2)	/* number of userland PML4 pages */
180 #define	NUPDPE		(NUPML4E*NPDPEPG)/* number of userland PDP pages */
181 #define	NUPDE		(NUPDPE*NPDEPG)	/* number of userland PD entries */
182 
183 /*
184  * NDMPML4E is the maximum number of PML4 entries that will be
185  * used to implement the direct map.  It must be a power of two,
186  * and should generally exceed NKPML4E.  The maximum possible
187  * value is 64; using 128 will make the direct map intrude into
188  * the recursive page table map.
189  */
190 #define	NDMPML4E	8
191 
192 /*
193  * These values control the layout of virtual memory.  The starting address
194  * of the direct map, which is controlled by DMPML4I, must be a multiple of
195  * its size.  (See the PHYS_TO_DMAP() and DMAP_TO_PHYS() macros.)
196  *
197  * Note: KPML4I is the index of the (single) level 4 page that maps
198  * the KVA that holds KERNBASE, while KPML4BASE is the index of the
199  * first level 4 page that maps VM_MIN_KERNEL_ADDRESS.  If NKPML4E
200  * is 1, these are the same, otherwise KPML4BASE < KPML4I and extra
201  * level 4 PDEs are needed to map from VM_MIN_KERNEL_ADDRESS up to
202  * KERNBASE.
203  *
204  * (KPML4I combines with KPDPI to choose where KERNBASE starts.
205  * Or, in other words, KPML4I provides bits 39..47 of KERNBASE,
206  * and KPDPI provides bits 30..38.)
207  */
208 #define	PML4PML4I	(NPML4EPG/2)	/* Index of recursive pml4 mapping */
209 
210 #define	KPML4BASE	(NPML4EPG-NKPML4E) /* KVM at highest addresses */
211 #define	DMPML4I		rounddown(KPML4BASE-NDMPML4E, NDMPML4E) /* Below KVM */
212 
213 #define	KPML4I		(NPML4EPG-1)
214 #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */
215 
216 /*
217  * XXX doesn't really belong here I guess...
218  */
219 #define ISA_HOLE_START    0xa0000
220 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
221 
222 #define	PMAP_PCID_NONE		0xffffffff
223 #define	PMAP_PCID_KERN		0
224 #define	PMAP_PCID_OVERMAX	0x1000
225 
226 #ifndef LOCORE
227 
228 #include <sys/queue.h>
229 #include <sys/_cpuset.h>
230 #include <sys/_lock.h>
231 #include <sys/_mutex.h>
232 
233 #include <vm/_vm_radix.h>
234 
235 typedef u_int64_t pd_entry_t;
236 typedef u_int64_t pt_entry_t;
237 typedef u_int64_t pdp_entry_t;
238 typedef u_int64_t pml4_entry_t;
239 
240 /*
241  * Address of current address space page table maps and directories.
242  */
243 #ifdef _KERNEL
244 #define	addr_PTmap	(KVADDR(PML4PML4I, 0, 0, 0))
245 #define	addr_PDmap	(KVADDR(PML4PML4I, PML4PML4I, 0, 0))
246 #define	addr_PDPmap	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
247 #define	addr_PML4map	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
248 #define	addr_PML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
249 #define	PTmap		((pt_entry_t *)(addr_PTmap))
250 #define	PDmap		((pd_entry_t *)(addr_PDmap))
251 #define	PDPmap		((pd_entry_t *)(addr_PDPmap))
252 #define	PML4map		((pd_entry_t *)(addr_PML4map))
253 #define	PML4pml4e	((pd_entry_t *)(addr_PML4pml4e))
254 
255 extern int nkpt;		/* Initial number of kernel page tables */
256 extern u_int64_t KPDPphys;	/* physical address of kernel level 3 */
257 extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */
258 
259 /*
260  * virtual address to page table entry and
261  * to physical address.
262  * Note: these work recursively, thus vtopte of a pte will give
263  * the corresponding pde that in turn maps it.
264  */
265 pt_entry_t *vtopte(vm_offset_t);
266 #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va)))
267 
268 #define	pte_load_store(ptep, pte)	atomic_swap_long(ptep, pte)
269 #define	pte_load_clear(ptep)		atomic_swap_long(ptep, 0)
270 #define	pte_store(ptep, pte) do { \
271 	*(u_long *)(ptep) = (u_long)(pte); \
272 } while (0)
273 #define	pte_clear(ptep)			pte_store(ptep, 0)
274 
275 #define	pde_store(pdep, pde)		pte_store(pdep, pde)
276 
277 extern pt_entry_t pg_nx;
278 
279 #endif /* _KERNEL */
280 
281 /*
282  * Pmap stuff
283  */
284 struct	pv_entry;
285 struct	pv_chunk;
286 
287 /*
288  * Locks
289  * (p) PV list lock
290  */
291 struct md_page {
292 	TAILQ_HEAD(, pv_entry)	pv_list;  /* (p) */
293 	int			pv_gen;   /* (p) */
294 	int			pat_mode;
295 };
296 
297 enum pmap_type {
298 	PT_X86,			/* regular x86 page tables */
299 	PT_EPT,			/* Intel's nested page tables */
300 	PT_RVI,			/* AMD's nested page tables */
301 };
302 
303 struct pmap_pcids {
304 	uint32_t	pm_pcid;
305 	uint32_t	pm_gen;
306 };
307 
308 /*
309  * The kernel virtual address (KVA) of the level 4 page table page is always
310  * within the direct map (DMAP) region.
311  */
312 struct pmap {
313 	struct mtx		pm_mtx;
314 	pml4_entry_t		*pm_pml4;	/* KVA of level 4 page table */
315 	uint64_t		pm_cr3;
316 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
317 	cpuset_t		pm_active;	/* active on cpus */
318 	enum pmap_type		pm_type;	/* regular or nested tables */
319 	struct pmap_statistics	pm_stats;	/* pmap statistics */
320 	struct vm_radix		pm_root;	/* spare page table pages */
321 	long			pm_eptgen;	/* EPT pmap generation id */
322 	int			pm_flags;
323 	struct pmap_pcids	pm_pcids[MAXCPU];
324 };
325 
326 /* flags */
327 #define	PMAP_NESTED_IPIMASK	0xff
328 #define	PMAP_PDE_SUPERPAGE	(1 << 8)	/* supports 2MB superpages */
329 #define	PMAP_EMULATE_AD_BITS	(1 << 9)	/* needs A/D bits emulation */
330 #define	PMAP_SUPPORTS_EXEC_ONLY	(1 << 10)	/* execute only mappings ok */
331 
332 typedef struct pmap	*pmap_t;
333 
334 #ifdef _KERNEL
335 extern struct pmap	kernel_pmap_store;
336 #define kernel_pmap	(&kernel_pmap_store)
337 
338 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
339 #define	PMAP_LOCK_ASSERT(pmap, type) \
340 				mtx_assert(&(pmap)->pm_mtx, (type))
341 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
342 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
343 				    NULL, MTX_DEF | MTX_DUPOK)
344 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
345 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
346 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
347 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
348 
349 int	pmap_pinit_type(pmap_t pmap, enum pmap_type pm_type, int flags);
350 int	pmap_emulate_accessed_dirty(pmap_t pmap, vm_offset_t va, int ftype);
351 #endif
352 
353 /*
354  * For each vm_page_t, there is a list of all currently valid virtual
355  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
356  */
357 typedef struct pv_entry {
358 	vm_offset_t	pv_va;		/* virtual address for mapping */
359 	TAILQ_ENTRY(pv_entry)	pv_next;
360 } *pv_entry_t;
361 
362 /*
363  * pv_entries are allocated in chunks per-process.  This avoids the
364  * need to track per-pmap assignments.
365  */
366 #define	_NPCM	3
367 #define	_NPCPV	168
368 struct pv_chunk {
369 	pmap_t			pc_pmap;
370 	TAILQ_ENTRY(pv_chunk)	pc_list;
371 	uint64_t		pc_map[_NPCM];	/* bitmap; 1 = free */
372 	TAILQ_ENTRY(pv_chunk)	pc_lru;
373 	struct pv_entry		pc_pventry[_NPCPV];
374 };
375 
376 #ifdef	_KERNEL
377 
378 extern caddr_t	CADDR1;
379 extern pt_entry_t *CMAP1;
380 extern vm_paddr_t phys_avail[];
381 extern vm_paddr_t dump_avail[];
382 extern vm_offset_t virtual_avail;
383 extern vm_offset_t virtual_end;
384 extern vm_paddr_t dmaplimit;
385 
386 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
387 #define	pmap_page_is_write_mapped(m)	(((m)->aflags & PGA_WRITEABLE) != 0)
388 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
389 
390 struct thread;
391 
392 void	pmap_activate_sw(struct thread *);
393 void	pmap_bootstrap(vm_paddr_t *);
394 int	pmap_cache_bits(pmap_t pmap, int mode, boolean_t is_pde);
395 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
396 void	pmap_demote_DMAP(vm_paddr_t base, vm_size_t len, boolean_t invalidate);
397 void	pmap_init_pat(void);
398 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
399 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
400 vm_paddr_t pmap_kextract(vm_offset_t);
401 void	pmap_kremove(vm_offset_t);
402 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
403 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
404 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
405 boolean_t pmap_page_is_mapped(vm_page_t m);
406 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
407 void	pmap_pinit_pml4(vm_page_t);
408 void	pmap_unmapdev(vm_offset_t, vm_size_t);
409 void	pmap_invalidate_page(pmap_t, vm_offset_t);
410 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
411 void	pmap_invalidate_all(pmap_t);
412 void	pmap_invalidate_cache(void);
413 void	pmap_invalidate_cache_pages(vm_page_t *pages, int count);
414 void	pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva,
415 	    boolean_t force);
416 void	pmap_get_mapping(pmap_t pmap, vm_offset_t va, uint64_t *ptr, int *num);
417 boolean_t pmap_map_io_transient(vm_page_t *, vm_offset_t *, int, boolean_t);
418 void	pmap_unmap_io_transient(vm_page_t *, vm_offset_t *, int, boolean_t);
419 #endif /* _KERNEL */
420 
421 /* Return various clipped indexes for a given VA */
422 static __inline vm_pindex_t
423 pmap_pte_index(vm_offset_t va)
424 {
425 
426 	return ((va >> PAGE_SHIFT) & ((1ul << NPTEPGSHIFT) - 1));
427 }
428 
429 static __inline vm_pindex_t
430 pmap_pde_index(vm_offset_t va)
431 {
432 
433 	return ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
434 }
435 
436 static __inline vm_pindex_t
437 pmap_pdpe_index(vm_offset_t va)
438 {
439 
440 	return ((va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1));
441 }
442 
443 static __inline vm_pindex_t
444 pmap_pml4e_index(vm_offset_t va)
445 {
446 
447 	return ((va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1));
448 }
449 
450 #endif /* !LOCORE */
451 
452 #endif /* !_MACHINE_PMAP_H_ */
453