kernel - Improve pmap hinting, improve performance
[dragonfly.git] / sys / platform / pc64 / include / pmap.h
blobd4ab89db98ed4bc6962da8dff4b1263ab97e7581
1 /*
2 * Copyright (c) 1991 Regents of the University of California.
3 * Copyright (c) 2003 Peter Wemm.
4 * Copyright (c) 2008 The DragonFly Project.
5 * All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * the Systems Programming Group of the University of Utah Computer
9 * Science Department and William Jolitz of UUNET Technologies Inc.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
35 * Derived from hp300 version by Mike Hibler, this version by William
36 * Jolitz uses a recursive map [a pde points to the page directory] to
37 * map the page tables using the pagetables themselves. This is done to
38 * reduce the impact on kernel virtual memory for lots of sparse address
39 * space, and to reduce the cost of memory to each process.
41 * from: hp300: @(#)pmap.h 7.2 (Berkeley) 12/16/90
42 * from: @(#)pmap.h 7.4 (Berkeley) 5/12/91
43 * $FreeBSD: src/sys/i386/include/pmap.h,v 1.65.2.3 2001/10/03 07:15:37 peter Exp $
46 #ifndef _MACHINE_PMAP_H_
47 #define _MACHINE_PMAP_H_
49 #include <cpu/pmap.h>
52 * Size of Kernel address space. This is the number of page table pages
53 * (2GB each) to use for the kernel. 256 pages == 512 Gigabytes.
54 * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
56 #ifndef KVA_PAGES
57 #define KVA_PAGES 256
58 #endif
61 * Pte related macros. This is complicated by having to deal with
62 * the sign extension of the 48th bit.
64 #define KVADDR(l4, l3, l2, l1) ( \
65 ((unsigned long)-1 << 47) | \
66 ((unsigned long)(l4) << PML4SHIFT) | \
67 ((unsigned long)(l3) << PDPSHIFT) | \
68 ((unsigned long)(l2) << PDRSHIFT) | \
69 ((unsigned long)(l1) << PAGE_SHIFT))
71 #define UVADDR(l4, l3, l2, l1) ( \
72 ((unsigned long)(l4) << PML4SHIFT) | \
73 ((unsigned long)(l3) << PDPSHIFT) | \
74 ((unsigned long)(l2) << PDRSHIFT) | \
75 ((unsigned long)(l1) << PAGE_SHIFT))
78 * NOTE: We no longer hardwire NKPT, it is calculated in create_pagetables()
80 #define NKPML4E 1 /* number of kernel PML4 slots */
81 /* NKPDPE defined in vmparam.h */
84 * NUPDPs 512 (256 user) number of PDPs in user page table
85 * NUPDs 512 * 512 number of PDs in user page table
86 * NUPTs 512 * 512 * 512 number of PTs in user page table
87 * NUPTEs 512 * 512 * 512 * 512 number of PTEs in user page table
89 * NUPDP_USER number of PDPs reserved for userland
90 * NUPTE_USER number of PTEs reserved for userland (big number)
92 #define NUPDP_USER (NPML4EPG/2)
93 #define NUPDP_TOTAL (NPML4EPG)
94 #define NUPD_TOTAL (NPDPEPG * NUPDP_TOTAL)
95 #define NUPT_TOTAL (NPDEPG * NUPD_TOTAL)
96 #define NUPTE_TOTAL ((vm_pindex_t)NPTEPG * NUPT_TOTAL)
97 #define NUPTE_USER ((vm_pindex_t)NPTEPG * NPDEPG * NPDPEPG * NUPDP_USER)
100 * Number of 512G dmap PML4 slots (max ~254 or so but don't go over 64,
101 * which gives us 32TB of ram). Because we cache free, empty pmaps the
102 * initialization overhead is minimal.
104 * It should be possible to bump this up to 255 (but not 256), which would
105 * be able to address a maximum of ~127TB of physical ram.
107 #define NDMPML4E 64
110 * The *PML4I values control the layout of virtual memory. Each PML4
111 * entry represents 512G.
113 #define PML4PML4I (NPML4EPG/2) /* Index of recursive pml4 mapping */
115 #define KPML4I (NPML4EPG-1) /* Top 512GB for KVM */
116 #define DMPML4I (KPML4I-NDMPML4E) /* Next 512GBxN down for dmap */
119 * The location of KERNBASE in the last PD of the kernel's KVM (KPML4I)
120 * space. Each PD represents 1GB. The kernel must be placed here
121 * for the compile/link options to work properly so absolute 32-bit
122 * addressing can be used to access stuff.
124 #define KPDPI (NPDPEPG-2) /* kernbase at -2GB */
127 * per-CPU data assume ~64K x SMP_MAXCPU, say up to 256 cpus
128 * in the future or 16MB of space. Each PD represents 2MB so
129 * use NPDEPG-8 to place the per-CPU data.
131 #define MPPML4I KPML4I
132 #define MPPDPI KPDPI
133 #define MPPTDI (NPDEPG-8)
136 * XXX doesn't really belong here I guess...
138 #define ISA_HOLE_START 0xa0000
139 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
141 #ifndef LOCORE
143 #ifndef _SYS_TYPES_H_
144 #include <sys/types.h>
145 #endif
146 #ifndef _SYS_QUEUE_H_
147 #include <sys/queue.h>
148 #endif
149 #ifndef _SYS_TREE_H_
150 #include <sys/tree.h>
151 #endif
152 #ifndef _SYS_SPINLOCK_H_
153 #include <sys/spinlock.h>
154 #endif
155 #ifndef _SYS_THREAD_H_
156 #include <sys/thread.h>
157 #endif
158 #ifndef _MACHINE_TYPES_H_
159 #include <machine/types.h>
160 #endif
161 #ifndef _MACHINE_PARAM_H_
162 #include <machine/param.h>
163 #endif
166 * Address of current and alternate address space page table maps
167 * and directories.
169 #ifdef _KERNEL
170 #define addr_PTmap (KVADDR(PML4PML4I, 0, 0, 0))
171 #define addr_PDmap (KVADDR(PML4PML4I, PML4PML4I, 0, 0))
172 #define addr_PDPmap (KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
173 #define addr_PML4map (KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
174 #define addr_PML4pml4e (addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
175 #define PTmap ((pt_entry_t *)(addr_PTmap))
176 #define PDmap ((pd_entry_t *)(addr_PDmap))
177 #define PDPmap ((pd_entry_t *)(addr_PDPmap))
178 #define PML4map ((pd_entry_t *)(addr_PML4map))
179 #define PML4pml4e ((pd_entry_t *)(addr_PML4pml4e))
181 extern u_int64_t KPML4phys; /* physical address of kernel level 4 */
182 extern int pmap_fast_kernel_cpusync;
183 #endif
185 #ifdef _KERNEL
188 * XXX
190 #define vtophys(va) pmap_kextract(((vm_offset_t)(va)))
191 #define vtophys_pte(va) ((pt_entry_t)pmap_kextract(((vm_offset_t)(va))))
193 #endif
195 #define pte_load_clear(pte) atomic_readandclear_long(pte)
197 static __inline void
198 pte_store(pt_entry_t *ptep, pt_entry_t pte)
200 *ptep = pte;
203 #define pde_store(pdep, pde) pte_store((pdep), (pde))
206 * Pmap stuff
208 struct pmap;
209 struct pv_entry;
210 struct vm_page;
211 struct vm_object;
212 struct vmspace;
214 TAILQ_HEAD(md_page_pv_list, pv_entry);
216 * vm_page structures embed a list of related pv_entry's
218 struct md_page {
219 struct md_page_pv_list pv_list;
223 * vm_object's representing large mappings can contain embedded pmaps
224 * to organize sharing at higher page table levels for PROT_READ and
225 * PROT_READ|PROT_WRITE maps.
227 struct md_object {
228 struct pmap *pmap_rw;
229 struct pmap *pmap_ro;
233 * Each machine dependent implementation is expected to
234 * keep certain statistics. They may do this anyway they
235 * so choose, but are expected to return the statistics
236 * in the following structure.
238 * NOTE: We try to match the size of the pc32 pmap with the vkernel pmap
239 * so the same utilities (like 'ps') can be used on both.
241 struct pmap_statistics {
242 long resident_count; /* # of pages mapped (total) */
243 long wired_count; /* # of pages wired */
245 typedef struct pmap_statistics *pmap_statistics_t;
247 struct pv_entry_rb_tree;
248 RB_PROTOTYPE2(pv_entry_rb_tree, pv_entry, pv_entry,
249 pv_entry_compare, vm_pindex_t);
251 /* Types of PMAP (regular, EPT Intel, NPT Amd) */
252 #define REGULAR_PMAP 0
253 #define EPT_PMAP 1
255 /* Bits indexes in pmap_bits */
256 #define TYPE_IDX 0
257 #define PG_V_IDX 1
258 #define PG_RW_IDX 2
259 #define PG_U_IDX 3
260 #define PG_A_IDX 4
261 #define PG_M_IDX 5
262 #define PG_PS_IDX 6
263 #define PG_G_IDX 7
264 #define PG_W_IDX 8
265 #define PG_MANAGED_IDX 9
266 #define PG_DEVICE_IDX 10
267 #define PG_N_IDX 11
268 #define PG_NX_IDX 12
269 #define PG_BITS_SIZE 13
271 #define PROTECTION_CODES_SIZE 8
272 #define PAT_INDEX_SIZE 8
274 #define PM_PLACEMARKS 64 /* 16 @ 4 zones */
275 #define PM_NOPLACEMARK ((vm_pindex_t)-1)
276 #define PM_PLACEMARK_WAKEUP ((vm_pindex_t)0x8000000000000000LLU)
278 struct pmap {
279 pml4_entry_t *pm_pml4; /* KVA of level 4 page table */
280 struct pv_entry *pm_pmlpv; /* PV entry for pml4 */
281 TAILQ_ENTRY(pmap) pm_pmnode; /* list of pmaps */
282 RB_HEAD(pv_entry_rb_tree, pv_entry) pm_pvroot;
283 int pm_count; /* reference count */
284 cpulock_t pm_active_lock; /* interlock */
285 cpumask_t pm_active; /* active on cpus */
286 int pm_flags;
287 struct pmap_statistics pm_stats; /* pmap statistics */
288 struct spinlock pm_spin;
289 struct pv_entry *pm_pvhint_pt; /* pv_entry lookup hint */
290 struct pv_entry *pm_pvhint_pte; /* pv_entry lookup hint */
291 vm_pindex_t pm_placemarks[PM_PLACEMARKS];
292 long pm_invgen;
293 uint64_t pmap_bits[PG_BITS_SIZE];
294 uint64_t protection_codes[PROTECTION_CODES_SIZE];
295 pt_entry_t pmap_cache_bits[PAT_INDEX_SIZE];
296 pt_entry_t pmap_cache_mask;
297 int (*copyinstr)(const void *, void *, size_t, size_t *);
298 int (*copyin)(const void *, void *, size_t);
299 int (*copyout)(const void *, void *, size_t);
300 int (*fubyte)(const uint8_t *); /* returns int for -1 err */
301 int (*subyte)(uint8_t *, uint8_t);
302 int32_t (*fuword32)(const uint32_t *);
303 int64_t (*fuword64)(const uint64_t *);
304 int (*suword64)(uint64_t *, uint64_t);
305 int (*suword32)(uint32_t *, int);
306 uint32_t (*swapu32)(volatile uint32_t *, uint32_t v);
307 uint64_t (*swapu64)(volatile uint64_t *, uint64_t v);
310 #define PMAP_FLAG_SIMPLE 0x00000001
311 #define PMAP_EMULATE_AD_BITS 0x00000002
312 #define PMAP_HVM 0x00000004
313 #define PMAP_SEGSHARED 0x00000008 /* segment shared opt */
315 #define pmap_resident_count(pmap) ((pmap)->pm_stats.resident_count)
316 #define pmap_resident_tlnw_count(pmap) ((pmap)->pm_stats.resident_count - \
317 (pmap)->pm_stats.wired_count)
319 typedef struct pmap *pmap_t;
321 #ifdef _KERNEL
322 extern struct pmap kernel_pmap;
323 #endif
326 * For each vm_page_t, there is a list of all currently valid virtual
327 * mappings of that page. An entry is a pv_entry_t, the list is pv_table.
329 typedef struct pv_entry {
330 pmap_t pv_pmap; /* pmap where mapping lies */
331 vm_pindex_t pv_pindex; /* PTE, PT, PD, PDP, or PML4 */
332 TAILQ_ENTRY(pv_entry) pv_list;
333 RB_ENTRY(pv_entry) pv_entry;
334 struct vm_page *pv_m; /* page being mapped */
335 u_int pv_hold; /* interlock action */
336 u_int pv_flags;
337 #ifdef PMAP_DEBUG
338 const char *pv_func;
339 int pv_line;
340 const char *pv_func_lastfree;
341 int pv_line_lastfree;
342 #endif
343 } *pv_entry_t;
345 #define PV_HOLD_LOCKED 0x80000000U
346 #define PV_HOLD_WAITING 0x40000000U
347 #define PV_HOLD_UNUSED2000 0x20000000U
348 #define PV_HOLD_MASK 0x1FFFFFFFU
350 #define PV_FLAG_VMOBJECT 0x00000001U /* shared pt in VM obj */
352 #ifdef _KERNEL
354 extern caddr_t CADDR1;
355 extern pt_entry_t *CMAP1;
356 extern vm_paddr_t avail_end;
357 extern vm_paddr_t avail_start;
358 extern vm_offset_t clean_eva;
359 extern vm_offset_t clean_sva;
360 extern char *ptvmmap; /* poor name! */
362 #ifndef __VM_PAGE_T_DEFINED__
363 #define __VM_PAGE_T_DEFINED__
364 typedef struct vm_page *vm_page_t;
365 #endif
366 #ifndef __VM_MEMATTR_T_DEFINED__
367 #define __VM_MEMATTR_T_DEFINED__
368 typedef char vm_memattr_t;
369 #endif
371 void pmap_release(struct pmap *pmap);
372 void pmap_interlock_wait (struct vmspace *);
373 void pmap_bootstrap (vm_paddr_t *);
374 void *pmap_mapbios(vm_paddr_t, vm_size_t);
375 void *pmap_mapdev (vm_paddr_t, vm_size_t);
376 void *pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
377 void *pmap_mapdev_uncacheable(vm_paddr_t, vm_size_t);
378 void pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
379 void pmap_unmapdev (vm_offset_t, vm_size_t);
380 struct vm_page *pmap_use_pt (pmap_t, vm_offset_t);
381 void pmap_set_opt (void);
382 void pmap_init_pat(void);
383 void pmap_invalidate_cache_pages(vm_page_t *pages, int count);
384 void pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva);
386 static __inline int
387 pmap_emulate_ad_bits(pmap_t pmap) {
388 return pmap->pm_flags & PMAP_EMULATE_AD_BITS;
391 /* Return various clipped indexes for a given VA */
394 * Returns the index of a PTE in a PT, representing a terminal
395 * page.
397 static __inline vm_pindex_t
398 pmap_pte_index(vm_offset_t va)
400 return ((va >> PAGE_SHIFT) & ((1ul << NPTEPGSHIFT) - 1));
404 * Returns the index of a PT in a PD
406 static __inline vm_pindex_t
407 pmap_pde_index(vm_offset_t va)
409 return ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
413 * Returns the index of a PD in a PDP
415 static __inline vm_pindex_t
416 pmap_pdpe_index(vm_offset_t va)
418 return ((va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1));
422 * Returns the index of a PDP in the PML4
424 static __inline vm_pindex_t
425 pmap_pml4e_index(vm_offset_t va)
427 return ((va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1));
430 #endif /* _KERNEL */
432 #endif /* !LOCORE */
434 #endif /* !_MACHINE_PMAP_H_ */