| 1 | /*- |
| 2 | * SPDX-License-Identifier: BSD-3-Clause |
| 3 | * |
| 4 | * Copyright (c) 2003 Peter Wemm. |
| 5 | * Copyright (c) 1991 Regents of the University of California. |
| 6 | * All rights reserved. |
| 7 | * |
| 8 | * This code is derived from software contributed to Berkeley by |
| 9 | * the Systems Programming Group of the University of Utah Computer |
| 10 | * Science Department and William Jolitz of UUNET Technologies Inc. |
| 11 | * |
| 12 | * Redistribution and use in source and binary forms, with or without |
| 13 | * modification, are permitted provided that the following conditions |
| 14 | * are met: |
| 15 | * 1. Redistributions of source code must retain the above copyright |
| 16 | * notice, this list of conditions and the following disclaimer. |
| 17 | * 2. Redistributions in binary form must reproduce the above copyright |
| 18 | * notice, this list of conditions and the following disclaimer in the |
| 19 | * documentation and/or other materials provided with the distribution. |
| 20 | * 3. Neither the name of the University nor the names of its contributors |
| 21 | * may be used to endorse or promote products derived from this software |
| 22 | * without specific prior written permission. |
| 23 | * |
| 24 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 25 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 26 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 27 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 28 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 29 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 30 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 31 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 32 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 33 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 34 | * SUCH DAMAGE. |
| 35 | * |
| 36 | * Derived from hp300 version by Mike Hibler, this version by William |
| 37 | * Jolitz uses a recursive map [a pde points to the page directory] to |
| 38 | * map the page tables using the pagetables themselves. This is done to |
| 39 | * reduce the impact on kernel virtual memory for lots of sparse address |
| 40 | * space, and to reduce the cost of memory to each process. |
| 41 | */ |
| 42 | |
| 43 | #ifdef __i386__ |
| 44 | #include <i386/pmap.h> |
| 45 | #else /* !__i386__ */ |
| 46 | |
| 47 | #ifndef _MACHINE_PMAP_H_ |
| 48 | #define	_MACHINE_PMAP_H_ |
| 49 | |
| 50 | #include <machine/pte.h> |
| 51 | |
| 52 | /* |
| 53 | * Define the PG_xx macros in terms of the bits on x86 PTEs. |
| 54 | */ |
| 55 | #define	PG_V		X86_PG_V |
| 56 | #define	PG_RW		X86_PG_RW |
| 57 | #define	PG_U		X86_PG_U |
| 58 | #define	PG_NC_PWT	X86_PG_NC_PWT |
| 59 | #define	PG_NC_PCD	X86_PG_NC_PCD |
| 60 | #define	PG_A		X86_PG_A |
| 61 | #define	PG_M		X86_PG_M |
| 62 | #define	PG_PS		X86_PG_PS |
| 63 | #define	PG_PTE_PAT	X86_PG_PTE_PAT |
| 64 | #define	PG_G		X86_PG_G |
| 65 | #define	PG_AVAIL1	X86_PG_AVAIL1 |
| 66 | #define	PG_AVAIL2	X86_PG_AVAIL2 |
| 67 | #define	PG_AVAIL3	X86_PG_AVAIL3 |
| 68 | #define	PG_PDE_PAT	X86_PG_PDE_PAT |
| 69 | #define	PG_NX		X86_PG_NX |
| 70 | #define	PG_PDE_CACHE	X86_PG_PDE_CACHE |
| 71 | #define	PG_PTE_CACHE	X86_PG_PTE_CACHE |
| 72 | |
| 73 | /* Our various interpretations of the above */ |
| 74 | #define	PG_W		X86_PG_AVAIL3	/* "Wired" pseudoflag */ |
| 75 | #define	PG_MANAGED	X86_PG_AVAIL2 |
| 76 | #define	EPT_PG_EMUL_V	X86_PG_AVAIL(52) |
| 77 | #define	EPT_PG_EMUL_RW	X86_PG_AVAIL(53) |
| 78 | #define	PG_PROMOTED	X86_PG_AVAIL(54)	/* PDE only */ |
| 79 | |
| 80 | /* |
| 81 | * Promotion to a 2MB (PDE) page mapping requires that the corresponding 4KB |
| 82 | * (PTE) page mappings have identical settings for the following fields: |
| 83 | */ |
| 84 | #define	PG_PTE_PROMOTE	(PG_NX | PG_MANAGED | PG_W | PG_G | PG_PTE_CACHE | \ |
| 85 | 	 PG_M | PG_U | PG_RW | PG_V | PG_PKU_MASK) |
| 86 | |
| 87 | /* |
| 88 | * undef the PG_xx macros that define bits in the regular x86 PTEs that |
| 89 | * have a different position in nested PTEs. This is done when compiling |
| 90 | * code that needs to be aware of the differences between regular x86 and |
| 91 | * nested PTEs. |
| 92 | * |
| 93 | * The appropriate bitmask will be calculated at runtime based on the pmap |
| 94 | * type. |
| 95 | */ |
| 96 | #ifdef AMD64_NPT_AWARE |
| 97 | #undef PG_AVAIL1		/* X86_PG_AVAIL1 aliases with EPT_PG_M */ |
| 98 | #undef PG_G |
| 99 | #undef PG_A |
| 100 | #undef PG_M |
| 101 | #undef PG_PDE_PAT |
| 102 | #undef PG_PDE_CACHE |
| 103 | #undef PG_PTE_PAT |
| 104 | #undef PG_PTE_CACHE |
| 105 | #undef PG_RW |
| 106 | #undef PG_V |
| 107 | #endif |
| 108 | |
| 109 | /* |
| 110 | * Pte related macros. This is complicated by having to deal with |
| 111 | * the sign extension of the 48th bit. |
| 112 | */ |
| 113 | #define KV4ADDR(l4, l3, l2, l1)		KV5ADDR(-1, l4, l3, l2, l1) |
| 114 | #define KV5ADDR(l5, l4, l3, l2, l1) (		\ |
| 115 | 	((unsigned long)-1 << 56) | \ |
| 116 | 	((unsigned long)(l5) << PML5SHIFT) | \ |
| 117 | 	((unsigned long)(l4) << PML4SHIFT) | \ |
| 118 | 	((unsigned long)(l3) << PDPSHIFT) | \ |
| 119 | 	((unsigned long)(l2) << PDRSHIFT) | \ |
| 120 | 	((unsigned long)(l1) << PAGE_SHIFT)) |
| 121 | |
| 122 | #define UVADDR(l5, l4, l3, l2, l1) (	 \ |
| 123 | 	((unsigned long)(l5) << PML5SHIFT) | \ |
| 124 | 	((unsigned long)(l4) << PML4SHIFT) | \ |
| 125 | 	((unsigned long)(l3) << PDPSHIFT) | \ |
| 126 | 	((unsigned long)(l2) << PDRSHIFT) | \ |
| 127 | 	((unsigned long)(l1) << PAGE_SHIFT)) |
| 128 | |
| 129 | /* |
| 130 | * Number of kernel PML4 slots. Can be anywhere from 1 to 64 or so, |
| 131 | * but setting it larger than NDMPML4E makes no sense. |
| 132 | * |
| 133 | * Each slot provides .5 TB of kernel virtual space. |
| 134 | */ |
| 135 | #define NKPML4E		4 |
| 136 | |
| 137 | /* |
| 138 | * Number of PML4 slots for the KASAN shadow map. It requires 1 byte of memory |
| 139 | * for every 8 bytes of the kernel address space. |
| 140 | */ |
| 141 | #define	NKASANPML4E	((NKPML4E + 7) / 8) |
| 142 | |
| 143 | /* |
| 144 | * Number of PML4 slots for the KMSAN shadow and origin maps. These are |
| 145 | * one-to-one with the kernel map. |
| 146 | */ |
| 147 | #define	NKMSANSHADPML4E	NKPML4E |
| 148 | #define	NKMSANORIGPML4E	NKPML4E |
| 149 | |
| 150 | /* |
| 151 | * We use the same numbering of the page table pages for 5-level and |
| 152 | * 4-level paging structures. |
| 153 | */ |
| 154 | #define	NUPML5E		(NPML5EPG / 2)		/* number of userland PML5 |
| 155 | 						 pages */ |
| 156 | #define	NUPML4E		(NUPML5E * NPML4EPG)	/* number of userland PML4 |
| 157 | 						 pages */ |
| 158 | #define	NUPDPE		(NUPML4E * NPDPEPG)	/* number of userland PDP |
| 159 | 						 pages */ |
| 160 | #define	NUPDE		(NUPDPE * NPDEPG)	/* number of userland PD |
| 161 | 						 entries */ |
| 162 | #define	NUP4ML4E	(NPML4EPG / 2) |
| 163 | |
| 164 | /* |
| 165 | * NDMPML4E is the maximum number of PML4 entries that will be |
| 166 | * used to implement the direct map. It must be a power of two, |
| 167 | * and should generally exceed NKPML4E. The maximum possible |
| 168 | * value is 64; using 128 will make the direct map intrude into |
| 169 | * the recursive page table map. |
| 170 | */ |
| 171 | #define	NDMPML4E	8 |
| 172 | #define	NDMPML5E	32 |
| 173 | |
| 174 | /* |
| 175 | * These values control the layout of virtual memory. The starting |
| 176 | * address of the direct map is controlled by DMPML4I on LA48 and |
| 177 | * DMPML5I on LA57. |
| 178 | * |
| 179 | * Note: KPML4I is the index of the (single) level 4 page that maps |
| 180 | * the KVA that holds KERNBASE, while KPML4BASE is the index of the |
| 181 | * first level 4 page that maps VM_MIN_KERNEL_ADDRESS. If NKPML4E |
| 182 | * is 1, these are the same, otherwise KPML4BASE < KPML4I and extra |
| 183 | * level 4 PDEs are needed to map from VM_MIN_KERNEL_ADDRESS up to |
| 184 | * KERNBASE. |
| 185 | * |
| 186 | * (KPML4I combines with KPDPI to choose where KERNBASE starts. |
| 187 | * Or, in other words, KPML4I provides bits 39..47 of KERNBASE, |
| 188 | * and KPDPI provides bits 30..38.) |
| 189 | */ |
| 190 | #define	PML4PML4I	(NPML4EPG / 2)	/* Index of recursive pml4 mapping */ |
| 191 | #define	PML5PML5I	(NPML5EPG / 2)	/* Index of recursive pml5 mapping */ |
| 192 | |
| 193 | #define	KPML4BASE	(NPML4EPG-NKPML4E) /* KVM at highest addresses */ |
| 194 | #define	DMPML4I		rounddown(KPML4BASE-NDMPML4E, NDMPML4E) /* Below KVM */ |
| 195 | #define	DMPML5I		(NPML5EPG / 2 + 1) |
| 196 | |
| 197 | #define	KPML4I		(NPML4EPG-1) |
| 198 | #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */ |
| 199 | |
| 200 | #define	KASANPML4I	(DMPML4I - NKASANPML4E) /* Below the direct map */ |
| 201 | |
| 202 | #define	KMSANSHADPML4I	(KPML4BASE - NKMSANSHADPML4E) |
| 203 | #define	KMSANORIGPML4I	(DMPML4I - NKMSANORIGPML4E) |
| 204 | |
| 205 | /* |
| 206 | * Large map: index of the first and max last pml4/la48 and pml5/la57 |
| 207 | * entry. |
| 208 | */ |
| 209 | #define	LMSPML4I	(PML4PML4I + 1) |
| 210 | #define	LMEPML4I	(KASANPML4I - 1) |
| 211 | #define	LMSPML5I	(DMPML5I + NDMPML5E) |
| 212 | #define	LMEPML5I	(LMSPML5I + 32 - 1)	/* 32 slots for large map */ |
| 213 | |
| 214 | /* |
| 215 | * XXX doesn't really belong here I guess... |
| 216 | */ |
| 217 | #define ISA_HOLE_START 0xa0000 |
| 218 | #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START) |
| 219 | |
| 220 | #define	PMAP_PCID_NONE		0xffffffff |
| 221 | #define	PMAP_PCID_KERN		0 |
| 222 | #define	PMAP_PCID_OVERMAX	0x1000 |
| 223 | #define	PMAP_PCID_OVERMAX_KERN	0x800 |
| 224 | #define	PMAP_PCID_USER_PT	0x800 |
| 225 | |
| 226 | #define	PMAP_NO_CR3		0xffffffffffffffff |
| 227 | #define	PMAP_UCR3_NOMASK	0xffffffffffffffff |
| 228 | |
| 229 | #ifndef LOCORE |
| 230 | |
| 231 | #include <sys/kassert.h> |
| 232 | #include <sys/queue.h> |
| 233 | #include <sys/_cpuset.h> |
| 234 | #include <sys/_lock.h> |
| 235 | #include <sys/_mutex.h> |
| 236 | #include <sys/_pctrie.h> |
| 237 | #include <machine/_pmap.h> |
| 238 | #include <sys/_pv_entry.h> |
| 239 | #include <sys/_rangeset.h> |
| 240 | #include <sys/_smr.h> |
| 241 | |
| 242 | #include <vm/_vm_radix.h> |
| 243 | |
| 244 | typedef u_int64_t pd_entry_t; |
| 245 | typedef u_int64_t pt_entry_t; |
| 246 | typedef u_int64_t pdp_entry_t; |
| 247 | typedef u_int64_t pml4_entry_t; |
| 248 | typedef u_int64_t pml5_entry_t; |
| 249 | |
| 250 | /* |
| 251 | * Address of current address space page table maps and directories. |
| 252 | */ |
| 253 | #ifdef _KERNEL |
| 254 | #define	addr_P4Tmap	(KV4ADDR(PML4PML4I, 0, 0, 0)) |
| 255 | #define	addr_P4Dmap	(KV4ADDR(PML4PML4I, PML4PML4I, 0, 0)) |
| 256 | #define	addr_P4DPmap	(KV4ADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0)) |
| 257 | #define	addr_P4ML4map	(KV4ADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I)) |
| 258 | #define	addr_P4ML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t))) |
| 259 | #define	P4Tmap		((pt_entry_t *)(addr_P4Tmap)) |
| 260 | #define	P4Dmap		((pd_entry_t *)(addr_P4Dmap)) |
| 261 | |
| 262 | #define	addr_P5Tmap	(KV5ADDR(PML5PML5I, 0, 0, 0, 0)) |
| 263 | #define	addr_P5Dmap	(KV5ADDR(PML5PML5I, PML5PML5I, 0, 0, 0)) |
| 264 | #define	addr_P5DPmap	(KV5ADDR(PML5PML5I, PML5PML5I, PML5PML5I, 0, 0)) |
| 265 | #define	addr_P5ML4map	(KV5ADDR(PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I, 0)) |
| 266 | #define	addr_P5ML5map	\ |
| 267 | (KVADDR(PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I)) |
| 268 | #define	addr_P5ML5pml5e	(addr_P5ML5map + (PML5PML5I * sizeof(pml5_entry_t))) |
| 269 | #define	P5Tmap		((pt_entry_t *)(addr_P5Tmap)) |
| 270 | #define	P5Dmap		((pd_entry_t *)(addr_P5Dmap)) |
| 271 | |
| 272 | extern int nkpt;		/* Initial number of kernel page tables */ |
| 273 | extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */ |
| 274 | extern u_int64_t KPML5phys;	/* physical address of kernel level 5 */ |
| 275 | |
| 276 | /* |
| 277 | * virtual address to page table entry and |
| 278 | * to physical address. |
| 279 | * Note: these work recursively, thus vtopte of a pte will give |
| 280 | * the corresponding pde that in turn maps it. |
| 281 | */ |
| 282 | pt_entry_t *vtopte(vm_offset_t); |
| 283 | #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va))) |
| 284 | |
| 285 | #define	pte_load_store(ptep, pte)	atomic_swap_long(ptep, pte) |
| 286 | #define	pte_load_clear(ptep)		atomic_swap_long(ptep, 0) |
| 287 | #define	pte_store(ptep, pte) do { \ |
| 288 | 	*(u_long *)(ptep) = (u_long)(pte); \ |
| 289 | } while (0) |
| 290 | #define	pte_clear(ptep)			pte_store(ptep, 0) |
| 291 | |
| 292 | #define	pde_store(pdep, pde)		pte_store(pdep, pde) |
| 293 | |
| 294 | extern pt_entry_t pg_nx; |
| 295 | |
| 296 | #endif /* _KERNEL */ |
| 297 | |
| 298 | /* |
| 299 | * Pmap stuff |
| 300 | */ |
| 301 | |
| 302 | /* |
| 303 | * Locks |
| 304 | * (p) PV list lock |
| 305 | */ |
| 306 | struct md_page { |
| 307 | 	TAILQ_HEAD(, pv_entry)	pv_list; /* (p) */ |
| 308 | 	int			pv_gen; /* (p) */ |
| 309 | 	int			pat_mode; |
| 310 | }; |
| 311 | |
| 312 | enum pmap_type { |
| 313 | 	PT_X86,			/* regular x86 page tables */ |
| 314 | 	PT_EPT,			/* Intel's nested page tables */ |
| 315 | 	PT_RVI,			/* AMD's nested page tables */ |
| 316 | }; |
| 317 | |
| 318 | /* |
| 319 | * The kernel virtual address (KVA) of the level 4 page table page is always |
| 320 | * within the direct map (DMAP) region. |
| 321 | */ |
| 322 | struct pmap { |
| 323 | 	struct mtx		pm_mtx; |
| 324 | 	pml4_entry_t		*pm_pmltop;	/* KVA of top level page table */ |
| 325 | 	pml4_entry_t		*pm_pmltopu;	/* KVA of user top page table */ |
| 326 | 	uint64_t		pm_cr3; |
| 327 | 	uint64_t		pm_ucr3; |
| 328 | 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */ |
| 329 | 	cpuset_t		pm_active;	/* active on cpus */ |
| 330 | 	enum pmap_type		pm_type;	/* regular or nested tables */ |
| 331 | 	struct pmap_statistics	pm_stats;	/* pmap statistics */ |
| 332 | 	struct vm_radix		pm_root;	/* spare page table pages */ |
| 333 | 	long			pm_eptgen;	/* EPT pmap generation id */ |
| 334 | 	smr_t			pm_eptsmr; |
| 335 | 	int			pm_flags; |
| 336 | 	struct pmap_pcid	*pm_pcidp; |
| 337 | 	struct rangeset		pm_pkru; |
| 338 | }; |
| 339 | |
| 340 | /* flags */ |
| 341 | #define	PMAP_NESTED_IPIMASK	0xff |
| 342 | #define	PMAP_PDE_SUPERPAGE	(1 << 8)	/* supports 2MB superpages */ |
| 343 | #define	PMAP_EMULATE_AD_BITS	(1 << 9)	/* needs A/D bits emulation */ |
| 344 | #define	PMAP_SUPPORTS_EXEC_ONLY	(1 << 10)	/* execute only mappings ok */ |
| 345 | |
| 346 | typedef struct pmap	*pmap_t; |
| 347 | |
| 348 | #ifdef _KERNEL |
| 349 | extern struct pmap	kernel_pmap_store; |
| 350 | #define kernel_pmap	(&kernel_pmap_store) |
| 351 | |
| 352 | #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx) |
| 353 | #define	PMAP_LOCK_ASSERT(pmap, type) \ |
| 354 | 				mtx_assert(&(pmap)->pm_mtx, (type)) |
| 355 | #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx) |
| 356 | #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \ |
| 357 | 				 NULL, MTX_DEF | MTX_DUPOK) |
| 358 | #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx) |
| 359 | #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx) |
| 360 | #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx) |
| 361 | #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx) |
| 362 | |
| 363 | int	pmap_pinit_type(pmap_t pmap, enum pmap_type pm_type, int flags); |
| 364 | int	pmap_emulate_accessed_dirty(pmap_t pmap, vm_offset_t va, int ftype); |
| 365 | |
| 366 | extern caddr_t	CADDR1; |
| 367 | extern pt_entry_t *CMAP1; |
| 368 | extern vm_offset_t virtual_avail; |
| 369 | extern vm_offset_t virtual_end; |
| 370 | extern vm_paddr_t dmaplimit; |
| 371 | extern int pmap_pcid_enabled; |
| 372 | extern int invpcid_works; |
| 373 | extern int invlpgb_works; |
| 374 | extern int invlpgb_maxcnt; |
| 375 | extern int pmap_pcid_invlpg_workaround; |
| 376 | extern int pmap_pcid_invlpg_workaround_uena; |
| 377 | |
| 378 | #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode) |
| 379 | #define	pmap_page_is_write_mapped(m)	(((m)->a.flags & PGA_WRITEABLE) != 0) |
| 380 | #define	pmap_unmapbios(va, sz)		pmap_unmapdev((va), (sz)) |
| 381 | |
| 382 | #define	pmap_vm_page_alloc_check(m)					\ |
| 383 | 	KASSERT(m->phys_addr < kernphys ||				\ |
| 384 | 	 m->phys_addr >= kernphys + (vm_offset_t)&_end - KERNSTART,	\ |
| 385 | 	 ("allocating kernel page %p pa %#lx kernphys %#lx end %p", \ |
| 386 | 	 m, m->phys_addr, kernphys, &_end)); |
| 387 | |
| 388 | struct thread; |
| 389 | |
| 390 | void	pmap_activate_boot(pmap_t pmap); |
| 391 | void	pmap_activate_sw(struct thread *); |
| 392 | void	pmap_allow_2m_x_ept_recalculate(void); |
| 393 | void	pmap_bootstrap(vm_paddr_t *); |
| 394 | int	pmap_cache_bits(pmap_t pmap, int mode, bool is_pde); |
| 395 | int	pmap_change_attr(vm_offset_t, vm_size_t, int); |
| 396 | int	pmap_change_prot(vm_offset_t, vm_size_t, vm_prot_t); |
| 397 | void	pmap_demote_DMAP(vm_paddr_t base, vm_size_t len, bool invalidate); |
| 398 | void	pmap_flush_cache_range(vm_offset_t, vm_offset_t); |
| 399 | void	pmap_flush_cache_phys_range(vm_paddr_t, vm_paddr_t, vm_memattr_t); |
| 400 | void	pmap_init_pat(void); |
| 401 | void	pmap_kenter(vm_offset_t va, vm_paddr_t pa); |
| 402 | void	*pmap_kenter_temporary(vm_paddr_t pa, int i); |
| 403 | vm_paddr_t pmap_kextract(vm_offset_t); |
| 404 | void	pmap_kremove(vm_offset_t); |
| 405 | int	pmap_large_map(vm_paddr_t, vm_size_t, void **, vm_memattr_t); |
| 406 | void	pmap_large_map_wb(void *sva, vm_size_t len); |
| 407 | void	pmap_large_unmap(void *sva, vm_size_t len); |
| 408 | void	*pmap_mapbios(vm_paddr_t, vm_size_t); |
| 409 | void	*pmap_mapdev(vm_paddr_t, vm_size_t); |
| 410 | void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int); |
| 411 | void	*pmap_mapdev_pciecfg(vm_paddr_t pa, vm_size_t size); |
| 412 | bool	pmap_not_in_di(void); |
| 413 | bool	pmap_page_is_mapped(vm_page_t m); |
| 414 | void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma); |
| 415 | void	pmap_page_set_memattr_noflush(vm_page_t m, vm_memattr_t ma); |
| 416 | void	pmap_pinit_pml4(vm_page_t); |
| 417 | void	pmap_pinit_pml5(vm_page_t); |
| 418 | bool	pmap_ps_enabled(pmap_t pmap); |
| 419 | void	pmap_unmapdev(void *, vm_size_t); |
| 420 | void	pmap_invalidate_page(pmap_t, vm_offset_t); |
| 421 | void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t); |
| 422 | void	pmap_invalidate_all(pmap_t); |
| 423 | void	pmap_invalidate_cache(void); |
| 424 | void	pmap_invalidate_cache_pages(vm_page_t *pages, int count); |
| 425 | void	pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva); |
| 426 | void	pmap_force_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva); |
| 427 | void	pmap_get_mapping(pmap_t pmap, vm_offset_t va, uint64_t *ptr, int *num); |
| 428 | bool	pmap_map_io_transient(vm_page_t *, vm_offset_t *, int, bool); |
| 429 | void	pmap_unmap_io_transient(vm_page_t *, vm_offset_t *, int, bool); |
| 430 | void	pmap_map_delete(pmap_t, vm_offset_t, vm_offset_t); |
| 431 | void	pmap_pti_add_kva(vm_offset_t sva, vm_offset_t eva, bool exec); |
| 432 | void	pmap_pti_remove_kva(vm_offset_t sva, vm_offset_t eva); |
| 433 | void	pmap_pti_pcid_invalidate(uint64_t ucr3, uint64_t kcr3); |
| 434 | void	pmap_pti_pcid_invlpg(uint64_t ucr3, uint64_t kcr3, vm_offset_t va); |
| 435 | void	pmap_pti_pcid_invlrng(uint64_t ucr3, uint64_t kcr3, vm_offset_t sva, |
| 436 | 	 vm_offset_t eva); |
| 437 | int	pmap_pkru_clear(pmap_t pmap, vm_offset_t sva, vm_offset_t eva); |
| 438 | int	pmap_pkru_set(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, |
| 439 | 	 u_int keyidx, int flags); |
| 440 | void	pmap_thread_init_invl_gen(struct thread *td); |
| 441 | int	pmap_vmspace_copy(pmap_t dst_pmap, pmap_t src_pmap); |
| 442 | void	pmap_page_array_startup(long count); |
| 443 | vm_page_t pmap_page_alloc_below_4g(bool zeroed); |
| 444 | |
| 445 | #if defined(KASAN) || defined(KMSAN) |
| 446 | void	pmap_san_enter(vm_offset_t); |
| 447 | #endif |
| 448 | |
| 449 | /* |
| 450 | * Returns a pointer to a set of CPUs on which the pmap is currently active. |
| 451 | * Note that the set can be modified without any mutual exclusion, so a copy |
| 452 | * must be made if a stable value is required. |
| 453 | */ |
| 454 | static __inline volatile cpuset_t * |
| 455 | pmap_invalidate_cpu_mask(pmap_t pmap) |
| 456 | { |
| 457 | 	return (&pmap->pm_active); |
| 458 | } |
| 459 | |
| 460 | #if defined(_SYS_PCPU_H_) && defined(_MACHINE_CPUFUNC_H_) |
| 461 | /* |
| 462 | * It seems that AlderLake+ small cores have some microarchitectural |
| 463 | * bug, which results in the INVLPG instruction failing to flush all |
| 464 | * global TLB entries when PCID is enabled. Work around it for now, |
| 465 | * by doing global invalidation on small cores instead of INVLPG. |
| 466 | */ |
| 467 | static __inline void |
| 468 | pmap_invlpg(pmap_t pmap, vm_offset_t va) |
| 469 | { |
| 470 | 	if (pmap == kernel_pmap && PCPU_GET(pcid_invlpg_workaround)) { |
| 471 | 		struct invpcid_descr d = { 0 }; |
| 472 | |
| 473 | 		invpcid(&d, INVPCID_CTXGLOB); |
| 474 | 	} else { |
| 475 | 		invlpg(va); |
| 476 | 	} |
| 477 | } |
| 478 | #endif /* sys/pcpu.h && machine/cpufunc.h */ |
| 479 | |
| 480 | #if defined(_SYS_PCPU_H_) |
| 481 | /* Return pcid for the pmap pmap on current cpu */ |
| 482 | static __inline uint32_t |
| 483 | pmap_get_pcid(pmap_t pmap) |
| 484 | { |
| 485 | 	struct pmap_pcid *pcidp; |
| 486 | |
| 487 | 	MPASS(pmap_pcid_enabled); |
| 488 | 	pcidp = zpcpu_get(pmap->pm_pcidp); |
| 489 | 	return (pcidp->pm_pcid); |
| 490 | } |
| 491 | #endif /* sys/pcpu.h */ |
| 492 | |
| 493 | /* |
| 494 | * Invalidation request. PCPU pc_smp_tlb_op uses u_int instead of the |
| 495 | * enum to avoid both namespace and ABI issues (with enums). |
| 496 | */ |
| 497 | enum invl_op_codes { |
| 498 | 	INVL_OP_TLB = 1, |
| 499 | 	INVL_OP_TLB_INVPCID = 2, |
| 500 | 	INVL_OP_TLB_INVPCID_PTI = 3, |
| 501 | 	INVL_OP_TLB_PCID = 4, |
| 502 | 	INVL_OP_PGRNG = 5, |
| 503 | 	INVL_OP_PGRNG_INVPCID = 6, |
| 504 | 	INVL_OP_PGRNG_PCID = 7, |
| 505 | 	INVL_OP_PG = 8, |
| 506 | 	INVL_OP_PG_INVPCID = 9, |
| 507 | 	INVL_OP_PG_PCID = 10, |
| 508 | 	INVL_OP_CACHE = 11, |
| 509 | }; |
| 510 | |
| 511 | typedef void (*smp_invl_local_cb_t)(struct pmap *, vm_offset_t addr1, |
| 512 | vm_offset_t addr2); |
| 513 | typedef void (*smp_targeted_tlb_shootdown_t)(pmap_t, vm_offset_t, vm_offset_t, |
| 514 | smp_invl_local_cb_t, enum invl_op_codes); |
| 515 | |
| 516 | void smp_targeted_tlb_shootdown_native(pmap_t, vm_offset_t, vm_offset_t, |
| 517 | smp_invl_local_cb_t, enum invl_op_codes); |
| 518 | extern smp_targeted_tlb_shootdown_t smp_targeted_tlb_shootdown; |
| 519 | |
| 520 | #endif /* _KERNEL */ |
| 521 | |
| 522 | /* Return various clipped indexes for a given VA */ |
| 523 | static __inline vm_pindex_t |
| 524 | pmap_pte_index(vm_offset_t va) |
| 525 | { |
| 526 | |
| 527 | 	return ((va >> PAGE_SHIFT) & ((1ul << NPTEPGSHIFT) - 1)); |
| 528 | } |
| 529 | |
| 530 | static __inline vm_pindex_t |
| 531 | pmap_pde_index(vm_offset_t va) |
| 532 | { |
| 533 | |
| 534 | 	return ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1)); |
| 535 | } |
| 536 | |
| 537 | static __inline vm_pindex_t |
| 538 | pmap_pdpe_index(vm_offset_t va) |
| 539 | { |
| 540 | |
| 541 | 	return ((va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1)); |
| 542 | } |
| 543 | |
| 544 | static __inline vm_pindex_t |
| 545 | pmap_pml4e_index(vm_offset_t va) |
| 546 | { |
| 547 | |
| 548 | 	return ((va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1)); |
| 549 | } |
| 550 | |
| 551 | static __inline vm_pindex_t |
| 552 | pmap_pml5e_index(vm_offset_t va) |
| 553 | { |
| 554 | |
| 555 | 	return ((va >> PML5SHIFT) & ((1ul << NPML5EPGSHIFT) - 1)); |
| 556 | } |
| 557 | |
| 558 | struct kva_layout_s { |
| 559 | 	vm_offset_t kva_min; |
| 560 | 	vm_offset_t kva_max; |
| 561 | 	vm_offset_t dmap_low;	/* DMAP_MIN_ADDRESS */ |
| 562 | 	vm_offset_t dmap_high;	/* DMAP_MAX_ADDRESS */ |
| 563 | 	vm_offset_t lm_low;	/* LARGEMAP_MIN_ADDRESS */ |
| 564 | 	vm_offset_t lm_high;	/* LARGEMAP_MAX_ADDRESS */ |
| 565 | 	vm_offset_t km_low;	/* VM_MIN_KERNEL_ADDRESS */ |
| 566 | 	vm_offset_t km_high;	/* VM_MAX_KERNEL_ADDRESS */ |
| 567 | 	vm_offset_t rec_pt; |
| 568 | 	vm_offset_t kasan_shadow_low;	/* KASAN_MIN_ADDRESS */ |
| 569 | 	vm_offset_t kasan_shadow_high;	/* KASAN_MAX_ADDRESS */ |
| 570 | 	vm_offset_t kmsan_shadow_low;	/* KMSAN_SHAD_MIN_ADDRESS */ |
| 571 | 	vm_offset_t kmsan_shadow_high;	/* KMSAN_SHAD_MAX_ADDRESS */ |
| 572 | 	vm_offset_t kmsan_origin_low;	/* KMSAN_ORIG_MIN_ADDRESS */ |
| 573 | 	vm_offset_t kmsan_origin_high;	/* KMSAN_ORIG_MAX_ADDRESS */ |
| 574 | }; |
| 575 | extern struct kva_layout_s kva_layout; |
| 576 | |
| 577 | #endif /* !LOCORE */ |
| 578 | |
| 579 | #endif /* !_MACHINE_PMAP_H_ */ |
| 580 | |
| 581 | #endif /* __i386__ */ |