X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/9bccf70c0258c7cac2dcb80011b2a964d884c552..99c3a10404e5d1ef94397ab4df5a8b74711fc4d3:/osfmk/i386/pmap.h diff --git a/osfmk/i386/pmap.h b/osfmk/i386/pmap.h index 0517c243c..5cc91b6e2 100644 --- a/osfmk/i386/pmap.h +++ b/osfmk/i386/pmap.h @@ -1,23 +1,29 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2007 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * - * The contents of this file constitute Original Code as defined in and - * are subject to the Apple Public Source License Version 1.1 (the - * "License"). You may not use this file except in compliance with the - * License. Please obtain a copy of the License at - * http://www.apple.com/publicsource and read it before using this file. + * This file contains Original Code and/or Modifications of Original Code + * as defined in and that are subject to the Apple Public Source License + * Version 2.0 (the 'License'). You may not use this file except in + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. * - * This Original Code and all software distributed under the License are - * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. + * + * The Original Code and all software distributed under the License are + * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the - * License for the specific language governing rights and limitations - * under the License. + * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. + * Please see the License for the specific language governing rights and + * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * @OSF_COPYRIGHT@ @@ -58,14 +64,13 @@ * * Machine-dependent structures for the physical map module. */ - +#ifdef KERNEL_PRIVATE #ifndef _PMAP_MACHINE_ #define _PMAP_MACHINE_ 1 #ifndef ASSEMBLER #include -#include #include #include @@ -73,8 +78,14 @@ #include #include #include -#include +#include #include +#include + +#include +#include + +#include /* * Define the generic in terms of the specific @@ -94,32 +105,259 @@ * i386/i486/i860 Page Table Entry */ -typedef unsigned int pt_entry_t; +#endif /* ASSEMBLER */ + +#define NPGPTD 4ULL +#define PDESHIFT 21ULL +#define PTEMASK 0x1ffULL +#define PTEINDX 3ULL + +#define PTESHIFT 12ULL + +#ifdef __i386__ +#define INITPT_SEG_BASE 0x100000 +#endif + +#ifdef __x86_64__ +#define LOW_4GB_MASK ((vm_offset_t)0x00000000FFFFFFFFUL) +#endif + +#define PDESIZE sizeof(pd_entry_t) /* for assembly files */ +#define PTESIZE sizeof(pt_entry_t) /* for assembly files */ + +#define INTEL_OFFMASK (I386_PGBYTES - 1) +#define INTEL_LOFFMASK (I386_LPGBYTES - 1) +#define PG_FRAME 0x000FFFFFFFFFF000ULL +#define NPTEPG (PAGE_SIZE/(sizeof (pt_entry_t))) +#define NPTDPG (PAGE_SIZE/(sizeof (pd_entry_t))) + +#define NBPTD (NPGPTD << PAGE_SHIFT) +#define NPDEPTD (NBPTD / (sizeof (pd_entry_t))) +#define NPDEPG (PAGE_SIZE/(sizeof (pd_entry_t))) +#define NBPDE (1ULL << PDESHIFT) +#define PDEMASK (NBPDE - 1) + +#define PTE_PER_PAGE 512 /* number of PTE's per page on any level */ + + /* cleanly define parameters for all the page table levels */ +typedef uint64_t pml4_entry_t; +#define NPML4PG (PAGE_SIZE/(sizeof (pml4_entry_t))) +#define PML4SHIFT 39 +#define PML4PGSHIFT 9 +#define NBPML4 (1ULL << PML4SHIFT) +#define PML4MASK (NBPML4-1) +#define PML4_ENTRY_NULL ((pml4_entry_t *) 0) + +typedef uint64_t pdpt_entry_t; +#define NPDPTPG (PAGE_SIZE/(sizeof (pdpt_entry_t))) +#define PDPTSHIFT 30 +#define PDPTPGSHIFT 9 +#define NBPDPT (1ULL << PDPTSHIFT) +#define PDPTMASK (NBPDPT-1) +#define PDPT_ENTRY_NULL ((pdpt_entry_t *) 0) + +typedef uint64_t pd_entry_t; +#define NPDPG (PAGE_SIZE/(sizeof (pd_entry_t))) +#define PDSHIFT 21 +#define PDPGSHIFT 9 +#define NBPD (1ULL << PDSHIFT) +#define PDMASK (NBPD-1) +#define PD_ENTRY_NULL ((pd_entry_t *) 0) + +typedef uint64_t pt_entry_t; +#define NPTPG (PAGE_SIZE/(sizeof (pt_entry_t))) +#define PTSHIFT 12 +#define PTPGSHIFT 9 +#define NBPT (1ULL << PTSHIFT) +#define PTMASK (NBPT-1) #define PT_ENTRY_NULL ((pt_entry_t *) 0) -#endif /* ASSEMBLER */ +typedef uint64_t pmap_paddr_t; + +#if DEBUG +#define PMAP_ASSERT 1 +#endif +#if PMAP_ASSERT +#define pmap_assert(ex) ((ex) ? (void)0 : Assert(__FILE__, __LINE__, # ex)) + +#define pmap_assert2(ex, fmt, args...) \ + do { \ + if (!(ex)) { \ + kprintf("Assertion %s failed (%s:%d, caller %p) " fmt , #ex, __FILE__, __LINE__, __builtin_return_address(0), ##args); \ + panic("Assertion %s failed (%s:%d, caller %p) " fmt , #ex, __FILE__, __LINE__, __builtin_return_address(0), ##args); \ + } \ + } while(0) +#else +#define pmap_assert(ex) +#define pmap_assert2(ex, fmt, args...) +#endif + +/* superpages */ +#ifdef __x86_64__ +#define SUPERPAGE_NBASEPAGES 512 +#else +#define SUPERPAGE_NBASEPAGES 1 /* we don't support superpages on i386 */ +#endif -#define INTEL_OFFMASK 0xfff /* offset within page */ -#define PDESHIFT 22 /* page descriptor shift */ -#define PDEMASK 0x3ff /* mask for page descriptor index */ -#define PTESHIFT 12 /* page table shift */ -#define PTEMASK 0x3ff /* mask for page table index */ +/* + * Atomic 64-bit store of a page table entry. + */ +static inline void +pmap_store_pte(pt_entry_t *entryp, pt_entry_t value) +{ +#ifdef __i386__ + /* + * Load the new value into %ecx:%ebx + * Load the old value into %edx:%eax + * Compare-exchange-8bytes at address entryp (loaded in %edi) + * If the compare succeeds, the new value will have been stored. + * Otherwise, the old value changed and reloaded, so try again. + */ + __asm__ volatile( + " movl (%0), %%eax \n\t" + " movl 4(%0), %%edx \n\t" + "1: \n\t" + " cmpxchg8b (%0) \n\t" + " jnz 1b" + : + : "D" (entryp), + "b" ((uint32_t)value), + "c" ((uint32_t)(value >> 32)) + : "eax", "edx", "memory"); +#else + /* + * In the 32-bit kernel a compare-and-exchange loop was + * required to provide atomicity. For K64, life is easier: + */ + *entryp = value; +#endif +} +/* in 64 bit spaces, the number of each type of page in the page tables */ +#define NPML4PGS (1ULL * (PAGE_SIZE/(sizeof (pml4_entry_t)))) +#define NPDPTPGS (NPML4PGS * (PAGE_SIZE/(sizeof (pdpt_entry_t)))) +#define NPDEPGS (NPDPTPGS * (PAGE_SIZE/(sizeof (pd_entry_t)))) +#define NPTEPGS (NPDEPGS * (PAGE_SIZE/(sizeof (pt_entry_t)))) +#ifdef __i386__ +/* + * The 64-bit kernel is remapped in uber-space which is at the base + * the highest 4th-level directory (KERNEL_UBER_PML4_INDEX). That is, + * 512GB from the top of virtual space (or zero). + */ +#define KERNEL_UBER_PML4_INDEX 511 +#define KERNEL_UBER_BASE (0ULL - NBPML4) +#define KERNEL_UBER_BASE_HI32 ((uint32_t)(KERNEL_UBER_BASE >> 32)) +#else +#define KERNEL_PML4_INDEX 511 +#define KERNEL_KEXTS_INDEX 510 /* Home of KEXTs - the basement */ +#define KERNEL_PHYSMAP_PML4_INDEX 509 /* virtual to physical map */ +#define KERNEL_BASE (0ULL - NBPML4) +#define KERNEL_BASEMENT (KERNEL_BASE - NBPML4) +#endif + +#define VM_WIMG_COPYBACK VM_MEM_COHERENT +#define VM_WIMG_COPYBACKLW VM_WIMG_COPYBACK #define VM_WIMG_DEFAULT VM_MEM_COHERENT +/* ?? intel ?? */ +#define VM_WIMG_IO (VM_MEM_COHERENT | \ + VM_MEM_NOT_CACHEABLE | VM_MEM_GUARDED) +#define VM_WIMG_WTHRU (VM_MEM_WRITE_THROUGH | VM_MEM_COHERENT | VM_MEM_GUARDED) +/* write combining mode, aka store gather */ +#define VM_WIMG_WCOMB (VM_MEM_NOT_CACHEABLE | VM_MEM_COHERENT) +#define VM_WIMG_INNERWBACK VM_MEM_COHERENT +/* + * Pte related macros + */ +#ifdef __i386__ +#define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<dirbase[(vm_offset_t)(v) >> PDESHIFT]))*/ + +#define HIGH_MEM_BASE ((uint32_t)( -NBPDE) ) /* shared gdt etc seg addr */ /* XXX64 ?? */ +#define pmap_index_to_virt(x) (HIGH_MEM_BASE | ((unsigned)(x) << PAGE_SHIFT)) +#endif /* * Convert address offset to page descriptor index */ -#define pdenum(pmap, a) (((((pmap) == kernel_pmap) ? \ - kvtolinear(a) : (a)) \ - >> PDESHIFT) & PDEMASK) +#define pdptnum(pmap, a) (((vm_offset_t)(a) >> PDPTSHIFT) & PDPTMASK) +#define pdenum(pmap, a) (((vm_offset_t)(a) >> PDESHIFT) & PDEMASK) +#define PMAP_INVALID_PDPTNUM (~0ULL) + +#define pdeidx(pmap, a) (((a) >> PDSHIFT) & ((1ULL<<(48 - PDSHIFT)) -1)) +#define pdptidx(pmap, a) (((a) >> PDPTSHIFT) & ((1ULL<<(48 - PDPTSHIFT)) -1)) +#define pml4idx(pmap, a) (((a) >> PML4SHIFT) & ((1ULL<<(48 - PML4SHIFT)) -1)) + /* * Convert page descriptor index to user virtual address @@ -129,30 +367,43 @@ typedef unsigned int pt_entry_t; /* * Convert address offset to page table index */ -#define ptenum(a) (((a) >> PTESHIFT) & PTEMASK) - -#define NPTES (intel_ptob(1)/sizeof(pt_entry_t)) -#define NPDES (intel_ptob(1)/sizeof(pt_entry_t)) +#define ptenum(a) (((vm_offset_t)(a) >> PTESHIFT) & PTEMASK) /* * Hardware pte bit definitions (to be used directly on the ptes * without using the bit fields). */ -#define INTEL_PTE_VALID 0x00000001 -#define INTEL_PTE_WRITE 0x00000002 -#define INTEL_PTE_USER 0x00000004 -#define INTEL_PTE_WTHRU 0x00000008 -#define INTEL_PTE_NCACHE 0x00000010 -#define INTEL_PTE_REF 0x00000020 -#define INTEL_PTE_MOD 0x00000040 -#define INTEL_PTE_WIRED 0x00000200 -#define INTEL_PTE_PFN 0xfffff000 - -#define pa_to_pte(a) ((a) & INTEL_PTE_PFN) -#define pte_to_pa(p) ((p) & INTEL_PTE_PFN) +#define INTEL_PTE_VALID 0x00000001ULL +#define INTEL_PTE_WRITE 0x00000002ULL +#define INTEL_PTE_RW 0x00000002ULL +#define INTEL_PTE_USER 0x00000004ULL +#define INTEL_PTE_WTHRU 0x00000008ULL +#define INTEL_PTE_NCACHE 0x00000010ULL +#define INTEL_PTE_REF 0x00000020ULL +#define INTEL_PTE_MOD 0x00000040ULL +#define INTEL_PTE_PS 0x00000080ULL +#define INTEL_PTE_PTA 0x00000080ULL +#define INTEL_PTE_GLOBAL 0x00000100ULL +#define INTEL_PTE_WIRED 0x00000200ULL +#define INTEL_PDPTE_NESTED 0x00000400ULL +#define INTEL_PTE_PFN PG_FRAME + +#define INTEL_PTE_NX (1ULL << 63) + +#define INTEL_PTE_INVALID 0 +/* This is conservative, but suffices */ +#define INTEL_PTE_RSVD ((1ULL << 10) | (1ULL << 11) | (0x1FFULL << 54)) + +#define pa_to_pte(a) ((a) & INTEL_PTE_PFN) /* XXX */ +#define pte_to_pa(p) ((p) & INTEL_PTE_PFN) /* XXX */ #define pte_increment_pa(p) ((p) += INTEL_OFFMASK+1) +#define pte_kernel_rw(p) ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_RW)) +#define pte_kernel_ro(p) ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID)) +#define pte_user_rw(p) ((pt_entry)t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_USER|INTEL_PTE_RW)) +#define pte_user_ro(p) ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_USER)) + #define PMAP_DEFAULT_CACHE 0 #define PMAP_INHIBIT_CACHE 1 #define PMAP_GUARDED_CACHE 2 @@ -160,279 +411,443 @@ typedef unsigned int pt_entry_t; #define PMAP_NO_GUARD_CACHE 8 -/* - * Convert page table entry to kernel virtual address - */ -#define ptetokv(a) (phystokv(pte_to_pa(a))) - #ifndef ASSEMBLER -typedef volatile long cpu_set; /* set of CPUs - must be <= 32 */ - /* changed by other processors */ -struct pmap { - pt_entry_t *dirbase; /* page directory pointer register */ - vm_offset_t pdirbase; /* phys. address of dirbase */ - int ref_count; /* reference count */ - decl_simple_lock_data(,lock) /* lock on map */ - struct pmap_statistics stats; /* map statistics */ - cpu_set cpus_using; /* bitmap of cpus using pmap */ -}; +#include -/* - * Optimization avoiding some TLB flushes when switching to - * kernel-loaded threads. This is effective only for i386: - * Since user task, kernel task and kernel loaded tasks share the - * same virtual space (with appropriate protections), any pmap - * allows mapping kernel and kernel loaded tasks. - * - * The idea is to avoid switching to another pmap unnecessarily when - * switching to a kernel-loaded task, or when switching to the kernel - * itself. - * - * We store the pmap we are really using (from which we fetched the - * dirbase value) in real_pmap[cpu_number()]. - * - * Invariant: - * current_pmap() == real_pmap[cpu_number()] || current_pmap() == kernel_pmap. +/* + * Address of current and alternate address space page table maps + * and directories. */ -extern struct pmap *real_pmap[NCPUS]; - -#include +#ifdef __i386__ +extern pt_entry_t PTmap[], APTmap[], Upte; +extern pd_entry_t PTD[], APTD[], PTDpde[], APTDpde[], Upde; +extern pmap_paddr_t lo_kernel_cr3; +extern pdpt_entry_t *IdlePDPT64; +extern pdpt_entry_t IdlePDPT[]; +extern pml4_entry_t IdlePML4[]; +#else +extern pt_entry_t *PTmap; +extern pdpt_entry_t *IdlePDPT; +extern pml4_entry_t *IdlePML4; +#endif +extern boolean_t no_shared_cr3; +extern addr64_t kernel64_cr3; +extern pd_entry_t *IdlePTD; /* physical addr of "Idle" state PTD */ + +extern uint64_t pmap_pv_hashlist_walks; +extern uint64_t pmap_pv_hashlist_cnts; +extern uint32_t pmap_pv_hashlist_max; +extern uint32_t pmap_kernel_text_ps; + +#ifdef __i386__ /* - * If switching to the kernel pmap, don't incur the TLB cost of switching - * to its page tables, since all maps include the kernel map as a subset. - * Simply record that this CPU is logically on the kernel pmap (see - * pmap_destroy). - * - * Similarly, if switching to a pmap (other than kernel_pmap that is already - * in use, don't do anything to the hardware, to avoid a TLB flush. + * ** i386 ** + * virtual address to page table entry and + * to physical address. Likewise for alternate address space. + * Note: these work recursively, thus vtopte of a pte will give + * the corresponding pde that in turn maps it. */ -#if NCPUS > 1 -#define PMAP_CPU_SET(pmap, my_cpu) i_bit_set(my_cpu, &((pmap)->cpus_using)) -#define PMAP_CPU_CLR(pmap, my_cpu) i_bit_clear(my_cpu, &((pmap)->cpus_using)) -#else /* NCPUS > 1 */ -#define PMAP_CPU_SET(pmap,my_cpu) (pmap)->cpus_using = TRUE -#define PMAP_CPU_CLR(pmap,my_cpu) (pmap)->cpus_using = FALSE -#endif /* NCPUS > 1 */ +#define vtopte(va) (PTmap + i386_btop((vm_offset_t)va)) +#endif -#define set_dirbase(mypmap, my_cpu) { \ - struct pmap **ppmap = &real_pmap[my_cpu]; \ - vm_offset_t pdirbase = (mypmap)->pdirbase; \ - \ - if (*ppmap == (vm_offset_t)NULL) { \ - *ppmap = (mypmap); \ - PMAP_CPU_SET((mypmap), my_cpu); \ - set_cr3(pdirbase); \ - } else if ((mypmap) != kernel_pmap && (mypmap) != *ppmap ) { \ - if (*ppmap != kernel_pmap) \ - PMAP_CPU_CLR(*ppmap, my_cpu); \ - *ppmap = (mypmap); \ - PMAP_CPU_SET((mypmap), my_cpu); \ - set_cr3(pdirbase); \ - } \ - assert((mypmap) == *ppmap || (mypmap) == kernel_pmap); \ +#ifdef __x86_64__ +#define ID_MAP_VTOP(x) ((void *)(((uint64_t)(x)) & LOW_4GB_MASK)) + +extern uint64_t physmap_base, physmap_max; + +#define NPHYSMAP (MAX(K64_MAXMEM/GB + 4, 4)) + +static inline boolean_t physmap_enclosed(addr64_t a) { + return (a < (NPHYSMAP * GB)); +} + +static inline void * PHYSMAP_PTOV_check(void *paddr) { + uint64_t pvaddr = (uint64_t)paddr + physmap_base; + + if (__improbable(pvaddr >= physmap_max)) + panic("PHYSMAP_PTOV bounds exceeded, 0x%qx, 0x%qx, 0x%qx", + pvaddr, physmap_base, physmap_max); + + return (void *)pvaddr; } -#if NCPUS > 1 +#define PHYSMAP_PTOV(x) (PHYSMAP_PTOV_check((void*) (x))) + /* - * List of cpus that are actively using mapped memory. Any - * pmap update operation must wait for all cpus in this list. - * Update operations must still be queued to cpus not in this - * list. + * For KASLR, we alias the master processor's IDT and GDT at fixed + * virtual addresses to defeat SIDT/SGDT address leakage. */ -extern cpu_set cpus_active; +#define MASTER_IDT_ALIAS (VM_MIN_KERNEL_ADDRESS + 0x0000) +#define MASTER_GDT_ALIAS (VM_MIN_KERNEL_ADDRESS + 0x1000) /* - * List of cpus that are idle, but still operating, and will want - * to see any kernel pmap updates when they become active. + * The low global vector page is mapped at a fixed alias also. */ -extern cpu_set cpus_idle; +#define LOWGLOBAL_ALIAS (VM_MIN_KERNEL_ADDRESS + 0x2000) +#endif /*__x86_64__ */ + +typedef volatile long cpu_set; /* set of CPUs - must be <= 32 */ + /* changed by other processors */ +#include + +/* + * For each vm_page_t, there is a list of all currently + * valid virtual mappings of that page. An entry is + * a pv_entry_t; the list is the pv_table. + */ + +struct pmap { + decl_simple_lock_data(,lock) /* lock on map */ + pmap_paddr_t pm_cr3; /* physical addr */ + boolean_t pm_shared; + pd_entry_t *dirbase; /* page directory pointer */ +#ifdef __i386__ + pmap_paddr_t pdirbase; /* phys. address of dirbase */ + vm_offset_t pm_hold; /* true pdpt zalloc addr */ +#endif + vm_object_t pm_obj; /* object to hold pde's */ + task_map_t pm_task_map; + pdpt_entry_t *pm_pdpt; /* KVA of 3rd level page */ + pml4_entry_t *pm_pml4; /* VKA of top level */ + vm_object_t pm_obj_pdpt; /* holds pdpt pages */ + vm_object_t pm_obj_pml4; /* holds pml4 pages */ +#define PMAP_PCID_MAX_CPUS (48) /* Must be a multiple of 8 */ + pcid_t pmap_pcid_cpus[PMAP_PCID_MAX_CPUS]; + volatile uint8_t pmap_pcid_coherency_vector[PMAP_PCID_MAX_CPUS]; + struct pmap_statistics stats; /* map statistics */ + int ref_count; /* reference count */ + int nx_enabled; + ledger_t ledger; /* ledger tracking phys mappings */ +}; + + +#if NCOPY_WINDOWS > 0 +#define PMAP_PDPT_FIRST_WINDOW 0 +#define PMAP_PDPT_NWINDOWS 4 +#define PMAP_PDE_FIRST_WINDOW (PMAP_PDPT_NWINDOWS) +#define PMAP_PDE_NWINDOWS 4 +#define PMAP_PTE_FIRST_WINDOW (PMAP_PDE_FIRST_WINDOW + PMAP_PDE_NWINDOWS) +#define PMAP_PTE_NWINDOWS 4 + +#define PMAP_NWINDOWS_FIRSTFREE (PMAP_PTE_FIRST_WINDOW + PMAP_PTE_NWINDOWS) +#define PMAP_WINDOW_SIZE 8 +#define PMAP_NWINDOWS (PMAP_NWINDOWS_FIRSTFREE + PMAP_WINDOW_SIZE) + +typedef struct { + pt_entry_t *prv_CMAP; + caddr_t prv_CADDR; +} mapwindow_t; + +typedef struct cpu_pmap { + int pdpt_window_index; + int pde_window_index; + int pte_window_index; + mapwindow_t mapwindow[PMAP_NWINDOWS]; +} cpu_pmap_t; + + +extern mapwindow_t *pmap_get_mapwindow(pt_entry_t pentry); +extern void pmap_put_mapwindow(mapwindow_t *map); +#endif + +typedef struct pmap_memory_regions { + ppnum_t base; + ppnum_t end; + ppnum_t alloc; + uint32_t type; + uint64_t attribute; +} pmap_memory_region_t; + +extern unsigned pmap_memory_region_count; +extern unsigned pmap_memory_region_current; + +#define PMAP_MEMORY_REGIONS_SIZE 128 + +extern pmap_memory_region_t pmap_memory_regions[]; +#include + +static inline void +set_dirbase(pmap_t tpmap, __unused thread_t thread) { + int ccpu = cpu_number(); + cpu_datap(ccpu)->cpu_task_cr3 = tpmap->pm_cr3; + cpu_datap(ccpu)->cpu_task_map = tpmap->pm_task_map; +#ifndef __i386__ + /* + * Switch cr3 if necessary + * - unless running with no_shared_cr3 debugging mode + * and we're not on the kernel's cr3 (after pre-empted copyio) + */ + if (__probable(!no_shared_cr3)) { + if (get_cr3_base() != tpmap->pm_cr3) { + if (pmap_pcid_ncpus) { + pmap_pcid_activate(tpmap, ccpu); + } + else + set_cr3_raw(tpmap->pm_cr3); + } + } else { + if (get_cr3_base() != cpu_datap(ccpu)->cpu_kernel_cr3) + set_cr3_raw(cpu_datap(ccpu)->cpu_kernel_cr3); + } +#endif +} /* * External declarations for PMAP_ACTIVATE. */ -extern void process_pmap_updates(struct pmap *pmap); +extern void process_pmap_updates(void); extern void pmap_update_interrupt(void); -#endif /* NCPUS > 1 */ - /* * Machine dependent routines that are used only for i386/i486/i860. */ -extern vm_offset_t (phystokv)( - vm_offset_t pa); -extern vm_offset_t (kvtophys)( +extern addr64_t (kvtophys)( vm_offset_t addr); +extern kern_return_t pmap_expand( + pmap_t pmap, + vm_map_offset_t addr, + unsigned int options); +#if !defined(__x86_64__) extern pt_entry_t *pmap_pte( struct pmap *pmap, - vm_offset_t addr); + vm_map_offset_t addr); +extern pd_entry_t *pmap_pde( + struct pmap *pmap, + vm_map_offset_t addr); + +extern pd_entry_t *pmap64_pde( + struct pmap *pmap, + vm_map_offset_t addr); + +extern pdpt_entry_t *pmap64_pdpt( + struct pmap *pmap, + vm_map_offset_t addr); +#endif extern vm_offset_t pmap_map( vm_offset_t virt, - vm_offset_t start, - vm_offset_t end, - vm_prot_t prot); + vm_map_offset_t start, + vm_map_offset_t end, + vm_prot_t prot, + unsigned int flags); extern vm_offset_t pmap_map_bd( vm_offset_t virt, - vm_offset_t start, - vm_offset_t end, - vm_prot_t prot); + vm_map_offset_t start, + vm_map_offset_t end, + vm_prot_t prot, + unsigned int flags); extern void pmap_bootstrap( - vm_offset_t load_start); + vm_offset_t load_start, + boolean_t IA32e); extern boolean_t pmap_valid_page( - vm_offset_t pa); + ppnum_t pn); extern int pmap_list_resident_pages( struct pmap *pmap, vm_offset_t *listp, int space); - -extern void flush_tlb(void); +extern void x86_filter_TLB_coherency_interrupts(boolean_t); +#ifdef __i386__ +extern void pmap_commpage32_init( + vm_offset_t kernel, + vm_offset_t user, + int count); +extern void pmap_commpage64_init( + vm_offset_t kernel, + vm_map_offset_t user, + int count); + +#endif +/* + * Get cache attributes (as pagetable bits) for the specified phys page + */ +extern unsigned pmap_get_cache_attributes(ppnum_t); +#if NCOPY_WINDOWS > 0 +extern struct cpu_pmap *pmap_cpu_alloc( + boolean_t is_boot_cpu); +extern void pmap_cpu_free( + struct cpu_pmap *cp); +#endif + +extern void pmap_map_block( + pmap_t pmap, + addr64_t va, + ppnum_t pa, + uint32_t size, + vm_prot_t prot, + int attr, + unsigned int flags); + extern void invalidate_icache(vm_offset_t addr, unsigned cnt, int phys); extern void flush_dcache(vm_offset_t addr, unsigned count, int phys); - +extern ppnum_t pmap_find_phys(pmap_t map, addr64_t va); + +extern void pmap_cpu_init(void); +extern void pmap_disable_NX(pmap_t pmap); +#ifdef __i386__ +extern void pmap_set_4GB_pagezero(pmap_t pmap); +extern void pmap_clear_4GB_pagezero(pmap_t pmap); +extern void pmap_load_kernel_cr3(void); +extern vm_offset_t pmap_cpu_high_map_vaddr(int, enum high_cpu_types); +extern vm_offset_t pmap_high_map_vaddr(enum high_cpu_types); +extern vm_offset_t pmap_high_map(pt_entry_t, enum high_cpu_types); +extern vm_offset_t pmap_cpu_high_shared_remap(int, enum high_cpu_types, vm_offset_t, int); +extern vm_offset_t pmap_high_shared_remap(enum high_fixed_addresses, vm_offset_t, int); +#endif + +extern void pt_fake_zone_init(int); +extern void pt_fake_zone_info(int *, vm_size_t *, vm_size_t *, vm_size_t *, vm_size_t *, + uint64_t *, int *, int *, int *); +extern void pmap_pagetable_corruption_msg_log(int (*)(const char * fmt, ...)__printflike(1,2)); /* * Macros for speed. */ -#if NCPUS > 1 #include -/* - * For multiple CPUS, PMAP_ACTIVATE and PMAP_DEACTIVATE must manage - * fields to control TLB invalidation on other CPUS. - */ - -#define PMAP_ACTIVATE_KERNEL(my_cpu) { \ - \ - /* \ - * Let pmap updates proceed while we wait for this pmap. \ - */ \ - i_bit_clear((my_cpu), &cpus_active); \ - \ - /* \ - * Lock the pmap to put this cpu in its active set. \ - * Wait for updates here. \ - */ \ - simple_lock(&kernel_pmap->lock); \ - \ - /* \ - * Mark that this cpu is using the pmap. \ - */ \ - i_bit_set((my_cpu), &kernel_pmap->cpus_using); \ - \ - /* \ - * Mark this cpu active - IPL will be lowered by \ - * load_context(). \ - */ \ - i_bit_set((my_cpu), &cpus_active); \ - \ - simple_unlock(&kernel_pmap->lock); \ + +#define PMAP_ACTIVATE_MAP(map, thread) { \ + register pmap_t tpmap; \ + \ + tpmap = vm_map_pmap(map); \ + set_dirbase(tpmap, thread); \ } -#define PMAP_DEACTIVATE_KERNEL(my_cpu) { \ - /* \ - * Mark pmap no longer in use by this cpu even if \ - * pmap is locked against updates. \ - */ \ - i_bit_clear((my_cpu), &kernel_pmap->cpus_using); \ -} +#ifdef __i386__ +#define PMAP_DEACTIVATE_MAP(map, thread) \ + if (vm_map_pmap(map)->pm_task_map == TASK_MAP_64BIT_SHARED) \ + pmap_load_kernel_cr3(); +#elif defined(__x86_64__) +#define PMAP_DEACTIVATE_MAP(map, thread) \ + pmap_assert(pmap_pcid_ncpus ? (pcid_for_pmap_cpu_tuple(map->pmap, cpu_number()) == (get_cr3_raw() & 0xFFF)) : TRUE); +#else +#define PMAP_DEACTIVATE_MAP(map, thread) +#endif -#define PMAP_ACTIVATE_MAP(map, my_cpu) { \ - register struct pmap *tpmap; \ - \ - tpmap = vm_map_pmap(map); \ - if (tpmap == kernel_pmap) { \ - /* \ - * If this is the kernel pmap, switch to its page tables. \ - */ \ - set_dirbase(kernel_pmap, my_cpu); \ - } \ - else { \ - /* \ - * Let pmap updates proceed while we wait for this pmap. \ - */ \ - i_bit_clear((my_cpu), &cpus_active); \ - \ - /* \ - * Lock the pmap to put this cpu in its active set. \ - * Wait for updates here. \ - */ \ - simple_lock(&tpmap->lock); \ - \ - /* \ - * No need to invalidate the TLB - the entire user pmap \ - * will be invalidated by reloading dirbase. \ - */ \ - set_dirbase(tpmap, my_cpu); \ - \ - /* \ - * Mark this cpu active - IPL will be lowered by \ - * load_context(). \ - */ \ - i_bit_set((my_cpu), &cpus_active); \ - \ - simple_unlock(&tpmap->lock); \ - } \ -} +#if defined(__i386__) -#define PMAP_DEACTIVATE_MAP(map, my_cpu) - -#define PMAP_ACTIVATE_USER(th, my_cpu) { \ - spl_t spl; \ - \ - spl = splhigh(); \ - PMAP_ACTIVATE_MAP(th->map, my_cpu) \ - splx(spl); \ -} - -#define PMAP_DEACTIVATE_USER(th, my_cpu) { \ +#define PMAP_SWITCH_CONTEXT(old_th, new_th, my_cpu) { \ spl_t spl; \ - \ - spl = splhigh(); \ - PMAP_DEACTIVATE_MAP(th->map, my_cpu) \ + pt_entry_t *kpdp; \ + pt_entry_t *updp; \ + int i; \ + int need_flush; \ + \ + need_flush = 0; \ + spl = splhigh(); \ + if ((old_th->map != new_th->map) || (new_th->task != old_th->task)) { \ + PMAP_DEACTIVATE_MAP(old_th->map, old_th); \ + PMAP_ACTIVATE_MAP(new_th->map, new_th); \ + } \ + kpdp = current_cpu_datap()->cpu_copywindow_pdp; \ + for (i = 0; i < NCOPY_WINDOWS; i++) { \ + if (new_th->machine.copy_window[i].user_base != (user_addr_t)-1) { \ + updp = pmap_pde(new_th->map->pmap, \ + new_th->machine.copy_window[i].user_base);\ + pmap_store_pte(kpdp, updp ? *updp : 0); \ + } \ + kpdp++; \ + } \ splx(spl); \ + if (new_th->machine.copyio_state == WINDOWS_OPENED) \ + need_flush = 1; \ + else \ + new_th->machine.copyio_state = WINDOWS_DIRTY; \ + if (new_th->machine.physwindow_pte) { \ + pmap_store_pte((current_cpu_datap()->cpu_physwindow_ptep), \ + new_th->machine.physwindow_pte); \ + if (need_flush == 0) \ + invlpg((uintptr_t)current_cpu_datap()->cpu_physwindow_base);\ + } \ + if (need_flush) \ + flush_tlb(); \ } +#else /* __x86_64__ */ #define PMAP_SWITCH_CONTEXT(old_th, new_th, my_cpu) { \ - spl_t spl; \ - \ + \ + pmap_assert(ml_get_interrupts_enabled() == FALSE); \ if (old_th->map != new_th->map) { \ - spl = splhigh(); \ - PMAP_DEACTIVATE_MAP(old_th->map, my_cpu); \ - PMAP_ACTIVATE_MAP(new_th->map, my_cpu); \ - splx(spl); \ + PMAP_DEACTIVATE_MAP(old_th->map, old_th); \ + PMAP_ACTIVATE_MAP(new_th->map, new_th); \ } \ } +#endif /* __i386__ */ +#if NCOPY_WINDOWS > 0 +#define PMAP_SWITCH_USER(th, new_map, my_cpu) { \ + spl_t spl; \ + \ + spl = splhigh(); \ + PMAP_DEACTIVATE_MAP(th->map, th); \ + th->map = new_map; \ + PMAP_ACTIVATE_MAP(th->map, th); \ + splx(spl); \ + inval_copy_windows(th); \ +} +#else #define PMAP_SWITCH_USER(th, new_map, my_cpu) { \ spl_t spl; \ \ - spl = splhigh(); \ - PMAP_DEACTIVATE_MAP(th->map, my_cpu); \ + spl = splhigh(); \ + PMAP_DEACTIVATE_MAP(th->map, th); \ th->map = new_map; \ - PMAP_ACTIVATE_MAP(th->map, my_cpu); \ + PMAP_ACTIVATE_MAP(th->map, th); \ splx(spl); \ } +#endif -#if MP_V1_1 -#define set_led(cpu) -#define clear_led(cpu) -#endif /* MP_V1_1 */ +/* + * Marking the current cpu's cr3 inactive is achieved by setting its lsb. + * Marking the current cpu's cr3 active once more involves clearng this bit. + * Note that valid page tables are page-aligned and so the bottom 12 bits + * are normally zero, modulo PCID. + * We can only mark the current cpu active/inactive but we can test any cpu. + */ +#define CPU_CR3_MARK_INACTIVE() \ + current_cpu_datap()->cpu_active_cr3 |= 1 + +#define CPU_CR3_MARK_ACTIVE() \ + current_cpu_datap()->cpu_active_cr3 &= ~1 + +#define CPU_CR3_IS_ACTIVE(cpu) \ + ((cpu_datap(cpu)->cpu_active_cr3 & 1) == 0) +#define CPU_GET_ACTIVE_CR3(cpu) \ + (cpu_datap(cpu)->cpu_active_cr3 & ~1) + +#define CPU_GET_TASK_CR3(cpu) \ + (cpu_datap(cpu)->cpu_task_cr3) + +/* + * Mark this cpu idle, and remove it from the active set, + * since it is not actively using any pmap. Signal_cpus + * will notice that it is idle, and avoid signaling it, + * but will queue the update request for when the cpu + * becomes active. + */ +#if defined(__x86_64__) +#define MARK_CPU_IDLE(my_cpu) { \ + assert(ml_get_interrupts_enabled() == FALSE); \ + CPU_CR3_MARK_INACTIVE(); \ + __asm__ volatile("mfence"); \ +} +#else /* __i386__ native */ #define MARK_CPU_IDLE(my_cpu) { \ + assert(ml_get_interrupts_enabled() == FALSE); \ /* \ * Mark this cpu idle, and remove it from the active set, \ * since it is not actively using any pmap. Signal_cpus \ @@ -440,16 +855,17 @@ extern void flush_dcache(vm_offset_t addr, unsigned count, int phys); * but will queue the update request for when the cpu \ * becomes active. \ */ \ - int s = splhigh(); \ - i_bit_set((my_cpu), &cpus_idle); \ - i_bit_clear((my_cpu), &cpus_active); \ - splx(s); \ - set_led(my_cpu); \ + if (!cpu_mode_is64bit() || no_shared_cr3) \ + process_pmap_updates(); \ + else \ + pmap_load_kernel_cr3(); \ + CPU_CR3_MARK_INACTIVE(); \ + __asm__ volatile("mfence"); \ } +#endif /* __i386__ */ -#define MARK_CPU_ACTIVE(my_cpu) { \ - \ - int s = splhigh(); \ +#define MARK_CPU_ACTIVE(my_cpu) { \ + assert(ml_get_interrupts_enabled() == FALSE); \ /* \ * If a kernel_pmap update was requested while this cpu \ * was idle, process it as if we got the interrupt. \ @@ -460,73 +876,42 @@ extern void flush_dcache(vm_offset_t addr, unsigned count, int phys); * set assures that we will receive another update \ * interrupt if this happens. \ */ \ - i_bit_clear((my_cpu), &cpus_idle); \ + CPU_CR3_MARK_ACTIVE(); \ + __asm__ volatile("mfence"); \ \ - /* \ - * Mark that this cpu is now active. \ - */ \ - i_bit_set((my_cpu), &cpus_active); \ - splx(s); \ - clear_led(my_cpu); \ -} - -#else /* NCPUS > 1 */ - -/* - * With only one CPU, we just have to indicate whether the pmap is - * in use. - */ - -#define PMAP_ACTIVATE_KERNEL(my_cpu) { \ - kernel_pmap->cpus_using = TRUE; \ -} - -#define PMAP_DEACTIVATE_KERNEL(my_cpu) { \ - kernel_pmap->cpus_using = FALSE; \ -} - -#define PMAP_ACTIVATE_MAP(map, my_cpu) \ - set_dirbase(vm_map_pmap(map), my_cpu) - -#define PMAP_DEACTIVATE_MAP(map, my_cpu) - -#define PMAP_ACTIVATE_USER(th, my_cpu) \ - PMAP_ACTIVATE_MAP(th->map, my_cpu) - -#define PMAP_DEACTIVATE_USER(th, my_cpu) \ - PMAP_DEACTIVATE_MAP(th->map, my_cpu) - -#define PMAP_SWITCH_CONTEXT(old_th, new_th, my_cpu) { \ - if (old_th->map != new_th->map) { \ - PMAP_DEACTIVATE_MAP(old_th->map, my_cpu); \ - PMAP_ACTIVATE_MAP(new_th->map, my_cpu); \ - } \ + if (current_cpu_datap()->cpu_tlb_invalid) \ + process_pmap_updates(); \ } -#define PMAP_SWITCH_USER(th, new_map, my_cpu) { \ - PMAP_DEACTIVATE_MAP(th->map, my_cpu); \ - th->map = new_map; \ - PMAP_ACTIVATE_MAP(th->map, my_cpu); \ -} - -#endif /* NCPUS > 1 */ - #define PMAP_CONTEXT(pmap, thread) #define pmap_kernel_va(VA) \ - (((VA) >= VM_MIN_KERNEL_ADDRESS) && ((VA) <= VM_MAX_KERNEL_ADDRESS)) + ((((vm_offset_t) (VA)) >= vm_min_kernel_address) && \ + (((vm_offset_t) (VA)) <= vm_max_kernel_address)) + #define pmap_resident_count(pmap) ((pmap)->stats.resident_count) -#define pmap_phys_address(frame) ((vm_offset_t) (intel_ptob(frame))) -#define pmap_phys_to_frame(phys) ((int) (intel_btop(phys))) +#define pmap_resident_max(pmap) ((pmap)->stats.resident_max) #define pmap_copy(dst_pmap,src_pmap,dst_addr,len,src_addr) #define pmap_attribute(pmap,addr,size,attr,value) \ (KERN_INVALID_ADDRESS) #define pmap_attribute_cache_sync(addr,size,attr,value) \ (KERN_INVALID_ADDRESS) -#define pmap_sync_caches_phys(pa) \ - (KERN_INVALID_ADDRESS) + +#define MACHINE_PMAP_IS_EMPTY 1 +extern boolean_t pmap_is_empty(pmap_t pmap, + vm_map_offset_t start, + vm_map_offset_t end); + +#define MACHINE_BOOTSTRAPPTD 1 /* Static bootstrap page-tables */ + +kern_return_t +pmap_permissions_verify(pmap_t, vm_map_t, vm_offset_t, vm_offset_t); #endif /* ASSEMBLER */ + #endif /* _PMAP_MACHINE_ */ + + +#endif /* KERNEL_PRIVATE */