/*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2012 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@
*
* Machine-dependent structures for the physical map module.
*/
-
+#ifdef KERNEL_PRIVATE
#ifndef _PMAP_MACHINE_
#define _PMAP_MACHINE_ 1
#ifndef ASSEMBLER
-#include <platforms.h>
-#include <mp_v1_1.h>
#include <mach/kern_return.h>
#include <mach/machine/vm_types.h>
#include <mach/vm_statistics.h>
#include <mach/machine/vm_param.h>
#include <kern/kern_types.h>
-#include <kern/thread_act.h>
-#include <kern/lock.h>
+#include <kern/thread.h>
+#include <kern/simple_lock.h>
+#include <mach/branch_predicates.h>
+
+#include <i386/mp.h>
+#include <i386/proc_reg.h>
+
+#include <i386/pal_routines.h>
/*
* Define the generic in terms of the specific
* 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 __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)
+{
+ /*
+ * In the 32-bit kernel a compare-and-exchange loop was
+ * required to provide atomicity. For K64, life is easier:
+ */
+ *entryp = value;
+}
+/* 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))))
+
+#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)
+
+#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
/*
- * Convert kernel virtual address to linear address
+ * Pte related macros
*/
+#define KVADDR(pmi, pdpi, pdi, pti) \
+ ((vm_offset_t) \
+ ((uint64_t) -1 << 47) | \
+ ((uint64_t)(pmi) << PML4SHIFT) | \
+ ((uint64_t)(pdpi) << PDPTSHIFT) | \
+ ((uint64_t)(pdi) << PDESHIFT) | \
+ ((uint64_t)(pti) << PTESHIFT))
+
+/*
+ * Size of Kernel address space. This is the number of page table pages
+ * (4MB each) to use for the kernel. 256 pages == 1 Gigabyte.
+ * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
+ */
+#ifndef KVA_PAGES
+#define KVA_PAGES 1024
+#endif
+
+#ifndef NKPT
+#define NKPT 500 /* actual number of kernel page tables */
+#endif
+#ifndef NKPDE
+#define NKPDE (KVA_PAGES - 1) /* addressable number of page tables/pde's */
+#endif
+
-#define kvtolinear(a) ((a)+LINEAR_KERNEL_ADDRESS)
/*
* 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
/*
* 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 0x00000400ULL
+#define INTEL_PDPTE_NESTED 0x00000800ULL
+#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 INTEL_COMPRESSED (1ULL << 62) /* marker, for invalid PTE only -- ignored by hardware for both regular/EPT entries*/
+
+#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_INVEPT_SINGLE_CONTEXT 1
+
+
+#define INTEL_EPTP_AD 0x00000040ULL
+
+#define INTEL_EPT_READ 0x00000001ULL
+#define INTEL_EPT_WRITE 0x00000002ULL
+#define INTEL_EPT_EX 0x00000004ULL
+#define INTEL_EPT_IPTA 0x00000040ULL
+#define INTEL_EPT_PS 0x00000080ULL
+#define INTEL_EPT_REF 0x00000100ULL
+#define INTEL_EPT_MOD 0x00000200ULL
+
+#define INTEL_EPT_CACHE_MASK 0x00000038ULL
+#define INTEL_EPT_NCACHE 0x00000000ULL
+#define INTEL_EPT_WC 0x00000008ULL
+#define INTEL_EPT_WTHRU 0x00000020ULL
+#define INTEL_EPT_WP 0x00000028ULL
+#define INTEL_EPT_WB 0x00000030ULL
+
/*
- * Convert page table entry to kernel virtual address
+ * Routines to filter correct bits depending on the pmap type
*/
-#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 */
+static inline pt_entry_t
+pte_remove_ex(pt_entry_t pte, boolean_t is_ept)
+{
+ if (__probable(!is_ept)) {
+ return (pte | INTEL_PTE_NX);
+ }
-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 */
-};
+ return (pte & (~INTEL_EPT_EX));
+}
-/*
- * 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.
+static inline pt_entry_t
+pte_set_ex(pt_entry_t pte, boolean_t is_ept)
+{
+ if (__probable(!is_ept)) {
+ return (pte & (~INTEL_PTE_NX));
+ }
+
+ return (pte | INTEL_EPT_EX);
+}
+
+static inline pt_entry_t
+physmap_refmod_to_ept(pt_entry_t physmap_pte)
+{
+ pt_entry_t ept_pte = 0;
+
+ if (physmap_pte & INTEL_PTE_MOD) {
+ ept_pte |= INTEL_EPT_MOD;
+ }
+
+ if (physmap_pte & INTEL_PTE_REF) {
+ ept_pte |= INTEL_EPT_REF;
+ }
+
+ return ept_pte;
+}
+
+static inline pt_entry_t
+ept_refmod_to_physmap(pt_entry_t ept_pte)
+{
+ pt_entry_t physmap_pte = 0;
+
+ assert((ept_pte & ~(INTEL_EPT_REF | INTEL_EPT_MOD)) == 0);
+
+ if (ept_pte & INTEL_EPT_REF) {
+ physmap_pte |= INTEL_PTE_REF;
+ }
+
+ if (ept_pte & INTEL_EPT_MOD) {
+ physmap_pte |= INTEL_PTE_MOD;
+ }
+
+ return physmap_pte;
+}
+
+/*
+ * Note: Not all Intel processors support EPT referenced access and dirty bits.
+ * During pmap_init() we check the VMX capability for the current hardware
+ * and update this variable accordingly.
*/
+extern boolean_t pmap_ept_support_ad;
+
+#define PTE_VALID_MASK(is_ept) ((is_ept) ? (INTEL_EPT_READ | INTEL_EPT_WRITE | INTEL_EPT_EX) : INTEL_PTE_VALID)
+#define PTE_READ(is_ept) ((is_ept) ? INTEL_EPT_READ : INTEL_PTE_VALID)
+#define PTE_WRITE(is_ept) ((is_ept) ? INTEL_EPT_WRITE : INTEL_PTE_WRITE)
+#define PTE_PS INTEL_PTE_PS
+#define PTE_COMPRESSED INTEL_COMPRESSED
+#define PTE_NCACHE(is_ept) ((is_ept) ? INTEL_EPT_NCACHE : INTEL_PTE_NCACHE)
+#define PTE_WTHRU(is_ept) ((is_ept) ? INTEL_EPT_WTHRU : INTEL_PTE_WTHRU)
+#define PTE_REF(is_ept) ((is_ept) ? INTEL_EPT_REF : INTEL_PTE_REF)
+#define PTE_MOD(is_ept) ((is_ept) ? INTEL_EPT_MOD : INTEL_PTE_MOD)
+#define PTE_WIRED INTEL_PTE_WIRED
+
+
+#define PMAP_DEFAULT_CACHE 0
+#define PMAP_INHIBIT_CACHE 1
+#define PMAP_GUARDED_CACHE 2
+#define PMAP_ACTIVATE_CACHE 4
+#define PMAP_NO_GUARD_CACHE 8
-extern struct pmap *real_pmap[NCPUS];
+#ifndef ASSEMBLER
+
+#include <sys/queue.h>
-#include <i386/proc_reg.h>
/*
- * 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.
+ * Address of current and alternate address space page table maps
+ * and directories.
*/
-#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 */
+extern pt_entry_t *PTmap;
+extern pdpt_entry_t *IdlePDPT;
+extern pml4_entry_t *IdlePML4;
+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;
-#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));
}
-#if NCPUS > 1
+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;
+}
+
+#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.
+ * And non-boot processor's GDT aliases likewise (skipping LOWGLOBAL_ALIAS)
+ * The low global vector page is mapped at a fixed alias also.
*/
-extern cpu_set cpus_active;
+#define MASTER_IDT_ALIAS (VM_MIN_KERNEL_ADDRESS + 0x0000)
+#define MASTER_GDT_ALIAS (VM_MIN_KERNEL_ADDRESS + 0x1000)
+#define LOWGLOBAL_ALIAS (VM_MIN_KERNEL_ADDRESS + 0x2000)
+#define CPU_GDT_ALIAS(_cpu) (LOWGLOBAL_ALIAS + (0x1000*(_cpu)))
+
+#endif /*__x86_64__ */
+
+#include <vm/vm_page.h>
/*
- * List of cpus that are idle, but still operating, and will want
- * to see any kernel pmap updates when they become active.
+ * 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.
*/
-extern cpu_set cpus_idle;
+struct pmap {
+ decl_simple_lock_data(,lock) /* lock on map */
+ pmap_paddr_t pm_cr3; /* physical addr */
+ pmap_paddr_t pm_eptp; /* EPTP */
+ boolean_t pm_shared;
+ pd_entry_t *dirbase; /* page directory pointer */
+ 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 MAX_CPUS /* 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 */
+};
+
+static inline boolean_t
+is_ept_pmap(pmap_t p)
+{
+ if (__probable(p->pm_cr3 != 0)) {
+ assert(p->pm_eptp == 0);
+ return FALSE;
+ }
+
+ assert(p->pm_eptp != 0);
+
+ return TRUE;
+}
+
+void hv_ept_pmap_create(void **ept_pmap, void **eptp);
+
+#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; /* first page of this region */
+ ppnum_t alloc_up; /* pages below this one have been "stolen" */
+ ppnum_t alloc_down; /* pages above this one have been "stolen" */
+ ppnum_t end; /* last page of this region */
+ 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 <i386/pmap_pcid.h>
+
+static inline void
+set_dirbase(pmap_t tpmap, __unused thread_t thread, int my_cpu) {
+ int ccpu = my_cpu;
+ cpu_datap(ccpu)->cpu_task_cr3 = tpmap->pm_cr3;
+ cpu_datap(ccpu)->cpu_task_map = tpmap->pm_task_map;
+ /*
+ * 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);
+ }
+}
/*
* 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);
+/*
+ * Get cache attributes (as pagetable bits) for the specified phys page
+ */
+extern unsigned pmap_get_cache_attributes(ppnum_t, boolean_t is_ept);
+#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);
+
+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 <kern/spl.h>
-/*
- * 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, my_cpu) { \
+ register pmap_t tpmap; \
+ \
+ tpmap = vm_map_pmap(map); \
+ set_dirbase(tpmap, thread, my_cpu); \
}
-#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); \
-}
+#if defined(__x86_64__)
+#define PMAP_DEACTIVATE_MAP(map, thread, ccpu) \
+ pmap_assert(pmap_pcid_ncpus ? (pcid_for_pmap_cpu_tuple(map->pmap, ccpu) == (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); \
+#define PMAP_SWITCH_CONTEXT(old_th, new_th, my_cpu) { \
+ \
+ pmap_assert(ml_get_interrupts_enabled() == FALSE); \
+ if (old_th->map != new_th->map) { \
+ PMAP_DEACTIVATE_MAP(old_th->map, old_th, my_cpu); \
+ PMAP_ACTIVATE_MAP(new_th->map, new_th, my_cpu); \
} \
}
-#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) { \
+#if NCOPY_WINDOWS > 0
+#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, th); \
splx(spl); \
+ inval_copy_windows(th); \
}
-
-#define PMAP_SWITCH_CONTEXT(old_th, new_th, my_cpu) { \
- spl_t spl; \
- \
- 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); \
- } \
-}
-
+#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, my_cpu); \
th->map = new_map; \
- PMAP_ACTIVATE_MAP(th->map, my_cpu); \
+ PMAP_ACTIVATE_MAP(th->map, th, my_cpu); \
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.
+ */
#define MARK_CPU_IDLE(my_cpu) { \
- /* \
- * 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. \
- */ \
- int s = splhigh(); \
- i_bit_set((my_cpu), &cpus_idle); \
- i_bit_clear((my_cpu), &cpus_active); \
- splx(s); \
- set_led(my_cpu); \
+ assert(ml_get_interrupts_enabled() == FALSE); \
+ CPU_CR3_MARK_INACTIVE(); \
+ mfence(); \
}
-#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. \
* set assures that we will receive another update \
* interrupt if this happens. \
*/ \
- i_bit_clear((my_cpu), &cpus_idle); \
+ CPU_CR3_MARK_ACTIVE(); \
+ 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); \
- } \
-}
-
-#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); \
+ if (current_cpu_datap()->cpu_tlb_invalid) \
+ process_pmap_updates(); \
}
-#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_compressed(pmap) ((pmap)->stats.compressed)
#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 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 */
projects / apple / xnu.git / blobdiff