-
/*
- * Copyright (c) 2000-2007 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- *
+ *
* 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
* 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.
- *
+ *
* 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,
* 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_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* Mach Operating System
* Copyright (c) 1991,1990,1989,1988 Carnegie Mellon University
* All Rights Reserved.
- *
+ *
* Permission to use, copy, modify and distribute this software and its
* documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
- *
+ *
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
* ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
- *
+ *
* Carnegie Mellon requests users of this software to return to
- *
+ *
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
- *
+ *
* any improvements or extensions that they make and grant Carnegie Mellon
* the rights to redistribute these changes.
*/
*/
#include <string.h>
-#include <norma_vm.h>
-#include <mach_kdb.h>
#include <mach_ldebug.h>
#include <libkern/OSAtomic.h>
#include <kern/thread.h>
#include <kern/zalloc.h>
#include <kern/queue.h>
+#include <kern/ledger.h>
+#include <kern/mach_param.h>
-#include <kern/lock.h>
#include <kern/kalloc.h>
#include <kern/spl.h>
#include <mach/machine/vm_param.h>
#include <machine/thread.h>
-#include <kern/misc_protos.h> /* prototyping */
+#include <kern/misc_protos.h> /* prototyping */
#include <i386/misc_protos.h>
+#include <i386/i386_lowmem.h>
#include <x86_64/lowglobals.h>
#include <i386/cpuid.h>
#include <i386/proc_reg.h>
#include <i386/tsc.h>
#include <i386/pmap_internal.h>
-
-#if MACH_KDB
-#include <ddb/db_command.h>
-#include <ddb/db_output.h>
-#include <ddb/db_sym.h>
-#include <ddb/db_print.h>
-#endif /* MACH_KDB */
+#include <i386/pmap_pcid.h>
+#if CONFIG_VMX
+#include <i386/vmx/vmx_cpu.h>
+#endif
#include <vm/vm_protos.h>
+#include <san/kasan.h>
#include <i386/mp.h>
#include <i386/mp_desc.h>
+#include <libkern/kernel_mach_header.h>
+#include <pexpert/i386/efi.h>
-/* #define DEBUGINTERRUPTS 1 uncomment to ensure pmap callers have interrupts enabled */
-#ifdef DEBUGINTERRUPTS
-#define pmap_intr_assert() { \
- if (processor_avail_count > 1 && !ml_get_interrupts_enabled()) \
- panic("pmap interrupt assert %s, %d",__FILE__, __LINE__); \
-}
-#else
-#define pmap_intr_assert()
-#endif
+#if MACH_ASSERT
+int pmap_stats_assert = 1;
+#endif /* MACH_ASSERT */
#ifdef IWANTTODEBUG
-#undef DEBUG
+#undef DEBUG
#define DEBUG 1
#define POSTCODE_DELAY 1
#include <i386/postcode.h>
#endif /* IWANTTODEBUG */
-boolean_t pmap_trace = FALSE;
-
-#if PMAP_DBG
+#ifdef PMAP_DEBUG
#define DBG(x...) kprintf("DBG: " x)
#else
#define DBG(x...)
#endif
-
-boolean_t no_shared_cr3 = DEBUG; /* TRUE for DEBUG by default */
-
-/*
- * Forward declarations for internal functions.
- */
-
-void pmap_remove_range(
- pmap_t pmap,
- vm_map_offset_t va,
- pt_entry_t *spte,
- pt_entry_t *epte);
-
-void phys_attribute_clear(
- ppnum_t phys,
- int bits);
-
-int phys_attribute_test(
- ppnum_t phys,
- int bits);
-
-void phys_attribute_set(
- ppnum_t phys,
- int bits);
-
-void pmap_set_reference(
- ppnum_t pn);
-
-boolean_t phys_page_exists(
- ppnum_t pn);
-
-
-int nx_enabled = 1; /* enable no-execute protection */
-int allow_data_exec = VM_ABI_32; /* 32-bit apps may execute data by default, 64-bit apps may not */
-int allow_stack_exec = 0; /* No apps may execute from the stack by default */
-
-const boolean_t cpu_64bit = TRUE; /* Mais oui! */
-
-/*
- * when spinning through pmap_remove
- * ensure that we don't spend too much
- * time with preemption disabled.
- * I'm setting the current threshold
- * to 20us
+/* Compile time assert to ensure adjacency/alignment of per-CPU data fields used
+ * in the trampolines for kernel/user boundary TLB coherency.
*/
-#define MAX_PREEMPTION_LATENCY_NS 20000
-
-uint64_t max_preemption_latency_tsc = 0;
-
-
-/*
- * Private data structures.
- */
-
-/*
- * For each vm_page_t, there is a list of all currently
- * valid virtual mappings of that page. An entry is
- * a pv_rooted_entry_t; the list is the pv_table.
- *
- * N.B. with the new combo rooted/hashed scheme it is
- * only possibly to remove individual non-rooted entries
- * if they are found via the hashed chains as there is no
- * way to unlink the singly linked hashed entries if navigated to
- * via the queue list off the rooted entries. Think of it as
- * hash/walk/pull, keeping track of the prev pointer while walking
- * the singly linked hash list. All of this is to save memory and
- * keep both types of pv_entries as small as possible.
- */
-
-/*
-
-PV HASHING Changes - JK 1/2007
-
-Pve's establish physical to virtual mappings. These are used for aliasing of a
-physical page to (potentially many) virtual addresses within pmaps. In the
-previous implementation the structure of the pv_entries (each 16 bytes in size) was
-
-typedef struct pv_entry {
- struct pv_entry_t next;
- pmap_t pmap;
- vm_map_offset_t va;
-} *pv_entry_t;
-
-An initial array of these is created at boot time, one per physical page of
-memory, indexed by the physical page number. Additionally, a pool of entries
-is created from a pv_zone to be used as needed by pmap_enter() when it is
-creating new mappings. Originally, we kept this pool around because the code
-in pmap_enter() was unable to block if it needed an entry and none were
-available - we'd panic. Some time ago I restructured the pmap_enter() code
-so that for user pmaps it can block while zalloc'ing a pv structure and restart,
-removing a panic from the code (in the case of the kernel pmap we cannot block
-and still panic, so, we keep a separate hot pool for use only on kernel pmaps).
-The pool has not been removed since there is a large performance gain keeping
-freed pv's around for reuse and not suffering the overhead of zalloc for every
-new pv we need.
-
-As pmap_enter() created new mappings it linked the new pve's for them off the
-fixed pv array for that ppn (off the next pointer). These pve's are accessed
-for several operations, one of them being address space teardown. In that case,
-we basically do this
-
- for (every page/pte in the space) {
- calc pve_ptr from the ppn in the pte
- for (every pv in the list for the ppn) {
- if (this pv is for this pmap/vaddr) {
- do housekeeping
- unlink/free the pv
- }
- }
- }
-
-The problem arose when we were running, say 8000 (or even 2000) apache or
-other processes and one or all terminate. The list hanging off each pv array
-entry could have thousands of entries. We were continuously linearly searching
-each of these lists as we stepped through the address space we were tearing
-down. Because of the locks we hold, likely taking a cache miss for each node,
-and interrupt disabling for MP issues the system became completely unresponsive
-for many seconds while we did this.
-
-Realizing that pve's are accessed in two distinct ways (linearly running the
-list by ppn for operations like pmap_page_protect and finding and
-modifying/removing a single pve as part of pmap_enter processing) has led to
-modifying the pve structures and databases.
-
-There are now two types of pve structures. A "rooted" structure which is
-basically the original structure accessed in an array by ppn, and a ''hashed''
-structure accessed on a hash list via a hash of [pmap, vaddr]. These have been
-designed with the two goals of minimizing wired memory and making the lookup of
-a ppn faster. Since a vast majority of pages in the system are not aliased
-and hence represented by a single pv entry I've kept the rooted entry size as
-small as possible because there is one of these dedicated for every physical
-page of memory. The hashed pve's are larger due to the addition of the hash
-link and the ppn entry needed for matching while running the hash list to find
-the entry we are looking for. This way, only systems that have lots of
-aliasing (like 2000+ httpd procs) will pay the extra memory price. Both
-structures have the same first three fields allowing some simplification in
-the code.
-
-They have these shapes
-
-typedef struct pv_rooted_entry {
- queue_head_t qlink;
- vm_map_offset_t va;
- pmap_t pmap;
-} *pv_rooted_entry_t;
-
-
-typedef struct pv_hashed_entry {
- queue_head_t qlink;
- vm_map_offset_t va;
- pmap_t pmap;
- ppnum_t ppn;
- struct pv_hashed_entry *nexth;
-} *pv_hashed_entry_t;
-
-The main flow difference is that the code is now aware of the rooted entry and
-the hashed entries. Code that runs the pv list still starts with the rooted
-entry and then continues down the qlink onto the hashed entries. Code that is
-looking up a specific pv entry first checks the rooted entry and then hashes
-and runs the hash list for the match. The hash list lengths are much smaller
-than the original pv lists that contained all aliases for the specific ppn.
-
-*/
-
-typedef struct pv_rooted_entry {
- /* first three entries must match pv_hashed_entry_t */
- queue_head_t qlink;
- vm_map_offset_t va; /* virtual address for mapping */
- pmap_t pmap; /* pmap where mapping lies */
-} *pv_rooted_entry_t;
-
-#define PV_ROOTED_ENTRY_NULL ((pv_rooted_entry_t) 0)
-
-pv_rooted_entry_t pv_head_table; /* array of entries, one per page */
-
-typedef struct pv_hashed_entry {
- /* first three entries must match pv_rooted_entry_t */
- queue_head_t qlink;
- vm_map_offset_t va;
- pmap_t pmap;
- ppnum_t ppn;
- struct pv_hashed_entry *nexth;
-} *pv_hashed_entry_t;
-
-#define PV_HASHED_ENTRY_NULL ((pv_hashed_entry_t)0)
-
-#define NPVHASH 4095 /* MUST BE 2^N - 1 */
-pv_hashed_entry_t *pv_hash_table; /* hash lists */
+char pmap_cpu_data_assert[(((offsetof(cpu_data_t, cpu_tlb_invalid) - offsetof(cpu_data_t, cpu_active_cr3)) == 8) && (offsetof(cpu_data_t, cpu_active_cr3) % 64 == 0)) ? 1 : -1];
+boolean_t pmap_trace = FALSE;
-uint32_t npvhash = 0;
+boolean_t no_shared_cr3 = DEBUG; /* TRUE for DEBUG by default */
-//#define PV_DEBUG 1 /* uncomment to enable some PV debugging code */
-#ifdef PV_DEBUG
-#define CHK_NPVHASH() if(0 == npvhash) panic("npvhash uninitialized");
+#if DEVELOPMENT || DEBUG
+int nx_enabled = 1; /* enable no-execute protection -- set during boot */
#else
-#define CHK_NPVHASH(x)
+const int nx_enabled = 1;
#endif
-pv_hashed_entry_t pv_hashed_free_list = PV_HASHED_ENTRY_NULL;
-pv_hashed_entry_t pv_hashed_kern_free_list = PV_HASHED_ENTRY_NULL;
-decl_simple_lock_data(,pv_hashed_free_list_lock)
-decl_simple_lock_data(,pv_hashed_kern_free_list_lock)
-decl_simple_lock_data(,pv_hash_table_lock)
-
-int pv_hashed_free_count = 0;
-int pv_hashed_kern_free_count = 0;
-#define PV_HASHED_LOW_WATER_MARK 5000
-#define PV_HASHED_KERN_LOW_WATER_MARK 100
-#define PV_HASHED_ALLOC_CHUNK 2000
-#define PV_HASHED_KERN_ALLOC_CHUNK 50
-thread_call_t mapping_adjust_call;
-static thread_call_data_t mapping_adjust_call_data;
-uint32_t mappingrecurse = 0;
-
-#define PV_HASHED_ALLOC(pvh_e) { \
- simple_lock(&pv_hashed_free_list_lock); \
- if ((pvh_e = pv_hashed_free_list) != 0) { \
- pv_hashed_free_list = (pv_hashed_entry_t)pvh_e->qlink.next; \
- pv_hashed_free_count--; \
- if (pv_hashed_free_count < PV_HASHED_LOW_WATER_MARK) \
- if (hw_compare_and_store(0,1,(u_int *)&mappingrecurse)) \
- thread_call_enter(mapping_adjust_call); \
- } \
- simple_unlock(&pv_hashed_free_list_lock); \
-}
-
-#define PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt) { \
- simple_lock(&pv_hashed_free_list_lock); \
- pvh_et->qlink.next = (queue_entry_t)pv_hashed_free_list; \
- pv_hashed_free_list = pvh_eh; \
- pv_hashed_free_count += pv_cnt; \
- simple_unlock(&pv_hashed_free_list_lock); \
-}
-
-#define PV_HASHED_KERN_ALLOC(pvh_e) { \
- simple_lock(&pv_hashed_kern_free_list_lock); \
- if ((pvh_e = pv_hashed_kern_free_list) != 0) { \
- pv_hashed_kern_free_list = (pv_hashed_entry_t)pvh_e->qlink.next; \
- pv_hashed_kern_free_count--; \
- if (pv_hashed_kern_free_count < PV_HASHED_KERN_LOW_WATER_MARK)\
- if (hw_compare_and_store(0,1,(u_int *)&mappingrecurse)) \
- thread_call_enter(mapping_adjust_call); \
- } \
- simple_unlock(&pv_hashed_kern_free_list_lock); \
-}
+#if DEBUG || DEVELOPMENT
+int allow_data_exec = VM_ABI_32; /* 32-bit apps may execute data by default, 64-bit apps may not */
+int allow_stack_exec = 0; /* No apps may execute from the stack by default */
+#else /* DEBUG || DEVELOPMENT */
+const int allow_data_exec = VM_ABI_32; /* 32-bit apps may execute data by default, 64-bit apps may not */
+const int allow_stack_exec = 0; /* No apps may execute from the stack by default */
+#endif /* DEBUG || DEVELOPMENT */
-#define PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt) { \
- simple_lock(&pv_hashed_kern_free_list_lock); \
- pvh_et->qlink.next = (queue_entry_t)pv_hashed_kern_free_list; \
- pv_hashed_kern_free_list = pvh_eh; \
- pv_hashed_kern_free_count += pv_cnt; \
- simple_unlock(&pv_hashed_kern_free_list_lock); \
-}
+uint64_t max_preemption_latency_tsc = 0;
-zone_t pv_hashed_list_zone; /* zone of pv_hashed_entry structures */
+pv_hashed_entry_t *pv_hash_table; /* hash lists */
-static zone_t pdpt_zone;
+uint32_t npvhashmask = 0, npvhashbuckets = 0;
-/*
- * Each entry in the pv_head_table is locked by a bit in the
- * pv_lock_table. The lock bits are accessed by the physical
- * address of the page they lock.
- */
+pv_hashed_entry_t pv_hashed_free_list = PV_HASHED_ENTRY_NULL;
+pv_hashed_entry_t pv_hashed_kern_free_list = PV_HASHED_ENTRY_NULL;
+decl_simple_lock_data(, pv_hashed_free_list_lock)
+decl_simple_lock_data(, pv_hashed_kern_free_list_lock)
+decl_simple_lock_data(, pv_hash_table_lock)
-char *pv_lock_table; /* pointer to array of bits */
-#define pv_lock_table_size(n) (((n)+BYTE_SIZE-1)/BYTE_SIZE)
+decl_simple_lock_data(, phys_backup_lock)
-char *pv_hash_lock_table;
-#define pv_hash_lock_table_size(n) (((n)+BYTE_SIZE-1)/BYTE_SIZE)
+zone_t pv_hashed_list_zone; /* zone of pv_hashed_entry structures */
/*
* First and last physical addresses that we maintain any information
* for. Initialized to zero so that pmap operations done before
* pmap_init won't touch any non-existent structures.
*/
-boolean_t pmap_initialized = FALSE;/* Has pmap_init completed? */
-
-static struct vm_object kptobj_object_store;
-static struct vm_object kpml4obj_object_store;
-static struct vm_object kpdptobj_object_store;
-
-/*
- * Index into pv_head table, its lock bits, and the modify/reference and managed bits
- */
-
-#define pa_index(pa) (i386_btop(pa))
-#define ppn_to_pai(ppn) ((int)ppn)
-
-#define pai_to_pvh(pai) (&pv_head_table[pai])
-#define lock_pvh_pai(pai) bit_lock(pai, (void *)pv_lock_table)
-#define unlock_pvh_pai(pai) bit_unlock(pai, (void *)pv_lock_table)
-
-static inline uint32_t
-pvhashidx(pmap_t pmap, vm_offset_t va)
-{
- return ((uint32_t)(uint64_t)pmap ^
- ((uint32_t)((uint64_t)va >> PAGE_SHIFT) & 0xFFFFFFFF)) &
- npvhash;
-}
-#define pvhash(idx) (&pv_hash_table[idx])
+boolean_t pmap_initialized = FALSE;/* Has pmap_init completed? */
-#define lock_hash_hash(hash) bit_lock(hash, (void *)pv_hash_lock_table)
-#define unlock_hash_hash(hash) bit_unlock(hash, (void *)pv_hash_lock_table)
+static struct vm_object kptobj_object_store __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT)));
+static struct vm_object kpml4obj_object_store __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT)));
+static struct vm_object kpdptobj_object_store __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT)));
/*
* Array of physical page attribites for managed pages.
* One byte per physical page.
*/
-char *pmap_phys_attributes;
-unsigned int last_managed_page = 0;
-#define IS_MANAGED_PAGE(x) \
- ((unsigned int)(x) <= last_managed_page && \
- (pmap_phys_attributes[x] & PHYS_MANAGED))
-
-/*
- * Physical page attributes. Copy bits from PTE definition.
- */
-#define PHYS_MODIFIED INTEL_PTE_MOD /* page modified */
-#define PHYS_REFERENCED INTEL_PTE_REF /* page referenced */
-#define PHYS_MANAGED INTEL_PTE_VALID /* page is managed */
-
-/*
- * Amount of virtual memory mapped by one
- * page-directory entry.
- */
-#define PDE_MAPPED_SIZE (pdetova(1))
-uint64_t pde_mapped_size = PDE_MAPPED_SIZE;
-
-/*
- * Locking and TLB invalidation
- */
-
-/*
- * Locking Protocols: (changed 2/2007 JK)
- *
- * There are two structures in the pmap module that need locking:
- * the pmaps themselves, and the per-page pv_lists (which are locked
- * by locking the pv_lock_table entry that corresponds to the pv_head
- * for the list in question.) Most routines want to lock a pmap and
- * then do operations in it that require pv_list locking -- however
- * pmap_remove_all and pmap_copy_on_write operate on a physical page
- * basis and want to do the locking in the reverse order, i.e. lock
- * a pv_list and then go through all the pmaps referenced by that list.
- *
- * The system wide pmap lock has been removed. Now, paths take a lock
- * on the pmap before changing its 'shape' and the reverse order lockers
- * (coming in by phys ppn) take a lock on the corresponding pv and then
- * retest to be sure nothing changed during the window before they locked
- * and can then run up/down the pv lists holding the list lock. This also
- * lets the pmap layer run (nearly completely) interrupt enabled, unlike
- * previously.
- */
-
-/*
- * PV locking
- */
-
-#define LOCK_PVH(index) { \
- mp_disable_preemption(); \
- lock_pvh_pai(index); \
-}
-
-#define UNLOCK_PVH(index) { \
- unlock_pvh_pai(index); \
- mp_enable_preemption(); \
-}
-/*
- * PV hash locking
- */
-
-#define LOCK_PV_HASH(hash) lock_hash_hash(hash)
-#define UNLOCK_PV_HASH(hash) unlock_hash_hash(hash)
+char *pmap_phys_attributes;
+ppnum_t last_managed_page = 0;
unsigned pmap_memory_region_count;
unsigned pmap_memory_region_current;
/*
* Other useful macros.
*/
-#define current_pmap() (vm_map_pmap(current_thread()->map))
-
-struct pmap kernel_pmap_store;
-pmap_t kernel_pmap;
-
-pd_entry_t high_shared_pde;
-pd_entry_t commpage64_pde;
-
-struct zone *pmap_zone; /* zone of pmap structures */
-
-int pmap_debug = 0; /* flag for debugging prints */
-
-unsigned int inuse_ptepages_count = 0;
-
-addr64_t kernel64_cr3;
-
-/*
- * Pmap cache. Cache is threaded through ref_count field of pmap.
- * Max will eventually be constant -- variable for experimentation.
- */
-int pmap_cache_max = 32;
-int pmap_alloc_chunk = 8;
-pmap_t pmap_cache_list;
-int pmap_cache_count;
-decl_simple_lock_data(,pmap_cache_lock)
-
-extern char end;
-
-static int nkpt;
-
-pt_entry_t *DMAP1, *DMAP2;
-caddr_t DADDR1;
-caddr_t DADDR2;
-
-/*
- * unlinks the pv_hashed_entry_t pvh from the singly linked hash chain.
- * properly deals with the anchor.
- * must be called with the hash locked, does not unlock it
- */
-
-static inline void
-pmap_pvh_unlink(pv_hashed_entry_t pvh)
-{
- pv_hashed_entry_t curh;
- pv_hashed_entry_t *pprevh;
- int pvhash_idx;
-
- CHK_NPVHASH();
- pvhash_idx = pvhashidx(pvh->pmap, pvh->va);
+#define current_pmap() (vm_map_pmap(current_thread()->map))
- pprevh = pvhash(pvhash_idx);
+struct pmap kernel_pmap_store;
+pmap_t kernel_pmap;
-#if PV_DEBUG
- if (NULL == *pprevh)
- panic("pvh_unlink null anchor"); /* JK DEBUG */
-#endif
- curh = *pprevh;
-
- while (PV_HASHED_ENTRY_NULL != curh) {
- if (pvh == curh)
- break;
- pprevh = &curh->nexth;
- curh = curh->nexth;
- }
- if (PV_HASHED_ENTRY_NULL == curh) panic("pmap_pvh_unlink no pvh");
- *pprevh = pvh->nexth;
- return;
-}
+struct zone *pmap_zone; /* zone of pmap structures */
-static inline void
-pv_hash_add(pv_hashed_entry_t pvh_e,
- pv_rooted_entry_t pv_h)
-{
- pv_hashed_entry_t *hashp;
- int pvhash_idx;
-
- CHK_NPVHASH();
- pvhash_idx = pvhashidx(pvh_e->pmap, pvh_e->va);
- LOCK_PV_HASH(pvhash_idx);
- insque(&pvh_e->qlink, &pv_h->qlink);
- hashp = pvhash(pvhash_idx);
-#if PV_DEBUG
- if (NULL==hashp)
- panic("pv_hash_add(%p) null hash bucket", pvh_e);
-#endif
- pvh_e->nexth = *hashp;
- *hashp = pvh_e;
- UNLOCK_PV_HASH(pvhash_idx);
-}
+struct zone *pmap_anchor_zone;
+struct zone *pmap_uanchor_zone;
+int pmap_debug = 0; /* flag for debugging prints */
-static inline void
-pv_hash_remove(pv_hashed_entry_t pvh_e)
-{
- int pvhash_idx;
+unsigned int inuse_ptepages_count = 0;
+long long alloc_ptepages_count __attribute__((aligned(8))) = 0; /* aligned for atomic access */
+unsigned int bootstrap_wired_pages = 0;
+int pt_fake_zone_index = -1;
- CHK_NPVHASH();
- pvhash_idx = pvhashidx(pvh_e->pmap,pvh_e->va);
- LOCK_PV_HASH(pvhash_idx);
- remque(&pvh_e->qlink);
- pmap_pvh_unlink(pvh_e);
- UNLOCK_PV_HASH(pvhash_idx);
-}
+extern long NMIPI_acks;
-/*
- * Remove pv list entry.
- * Called with pv_head_table entry locked.
- * Returns pv entry to be freed (or NULL).
- */
-static inline pv_hashed_entry_t
-pmap_pv_remove(pmap_t pmap,
- vm_map_offset_t vaddr,
- ppnum_t ppn)
-{
- pv_hashed_entry_t pvh_e;
- pv_rooted_entry_t pv_h;
- pv_hashed_entry_t *pprevh;
- int pvhash_idx;
- uint32_t pv_cnt;
-
- pvh_e = PV_HASHED_ENTRY_NULL;
- pv_h = pai_to_pvh(ppn_to_pai(ppn));
- if (pv_h->pmap == PMAP_NULL)
- panic("pmap_pv_remove(%p,%llu,%u): null pv_list!",
- pmap, vaddr, ppn);
-
- if (pv_h->va == vaddr && pv_h->pmap == pmap) {
- /*
- * Header is the pv_rooted_entry.
- * We can't free that. If there is a queued
- * entry after this one we remove that
- * from the ppn queue, we remove it from the hash chain
- * and copy it to the rooted entry. Then free it instead.
- */
- pvh_e = (pv_hashed_entry_t) queue_next(&pv_h->qlink);
- if (pv_h != (pv_rooted_entry_t) pvh_e) {
- /*
- * Entry queued to root, remove this from hash
- * and install as nem root.
- */
- CHK_NPVHASH();
- pvhash_idx = pvhashidx(pvh_e->pmap, pvh_e->va);
- LOCK_PV_HASH(pvhash_idx);
- remque(&pvh_e->qlink);
- pprevh = pvhash(pvhash_idx);
- if (PV_HASHED_ENTRY_NULL == *pprevh) {
- panic("pmap_pv_remove(%p,%llu,%u): "
- "empty hash, removing rooted",
- pmap, vaddr, ppn);
- }
- pmap_pvh_unlink(pvh_e);
- UNLOCK_PV_HASH(pvhash_idx);
- pv_h->pmap = pvh_e->pmap;
- pv_h->va = pvh_e->va; /* dispose of pvh_e */
- } else {
- /* none queued after rooted */
- pv_h->pmap = PMAP_NULL;
- pvh_e = PV_HASHED_ENTRY_NULL;
- }
- } else {
- /*
- * not removing rooted pv. find it on hash chain, remove from
- * ppn queue and hash chain and free it
- */
- CHK_NPVHASH();
- pvhash_idx = pvhashidx(pmap, vaddr);
- LOCK_PV_HASH(pvhash_idx);
- pprevh = pvhash(pvhash_idx);
- if (PV_HASHED_ENTRY_NULL == *pprevh) {
- panic("pmap_pv_remove(%p,%llu,%u): empty hash",
- pmap, vaddr, ppn);
- }
- pvh_e = *pprevh;
- pmap_pv_hashlist_walks++;
- pv_cnt = 0;
- while (PV_HASHED_ENTRY_NULL != pvh_e) {
- pv_cnt++;
- if (pvh_e->pmap == pmap &&
- pvh_e->va == vaddr &&
- pvh_e->ppn == ppn)
- break;
- pprevh = &pvh_e->nexth;
- pvh_e = pvh_e->nexth;
- }
- if (PV_HASHED_ENTRY_NULL == pvh_e)
- panic("pmap_pv_remove(%p,%llu,%u): pv not on hash",
- pmap, vaddr, ppn);
- pmap_pv_hashlist_cnts += pv_cnt;
- if (pmap_pv_hashlist_max < pv_cnt)
- pmap_pv_hashlist_max = pv_cnt;
- *pprevh = pvh_e->nexth;
- remque(&pvh_e->qlink);
- UNLOCK_PV_HASH(pvhash_idx);
- }
+boolean_t kernel_text_ps_4K = TRUE;
- return pvh_e;
-}
+extern char end;
-/*
- * for legacy, returns the address of the pde entry.
- * for 64 bit, causes the pdpt page containing the pde entry to be mapped,
- * then returns the mapped address of the pde entry in that page
- */
-pd_entry_t *
-pmap_pde(pmap_t m, vm_map_offset_t v)
-{
- pd_entry_t *pde;
+static int nkpt;
- assert(m);
-#if 0
- if (m == kernel_pmap)
- pde = (&((m)->dirbase[(vm_offset_t)(v) >> PDESHIFT]));
- else
+#if DEVELOPMENT || DEBUG
+boolean_t pmap_disable_kheap_nx = FALSE;
+boolean_t pmap_disable_kstack_nx = FALSE;
+boolean_t wpkernel = TRUE;
+#else
+const boolean_t wpkernel = TRUE;
#endif
- pde = pmap64_pde(m, v);
-
- return pde;
-}
-
-/*
- * the single pml4 page per pmap is allocated at pmap create time and exists
- * for the duration of the pmap. we allocate this page in kernel vm.
- * this returns the address of the requested pml4 entry in the top level page.
- */
-static inline
-pml4_entry_t *
-pmap64_pml4(pmap_t pmap, vm_map_offset_t vaddr)
-{
- return &pmap->pm_pml4[(vaddr >> PML4SHIFT) & (NPML4PG-1)];
-}
-
-/*
- * maps in the pml4 page, if any, containing the pdpt entry requested
- * and returns the address of the pdpt entry in that mapped page
- */
-pdpt_entry_t *
-pmap64_pdpt(pmap_t pmap, vm_map_offset_t vaddr)
-{
- pml4_entry_t newpf;
- pml4_entry_t *pml4;
-
- assert(pmap);
- if ((vaddr > 0x00007FFFFFFFFFFFULL) &&
- (vaddr < 0xFFFF800000000000ULL)) {
- return (0);
- }
- pml4 = pmap64_pml4(pmap, vaddr);
- if (pml4 && ((*pml4 & INTEL_PTE_VALID))) {
- newpf = *pml4 & PG_FRAME;
- return &((pdpt_entry_t *) PHYSMAP_PTOV(newpf))
- [(vaddr >> PDPTSHIFT) & (NPDPTPG-1)];
- }
- return (NULL);
-}
-/*
- * maps in the pdpt page, if any, containing the pde entry requested
- * and returns the address of the pde entry in that mapped page
- */
-pd_entry_t *
-pmap64_pde(pmap_t pmap, vm_map_offset_t vaddr)
-{
- pdpt_entry_t newpf;
- pdpt_entry_t *pdpt;
-
- assert(pmap);
- if ((vaddr > 0x00007FFFFFFFFFFFULL) &&
- (vaddr < 0xFFFF800000000000ULL)) {
- return (0);
- }
-
- pdpt = pmap64_pdpt(pmap, vaddr);
-
- if (pdpt && ((*pdpt & INTEL_PTE_VALID))) {
- newpf = *pdpt & PG_FRAME;
- return &((pd_entry_t *) PHYSMAP_PTOV(newpf))
- [(vaddr >> PDSHIFT) & (NPDPG-1)];
- }
- return (NULL);
-}
-
-/*
- * return address of mapped pte for vaddr va in pmap pmap.
- *
- * physically maps the pde page, if any, containing the pte in and returns
- * the address of the pte in that mapped page
- *
- * In case the pde maps a superpage, return the pde, which, in this case
- * is the actual page table entry.
- */
-pt_entry_t *
-pmap_pte(pmap_t pmap, vm_map_offset_t vaddr)
-{
- pd_entry_t *pde;
- pd_entry_t newpf;
-
- assert(pmap);
- pde = pmap_pde(pmap, vaddr);
+extern long __stack_chk_guard[];
- if (pde && ((*pde & INTEL_PTE_VALID))) {
- if (*pde & INTEL_PTE_PS)
- return pde;
- newpf = *pde & PG_FRAME;
- return &((pt_entry_t *)PHYSMAP_PTOV(newpf))
- [i386_btop(vaddr) & (ppnum_t)(NPTEPG-1)];
- }
- return (NULL);
-}
+static uint64_t pmap_eptp_flags = 0;
+boolean_t pmap_ept_support_ad = FALSE;
+static void process_pmap_updates(pmap_t, bool, addr64_t, addr64_t);
/*
* Map memory at initialization. The physical addresses being
* mapped are not managed and are never unmapped.
*/
vm_offset_t
pmap_map(
- vm_offset_t virt,
- vm_map_offset_t start_addr,
- vm_map_offset_t end_addr,
- vm_prot_t prot,
- unsigned int flags)
+ vm_offset_t virt,
+ vm_map_offset_t start_addr,
+ vm_map_offset_t end_addr,
+ vm_prot_t prot,
+ unsigned int flags)
{
- int ps;
+ kern_return_t kr;
+ int ps;
ps = PAGE_SIZE;
while (start_addr < end_addr) {
- pmap_enter(kernel_pmap, (vm_map_offset_t)virt,
- (ppnum_t) i386_btop(start_addr), prot, flags, FALSE);
+ kr = pmap_enter(kernel_pmap, (vm_map_offset_t)virt,
+ (ppnum_t) i386_btop(start_addr), prot, VM_PROT_NONE, flags, TRUE);
+
+ if (kr != KERN_SUCCESS) {
+ panic("%s: failed pmap_enter, "
+ "virt=%p, start_addr=%p, end_addr=%p, prot=%#x, flags=%#x",
+ __FUNCTION__,
+ (void *)virt, (void *)start_addr, (void *)end_addr, prot, flags);
+ }
+
virt += ps;
start_addr += ps;
}
- return(virt);
+ return virt;
}
-/*
- * Back-door routine for mapping kernel VM at initialization.
- * Useful for mapping memory outside the range
- * Sets no-cache, A, D.
- * Otherwise like pmap_map.
- */
-vm_offset_t
-pmap_map_bd(
- vm_offset_t virt,
- vm_map_offset_t start_addr,
- vm_map_offset_t end_addr,
- vm_prot_t prot,
- unsigned int flags)
-{
- pt_entry_t template;
- pt_entry_t *pte;
- spl_t spl;
-
- template = pa_to_pte(start_addr)
- | INTEL_PTE_REF
- | INTEL_PTE_MOD
- | INTEL_PTE_WIRED
- | INTEL_PTE_VALID;
-
- if (flags & (VM_MEM_NOT_CACHEABLE | VM_WIMG_USE_DEFAULT)) {
- template |= INTEL_PTE_NCACHE;
- if (!(flags & (VM_MEM_GUARDED | VM_WIMG_USE_DEFAULT)))
- template |= INTEL_PTE_PTA;
- }
- if (prot & VM_PROT_WRITE)
- template |= INTEL_PTE_WRITE;
-
-
- while (start_addr < end_addr) {
- spl = splhigh();
- pte = pmap_pte(kernel_pmap, (vm_map_offset_t)virt);
- if (pte == PT_ENTRY_NULL) {
- panic("pmap_map_bd: Invalid kernel address\n");
- }
- pmap_store_pte(pte, template);
- splx(spl);
- pte_increment_pa(template);
- virt += PAGE_SIZE;
- start_addr += PAGE_SIZE;
- }
-
+extern char *first_avail;
+extern vm_offset_t virtual_avail, virtual_end;
+extern pmap_paddr_t avail_start, avail_end;
+extern vm_offset_t sHIB;
+extern vm_offset_t eHIB;
+extern vm_offset_t stext;
+extern vm_offset_t etext;
+extern vm_offset_t sdata, edata;
+extern vm_offset_t sconst, econst;
- flush_tlb();
- return(virt);
-}
+extern void *KPTphys;
-extern char *first_avail;
-extern vm_offset_t virtual_avail, virtual_end;
-extern pmap_paddr_t avail_start, avail_end;
-extern vm_offset_t sHIB;
-extern vm_offset_t eHIB;
-extern vm_offset_t stext;
-extern vm_offset_t etext;
-extern vm_offset_t sdata;
+boolean_t pmap_smep_enabled = FALSE;
+boolean_t pmap_smap_enabled = FALSE;
void
pmap_cpu_init(void)
{
- /*
- * Here early in the life of a processor (from cpu_mode_init()).
- * Ensure global page feature is disabled.
- */
- set_cr4(get_cr4() &~ CR4_PGE);
+ cpu_data_t *cdp = current_cpu_datap();
+
+ set_cr4(get_cr4() | CR4_PGE);
/*
* Initialize the per-cpu, TLB-related fields.
*/
- current_cpu_datap()->cpu_kernel_cr3 = kernel_pmap->pm_cr3;
- current_cpu_datap()->cpu_active_cr3 = kernel_pmap->pm_cr3;
- current_cpu_datap()->cpu_tlb_invalid = FALSE;
+ cdp->cpu_kernel_cr3 = kernel_pmap->pm_cr3;
+ cpu_shadowp(cdp->cpu_number)->cpu_kernel_cr3 = cdp->cpu_kernel_cr3;
+ cdp->cpu_active_cr3 = kernel_pmap->pm_cr3;
+ cdp->cpu_tlb_invalid = 0;
+ cdp->cpu_task_map = TASK_MAP_64BIT;
+
+ pmap_pcid_configure();
+ if (cpuid_leaf7_features() & CPUID_LEAF7_FEATURE_SMEP) {
+ pmap_smep_enabled = TRUE;
+#if DEVELOPMENT || DEBUG
+ boolean_t nsmep;
+ if (PE_parse_boot_argn("-pmap_smep_disable", &nsmep, sizeof(nsmep))) {
+ pmap_smep_enabled = FALSE;
+ }
+#endif
+ if (pmap_smep_enabled) {
+ set_cr4(get_cr4() | CR4_SMEP);
+ }
+ }
+ if (cpuid_leaf7_features() & CPUID_LEAF7_FEATURE_SMAP) {
+ pmap_smap_enabled = TRUE;
+#if DEVELOPMENT || DEBUG
+ boolean_t nsmap;
+ if (PE_parse_boot_argn("-pmap_smap_disable", &nsmap, sizeof(nsmap))) {
+ pmap_smap_enabled = FALSE;
+ }
+#endif
+ if (pmap_smap_enabled) {
+ set_cr4(get_cr4() | CR4_SMAP);
+ }
+ }
+
+#if !MONOTONIC
+ if (cdp->cpu_fixed_pmcs_enabled) {
+ boolean_t enable = TRUE;
+ cpu_pmc_control(&enable);
+ }
+#endif /* !MONOTONIC */
}
+static uint32_t
+pmap_scale_shift(void)
+{
+ uint32_t scale = 0;
+
+ if (sane_size <= 8 * GB) {
+ scale = (uint32_t)(sane_size / (2 * GB));
+ } else if (sane_size <= 32 * GB) {
+ scale = 4 + (uint32_t)((sane_size - (8 * GB)) / (4 * GB));
+ } else {
+ scale = 10 + (uint32_t)MIN(4, ((sane_size - (32 * GB)) / (8 * GB)));
+ }
+ return scale;
+}
+lck_grp_t pmap_lck_grp;
+lck_grp_attr_t pmap_lck_grp_attr;
+lck_attr_t pmap_lck_rw_attr;
/*
* Bootstrap the system enough to run with virtual memory.
void
pmap_bootstrap(
- __unused vm_offset_t load_start,
- __unused boolean_t IA32e)
+ __unused vm_offset_t load_start,
+ __unused boolean_t IA32e)
{
#if NCOPY_WINDOWS > 0
- vm_offset_t va;
+ vm_offset_t va;
int i;
#endif
-
assert(IA32e);
- vm_last_addr = VM_MAX_KERNEL_ADDRESS; /* Set the highest address
- * known to VM */
+ vm_last_addr = VM_MAX_KERNEL_ADDRESS; /* Set the highest address
+ * known to VM */
/*
* The kernel's pmap is statically allocated so we don't
* have to use pmap_create, which is unlikely to work
kernel_pmap = &kernel_pmap_store;
kernel_pmap->ref_count = 1;
- kernel_pmap->nx_enabled = FALSE;
+#if DEVELOPMENT || DEBUG
+ kernel_pmap->nx_enabled = TRUE;
+#endif
kernel_pmap->pm_task_map = TASK_MAP_64BIT;
kernel_pmap->pm_obj = (vm_object_t) NULL;
- kernel_pmap->dirbase = (pd_entry_t *)((uintptr_t)IdlePTD);
- kernel_pmap->pm_pdpt = (pd_entry_t *) ((uintptr_t)IdlePDPT);
kernel_pmap->pm_pml4 = IdlePML4;
+ kernel_pmap->pm_upml4 = IdlePML4;
kernel_pmap->pm_cr3 = (uintptr_t)ID_MAP_VTOP(IdlePML4);
+ kernel_pmap->pm_ucr3 = (uintptr_t)ID_MAP_VTOP(IdlePML4);
+ kernel_pmap->pm_eptp = 0;
+ pmap_pcid_initialize_kernel(kernel_pmap);
- current_cpu_datap()->cpu_kernel_cr3 = (addr64_t) kernel_pmap->pm_cr3;
+ current_cpu_datap()->cpu_kernel_cr3 = cpu_shadowp(cpu_number())->cpu_kernel_cr3 = (addr64_t) kernel_pmap->pm_cr3;
nkpt = NKPT;
- OSAddAtomic(NKPT, &inuse_ptepages_count);
+ OSAddAtomic(NKPT, &inuse_ptepages_count);
+ OSAddAtomic64(NKPT, &alloc_ptepages_count);
+ bootstrap_wired_pages = NKPT;
virtual_avail = (vm_offset_t)(VM_MIN_KERNEL_ADDRESS) + (vm_offset_t)first_avail;
virtual_end = (vm_offset_t)(VM_MAX_KERNEL_ADDRESS);
* Reserve some special page table entries/VA space for temporary
* mapping of pages.
*/
-#define SYSMAP(c, p, v, n) \
+#define SYSMAP(c, p, v, n) \
v = (c)va; va += ((n)*INTEL_PGBYTES);
va = virtual_avail;
- for (i=0; i<PMAP_NWINDOWS; i++) {
+ for (i = 0; i < PMAP_NWINDOWS; i++) {
#if 1
- kprintf("trying to do SYSMAP idx %d %p\n", i,
- current_cpu_datap());
- kprintf("cpu_pmap %p\n", current_cpu_datap()->cpu_pmap);
- kprintf("mapwindow %p\n", current_cpu_datap()->cpu_pmap->mapwindow);
- kprintf("two stuff %p %p\n",
- (void *)(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP),
- (void *)(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CADDR));
+ kprintf("trying to do SYSMAP idx %d %p\n", i,
+ current_cpu_datap());
+ kprintf("cpu_pmap %p\n", current_cpu_datap()->cpu_pmap);
+ kprintf("mapwindow %p\n", current_cpu_datap()->cpu_pmap->mapwindow);
+ kprintf("two stuff %p %p\n",
+ (void *)(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP),
+ (void *)(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CADDR));
#endif
- SYSMAP(caddr_t,
- (current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP),
- (current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CADDR),
- 1);
- current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP =
- &(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP_store);
- *current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP = 0;
- }
-
- /* DMAP user for debugger */
- SYSMAP(caddr_t, DMAP1, DADDR1, 1);
- SYSMAP(caddr_t, DMAP2, DADDR2, 1); /* XXX temporary - can remove */
+ SYSMAP(caddr_t,
+ (current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP),
+ (current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CADDR),
+ 1);
+ current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP =
+ &(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP_store);
+ *current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP = 0;
+ }
+
virtual_avail = va;
#endif
+ if (!PE_parse_boot_argn("npvhash", &npvhashmask, sizeof(npvhashmask))) {
+ npvhashmask = ((NPVHASHBUCKETS) << pmap_scale_shift()) - 1;
+ }
- if (PE_parse_boot_argn("npvhash", &npvhash, sizeof (npvhash))) {
- if (0 != ((npvhash + 1) & npvhash)) {
- kprintf("invalid hash %d, must be ((2^N)-1), "
- "using default %d\n", npvhash, NPVHASH);
- npvhash = NPVHASH;
- }
- } else {
- npvhash = NPVHASH;
+ npvhashbuckets = npvhashmask + 1;
+
+ if (0 != ((npvhashbuckets) & npvhashmask)) {
+ panic("invalid hash %d, must be ((2^N)-1), "
+ "using default %d\n", npvhashmask, NPVHASHMASK);
}
- printf("npvhash=%d\n", npvhash);
+ lck_grp_attr_setdefault(&pmap_lck_grp_attr);
+ lck_grp_init(&pmap_lck_grp, "pmap", &pmap_lck_grp_attr);
+
+ lck_attr_setdefault(&pmap_lck_rw_attr);
+ lck_attr_cleardebug(&pmap_lck_rw_attr);
+
+ lck_rw_init(&kernel_pmap->pmap_rwl, &pmap_lck_grp, &pmap_lck_rw_attr);
+ kernel_pmap->pmap_rwl.lck_rw_can_sleep = FALSE;
- simple_lock_init(&kernel_pmap->lock, 0);
simple_lock_init(&pv_hashed_free_list_lock, 0);
simple_lock_init(&pv_hashed_kern_free_list_lock, 0);
- simple_lock_init(&pv_hash_table_lock,0);
+ simple_lock_init(&pv_hash_table_lock, 0);
+ simple_lock_init(&phys_backup_lock, 0);
pmap_cpu_init();
+ if (pmap_pcid_ncpus) {
+ printf("PMAP: PCID enabled\n");
+ }
+
+ if (pmap_smep_enabled) {
+ printf("PMAP: Supervisor Mode Execute Protection enabled\n");
+ }
+ if (pmap_smap_enabled) {
+ printf("PMAP: Supervisor Mode Access Protection enabled\n");
+ }
+
+#if DEBUG
+ printf("Stack canary: 0x%lx\n", __stack_chk_guard[0]);
+ printf("early_random(): 0x%qx\n", early_random());
+#endif
+#if DEVELOPMENT || DEBUG
+ boolean_t ptmp;
+ /* Check if the user has requested disabling stack or heap no-execute
+ * enforcement. These are "const" variables; that qualifier is cast away
+ * when altering them. The TEXT/DATA const sections are marked
+ * write protected later in the kernel startup sequence, so altering
+ * them is possible at this point, in pmap_bootstrap().
+ */
+ if (PE_parse_boot_argn("-pmap_disable_kheap_nx", &ptmp, sizeof(ptmp))) {
+ boolean_t *pdknxp = (boolean_t *) &pmap_disable_kheap_nx;
+ *pdknxp = TRUE;
+ }
+
+ if (PE_parse_boot_argn("-pmap_disable_kstack_nx", &ptmp, sizeof(ptmp))) {
+ boolean_t *pdknhp = (boolean_t *) &pmap_disable_kstack_nx;
+ *pdknhp = TRUE;
+ }
+#endif /* DEVELOPMENT || DEBUG */
+
+ boot_args *args = (boot_args *)PE_state.bootArgs;
+ if (args->efiMode == kBootArgsEfiMode32) {
+ printf("EFI32: kernel virtual space limited to 4GB\n");
+ virtual_end = VM_MAX_KERNEL_ADDRESS_EFI32;
+ }
kprintf("Kernel virtual space from 0x%lx to 0x%lx.\n",
- (long)KERNEL_BASE, (long)virtual_end);
+ (long)KERNEL_BASE, (long)virtual_end);
kprintf("Available physical space from 0x%llx to 0x%llx\n",
- avail_start, avail_end);
+ avail_start, avail_end);
/*
* The -no_shared_cr3 boot-arg is a debugging feature (set by default
* in the DEBUG kernel) to force the kernel to switch to its own map
* (and cr3) when control is in kernelspace. The kernel's map does not
* include (i.e. share) userspace so wild references will cause
- * a panic. Only copyin and copyout are exempt from this.
+ * a panic. Only copyin and copyout are exempt from this.
*/
(void) PE_parse_boot_argn("-no_shared_cr3",
- &no_shared_cr3, sizeof (no_shared_cr3));
- if (no_shared_cr3)
+ &no_shared_cr3, sizeof(no_shared_cr3));
+ if (no_shared_cr3) {
kprintf("Kernel not sharing user map\n");
-
-#ifdef PMAP_TRACES
- if (PE_parse_boot_argn("-pmap_trace", &pmap_trace, sizeof (pmap_trace))) {
+ }
+
+#ifdef PMAP_TRACES
+ if (PE_parse_boot_argn("-pmap_trace", &pmap_trace, sizeof(pmap_trace))) {
kprintf("Kernel traces for pmap operations enabled\n");
- }
-#endif /* PMAP_TRACES */
+ }
+#endif /* PMAP_TRACES */
+
+#if MACH_ASSERT
+ PE_parse_boot_argn("pmap_asserts", &pmap_asserts_enabled, sizeof(pmap_asserts_enabled));
+ PE_parse_boot_argn("pmap_stats_assert",
+ &pmap_stats_assert,
+ sizeof(pmap_stats_assert));
+#endif /* MACH_ASSERT */
}
void
*endp = virtual_end;
}
-/*
- * Initialize the pmap module.
- * Called by vm_init, to initialize any structures that the pmap
- * system needs to map virtual memory.
- */
-void
-pmap_init(void)
-{
- long npages;
- vm_offset_t addr;
- vm_size_t s;
- vm_map_offset_t vaddr;
- ppnum_t ppn;
- kernel_pmap->pm_obj_pml4 = &kpml4obj_object_store;
- _vm_object_allocate((vm_object_size_t)NPML4PGS, &kpml4obj_object_store);
- kernel_pmap->pm_obj_pdpt = &kpdptobj_object_store;
- _vm_object_allocate((vm_object_size_t)NPDPTPGS, &kpdptobj_object_store);
+#if HIBERNATION
- kernel_pmap->pm_obj = &kptobj_object_store;
- _vm_object_allocate((vm_object_size_t)NPDEPGS, &kptobj_object_store);
+#include <IOKit/IOHibernatePrivate.h>
- /*
- * Allocate memory for the pv_head_table and its lock bits,
- * the modify bit array, and the pte_page table.
+int32_t pmap_npages;
+int32_t pmap_teardown_last_valid_compact_indx = -1;
+
+
+void hibernate_rebuild_pmap_structs(void);
+void hibernate_teardown_pmap_structs(addr64_t *, addr64_t *);
+void pmap_pack_index(uint32_t);
+int32_t pmap_unpack_index(pv_rooted_entry_t);
+
+
+int32_t
+pmap_unpack_index(pv_rooted_entry_t pv_h)
+{
+ int32_t indx = 0;
+
+ indx = (int32_t)(*((uint64_t *)(&pv_h->qlink.next)) >> 48);
+ indx = indx << 16;
+ indx |= (int32_t)(*((uint64_t *)(&pv_h->qlink.prev)) >> 48);
+
+ *((uint64_t *)(&pv_h->qlink.next)) |= ((uint64_t)0xffff << 48);
+ *((uint64_t *)(&pv_h->qlink.prev)) |= ((uint64_t)0xffff << 48);
+
+ return indx;
+}
+
+
+void
+pmap_pack_index(uint32_t indx)
+{
+ pv_rooted_entry_t pv_h;
+
+ pv_h = &pv_head_table[indx];
+
+ *((uint64_t *)(&pv_h->qlink.next)) &= ~((uint64_t)0xffff << 48);
+ *((uint64_t *)(&pv_h->qlink.prev)) &= ~((uint64_t)0xffff << 48);
+
+ *((uint64_t *)(&pv_h->qlink.next)) |= ((uint64_t)(indx >> 16)) << 48;
+ *((uint64_t *)(&pv_h->qlink.prev)) |= ((uint64_t)(indx & 0xffff)) << 48;
+}
+
+
+void
+hibernate_teardown_pmap_structs(addr64_t *unneeded_start, addr64_t *unneeded_end)
+{
+ int32_t i;
+ int32_t compact_target_indx;
+
+ compact_target_indx = 0;
+
+ for (i = 0; i < pmap_npages; i++) {
+ if (pv_head_table[i].pmap == PMAP_NULL) {
+ if (pv_head_table[compact_target_indx].pmap != PMAP_NULL) {
+ compact_target_indx = i;
+ }
+ } else {
+ pmap_pack_index((uint32_t)i);
+
+ if (pv_head_table[compact_target_indx].pmap == PMAP_NULL) {
+ /*
+ * we've got a hole to fill, so
+ * move this pv_rooted_entry_t to it's new home
+ */
+ pv_head_table[compact_target_indx] = pv_head_table[i];
+ pv_head_table[i].pmap = PMAP_NULL;
+
+ pmap_teardown_last_valid_compact_indx = compact_target_indx;
+ compact_target_indx++;
+ } else {
+ pmap_teardown_last_valid_compact_indx = i;
+ }
+ }
+ }
+ *unneeded_start = (addr64_t)&pv_head_table[pmap_teardown_last_valid_compact_indx + 1];
+ *unneeded_end = (addr64_t)&pv_head_table[pmap_npages - 1];
+
+ HIBLOG("hibernate_teardown_pmap_structs done: last_valid_compact_indx %d\n", pmap_teardown_last_valid_compact_indx);
+}
+
+
+void
+hibernate_rebuild_pmap_structs(void)
+{
+ int32_t cindx, eindx, rindx = 0;
+ pv_rooted_entry_t pv_h;
+
+ eindx = (int32_t)pmap_npages;
+
+ for (cindx = pmap_teardown_last_valid_compact_indx; cindx >= 0; cindx--) {
+ pv_h = &pv_head_table[cindx];
+
+ rindx = pmap_unpack_index(pv_h);
+ assert(rindx < pmap_npages);
+
+ if (rindx != cindx) {
+ /*
+ * this pv_rooted_entry_t was moved by hibernate_teardown_pmap_structs,
+ * so move it back to its real location
+ */
+ pv_head_table[rindx] = pv_head_table[cindx];
+ }
+ if (rindx + 1 != eindx) {
+ /*
+ * the 'hole' between this vm_rooted_entry_t and the previous
+ * vm_rooted_entry_t we moved needs to be initialized as
+ * a range of zero'd vm_rooted_entry_t's
+ */
+ bzero((char *)&pv_head_table[rindx + 1], (eindx - rindx - 1) * sizeof(struct pv_rooted_entry));
+ }
+ eindx = rindx;
+ }
+ if (rindx) {
+ bzero((char *)&pv_head_table[0], rindx * sizeof(struct pv_rooted_entry));
+ }
+
+ HIBLOG("hibernate_rebuild_pmap_structs done: last_valid_compact_indx %d\n", pmap_teardown_last_valid_compact_indx);
+}
+
+#endif
+
+/*
+ * Initialize the pmap module.
+ * Called by vm_init, to initialize any structures that the pmap
+ * system needs to map virtual memory.
+ */
+void
+pmap_init(void)
+{
+ long npages;
+ vm_offset_t addr;
+ vm_size_t s, vsize;
+ vm_map_offset_t vaddr;
+ ppnum_t ppn;
+
+
+ kernel_pmap->pm_obj_pml4 = &kpml4obj_object_store;
+ _vm_object_allocate((vm_object_size_t)NPML4PGS * PAGE_SIZE, &kpml4obj_object_store);
+
+ kernel_pmap->pm_obj_pdpt = &kpdptobj_object_store;
+ _vm_object_allocate((vm_object_size_t)NPDPTPGS * PAGE_SIZE, &kpdptobj_object_store);
+
+ kernel_pmap->pm_obj = &kptobj_object_store;
+ _vm_object_allocate((vm_object_size_t)NPDEPGS * PAGE_SIZE, &kptobj_object_store);
+
+ /*
+ * Allocate memory for the pv_head_table and its lock bits,
+ * the modify bit array, and the pte_page table.
*/
/*
*/
npages = i386_btop(avail_end);
+#if HIBERNATION
+ pmap_npages = (uint32_t)npages;
+#endif
s = (vm_size_t) (sizeof(struct pv_rooted_entry) * npages
- + (sizeof (struct pv_hashed_entry_t *) * (npvhash+1))
- + pv_lock_table_size(npages)
- + pv_hash_lock_table_size((npvhash+1))
- + npages);
-
+ + (sizeof(struct pv_hashed_entry_t *) * (npvhashbuckets))
+ + pv_lock_table_size(npages)
+ + pv_hash_lock_table_size((npvhashbuckets))
+ + npages);
s = round_page(s);
if (kernel_memory_allocate(kernel_map, &addr, s, 0,
- KMA_KOBJECT | KMA_PERMANENT)
- != KERN_SUCCESS)
+ KMA_KOBJECT | KMA_PERMANENT, VM_KERN_MEMORY_PMAP)
+ != KERN_SUCCESS) {
panic("pmap_init");
+ }
memset((char *)addr, 0, s);
+ vaddr = addr;
+ vsize = s;
+
#if PV_DEBUG
- if (0 == npvhash) panic("npvhash not initialized");
+ if (0 == npvhashmask) {
+ panic("npvhashmask not initialized");
+ }
#endif
/*
addr = (vm_offset_t) (pv_head_table + npages);
pv_hash_table = (pv_hashed_entry_t *)addr;
- addr = (vm_offset_t) (pv_hash_table + (npvhash + 1));
+ addr = (vm_offset_t) (pv_hash_table + (npvhashbuckets));
pv_lock_table = (char *) addr;
addr = (vm_offset_t) (pv_lock_table + pv_lock_table_size(npages));
pv_hash_lock_table = (char *) addr;
- addr = (vm_offset_t) (pv_hash_lock_table + pv_hash_lock_table_size((npvhash+1)));
+ addr = (vm_offset_t) (pv_hash_lock_table + pv_hash_lock_table_size((npvhashbuckets)));
pmap_phys_attributes = (char *) addr;
ppnum_t last_pn = i386_btop(avail_end);
- unsigned int i;
+ unsigned int i;
pmap_memory_region_t *pmptr = pmap_memory_regions;
for (i = 0; i < pmap_memory_region_count; i++, pmptr++) {
- if (pmptr->type != kEfiConventionalMemory)
+ if (pmptr->type != kEfiConventionalMemory) {
continue;
- unsigned int pn;
+ }
+ ppnum_t pn;
for (pn = pmptr->base; pn <= pmptr->end; pn++) {
if (pn < last_pn) {
pmap_phys_attributes[pn] |= PHYS_MANAGED;
- if (pn > last_managed_page)
+
+ if (pn > last_managed_page) {
last_managed_page = pn;
+ }
+
+ if (pn >= lowest_hi && pn <= highest_hi) {
+ pmap_phys_attributes[pn] |= PHYS_NOENCRYPT;
+ }
}
}
}
+ while (vsize) {
+ ppn = pmap_find_phys(kernel_pmap, vaddr);
+ pmap_phys_attributes[ppn] |= PHYS_NOENCRYPT;
+
+ vaddr += PAGE_SIZE;
+ vsize -= PAGE_SIZE;
+ }
/*
* Create the zone of physical maps,
* and of the physical-to-virtual entries.
*/
s = (vm_size_t) sizeof(struct pmap);
- pmap_zone = zinit(s, 400*s, 4096, "pmap"); /* XXX */
- s = (vm_size_t) sizeof(struct pv_hashed_entry);
- pv_hashed_list_zone = zinit(s, 10000*s, 4096, "pv_list"); /* XXX */
- s = 63;
- pdpt_zone = zinit(s, 400*s, 4096, "pdpt"); /* XXX */
+ pmap_zone = zinit(s, 400 * s, 4096, "pmap"); /* XXX */
+ zone_change(pmap_zone, Z_NOENCRYPT, TRUE);
+
+ pmap_anchor_zone = zinit(PAGE_SIZE, task_max, PAGE_SIZE, "pagetable anchors");
+ zone_change(pmap_anchor_zone, Z_NOENCRYPT, TRUE);
+
+ /* The anchor is required to be page aligned. Zone debugging adds
+ * padding which may violate that requirement. Tell the zone
+ * subsystem that alignment is required.
+ */
+
+ zone_change(pmap_anchor_zone, Z_ALIGNMENT_REQUIRED, TRUE);
+/* TODO: possible general optimisation...pre-allocate via zones commonly created
+ * level3/2 pagetables
+ */
+ pmap_uanchor_zone = zinit(PAGE_SIZE, task_max, PAGE_SIZE, "pagetable user anchors");
+ zone_change(pmap_uanchor_zone, Z_NOENCRYPT, TRUE);
+
+ /* The anchor is required to be page aligned. Zone debugging adds
+ * padding which may violate that requirement. Tell the zone
+ * subsystem that alignment is required.
+ */
+ zone_change(pmap_uanchor_zone, Z_ALIGNMENT_REQUIRED, TRUE);
- /* create pv entries for kernel pages mapped by low level
- startup code. these have to exist so we can pmap_remove()
- e.g. kext pages from the middle of our addr space */
+ s = (vm_size_t) sizeof(struct pv_hashed_entry);
+ pv_hashed_list_zone = zinit(s, 10000 * s /* Expandable zone */,
+ 4096 * 3 /* LCM x86_64*/, "pv_list");
+ zone_change(pv_hashed_list_zone, Z_NOENCRYPT, TRUE);
+ zone_change(pv_hashed_list_zone, Z_GZALLOC_EXEMPT, TRUE);
+ /* create pv entries for kernel pages that might get pmap_remove()ed */
vaddr = (vm_map_offset_t) VM_MIN_KERNEL_ADDRESS;
- for (ppn = 0; ppn < i386_btop(avail_start); ppn++) {
- pv_rooted_entry_t pv_e;
+ for (ppn = VM_MIN_KERNEL_PAGE; ppn < i386_btop(avail_start); ppn++) {
+ pv_rooted_entry_t pv_h;
- pv_e = pai_to_pvh(ppn);
- pv_e->va = vaddr;
+ pv_h = pai_to_pvh(ppn);
+ assert(pv_h->qlink.next == 0); /* shouldn't be init'd yet */
+ assert(pv_h->pmap == NULL);
+ pv_h->va_and_flags = vaddr;
vaddr += PAGE_SIZE;
- pv_e->pmap = kernel_pmap;
- queue_init(&pv_e->qlink);
+ pv_h->pmap = kernel_pmap;
+ queue_init(&pv_h->qlink);
}
pmap_initialized = TRUE;
- /*
- * Initialize pmap cache.
- */
- pmap_cache_list = PMAP_NULL;
- pmap_cache_count = 0;
- simple_lock_init(&pmap_cache_lock, 0);
-
max_preemption_latency_tsc = tmrCvt((uint64_t)MAX_PREEMPTION_LATENCY_NS, tscFCvtn2t);
/*
* Ensure the kernel's PML4 entry exists for the basement
* before this is shared with any user.
*/
- pmap_expand_pml4(kernel_pmap, KERNEL_BASEMENT);
+ pmap_expand_pml4(kernel_pmap, KERNEL_BASEMENT, PMAP_EXPAND_OPTIONS_NONE);
+
+#if CONFIG_VMX
+ pmap_ept_support_ad = vmx_hv_support() && (VMX_CAP(MSR_IA32_VMX_EPT_VPID_CAP, MSR_IA32_VMX_EPT_VPID_CAP_AD_SHIFT, 1) ? TRUE : FALSE);
+ pmap_eptp_flags = HV_VMX_EPTP_MEMORY_TYPE_WB | HV_VMX_EPTP_WALK_LENGTH(4) | (pmap_ept_support_ad ? HV_VMX_EPTP_ENABLE_AD_FLAGS : 0);
+#endif /* CONFIG_VMX */
+}
+
+/*
+ * Create pv entries for kernel pages mapped by low level
+ * startup code. These have to exist so we can pmap_remove() them.
+ */
+void
+pmap_pv_fixup(vm_offset_t start, vm_size_t length)
+{
+ ppnum_t ppn;
+ pv_rooted_entry_t pv_h;
+
+ while (length != 0) {
+ ppn = pmap_find_phys(kernel_pmap, start);
+ if (ppn != 0) {
+ pv_h = pai_to_pvh(ppn);
+ assert(pv_h->qlink.next == 0); /* shouldn't be init'd yet */
+ assert(pv_h->pmap == 0);
+ pv_h->va_and_flags = start;
+ pv_h->pmap = kernel_pmap;
+ queue_init(&pv_h->qlink);
+ }
+ start += PAGE_SIZE;
+ length -= PAGE_SIZE;
+ }
+}
+
+static
+void
+pmap_mark_range(pmap_t npmap, uint64_t sv, uint64_t nxrosz, boolean_t NX, boolean_t ro)
+{
+ uint64_t ev = sv + nxrosz, cv = sv;
+ pd_entry_t *pdep;
+ pt_entry_t *ptep = NULL;
+
+ assert(!is_ept_pmap(npmap));
+
+ assert(((sv & 0xFFFULL) | (nxrosz & 0xFFFULL)) == 0);
+
+ for (pdep = pmap_pde(npmap, cv); pdep != NULL && (cv < ev);) {
+ uint64_t pdev = (cv & ~((uint64_t)PDEMASK));
+
+ if (*pdep & INTEL_PTE_PS) {
+ if (NX) {
+ *pdep |= INTEL_PTE_NX;
+ }
+ if (ro) {
+ *pdep &= ~INTEL_PTE_WRITE;
+ }
+ cv += NBPD;
+ cv &= ~((uint64_t) PDEMASK);
+ pdep = pmap_pde(npmap, cv);
+ continue;
+ }
+
+ for (ptep = pmap_pte(npmap, cv); ptep != NULL && (cv < (pdev + NBPD)) && (cv < ev);) {
+ if (NX) {
+ *ptep |= INTEL_PTE_NX;
+ }
+ if (ro) {
+ *ptep &= ~INTEL_PTE_WRITE;
+ }
+ cv += NBPT;
+ ptep = pmap_pte(npmap, cv);
+ }
+ }
+ DPRINTF("%s(0x%llx, 0x%llx, %u, %u): 0x%llx, 0x%llx\n", __FUNCTION__, sv, nxrosz, NX, ro, cv, ptep ? *ptep: 0);
}
+/*
+ * Called once VM is fully initialized so that we can release unused
+ * sections of low memory to the general pool.
+ * Also complete the set-up of identity-mapped sections of the kernel:
+ * 1) write-protect kernel text
+ * 2) map kernel text using large pages if possible
+ * 3) read and write-protect page zero (for K32)
+ * 4) map the global page at the appropriate virtual address.
+ *
+ * Use of large pages
+ * ------------------
+ * To effectively map and write-protect all kernel text pages, the text
+ * must be 2M-aligned at the base, and the data section above must also be
+ * 2M-aligned. That is, there's padding below and above. This is achieved
+ * through linker directives. Large pages are used only if this alignment
+ * exists (and not overriden by the -kernel_text_page_4K boot-arg). The
+ * memory layout is:
+ *
+ * : :
+ * | __DATA |
+ * sdata: ================== 2Meg
+ * | |
+ * | zero-padding |
+ * | |
+ * etext: ------------------
+ * | |
+ * : :
+ * | |
+ * | __TEXT |
+ * | |
+ * : :
+ * | |
+ * stext: ================== 2Meg
+ * | |
+ * | zero-padding |
+ * | |
+ * eHIB: ------------------
+ * | __HIB |
+ * : :
+ *
+ * Prior to changing the mapping from 4K to 2M, the zero-padding pages
+ * [eHIB,stext] and [etext,sdata] are ml_static_mfree()'d. Then all the
+ * 4K pages covering [stext,etext] are coalesced as 2M large pages.
+ * The now unused level-1 PTE pages are also freed.
+ */
+extern ppnum_t vm_kernel_base_page;
+static uint32_t constptes = 0, dataptes = 0;
+
+void
+pmap_lowmem_finalize(void)
+{
+ spl_t spl;
+ int i;
+
+ /*
+ * Update wired memory statistics for early boot pages
+ */
+ PMAP_ZINFO_PALLOC(kernel_pmap, bootstrap_wired_pages * PAGE_SIZE);
+
+ /*
+ * Free pages in pmap regions below the base:
+ * rdar://6332712
+ * We can't free all the pages to VM that EFI reports available.
+ * Pages in the range 0xc0000-0xff000 aren't safe over sleep/wake.
+ * There's also a size miscalculation here: pend is one page less
+ * than it should be but this is not fixed to be backwards
+ * compatible.
+ * This is important for KASLR because up to 256*2MB = 512MB of space
+ * needs has to be released to VM.
+ */
+ for (i = 0;
+ pmap_memory_regions[i].end < vm_kernel_base_page;
+ i++) {
+ vm_offset_t pbase = i386_ptob(pmap_memory_regions[i].base);
+ vm_offset_t pend = i386_ptob(pmap_memory_regions[i].end + 1);
+
+ DBG("pmap region %d [%p..[%p\n",
+ i, (void *) pbase, (void *) pend);
+
+ if (pmap_memory_regions[i].attribute & EFI_MEMORY_KERN_RESERVED) {
+ continue;
+ }
+ /*
+ * rdar://6332712
+ * Adjust limits not to free pages in range 0xc0000-0xff000.
+ */
+ if (pbase >= 0xc0000 && pend <= 0x100000) {
+ continue;
+ }
+ if (pbase < 0xc0000 && pend > 0x100000) {
+ /* page range entirely within region, free lower part */
+ DBG("- ml_static_mfree(%p,%p)\n",
+ (void *) ml_static_ptovirt(pbase),
+ (void *) (0xc0000 - pbase));
+ ml_static_mfree(ml_static_ptovirt(pbase), 0xc0000 - pbase);
+ pbase = 0x100000;
+ }
+ if (pbase < 0xc0000) {
+ pend = MIN(pend, 0xc0000);
+ }
+ if (pend > 0x100000) {
+ pbase = MAX(pbase, 0x100000);
+ }
+ DBG("- ml_static_mfree(%p,%p)\n",
+ (void *) ml_static_ptovirt(pbase),
+ (void *) (pend - pbase));
+ ml_static_mfree(ml_static_ptovirt(pbase), pend - pbase);
+ }
+
+ /* A final pass to get rid of all initial identity mappings to
+ * low pages.
+ */
+ DPRINTF("%s: Removing mappings from 0->0x%lx\n", __FUNCTION__, vm_kernel_base);
+
+ /*
+ * Remove all mappings past the boot-cpu descriptor aliases and low globals.
+ * Non-boot-cpu GDT aliases will be remapped later as needed.
+ */
+ pmap_remove(kernel_pmap, LOWGLOBAL_ALIAS + PAGE_SIZE, vm_kernel_base);
+
+ /*
+ * If text and data are both 2MB-aligned,
+ * we can map text with large-pages,
+ * unless the -kernel_text_ps_4K boot-arg overrides.
+ */
+ if ((stext & I386_LPGMASK) == 0 && (sdata & I386_LPGMASK) == 0) {
+ kprintf("Kernel text is 2MB aligned");
+ kernel_text_ps_4K = FALSE;
+ if (PE_parse_boot_argn("-kernel_text_ps_4K",
+ &kernel_text_ps_4K,
+ sizeof(kernel_text_ps_4K))) {
+ kprintf(" but will be mapped with 4K pages\n");
+ } else {
+ kprintf(" and will be mapped with 2M pages\n");
+ }
+ }
+#if DEVELOPMENT || DEBUG
+ (void) PE_parse_boot_argn("wpkernel", &wpkernel, sizeof(wpkernel));
+#endif
+ if (wpkernel) {
+ kprintf("Kernel text %p-%p to be write-protected\n",
+ (void *) stext, (void *) etext);
+ }
+
+ spl = splhigh();
+
+ /*
+ * Scan over text if mappings are to be changed:
+ * - Remap kernel text readonly unless the "wpkernel" boot-arg is 0
+ * - Change to large-pages if possible and not overriden.
+ */
+ if (kernel_text_ps_4K && wpkernel) {
+ vm_offset_t myva;
+ for (myva = stext; myva < etext; myva += PAGE_SIZE) {
+ pt_entry_t *ptep;
+
+ ptep = pmap_pte(kernel_pmap, (vm_map_offset_t)myva);
+ if (ptep) {
+ pmap_store_pte(ptep, *ptep & ~INTEL_PTE_WRITE);
+ }
+ }
+ }
+
+ if (!kernel_text_ps_4K) {
+ vm_offset_t myva;
+
+ /*
+ * Release zero-filled page padding used for 2M-alignment.
+ */
+ DBG("ml_static_mfree(%p,%p) for padding below text\n",
+ (void *) eHIB, (void *) (stext - eHIB));
+ ml_static_mfree(eHIB, stext - eHIB);
+ DBG("ml_static_mfree(%p,%p) for padding above text\n",
+ (void *) etext, (void *) (sdata - etext));
+ ml_static_mfree(etext, sdata - etext);
+
+ /*
+ * Coalesce text pages into large pages.
+ */
+ for (myva = stext; myva < sdata; myva += I386_LPGBYTES) {
+ pt_entry_t *ptep;
+ vm_offset_t pte_phys;
+ pt_entry_t *pdep;
+ pt_entry_t pde;
+
+ pdep = pmap_pde(kernel_pmap, (vm_map_offset_t)myva);
+ ptep = pmap_pte(kernel_pmap, (vm_map_offset_t)myva);
+ DBG("myva: %p pdep: %p ptep: %p\n",
+ (void *) myva, (void *) pdep, (void *) ptep);
+ if ((*ptep & INTEL_PTE_VALID) == 0) {
+ continue;
+ }
+ pte_phys = (vm_offset_t)(*ptep & PG_FRAME);
+ pde = *pdep & PTMASK; /* page attributes from pde */
+ pde |= INTEL_PTE_PS; /* make it a 2M entry */
+ pde |= pte_phys; /* take page frame from pte */
+
+ if (wpkernel) {
+ pde &= ~INTEL_PTE_WRITE;
+ }
+ DBG("pmap_store_pte(%p,0x%llx)\n",
+ (void *)pdep, pde);
+ pmap_store_pte(pdep, pde);
+
+ /*
+ * Free the now-unused level-1 pte.
+ * Note: ptep is a virtual address to the pte in the
+ * recursive map. We can't use this address to free
+ * the page. Instead we need to compute its address
+ * in the Idle PTEs in "low memory".
+ */
+ vm_offset_t vm_ptep = (vm_offset_t) KPTphys
+ + (pte_phys >> PTPGSHIFT);
+ DBG("ml_static_mfree(%p,0x%x) for pte\n",
+ (void *) vm_ptep, PAGE_SIZE);
+ ml_static_mfree(vm_ptep, PAGE_SIZE);
+ }
+
+ /* Change variable read by sysctl machdep.pmap */
+ pmap_kernel_text_ps = I386_LPGBYTES;
+ }
+
+ boolean_t doconstro = TRUE;
+#if DEVELOPMENT || DEBUG
+ (void) PE_parse_boot_argn("dataconstro", &doconstro, sizeof(doconstro));
+#endif
+ if (doconstro) {
+ if (sconst & PAGE_MASK) {
+ panic("CONST segment misaligned 0x%lx 0x%lx\n",
+ sconst, econst);
+ }
+ kprintf("Marking const DATA read-only\n");
+ }
+
+ vm_offset_t dva;
+
+ for (dva = sdata; dva < edata; dva += I386_PGBYTES) {
+ assert(((sdata | edata) & PAGE_MASK) == 0);
+ pt_entry_t dpte, *dptep = pmap_pte(kernel_pmap, dva);
+
+ dpte = *dptep;
+ assert((dpte & INTEL_PTE_VALID));
+ dpte |= INTEL_PTE_NX;
+ pmap_store_pte(dptep, dpte);
+ dataptes++;
+ }
+ assert(dataptes > 0);
+
+ for (dva = sconst; dva < econst; dva += I386_PGBYTES) {
+ pt_entry_t dpte, *dptep = pmap_pte(kernel_pmap, dva);
+
+ dpte = *dptep;
+
+ assert((dpte & INTEL_PTE_VALID));
+ dpte |= INTEL_PTE_NX;
+ dpte &= ~INTEL_PTE_WRITE;
+ constptes++;
+ pmap_store_pte(dptep, dpte);
+ }
+
+ assert(constptes > 0);
+
+ kernel_segment_command_t * seg;
+ kernel_section_t * sec;
+
+ for (seg = firstseg(); seg != NULL; seg = nextsegfromheader(&_mh_execute_header, seg)) {
+ if (!strcmp(seg->segname, "__TEXT") ||
+ !strcmp(seg->segname, "__DATA")) {
+ continue;
+ }
+ //XXX
+ if (!strcmp(seg->segname, "__KLD")) {
+ continue;
+ }
+ if (!strcmp(seg->segname, "__HIB")) {
+ for (sec = firstsect(seg); sec != NULL; sec = nextsect(seg, sec)) {
+ if (sec->addr & PAGE_MASK) {
+ panic("__HIB segment's sections misaligned");
+ }
+ if (!strcmp(sec->sectname, "__text")) {
+ pmap_mark_range(kernel_pmap, sec->addr, round_page(sec->size), FALSE, TRUE);
+ } else {
+ pmap_mark_range(kernel_pmap, sec->addr, round_page(sec->size), TRUE, FALSE);
+ }
+ }
+ } else {
+ pmap_mark_range(kernel_pmap, seg->vmaddr, round_page_64(seg->vmsize), TRUE, FALSE);
+ }
+ }
+
+ /*
+ * If we're debugging, map the low global vector page at the fixed
+ * virtual address. Otherwise, remove the mapping for this.
+ */
+ if (debug_boot_arg) {
+ pt_entry_t *pte = NULL;
+ if (0 == (pte = pmap_pte(kernel_pmap, LOWGLOBAL_ALIAS))) {
+ panic("lowmem pte");
+ }
+ /* make sure it is defined on page boundary */
+ assert(0 == ((vm_offset_t) &lowGlo & PAGE_MASK));
+ pmap_store_pte(pte, kvtophys((vm_offset_t)&lowGlo)
+ | INTEL_PTE_REF
+ | INTEL_PTE_MOD
+ | INTEL_PTE_WIRED
+ | INTEL_PTE_VALID
+ | INTEL_PTE_WRITE
+ | INTEL_PTE_NX);
+ } else {
+ pmap_remove(kernel_pmap,
+ LOWGLOBAL_ALIAS, LOWGLOBAL_ALIAS + PAGE_SIZE);
+ }
+ pmap_tlbi_range(0, ~0ULL, true, 0);
+ splx(spl);
+}
/*
* this function is only used for debugging fron the vm layer
*/
boolean_t
pmap_verify_free(
- ppnum_t pn)
+ ppnum_t pn)
{
- pv_rooted_entry_t pv_h;
- int pai;
- boolean_t result;
+ pv_rooted_entry_t pv_h;
+ int pai;
+ boolean_t result;
assert(pn != vm_page_fictitious_addr);
- if (!pmap_initialized)
- return(TRUE);
+ if (!pmap_initialized) {
+ return TRUE;
+ }
- if (pn == vm_page_guard_addr)
+ if (pn == vm_page_guard_addr) {
return TRUE;
+ }
pai = ppn_to_pai(pn);
- if (!IS_MANAGED_PAGE(pai))
- return(FALSE);
+ if (!IS_MANAGED_PAGE(pai)) {
+ return FALSE;
+ }
pv_h = pai_to_pvh(pn);
result = (pv_h->pmap == PMAP_NULL);
- return(result);
+ return result;
+}
+
+
+#if MACH_ASSERT
+void
+pmap_assert_free(ppnum_t pn)
+{
+ int pai;
+ pv_rooted_entry_t pv_h = NULL;
+ pmap_t pmap = NULL;
+ vm_offset_t va = 0;
+ static char buffer[32];
+ static char *pr_name = "not managed pn";
+ uint_t attr;
+ pt_entry_t *ptep;
+ pt_entry_t pte = -1ull;
+
+ if (pmap_verify_free(pn)) {
+ return;
+ }
+
+ if (pn > last_managed_page) {
+ attr = 0xff;
+ goto done;
+ }
+
+ pai = ppn_to_pai(pn);
+ attr = pmap_phys_attributes[pai];
+ pv_h = pai_to_pvh(pai);
+ va = pv_h->va_and_flags;
+ pmap = pv_h->pmap;
+ if (pmap == kernel_pmap) {
+ pr_name = "kernel";
+ } else if (pmap == NULL) {
+ pr_name = "pmap NULL";
+ } else if (pmap->pmap_procname[0] != 0) {
+ pr_name = &pmap->pmap_procname[0];
+ } else {
+ snprintf(buffer, sizeof(buffer), "pmap %p", pv_h->pmap);
+ pr_name = buffer;
+ }
+
+ if (pmap != NULL) {
+ ptep = pmap_pte(pmap, va);
+ if (ptep != NULL) {
+ pte = (uintptr_t)*ptep;
+ }
+ }
+
+done:
+ panic("page not FREE page: 0x%lx attr: 0x%x %s va: 0x%lx PTE: 0x%llx",
+ (ulong_t)pn, attr, pr_name, va, pte);
}
+#endif /* MACH_ASSERT */
boolean_t
pmap_is_empty(
- pmap_t pmap,
- vm_map_offset_t va_start,
- vm_map_offset_t va_end)
+ pmap_t pmap,
+ vm_map_offset_t va_start,
+ vm_map_offset_t va_end)
{
vm_map_offset_t offset;
ppnum_t phys_page;
* This assumes the count is correct
* .. the debug kernel ought to be checking perhaps by page table walk.
*/
- if (pmap->stats.resident_count == 0)
+ if (pmap->stats.resident_count == 0) {
return TRUE;
+ }
for (offset = va_start;
- offset < va_end;
- offset += PAGE_SIZE_64) {
+ offset < va_end;
+ offset += PAGE_SIZE_64) {
phys_page = pmap_find_phys(pmap, offset);
if (phys_page) {
kprintf("pmap_is_empty(%p,0x%llx,0x%llx): "
- "page %d at 0x%llx\n",
- pmap, va_start, va_end, phys_page, offset);
+ "page %d at 0x%llx\n",
+ pmap, va_start, va_end, phys_page, offset);
return FALSE;
}
}
return TRUE;
}
+void
+hv_ept_pmap_create(void **ept_pmap, void **eptp)
+{
+ pmap_t p;
-/*
- * Create and return a physical map.
- *
- * If the size specified for the map
- * is zero, the map is an actual physical
+ if ((ept_pmap == NULL) || (eptp == NULL)) {
+ return;
+ }
+
+ p = pmap_create_options(get_task_ledger(current_task()), 0, (PMAP_CREATE_64BIT | PMAP_CREATE_EPT));
+ if (p == PMAP_NULL) {
+ *ept_pmap = NULL;
+ *eptp = NULL;
+ return;
+ }
+
+ assert(is_ept_pmap(p));
+
+ *ept_pmap = (void*)p;
+ *eptp = (void*)(p->pm_eptp);
+ return;
+}
+
+/*
+ * Create and return a physical map.
+ *
+ * If the size specified for the map
+ * is zero, the map is an actual physical
* map, and may be referenced by the
* hardware.
*
* the map will be used in software only, and
* is bounded by that size.
*/
+
pmap_t
-pmap_create(
- vm_map_size_t sz,
- boolean_t is_64bit)
+pmap_create_options(
+ ledger_t ledger,
+ vm_map_size_t sz,
+ int flags)
{
- pmap_t p;
- vm_size_t size;
+ pmap_t p;
+ vm_size_t size;
pml4_entry_t *pml4;
pml4_entry_t *kpml4;
+ int i;
- PMAP_TRACE(PMAP_CODE(PMAP__CREATE) | DBG_FUNC_START,
- (uint32_t) (sz>>32), (uint32_t) sz, is_64bit, 0, 0);
+ PMAP_TRACE(PMAP_CODE(PMAP__CREATE) | DBG_FUNC_START, sz, flags);
size = (vm_size_t) sz;
*/
if (size != 0) {
- return(PMAP_NULL);
+ return PMAP_NULL;
+ }
+
+ /*
+ * Return error when unrecognized flags are passed.
+ */
+ if (__improbable((flags & ~(PMAP_CREATE_KNOWN_FLAGS)) != 0)) {
+ return PMAP_NULL;
}
p = (pmap_t) zalloc(pmap_zone);
- if (PMAP_NULL == p)
+ if (PMAP_NULL == p) {
panic("pmap_create zalloc");
+ }
+
+ /* Zero all fields */
+ bzero(p, sizeof(*p));
+
+ lck_rw_init(&p->pmap_rwl, &pmap_lck_grp, &pmap_lck_rw_attr);
+ p->pmap_rwl.lck_rw_can_sleep = FALSE;
+
+ bzero(&p->stats, sizeof(p->stats));
- /* init counts now since we'll be bumping some */
- simple_lock_init(&p->lock, 0);
- p->stats.resident_count = 0;
- p->stats.resident_max = 0;
- p->stats.wired_count = 0;
p->ref_count = 1;
+#if DEVELOPMENT || DEBUG
p->nx_enabled = 1;
+#endif
p->pm_shared = FALSE;
+ ledger_reference(ledger);
+ p->ledger = ledger;
- p->pm_task_map = is_64bit ? TASK_MAP_64BIT : TASK_MAP_32BIT;;
+ p->pm_task_map = ((flags & PMAP_CREATE_64BIT) ? TASK_MAP_64BIT : TASK_MAP_32BIT);
+
+ p->pagezero_accessible = FALSE;
+
+ if (pmap_pcid_ncpus) {
+ pmap_pcid_initialize(p);
+ }
- /* alloc the pml4 page in kernel vm */
- if (KERN_SUCCESS != kmem_alloc_kobject(kernel_map, (vm_offset_t *)(&p->pm_pml4), PAGE_SIZE))
- panic("pmap_create kmem_alloc_kobject pml4");
+ p->pm_pml4 = zalloc(pmap_anchor_zone);
+ p->pm_upml4 = zalloc(pmap_uanchor_zone); //cleanup for EPT
- memset((char *)p->pm_pml4, 0, PAGE_SIZE);
- p->pm_cr3 = (pmap_paddr_t)kvtophys((vm_offset_t)p->pm_pml4);
+ pmap_assert((((uintptr_t)p->pm_pml4) & PAGE_MASK) == 0);
+ pmap_assert((((uintptr_t)p->pm_upml4) & PAGE_MASK) == 0);
- OSAddAtomic(1, &inuse_ptepages_count);
+ memset((char *)p->pm_pml4, 0, PAGE_SIZE);
+ memset((char *)p->pm_upml4, 0, PAGE_SIZE);
+
+ if (flags & PMAP_CREATE_EPT) {
+ p->pm_eptp = (pmap_paddr_t)kvtophys((vm_offset_t)p->pm_pml4) | pmap_eptp_flags;
+ p->pm_cr3 = 0;
+ } else {
+ p->pm_eptp = 0;
+ p->pm_cr3 = (pmap_paddr_t)kvtophys((vm_offset_t)p->pm_pml4);
+ p->pm_ucr3 = (pmap_paddr_t)kvtophys((vm_offset_t)p->pm_upml4);
+ }
/* allocate the vm_objs to hold the pdpt, pde and pte pages */
- p->pm_obj_pml4 = vm_object_allocate((vm_object_size_t)(NPML4PGS));
- if (NULL == p->pm_obj_pml4)
+ p->pm_obj_pml4 = vm_object_allocate((vm_object_size_t)(NPML4PGS) *PAGE_SIZE);
+ if (NULL == p->pm_obj_pml4) {
panic("pmap_create pdpt obj");
+ }
- p->pm_obj_pdpt = vm_object_allocate((vm_object_size_t)(NPDPTPGS));
- if (NULL == p->pm_obj_pdpt)
+ p->pm_obj_pdpt = vm_object_allocate((vm_object_size_t)(NPDPTPGS) *PAGE_SIZE);
+ if (NULL == p->pm_obj_pdpt) {
panic("pmap_create pdpt obj");
+ }
- p->pm_obj = vm_object_allocate((vm_object_size_t)(NPDEPGS));
- if (NULL == p->pm_obj)
+ p->pm_obj = vm_object_allocate((vm_object_size_t)(NPDEPGS) *PAGE_SIZE);
+ if (NULL == p->pm_obj) {
panic("pmap_create pte obj");
+ }
+
+ if (!(flags & PMAP_CREATE_EPT)) {
+ /* All host pmaps share the kernel's pml4 */
+ pml4 = pmap64_pml4(p, 0ULL);
+ kpml4 = kernel_pmap->pm_pml4;
+ for (i = KERNEL_PML4_INDEX; i < (KERNEL_PML4_INDEX + KERNEL_PML4_COUNT); i++) {
+ pml4[i] = kpml4[i];
+ }
+ pml4[KERNEL_KEXTS_INDEX] = kpml4[KERNEL_KEXTS_INDEX];
+ for (i = KERNEL_PHYSMAP_PML4_INDEX; i < (KERNEL_PHYSMAP_PML4_INDEX + KERNEL_PHYSMAP_PML4_COUNT); i++) {
+ pml4[i] = kpml4[i];
+ }
+ pml4[KERNEL_DBLMAP_PML4_INDEX] = kpml4[KERNEL_DBLMAP_PML4_INDEX];
+#if KASAN
+ for (i = KERNEL_KASAN_PML4_FIRST; i <= KERNEL_KASAN_PML4_LAST; i++) {
+ pml4[i] = kpml4[i];
+ }
+#endif
+ pml4_entry_t *pml4u = pmap64_user_pml4(p, 0ULL);
+ pml4u[KERNEL_DBLMAP_PML4_INDEX] = kpml4[KERNEL_DBLMAP_PML4_INDEX];
+ }
+
+#if MACH_ASSERT
+ p->pmap_stats_assert = TRUE;
+ p->pmap_pid = 0;
+ strlcpy(p->pmap_procname, "<nil>", sizeof(p->pmap_procname));
+#endif /* MACH_ASSERT */
+
+ PMAP_TRACE(PMAP_CODE(PMAP__CREATE) | DBG_FUNC_END,
+ VM_KERNEL_ADDRHIDE(p));
+
+ return p;
+}
- /* All pmaps share the kennel's pml4 */
- pml4 = pmap64_pml4(p, 0ULL);
- kpml4 = kernel_pmap->pm_pml4;
- pml4[KERNEL_PML4_INDEX] = kpml4[KERNEL_PML4_INDEX];
- pml4[KERNEL_KEXTS_INDEX] = kpml4[KERNEL_KEXTS_INDEX];
- pml4[KERNEL_PHYSMAP_INDEX] = kpml4[KERNEL_PHYSMAP_INDEX];
+pmap_t
+pmap_create(
+ ledger_t ledger,
+ vm_map_size_t sz,
+ boolean_t is_64bit)
+{
+ return pmap_create_options(ledger, sz, ((is_64bit) ? PMAP_CREATE_64BIT : 0));
+}
- PMAP_TRACE(PMAP_CODE(PMAP__CREATE) | DBG_FUNC_START,
- p, is_64bit, 0, 0, 0);
+/*
+ * We maintain stats and ledgers so that a task's physical footprint is:
+ * phys_footprint = ((internal - alternate_accounting)
+ * + (internal_compressed - alternate_accounting_compressed)
+ * + iokit_mapped
+ * + purgeable_nonvolatile
+ * + purgeable_nonvolatile_compressed
+ * + page_table)
+ * where "alternate_accounting" includes "iokit" and "purgeable" memory.
+ */
- return(p);
+#if MACH_ASSERT
+struct {
+ uint64_t num_pmaps_checked;
+
+ int phys_footprint_over;
+ ledger_amount_t phys_footprint_over_total;
+ ledger_amount_t phys_footprint_over_max;
+ int phys_footprint_under;
+ ledger_amount_t phys_footprint_under_total;
+ ledger_amount_t phys_footprint_under_max;
+
+ int internal_over;
+ ledger_amount_t internal_over_total;
+ ledger_amount_t internal_over_max;
+ int internal_under;
+ ledger_amount_t internal_under_total;
+ ledger_amount_t internal_under_max;
+
+ int internal_compressed_over;
+ ledger_amount_t internal_compressed_over_total;
+ ledger_amount_t internal_compressed_over_max;
+ int internal_compressed_under;
+ ledger_amount_t internal_compressed_under_total;
+ ledger_amount_t internal_compressed_under_max;
+
+ int iokit_mapped_over;
+ ledger_amount_t iokit_mapped_over_total;
+ ledger_amount_t iokit_mapped_over_max;
+ int iokit_mapped_under;
+ ledger_amount_t iokit_mapped_under_total;
+ ledger_amount_t iokit_mapped_under_max;
+
+ int alternate_accounting_over;
+ ledger_amount_t alternate_accounting_over_total;
+ ledger_amount_t alternate_accounting_over_max;
+ int alternate_accounting_under;
+ ledger_amount_t alternate_accounting_under_total;
+ ledger_amount_t alternate_accounting_under_max;
+
+ int alternate_accounting_compressed_over;
+ ledger_amount_t alternate_accounting_compressed_over_total;
+ ledger_amount_t alternate_accounting_compressed_over_max;
+ int alternate_accounting_compressed_under;
+ ledger_amount_t alternate_accounting_compressed_under_total;
+ ledger_amount_t alternate_accounting_compressed_under_max;
+
+ int page_table_over;
+ ledger_amount_t page_table_over_total;
+ ledger_amount_t page_table_over_max;
+ int page_table_under;
+ ledger_amount_t page_table_under_total;
+ ledger_amount_t page_table_under_max;
+
+ int purgeable_volatile_over;
+ ledger_amount_t purgeable_volatile_over_total;
+ ledger_amount_t purgeable_volatile_over_max;
+ int purgeable_volatile_under;
+ ledger_amount_t purgeable_volatile_under_total;
+ ledger_amount_t purgeable_volatile_under_max;
+
+ int purgeable_nonvolatile_over;
+ ledger_amount_t purgeable_nonvolatile_over_total;
+ ledger_amount_t purgeable_nonvolatile_over_max;
+ int purgeable_nonvolatile_under;
+ ledger_amount_t purgeable_nonvolatile_under_total;
+ ledger_amount_t purgeable_nonvolatile_under_max;
+
+ int purgeable_volatile_compressed_over;
+ ledger_amount_t purgeable_volatile_compressed_over_total;
+ ledger_amount_t purgeable_volatile_compressed_over_max;
+ int purgeable_volatile_compressed_under;
+ ledger_amount_t purgeable_volatile_compressed_under_total;
+ ledger_amount_t purgeable_volatile_compressed_under_max;
+
+ int purgeable_nonvolatile_compressed_over;
+ ledger_amount_t purgeable_nonvolatile_compressed_over_total;
+ ledger_amount_t purgeable_nonvolatile_compressed_over_max;
+ int purgeable_nonvolatile_compressed_under;
+ ledger_amount_t purgeable_nonvolatile_compressed_under_total;
+ ledger_amount_t purgeable_nonvolatile_compressed_under_max;
+
+ int network_volatile_over;
+ ledger_amount_t network_volatile_over_total;
+ ledger_amount_t network_volatile_over_max;
+ int network_volatile_under;
+ ledger_amount_t network_volatile_under_total;
+ ledger_amount_t network_volatile_under_max;
+
+ int network_nonvolatile_over;
+ ledger_amount_t network_nonvolatile_over_total;
+ ledger_amount_t network_nonvolatile_over_max;
+ int network_nonvolatile_under;
+ ledger_amount_t network_nonvolatile_under_total;
+ ledger_amount_t network_nonvolatile_under_max;
+
+ int network_volatile_compressed_over;
+ ledger_amount_t network_volatile_compressed_over_total;
+ ledger_amount_t network_volatile_compressed_over_max;
+ int network_volatile_compressed_under;
+ ledger_amount_t network_volatile_compressed_under_total;
+ ledger_amount_t network_volatile_compressed_under_max;
+
+ int network_nonvolatile_compressed_over;
+ ledger_amount_t network_nonvolatile_compressed_over_total;
+ ledger_amount_t network_nonvolatile_compressed_over_max;
+ int network_nonvolatile_compressed_under;
+ ledger_amount_t network_nonvolatile_compressed_under_total;
+ ledger_amount_t network_nonvolatile_compressed_under_max;
+} pmap_ledgers_drift;
+static void pmap_check_ledgers(pmap_t pmap);
+#else /* MACH_ASSERT */
+static inline void
+pmap_check_ledgers(__unused pmap_t pmap)
+{
}
+#endif /* MACH_ASSERT */
/*
* Retire the given physical map from service.
* Should only be called if the map contains
* no valid mappings.
*/
+extern int vm_wired_objects_page_count;
void
-pmap_destroy(
- register pmap_t p)
+pmap_destroy(pmap_t p)
{
- register int c;
+ int c;
- if (p == PMAP_NULL)
+ if (p == PMAP_NULL) {
return;
+ }
PMAP_TRACE(PMAP_CODE(PMAP__DESTROY) | DBG_FUNC_START,
- p, 0, 0, 0, 0);
+ VM_KERNEL_ADDRHIDe(p));
- PMAP_LOCK(p);
+ PMAP_LOCK_EXCLUSIVE(p);
c = --p->ref_count;
+ pmap_assert((current_thread() && (current_thread()->map)) ? (current_thread()->map->pmap != p) : TRUE);
+
if (c == 0) {
- /*
+ /*
* If some cpu is not using the physical pmap pointer that it
* is supposed to be (see set_dirbase), we might be using the
* pmap that is being destroyed! Make sure we are
* physically on the right pmap:
*/
PMAP_UPDATE_TLBS(p, 0x0ULL, 0xFFFFFFFFFFFFF000ULL);
+ if (pmap_pcid_ncpus) {
+ pmap_destroy_pcid_sync(p);
+ }
}
- PMAP_UNLOCK(p);
+ PMAP_UNLOCK_EXCLUSIVE(p);
if (c != 0) {
- PMAP_TRACE(PMAP_CODE(PMAP__DESTROY) | DBG_FUNC_END,
- p, 1, 0, 0, 0);
- return; /* still in use */
+ PMAP_TRACE(PMAP_CODE(PMAP__DESTROY) | DBG_FUNC_END);
+ pmap_assert(p == kernel_pmap);
+ return; /* still in use */
}
/*
*/
int inuse_ptepages = 0;
- inuse_ptepages++;
- kmem_free(kernel_map, (vm_offset_t)p->pm_pml4, PAGE_SIZE);
+ zfree(pmap_anchor_zone, p->pm_pml4);
+ zfree(pmap_uanchor_zone, p->pm_upml4);
inuse_ptepages += p->pm_obj_pml4->resident_page_count;
vm_object_deallocate(p->pm_obj_pml4);
inuse_ptepages += p->pm_obj->resident_page_count;
vm_object_deallocate(p->pm_obj);
- OSAddAtomic(-inuse_ptepages, &inuse_ptepages_count);
+ OSAddAtomic(-inuse_ptepages, &inuse_ptepages_count);
+ PMAP_ZINFO_PFREE(p, inuse_ptepages * PAGE_SIZE);
+ pmap_check_ledgers(p);
+ ledger_dereference(p->ledger);
zfree(pmap_zone, p);
- PMAP_TRACE(PMAP_CODE(PMAP__DESTROY) | DBG_FUNC_END,
- 0, 0, 0, 0, 0);
+ PMAP_TRACE(PMAP_CODE(PMAP__DESTROY) | DBG_FUNC_END);
}
/*
*/
void
-pmap_reference(pmap_t p)
+pmap_reference(pmap_t p)
{
if (p != PMAP_NULL) {
- PMAP_LOCK(p);
+ PMAP_LOCK_EXCLUSIVE(p);
p->ref_count++;
- PMAP_UNLOCK(p);;
+ PMAP_UNLOCK_EXCLUSIVE(p);;
}
}
-/*
- * Remove a range of hardware page-table entries.
- * The entries given are the first (inclusive)
- * and last (exclusive) entries for the VM pages.
- * The virtual address is the va for the first pte.
- *
- * The pmap must be locked.
- * If the pmap is not the kernel pmap, the range must lie
- * entirely within one pte-page. This is NOT checked.
- * Assumes that the pte-page exists.
- */
-
-void
-pmap_remove_range(
- pmap_t pmap,
- vm_map_offset_t start_vaddr,
- pt_entry_t *spte,
- pt_entry_t *epte)
-{
- pt_entry_t *cpte;
- pv_hashed_entry_t pvh_et = PV_HASHED_ENTRY_NULL;
- pv_hashed_entry_t pvh_eh = PV_HASHED_ENTRY_NULL;
- pv_hashed_entry_t pvh_e;
- int pvh_cnt = 0;
- int num_removed, num_unwired, num_found;
- int pai;
- pmap_paddr_t pa;
- vm_map_offset_t vaddr;
-
- num_removed = 0;
- num_unwired = 0;
- num_found = 0;
-
- /* invalidate the PTEs first to "freeze" them */
- for (cpte = spte, vaddr = start_vaddr;
- cpte < epte;
- cpte++, vaddr += PAGE_SIZE_64) {
-
- pa = pte_to_pa(*cpte);
- if (pa == 0)
- continue;
- num_found++;
-
- if (iswired(*cpte))
- num_unwired++;
-
- pai = pa_index(pa);
-
- if (!IS_MANAGED_PAGE(pai)) {
- /*
- * Outside range of managed physical memory.
- * Just remove the mappings.
- */
- pmap_store_pte(cpte, 0);
- continue;
- }
-
- /* invalidate the PTE */
- pmap_update_pte(cpte, *cpte, (*cpte & ~INTEL_PTE_VALID));
- }
-
- if (num_found == 0) {
- /* nothing was changed: we're done */
- goto update_counts;
- }
-
- /* propagate the invalidates to other CPUs */
-
- PMAP_UPDATE_TLBS(pmap, start_vaddr, vaddr);
-
- for (cpte = spte, vaddr = start_vaddr;
- cpte < epte;
- cpte++, vaddr += PAGE_SIZE_64) {
-
- pa = pte_to_pa(*cpte);
- if (pa == 0)
- continue;
-
- pai = pa_index(pa);
-
- LOCK_PVH(pai);
-
- pa = pte_to_pa(*cpte);
- if (pa == 0) {
- UNLOCK_PVH(pai);
- continue;
- }
- num_removed++;
-
- /*
- * Get the modify and reference bits, then
- * nuke the entry in the page table
- */
- /* remember reference and change */
- pmap_phys_attributes[pai] |=
- (char) (*cpte & (PHYS_MODIFIED | PHYS_REFERENCED));
- /* completely invalidate the PTE */
- pmap_store_pte(cpte, 0);
-
- /*
- * Remove the mapping from the pvlist for this physical page.
- */
- pvh_e = pmap_pv_remove(pmap, vaddr, (ppnum_t) pai);
-
- UNLOCK_PVH(pai);
-
- if (pvh_e != PV_HASHED_ENTRY_NULL) {
- pvh_e->qlink.next = (queue_entry_t) pvh_eh;
- pvh_eh = pvh_e;
-
- if (pvh_et == PV_HASHED_ENTRY_NULL) {
- pvh_et = pvh_e;
- }
- pvh_cnt++;
- }
- } /* for loop */
-
- if (pvh_eh != PV_HASHED_ENTRY_NULL) {
- PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pvh_cnt);
- }
-update_counts:
- /*
- * Update the counts
- */
-#if TESTING
- if (pmap->stats.resident_count < num_removed)
- panic("pmap_remove_range: resident_count");
-#endif
- assert(pmap->stats.resident_count >= num_removed);
- OSAddAtomic(-num_removed, &pmap->stats.resident_count);
-
-#if TESTING
- if (pmap->stats.wired_count < num_unwired)
- panic("pmap_remove_range: wired_count");
-#endif
- assert(pmap->stats.wired_count >= num_unwired);
- OSAddAtomic(-num_unwired, &pmap->stats.wired_count);
-
- return;
-}
-
/*
* Remove phys addr if mapped in specified map
*
*/
void
pmap_remove_some_phys(
- __unused pmap_t map,
+ __unused pmap_t map,
__unused ppnum_t pn)
{
-
/* Implement to support working set code */
-
}
-/*
- * Remove the given range of addresses
- * from the specified map.
- *
- * It is assumed that the start and end are properly
- * rounded to the hardware page size.
- */
-void
-pmap_remove(
- pmap_t map,
- addr64_t s64,
- addr64_t e64)
-{
- pt_entry_t *pde;
- pt_entry_t *spte, *epte;
- addr64_t l64;
- uint64_t deadline;
-
- pmap_intr_assert();
-
- if (map == PMAP_NULL || s64 == e64)
- return;
-
- PMAP_TRACE(PMAP_CODE(PMAP__REMOVE) | DBG_FUNC_START,
- map,
- (uint32_t) (s64 >> 32), s64,
- (uint32_t) (e64 >> 32), e64);
-
-
- PMAP_LOCK(map);
-
-#if 0
- /*
- * Check that address range in the kernel does not overlap the stacks.
- * We initialize local static min/max variables once to avoid making
- * 2 function calls for every remove. Note also that these functions
- * both return 0 before kernel stacks have been initialized, and hence
- * the panic is not triggered in this case.
- */
- if (map == kernel_pmap) {
- static vm_offset_t kernel_stack_min = 0;
- static vm_offset_t kernel_stack_max = 0;
-
- if (kernel_stack_min == 0) {
- kernel_stack_min = min_valid_stack_address();
- kernel_stack_max = max_valid_stack_address();
- }
- if ((kernel_stack_min <= s64 && s64 < kernel_stack_max) ||
- (kernel_stack_min < e64 && e64 <= kernel_stack_max))
- panic("pmap_remove() attempted in kernel stack");
- }
-#else
-
- /*
- * The values of kernel_stack_min and kernel_stack_max are no longer
- * relevant now that we allocate kernel stacks in the kernel map,
- * so the old code above no longer applies. If we wanted to check that
- * we weren't removing a mapping of a page in a kernel stack we'd
- * mark the PTE with an unused bit and check that here.
- */
-
-#endif
- deadline = rdtsc64() + max_preemption_latency_tsc;
-
- while (s64 < e64) {
- l64 = (s64 + pde_mapped_size) & ~(pde_mapped_size - 1);
- if (l64 > e64)
- l64 = e64;
- pde = pmap_pde(map, s64);
-
- if (pde && (*pde & INTEL_PTE_VALID)) {
- if (*pde & INTEL_PTE_PS) {
- /*
- * If we're removing a superpage, pmap_remove_range()
- * must work on level 2 instead of level 1; and we're
- * only passing a single level 2 entry instead of a
- * level 1 range.
- */
- spte = pde;
- epte = spte+1; /* excluded */
- } else {
- spte = pmap_pte(map, (s64 & ~(pde_mapped_size - 1)));
- spte = &spte[ptenum(s64)];
- epte = &spte[intel_btop(l64 - s64)];
- }
- pmap_remove_range(map, s64, spte, epte);
- }
- s64 = l64;
- pde++;
-
- if (s64 < e64 && rdtsc64() >= deadline) {
- PMAP_UNLOCK(map)
- PMAP_LOCK(map)
- deadline = rdtsc64() + max_preemption_latency_tsc;
- }
- }
-
- PMAP_UNLOCK(map);
-
- PMAP_TRACE(PMAP_CODE(PMAP__REMOVE) | DBG_FUNC_END,
- map, 0, 0, 0, 0);
-
-}
-
-/*
- * Routine: pmap_page_protect
- *
- * Function:
- * Lower the permission for all mappings to a given
- * page.
- */
void
-pmap_page_protect(
- ppnum_t pn,
- vm_prot_t prot)
+pmap_protect(
+ pmap_t map,
+ vm_map_offset_t sva,
+ vm_map_offset_t eva,
+ vm_prot_t prot)
{
- pv_hashed_entry_t pvh_eh = PV_HASHED_ENTRY_NULL;
- pv_hashed_entry_t pvh_et = PV_HASHED_ENTRY_NULL;
- pv_hashed_entry_t nexth;
- int pvh_cnt = 0;
- pv_rooted_entry_t pv_h;
- pv_rooted_entry_t pv_e;
- pv_hashed_entry_t pvh_e;
- pt_entry_t *pte;
- int pai;
- pmap_t pmap;
- boolean_t remove;
-
- pmap_intr_assert();
- assert(pn != vm_page_fictitious_addr);
- if (pn == vm_page_guard_addr)
- return;
-
- pai = ppn_to_pai(pn);
-
- if (!IS_MANAGED_PAGE(pai)) {
- /*
- * Not a managed page.
- */
- return;
- }
- PMAP_TRACE(PMAP_CODE(PMAP__PAGE_PROTECT) | DBG_FUNC_START,
- pn, prot, 0, 0, 0);
-
- /*
- * Determine the new protection.
- */
- switch (prot) {
- case VM_PROT_READ:
- case VM_PROT_READ | VM_PROT_EXECUTE:
- remove = FALSE;
- break;
- case VM_PROT_ALL:
- return; /* nothing to do */
- default:
- remove = TRUE;
- break;
- }
-
- pv_h = pai_to_pvh(pai);
-
- LOCK_PVH(pai);
-
-
- /*
- * Walk down PV list, if any, changing or removing all mappings.
- */
- if (pv_h->pmap == PMAP_NULL)
- goto done;
-
- pv_e = pv_h;
- pvh_e = (pv_hashed_entry_t) pv_e; /* cheat */
-
- do {
- vm_map_offset_t vaddr;
-
- pmap = pv_e->pmap;
- vaddr = pv_e->va;
- pte = pmap_pte(pmap, vaddr);
- if (0 == pte) {
- panic("pmap_page_protect() "
- "pmap=%p pn=0x%x vaddr=0x%llx\n",
- pmap, pn, vaddr);
- }
- nexth = (pv_hashed_entry_t) queue_next(&pvh_e->qlink);
-
- /*
- * Remove the mapping if new protection is NONE
- * or if write-protecting a kernel mapping.
- */
- if (remove || pmap == kernel_pmap) {
- /*
- * Remove the mapping, collecting dirty bits.
- */
- pmap_update_pte(pte, *pte, *pte & ~INTEL_PTE_VALID);
- PMAP_UPDATE_TLBS(pmap, vaddr, vaddr+PAGE_SIZE);
- pmap_phys_attributes[pai] |=
- *pte & (PHYS_MODIFIED|PHYS_REFERENCED);
- pmap_store_pte(pte, 0);
-
-#if TESTING
- if (pmap->stats.resident_count < 1)
- panic("pmap_page_protect: resident_count");
-#endif
- assert(pmap->stats.resident_count >= 1);
- OSAddAtomic(-1, &pmap->stats.resident_count);
-
- /*
- * Deal with the pv_rooted_entry.
- */
-
- if (pv_e == pv_h) {
- /*
- * Fix up head later.
- */
- pv_h->pmap = PMAP_NULL;
- } else {
- /*
- * Delete this entry.
- */
- pv_hash_remove(pvh_e);
- pvh_e->qlink.next = (queue_entry_t) pvh_eh;
- pvh_eh = pvh_e;
-
- if (pvh_et == PV_HASHED_ENTRY_NULL)
- pvh_et = pvh_e;
- pvh_cnt++;
- }
- } else {
- /*
- * Write-protect.
- */
- pmap_update_pte(pte, *pte, *pte & ~INTEL_PTE_WRITE);
- PMAP_UPDATE_TLBS(pmap, vaddr, vaddr+PAGE_SIZE);
- }
- pvh_e = nexth;
- } while ((pv_e = (pv_rooted_entry_t) nexth) != pv_h);
-
-
- /*
- * If pv_head mapping was removed, fix it up.
- */
- if (pv_h->pmap == PMAP_NULL) {
- pvh_e = (pv_hashed_entry_t) queue_next(&pv_h->qlink);
-
- if (pvh_e != (pv_hashed_entry_t) pv_h) {
- pv_hash_remove(pvh_e);
- pv_h->pmap = pvh_e->pmap;
- pv_h->va = pvh_e->va;
- pvh_e->qlink.next = (queue_entry_t) pvh_eh;
- pvh_eh = pvh_e;
-
- if (pvh_et == PV_HASHED_ENTRY_NULL)
- pvh_et = pvh_e;
- pvh_cnt++;
- }
- }
- if (pvh_eh != PV_HASHED_ENTRY_NULL) {
- PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pvh_cnt);
- }
-done:
- UNLOCK_PVH(pai);
-
- PMAP_TRACE(PMAP_CODE(PMAP__PAGE_PROTECT) | DBG_FUNC_END,
- 0, 0, 0, 0, 0);
+ pmap_protect_options(map, sva, eva, prot, 0, NULL);
}
-/*
- * Routine:
- * pmap_disconnect
- *
- * Function:
- * Disconnect all mappings for this page and return reference and change status
- * in generic format.
- *
- */
-unsigned int pmap_disconnect(
- ppnum_t pa)
-{
- pmap_page_protect(pa, 0); /* disconnect the page */
- return (pmap_get_refmod(pa)); /* return ref/chg status */
-}
-
/*
* Set the physical protection on the
* specified range of this map as requested.
- * Will not increase permissions.
+ *
+ * VERY IMPORTANT: Will *NOT* increase permissions.
+ * pmap_protect_options() should protect the range against any access types
+ * that are not in "prot" but it should never grant extra access.
+ * For example, if "prot" is READ|EXECUTE, that means "remove write
+ * access" but it does *not* mean "add read and execute" access.
+ * VM relies on getting soft-faults to enforce extra checks (code
+ * signing, for example), for example.
+ * New access permissions are granted via pmap_enter() only.
*/
void
-pmap_protect(
- pmap_t map,
- vm_map_offset_t sva,
- vm_map_offset_t eva,
- vm_prot_t prot)
-{
- pt_entry_t *pde;
- pt_entry_t *spte, *epte;
+pmap_protect_options(
+ pmap_t map,
+ vm_map_offset_t sva,
+ vm_map_offset_t eva,
+ vm_prot_t prot,
+ unsigned int options,
+ void *arg)
+{
+ pt_entry_t *pde;
+ pt_entry_t *spte, *epte;
vm_map_offset_t lva;
vm_map_offset_t orig_sva;
boolean_t set_NX;
int num_found = 0;
+ boolean_t is_ept;
pmap_intr_assert();
- if (map == PMAP_NULL)
+ if (map == PMAP_NULL) {
return;
+ }
if (prot == VM_PROT_NONE) {
- pmap_remove(map, sva, eva);
+ pmap_remove_options(map, sva, eva, options);
return;
}
+
PMAP_TRACE(PMAP_CODE(PMAP__PROTECT) | DBG_FUNC_START,
- map,
- (uint32_t) (sva >> 32), (uint32_t) sva,
- (uint32_t) (eva >> 32), (uint32_t) eva);
+ VM_KERNEL_ADDRHIDE(map), VM_KERNEL_ADDRHIDE(sva),
+ VM_KERNEL_ADDRHIDE(eva));
- if ((prot & VM_PROT_EXECUTE) || !nx_enabled || !map->nx_enabled)
+ if (prot & VM_PROT_EXECUTE) {
set_NX = FALSE;
- else
+ } else {
set_NX = TRUE;
+ }
+
+#if DEVELOPMENT || DEBUG
+ if (__improbable(set_NX && (!nx_enabled || !map->nx_enabled))) {
+ set_NX = FALSE;
+ }
+#endif
+ is_ept = is_ept_pmap(map);
- PMAP_LOCK(map);
+ PMAP_LOCK_EXCLUSIVE(map);
orig_sva = sva;
while (sva < eva) {
- lva = (sva + pde_mapped_size) & ~(pde_mapped_size - 1);
- if (lva > eva)
+ lva = (sva + PDE_MAPPED_SIZE) & ~(PDE_MAPPED_SIZE - 1);
+ if (lva > eva) {
lva = eva;
+ }
pde = pmap_pde(map, sva);
- if (pde && (*pde & INTEL_PTE_VALID)) {
- if (*pde & INTEL_PTE_PS) {
+ if (pde && (*pde & PTE_VALID_MASK(is_ept))) {
+ if (*pde & PTE_PS) {
/* superpage */
spte = pde;
- epte = spte+1; /* excluded */
+ epte = spte + 1; /* excluded */
} else {
- spte = pmap_pte(map, (sva & ~(pde_mapped_size - 1)));
+ spte = pmap_pte(map, (sva & ~(PDE_MAPPED_SIZE - 1)));
spte = &spte[ptenum(sva)];
epte = &spte[intel_btop(lva - sva)];
}
for (; spte < epte; spte++) {
- if (!(*spte & INTEL_PTE_VALID))
+ if (!(*spte & PTE_VALID_MASK(is_ept))) {
continue;
+ }
- if (prot & VM_PROT_WRITE)
- pmap_update_pte(spte, *spte,
- *spte | INTEL_PTE_WRITE);
- else
- pmap_update_pte(spte, *spte,
- *spte & ~INTEL_PTE_WRITE);
-
- if (set_NX)
- pmap_update_pte(spte, *spte,
- *spte | INTEL_PTE_NX);
- else
- pmap_update_pte(spte, *spte,
- *spte & ~INTEL_PTE_NX);
-
- num_found++;
- }
- }
- sva = lva;
- }
- if (num_found)
- PMAP_UPDATE_TLBS(map, orig_sva, eva);
-
- PMAP_UNLOCK(map);
-
- PMAP_TRACE(PMAP_CODE(PMAP__PROTECT) | DBG_FUNC_END,
- 0, 0, 0, 0, 0);
-
-}
-
-/* Map a (possibly) autogenned block */
-void
-pmap_map_block(
- pmap_t pmap,
- addr64_t va,
- ppnum_t pa,
- uint32_t size,
- vm_prot_t prot,
- int attr,
- __unused unsigned int flags)
-{
- uint32_t page;
- int cur_page_size;
-
- if (attr & VM_MEM_SUPERPAGE)
- cur_page_size = SUPERPAGE_SIZE;
- else
- cur_page_size = PAGE_SIZE;
-
- for (page = 0; page < size; page+=cur_page_size/PAGE_SIZE) {
- pmap_enter(pmap, va, pa, prot, attr, TRUE);
- va += cur_page_size;
- pa+=cur_page_size/PAGE_SIZE;
- }
-}
-
-
-/*
- * Insert the given physical page (p) at
- * the specified virtual address (v) in the
- * target physical map with the protection requested.
- *
- * If specified, the page will be wired down, meaning
- * that the related pte cannot be reclaimed.
- *
- * NB: This is the only routine which MAY NOT lazy-evaluate
- * or lose information. That is, this routine must actually
- * insert this page into the given map NOW.
- */
-void
-pmap_enter(
- register pmap_t pmap,
- vm_map_offset_t vaddr,
- ppnum_t pn,
- vm_prot_t prot,
- unsigned int flags,
- boolean_t wired)
-{
- pt_entry_t *pte;
- pv_rooted_entry_t pv_h;
- int pai;
- pv_hashed_entry_t pvh_e;
- pv_hashed_entry_t pvh_new;
- pt_entry_t template;
- pmap_paddr_t old_pa;
- pmap_paddr_t pa = (pmap_paddr_t) i386_ptob(pn);
- boolean_t need_tlbflush = FALSE;
- boolean_t set_NX;
- char oattr;
- boolean_t old_pa_locked;
- boolean_t superpage = flags & VM_MEM_SUPERPAGE;
- vm_object_t delpage_pm_obj = NULL;
- int delpage_pde_index = 0;
-
-
- pmap_intr_assert();
- assert(pn != vm_page_fictitious_addr);
- if (pmap_debug)
- kprintf("pmap_enter(%p,%llu,%u)\n", pmap, vaddr, pn);
- if (pmap == PMAP_NULL)
- return;
- if (pn == vm_page_guard_addr)
- return;
-
- PMAP_TRACE(PMAP_CODE(PMAP__ENTER) | DBG_FUNC_START,
- pmap,
- (uint32_t) (vaddr >> 32), (uint32_t) vaddr,
- pn, prot);
-
- if ((prot & VM_PROT_EXECUTE) || !nx_enabled || !pmap->nx_enabled)
- set_NX = FALSE;
- else
- set_NX = TRUE;
-
- /*
- * Must allocate a new pvlist entry while we're unlocked;
- * zalloc may cause pageout (which will lock the pmap system).
- * If we determine we need a pvlist entry, we will unlock
- * and allocate one. Then we will retry, throughing away
- * the allocated entry later (if we no longer need it).
- */
-
- pvh_new = PV_HASHED_ENTRY_NULL;
-Retry:
- pvh_e = PV_HASHED_ENTRY_NULL;
-
- PMAP_LOCK(pmap);
-
- /*
- * Expand pmap to include this pte. Assume that
- * pmap is always expanded to include enough hardware
- * pages to map one VM page.
- */
- if(superpage) {
- while ((pte = pmap64_pde(pmap, vaddr)) == PD_ENTRY_NULL) {
- /* need room for another pde entry */
- PMAP_UNLOCK(pmap);
- pmap_expand_pdpt(pmap, vaddr);
- PMAP_LOCK(pmap);
- }
- } else {
- while ((pte = pmap_pte(pmap, vaddr)) == PT_ENTRY_NULL) {
- /*
- * Must unlock to expand the pmap
- * going to grow pde level page(s)
- */
- PMAP_UNLOCK(pmap);
- pmap_expand(pmap, vaddr);
- PMAP_LOCK(pmap);
- }
- }
-
- if (superpage && *pte && !(*pte & INTEL_PTE_PS)) {
- /*
- * There is still an empty page table mapped that
- * was used for a previous base page mapping.
- * Remember the PDE and the PDE index, so that we
- * can free the page at the end of this function.
- */
- delpage_pde_index = (int)pdeidx(pmap, vaddr);
- delpage_pm_obj = pmap->pm_obj;
- *pte = 0;
- }
-
- old_pa = pte_to_pa(*pte);
- pai = pa_index(old_pa);
- old_pa_locked = FALSE;
-
- /*
- * if we have a previous managed page, lock the pv entry now. after
- * we lock it, check to see if someone beat us to the lock and if so
- * drop the lock
- */
- if ((0 != old_pa) && IS_MANAGED_PAGE(pai)) {
- LOCK_PVH(pai);
- old_pa_locked = TRUE;
- old_pa = pte_to_pa(*pte);
- if (0 == old_pa) {
- UNLOCK_PVH(pai); /* another path beat us to it */
- old_pa_locked = FALSE;
- }
- }
-
- /*
- * Special case if the incoming physical page is already mapped
- * at this address.
- */
- if (old_pa == pa) {
-
- /*
- * May be changing its wired attribute or protection
- */
-
- template = pa_to_pte(pa) | INTEL_PTE_VALID;
-
- if (VM_MEM_NOT_CACHEABLE ==
- (flags & (VM_MEM_NOT_CACHEABLE | VM_WIMG_USE_DEFAULT))) {
- if (!(flags & VM_MEM_GUARDED))
- template |= INTEL_PTE_PTA;
- template |= INTEL_PTE_NCACHE;
- }
- if (pmap != kernel_pmap)
- template |= INTEL_PTE_USER;
- if (prot & VM_PROT_WRITE)
- template |= INTEL_PTE_WRITE;
-
- if (set_NX)
- template |= INTEL_PTE_NX;
-
- if (wired) {
- template |= INTEL_PTE_WIRED;
- if (!iswired(*pte))
- OSAddAtomic(+1,
- &pmap->stats.wired_count);
- } else {
- if (iswired(*pte)) {
- assert(pmap->stats.wired_count >= 1);
- OSAddAtomic(-1,
- &pmap->stats.wired_count);
- }
- }
- if (superpage) /* this path can not be used */
- template |= INTEL_PTE_PS; /* to change the page size! */
-
- /* store modified PTE and preserve RC bits */
- pmap_update_pte(pte, *pte,
- template | (*pte & (INTEL_PTE_REF | INTEL_PTE_MOD)));
- if (old_pa_locked) {
- UNLOCK_PVH(pai);
- old_pa_locked = FALSE;
- }
- need_tlbflush = TRUE;
- goto Done;
- }
-
- /*
- * Outline of code from here:
- * 1) If va was mapped, update TLBs, remove the mapping
- * and remove old pvlist entry.
- * 2) Add pvlist entry for new mapping
- * 3) Enter new mapping.
- *
- * If the old physical page is not managed step 1) is skipped
- * (except for updating the TLBs), and the mapping is
- * overwritten at step 3). If the new physical page is not
- * managed, step 2) is skipped.
- */
-
- if (old_pa != (pmap_paddr_t) 0) {
-
- /*
- * Don't do anything to pages outside valid memory here.
- * Instead convince the code that enters a new mapping
- * to overwrite the old one.
- */
-
- /* invalidate the PTE */
- pmap_update_pte(pte, *pte, (*pte & ~INTEL_PTE_VALID));
- /* propagate invalidate everywhere */
- PMAP_UPDATE_TLBS(pmap, vaddr, vaddr + PAGE_SIZE);
- /* remember reference and change */
- oattr = (char) (*pte & (PHYS_MODIFIED | PHYS_REFERENCED));
- /* completely invalidate the PTE */
- pmap_store_pte(pte, 0);
-
- if (IS_MANAGED_PAGE(pai)) {
-#if TESTING
- if (pmap->stats.resident_count < 1)
- panic("pmap_enter: resident_count");
-#endif
- assert(pmap->stats.resident_count >= 1);
- OSAddAtomic(-1,
- &pmap->stats.resident_count);
-
- if (iswired(*pte)) {
-#if TESTING
- if (pmap->stats.wired_count < 1)
- panic("pmap_enter: wired_count");
-#endif
- assert(pmap->stats.wired_count >= 1);
- OSAddAtomic(-1,
- &pmap->stats.wired_count);
- }
- pmap_phys_attributes[pai] |= oattr;
-
- /*
- * Remove the mapping from the pvlist for
- * this physical page.
- * We'll end up with either a rooted pv or a
- * hashed pv
- */
- pvh_e = pmap_pv_remove(pmap, vaddr, (ppnum_t) pai);
-
- } else {
-
- /*
- * old_pa is not managed.
- * Do removal part of accounting.
- */
-
- if (iswired(*pte)) {
- assert(pmap->stats.wired_count >= 1);
- OSAddAtomic(-1,
- &pmap->stats.wired_count);
- }
- }
- }
-
- /*
- * if we had a previously managed paged locked, unlock it now
- */
- if (old_pa_locked) {
- UNLOCK_PVH(pai);
- old_pa_locked = FALSE;
- }
-
- pai = pa_index(pa); /* now working with new incoming phys page */
- if (IS_MANAGED_PAGE(pai)) {
-
- /*
- * Step 2) Enter the mapping in the PV list for this
- * physical page.
- */
- pv_h = pai_to_pvh(pai);
-
- LOCK_PVH(pai);
+ if (is_ept) {
+ if (!(prot & VM_PROT_READ)) {
+ pmap_update_pte(spte, PTE_READ(is_ept), 0);
+ }
+ }
+ if (!(prot & VM_PROT_WRITE)) {
+ pmap_update_pte(spte, PTE_WRITE(is_ept), 0);
+ }
+#if DEVELOPMENT || DEBUG
+ else if ((options & PMAP_OPTIONS_PROTECT_IMMEDIATE) &&
+ map == kernel_pmap) {
+ pmap_update_pte(spte, 0, PTE_WRITE(is_ept));
+ }
+#endif /* DEVELOPMENT || DEBUG */
- if (pv_h->pmap == PMAP_NULL) {
- /*
- * No mappings yet, use rooted pv
- */
- pv_h->va = vaddr;
- pv_h->pmap = pmap;
- queue_init(&pv_h->qlink);
- } else {
- /*
- * Add new pv_hashed_entry after header.
- */
- if ((PV_HASHED_ENTRY_NULL == pvh_e) && pvh_new) {
- pvh_e = pvh_new;
- pvh_new = PV_HASHED_ENTRY_NULL;
- } else if (PV_HASHED_ENTRY_NULL == pvh_e) {
- PV_HASHED_ALLOC(pvh_e);
- if (PV_HASHED_ENTRY_NULL == pvh_e) {
- /*
- * the pv list is empty. if we are on
- * the kernel pmap we'll use one of
- * the special private kernel pv_e's,
- * else, we need to unlock
- * everything, zalloc a pv_e, and
- * restart bringing in the pv_e with
- * us.
- */
- if (kernel_pmap == pmap) {
- PV_HASHED_KERN_ALLOC(pvh_e);
+ if (set_NX) {
+ if (!is_ept) {
+ pmap_update_pte(spte, 0, INTEL_PTE_NX);
} else {
- UNLOCK_PVH(pai);
- PMAP_UNLOCK(pmap);
- pvh_new = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
- goto Retry;
+ pmap_update_pte(spte, INTEL_EPT_EX, 0);
}
}
+ num_found++;
}
- if (PV_HASHED_ENTRY_NULL == pvh_e)
- panic("pvh_e exhaustion");
-
- pvh_e->va = vaddr;
- pvh_e->pmap = pmap;
- pvh_e->ppn = pn;
- pv_hash_add(pvh_e, pv_h);
-
- /*
- * Remember that we used the pvlist entry.
- */
- pvh_e = PV_HASHED_ENTRY_NULL;
}
-
- /*
- * only count the mapping
- * for 'managed memory'
- */
- OSAddAtomic(+1, & pmap->stats.resident_count);
- if (pmap->stats.resident_count > pmap->stats.resident_max) {
- pmap->stats.resident_max = pmap->stats.resident_count;
+ sva = lva;
+ }
+ if (num_found) {
+ if (options & PMAP_OPTIONS_NOFLUSH) {
+ PMAP_UPDATE_TLBS_DELAYED(map, orig_sva, eva, (pmap_flush_context *)arg);
+ } else {
+ PMAP_UPDATE_TLBS(map, orig_sva, eva);
}
}
- /*
- * Step 3) Enter the mapping.
- *
- * Build a template to speed up entering -
- * only the pfn changes.
- */
- template = pa_to_pte(pa) | INTEL_PTE_VALID;
-
- if (flags & VM_MEM_NOT_CACHEABLE) {
- if (!(flags & VM_MEM_GUARDED))
- template |= INTEL_PTE_PTA;
- template |= INTEL_PTE_NCACHE;
- }
- if (pmap != kernel_pmap)
- template |= INTEL_PTE_USER;
- if (prot & VM_PROT_WRITE)
- template |= INTEL_PTE_WRITE;
- if (set_NX)
- template |= INTEL_PTE_NX;
- if (wired) {
- template |= INTEL_PTE_WIRED;
- OSAddAtomic(+1, & pmap->stats.wired_count);
- }
- if (superpage)
- template |= INTEL_PTE_PS;
- pmap_store_pte(pte, template);
- /*
- * if this was a managed page we delayed unlocking the pv until here
- * to prevent pmap_page_protect et al from finding it until the pte
- * has been stored
- */
- if (IS_MANAGED_PAGE(pai)) {
- UNLOCK_PVH(pai);
- }
-Done:
- if (need_tlbflush == TRUE)
- PMAP_UPDATE_TLBS(pmap, vaddr, vaddr + PAGE_SIZE);
+ PMAP_UNLOCK_EXCLUSIVE(map);
- if (pvh_e != PV_HASHED_ENTRY_NULL) {
- PV_HASHED_FREE_LIST(pvh_e, pvh_e, 1);
- }
- if (pvh_new != PV_HASHED_ENTRY_NULL) {
- PV_HASHED_KERN_FREE_LIST(pvh_new, pvh_new, 1);
- }
- PMAP_UNLOCK(pmap);
+ PMAP_TRACE(PMAP_CODE(PMAP__PROTECT) | DBG_FUNC_END);
+}
- if (delpage_pm_obj) {
- vm_page_t m;
+/* Map a (possibly) autogenned block */
+kern_return_t
+pmap_map_block(
+ pmap_t pmap,
+ addr64_t va,
+ ppnum_t pa,
+ uint32_t size,
+ vm_prot_t prot,
+ int attr,
+ __unused unsigned int flags)
+{
+ kern_return_t kr;
+ addr64_t original_va = va;
+ uint32_t page;
+ int cur_page_size;
- vm_object_lock(delpage_pm_obj);
- m = vm_page_lookup(delpage_pm_obj, delpage_pde_index);
- if (m == VM_PAGE_NULL)
- panic("pmap_enter: pte page not in object");
- VM_PAGE_FREE(m);
- OSAddAtomic(-1, &inuse_ptepages_count);
- vm_object_unlock(delpage_pm_obj);
+ if (attr & VM_MEM_SUPERPAGE) {
+ cur_page_size = SUPERPAGE_SIZE;
+ } else {
+ cur_page_size = PAGE_SIZE;
}
- PMAP_TRACE(PMAP_CODE(PMAP__ENTER) | DBG_FUNC_END, 0, 0, 0, 0, 0);
-}
-
-/*
- * Routine: pmap_change_wiring
- * Function: Change the wiring attribute for a map/virtual-address
- * pair.
- * In/out conditions:
- * The mapping must already exist in the pmap.
- */
-void
-pmap_change_wiring(
- pmap_t map,
- vm_map_offset_t vaddr,
- boolean_t wired)
-{
- pt_entry_t *pte;
+ for (page = 0; page < size; page += cur_page_size / PAGE_SIZE) {
+ kr = pmap_enter(pmap, va, pa, prot, VM_PROT_NONE, attr, TRUE);
- PMAP_LOCK(map);
+ if (kr != KERN_SUCCESS) {
+ /*
+ * This will panic for now, as it is unclear that
+ * removing the mappings is correct.
+ */
+ panic("%s: failed pmap_enter, "
+ "pmap=%p, va=%#llx, pa=%u, size=%u, prot=%#x, flags=%#x",
+ __FUNCTION__,
+ pmap, va, pa, size, prot, flags);
- if ((pte = pmap_pte(map, vaddr)) == PT_ENTRY_NULL)
- panic("pmap_change_wiring: pte missing");
+ pmap_remove(pmap, original_va, va - original_va);
+ return kr;
+ }
- if (wired && !iswired(*pte)) {
- /*
- * wiring down mapping
- */
- OSAddAtomic(+1, &map->stats.wired_count);
- pmap_update_pte(pte, *pte, (*pte | INTEL_PTE_WIRED));
- }
- else if (!wired && iswired(*pte)) {
- /*
- * unwiring mapping
- */
- assert(map->stats.wired_count >= 1);
- OSAddAtomic(-1, &map->stats.wired_count);
- pmap_update_pte(pte, *pte, (*pte & ~INTEL_PTE_WIRED));
+ va += cur_page_size;
+ pa += cur_page_size / PAGE_SIZE;
}
- PMAP_UNLOCK(map);
+ return KERN_SUCCESS;
}
-void
+kern_return_t
pmap_expand_pml4(
- pmap_t map,
- vm_map_offset_t vaddr)
+ pmap_t map,
+ vm_map_offset_t vaddr,
+ unsigned int options)
{
- vm_page_t m;
- pmap_paddr_t pa;
- uint64_t i;
- ppnum_t pn;
- pml4_entry_t *pml4p;
+ vm_page_t m;
+ pmap_paddr_t pa;
+ uint64_t i;
+ ppnum_t pn;
+ pml4_entry_t *pml4p;
+ boolean_t is_ept = is_ept_pmap(map);
DBG("pmap_expand_pml4(%p,%p)\n", map, (void *)vaddr);
+ /* With the exception of the kext "basement", the kernel's level 4
+ * pagetables must not be dynamically expanded.
+ */
+ assert(map != kernel_pmap || (vaddr == KERNEL_BASEMENT));
/*
* Allocate a VM page for the pml4 page
*/
- while ((m = vm_page_grab()) == VM_PAGE_NULL)
+ while ((m = vm_page_grab()) == VM_PAGE_NULL) {
+ if (options & PMAP_EXPAND_OPTIONS_NOWAIT) {
+ return KERN_RESOURCE_SHORTAGE;
+ }
VM_PAGE_WAIT();
-
+ }
/*
* put the page into the pmap's obj list so it
* can be found later.
*/
- pn = m->phys_page;
+ pn = VM_PAGE_GET_PHYS_PAGE(m);
pa = i386_ptob(pn);
i = pml4idx(map, vaddr);
pmap_zero_page(pn);
vm_page_lockspin_queues();
- vm_page_wire(m);
+ vm_page_wire(m, VM_KERN_MEMORY_PTE, TRUE);
vm_page_unlock_queues();
- OSAddAtomic(1, &inuse_ptepages_count);
+ OSAddAtomic(1, &inuse_ptepages_count);
+ OSAddAtomic64(1, &alloc_ptepages_count);
+ PMAP_ZINFO_PALLOC(map, PAGE_SIZE);
/* Take the oject lock (mutex) before the PMAP_LOCK (spinlock) */
vm_object_lock(map->pm_obj_pml4);
- PMAP_LOCK(map);
+ PMAP_LOCK_EXCLUSIVE(map);
/*
* See if someone else expanded us first
*/
if (pmap64_pdpt(map, vaddr) != PDPT_ENTRY_NULL) {
- PMAP_UNLOCK(map);
+ PMAP_UNLOCK_EXCLUSIVE(map);
vm_object_unlock(map->pm_obj_pml4);
VM_PAGE_FREE(m);
- OSAddAtomic(-1, &inuse_ptepages_count);
- return;
+ OSAddAtomic(-1, &inuse_ptepages_count);
+ PMAP_ZINFO_PFREE(map, PAGE_SIZE);
+ return KERN_SUCCESS;
}
#if 0 /* DEBUG */
- if (0 != vm_page_lookup(map->pm_obj_pml4, (vm_object_offset_t)i)) {
- panic("pmap_expand_pml4: obj not empty, pmap %p pm_obj %p vaddr 0x%llx i 0x%llx\n",
- map, map->pm_obj_pml4, vaddr, i);
- }
+ if (0 != vm_page_lookup(map->pm_obj_pml4, (vm_object_offset_t)i * PAGE_SIZE)) {
+ panic("pmap_expand_pml4: obj not empty, pmap %p pm_obj %p vaddr 0x%llx i 0x%llx\n",
+ map, map->pm_obj_pml4, vaddr, i);
+ }
#endif
- vm_page_insert(m, map->pm_obj_pml4, (vm_object_offset_t)i);
+ vm_page_insert_wired(m, map->pm_obj_pml4, (vm_object_offset_t)i * PAGE_SIZE, VM_KERN_MEMORY_PTE);
vm_object_unlock(map->pm_obj_pml4);
/*
pml4p = pmap64_pml4(map, vaddr); /* refetch under lock */
pmap_store_pte(pml4p, pa_to_pte(pa)
- | INTEL_PTE_VALID
- | INTEL_PTE_USER
- | INTEL_PTE_WRITE);
+ | PTE_READ(is_ept)
+ | (is_ept ? INTEL_EPT_EX : INTEL_PTE_USER)
+ | PTE_WRITE(is_ept));
+ pml4_entry_t *upml4p;
- PMAP_UNLOCK(map);
+ upml4p = pmap64_user_pml4(map, vaddr);
+ pmap_store_pte(upml4p, pa_to_pte(pa)
+ | PTE_READ(is_ept)
+ | (is_ept ? INTEL_EPT_EX : INTEL_PTE_USER)
+ | PTE_WRITE(is_ept));
- return;
+ PMAP_UNLOCK_EXCLUSIVE(map);
+
+ return KERN_SUCCESS;
}
-void
-pmap_expand_pdpt(
- pmap_t map,
- vm_map_offset_t vaddr)
+kern_return_t
+pmap_expand_pdpt(pmap_t map, vm_map_offset_t vaddr, unsigned int options)
{
- vm_page_t m;
- pmap_paddr_t pa;
- uint64_t i;
- ppnum_t pn;
- pdpt_entry_t *pdptp;
+ vm_page_t m;
+ pmap_paddr_t pa;
+ uint64_t i;
+ ppnum_t pn;
+ pdpt_entry_t *pdptp;
+ boolean_t is_ept = is_ept_pmap(map);
DBG("pmap_expand_pdpt(%p,%p)\n", map, (void *)vaddr);
while ((pdptp = pmap64_pdpt(map, vaddr)) == PDPT_ENTRY_NULL) {
- pmap_expand_pml4(map, vaddr);
+ kern_return_t pep4kr = pmap_expand_pml4(map, vaddr, options);
+ if (pep4kr != KERN_SUCCESS) {
+ return pep4kr;
+ }
}
/*
* Allocate a VM page for the pdpt page
*/
- while ((m = vm_page_grab()) == VM_PAGE_NULL)
+ while ((m = vm_page_grab()) == VM_PAGE_NULL) {
+ if (options & PMAP_EXPAND_OPTIONS_NOWAIT) {
+ return KERN_RESOURCE_SHORTAGE;
+ }
VM_PAGE_WAIT();
+ }
/*
* put the page into the pmap's obj list so it
* can be found later.
*/
- pn = m->phys_page;
+ pn = VM_PAGE_GET_PHYS_PAGE(m);
pa = i386_ptob(pn);
i = pdptidx(map, vaddr);
pmap_zero_page(pn);
vm_page_lockspin_queues();
- vm_page_wire(m);
+ vm_page_wire(m, VM_KERN_MEMORY_PTE, TRUE);
vm_page_unlock_queues();
- OSAddAtomic(1, &inuse_ptepages_count);
+ OSAddAtomic(1, &inuse_ptepages_count);
+ OSAddAtomic64(1, &alloc_ptepages_count);
+ PMAP_ZINFO_PALLOC(map, PAGE_SIZE);
/* Take the oject lock (mutex) before the PMAP_LOCK (spinlock) */
vm_object_lock(map->pm_obj_pdpt);
- PMAP_LOCK(map);
+ PMAP_LOCK_EXCLUSIVE(map);
/*
* See if someone else expanded us first
*/
- if (pmap64_pde(map, vaddr) != PD_ENTRY_NULL) {
- PMAP_UNLOCK(map);
+ if (pmap_pde(map, vaddr) != PD_ENTRY_NULL) {
+ PMAP_UNLOCK_EXCLUSIVE(map);
vm_object_unlock(map->pm_obj_pdpt);
VM_PAGE_FREE(m);
- OSAddAtomic(-1, &inuse_ptepages_count);
- return;
+ OSAddAtomic(-1, &inuse_ptepages_count);
+ PMAP_ZINFO_PFREE(map, PAGE_SIZE);
+ return KERN_SUCCESS;
}
#if 0 /* DEBUG */
- if (0 != vm_page_lookup(map->pm_obj_pdpt, (vm_object_offset_t)i)) {
- panic("pmap_expand_pdpt: obj not empty, pmap %p pm_obj %p vaddr 0x%llx i 0x%llx\n",
- map, map->pm_obj_pdpt, vaddr, i);
- }
+ if (0 != vm_page_lookup(map->pm_obj_pdpt, (vm_object_offset_t)i * PAGE_SIZE)) {
+ panic("pmap_expand_pdpt: obj not empty, pmap %p pm_obj %p vaddr 0x%llx i 0x%llx\n",
+ map, map->pm_obj_pdpt, vaddr, i);
+ }
#endif
- vm_page_insert(m, map->pm_obj_pdpt, (vm_object_offset_t)i);
+ vm_page_insert_wired(m, map->pm_obj_pdpt, (vm_object_offset_t)i * PAGE_SIZE, VM_KERN_MEMORY_PTE);
vm_object_unlock(map->pm_obj_pdpt);
/*
pdptp = pmap64_pdpt(map, vaddr); /* refetch under lock */
pmap_store_pte(pdptp, pa_to_pte(pa)
- | INTEL_PTE_VALID
- | INTEL_PTE_USER
- | INTEL_PTE_WRITE);
-
- PMAP_UNLOCK(map);
+ | PTE_READ(is_ept)
+ | (is_ept ? INTEL_EPT_EX : INTEL_PTE_USER)
+ | PTE_WRITE(is_ept));
- return;
+ PMAP_UNLOCK_EXCLUSIVE(map);
+ return KERN_SUCCESS;
}
* has been expanded enough.
* (We won't loop forever, since page tables aren't shrunk.)
*/
-void
+kern_return_t
pmap_expand(
- pmap_t map,
- vm_map_offset_t vaddr)
-{
- pt_entry_t *pdp;
- register vm_page_t m;
- register pmap_paddr_t pa;
- uint64_t i;
+ pmap_t map,
+ vm_map_offset_t vaddr,
+ unsigned int options)
+{
+ pt_entry_t *pdp;
+ vm_page_t m;
+ pmap_paddr_t pa;
+ uint64_t i;
ppnum_t pn;
+ boolean_t is_ept = is_ept_pmap(map);
/*
- * For the kernel, the virtual address must be in or above the basement
+ * For the kernel, the virtual address must be in or above the basement
* which is for kexts and is in the 512GB immediately below the kernel..
* XXX - should use VM_MIN_KERNEL_AND_KEXT_ADDRESS not KERNEL_BASEMENT
*/
- if (map == kernel_pmap &&
- !(vaddr >= KERNEL_BASEMENT && vaddr <= VM_MAX_KERNEL_ADDRESS))
- panic("pmap_expand: bad vaddr 0x%llx for kernel pmap", vaddr);
-
+ if (__improbable(map == kernel_pmap &&
+ !(vaddr >= KERNEL_BASEMENT && vaddr <= VM_MAX_KERNEL_ADDRESS))) {
+ if ((options & PMAP_EXPAND_OPTIONS_ALIASMAP) == 0) {
+ panic("pmap_expand: bad vaddr 0x%llx for kernel pmap", vaddr);
+ }
+ }
- while ((pdp = pmap64_pde(map, vaddr)) == PD_ENTRY_NULL) {
- /* need room for another pde entry */
- pmap_expand_pdpt(map, vaddr);
+ while ((pdp = pmap_pde(map, vaddr)) == PD_ENTRY_NULL) {
+ assert((options & PMAP_EXPAND_OPTIONS_ALIASMAP) == 0);
+ kern_return_t pepkr = pmap_expand_pdpt(map, vaddr, options);
+ if (pepkr != KERN_SUCCESS) {
+ return pepkr;
+ }
}
/*
* Allocate a VM page for the pde entries.
*/
- while ((m = vm_page_grab()) == VM_PAGE_NULL)
+ while ((m = vm_page_grab()) == VM_PAGE_NULL) {
+ if (options & PMAP_EXPAND_OPTIONS_NOWAIT) {
+ return KERN_RESOURCE_SHORTAGE;
+ }
VM_PAGE_WAIT();
+ }
/*
* put the page into the pmap's obj list so it
* can be found later.
*/
- pn = m->phys_page;
+ pn = VM_PAGE_GET_PHYS_PAGE(m);
pa = i386_ptob(pn);
i = pdeidx(map, vaddr);
pmap_zero_page(pn);
vm_page_lockspin_queues();
- vm_page_wire(m);
+ vm_page_wire(m, VM_KERN_MEMORY_PTE, TRUE);
vm_page_unlock_queues();
- OSAddAtomic(1, &inuse_ptepages_count);
+ OSAddAtomic(1, &inuse_ptepages_count);
+ OSAddAtomic64(1, &alloc_ptepages_count);
+ PMAP_ZINFO_PALLOC(map, PAGE_SIZE);
/* Take the oject lock (mutex) before the PMAP_LOCK (spinlock) */
vm_object_lock(map->pm_obj);
- PMAP_LOCK(map);
+ PMAP_LOCK_EXCLUSIVE(map);
/*
* See if someone else expanded us first
*/
if (pmap_pte(map, vaddr) != PT_ENTRY_NULL) {
- PMAP_UNLOCK(map);
+ PMAP_UNLOCK_EXCLUSIVE(map);
vm_object_unlock(map->pm_obj);
VM_PAGE_FREE(m);
- OSAddAtomic(-1, &inuse_ptepages_count);
- return;
+ OSAddAtomic(-1, &inuse_ptepages_count); //todo replace all with inlines
+ PMAP_ZINFO_PFREE(map, PAGE_SIZE);
+ return KERN_SUCCESS;
}
#if 0 /* DEBUG */
- if (0 != vm_page_lookup(map->pm_obj, (vm_object_offset_t)i)) {
- panic("pmap_expand: obj not empty, pmap 0x%x pm_obj 0x%x vaddr 0x%llx i 0x%llx\n",
- map, map->pm_obj, vaddr, i);
- }
+ if (0 != vm_page_lookup(map->pm_obj, (vm_object_offset_t)i * PAGE_SIZE)) {
+ panic("pmap_expand: obj not empty, pmap 0x%x pm_obj 0x%x vaddr 0x%llx i 0x%llx\n",
+ map, map->pm_obj, vaddr, i);
+ }
#endif
- vm_page_insert(m, map->pm_obj, (vm_object_offset_t)i);
+ vm_page_insert_wired(m, map->pm_obj, (vm_object_offset_t)i * PAGE_SIZE, VM_KERN_MEMORY_PTE);
vm_object_unlock(map->pm_obj);
/*
*/
pdp = pmap_pde(map, vaddr);
pmap_store_pte(pdp, pa_to_pte(pa)
- | INTEL_PTE_VALID
- | INTEL_PTE_USER
- | INTEL_PTE_WRITE);
+ | PTE_READ(is_ept)
+ | (is_ept ? INTEL_EPT_EX : INTEL_PTE_USER)
+ | PTE_WRITE(is_ept));
- PMAP_UNLOCK(map);
+ PMAP_UNLOCK_EXCLUSIVE(map);
- return;
+ return KERN_SUCCESS;
}
/* On K64 machines with more than 32GB of memory, pmap_steal_memory
* that pmap_steal_memory uses, rather than calling vm_page_grab (which
* isn't available yet). */
void
-pmap_pre_expand(pmap_t pmap, vm_map_offset_t vaddr) {
+pmap_pre_expand(pmap_t pmap, vm_map_offset_t vaddr)
+{
ppnum_t pn;
- pt_entry_t *pte;
+ pt_entry_t *pte;
+ boolean_t is_ept = is_ept_pmap(pmap);
- PMAP_LOCK(pmap);
+ PMAP_LOCK_EXCLUSIVE(pmap);
- if(pmap64_pdpt(pmap, vaddr) == PDPT_ENTRY_NULL) {
- if (!pmap_next_page_k64(&pn))
+ if (pmap64_pdpt(pmap, vaddr) == PDPT_ENTRY_NULL) {
+ if (!pmap_next_page_hi(&pn)) {
panic("pmap_pre_expand");
+ }
pmap_zero_page(pn);
pte = pmap64_pml4(pmap, vaddr);
pmap_store_pte(pte, pa_to_pte(i386_ptob(pn))
- | INTEL_PTE_VALID
- | INTEL_PTE_USER
- | INTEL_PTE_WRITE);
+ | PTE_READ(is_ept)
+ | (is_ept ? INTEL_EPT_EX : INTEL_PTE_USER)
+ | PTE_WRITE(is_ept));
+
+ pte = pmap64_user_pml4(pmap, vaddr);
+
+ pmap_store_pte(pte, pa_to_pte(i386_ptob(pn))
+ | PTE_READ(is_ept)
+ | (is_ept ? INTEL_EPT_EX : INTEL_PTE_USER)
+ | PTE_WRITE(is_ept));
}
- if(pmap64_pde(pmap, vaddr) == PD_ENTRY_NULL) {
- if (!pmap_next_page_k64(&pn))
+ if (pmap_pde(pmap, vaddr) == PD_ENTRY_NULL) {
+ if (!pmap_next_page_hi(&pn)) {
panic("pmap_pre_expand");
+ }
pmap_zero_page(pn);
pte = pmap64_pdpt(pmap, vaddr);
pmap_store_pte(pte, pa_to_pte(i386_ptob(pn))
- | INTEL_PTE_VALID
- | INTEL_PTE_USER
- | INTEL_PTE_WRITE);
+ | PTE_READ(is_ept)
+ | (is_ept ? INTEL_EPT_EX : INTEL_PTE_USER)
+ | PTE_WRITE(is_ept));
}
- if(pmap_pte(pmap, vaddr) == PT_ENTRY_NULL) {
- if (!pmap_next_page_k64(&pn))
+ if (pmap_pte(pmap, vaddr) == PT_ENTRY_NULL) {
+ if (!pmap_next_page_hi(&pn)) {
panic("pmap_pre_expand");
+ }
pmap_zero_page(pn);
- pte = pmap64_pde(pmap, vaddr);
+ pte = pmap_pde(pmap, vaddr);
pmap_store_pte(pte, pa_to_pte(i386_ptob(pn))
- | INTEL_PTE_VALID
- | INTEL_PTE_USER
- | INTEL_PTE_WRITE);
+ | PTE_READ(is_ept)
+ | (is_ept ? INTEL_EPT_EX : INTEL_PTE_USER)
+ | PTE_WRITE(is_ept));
}
- PMAP_UNLOCK(pmap);
+ PMAP_UNLOCK_EXCLUSIVE(pmap);
}
/*
* pmap_sync_page_data_phys(ppnum_t pa)
- *
+ *
* Invalidates all of the instruction cache on a physical page and
* pushes any dirty data from the data cache for the same physical page
* Not required in i386.
/*
* pmap_sync_page_attributes_phys(ppnum_t pa)
- *
+ *
* Write back and invalidate all cachelines on a physical page.
*/
void
#ifdef CURRENTLY_UNUSED_AND_UNTESTED
-int collect_ref;
-int collect_unref;
+int collect_ref;
+int collect_unref;
/*
* Routine: pmap_collect
*/
void
pmap_collect(
- pmap_t p)
+ pmap_t p)
{
- register pt_entry_t *pdp, *ptp;
- pt_entry_t *eptp;
- int wired;
+ pt_entry_t *pdp, *ptp;
+ pt_entry_t *eptp;
+ int wired;
+ boolean_t is_ept;
- if (p == PMAP_NULL)
+ if (p == PMAP_NULL) {
return;
+ }
- if (p == kernel_pmap)
+ if (p == kernel_pmap) {
return;
+ }
+
+ is_ept = is_ept_pmap(p);
/*
* Garbage collect map.
PMAP_LOCK(p);
for (pdp = (pt_entry_t *)p->dirbase;
- pdp < (pt_entry_t *)&p->dirbase[(UMAXPTDI+1)];
- pdp++)
- {
- if (*pdp & INTEL_PTE_VALID) {
- if(*pdp & INTEL_PTE_REF) {
- pmap_store_pte(pdp, *pdp & ~INTEL_PTE_REF);
- collect_ref++;
- } else {
- collect_unref++;
- ptp = pmap_pte(p, pdetova(pdp - (pt_entry_t *)p->dirbase));
- eptp = ptp + NPTEPG;
+ pdp < (pt_entry_t *)&p->dirbase[(UMAXPTDI + 1)];
+ pdp++) {
+ if (*pdp & PTE_VALID_MASK(is_ept)) {
+ if (*pdp & PTE_REF(is_ept)) {
+ pmap_store_pte(pdp, *pdp & ~PTE_REF(is_ept));
+ collect_ref++;
+ } else {
+ collect_unref++;
+ ptp = pmap_pte(p, pdetova(pdp - (pt_entry_t *)p->dirbase));
+ eptp = ptp + NPTEPG;
- /*
- * If the pte page has any wired mappings, we cannot
- * free it.
- */
- wired = 0;
- {
- register pt_entry_t *ptep;
- for (ptep = ptp; ptep < eptp; ptep++) {
- if (iswired(*ptep)) {
- wired = 1;
- break;
- }
- }
- }
- if (!wired) {
- /*
- * Remove the virtual addresses mapped by this pte page.
- */
- pmap_remove_range(p,
- pdetova(pdp - (pt_entry_t *)p->dirbase),
- ptp,
- eptp);
+ /*
+ * If the pte page has any wired mappings, we cannot
+ * free it.
+ */
+ wired = 0;
+ {
+ pt_entry_t *ptep;
+ for (ptep = ptp; ptep < eptp; ptep++) {
+ if (iswired(*ptep)) {
+ wired = 1;
+ break;
+ }
+ }
+ }
+ if (!wired) {
+ /*
+ * Remove the virtual addresses mapped by this pte page.
+ */
+ pmap_remove_range(p,
+ pdetova(pdp - (pt_entry_t *)p->dirbase),
+ ptp,
+ eptp);
+
+ /*
+ * Invalidate the page directory pointer.
+ */
+ pmap_store_pte(pdp, 0x0);
- /*
- * Invalidate the page directory pointer.
- */
- pmap_store_pte(pdp, 0x0);
-
- PMAP_UNLOCK(p);
+ PMAP_UNLOCK(p);
- /*
- * And free the pte page itself.
- */
- {
- register vm_page_t m;
+ /*
+ * And free the pte page itself.
+ */
+ {
+ vm_page_t m;
- vm_object_lock(p->pm_obj);
+ vm_object_lock(p->pm_obj);
- m = vm_page_lookup(p->pm_obj,(vm_object_offset_t)(pdp - (pt_entry_t *)&p->dirbase[0]));
- if (m == VM_PAGE_NULL)
- panic("pmap_collect: pte page not in object");
+ m = vm_page_lookup(p->pm_obj, (vm_object_offset_t)(pdp - (pt_entry_t *)&p->dirbase[0]) * PAGE_SIZE);
+ if (m == VM_PAGE_NULL) {
+ panic("pmap_collect: pte page not in object");
+ }
- VM_PAGE_FREE(m);
+ vm_object_unlock(p->pm_obj);
- OSAddAtomic(-1, &inuse_ptepages_count);
+ VM_PAGE_FREE(m);
- vm_object_unlock(p->pm_obj);
- }
+ OSAddAtomic(-1, &inuse_ptepages_count);
+ PMAP_ZINFO_PFREE(p, PAGE_SIZE);
+ }
- PMAP_LOCK(p);
+ PMAP_LOCK(p);
+ }
+ }
}
- }
- }
}
PMAP_UPDATE_TLBS(p, 0x0, 0xFFFFFFFFFFFFF000ULL);
PMAP_UNLOCK(p);
return;
-
}
#endif
pmap_copy_page(ppnum_t src, ppnum_t dst)
{
bcopy_phys((addr64_t)i386_ptob(src),
- (addr64_t)i386_ptob(dst),
- PAGE_SIZE);
+ (addr64_t)i386_ptob(dst),
+ PAGE_SIZE);
}
*/
void
pmap_pageable(
- __unused pmap_t pmap,
- __unused vm_map_offset_t start_addr,
- __unused vm_map_offset_t end_addr,
- __unused boolean_t pageable)
+ __unused pmap_t pmap,
+ __unused vm_map_offset_t start_addr,
+ __unused vm_map_offset_t end_addr,
+ __unused boolean_t pageable)
{
-#ifdef lint
+#ifdef lint
pmap++; start_addr++; end_addr++; pageable++;
-#endif /* lint */
-}
-
-/*
- * Clear specified attribute bits.
- */
-void
-phys_attribute_clear(
- ppnum_t pn,
- int bits)
-{
- pv_rooted_entry_t pv_h;
- pv_hashed_entry_t pv_e;
- pt_entry_t *pte;
- int pai;
- pmap_t pmap;
-
- pmap_intr_assert();
- assert(pn != vm_page_fictitious_addr);
- if (pn == vm_page_guard_addr)
- return;
-
- pai = ppn_to_pai(pn);
-
- if (!IS_MANAGED_PAGE(pai)) {
- /*
- * Not a managed page.
- */
- return;
- }
-
-
- PMAP_TRACE(PMAP_CODE(PMAP__ATTRIBUTE_CLEAR) | DBG_FUNC_START,
- pn, bits, 0, 0, 0);
-
- pv_h = pai_to_pvh(pai);
-
- LOCK_PVH(pai);
-
- /*
- * Walk down PV list, clearing all modify or reference bits.
- * We do not have to lock the pv_list because we have
- * the entire pmap system locked.
- */
- if (pv_h->pmap != PMAP_NULL) {
- /*
- * There are some mappings.
- */
-
- pv_e = (pv_hashed_entry_t)pv_h;
-
- do {
- vm_map_offset_t va;
-
- pmap = pv_e->pmap;
- va = pv_e->va;
-
- /*
- * Clear modify and/or reference bits.
- */
- pte = pmap_pte(pmap, va);
- pmap_update_pte(pte, *pte, (*pte & ~bits));
- /* Ensure all processors using this translation
- * invalidate this TLB entry. The invalidation *must*
- * follow the PTE update, to ensure that the TLB
- * shadow of the 'D' bit (in particular) is
- * synchronized with the updated PTE.
- */
- PMAP_UPDATE_TLBS(pmap, va, va + PAGE_SIZE);
-
- pv_e = (pv_hashed_entry_t)queue_next(&pv_e->qlink);
-
- } while (pv_e != (pv_hashed_entry_t)pv_h);
- }
- pmap_phys_attributes[pai] &= ~bits;
-
- UNLOCK_PVH(pai);
-
- PMAP_TRACE(PMAP_CODE(PMAP__ATTRIBUTE_CLEAR) | DBG_FUNC_END,
- 0, 0, 0, 0, 0);
-}
-
-/*
- * Check specified attribute bits.
- */
-int
-phys_attribute_test(
- ppnum_t pn,
- int bits)
-{
- pv_rooted_entry_t pv_h;
- pv_hashed_entry_t pv_e;
- pt_entry_t *pte;
- int pai;
- pmap_t pmap;
- int attributes = 0;
-
- pmap_intr_assert();
- assert(pn != vm_page_fictitious_addr);
- if (pn == vm_page_guard_addr)
- return 0;
-
- pai = ppn_to_pai(pn);
-
- if (!IS_MANAGED_PAGE(pai)) {
- /*
- * Not a managed page.
- */
- return 0;
- }
-
- /*
- * super fast check... if bits already collected
- * no need to take any locks...
- * if not set, we need to recheck after taking
- * the lock in case they got pulled in while
- * we were waiting for the lock
- */
- if ((pmap_phys_attributes[pai] & bits) == bits)
- return bits;
-
- pv_h = pai_to_pvh(pai);
-
- LOCK_PVH(pai);
-
- attributes = pmap_phys_attributes[pai] & bits;
-
-
- /*
- * Walk down PV list, checking the mappings until we
- * reach the end or we've found the attributes we've asked for
- * We do not have to lock the pv_list because we have
- * the entire pmap system locked.
- */
- if (attributes != bits &&
- pv_h->pmap != PMAP_NULL) {
- /*
- * There are some mappings.
- */
- pv_e = (pv_hashed_entry_t)pv_h;
- do {
- vm_map_offset_t va;
-
- pmap = pv_e->pmap;
- va = pv_e->va;
- /*
- * first make sure any processor actively
- * using this pmap, flushes its TLB state
- */
- PMAP_UPDATE_TLBS(pmap, va, va + PAGE_SIZE);
-
- /*
- * pick up modify and/or reference bits from mapping
- */
-
- pte = pmap_pte(pmap, va);
- attributes |= (int)(*pte & bits);
-
- pv_e = (pv_hashed_entry_t)queue_next(&pv_e->qlink);
-
- } while ((attributes != bits) &&
- (pv_e != (pv_hashed_entry_t)pv_h));
- }
-
- UNLOCK_PVH(pai);
- return (attributes);
-}
-
-/*
- * Set specified attribute bits.
- */
-void
-phys_attribute_set(
- ppnum_t pn,
- int bits)
-{
- int pai;
-
- pmap_intr_assert();
- assert(pn != vm_page_fictitious_addr);
- if (pn == vm_page_guard_addr)
- return;
-
- pai = ppn_to_pai(pn);
-
- if (!IS_MANAGED_PAGE(pai)) {
- /* Not a managed page. */
- return;
- }
-
- LOCK_PVH(pai);
- pmap_phys_attributes[pai] |= bits;
- UNLOCK_PVH(pai);
-}
-
-/*
- * Set the modify bit on the specified physical page.
- */
-
-void
-pmap_set_modify(ppnum_t pn)
-{
- phys_attribute_set(pn, PHYS_MODIFIED);
-}
-
-/*
- * Clear the modify bits on the specified physical page.
- */
-
-void
-pmap_clear_modify(ppnum_t pn)
-{
- phys_attribute_clear(pn, PHYS_MODIFIED);
-}
-
-/*
- * pmap_is_modified:
- *
- * Return whether or not the specified physical page is modified
- * by any physical maps.
- */
-
-boolean_t
-pmap_is_modified(ppnum_t pn)
-{
- if (phys_attribute_test(pn, PHYS_MODIFIED))
- return TRUE;
- return FALSE;
-}
-
-/*
- * pmap_clear_reference:
- *
- * Clear the reference bit on the specified physical page.
- */
-
-void
-pmap_clear_reference(ppnum_t pn)
-{
- phys_attribute_clear(pn, PHYS_REFERENCED);
-}
-
-void
-pmap_set_reference(ppnum_t pn)
-{
- phys_attribute_set(pn, PHYS_REFERENCED);
-}
-
-/*
- * pmap_is_referenced:
- *
- * Return whether or not the specified physical page is referenced
- * by any physical maps.
- */
-
-boolean_t
-pmap_is_referenced(ppnum_t pn)
-{
- if (phys_attribute_test(pn, PHYS_REFERENCED))
- return TRUE;
- return FALSE;
-}
-
-/*
- * pmap_get_refmod(phys)
- * returns the referenced and modified bits of the specified
- * physical page.
- */
-unsigned int
-pmap_get_refmod(ppnum_t pn)
-{
- int refmod;
- unsigned int retval = 0;
-
- refmod = phys_attribute_test(pn, PHYS_MODIFIED | PHYS_REFERENCED);
-
- if (refmod & PHYS_MODIFIED)
- retval |= VM_MEM_MODIFIED;
- if (refmod & PHYS_REFERENCED)
- retval |= VM_MEM_REFERENCED;
-
- return (retval);
+#endif /* lint */
}
-/*
- * pmap_clear_refmod(phys, mask)
- * clears the referenced and modified bits as specified by the mask
- * of the specified physical page.
- */
void
-pmap_clear_refmod(ppnum_t pn, unsigned int mask)
-{
- unsigned int x86Mask;
-
- x86Mask = ( ((mask & VM_MEM_MODIFIED)? PHYS_MODIFIED : 0)
- | ((mask & VM_MEM_REFERENCED)? PHYS_REFERENCED : 0));
- phys_attribute_clear(pn, x86Mask);
-}
-
-void
-invalidate_icache(__unused vm_offset_t addr,
- __unused unsigned cnt,
- __unused int phys)
+invalidate_icache(__unused vm_offset_t addr,
+ __unused unsigned cnt,
+ __unused int phys)
{
return;
}
-void
-flush_dcache(__unused vm_offset_t addr,
- __unused unsigned count,
- __unused int phys)
+void
+flush_dcache(__unused vm_offset_t addr,
+ __unused unsigned count,
+ __unused int phys)
{
return;
}
extern kern_return_t dtrace_copyio_preflight(addr64_t);
extern kern_return_t dtrace_copyio_postflight(addr64_t);
-kern_return_t dtrace_copyio_preflight(__unused addr64_t va)
+kern_return_t
+dtrace_copyio_preflight(__unused addr64_t va)
{
thread_t thread = current_thread();
-
- if (current_map() == kernel_map)
+ uint64_t ccr3;
+ if (current_map() == kernel_map) {
return KERN_FAILURE;
- else if (get_cr3() != thread->map->pmap->pm_cr3)
+ } else if (((ccr3 = get_cr3_base()) != thread->map->pmap->pm_cr3) && (no_shared_cr3 == FALSE)) {
return KERN_FAILURE;
- else if (thread->machine.specFlags & CopyIOActive)
+ } else if (no_shared_cr3 && (ccr3 != kernel_pmap->pm_cr3)) {
return KERN_FAILURE;
- else
+ } else {
return KERN_SUCCESS;
+ }
}
-
-kern_return_t dtrace_copyio_postflight(__unused addr64_t va)
+
+kern_return_t
+dtrace_copyio_postflight(__unused addr64_t va)
{
return KERN_SUCCESS;
}
#endif /* CONFIG_DTRACE */
#include <mach_vm_debug.h>
-#if MACH_VM_DEBUG
+#if MACH_VM_DEBUG
#include <vm/vm_debug.h>
int
pmap_list_resident_pages(
- __unused pmap_t pmap,
- __unused vm_offset_t *listp,
- __unused int space)
+ __unused pmap_t pmap,
+ __unused vm_offset_t *listp,
+ __unused int space)
{
return 0;
}
-#endif /* MACH_VM_DEBUG */
-
+#endif /* MACH_VM_DEBUG */
+#if CONFIG_COREDUMP
/* temporary workaround */
boolean_t
coredumpok(__unused vm_map_t map, __unused vm_offset_t va)
pt_entry_t *ptep;
ptep = pmap_pte(map->pmap, va);
- if (0 == ptep)
+ if (0 == ptep) {
return FALSE;
- return ((*ptep & (INTEL_PTE_NCACHE | INTEL_PTE_WIRED)) != (INTEL_PTE_NCACHE | INTEL_PTE_WIRED));
+ }
+ return (*ptep & (INTEL_PTE_NCACHE | INTEL_PTE_WIRED)) != (INTEL_PTE_NCACHE | INTEL_PTE_WIRED);
#else
return TRUE;
#endif
}
-
+#endif
boolean_t
phys_page_exists(ppnum_t pn)
{
assert(pn != vm_page_fictitious_addr);
- if (!pmap_initialized)
+ if (!pmap_initialized) {
return TRUE;
+ }
- if (pn == vm_page_guard_addr)
+ if (pn == vm_page_guard_addr) {
return FALSE;
+ }
- if (!IS_MANAGED_PAGE(ppn_to_pai(pn)))
+ if (!IS_MANAGED_PAGE(ppn_to_pai(pn))) {
return FALSE;
+ }
return TRUE;
}
+
+
+void
+pmap_switch(pmap_t tpmap)
+{
+ PMAP_TRACE_CONSTANT(PMAP_CODE(PMAP__SWITCH) | DBG_FUNC_START, VM_KERNEL_ADDRHIDE(tpmap));
+ assert(ml_get_interrupts_enabled() == FALSE);
+ set_dirbase(tpmap, current_thread(), cpu_number());
+ PMAP_TRACE_CONSTANT(PMAP_CODE(PMAP__SWITCH) | DBG_FUNC_END);
+}
+
+
+/*
+ * disable no-execute capability on
+ * the specified pmap
+ */
void
-mapping_free_prime(void)
+pmap_disable_NX(__unused pmap_t pmap)
{
- int i;
- pv_hashed_entry_t pvh_e;
- pv_hashed_entry_t pvh_eh;
- pv_hashed_entry_t pvh_et;
- int pv_cnt;
+#if DEVELOPMENT || DEBUG
+ pmap->nx_enabled = 0;
+#endif
+}
- pv_cnt = 0;
- pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
- for (i = 0; i < (5 * PV_HASHED_ALLOC_CHUNK); i++) {
- pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
+void
+pt_fake_zone_init(int zone_index)
+{
+ pt_fake_zone_index = zone_index;
+}
- pvh_e->qlink.next = (queue_entry_t)pvh_eh;
- pvh_eh = pvh_e;
+void
+pt_fake_zone_info(
+ int *count,
+ vm_size_t *cur_size,
+ vm_size_t *max_size,
+ vm_size_t *elem_size,
+ vm_size_t *alloc_size,
+ uint64_t *sum_size,
+ int *collectable,
+ int *exhaustable,
+ int *caller_acct)
+{
+ *count = inuse_ptepages_count;
+ *cur_size = PAGE_SIZE * inuse_ptepages_count;
+ *max_size = PAGE_SIZE * (inuse_ptepages_count +
+ vm_page_inactive_count +
+ vm_page_active_count +
+ vm_page_free_count);
+ *elem_size = PAGE_SIZE;
+ *alloc_size = PAGE_SIZE;
+ *sum_size = alloc_ptepages_count * PAGE_SIZE;
+
+ *collectable = 1;
+ *exhaustable = 0;
+ *caller_acct = 1;
+}
- if (pvh_et == PV_HASHED_ENTRY_NULL)
- pvh_et = pvh_e;
- pv_cnt++;
- }
- PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
- pv_cnt = 0;
- pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
- for (i = 0; i < PV_HASHED_KERN_ALLOC_CHUNK; i++) {
- pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
+void
+pmap_flush_context_init(pmap_flush_context *pfc)
+{
+ pfc->pfc_cpus = 0;
+ pfc->pfc_invalid_global = 0;
+}
- pvh_e->qlink.next = (queue_entry_t)pvh_eh;
- pvh_eh = pvh_e;
+static bool
+pmap_tlbi_response(uint32_t lcpu, uint32_t rcpu, bool ngflush)
+{
+ bool responded = false;
+ bool gflushed = (cpu_datap(rcpu)->cpu_tlb_invalid_global_count !=
+ cpu_datap(lcpu)->cpu_tlb_gen_counts_global[rcpu]);
- if (pvh_et == PV_HASHED_ENTRY_NULL)
- pvh_et = pvh_e;
- pv_cnt++;
+ if (ngflush) {
+ if (gflushed) {
+ responded = true;
+ }
+ } else {
+ if (gflushed) {
+ responded = true;
+ } else {
+ bool lflushed = (cpu_datap(rcpu)->cpu_tlb_invalid_local_count !=
+ cpu_datap(lcpu)->cpu_tlb_gen_counts_local[rcpu]);
+ if (lflushed) {
+ responded = true;
+ }
+ }
}
- PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
+ if (responded == false) {
+ if ((cpu_datap(rcpu)->cpu_tlb_invalid == 0) ||
+ !CPU_CR3_IS_ACTIVE(rcpu) ||
+ !cpu_is_running(rcpu)) {
+ responded = true;
+ }
+ }
+ return responded;
}
+extern uint64_t TLBTimeOut;
void
-mapping_adjust(void)
-{
- pv_hashed_entry_t pvh_e;
- pv_hashed_entry_t pvh_eh;
- pv_hashed_entry_t pvh_et;
- int pv_cnt;
- int i;
+pmap_flush(
+ pmap_flush_context *pfc)
+{
+ unsigned int my_cpu;
+ unsigned int cpu;
+ cpumask_t cpu_bit;
+ cpumask_t cpus_to_respond = 0;
+ cpumask_t cpus_to_signal = 0;
+ cpumask_t cpus_signaled = 0;
+ boolean_t flush_self = FALSE;
+ uint64_t deadline;
+ bool need_global_flush = false;
- if (mapping_adjust_call == NULL) {
- thread_call_setup(&mapping_adjust_call_data,
- (thread_call_func_t) mapping_adjust,
- (thread_call_param_t) NULL);
- mapping_adjust_call = &mapping_adjust_call_data;
- }
+ mp_disable_preemption();
- pv_cnt = 0;
- pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
- if (pv_hashed_kern_free_count < PV_HASHED_KERN_LOW_WATER_MARK) {
- for (i = 0; i < PV_HASHED_KERN_ALLOC_CHUNK; i++) {
- pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
+ my_cpu = cpu_number();
+ cpus_to_signal = pfc->pfc_cpus;
- pvh_e->qlink.next = (queue_entry_t)pvh_eh;
- pvh_eh = pvh_e;
+ PMAP_TRACE_CONSTANT(PMAP_CODE(PMAP__FLUSH_DELAYED_TLBS) | DBG_FUNC_START,
+ NULL, cpus_to_signal);
- if (pvh_et == PV_HASHED_ENTRY_NULL)
- pvh_et = pvh_e;
- pv_cnt++;
- }
- PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
- }
+ for (cpu = 0, cpu_bit = 1; cpu < real_ncpus && cpus_to_signal; cpu++, cpu_bit <<= 1) {
+ if (cpus_to_signal & cpu_bit) {
+ cpus_to_signal &= ~cpu_bit;
- pv_cnt = 0;
- pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
- if (pv_hashed_free_count < PV_HASHED_LOW_WATER_MARK) {
- for (i = 0; i < PV_HASHED_ALLOC_CHUNK; i++) {
- pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
+ if (!cpu_is_running(cpu)) {
+ continue;
+ }
- pvh_e->qlink.next = (queue_entry_t)pvh_eh;
- pvh_eh = pvh_e;
+ if (pfc->pfc_invalid_global & cpu_bit) {
+ cpu_datap(cpu)->cpu_tlb_invalid_global = 1;
+ need_global_flush = true;
+ } else {
+ cpu_datap(cpu)->cpu_tlb_invalid_local = 1;
+ }
+ cpu_datap(my_cpu)->cpu_tlb_gen_counts_global[cpu] = cpu_datap(cpu)->cpu_tlb_invalid_global_count;
+ cpu_datap(my_cpu)->cpu_tlb_gen_counts_local[cpu] = cpu_datap(cpu)->cpu_tlb_invalid_local_count;
+ mfence();
- if (pvh_et == PV_HASHED_ENTRY_NULL)
- pvh_et = pvh_e;
- pv_cnt++;
+ if (cpu == my_cpu) {
+ flush_self = TRUE;
+ continue;
+ }
+ if (CPU_CR3_IS_ACTIVE(cpu)) {
+ cpus_to_respond |= cpu_bit;
+ i386_signal_cpu(cpu, MP_TLB_FLUSH, ASYNC);
+ }
}
- PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
}
- mappingrecurse = 0;
-}
+ cpus_signaled = cpus_to_respond;
+
+ /*
+ * Flush local tlb if required.
+ * Do this now to overlap with other processors responding.
+ */
+ if (flush_self) {
+ process_pmap_updates(NULL, (pfc->pfc_invalid_global != 0), 0ULL, ~0ULL);
+ }
+ if (cpus_to_respond) {
+ deadline = mach_absolute_time() +
+ (TLBTimeOut ? TLBTimeOut : LockTimeOut);
+ boolean_t is_timeout_traced = FALSE;
-void
-pmap_switch(pmap_t tpmap)
-{
- spl_t s;
+ /*
+ * Wait for those other cpus to acknowledge
+ */
+ while (cpus_to_respond != 0) {
+ long orig_acks = 0;
- s = splhigh(); /* Make sure interruptions are disabled */
- set_dirbase(tpmap, current_thread());
- splx(s);
-}
+ for (cpu = 0, cpu_bit = 1; cpu < real_ncpus; cpu++, cpu_bit <<= 1) {
+ bool responded = false;
+ if ((cpus_to_respond & cpu_bit) != 0) {
+ responded = pmap_tlbi_response(my_cpu, cpu, need_global_flush);
+ if (responded) {
+ cpus_to_respond &= ~cpu_bit;
+ }
+ cpu_pause();
+ }
+ if (cpus_to_respond == 0) {
+ break;
+ }
+ }
+ if (cpus_to_respond && (mach_absolute_time() > deadline)) {
+ if (machine_timeout_suspended()) {
+ continue;
+ }
+ if (TLBTimeOut == 0) {
+ if (is_timeout_traced) {
+ continue;
+ }
-/*
- * disable no-execute capability on
- * the specified pmap
- */
-void
-pmap_disable_NX(pmap_t pmap)
-{
- pmap->nx_enabled = 0;
-}
+ PMAP_TRACE_CONSTANT(PMAP_CODE(PMAP__FLUSH_TLBS_TO),
+ NULL, cpus_to_signal, cpus_to_respond);
-void
-pt_fake_zone_info(
- int *count,
- vm_size_t *cur_size,
- vm_size_t *max_size,
- vm_size_t *elem_size,
- vm_size_t *alloc_size,
- int *collectable,
- int *exhaustable)
-{
- *count = inuse_ptepages_count;
- *cur_size = PAGE_SIZE * inuse_ptepages_count;
- *max_size = PAGE_SIZE * (inuse_ptepages_count +
- vm_page_inactive_count +
- vm_page_active_count +
- vm_page_free_count);
- *elem_size = PAGE_SIZE;
- *alloc_size = PAGE_SIZE;
+ is_timeout_traced = TRUE;
+ continue;
+ }
+ orig_acks = NMIPI_acks;
+ NMIPI_panic(cpus_to_respond, TLB_FLUSH_TIMEOUT);
+ panic("Uninterruptible processor(s): CPU bitmap: 0x%llx, NMIPI acks: 0x%lx, now: 0x%lx, deadline: %llu",
+ cpus_to_respond, orig_acks, NMIPI_acks, deadline);
+ }
+ }
+ }
- *collectable = 1;
- *exhaustable = 0;
+ PMAP_TRACE_CONSTANT(PMAP_CODE(PMAP__FLUSH_DELAYED_TLBS) | DBG_FUNC_END,
+ NULL, cpus_signaled, flush_self);
+
+ mp_enable_preemption();
}
-static inline void
-pmap_cpuset_NMIPI(cpu_set cpu_mask) {
- unsigned int cpu, cpu_bit;
- uint64_t deadline;
- for (cpu = 0, cpu_bit = 1; cpu < real_ncpus; cpu++, cpu_bit <<= 1) {
- if (cpu_mask & cpu_bit)
- cpu_NMI_interrupt(cpu);
- }
- deadline = mach_absolute_time() + (LockTimeOut);
- while (mach_absolute_time() < deadline)
- cpu_pause();
+static void
+invept(void *eptp)
+{
+ struct {
+ uint64_t eptp;
+ uint64_t reserved;
+ } __attribute__((aligned(16), packed)) invept_descriptor = {(uint64_t)eptp, 0};
+
+ __asm__ volatile ("invept (%%rax), %%rcx"
+ : : "c" (PMAP_INVEPT_SINGLE_CONTEXT), "a" (&invept_descriptor)
+ : "cc", "memory");
}
/*
* - flush the local tlb if active for this pmap
* - return ... the caller will unlock the pmap
*/
+
void
-pmap_flush_tlbs(pmap_t pmap)
-{
- unsigned int cpu;
- unsigned int cpu_bit;
- cpu_set cpus_to_signal;
- unsigned int my_cpu = cpu_number();
- pmap_paddr_t pmap_cr3 = pmap->pm_cr3;
- boolean_t flush_self = FALSE;
- uint64_t deadline;
+pmap_flush_tlbs(pmap_t pmap, vm_map_offset_t startv, vm_map_offset_t endv, int options, pmap_flush_context *pfc)
+{
+ unsigned int cpu;
+ cpumask_t cpu_bit;
+ cpumask_t cpus_to_signal = 0;
+ unsigned int my_cpu = cpu_number();
+ pmap_paddr_t pmap_cr3 = pmap->pm_cr3;
+ boolean_t flush_self = FALSE;
+ uint64_t deadline;
+ boolean_t pmap_is_shared = (pmap->pm_shared || (pmap == kernel_pmap));
+ bool need_global_flush = false;
+ uint32_t event_code;
+ vm_map_offset_t event_startv, event_endv;
+ boolean_t is_ept = is_ept_pmap(pmap);
assert((processor_avail_count < 2) ||
- (ml_get_interrupts_enabled() && get_preemption_level() != 0));
+ (ml_get_interrupts_enabled() && get_preemption_level() != 0));
+
+ assert((endv - startv) >= PAGE_SIZE);
+ assert(((endv | startv) & PAGE_MASK) == 0);
+
+ if (__improbable(kdebug_enable)) {
+ if (pmap == kernel_pmap) {
+ event_code = PMAP_CODE(PMAP__FLUSH_KERN_TLBS);
+ event_startv = VM_KERNEL_UNSLIDE_OR_PERM(startv);
+ event_endv = VM_KERNEL_UNSLIDE_OR_PERM(endv);
+ } else if (__improbable(is_ept)) {
+ event_code = PMAP_CODE(PMAP__FLUSH_EPT);
+ event_startv = startv;
+ event_endv = endv;
+ } else {
+ event_code = PMAP_CODE(PMAP__FLUSH_TLBS);
+ event_startv = startv;
+ event_endv = endv;
+ }
+ }
+
+ PMAP_TRACE_CONSTANT(event_code | DBG_FUNC_START,
+ VM_KERNEL_UNSLIDE_OR_PERM(pmap), options,
+ event_startv, event_endv);
+
+ if (__improbable(is_ept)) {
+ mp_cpus_call(CPUMASK_ALL, ASYNC, invept, (void*)pmap->pm_eptp);
+ goto out;
+ }
/*
* Scan other cpus for matching active or task CR3.
* For idle cpus (with no active map) we mark them invalid but
* don't signal -- they'll check as they go busy.
*/
- cpus_to_signal = 0;
+ if (pmap_pcid_ncpus) {
+ if (pmap_is_shared) {
+ need_global_flush = true;
+ }
+ pmap_pcid_invalidate_all_cpus(pmap);
+ mfence();
+ }
+
for (cpu = 0, cpu_bit = 1; cpu < real_ncpus; cpu++, cpu_bit <<= 1) {
- if (!cpu_datap(cpu)->cpu_running)
+ if (!cpu_is_running(cpu)) {
continue;
- uint64_t cpu_active_cr3 = CPU_GET_ACTIVE_CR3(cpu);
- uint64_t cpu_task_cr3 = CPU_GET_TASK_CR3(cpu);
+ }
+ uint64_t cpu_active_cr3 = CPU_GET_ACTIVE_CR3(cpu);
+ uint64_t cpu_task_cr3 = CPU_GET_TASK_CR3(cpu);
if ((pmap_cr3 == cpu_task_cr3) ||
(pmap_cr3 == cpu_active_cr3) ||
- (pmap->pm_shared) ||
- (pmap == kernel_pmap)) {
+ (pmap_is_shared)) {
+ if (options & PMAP_DELAY_TLB_FLUSH) {
+ if (need_global_flush == true) {
+ pfc->pfc_invalid_global |= cpu_bit;
+ }
+ pfc->pfc_cpus |= cpu_bit;
+
+ continue;
+ }
+ if (need_global_flush == true) {
+ cpu_datap(my_cpu)->cpu_tlb_gen_counts_global[cpu] = cpu_datap(cpu)->cpu_tlb_invalid_global_count;
+ cpu_datap(cpu)->cpu_tlb_invalid_global = 1;
+ } else {
+ cpu_datap(my_cpu)->cpu_tlb_gen_counts_local[cpu] = cpu_datap(cpu)->cpu_tlb_invalid_local_count;
+ cpu_datap(cpu)->cpu_tlb_invalid_local = 1;
+ }
+
if (cpu == my_cpu) {
flush_self = TRUE;
continue;
}
- cpu_datap(cpu)->cpu_tlb_invalid = TRUE;
- __asm__ volatile("mfence");
+
+ mfence();
/*
* We don't need to signal processors which will flush
}
}
- PMAP_TRACE(PMAP_CODE(PMAP__FLUSH_TLBS) | DBG_FUNC_START,
- pmap, cpus_to_signal, flush_self, 0, 0);
+ if ((options & PMAP_DELAY_TLB_FLUSH)) {
+ goto out;
+ }
/*
* Flush local tlb if required.
* Do this now to overlap with other processors responding.
*/
- if (flush_self)
- flush_tlb();
+ if (flush_self) {
+ process_pmap_updates(pmap, pmap_is_shared, startv, endv);
+ }
if (cpus_to_signal) {
- cpu_set cpus_to_respond = cpus_to_signal;
+ cpumask_t cpus_to_respond = cpus_to_signal;
+
+ deadline = mach_absolute_time() +
+ (TLBTimeOut ? TLBTimeOut : LockTimeOut);
+ boolean_t is_timeout_traced = FALSE;
- deadline = mach_absolute_time() + LockTimeOut;
/*
* Wait for those other cpus to acknowledge
*/
while (cpus_to_respond != 0) {
- if (mach_absolute_time() > deadline) {
- if (mp_recent_debugger_activity())
- continue;
- if (!panic_active()) {
- pmap_tlb_flush_timeout = TRUE;
- pmap_cpuset_NMIPI(cpus_to_respond);
- }
- panic("pmap_flush_tlbs() timeout: "
- "cpu(s) failing to respond to interrupts, pmap=%p cpus_to_respond=0x%lx",
- pmap, cpus_to_respond);
- }
+ long orig_acks = 0;
for (cpu = 0, cpu_bit = 1; cpu < real_ncpus; cpu++, cpu_bit <<= 1) {
+ bool responded = false;
if ((cpus_to_respond & cpu_bit) != 0) {
- if (!cpu_datap(cpu)->cpu_running ||
- cpu_datap(cpu)->cpu_tlb_invalid == FALSE ||
- !CPU_CR3_IS_ACTIVE(cpu)) {
+ responded = pmap_tlbi_response(my_cpu, cpu, need_global_flush);
+ if (responded) {
cpus_to_respond &= ~cpu_bit;
}
cpu_pause();
}
- if (cpus_to_respond == 0)
+ if (cpus_to_respond == 0) {
break;
+ }
+ }
+ if (cpus_to_respond && (mach_absolute_time() > deadline)) {
+ if (machine_timeout_suspended()) {
+ continue;
+ }
+ if (TLBTimeOut == 0) {
+ /* cut tracepoint but don't panic */
+ if (is_timeout_traced) {
+ continue;
+ }
+
+ PMAP_TRACE_CONSTANT(PMAP_CODE(PMAP__FLUSH_TLBS_TO),
+ VM_KERNEL_UNSLIDE_OR_PERM(pmap),
+ cpus_to_signal,
+ cpus_to_respond);
+
+ is_timeout_traced = TRUE;
+ continue;
+ }
+ orig_acks = NMIPI_acks;
+ uint64_t tstamp1 = mach_absolute_time();
+ NMIPI_panic(cpus_to_respond, TLB_FLUSH_TIMEOUT);
+ uint64_t tstamp2 = mach_absolute_time();
+ panic("IPI timeout, unresponsive CPU bitmap: 0x%llx, NMIPI acks: 0x%lx, now: 0x%lx, deadline: %llu, pre-NMIPI time: 0x%llx, current: 0x%llx, global: %d",
+ cpus_to_respond, orig_acks, NMIPI_acks, deadline, tstamp1, tstamp2, need_global_flush);
}
}
}
- PMAP_TRACE(PMAP_CODE(PMAP__FLUSH_TLBS) | DBG_FUNC_END,
- pmap, cpus_to_signal, flush_self, 0, 0);
+ if (__improbable((pmap == kernel_pmap) && (flush_self != TRUE))) {
+ panic("pmap_flush_tlbs: pmap == kernel_pmap && flush_self != TRUE; kernel CR3: 0x%llX, pmap_cr3: 0x%llx, CPU active CR3: 0x%llX, CPU Task Map: %d", kernel_pmap->pm_cr3, pmap_cr3, current_cpu_datap()->cpu_active_cr3, current_cpu_datap()->cpu_task_map);
+ }
+
+out:
+ PMAP_TRACE_CONSTANT(event_code | DBG_FUNC_END,
+ VM_KERNEL_UNSLIDE_OR_PERM(pmap), cpus_to_signal,
+ event_startv, event_endv);
}
-void
-process_pmap_updates(void)
+static void
+process_pmap_updates(pmap_t p, bool pshared, addr64_t istart, addr64_t iend)
{
- assert(ml_get_interrupts_enabled() == 0 || get_preemption_level() != 0);
+ int ccpu = cpu_number();
+ bool gtlbf = false;
+
+ pmap_assert(ml_get_interrupts_enabled() == 0 ||
+ get_preemption_level() != 0);
- flush_tlb();
+ if (cpu_datap(ccpu)->cpu_tlb_invalid_global) {
+ cpu_datap(ccpu)->cpu_tlb_invalid_global_count++;
+ cpu_datap(ccpu)->cpu_tlb_invalid = 0;
+ gtlbf = true;
+ } else {
+ cpu_datap(ccpu)->cpu_tlb_invalid_local_count++;
+ cpu_datap(ccpu)->cpu_tlb_invalid_local = 0;
+ }
- current_cpu_datap()->cpu_tlb_invalid = FALSE;
- __asm__ volatile("mfence");
+ if (pmap_pcid_ncpus) {
+ if (p) {
+ /* TODO global generation count to
+ * avoid potentially redundant
+ * csw invalidations post-global invalidation
+ */
+ pmap_pcid_validate_cpu(p, ccpu);
+ pmap_tlbi_range(istart, iend, (pshared || gtlbf), p->pmap_pcid_cpus[ccpu]);
+ } else {
+ pmap_pcid_validate_current();
+ pmap_tlbi_range(istart, iend, true, 0);
+ }
+ } else {
+ pmap_tlbi_range(0, ~0ULL, true, 0);
+ }
}
void
pmap_update_interrupt(void)
{
- PMAP_TRACE(PMAP_CODE(PMAP__UPDATE_INTERRUPT) | DBG_FUNC_START,
- 0, 0, 0, 0, 0);
+ PMAP_TRACE(PMAP_CODE(PMAP__UPDATE_INTERRUPT) | DBG_FUNC_START);
+
+ if (current_cpu_datap()->cpu_tlb_invalid) {
+ process_pmap_updates(NULL, true, 0ULL, ~0ULL);
+ }
+
+ PMAP_TRACE(PMAP_CODE(PMAP__UPDATE_INTERRUPT) | DBG_FUNC_END);
+}
+
+#include <mach/mach_vm.h> /* mach_vm_region_recurse() */
+/* Scan kernel pmap for W+X PTEs, scan kernel VM map for W+X map entries
+ * and identify ranges with mismatched VM permissions and PTE permissions
+ */
+kern_return_t
+pmap_permissions_verify(pmap_t ipmap, vm_map_t ivmmap, vm_offset_t sv, vm_offset_t ev)
+{
+ vm_offset_t cv = sv;
+ kern_return_t rv = KERN_SUCCESS;
+ uint64_t skip4 = 0, skip2 = 0;
+
+ assert(!is_ept_pmap(ipmap));
+
+ sv &= ~PAGE_MASK_64;
+ ev &= ~PAGE_MASK_64;
+ while (cv < ev) {
+ if (__improbable((cv > 0x00007FFFFFFFFFFFULL) &&
+ (cv < 0xFFFF800000000000ULL))) {
+ cv = 0xFFFF800000000000ULL;
+ }
+ /* Potential inconsistencies from not holding pmap lock
+ * but harmless for the moment.
+ */
+ if (((cv & PML4MASK) == 0) && (pmap64_pml4(ipmap, cv) == 0)) {
+ if ((cv + NBPML4) > cv) {
+ cv += NBPML4;
+ } else {
+ break;
+ }
+ skip4++;
+ continue;
+ }
+ if (((cv & PDMASK) == 0) && (pmap_pde(ipmap, cv) == 0)) {
+ if ((cv + NBPD) > cv) {
+ cv += NBPD;
+ } else {
+ break;
+ }
+ skip2++;
+ continue;
+ }
+
+ pt_entry_t *ptep = pmap_pte(ipmap, cv);
+ if (ptep && (*ptep & INTEL_PTE_VALID)) {
+ if (*ptep & INTEL_PTE_WRITE) {
+ if (!(*ptep & INTEL_PTE_NX)) {
+ kprintf("W+X PTE at 0x%lx, P4: 0x%llx, P3: 0x%llx, P2: 0x%llx, PT: 0x%llx, VP: %u\n", cv, *pmap64_pml4(ipmap, cv), *pmap64_pdpt(ipmap, cv), *pmap_pde(ipmap, cv), *ptep, pmap_valid_page((ppnum_t)(i386_btop(pte_to_pa(*ptep)))));
+ rv = KERN_FAILURE;
+ }
+ }
+ }
+ cv += PAGE_SIZE;
+ }
+ kprintf("Completed pmap scan\n");
+ cv = sv;
+
+ struct vm_region_submap_info_64 vbr;
+ mach_msg_type_number_t vbrcount = 0;
+ mach_vm_size_t vmsize;
+ vm_prot_t prot;
+ uint32_t nesting_depth = 0;
+ kern_return_t kret;
+
+ while (cv < ev) {
+ for (;;) {
+ vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
+ if ((kret = mach_vm_region_recurse(ivmmap,
+ (mach_vm_address_t *) &cv, &vmsize, &nesting_depth,
+ (vm_region_recurse_info_t)&vbr,
+ &vbrcount)) != KERN_SUCCESS) {
+ break;
+ }
+
+ if (vbr.is_submap) {
+ nesting_depth++;
+ continue;
+ } else {
+ break;
+ }
+ }
+
+ if (kret != KERN_SUCCESS) {
+ break;
+ }
+
+ prot = vbr.protection;
+
+ if ((prot & (VM_PROT_WRITE | VM_PROT_EXECUTE)) == (VM_PROT_WRITE | VM_PROT_EXECUTE)) {
+ kprintf("W+X map entry at address 0x%lx\n", cv);
+ rv = KERN_FAILURE;
+ }
+
+ if (prot) {
+ vm_offset_t pcv;
+ for (pcv = cv; pcv < cv + vmsize; pcv += PAGE_SIZE) {
+ pt_entry_t *ptep = pmap_pte(ipmap, pcv);
+ vm_prot_t tprot;
+
+ if ((ptep == NULL) || !(*ptep & INTEL_PTE_VALID)) {
+ continue;
+ }
+ tprot = VM_PROT_READ;
+ if (*ptep & INTEL_PTE_WRITE) {
+ tprot |= VM_PROT_WRITE;
+ }
+ if ((*ptep & INTEL_PTE_NX) == 0) {
+ tprot |= VM_PROT_EXECUTE;
+ }
+ if (tprot != prot) {
+ kprintf("PTE/map entry permissions mismatch at address 0x%lx, pte: 0x%llx, protection: 0x%x\n", pcv, *ptep, prot);
+ rv = KERN_FAILURE;
+ }
+ }
+ }
+ cv += vmsize;
+ }
+ return rv;
+}
+
+#if MACH_ASSERT
+extern int pmap_ledgers_panic;
+extern int pmap_ledgers_panic_leeway;
+
+static void
+pmap_check_ledgers(
+ pmap_t pmap)
+{
+ ledger_amount_t bal;
+ int pid;
+ char *procname;
+ boolean_t do_panic;
+
+ if (pmap->pmap_pid == 0) {
+ /*
+ * This pmap was not or is no longer fully associated
+ * with a task (e.g. the old pmap after a fork()/exec() or
+ * spawn()). Its "ledger" still points at a task that is
+ * now using a different (and active) address space, so
+ * we can't check that all the pmap ledgers are balanced here.
+ *
+ * If the "pid" is set, that means that we went through
+ * pmap_set_process() in task_terminate_internal(), so
+ * this task's ledger should not have been re-used and
+ * all the pmap ledgers should be back to 0.
+ */
+ return;
+ }
+
+ do_panic = FALSE;
+ pid = pmap->pmap_pid;
+ procname = pmap->pmap_procname;
+
+ pmap_ledgers_drift.num_pmaps_checked++;
+
+#define LEDGER_CHECK_BALANCE(__LEDGER) \
+MACRO_BEGIN \
+ int panic_on_negative = TRUE; \
+ ledger_get_balance(pmap->ledger, \
+ task_ledgers.__LEDGER, \
+ &bal); \
+ ledger_get_panic_on_negative(pmap->ledger, \
+ task_ledgers.__LEDGER, \
+ &panic_on_negative); \
+ if (bal != 0) { \
+ if (panic_on_negative || \
+ (pmap_ledgers_panic && \
+ pmap_ledgers_panic_leeway > 0 && \
+ (bal > (pmap_ledgers_panic_leeway * PAGE_SIZE) || \
+ bal < (pmap_ledgers_panic_leeway * PAGE_SIZE)))) { \
+ do_panic = TRUE; \
+ } \
+ printf("LEDGER BALANCE proc %d (%s) " \
+ "\"%s\" = %lld\n", \
+ pid, procname, #__LEDGER, bal); \
+ if (bal > 0) { \
+ pmap_ledgers_drift.__LEDGER##_over++; \
+ pmap_ledgers_drift.__LEDGER##_over_total += bal; \
+ if (bal > pmap_ledgers_drift.__LEDGER##_over_max) { \
+ pmap_ledgers_drift.__LEDGER##_over_max = bal; \
+ } \
+ } else if (bal < 0) { \
+ pmap_ledgers_drift.__LEDGER##_under++; \
+ pmap_ledgers_drift.__LEDGER##_under_total += bal; \
+ if (bal < pmap_ledgers_drift.__LEDGER##_under_max) { \
+ pmap_ledgers_drift.__LEDGER##_under_max = bal; \
+ } \
+ } \
+ } \
+MACRO_END
+
+ LEDGER_CHECK_BALANCE(phys_footprint);
+ LEDGER_CHECK_BALANCE(internal);
+ LEDGER_CHECK_BALANCE(internal_compressed);
+ LEDGER_CHECK_BALANCE(iokit_mapped);
+ LEDGER_CHECK_BALANCE(alternate_accounting);
+ LEDGER_CHECK_BALANCE(alternate_accounting_compressed);
+ LEDGER_CHECK_BALANCE(page_table);
+ LEDGER_CHECK_BALANCE(purgeable_volatile);
+ LEDGER_CHECK_BALANCE(purgeable_nonvolatile);
+ LEDGER_CHECK_BALANCE(purgeable_volatile_compressed);
+ LEDGER_CHECK_BALANCE(purgeable_nonvolatile_compressed);
+ LEDGER_CHECK_BALANCE(network_volatile);
+ LEDGER_CHECK_BALANCE(network_nonvolatile);
+ LEDGER_CHECK_BALANCE(network_volatile_compressed);
+ LEDGER_CHECK_BALANCE(network_nonvolatile_compressed);
+
+ if (do_panic) {
+ if (pmap_ledgers_panic) {
+ panic("pmap_destroy(%p) %d[%s] has imbalanced ledgers\n",
+ pmap, pid, procname);
+ } else {
+ printf("pmap_destroy(%p) %d[%s] has imbalanced ledgers\n",
+ pmap, pid, procname);
+ }
+ }
+
+ if (pmap->stats.resident_count != 0 ||
+#if 35156815
+ /*
+ * "wired_count" is unfortunately a bit inaccurate, so let's
+ * tolerate some slight deviation to limit the amount of
+ * somewhat-spurious assertion failures.
+ */
+ pmap->stats.wired_count > 10 ||
+#else /* 35156815 */
+ pmap->stats.wired_count != 0 ||
+#endif /* 35156815 */
+ pmap->stats.device != 0 ||
+ pmap->stats.internal != 0 ||
+ pmap->stats.external != 0 ||
+ pmap->stats.reusable != 0 ||
+ pmap->stats.compressed != 0) {
+ if (pmap_stats_assert &&
+ pmap->pmap_stats_assert) {
+ panic("pmap_destroy(%p) %d[%s] imbalanced stats: resident=%d wired=%d device=%d internal=%d external=%d reusable=%d compressed=%lld",
+ pmap, pid, procname,
+ pmap->stats.resident_count,
+ pmap->stats.wired_count,
+ pmap->stats.device,
+ pmap->stats.internal,
+ pmap->stats.external,
+ pmap->stats.reusable,
+ pmap->stats.compressed);
+ } else {
+ printf("pmap_destroy(%p) %d[%s] imbalanced stats: resident=%d wired=%d device=%d internal=%d external=%d reusable=%d compressed=%lld",
+ pmap, pid, procname,
+ pmap->stats.resident_count,
+ pmap->stats.wired_count,
+ pmap->stats.device,
+ pmap->stats.internal,
+ pmap->stats.external,
+ pmap->stats.reusable,
+ pmap->stats.compressed);
+ }
+ }
+}
+
+void
+pmap_set_process(
+ pmap_t pmap,
+ int pid,
+ char *procname)
+{
+ if (pmap == NULL) {
+ return;
+ }
+
+ pmap->pmap_pid = pid;
+ strlcpy(pmap->pmap_procname, procname, sizeof(pmap->pmap_procname));
+ if (pmap_ledgers_panic_leeway) {
+ /*
+ * XXX FBDP
+ * Some processes somehow trigger some issues that make
+ * the pmap stats and ledgers go off track, causing
+ * some assertion failures and ledger panics.
+ * Turn off the sanity checks if we allow some ledger leeway
+ * because of that. We'll still do a final check in
+ * pmap_check_ledgers() for discrepancies larger than the
+ * allowed leeway after the address space has been fully
+ * cleaned up.
+ */
+ pmap->pmap_stats_assert = FALSE;
+ ledger_disable_panic_on_negative(pmap->ledger,
+ task_ledgers.phys_footprint);
+ ledger_disable_panic_on_negative(pmap->ledger,
+ task_ledgers.internal);
+ ledger_disable_panic_on_negative(pmap->ledger,
+ task_ledgers.internal_compressed);
+ ledger_disable_panic_on_negative(pmap->ledger,
+ task_ledgers.iokit_mapped);
+ ledger_disable_panic_on_negative(pmap->ledger,
+ task_ledgers.alternate_accounting);
+ ledger_disable_panic_on_negative(pmap->ledger,
+ task_ledgers.alternate_accounting_compressed);
+ }
+}
+#endif /* MACH_ASSERT */
+
+
+#if DEVELOPMENT || DEBUG
+int pmap_pagezero_mitigation = 1;
+#endif
- process_pmap_updates();
+void
+pmap_advise_pagezero_range(pmap_t lpmap, uint64_t low_bound)
+{
+#if DEVELOPMENT || DEBUG
+ if (pmap_pagezero_mitigation == 0) {
+ lpmap->pagezero_accessible = FALSE;
+ return;
+ }
+#endif
+ lpmap->pagezero_accessible = ((pmap_smap_enabled == FALSE) && (low_bound < 0x1000));
+ if (lpmap == current_pmap()) {
+ mp_disable_preemption();
+ current_cpu_datap()->cpu_pagezero_mapped = lpmap->pagezero_accessible;
+ mp_enable_preemption();
+ }
+}
- PMAP_TRACE(PMAP_CODE(PMAP__UPDATE_INTERRUPT) | DBG_FUNC_END,
- 0, 0, 0, 0, 0);
+uintptr_t
+pmap_verify_noncacheable(uintptr_t vaddr)
+{
+ pt_entry_t *ptep = NULL;
+ ptep = pmap_pte(kernel_pmap, vaddr);
+ if (ptep == NULL) {
+ panic("pmap_verify_noncacheable: no translation for 0x%lx", vaddr);
+ }
+ /* Non-cacheable OK */
+ if (*ptep & (INTEL_PTE_NCACHE)) {
+ return pte_to_pa(*ptep) | (vaddr & INTEL_OFFMASK);
+ }
+ /* Write-combined OK */
+ if (*ptep & (INTEL_PTE_PAT)) {
+ return pte_to_pa(*ptep) | (vaddr & INTEL_OFFMASK);
+ }
+ panic("pmap_verify_noncacheable: IO read from a cacheable address? address: 0x%lx, PTE: %p, *PTE: 0x%llx", vaddr, ptep, *ptep);
+ /*NOTREACHED*/
+ return 0;
}
+void
+trust_cache_init(void)
+{
+ // Unsupported on this architecture.
+}
-unsigned int
-pmap_cache_attributes(ppnum_t pn)
+kern_return_t
+pmap_load_legacy_trust_cache(struct pmap_legacy_trust_cache __unused *trust_cache,
+ const vm_size_t __unused trust_cache_len)
{
- return IS_MANAGED_PAGE(ppn_to_pai(pn)) ? VM_WIMG_COPYBACK
- : VM_WIMG_IO;
+ // Unsupported on this architecture.
+ return KERN_NOT_SUPPORTED;
}
+pmap_tc_ret_t
+pmap_load_image4_trust_cache(struct pmap_image4_trust_cache __unused *trust_cache,
+ const vm_size_t __unused trust_cache_len,
+ uint8_t const * __unused img4_manifest,
+ const vm_size_t __unused img4_manifest_buffer_len,
+ const vm_size_t __unused img4_manifest_actual_len,
+ bool __unused dry_run)
+{
+ // Unsupported on this architecture.
+ return PMAP_TC_UNKNOWN_FORMAT;
+}
projects / apple / xnu.git / blobdiff