]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/stack.c
xnu-4903.270.47.tar.gz
[apple/xnu.git] / osfmk / kern / stack.c
index 840ff67434b512402342c1f92481f15f93496823..2874583643d8c1462f701c45ca7aa9ad48d31305 100644 (file)
@@ -1,8 +1,8 @@
 /*
- * Copyright (c) 2003-2004 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2003-2007 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,
@@ -22,7 +22,7 @@
  * 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@
  */
 /*
 #include <mach/processor_set.h>
 
 #include <kern/kern_types.h>
+#include <kern/lock_group.h>
 #include <kern/mach_param.h>
 #include <kern/processor.h>
 #include <kern/thread.h>
 #include <kern/zalloc.h>
 #include <kern/kalloc.h>
+#include <kern/ledger.h>
 
 #include <vm/vm_map.h>
 #include <vm/vm_kern.h>
 
 #include <mach_debug.h>
+#include <san/kasan.h>
 
 /*
  *     We allocate stacks from generic kernel VM.
  *     because stack_alloc_try/thread_invoke operate at splsched.
  */
 
-decl_simple_lock_data(static,stack_lock_data)
-#define stack_lock()           simple_lock(&stack_lock_data)
-#define stack_unlock()         simple_unlock(&stack_lock_data)
+decl_simple_lock_data(static, stack_lock_data)
+#define stack_lock()            simple_lock(&stack_lock_data, LCK_GRP_NULL)
+#define stack_unlock()          simple_unlock(&stack_lock_data)
+
+#define STACK_CACHE_SIZE        2
 
-#define STACK_CACHE_SIZE       2
+static vm_offset_t              stack_free_list;
 
-static vm_map_t                        stack_map;
-static vm_offset_t             stack_free_list;
+static unsigned int             stack_free_count, stack_free_hiwat;             /* free list count */
+static unsigned int             stack_hiwat;
+unsigned int                    stack_total;                            /* current total count */
+unsigned long long              stack_allocs;                           /* total count of allocations */
 
-static unsigned int            stack_free_count, stack_free_hiwat;             /* free list count */
-static unsigned int            stack_total, stack_hiwat;                               /* current total count */
+static int                      stack_fake_zone_index = -1;     /* index in zone_info array */
 
-static unsigned int            stack_free_target;
-static int                             stack_free_delta;
+static unsigned int             stack_free_target;
+static int                              stack_free_delta;
 
-static unsigned int            stack_new_count;                                                /* total new stack allocations */
+static unsigned int             stack_new_count;                                                /* total new stack allocations */
 
-static vm_offset_t             stack_addr_mask;
+static vm_offset_t              stack_addr_mask;
+
+unsigned int                    kernel_stack_pages;
+vm_offset_t                     kernel_stack_size;
+vm_offset_t                     kernel_stack_mask;
+vm_offset_t                     kernel_stack_depth_max;
 
 /*
  *     The next field is at the base of the stack,
  *     so the low end is left unsullied.
  */
-#define stack_next(stack)      \
-                       (*((vm_offset_t *)((stack) + KERNEL_STACK_SIZE) - 1))
+#define stack_next(stack)       \
+       (*((vm_offset_t *)((stack) + kernel_stack_size) - 1))
+
+static inline int
+log2(vm_offset_t size)
+{
+       int     result;
+       for (result = 0; size > 0; result++) {
+               size >>= 1;
+       }
+       return result;
+}
+
+static inline vm_offset_t
+roundup_pow2(vm_offset_t size)
+{
+       return 1UL << (log2(size - 1) + 1);
+}
+
+static vm_offset_t stack_alloc_internal(void);
+static void stack_free_stack(vm_offset_t);
 
 void
 stack_init(void)
 {
-       vm_offset_t                     stacks, boundary;
-       vm_map_offset_t         map_addr;
-
        simple_lock_init(&stack_lock_data, 0);
-       
-       if (KERNEL_STACK_SIZE < round_page(KERNEL_STACK_SIZE))
-               panic("stack_init: stack size %d not a multiple of page size %d\n",     KERNEL_STACK_SIZE, PAGE_SIZE);
-       
-       for (boundary = PAGE_SIZE; boundary <= KERNEL_STACK_SIZE; )
-               boundary <<= 1;
-
-       stack_addr_mask = boundary - 1;
-
-       if (kmem_suballoc(kernel_map, &stacks, (boundary * (2 * THREAD_MAX + 64)),
-                                                               FALSE, VM_FLAGS_ANYWHERE, &stack_map) != KERN_SUCCESS)
-               panic("stack_init: kmem_suballoc");
-
-       map_addr = vm_map_min(stack_map);
-       if (vm_map_enter(stack_map, &map_addr, vm_map_round_page(PAGE_SIZE), 0, VM_FLAGS_FIXED,
-                                               VM_OBJECT_NULL, 0, FALSE, VM_PROT_NONE, VM_PROT_NONE, VM_INHERIT_DEFAULT) != KERN_SUCCESS)
-               panic("stack_init: vm_map_enter");
+
+       kernel_stack_pages = KERNEL_STACK_SIZE / PAGE_SIZE;
+       kernel_stack_size = KERNEL_STACK_SIZE;
+       kernel_stack_mask = -KERNEL_STACK_SIZE;
+       kernel_stack_depth_max = 0;
+
+       if (PE_parse_boot_argn("kernel_stack_pages",
+           &kernel_stack_pages,
+           sizeof(kernel_stack_pages))) {
+               kernel_stack_size = kernel_stack_pages * PAGE_SIZE;
+               printf("stack_init: kernel_stack_pages=%d kernel_stack_size=%p\n",
+                   kernel_stack_pages, (void *) kernel_stack_size);
+       }
+
+       if (kernel_stack_size < round_page(kernel_stack_size)) {
+               panic("stack_init: stack size %p not a multiple of page size %d\n",
+                   (void *) kernel_stack_size, PAGE_SIZE);
+       }
+
+       stack_addr_mask = roundup_pow2(kernel_stack_size) - 1;
+       kernel_stack_mask = ~stack_addr_mask;
 }
 
 /*
@@ -110,37 +141,72 @@ stack_init(void)
  *     Allocate a stack for a thread, may
  *     block.
  */
-void
-stack_alloc(
-       thread_t        thread)
-{
-       vm_offset_t             stack;
-       spl_t                   s;
 
-       assert(thread->kernel_stack == 0);
+static vm_offset_t
+stack_alloc_internal(void)
+{
+       vm_offset_t             stack = 0;
+       spl_t                   s;
+       int                     flags = 0;
+       kern_return_t           kr = KERN_SUCCESS;
 
        s = splsched();
        stack_lock();
+       stack_allocs++;
        stack = stack_free_list;
        if (stack != 0) {
                stack_free_list = stack_next(stack);
                stack_free_count--;
-       }
-       else {
-               if (++stack_total > stack_hiwat)
+       } else {
+               if (++stack_total > stack_hiwat) {
                        stack_hiwat = stack_total;
+               }
                stack_new_count++;
        }
        stack_free_delta--;
        stack_unlock();
        splx(s);
-               
+
        if (stack == 0) {
-               if (kernel_memory_allocate(stack_map, &stack, KERNEL_STACK_SIZE, stack_addr_mask, KMA_KOBJECT) != KERN_SUCCESS)
-                       panic("stack_alloc: kernel_memory_allocate");
+               /*
+                * Request guard pages on either side of the stack.  Ask
+                * kernel_memory_allocate() for two extra pages to account
+                * for these.
+                */
+
+               flags = KMA_GUARD_FIRST | KMA_GUARD_LAST | KMA_KSTACK | KMA_KOBJECT | KMA_ZERO;
+               kr = kernel_memory_allocate(kernel_map, &stack,
+                   kernel_stack_size + (2 * PAGE_SIZE),
+                   stack_addr_mask,
+                   flags,
+                   VM_KERN_MEMORY_STACK);
+               if (kr != KERN_SUCCESS) {
+                       panic("stack_alloc: kernel_memory_allocate(size:0x%llx, mask: 0x%llx, flags: 0x%x) failed with %d\n", (uint64_t)(kernel_stack_size + (2 * PAGE_SIZE)), (uint64_t)stack_addr_mask, flags, kr);
+               }
+
+               /*
+                * The stack address that comes back is the address of the lower
+                * guard page.  Skip past it to get the actual stack base address.
+                */
+
+               stack += PAGE_SIZE;
        }
+       return stack;
+}
 
-       machine_stack_attach(thread, stack);
+void
+stack_alloc(
+       thread_t        thread)
+{
+       assert(thread->kernel_stack == 0);
+       machine_stack_attach(thread, stack_alloc_internal());
+}
+
+void
+stack_handoff(thread_t from, thread_t to)
+{
+       assert(from == current_thread());
+       machine_stack_handoff(from, to);
 }
 
 /*
@@ -150,41 +216,36 @@ stack_alloc(
  */
 void
 stack_free(
-       thread_t        thread)
+       thread_t        thread)
 {
-    vm_offset_t                stack = machine_stack_detach(thread);
+       vm_offset_t         stack = machine_stack_detach(thread);
 
        assert(stack);
        if (stack != thread->reserved_stack) {
-               struct stack_cache      *cache;
-               spl_t                           s;
-
-               s = splsched();
-               cache = &PROCESSOR_DATA(current_processor(), stack_cache);
-               if (cache->count < STACK_CACHE_SIZE) {
-                       stack_next(stack) = cache->free;
-                       cache->free = stack;
-                       cache->count++;
-               }
-               else {
-                       stack_lock();
-                       stack_next(stack) = stack_free_list;
-                       stack_free_list = stack;
-                       if (++stack_free_count > stack_free_hiwat)
-                               stack_free_hiwat = stack_free_count;
-                       stack_free_delta++;
-                       stack_unlock();
-               }
-               splx(s);
+               stack_free_stack(stack);
        }
 }
 
 void
+stack_free_reserved(
+       thread_t        thread)
+{
+       if (thread->reserved_stack != thread->kernel_stack) {
+               stack_free_stack(thread->reserved_stack);
+       }
+}
+
+static void
 stack_free_stack(
-       vm_offset_t             stack)
+       vm_offset_t             stack)
 {
-       struct stack_cache      *cache;
-       spl_t                           s;
+       struct stack_cache      *cache;
+       spl_t                           s;
+
+#if KASAN_DEBUG
+       /* Sanity check - stack should be unpoisoned by now */
+       assert(kasan_check_shadow(stack, kernel_stack_size, 0));
+#endif
 
        s = splsched();
        cache = &PROCESSOR_DATA(current_processor(), stack_cache);
@@ -192,13 +253,13 @@ stack_free_stack(
                stack_next(stack) = cache->free;
                cache->free = stack;
                cache->count++;
-       }
-       else {
+       } else {
                stack_lock();
                stack_next(stack) = stack_free_list;
                stack_free_list = stack;
-               if (++stack_free_count > stack_free_hiwat)
+               if (++stack_free_count > stack_free_hiwat) {
                        stack_free_hiwat = stack_free_count;
+               }
                stack_free_delta++;
                stack_unlock();
        }
@@ -217,18 +278,17 @@ stack_free_stack(
  */
 boolean_t
 stack_alloc_try(
-       thread_t                thread)
+       thread_t                thread)
 {
-       struct stack_cache      *cache;
-       vm_offset_t                     stack;
+       struct stack_cache      *cache;
+       vm_offset_t                     stack;
 
        cache = &PROCESSOR_DATA(current_processor(), stack_cache);
        stack = cache->free;
        if (stack != 0) {
                cache->free = stack_next(stack);
                cache->count--;
-       }
-       else {
+       } else {
                if (stack_free_list != 0) {
                        stack_lock();
                        stack = stack_free_list;
@@ -243,13 +303,13 @@ stack_alloc_try(
 
        if (stack != 0 || (stack = thread->reserved_stack) != 0) {
                machine_stack_attach(thread, stack);
-               return (TRUE);
+               return TRUE;
        }
 
-       return (FALSE);
+       return FALSE;
 }
 
-static unsigned int            stack_collect_tick, last_stack_tick;
+static unsigned int             stack_collect_tick, last_stack_tick;
 
 /*
  *     stack_collect:
@@ -261,9 +321,9 @@ void
 stack_collect(void)
 {
        if (stack_collect_tick != last_stack_tick) {
-               unsigned int    target;
-               vm_offset_t             stack;
-               spl_t                   s;
+               unsigned int    target;
+               vm_offset_t             stack;
+               spl_t                   s;
 
                s = splsched();
                stack_lock();
@@ -278,9 +338,27 @@ stack_collect(void)
                        stack_unlock();
                        splx(s);
 
-                       if (vm_map_remove(stack_map, vm_map_trunc_page(stack),
-                                                               vm_map_round_page(stack + KERNEL_STACK_SIZE), VM_MAP_REMOVE_KUNWIRE) != KERN_SUCCESS)
+                       /*
+                        * Get the stack base address, then decrement by one page
+                        * to account for the lower guard page.  Add two extra pages
+                        * to the size to account for the guard pages on both ends
+                        * that were originally requested when the stack was allocated
+                        * back in stack_alloc().
+                        */
+
+                       stack = (vm_offset_t)vm_map_trunc_page(
+                               stack,
+                               VM_MAP_PAGE_MASK(kernel_map));
+                       stack -= PAGE_SIZE;
+                       if (vm_map_remove(
+                                   kernel_map,
+                                   stack,
+                                   stack + kernel_stack_size + (2 * PAGE_SIZE),
+                                   VM_MAP_REMOVE_KUNWIRE)
+                           != KERN_SUCCESS) {
                                panic("stack_collect: vm_map_remove");
+                       }
+                       stack = 0;
 
                        s = splsched();
                        stack_lock();
@@ -307,18 +385,18 @@ stack_collect(void)
  */
 void
 compute_stack_target(
-__unused void          *arg)
+       __unused void           *arg)
 {
-       spl_t           s;
+       spl_t           s;
 
        s = splsched();
        stack_lock();
 
-       if (stack_free_target > 5)
+       if (stack_free_target > 5) {
                stack_free_target = (4 * stack_free_target) / 5;
-       else
-       if (stack_free_target > 0)
+       } else if (stack_free_target > 0) {
                stack_free_target--;
+       }
 
        stack_free_target += (stack_free_delta >= 0)? stack_free_delta: -stack_free_delta;
 
@@ -330,14 +408,23 @@ __unused void             *arg)
 }
 
 void
-stack_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)
+stack_fake_zone_init(int zone_index)
+{
+       stack_fake_zone_index = zone_index;
+}
+
+void
+stack_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)
 {
-       unsigned int    total, hiwat, free;
-       spl_t                   s;
+       unsigned int    total, hiwat, free;
+       unsigned long long all;
+       spl_t                   s;
 
        s = splsched();
        stack_lock();
+       all = stack_allocs;
        total = stack_total;
        hiwat = stack_hiwat;
        free = stack_free_count;
@@ -345,21 +432,24 @@ stack_fake_zone_info(int *count, vm_size_t *cur_size, vm_size_t *max_size, vm_si
        splx(s);
 
        *count      = total - free;
-       *cur_size   = KERNEL_STACK_SIZE * total;
-       *max_size   = KERNEL_STACK_SIZE * hiwat;
-       *elem_size  = KERNEL_STACK_SIZE;
-       *alloc_size = KERNEL_STACK_SIZE;
+       *cur_size   = kernel_stack_size * total;
+       *max_size   = kernel_stack_size * hiwat;
+       *elem_size  = kernel_stack_size;
+       *alloc_size = kernel_stack_size;
+       *sum_size = all * kernel_stack_size;
+
        *collectable = 1;
        *exhaustable = 0;
+       *caller_acct = 1;
 }
 
 /* OBSOLETE */
-void   stack_privilege(
-                       thread_t        thread);
+void    stack_privilege(
+       thread_t        thread);
 
 void
 stack_privilege(
-       __unused thread_t       thread)
+       __unused thread_t       thread)
 {
        /* OBSOLETE */
 }
@@ -369,75 +459,74 @@ stack_privilege(
  */
 kern_return_t
 processor_set_stack_usage(
-       processor_set_t pset,
-       unsigned int    *totalp,
-       vm_size_t       *spacep,
-       vm_size_t       *residentp,
-       vm_size_t       *maxusagep,
-       vm_offset_t     *maxstackp)
+       processor_set_t pset,
+       unsigned int    *totalp,
+       vm_size_t       *spacep,
+       vm_size_t       *residentp,
+       vm_size_t       *maxusagep,
+       vm_offset_t     *maxstackp)
 {
 #if !MACH_DEBUG
-        return KERN_NOT_SUPPORTED;
+       return KERN_NOT_SUPPORTED;
 #else
        unsigned int total;
        vm_size_t maxusage;
        vm_offset_t maxstack;
 
-       register thread_t *threads;
-       register thread_t thread;
+       thread_t *thread_list;
+       thread_t thread;
 
-       unsigned int actual;    /* this many things */
+       unsigned int actual;    /* this many things */
        unsigned int i;
 
        vm_size_t size, size_needed;
        void *addr;
 
-       if (pset == PROCESSOR_SET_NULL)
+       if (pset == PROCESSOR_SET_NULL || pset != &pset0) {
                return KERN_INVALID_ARGUMENT;
+       }
 
-       size = 0; addr = 0;
+       size = 0;
+       addr = NULL;
 
        for (;;) {
-               pset_lock(pset);
-               if (!pset->active) {
-                       pset_unlock(pset);
-                       return KERN_INVALID_ARGUMENT;
-               }
+               lck_mtx_lock(&tasks_threads_lock);
 
-               actual = pset->thread_count;
+               actual = threads_count;
 
                /* do we have the memory we need? */
 
                size_needed = actual * sizeof(thread_t);
-               if (size_needed <= size)
+               if (size_needed <= size) {
                        break;
+               }
 
-               /* unlock the pset and allocate more memory */
-               pset_unlock(pset);
+               lck_mtx_unlock(&tasks_threads_lock);
 
-               if (size != 0)
+               if (size != 0) {
                        kfree(addr, size);
+               }
 
                assert(size_needed > 0);
                size = size_needed;
 
                addr = kalloc(size);
-               if (addr == 0)
+               if (addr == 0) {
                        return KERN_RESOURCE_SHORTAGE;
+               }
        }
 
-       /* OK, have memory and the processor_set is locked & active */
-       threads = (thread_t *) addr;
-       for (i = 0, thread = (thread_t) queue_first(&pset->threads);
-                                       !queue_end(&pset->threads, (queue_entry_t) thread);
-                                       thread = (thread_t) queue_next(&thread->pset_threads)) {
+       /* OK, have memory and list is locked */
+       thread_list = (thread_t *) addr;
+       for (i = 0, thread = (thread_t)(void *) queue_first(&threads);
+           !queue_end(&threads, (queue_entry_t) thread);
+           thread = (thread_t)(void *) queue_next(&thread->threads)) {
                thread_reference_internal(thread);
-               threads[i++] = thread;
+               thread_list[i++] = thread;
        }
        assert(i <= actual);
 
-       /* can unlock processor set now that we have the thread refs */
-       pset_unlock(pset);
+       lck_mtx_unlock(&tasks_threads_lock);
 
        /* calculate maxusage and free thread references */
 
@@ -445,32 +534,36 @@ processor_set_stack_usage(
        maxusage = 0;
        maxstack = 0;
        while (i > 0) {
-               thread_t threadref = threads[--i];
+               thread_t threadref = thread_list[--i];
 
-               if (threadref->kernel_stack != 0)
+               if (threadref->kernel_stack != 0) {
                        total++;
+               }
 
                thread_deallocate(threadref);
        }
 
-       if (size != 0)
+       if (size != 0) {
                kfree(addr, size);
+       }
 
        *totalp = total;
-       *residentp = *spacep = total * round_page(KERNEL_STACK_SIZE);
+       *residentp = *spacep = total * round_page(kernel_stack_size);
        *maxusagep = maxusage;
        *maxstackp = maxstack;
        return KERN_SUCCESS;
 
-#endif /* MACH_DEBUG */
+#endif  /* MACH_DEBUG */
 }
 
-vm_offset_t min_valid_stack_address(void)
+vm_offset_t
+min_valid_stack_address(void)
 {
-       return vm_map_min(stack_map);
+       return (vm_offset_t)vm_map_min(kernel_map);
 }
 
-vm_offset_t max_valid_stack_address(void)
+vm_offset_t
+max_valid_stack_address(void)
 {
-       return vm_map_max(stack_map);
+       return (vm_offset_t)vm_map_max(kernel_map);
 }