]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/stack.c
xnu-3248.50.21.tar.gz
[apple/xnu.git] / osfmk / kern / stack.c
index f9fb6682c9680c55a074ca5a2aecdbf48158659d..400cedcf99491ea95e5dfdb2bbf0eaa54ce1147d 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003-2004 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2003-2007 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
@@ -39,6 +39,7 @@
 #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>
@@ -58,11 +59,14 @@ decl_simple_lock_data(static,stack_lock_data)
 
 #define STACK_CACHE_SIZE       2
 
-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_total, stack_hiwat;                               /* current total count */
+static unsigned int            stack_hiwat;
+unsigned int                   stack_total;                            /* current total count */
+unsigned long long             stack_allocs;                           /* total count of allocations */
+
+static int                     stack_fake_zone_index = -1;     /* index in zone_info array */
 
 static unsigned int            stack_free_target;
 static int                             stack_free_delta;
@@ -71,37 +75,108 @@ static unsigned int                stack_new_count;                                                /* total new stack allocations */
 
 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;
+
+static inline void
+STACK_ZINFO_PALLOC(thread_t thread)
+{
+       task_t task;
+       zinfo_usage_t zinfo;
+
+       ledger_credit(thread->t_ledger, task_ledgers.tkm_private, kernel_stack_size);
+
+       if (stack_fake_zone_index != -1 &&
+           (task = thread->task) != NULL && (zinfo = task->tkm_zinfo) != NULL)
+               OSAddAtomic64(kernel_stack_size,
+                             (int64_t *)&zinfo[stack_fake_zone_index].alloc);
+}
+
+static inline void
+STACK_ZINFO_PFREE(thread_t thread)
+{
+       task_t task;
+       zinfo_usage_t zinfo;
+
+       ledger_debit(thread->t_ledger, task_ledgers.tkm_private, kernel_stack_size);
+
+       if (stack_fake_zone_index != -1 &&
+           (task = thread->task) != NULL && (zinfo = task->tkm_zinfo) != NULL)
+               OSAddAtomic64(kernel_stack_size, 
+                             (int64_t *)&zinfo[stack_fake_zone_index].free);
+}
+
+static inline void
+STACK_ZINFO_HANDOFF(thread_t from, thread_t to)
+{
+       ledger_debit(from->t_ledger, task_ledgers.tkm_private, kernel_stack_size);
+       ledger_credit(to->t_ledger, task_ledgers.tkm_private, kernel_stack_size);
+
+       if (stack_fake_zone_index != -1) {
+               task_t task;
+               zinfo_usage_t zinfo;
+       
+               if ((task = from->task) != NULL && (zinfo = task->tkm_zinfo) != NULL)
+                       OSAddAtomic64(kernel_stack_size, 
+                                     (int64_t *)&zinfo[stack_fake_zone_index].free);
+
+               if ((task = to->task) != NULL && (zinfo = task->tkm_zinfo) != NULL)
+                       OSAddAtomic64(kernel_stack_size, 
+                                     (int64_t *)&zinfo[stack_fake_zone_index].alloc);
+       }
+}
+
 /*
  *     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))
+       (*((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");
+       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);
+       }
 
-       map_addr = vm_map_min(stack_map);
-       if (vm_map_enter(stack_map, &map_addr, vm_map_round_page(PAGE_SIZE), 0, (VM_MAKE_TAG(VM_MEMORY_STACK) | 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");
+       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;
 }
 
 /*
@@ -110,17 +185,17 @@ stack_init(void)
  *     Allocate a stack for a thread, may
  *     block.
  */
-void
-stack_alloc(
-       thread_t        thread)
+
+static vm_offset_t 
+stack_alloc_internal(void)
 {
        vm_offset_t             stack;
        spl_t                   s;
-
-       assert(thread->kernel_stack == 0);
+       int                     guard_flags;
 
        s = splsched();
        stack_lock();
+       stack_allocs++;
        stack = stack_free_list;
        if (stack != 0) {
                stack_free_list = stack_next(stack);
@@ -136,11 +211,48 @@ stack_alloc(
        splx(s);
                
        if (stack == 0) {
-               if (kernel_memory_allocate(stack_map, &stack, KERNEL_STACK_SIZE, stack_addr_mask, KMA_KOBJECT) != KERN_SUCCESS)
+
+               /*
+                * Request guard pages on either side of the stack.  Ask
+                * kernel_memory_allocate() for two extra pages to account
+                * for these.
+                */
+
+               guard_flags = KMA_GUARD_FIRST | KMA_GUARD_LAST;
+               if (kernel_memory_allocate(kernel_map, &stack,
+                                          kernel_stack_size + (2*PAGE_SIZE),
+                                          stack_addr_mask,
+                                          KMA_KSTACK | KMA_KOBJECT | guard_flags,
+                                          VM_KERN_MEMORY_STACK)
+                   != KERN_SUCCESS)
                        panic("stack_alloc: kernel_memory_allocate");
+
+               /*
+                * 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;
+}
+
+void
+stack_alloc(
+       thread_t        thread)
+{
 
-       machine_stack_attach(thread, stack);
+       assert(thread->kernel_stack == 0);
+       machine_stack_attach(thread, stack_alloc_internal());
+       STACK_ZINFO_PALLOC(thread);
+}
+
+void
+stack_handoff(thread_t from, thread_t to)
+{
+       assert(from == current_thread());
+       machine_stack_handoff(from, to);
+       STACK_ZINFO_HANDOFF(from, to);
 }
 
 /*
@@ -156,30 +268,22 @@ stack_free(
 
        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_ZINFO_PFREE(thread);
+               stack_free_stack(stack);
        }
 }
 
 void
+stack_free_reserved(
+       thread_t        thread)
+{
+       if (thread->reserved_stack != thread->kernel_stack) {
+               stack_free_stack(thread->reserved_stack);
+               STACK_ZINFO_PFREE(thread);
+       }
+}
+
+static void
 stack_free_stack(
        vm_offset_t             stack)
 {
@@ -225,6 +329,7 @@ stack_alloc_try(
        cache = &PROCESSOR_DATA(current_processor(), stack_cache);
        stack = cache->free;
        if (stack != 0) {
+               STACK_ZINFO_PALLOC(thread);
                cache->free = stack_next(stack);
                cache->count--;
        }
@@ -233,6 +338,7 @@ stack_alloc_try(
                        stack_lock();
                        stack = stack_free_list;
                        if (stack != 0) {
+                               STACK_ZINFO_PALLOC(thread);
                                stack_free_list = stack_next(stack);
                                stack_free_count--;
                                stack_free_delta--;
@@ -278,9 +384,26 @@ 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();
@@ -330,14 +453,23 @@ __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;
+       unsigned long long all;
        spl_t                   s;
 
        s = splsched();
        stack_lock();
+       all = stack_allocs;
        total = stack_total;
        hiwat = stack_hiwat;
        free = stack_free_count;
@@ -345,12 +477,15 @@ 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 */
@@ -383,7 +518,7 @@ processor_set_stack_usage(
        vm_size_t maxusage;
        vm_offset_t maxstack;
 
-       register thread_t *threads;
+       register thread_t *thread_list;
        register thread_t thread;
 
        unsigned int actual;    /* this many things */
@@ -392,19 +527,16 @@ processor_set_stack_usage(
        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? */
 
@@ -412,8 +544,7 @@ processor_set_stack_usage(
                if (size_needed <= size)
                        break;
 
-               /* unlock the pset and allocate more memory */
-               pset_unlock(pset);
+               lck_mtx_unlock(&tasks_threads_lock);
 
                if (size != 0)
                        kfree(addr, size);
@@ -426,18 +557,17 @@ processor_set_stack_usage(
                        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 */
 
@@ -445,7 +575,7 @@ 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)
                        total++;
@@ -457,7 +587,7 @@ processor_set_stack_usage(
                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;
@@ -467,10 +597,10 @@ processor_set_stack_usage(
 
 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)
 {
-       return vm_map_max(stack_map);
+       return (vm_offset_t)vm_map_max(kernel_map);
 }