mm: page_alloc: remove trailing whitespace
[linux-2.6.git] / mm / kmemleak.c
index e094c4d..45eb621 100644 (file)
 #include <linux/sched.h>
 #include <linux/jiffies.h>
 #include <linux/delay.h>
-#include <linux/module.h>
+#include <linux/export.h>
 #include <linux/kthread.h>
 #include <linux/prio_tree.h>
-#include <linux/gfp.h>
 #include <linux/fs.h>
 #include <linux/debugfs.h>
 #include <linux/seq_file.h>
 #include <linux/string.h>
 #include <linux/nodemask.h>
 #include <linux/mm.h>
+#include <linux/workqueue.h>
+#include <linux/crc32.h>
 
 #include <asm/sections.h>
 #include <asm/processor.h>
-#include <asm/atomic.h>
+#include <linux/atomic.h>
 
+#include <linux/kmemcheck.h>
 #include <linux/kmemleak.h>
+#include <linux/memory_hotplug.h>
 
 /*
  * Kmemleak configuration and common defines.
  */
 #define MAX_TRACE              16      /* stack trace length */
-#define REPORTS_NR             50      /* maximum number of reported leaks */
 #define MSECS_MIN_AGE          5000    /* minimum object age for reporting */
-#define MSECS_SCAN_YIELD       10      /* CPU yielding period */
 #define SECS_FIRST_SCAN                60      /* delay before the first scan */
 #define SECS_SCAN_WAIT         600     /* subsequent auto scanning delay */
+#define MAX_SCAN_SIZE          4096    /* maximum size of a scanned block */
 
 #define BYTES_PER_POINTER      sizeof(void *)
 
 /* GFP bitmask for kmemleak internal allocations */
-#define GFP_KMEMLEAK_MASK      (GFP_KERNEL | GFP_ATOMIC)
+#define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
+                                __GFP_NORETRY | __GFP_NOMEMALLOC | \
+                                __GFP_NOWARN)
 
 /* scanning area inside a memory block */
 struct kmemleak_scan_area {
        struct hlist_node node;
-       unsigned long offset;
-       size_t length;
+       unsigned long start;
+       size_t size;
 };
 
+#define KMEMLEAK_GREY  0
+#define KMEMLEAK_BLACK -1
+
 /*
  * Structure holding the metadata for each allocated memory block.
  * Modifications to such objects should be made while holding the
@@ -144,6 +151,8 @@ struct kmemleak_object {
        int min_count;
        /* the total number of pointers found pointing to this object */
        int count;
+       /* checksum for detecting modified objects */
+       u32 checksum;
        /* memory ranges to be scanned inside an object (empty for all) */
        struct hlist_head area_list;
        unsigned long trace[MAX_TRACE];
@@ -160,6 +169,15 @@ struct kmemleak_object {
 /* flag set to not scan the object */
 #define OBJECT_NO_SCAN         (1 << 2)
 
+/* number of bytes to print per line; must be 16 or 32 */
+#define HEX_ROW_SIZE           16
+/* number of bytes to print at a time (1, 2, 4, 8) */
+#define HEX_GROUP_SIZE         1
+/* include ASCII after the hex output */
+#define HEX_ASCII              1
+/* max number of lines to be printed */
+#define HEX_MAX_LINES          2
+
 /* the list of all allocated objects */
 static LIST_HEAD(object_list);
 /* the list of gray-colored objects (see color_gray comment below) */
@@ -179,17 +197,16 @@ static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
 static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
 /* enables or disables early logging of the memory operations */
 static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
-/* set if a fata kmemleak error has occurred */
+/* set if a kmemleak warning was issued */
+static atomic_t kmemleak_warning = ATOMIC_INIT(0);
+/* set if a fatal kmemleak error has occurred */
 static atomic_t kmemleak_error = ATOMIC_INIT(0);
 
 /* minimum and maximum address that may be valid pointers */
 static unsigned long min_addr = ULONG_MAX;
 static unsigned long max_addr;
 
-/* used for yielding the CPU to other tasks during scanning */
-static unsigned long next_scan_yield;
 static struct task_struct *scan_thread;
-static unsigned long jiffies_scan_yield;
 /* used to avoid reporting of recently allocated objects */
 static unsigned long jiffies_min_age;
 static unsigned long jiffies_last_scan;
@@ -199,9 +216,9 @@ static signed long jiffies_scan_wait;
 static int kmemleak_stack_scan = 1;
 /* protects the memory scanning, parameters and debug/kmemleak file access */
 static DEFINE_MUTEX(scan_mutex);
+/* setting kmemleak=on, will set this var, skipping the disable */
+static int kmemleak_skip_disable;
 
-/* number of leaks reported (for limitation purposes) */
-static int reported_leaks;
 
 /*
  * Early object allocation/freeing logging. Kmemleak is initialized after the
@@ -214,7 +231,10 @@ static int reported_leaks;
 /* kmemleak operation type for early logging */
 enum {
        KMEMLEAK_ALLOC,
+       KMEMLEAK_ALLOC_PERCPU,
        KMEMLEAK_FREE,
+       KMEMLEAK_FREE_PART,
+       KMEMLEAK_FREE_PERCPU,
        KMEMLEAK_NOT_LEAK,
        KMEMLEAK_IGNORE,
        KMEMLEAK_SCAN_AREA,
@@ -230,26 +250,28 @@ struct early_log {
        const void *ptr;                /* allocated/freed memory block */
        size_t size;                    /* memory block size */
        int min_count;                  /* minimum reference count */
-       unsigned long offset;           /* scan area offset */
-       size_t length;                  /* scan area length */
+       unsigned long trace[MAX_TRACE]; /* stack trace */
+       unsigned int trace_len;         /* stack trace length */
 };
 
 /* early logging buffer and current position */
-static struct early_log early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE];
-static int crt_early_log;
+static struct early_log
+       early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
+static int crt_early_log __initdata;
 
 static void kmemleak_disable(void);
 
 /*
  * Print a warning and dump the stack trace.
  */
-#define kmemleak_warn(x...)    do {    \
-       pr_warning(x);                  \
-       dump_stack();                   \
+#define kmemleak_warn(x...)    do {            \
+       pr_warning(x);                          \
+       dump_stack();                           \
+       atomic_set(&kmemleak_warning, 1);       \
 } while (0)
 
 /*
- * Macro invoked when a serious kmemleak condition occured and cannot be
+ * Macro invoked when a serious kmemleak condition occurred and cannot be
  * recovered from. Kmemleak will be disabled and further allocation/freeing
  * tracing no longer available.
  */
@@ -259,6 +281,35 @@ static void kmemleak_disable(void);
 } while (0)
 
 /*
+ * Printing of the objects hex dump to the seq file. The number of lines to be
+ * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
+ * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
+ * with the object->lock held.
+ */
+static void hex_dump_object(struct seq_file *seq,
+                           struct kmemleak_object *object)
+{
+       const u8 *ptr = (const u8 *)object->pointer;
+       int i, len, remaining;
+       unsigned char linebuf[HEX_ROW_SIZE * 5];
+
+       /* limit the number of lines to HEX_MAX_LINES */
+       remaining = len =
+               min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE));
+
+       seq_printf(seq, "  hex dump (first %d bytes):\n", len);
+       for (i = 0; i < len; i += HEX_ROW_SIZE) {
+               int linelen = min(remaining, HEX_ROW_SIZE);
+
+               remaining -= HEX_ROW_SIZE;
+               hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE,
+                                  HEX_GROUP_SIZE, linebuf, sizeof(linebuf),
+                                  HEX_ASCII);
+               seq_printf(seq, "    %s\n", linebuf);
+       }
+}
+
+/*
  * Object colors, encoded with count and min_count:
  * - white - orphan object, not enough references to it (count < min_count)
  * - gray  - not orphan, not marked as false positive (min_count == 0) or
@@ -268,14 +319,16 @@ static void kmemleak_disable(void);
  * Newly created objects don't have any color assigned (object->count == -1)
  * before the next memory scan when they become white.
  */
-static int color_white(const struct kmemleak_object *object)
+static bool color_white(const struct kmemleak_object *object)
 {
-       return object->count != -1 && object->count < object->min_count;
+       return object->count != KMEMLEAK_BLACK &&
+               object->count < object->min_count;
 }
 
-static int color_gray(const struct kmemleak_object *object)
+static bool color_gray(const struct kmemleak_object *object)
 {
-       return object->min_count != -1 && object->count >= object->min_count;
+       return object->min_count != KMEMLEAK_BLACK &&
+               object->count >= object->min_count;
 }
 
 /*
@@ -283,9 +336,9 @@ static int color_gray(const struct kmemleak_object *object)
  * not be deleted and have a minimum age to avoid false positives caused by
  * pointers temporarily stored in CPU registers.
  */
-static int unreferenced_object(struct kmemleak_object *object)
+static bool unreferenced_object(struct kmemleak_object *object)
 {
-       return (object->flags & OBJECT_ALLOCATED) && color_white(object) &&
+       return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
                time_before_eq(object->jiffies + jiffies_min_age,
                               jiffies_last_scan);
 }
@@ -298,11 +351,14 @@ static void print_unreferenced(struct seq_file *seq,
                               struct kmemleak_object *object)
 {
        int i;
+       unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
 
        seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
                   object->pointer, object->size);
-       seq_printf(seq, "  comm \"%s\", pid %d, jiffies %lu\n",
-                  object->comm, object->pid, object->jiffies);
+       seq_printf(seq, "  comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
+                  object->comm, object->pid, object->jiffies,
+                  msecs_age / 1000, msecs_age % 1000);
+       hex_dump_object(seq, object);
        seq_printf(seq, "  backtrace:\n");
 
        for (i = 0; i < object->trace_len; i++) {
@@ -329,6 +385,8 @@ static void dump_object_info(struct kmemleak_object *object)
                  object->comm, object->pid, object->jiffies);
        pr_notice("  min_count = %d\n", object->min_count);
        pr_notice("  count = %d\n", object->count);
+       pr_notice("  flags = 0x%lx\n", object->flags);
+       pr_notice("  checksum = %d\n", object->checksum);
        pr_notice("  backtrace:\n");
        print_stack_trace(&trace, 4);
 }
@@ -351,7 +409,9 @@ static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
                object = prio_tree_entry(node, struct kmemleak_object,
                                         tree_node);
                if (!alias && object->pointer != ptr) {
-                       kmemleak_warn("Found object by alias");
+                       kmemleak_warn("Found object by alias at 0x%08lx\n",
+                                     ptr);
+                       dump_object_info(object);
                        object = NULL;
                }
        } else
@@ -433,21 +493,37 @@ static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
 }
 
 /*
+ * Save stack trace to the given array of MAX_TRACE size.
+ */
+static int __save_stack_trace(unsigned long *trace)
+{
+       struct stack_trace stack_trace;
+
+       stack_trace.max_entries = MAX_TRACE;
+       stack_trace.nr_entries = 0;
+       stack_trace.entries = trace;
+       stack_trace.skip = 2;
+       save_stack_trace(&stack_trace);
+
+       return stack_trace.nr_entries;
+}
+
+/*
  * Create the metadata (struct kmemleak_object) corresponding to an allocated
  * memory block and add it to the object_list and object_tree_root.
  */
-static void create_object(unsigned long ptr, size_t size, int min_count,
-                         gfp_t gfp)
+static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
+                                            int min_count, gfp_t gfp)
 {
        unsigned long flags;
        struct kmemleak_object *object;
        struct prio_tree_node *node;
-       struct stack_trace trace;
 
-       object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK);
+       object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
        if (!object) {
-               kmemleak_stop("Cannot allocate a kmemleak_object structure\n");
-               return;
+               pr_warning("Cannot allocate a kmemleak_object structure\n");
+               kmemleak_disable();
+               return NULL;
        }
 
        INIT_LIST_HEAD(&object->object_list);
@@ -459,8 +535,9 @@ static void create_object(unsigned long ptr, size_t size, int min_count,
        object->pointer = ptr;
        object->size = size;
        object->min_count = min_count;
-       object->count = -1;                     /* no color initially */
+       object->count = 0;                      /* white color initially */
        object->jiffies = jiffies;
+       object->checksum = 0;
 
        /* task information */
        if (in_irq()) {
@@ -481,18 +558,14 @@ static void create_object(unsigned long ptr, size_t size, int min_count,
        }
 
        /* kernel backtrace */
-       trace.max_entries = MAX_TRACE;
-       trace.nr_entries = 0;
-       trace.entries = object->trace;
-       trace.skip = 1;
-       save_stack_trace(&trace);
-       object->trace_len = trace.nr_entries;
+       object->trace_len = __save_stack_trace(object->trace);
 
        INIT_PRIO_TREE_NODE(&object->tree_node);
        object->tree_node.start = ptr;
        object->tree_node.last = ptr + size - 1;
 
        write_lock_irqsave(&kmemleak_lock, flags);
+
        min_addr = min(min_addr, ptr);
        max_addr = max(max_addr, ptr + size);
        node = prio_tree_insert(&object_tree_root, &object->tree_node);
@@ -503,45 +576,36 @@ static void create_object(unsigned long ptr, size_t size, int min_count,
         * random memory blocks.
         */
        if (node != &object->tree_node) {
-               unsigned long flags;
-
                kmemleak_stop("Cannot insert 0x%lx into the object search tree "
                              "(already existing)\n", ptr);
                object = lookup_object(ptr, 1);
-               spin_lock_irqsave(&object->lock, flags);
+               spin_lock(&object->lock);
                dump_object_info(object);
-               spin_unlock_irqrestore(&object->lock, flags);
+               spin_unlock(&object->lock);
 
                goto out;
        }
        list_add_tail_rcu(&object->object_list, &object_list);
 out:
        write_unlock_irqrestore(&kmemleak_lock, flags);
+       return object;
 }
 
 /*
  * Remove the metadata (struct kmemleak_object) for a memory block from the
  * object_list and object_tree_root and decrement its use_count.
  */
-static void delete_object(unsigned long ptr)
+static void __delete_object(struct kmemleak_object *object)
 {
        unsigned long flags;
-       struct kmemleak_object *object;
 
        write_lock_irqsave(&kmemleak_lock, flags);
-       object = lookup_object(ptr, 0);
-       if (!object) {
-               kmemleak_warn("Freeing unknown object at 0x%08lx\n",
-                             ptr);
-               write_unlock_irqrestore(&kmemleak_lock, flags);
-               return;
-       }
        prio_tree_remove(&object_tree_root, &object->tree_node);
        list_del_rcu(&object->object_list);
        write_unlock_irqrestore(&kmemleak_lock, flags);
 
        WARN_ON(!(object->flags & OBJECT_ALLOCATED));
-       WARN_ON(atomic_read(&object->use_count) < 1);
+       WARN_ON(atomic_read(&object->use_count) < 2);
 
        /*
         * Locking here also ensures that the corresponding memory block
@@ -554,73 +618,139 @@ static void delete_object(unsigned long ptr)
 }
 
 /*
- * Make a object permanently as gray-colored so that it can no longer be
- * reported as a leak. This is used in general to mark a false positive.
+ * Look up the metadata (struct kmemleak_object) corresponding to ptr and
+ * delete it.
  */
-static void make_gray_object(unsigned long ptr)
+static void delete_object_full(unsigned long ptr)
 {
-       unsigned long flags;
        struct kmemleak_object *object;
 
        object = find_and_get_object(ptr, 0);
        if (!object) {
-               kmemleak_warn("Graying unknown object at 0x%08lx\n", ptr);
+#ifdef DEBUG
+               kmemleak_warn("Freeing unknown object at 0x%08lx\n",
+                             ptr);
+#endif
                return;
        }
-
-       spin_lock_irqsave(&object->lock, flags);
-       object->min_count = 0;
-       spin_unlock_irqrestore(&object->lock, flags);
+       __delete_object(object);
        put_object(object);
 }
 
 /*
- * Mark the object as black-colored so that it is ignored from scans and
- * reporting.
+ * Look up the metadata (struct kmemleak_object) corresponding to ptr and
+ * delete it. If the memory block is partially freed, the function may create
+ * additional metadata for the remaining parts of the block.
  */
-static void make_black_object(unsigned long ptr)
+static void delete_object_part(unsigned long ptr, size_t size)
 {
-       unsigned long flags;
        struct kmemleak_object *object;
+       unsigned long start, end;
 
-       object = find_and_get_object(ptr, 0);
+       object = find_and_get_object(ptr, 1);
        if (!object) {
-               kmemleak_warn("Blacking unknown object at 0x%08lx\n", ptr);
+#ifdef DEBUG
+               kmemleak_warn("Partially freeing unknown object at 0x%08lx "
+                             "(size %zu)\n", ptr, size);
+#endif
                return;
        }
+       __delete_object(object);
+
+       /*
+        * Create one or two objects that may result from the memory block
+        * split. Note that partial freeing is only done by free_bootmem() and
+        * this happens before kmemleak_init() is called. The path below is
+        * only executed during early log recording in kmemleak_init(), so
+        * GFP_KERNEL is enough.
+        */
+       start = object->pointer;
+       end = object->pointer + object->size;
+       if (ptr > start)
+               create_object(start, ptr - start, object->min_count,
+                             GFP_KERNEL);
+       if (ptr + size < end)
+               create_object(ptr + size, end - ptr - size, object->min_count,
+                             GFP_KERNEL);
+
+       put_object(object);
+}
+
+static void __paint_it(struct kmemleak_object *object, int color)
+{
+       object->min_count = color;
+       if (color == KMEMLEAK_BLACK)
+               object->flags |= OBJECT_NO_SCAN;
+}
+
+static void paint_it(struct kmemleak_object *object, int color)
+{
+       unsigned long flags;
 
        spin_lock_irqsave(&object->lock, flags);
-       object->min_count = -1;
+       __paint_it(object, color);
        spin_unlock_irqrestore(&object->lock, flags);
+}
+
+static void paint_ptr(unsigned long ptr, int color)
+{
+       struct kmemleak_object *object;
+
+       object = find_and_get_object(ptr, 0);
+       if (!object) {
+               kmemleak_warn("Trying to color unknown object "
+                             "at 0x%08lx as %s\n", ptr,
+                             (color == KMEMLEAK_GREY) ? "Grey" :
+                             (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
+               return;
+       }
+       paint_it(object, color);
        put_object(object);
 }
 
 /*
+ * Mark an object permanently as gray-colored so that it can no longer be
+ * reported as a leak. This is used in general to mark a false positive.
+ */
+static void make_gray_object(unsigned long ptr)
+{
+       paint_ptr(ptr, KMEMLEAK_GREY);
+}
+
+/*
+ * Mark the object as black-colored so that it is ignored from scans and
+ * reporting.
+ */
+static void make_black_object(unsigned long ptr)
+{
+       paint_ptr(ptr, KMEMLEAK_BLACK);
+}
+
+/*
  * Add a scanning area to the object. If at least one such area is added,
  * kmemleak will only scan these ranges rather than the whole memory block.
  */
-static void add_scan_area(unsigned long ptr, unsigned long offset,
-                         size_t length, gfp_t gfp)
+static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
 {
        unsigned long flags;
        struct kmemleak_object *object;
        struct kmemleak_scan_area *area;
 
-       object = find_and_get_object(ptr, 0);
+       object = find_and_get_object(ptr, 1);
        if (!object) {
                kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
                              ptr);
                return;
        }
 
-       area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK);
+       area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
        if (!area) {
-               kmemleak_warn("Cannot allocate a scan area\n");
+               pr_warning("Cannot allocate a scan area\n");
                goto out;
        }
 
        spin_lock_irqsave(&object->lock, flags);
-       if (offset + length > object->size) {
+       if (ptr + size > object->pointer + object->size) {
                kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
                dump_object_info(object);
                kmem_cache_free(scan_area_cache, area);
@@ -628,8 +758,8 @@ static void add_scan_area(unsigned long ptr, unsigned long offset,
        }
 
        INIT_HLIST_NODE(&area->node);
-       area->offset = offset;
-       area->length = length;
+       area->start = ptr;
+       area->size = size;
 
        hlist_add_head(&area->node, &object->area_list);
 out_unlock:
@@ -664,14 +794,19 @@ static void object_no_scan(unsigned long ptr)
  * Log an early kmemleak_* call to the early_log buffer. These calls will be
  * processed later once kmemleak is fully initialized.
  */
-static void log_early(int op_type, const void *ptr, size_t size,
-                     int min_count, unsigned long offset, size_t length)
+static void __init log_early(int op_type, const void *ptr, size_t size,
+                            int min_count)
 {
        unsigned long flags;
        struct early_log *log;
 
+       if (atomic_read(&kmemleak_error)) {
+               /* kmemleak stopped recording, just count the requests */
+               crt_early_log++;
+               return;
+       }
+
        if (crt_early_log >= ARRAY_SIZE(early_log)) {
-               pr_warning("Early log buffer exceeded\n");
                kmemleak_disable();
                return;
        }
@@ -686,121 +821,265 @@ static void log_early(int op_type, const void *ptr, size_t size,
        log->ptr = ptr;
        log->size = size;
        log->min_count = min_count;
-       log->offset = offset;
-       log->length = length;
+       log->trace_len = __save_stack_trace(log->trace);
        crt_early_log++;
        local_irq_restore(flags);
 }
 
 /*
- * Memory allocation function callback. This function is called from the
- * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc,
- * vmalloc etc.).
+ * Log an early allocated block and populate the stack trace.
+ */
+static void early_alloc(struct early_log *log)
+{
+       struct kmemleak_object *object;
+       unsigned long flags;
+       int i;
+
+       if (!atomic_read(&kmemleak_enabled) || !log->ptr || IS_ERR(log->ptr))
+               return;
+
+       /*
+        * RCU locking needed to ensure object is not freed via put_object().
+        */
+       rcu_read_lock();
+       object = create_object((unsigned long)log->ptr, log->size,
+                              log->min_count, GFP_ATOMIC);
+       if (!object)
+               goto out;
+       spin_lock_irqsave(&object->lock, flags);
+       for (i = 0; i < log->trace_len; i++)
+               object->trace[i] = log->trace[i];
+       object->trace_len = log->trace_len;
+       spin_unlock_irqrestore(&object->lock, flags);
+out:
+       rcu_read_unlock();
+}
+
+/*
+ * Log an early allocated block and populate the stack trace.
  */
-void kmemleak_alloc(const void *ptr, size_t size, int min_count, gfp_t gfp)
+static void early_alloc_percpu(struct early_log *log)
+{
+       unsigned int cpu;
+       const void __percpu *ptr = log->ptr;
+
+       for_each_possible_cpu(cpu) {
+               log->ptr = per_cpu_ptr(ptr, cpu);
+               early_alloc(log);
+       }
+}
+
+/**
+ * kmemleak_alloc - register a newly allocated object
+ * @ptr:       pointer to beginning of the object
+ * @size:      size of the object
+ * @min_count: minimum number of references to this object. If during memory
+ *             scanning a number of references less than @min_count is found,
+ *             the object is reported as a memory leak. If @min_count is 0,
+ *             the object is never reported as a leak. If @min_count is -1,
+ *             the object is ignored (not scanned and not reported as a leak)
+ * @gfp:       kmalloc() flags used for kmemleak internal memory allocations
+ *
+ * This function is called from the kernel allocators when a new object
+ * (memory block) is allocated (kmem_cache_alloc, kmalloc, vmalloc etc.).
+ */
+void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
+                         gfp_t gfp)
 {
        pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                create_object((unsigned long)ptr, size, min_count, gfp);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0);
+               log_early(KMEMLEAK_ALLOC, ptr, size, min_count);
 }
 EXPORT_SYMBOL_GPL(kmemleak_alloc);
 
-/*
- * Memory freeing function callback. This function is called from the kernel
- * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.).
+/**
+ * kmemleak_alloc_percpu - register a newly allocated __percpu object
+ * @ptr:       __percpu pointer to beginning of the object
+ * @size:      size of the object
+ *
+ * This function is called from the kernel percpu allocator when a new object
+ * (memory block) is allocated (alloc_percpu). It assumes GFP_KERNEL
+ * allocation.
  */
-void kmemleak_free(const void *ptr)
+void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size)
+{
+       unsigned int cpu;
+
+       pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
+
+       /*
+        * Percpu allocations are only scanned and not reported as leaks
+        * (min_count is set to 0).
+        */
+       if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
+               for_each_possible_cpu(cpu)
+                       create_object((unsigned long)per_cpu_ptr(ptr, cpu),
+                                     size, 0, GFP_KERNEL);
+       else if (atomic_read(&kmemleak_early_log))
+               log_early(KMEMLEAK_ALLOC_PERCPU, ptr, size, 0);
+}
+EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
+
+/**
+ * kmemleak_free - unregister a previously registered object
+ * @ptr:       pointer to beginning of the object
+ *
+ * This function is called from the kernel allocators when an object (memory
+ * block) is freed (kmem_cache_free, kfree, vfree etc.).
+ */
+void __ref kmemleak_free(const void *ptr)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
-               delete_object((unsigned long)ptr);
+               delete_object_full((unsigned long)ptr);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0);
+               log_early(KMEMLEAK_FREE, ptr, 0, 0);
 }
 EXPORT_SYMBOL_GPL(kmemleak_free);
 
-/*
- * Mark an already allocated memory block as a false positive. This will cause
- * the block to no longer be reported as leak and always be scanned.
+/**
+ * kmemleak_free_part - partially unregister a previously registered object
+ * @ptr:       pointer to the beginning or inside the object. This also
+ *             represents the start of the range to be freed
+ * @size:      size to be unregistered
+ *
+ * This function is called when only a part of a memory block is freed
+ * (usually from the bootmem allocator).
  */
-void kmemleak_not_leak(const void *ptr)
+void __ref kmemleak_free_part(const void *ptr, size_t size)
+{
+       pr_debug("%s(0x%p)\n", __func__, ptr);
+
+       if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
+               delete_object_part((unsigned long)ptr, size);
+       else if (atomic_read(&kmemleak_early_log))
+               log_early(KMEMLEAK_FREE_PART, ptr, size, 0);
+}
+EXPORT_SYMBOL_GPL(kmemleak_free_part);
+
+/**
+ * kmemleak_free_percpu - unregister a previously registered __percpu object
+ * @ptr:       __percpu pointer to beginning of the object
+ *
+ * This function is called from the kernel percpu allocator when an object
+ * (memory block) is freed (free_percpu).
+ */
+void __ref kmemleak_free_percpu(const void __percpu *ptr)
+{
+       unsigned int cpu;
+
+       pr_debug("%s(0x%p)\n", __func__, ptr);
+
+       if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
+               for_each_possible_cpu(cpu)
+                       delete_object_full((unsigned long)per_cpu_ptr(ptr,
+                                                                     cpu));
+       else if (atomic_read(&kmemleak_early_log))
+               log_early(KMEMLEAK_FREE_PERCPU, ptr, 0, 0);
+}
+EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
+
+/**
+ * kmemleak_not_leak - mark an allocated object as false positive
+ * @ptr:       pointer to beginning of the object
+ *
+ * Calling this function on an object will cause the memory block to no longer
+ * be reported as leak and always be scanned.
+ */
+void __ref kmemleak_not_leak(const void *ptr)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                make_gray_object((unsigned long)ptr);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0);
+               log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0);
 }
 EXPORT_SYMBOL(kmemleak_not_leak);
 
-/*
- * Ignore a memory block. This is usually done when it is known that the
- * corresponding block is not a leak and does not contain any references to
- * other allocated memory blocks.
+/**
+ * kmemleak_ignore - ignore an allocated object
+ * @ptr:       pointer to beginning of the object
+ *
+ * Calling this function on an object will cause the memory block to be
+ * ignored (not scanned and not reported as a leak). This is usually done when
+ * it is known that the corresponding block is not a leak and does not contain
+ * any references to other allocated memory blocks.
  */
-void kmemleak_ignore(const void *ptr)
+void __ref kmemleak_ignore(const void *ptr)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                make_black_object((unsigned long)ptr);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0);
+               log_early(KMEMLEAK_IGNORE, ptr, 0, 0);
 }
 EXPORT_SYMBOL(kmemleak_ignore);
 
-/*
- * Limit the range to be scanned in an allocated memory block.
+/**
+ * kmemleak_scan_area - limit the range to be scanned in an allocated object
+ * @ptr:       pointer to beginning or inside the object. This also
+ *             represents the start of the scan area
+ * @size:      size of the scan area
+ * @gfp:       kmalloc() flags used for kmemleak internal memory allocations
+ *
+ * This function is used when it is known that only certain parts of an object
+ * contain references to other objects. Kmemleak will only scan these areas
+ * reducing the number false negatives.
  */
-void kmemleak_scan_area(const void *ptr, unsigned long offset, size_t length,
-                       gfp_t gfp)
+void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
-       if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
-               add_scan_area((unsigned long)ptr, offset, length, gfp);
+       if (atomic_read(&kmemleak_enabled) && ptr && size && !IS_ERR(ptr))
+               add_scan_area((unsigned long)ptr, size, gfp);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length);
+               log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0);
 }
 EXPORT_SYMBOL(kmemleak_scan_area);
 
-/*
- * Inform kmemleak not to scan the given memory block.
+/**
+ * kmemleak_no_scan - do not scan an allocated object
+ * @ptr:       pointer to beginning of the object
+ *
+ * This function notifies kmemleak not to scan the given memory block. Useful
+ * in situations where it is known that the given object does not contain any
+ * references to other objects. Kmemleak will not scan such objects reducing
+ * the number of false negatives.
  */
-void kmemleak_no_scan(const void *ptr)
+void __ref kmemleak_no_scan(const void *ptr)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                object_no_scan((unsigned long)ptr);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0);
+               log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0);
 }
 EXPORT_SYMBOL(kmemleak_no_scan);
 
 /*
- * Yield the CPU so that other tasks get a chance to run.  The yielding is
- * rate-limited to avoid excessive number of calls to the schedule() function
- * during memory scanning.
+ * Update an object's checksum and return true if it was modified.
  */
-static void scan_yield(void)
+static bool update_checksum(struct kmemleak_object *object)
 {
-       might_sleep();
+       u32 old_csum = object->checksum;
 
-       if (time_is_before_eq_jiffies(next_scan_yield)) {
-               schedule();
-               next_scan_yield = jiffies + jiffies_scan_yield;
-       }
+       if (!kmemcheck_is_obj_initialized(object->pointer, object->size))
+               return false;
+
+       object->checksum = crc32(0, (void *)object->pointer, object->size);
+       return object->checksum != old_csum;
 }
 
 /*
  * Memory scanning is a long process and it needs to be interruptable. This
- * function checks whether such interrupt condition occured.
+ * function checks whether such interrupt condition occurred.
  */
 static int scan_should_stop(void)
 {
@@ -824,28 +1103,28 @@ static int scan_should_stop(void)
  * found to the gray list.
  */
 static void scan_block(void *_start, void *_end,
-                      struct kmemleak_object *scanned)
+                      struct kmemleak_object *scanned, int allow_resched)
 {
        unsigned long *ptr;
        unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
        unsigned long *end = _end - (BYTES_PER_POINTER - 1);
 
        for (ptr = start; ptr < end; ptr++) {
-               unsigned long flags;
-               unsigned long pointer = *ptr;
                struct kmemleak_object *object;
+               unsigned long flags;
+               unsigned long pointer;
 
+               if (allow_resched)
+                       cond_resched();
                if (scan_should_stop())
                        break;
 
-               /*
-                * When scanning a memory block with a corresponding
-                * kmemleak_object, the CPU yielding is handled in the calling
-                * code since it holds the object->lock to avoid the block
-                * freeing.
-                */
-               if (!scanned)
-                       scan_yield();
+               /* don't scan uninitialized memory */
+               if (!kmemcheck_is_obj_initialized((unsigned long)ptr,
+                                                 BYTES_PER_POINTER))
+                       continue;
+
+               pointer = *ptr;
 
                object = find_and_get_object(pointer, 1);
                if (!object)
@@ -877,11 +1156,14 @@ static void scan_block(void *_start, void *_end,
                 * added to the gray_list.
                 */
                object->count++;
-               if (color_gray(object))
+               if (color_gray(object)) {
                        list_add_tail(&object->gray_list, &gray_list);
-               else
-                       put_object(object);
+                       spin_unlock_irqrestore(&object->lock, flags);
+                       continue;
+               }
+
                spin_unlock_irqrestore(&object->lock, flags);
+               put_object(object);
        }
 }
 
@@ -896,8 +1178,8 @@ static void scan_object(struct kmemleak_object *object)
        unsigned long flags;
 
        /*
-        * Once the object->lock is aquired, the corresponding memory block
-        * cannot be freed (the same lock is aquired in delete_object).
+        * Once the object->lock is acquired, the corresponding memory block
+        * cannot be freed (the same lock is acquired in delete_object).
         */
        spin_lock_irqsave(&object->lock, flags);
        if (object->flags & OBJECT_NO_SCAN)
@@ -905,19 +1187,63 @@ static void scan_object(struct kmemleak_object *object)
        if (!(object->flags & OBJECT_ALLOCATED))
                /* already freed object */
                goto out;
-       if (hlist_empty(&object->area_list))
-               scan_block((void *)object->pointer,
-                          (void *)(object->pointer + object->size), object);
-       else
+       if (hlist_empty(&object->area_list)) {
+               void *start = (void *)object->pointer;
+               void *end = (void *)(object->pointer + object->size);
+
+               while (start < end && (object->flags & OBJECT_ALLOCATED) &&
+                      !(object->flags & OBJECT_NO_SCAN)) {
+                       scan_block(start, min(start + MAX_SCAN_SIZE, end),
+                                  object, 0);
+                       start += MAX_SCAN_SIZE;
+
+                       spin_unlock_irqrestore(&object->lock, flags);
+                       cond_resched();
+                       spin_lock_irqsave(&object->lock, flags);
+               }
+       } else
                hlist_for_each_entry(area, elem, &object->area_list, node)
-                       scan_block((void *)(object->pointer + area->offset),
-                                  (void *)(object->pointer + area->offset
-                                           + area->length), object);
+                       scan_block((void *)area->start,
+                                  (void *)(area->start + area->size),
+                                  object, 0);
 out:
        spin_unlock_irqrestore(&object->lock, flags);
 }
 
 /*
+ * Scan the objects already referenced (gray objects). More objects will be
+ * referenced and, if there are no memory leaks, all the objects are scanned.
+ */
+static void scan_gray_list(void)
+{
+       struct kmemleak_object *object, *tmp;
+
+       /*
+        * The list traversal is safe for both tail additions and removals
+        * from inside the loop. The kmemleak objects cannot be freed from
+        * outside the loop because their use_count was incremented.
+        */
+       object = list_entry(gray_list.next, typeof(*object), gray_list);
+       while (&object->gray_list != &gray_list) {
+               cond_resched();
+
+               /* may add new objects to the list */
+               if (!scan_should_stop())
+                       scan_object(object);
+
+               tmp = list_entry(object->gray_list.next, typeof(*object),
+                                gray_list);
+
+               /* remove the object from the list and release it */
+               list_del(&object->gray_list);
+               put_object(object);
+
+               object = tmp;
+       }
+       WARN_ON(!list_empty(&gray_list));
+}
+
+/*
  * Scan data sections and all the referenced memory blocks allocated via the
  * kernel's standard allocators. This function must be called with the
  * scan_mutex held.
@@ -925,8 +1251,7 @@ out:
 static void kmemleak_scan(void)
 {
        unsigned long flags;
-       struct kmemleak_object *object, *tmp;
-       struct task_struct *task;
+       struct kmemleak_object *object;
        int i;
        int new_leaks = 0;
 
@@ -957,20 +1282,20 @@ static void kmemleak_scan(void)
        rcu_read_unlock();
 
        /* data/bss scanning */
-       scan_block(_sdata, _edata, NULL);
-       scan_block(__bss_start, __bss_stop, NULL);
+       scan_block(_sdata, _edata, NULL, 1);
+       scan_block(__bss_start, __bss_stop, NULL, 1);
 
 #ifdef CONFIG_SMP
        /* per-cpu sections scanning */
        for_each_possible_cpu(i)
                scan_block(__per_cpu_start + per_cpu_offset(i),
-                          __per_cpu_end + per_cpu_offset(i), NULL);
+                          __per_cpu_end + per_cpu_offset(i), NULL, 1);
 #endif
 
        /*
-        * Struct page scanning for each node. The code below is not yet safe
-        * with MEMORY_HOTPLUG.
+        * Struct page scanning for each node.
         */
+       lock_memory_hotplug();
        for_each_online_node(i) {
                pg_data_t *pgdat = NODE_DATA(i);
                unsigned long start_pfn = pgdat->node_start_pfn;
@@ -986,48 +1311,52 @@ static void kmemleak_scan(void)
                        /* only scan if page is in use */
                        if (page_count(page) == 0)
                                continue;
-                       scan_block(page, page + 1, NULL);
+                       scan_block(page, page + 1, NULL, 1);
                }
        }
+       unlock_memory_hotplug();
 
        /*
-        * Scanning the task stacks may introduce false negatives and it is
-        * not enabled by default.
+        * Scanning the task stacks (may introduce false negatives).
         */
        if (kmemleak_stack_scan) {
+               struct task_struct *p, *g;
+
                read_lock(&tasklist_lock);
-               for_each_process(task)
-                       scan_block(task_stack_page(task),
-                                  task_stack_page(task) + THREAD_SIZE, NULL);
+               do_each_thread(g, p) {
+                       scan_block(task_stack_page(p), task_stack_page(p) +
+                                  THREAD_SIZE, NULL, 0);
+               } while_each_thread(g, p);
                read_unlock(&tasklist_lock);
        }
 
        /*
         * Scan the objects already referenced from the sections scanned
-        * above. More objects will be referenced and, if there are no memory
-        * leaks, all the objects will be scanned. The list traversal is safe
-        * for both tail additions and removals from inside the loop. The
-        * kmemleak objects cannot be freed from outside the loop because their
-        * use_count was increased.
+        * above.
         */
-       object = list_entry(gray_list.next, typeof(*object), gray_list);
-       while (&object->gray_list != &gray_list) {
-               scan_yield();
-
-               /* may add new objects to the list */
-               if (!scan_should_stop())
-                       scan_object(object);
-
-               tmp = list_entry(object->gray_list.next, typeof(*object),
-                                gray_list);
+       scan_gray_list();
 
-               /* remove the object from the list and release it */
-               list_del(&object->gray_list);
-               put_object(object);
-
-               object = tmp;
+       /*
+        * Check for new or unreferenced objects modified since the previous
+        * scan and color them gray until the next scan.
+        */
+       rcu_read_lock();
+       list_for_each_entry_rcu(object, &object_list, object_list) {
+               spin_lock_irqsave(&object->lock, flags);
+               if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
+                   && update_checksum(object) && get_object(object)) {
+                       /* color it gray temporarily */
+                       object->count = object->min_count;
+                       list_add_tail(&object->gray_list, &gray_list);
+               }
+               spin_unlock_irqrestore(&object->lock, flags);
        }
-       WARN_ON(!list_empty(&gray_list));
+       rcu_read_unlock();
+
+       /*
+        * Re-scan the gray list for modified unreferenced objects.
+        */
+       scan_gray_list();
 
        /*
         * If scanning was stopped do not report any new unreferenced objects.
@@ -1065,6 +1394,7 @@ static int kmemleak_scan_thread(void *arg)
        static int first_run = 1;
 
        pr_info("Automatic memory scanning thread started\n");
+       set_user_nice(current, 10);
 
        /*
         * Wait before the first scan to allow the system to fully initialize.
@@ -1095,7 +1425,7 @@ static int kmemleak_scan_thread(void *arg)
  * Start the automatic memory scanning thread. This function must be called
  * with the scan_mutex held.
  */
-void start_scan_thread(void)
+static void start_scan_thread(void)
 {
        if (scan_thread)
                return;
@@ -1110,7 +1440,7 @@ void start_scan_thread(void)
  * Stop the automatic memory scanning thread. This function must be called
  * with the scan_mutex held.
  */
-void stop_scan_thread(void)
+static void stop_scan_thread(void)
 {
        if (scan_thread) {
                kthread_stop(scan_thread);
@@ -1127,11 +1457,11 @@ static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
 {
        struct kmemleak_object *object;
        loff_t n = *pos;
+       int err;
 
-       if (!n)
-               reported_leaks = 0;
-       if (reported_leaks >= REPORTS_NR)
-               return NULL;
+       err = mutex_lock_interruptible(&scan_mutex);
+       if (err < 0)
+               return ERR_PTR(err);
 
        rcu_read_lock();
        list_for_each_entry_rcu(object, &object_list, object_list) {
@@ -1142,7 +1472,6 @@ static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
        }
        object = NULL;
 out:
-       rcu_read_unlock();
        return object;
 }
 
@@ -1157,17 +1486,16 @@ static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
        struct list_head *n = &prev_obj->object_list;
 
        ++(*pos);
-       if (reported_leaks >= REPORTS_NR)
-               goto out;
 
-       rcu_read_lock();
        list_for_each_continue_rcu(n, &object_list) {
-               next_obj = list_entry(n, struct kmemleak_object, object_list);
-               if (get_object(next_obj))
+               struct kmemleak_object *obj =
+                       list_entry(n, struct kmemleak_object, object_list);
+               if (get_object(obj)) {
+                       next_obj = obj;
                        break;
+               }
        }
-       rcu_read_unlock();
-out:
+
        put_object(prev_obj);
        return next_obj;
 }
@@ -1177,8 +1505,16 @@ out:
  */
 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
 {
-       if (v)
-               put_object(v);
+       if (!IS_ERR(v)) {
+               /*
+                * kmemleak_seq_start may return ERR_PTR if the scan_mutex
+                * waiting was interrupted, so only release it if !IS_ERR.
+                */
+               rcu_read_unlock();
+               mutex_unlock(&scan_mutex);
+               if (v)
+                       put_object(v);
+       }
 }
 
 /*
@@ -1190,10 +1526,8 @@ static int kmemleak_seq_show(struct seq_file *seq, void *v)
        unsigned long flags;
 
        spin_lock_irqsave(&object->lock, flags);
-       if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object)) {
+       if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
                print_unreferenced(seq, object);
-               reported_leaks++;
-       }
        spin_unlock_irqrestore(&object->lock, flags);
        return 0;
 }
@@ -1207,36 +1541,55 @@ static const struct seq_operations kmemleak_seq_ops = {
 
 static int kmemleak_open(struct inode *inode, struct file *file)
 {
-       int ret = 0;
+       return seq_open(file, &kmemleak_seq_ops);
+}
 
-       if (!atomic_read(&kmemleak_enabled))
-               return -EBUSY;
+static int kmemleak_release(struct inode *inode, struct file *file)
+{
+       return seq_release(inode, file);
+}
 
-       ret = mutex_lock_interruptible(&scan_mutex);
-       if (ret < 0)
-               goto out;
-       if (file->f_mode & FMODE_READ) {
-               ret = seq_open(file, &kmemleak_seq_ops);
-               if (ret < 0)
-                       goto scan_unlock;
+static int dump_str_object_info(const char *str)
+{
+       unsigned long flags;
+       struct kmemleak_object *object;
+       unsigned long addr;
+
+       addr= simple_strtoul(str, NULL, 0);
+       object = find_and_get_object(addr, 0);
+       if (!object) {
+               pr_info("Unknown object at 0x%08lx\n", addr);
+               return -EINVAL;
        }
-       return ret;
 
-scan_unlock:
-       mutex_unlock(&scan_mutex);
-out:
-       return ret;
+       spin_lock_irqsave(&object->lock, flags);
+       dump_object_info(object);
+       spin_unlock_irqrestore(&object->lock, flags);
+
+       put_object(object);
+       return 0;
 }
 
-static int kmemleak_release(struct inode *inode, struct file *file)
+/*
+ * We use grey instead of black to ensure we can do future scans on the same
+ * objects. If we did not do future scans these black objects could
+ * potentially contain references to newly allocated objects in the future and
+ * we'd end up with false positives.
+ */
+static void kmemleak_clear(void)
 {
-       int ret = 0;
-
-       if (file->f_mode & FMODE_READ)
-               seq_release(inode, file);
-       mutex_unlock(&scan_mutex);
+       struct kmemleak_object *object;
+       unsigned long flags;
 
-       return ret;
+       rcu_read_lock();
+       list_for_each_entry_rcu(object, &object_list, object_list) {
+               spin_lock_irqsave(&object->lock, flags);
+               if ((object->flags & OBJECT_REPORTED) &&
+                   unreferenced_object(object))
+                       __paint_it(object, KMEMLEAK_GREY);
+               spin_unlock_irqrestore(&object->lock, flags);
+       }
+       rcu_read_unlock();
 }
 
 /*
@@ -1250,12 +1603,16 @@ static int kmemleak_release(struct inode *inode, struct file *file)
  *   scan=...  - set the automatic memory scanning period in seconds (0 to
  *               disable it)
  *   scan      - trigger a memory scan
+ *   clear     - mark all current reported unreferenced kmemleak objects as
+ *               grey to ignore printing them
+ *   dump=...  - dump information about the object found at the given address
  */
 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
                              size_t size, loff_t *ppos)
 {
        char buf[64];
        int buf_size;
+       int ret;
 
        if (!atomic_read(&kmemleak_enabled))
                return -EBUSY;
@@ -1265,6 +1622,10 @@ static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
                return -EFAULT;
        buf[buf_size] = 0;
 
+       ret = mutex_lock_interruptible(&scan_mutex);
+       if (ret < 0)
+               return ret;
+
        if (strncmp(buf, "off", 3) == 0)
                kmemleak_disable();
        else if (strncmp(buf, "stack=on", 8) == 0)
@@ -1277,11 +1638,10 @@ static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
                stop_scan_thread();
        else if (strncmp(buf, "scan=", 5) == 0) {
                unsigned long secs;
-               int err;
 
-               err = strict_strtoul(buf + 5, 0, &secs);
-               if (err < 0)
-                       return err;
+               ret = strict_strtoul(buf + 5, 0, &secs);
+               if (ret < 0)
+                       goto out;
                stop_scan_thread();
                if (secs) {
                        jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
@@ -1289,8 +1649,17 @@ static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
                }
        } else if (strncmp(buf, "scan", 4) == 0)
                kmemleak_scan();
+       else if (strncmp(buf, "clear", 5) == 0)
+               kmemleak_clear();
+       else if (strncmp(buf, "dump=", 5) == 0)
+               ret = dump_str_object_info(buf + 5);
        else
-               return -EINVAL;
+               ret = -EINVAL;
+
+out:
+       mutex_unlock(&scan_mutex);
+       if (ret < 0)
+               return ret;
 
        /* ignore the rest of the buffer, only one command at a time */
        *ppos += size;
@@ -1307,37 +1676,28 @@ static const struct file_operations kmemleak_fops = {
 };
 
 /*
- * Perform the freeing of the kmemleak internal objects after waiting for any
- * current memory scan to complete.
+ * Stop the memory scanning thread and free the kmemleak internal objects if
+ * no previous scan thread (otherwise, kmemleak may still have some useful
+ * information on memory leaks).
  */
-static int kmemleak_cleanup_thread(void *arg)
+static void kmemleak_do_cleanup(struct work_struct *work)
 {
        struct kmemleak_object *object;
+       bool cleanup = scan_thread == NULL;
 
        mutex_lock(&scan_mutex);
        stop_scan_thread();
 
-       rcu_read_lock();
-       list_for_each_entry_rcu(object, &object_list, object_list)
-               delete_object(object->pointer);
-       rcu_read_unlock();
+       if (cleanup) {
+               rcu_read_lock();
+               list_for_each_entry_rcu(object, &object_list, object_list)
+                       delete_object_full(object->pointer);
+               rcu_read_unlock();
+       }
        mutex_unlock(&scan_mutex);
-
-       return 0;
 }
 
-/*
- * Start the clean-up thread.
- */
-static void kmemleak_cleanup(void)
-{
-       struct task_struct *cleanup_thread;
-
-       cleanup_thread = kthread_run(kmemleak_cleanup_thread, NULL,
-                                    "kmemleak-clean");
-       if (IS_ERR(cleanup_thread))
-               pr_warning("Failed to create the clean-up thread\n");
-}
+static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
 
 /*
  * Disable kmemleak. No memory allocation/freeing will be traced once this
@@ -1350,12 +1710,11 @@ static void kmemleak_disable(void)
                return;
 
        /* stop any memory operation tracing */
-       atomic_set(&kmemleak_early_log, 0);
        atomic_set(&kmemleak_enabled, 0);
 
        /* check whether it is too early for a kernel thread */
        if (atomic_read(&kmemleak_initialized))
-               kmemleak_cleanup();
+               schedule_work(&cleanup_work);
 
        pr_info("Kernel memory leak detector disabled\n");
 }
@@ -1369,12 +1728,25 @@ static int kmemleak_boot_config(char *str)
                return -EINVAL;
        if (strcmp(str, "off") == 0)
                kmemleak_disable();
-       else if (strcmp(str, "on") != 0)
+       else if (strcmp(str, "on") == 0)
+               kmemleak_skip_disable = 1;
+       else
                return -EINVAL;
        return 0;
 }
 early_param("kmemleak", kmemleak_boot_config);
 
+static void __init print_log_trace(struct early_log *log)
+{
+       struct stack_trace trace;
+
+       trace.nr_entries = log->trace_len;
+       trace.entries = log->trace;
+
+       pr_notice("Early log backtrace:\n");
+       print_stack_trace(&trace, 2);
+}
+
 /*
  * Kmemleak initialization.
  */
@@ -1383,7 +1755,14 @@ void __init kmemleak_init(void)
        int i;
        unsigned long flags;
 
-       jiffies_scan_yield = msecs_to_jiffies(MSECS_SCAN_YIELD);
+#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
+       if (!kmemleak_skip_disable) {
+               atomic_set(&kmemleak_early_log, 0);
+               kmemleak_disable();
+               return;
+       }
+#endif
+
        jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
        jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
 
@@ -1391,12 +1770,18 @@ void __init kmemleak_init(void)
        scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
        INIT_PRIO_TREE_ROOT(&object_tree_root);
 
+       if (crt_early_log >= ARRAY_SIZE(early_log))
+               pr_warning("Early log buffer exceeded (%d), please increase "
+                          "DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n", crt_early_log);
+
        /* the kernel is still in UP mode, so disabling the IRQs is enough */
        local_irq_save(flags);
-       if (!atomic_read(&kmemleak_error)) {
+       atomic_set(&kmemleak_early_log, 0);
+       if (atomic_read(&kmemleak_error)) {
+               local_irq_restore(flags);
+               return;
+       } else
                atomic_set(&kmemleak_enabled, 1);
-               atomic_set(&kmemleak_early_log, 0);
-       }
        local_irq_restore(flags);
 
        /*
@@ -1409,12 +1794,20 @@ void __init kmemleak_init(void)
 
                switch (log->op_type) {
                case KMEMLEAK_ALLOC:
-                       kmemleak_alloc(log->ptr, log->size, log->min_count,
-                                      GFP_KERNEL);
+                       early_alloc(log);
+                       break;
+               case KMEMLEAK_ALLOC_PERCPU:
+                       early_alloc_percpu(log);
                        break;
                case KMEMLEAK_FREE:
                        kmemleak_free(log->ptr);
                        break;
+               case KMEMLEAK_FREE_PART:
+                       kmemleak_free_part(log->ptr, log->size);
+                       break;
+               case KMEMLEAK_FREE_PERCPU:
+                       kmemleak_free_percpu(log->ptr);
+                       break;
                case KMEMLEAK_NOT_LEAK:
                        kmemleak_not_leak(log->ptr);
                        break;
@@ -1422,14 +1815,19 @@ void __init kmemleak_init(void)
                        kmemleak_ignore(log->ptr);
                        break;
                case KMEMLEAK_SCAN_AREA:
-                       kmemleak_scan_area(log->ptr, log->offset, log->length,
-                                          GFP_KERNEL);
+                       kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL);
                        break;
                case KMEMLEAK_NO_SCAN:
                        kmemleak_no_scan(log->ptr);
                        break;
                default:
-                       WARN_ON(1);
+                       kmemleak_warn("Unknown early log operation: %d\n",
+                                     log->op_type);
+               }
+
+               if (atomic_read(&kmemleak_warning)) {
+                       print_log_trace(log);
+                       atomic_set(&kmemleak_warning, 0);
                }
        }
 }
@@ -1445,12 +1843,12 @@ static int __init kmemleak_late_init(void)
 
        if (atomic_read(&kmemleak_error)) {
                /*
-                * Some error occured and kmemleak was disabled. There is a
+                * Some error occurred and kmemleak was disabled. There is a
                 * small chance that kmemleak_disable() was called immediately
                 * after setting kmemleak_initialized and we may end up with
                 * two clean-up threads but serialized by scan_mutex.
                 */
-               kmemleak_cleanup();
+               schedule_work(&cleanup_work);
                return -ENOMEM;
        }