Message-ID: <20100830203037.GJ2420@linux.vnet.ibm.com>
Date:	Mon, 30 Aug 2010 13:30:37 -0700
From:	"Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
To:	Hiroshi DOYU <Hiroshi.DOYU@...ia.com>
Cc:	catalin.marinas@....com, linux-kernel@...r.kernel.org,
	ext-phil.2.carmody@...ia.com, linux-omap@...r.kernel.org
Subject: Re: [RFC][PATCH 0/1] kmemleak: Fix false positive with alias

On Fri, Aug 27, 2010 at 09:12:24AM +0300, Hiroshi DOYU wrote:
> Hi Catalin,
> 
> From: Hiroshi DOYU <Hiroshi.DOYU@...ia.com>
> Subject: Re: [RFC][PATCH 0/1] kmemleak: Fix false positive with alias
> Date: Tue, 10 Aug 2010 18:49:03 +0300 (EEST)
> 
> <snip>
> >>> The performance impact is indeed pretty high, though some parts of the
> >>> code look over-engineered to me (the __scan_block function with a loop
> >>> going through an array of two function pointers - I think the compiler
> >>> cannot figure out what to inline). You could just extend the
> >>> find_and_get_object() to search both trees under a single spinlock
> >>> region (as locking also takes time).
> >> 
> >> Ok, a good point.
> > 
> > Now there's not much difference with the attached patch, a new version
> > of alias.
> > 
> > / # modprobe kmemleak-special-test use_alias=0
> > / # time echo scan > /sys/kernel/debug/kmemleak
> > real    0m 2.30s
> > user    0m 0.00s
> > sys     0m 2.30s
> > 
> > / # modprobe kmemleak-special-test use_alias=1
> > / # time echo scan > /sys/kernel/debug/kmemleak
> > real    0m 3.91s
> > user    0m 0.00s
> > sys     0m 3.91s
> 
> It would be nice if you could have some time to take a look at this
> patch and give some comments.
> 
> > From a5670d69b2cafe85f6f26f6951097210d3b9917f Mon Sep 17 00:00:00 2001
> > From: Hiroshi DOYU <Hiroshi.DOYU@...ia.com>
> > Date: Thu, 17 Jun 2010 13:36:45 +0300
> > Subject: [PATCH 1/1] kmemleak: Fix false positive with address alias
> > 
> > There is a false positive case that a pointer is calculated by other
> > methods than the usual container_of macro. "kmemleak_ignore" can cover
> > such a false positive, but it would loose the advantage of memory leak
> > detection. This patch allows kmemleak to work with such false
> > positives by aliasing of address. A client module can register an
> > alias address to an original pointer.
> > 
> > A typical use case could be the IOMMU pagetable allocation which
> > stores pointers to the second level of page tables with some
> > conversion, for example, a physical address with attribute bits. Right
> > now I don't have other use cases but I hope that there could be some
> > that this special scan works with.

A few questions below...

							Thanx, Paul

> > Signed-off-by: Hiroshi DOYU <Hiroshi.DOYU@...ia.com>
> > Cc: Phil Carmody <ext-phil.2.carmody@...ia.com>
> > ---
> >  include/linux/kmemleak.h |    8 ++
> >  mm/kmemleak.c            |  208 +++++++++++++++++++++++++++++++++++++++++++---
> >  2 files changed, 204 insertions(+), 12 deletions(-)
> > 
> > diff --git a/include/linux/kmemleak.h b/include/linux/kmemleak.h
> > index 99d9a67..9e2af3a 100644
> > --- a/include/linux/kmemleak.h
> > +++ b/include/linux/kmemleak.h
> > @@ -34,6 +34,8 @@ extern void kmemleak_not_leak(const void *ptr) __ref;
> >  extern void kmemleak_ignore(const void *ptr) __ref;
> >  extern void kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp) __ref;
> >  extern void kmemleak_no_scan(const void *ptr) __ref;
> > +extern void kmemleak_add_alias(const void *ptr, const void *new) __ref;
> > +extern void kmemleak_unalias(const void *alias) __ref;
> >  
> >  static inline void kmemleak_alloc_recursive(const void *ptr, size_t size,
> >  					    int min_count, unsigned long flags,
> > @@ -92,6 +94,12 @@ static inline void kmemleak_erase(void **ptr)
> >  static inline void kmemleak_no_scan(const void *ptr)
> >  {
> >  }
> > +static inline void kmemleak_add_alias(const void *ptr, const void *new)
> > +{
> > +}
> > +static inline void kmemleak_unalias(const void *alias)
> > +{
> > +}
> >  
> >  #endif	/* CONFIG_DEBUG_KMEMLEAK */
> >  
> > diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> > index 2c0d032..3875cb7 100644
> > --- a/mm/kmemleak.c
> > +++ b/mm/kmemleak.c
> > @@ -157,6 +157,13 @@ struct kmemleak_object {
> >  	unsigned long jiffies;		/* creation timestamp */
> >  	pid_t pid;			/* pid of the current task */
> >  	char comm[TASK_COMM_LEN];	/* executable name */
> > +	struct kmemleak_alias *alias;	/* if a pointer is modified */
> > +};
> > +
> > +struct kmemleak_alias {
> > +	struct list_head alias_list;
> > +	struct prio_tree_node tree_node;
> > +	struct kmemleak_object *object;
> >  };
> >  
> >  /* flag representing the memory block allocation status */
> > @@ -179,13 +186,18 @@ struct kmemleak_object {
> >  static LIST_HEAD(object_list);
> >  /* the list of gray-colored objects (see color_gray comment below) */
> >  static LIST_HEAD(gray_list);
> > +/* the list of objects with alias (see alias comment below) */
> > +static LIST_HEAD(alias_list);
> >  /* prio search tree for object boundaries */
> >  static struct prio_tree_root object_tree_root;
> > +/* prio search tree for alias object boundaries */
> > +static struct prio_tree_root alias_tree_root;
> >  /* rw_lock protecting the access to object_list and prio_tree_root */
> >  static DEFINE_RWLOCK(kmemleak_lock);
> >  
> >  /* allocation caches for kmemleak internal data */
> >  static struct kmem_cache *object_cache;
> > +static struct kmem_cache *alias_cache;
> >  static struct kmem_cache *scan_area_cache;
> >  
> >  /* set if tracing memory operations is enabled */
> > @@ -269,6 +281,8 @@ static void kmemleak_disable(void);
> >  	kmemleak_disable();		\
> >  } while (0)
> >  
> > +#define to_address(obj) ((obj)->tree_node.start)
> > +
> >  /*
> >   * 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
> > @@ -369,7 +383,7 @@ static void dump_object_info(struct kmemleak_object *object)
> >  	trace.entries = object->trace;
> >  
> >  	pr_notice("Object 0x%08lx (size %zu):\n",
> > -		  object->tree_node.start, object->size);
> > +		  to_address(object), object->size);
> >  	pr_notice("  comm \"%s\", pid %d, jiffies %lu\n",
> >  		  object->comm, object->pid, object->jiffies);
> >  	pr_notice("  min_count = %d\n", object->min_count);
> > @@ -436,6 +450,8 @@ static void free_object_rcu(struct rcu_head *rcu)
> >  		hlist_del(elem);
> >  		kmem_cache_free(scan_area_cache, area);
> >  	}
> > +	if (object->alias)
> > +		kmem_cache_free(alias_cache, object->alias);
> >  	kmem_cache_free(object_cache, object);
> >  }
> >  
> > @@ -460,12 +476,11 @@ static void put_object(struct kmemleak_object *object)
> >  /*
> >   * Look up an object in the prio search tree and increase its use_count.
> >   */
> > -static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
> > +static struct kmemleak_object *__find_and_get_object(unsigned long ptr, int alias)
> >  {
> >  	unsigned long flags;
> >  	struct kmemleak_object *object = NULL;
> >  
> > -	rcu_read_lock();
> >  	read_lock_irqsave(&kmemleak_lock, flags);
> >  	if (ptr >= min_addr && ptr < max_addr)
> >  		object = lookup_object(ptr, alias);
> > @@ -474,6 +489,75 @@ static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
> >  	/* check whether the object is still available */
> >  	if (object && !get_object(object))
> >  		object = NULL;
> > +
> > +	return object;
> > +}
> > +
> > +static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
> > +{
> > +	struct kmemleak_object *object;
> > +
> > +	rcu_read_lock();
> > +	object = __find_and_get_object(ptr, alias);
> > +	rcu_read_unlock();
> > +
> > +	return object;
> > +}
> > +
> > +static struct kmemleak_object *__find_and_get_alias(unsigned long ptr, int alias)
> > +{
> > +	struct kmemleak_object *object = NULL;
> > +	struct kmemleak_alias *ao = NULL;
> > +	struct prio_tree_node *node;
> > +	struct prio_tree_iter iter;
> > +	unsigned long flags;
> > +
> > +	read_lock_irqsave(&kmemleak_lock, flags);

If we hold this readlock, how is RCU helping us?  Or are there updates
that do not write-hold kmemleak_lock?

> > +
> > +	prio_tree_iter_init(&iter, &alias_tree_root, ptr, ptr);
> > +	node = prio_tree_next(&iter);
> > +	if (node) {
> > +		ao = prio_tree_entry(node, struct kmemleak_alias, tree_node);
> > +		if (!alias && to_address(ao) != ptr) {
> > +			kmemleak_warn("Found object by alias");
> > +			ao = NULL;
> > +		}
> > +	}
> > +
> > +	read_unlock_irqrestore(&kmemleak_lock, flags);
> > +
> > +	if (ao && get_object(ao->object))
> > +		object = ao->object;
> > +
> > +	return object;
> > +}
> > +
> > +static struct kmemleak_object *find_and_get_alias(unsigned long ptr, int alias)
> > +{
> > +	struct kmemleak_object *object = NULL;
> > +
> > +	rcu_read_lock();
> > +	object = __find_and_get_alias(ptr, alias);
> > +	rcu_read_unlock();
> > +
> > +	return object;
> > +}
> > +
> > +/*
> > + * Try to find object first, and then with alias address if not found.
> > + */
> > +static struct kmemleak_object *find_and_get_object_with_alias(unsigned long ptr,
> > +							      int alias)
> > +{
> > +	struct kmemleak_object *object = NULL;
> > +
> > +	rcu_read_lock();
> > +
> > +	object = __find_and_get_object(ptr, alias);
> > +	if (!object &&
> > +	    !prio_tree_empty(&alias_tree_root))
> > +		object = __find_and_get_alias(ptr, alias);
> > +
> >  	rcu_read_unlock();
> >  
> >  	return object;
> > @@ -524,6 +608,7 @@ static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
> >  	object->count = 0;			/* white color initially */
> >  	object->jiffies = jiffies;
> >  	object->checksum = 0;
> > +	object->alias = NULL;
> >  
> >  	/* task information */
> >  	if (in_irq()) {
> > @@ -547,7 +632,7 @@ static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
> >  	object->trace_len = __save_stack_trace(object->trace);
> >  
> >  	INIT_PRIO_TREE_NODE(&object->tree_node);
> > -	object->tree_node.start = ptr;
> > +	to_address(object) = ptr;
> >  	object->tree_node.last = ptr + size - 1;
> >  
> >  	write_lock_irqsave(&kmemleak_lock, flags);
> > @@ -577,6 +662,57 @@ out:
> >  	return object;
> >  }
> >  
> > +static void create_alias_object(struct kmemleak_object *object,
> > +				unsigned long ptr)
> > +{
> > +	struct kmemleak_alias *alias;
> > +	struct prio_tree_node *node;
> > +	unsigned long flags;
> > +
> > +	alias = kmem_cache_alloc(alias_cache, GFP_KERNEL);
> > +	if (!alias) {
> > +		kmemleak_stop("Cannot allocate a kmemleak_alias structure\n");
> > +		return;
> > +	}
> > +	INIT_LIST_HEAD(&alias->alias_list);
> > +	INIT_PRIO_TREE_NODE(&alias->tree_node);
> > +	to_address(alias) = ptr;
> > +
> > +	spin_lock_irqsave(&object->lock, flags);
> > +	alias->tree_node.last = ptr + object->size - 1;
> > +	alias->object = object;
> > +	object->alias = alias;
> > +	spin_unlock_irqrestore(&object->lock, flags);
> > +
> > +	write_lock_irqsave(&kmemleak_lock, flags);
> > +
> > +	node = prio_tree_insert(&alias_tree_root, &alias->tree_node);
> > +	if (!node) {
> > +		kmemleak_warn("Cannot allocate a kmemleak_alias structure\n");
> > +		kmem_cache_free(alias_cache, alias);
> > +	}
> > +	list_add_tail_rcu(&alias->alias_list, &alias_list);
> > +
> > +	write_unlock_irqrestore(&kmemleak_lock, flags);
> > +}
> > +
> > +static void __delete_alias_object(struct kmemleak_object *object)
> > +{
> > +	prio_tree_remove(&alias_tree_root, &object->alias->tree_node);
> > +	list_del_rcu(&object->alias->alias_list);

Don't we need an RCU grace period here, based on either synchronize_rcu()
or call_rcu()?  Perhaps calling free_object_rcu(), though perhaps it frees
up more than you would like.

> > +	kmem_cache_free(alias_cache, object->alias);
> > +	object->alias = NULL;
> > +}
> > +
> > +static void delete_alias_object(struct kmemleak_object *object)
> > +{
> > +	unsigned long flags;
> > +
> > +	write_lock_irqsave(&kmemleak_lock, flags);
> > +	__delete_alias_object(object);
> > +	write_unlock_irqrestore(&kmemleak_lock, flags);
> > +}
> > +
> >  /*
> >   * Remove the metadata (struct kmemleak_object) for a memory block from the
> >   * object_list and object_tree_root and decrement its use_count.
> > @@ -588,6 +724,8 @@ static void __delete_object(struct kmemleak_object *object)
> >  	write_lock_irqsave(&kmemleak_lock, flags);
> >  	prio_tree_remove(&object_tree_root, &object->tree_node);
> >  	list_del_rcu(&object->object_list);
> > +	if (object->alias)
> > +		__delete_alias_object(object);
> >  	write_unlock_irqrestore(&kmemleak_lock, flags);
> >  
> >  	WARN_ON(!(object->flags & OBJECT_ALLOCATED));
> > @@ -630,7 +768,7 @@ static void delete_object_full(unsigned long ptr)
> >   */
> >  static void delete_object_part(unsigned long ptr, size_t size)
> >  {
> > -	struct kmemleak_object *object;
> > +	struct kmemleak_object *object, *new;
> >  	unsigned long start, end;
> >  
> >  	object = find_and_get_object(ptr, 1);
> > @@ -652,12 +790,24 @@ static void delete_object_part(unsigned long ptr, size_t size)
> >  	 */
> >  	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);
> > +	if (ptr > start) {
> > +		new = create_object(start, ptr - start, object->min_count,
> > +				    GFP_KERNEL);
> > +		if (new && unlikely(object->alias))
> > +			create_alias_object(new, to_address(object->alias));
> > +	}
> > +	if (ptr + size < end) {
> > +		new = create_object(ptr + size, end - ptr - size,
> > +				    object->min_count, GFP_KERNEL);
> > +		if (new && unlikely(object->alias)) {
> > +			unsigned long alias_ptr;
> > +
> > +			alias_ptr = to_address(object->alias);
> > +			alias_ptr += ptr - start + size;
> > +
> > +			create_alias_object(new, alias_ptr);
> > +		}
> > +	}
> >  
> >  	put_object(object);
> >  }
> > @@ -944,6 +1094,38 @@ void __ref kmemleak_no_scan(const void *ptr)
> >  }
> >  EXPORT_SYMBOL(kmemleak_no_scan);
> >  
> > +void kmemleak_add_alias(const void *ptr, const void *alias)
> > +{
> > +	struct kmemleak_object *object;
> > +
> > +	pr_debug("%s(0x%p -> 0x%p)\n", __func__, ptr, alias);
> > +
> > +	object = find_and_get_object((unsigned long)ptr, 0);
> > +	if (!object) {
> > +		kmemleak_warn("Aliasing unknown object at 0x%p\n", ptr);
> > +		return;
> > +	}
> > +	create_alias_object(object, (unsigned long)alias);
> > +	put_object(object);
> > +}
> > +EXPORT_SYMBOL(kmemleak_add_alias);
> > +
> > +void kmemleak_unalias(const void *alias)
> > +{
> > +	struct kmemleak_object *object;
> > +
> > +	pr_debug("%s(0x%p)\n", __func__, alias);
> > +
> > +	object = find_and_get_alias((unsigned long)alias, 0);
> > +	if (!object) {
> > +		kmemleak_warn("Aliasing unknown object at 0x%p\n", alias);
> > +		return;
> > +	}
> > +	delete_alias_object(object);
> > +	put_object(object);
> > +}
> > +EXPORT_SYMBOL(kmemleak_unalias);
> > +
> >  /*
> >   * Update an object's checksum and return true if it was modified.
> >   */
> > @@ -1007,7 +1189,7 @@ static void scan_block(void *_start, void *_end,
> >  
> >  		pointer = *ptr;
> >  
> > -		object = find_and_get_object(pointer, 1);
> > +		object = find_and_get_object_with_alias(pointer, 1);
> >  		if (!object)
> >  			continue;
> >  		if (object == scanned) {
> > @@ -1620,8 +1802,10 @@ void __init kmemleak_init(void)
> >  	jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
> >  
> >  	object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
> > +	alias_cache = KMEM_CACHE(kmemleak_alias, SLAB_NOLEAKTRACE);
> >  	scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
> >  	INIT_PRIO_TREE_ROOT(&object_tree_root);
> > +	INIT_PRIO_TREE_ROOT(&alias_tree_root);
> >  
> >  	/* the kernel is still in UP mode, so disabling the IRQs is enough */
> >  	local_irq_save(flags);
> > -- 
> > 1.7.1.rc2
> > 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@...r.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists