Re: [PATCH v2 3/6] mm/slub: use stackdepot to save stack trace in objects

[Vlastimil Babka]

On 3/4/22 12:25, Hyeonggon Yoo wrote:
> On Wed, Mar 02, 2022 at 06:31:19PM +0100, Vlastimil Babka wrote:
>> From: Oliver Glitta <glittao@xxxxxxxxx>
>> 
>> Many stack traces are similar so there are many similar arrays.
>> Stackdepot saves each unique stack only once.
>> 
>> Replace field addrs in struct track with depot_stack_handle_t handle.  Use
>> stackdepot to save stack trace.
>> 
>> The benefits are smaller memory overhead and possibility to aggregate
>> per-cache statistics in the following patch using the stackdepot handle
>> instead of matching stacks manually.
>> 
>> [ vbabka@xxxxxxx: rebase to 5.17-rc1 and adjust accordingly ]
>> 
>> This was initially merged as commit 788691464c29 and reverted by commit
>> ae14c63a9f20 due to several issues, that should now be fixed.
>> The problem of unconditional memory overhead by stackdepot has been
>> addressed by commit 2dba5eb1c73b ("lib/stackdepot: allow optional init
>> and stack_table allocation by kvmalloc()"), so the dependency on
>> stackdepot will result in extra memory usage only when a slab cache
>> tracking is actually enabled, and not for all CONFIG_SLUB_DEBUG builds.
>> The build failures on some architectures were also addressed, and the
>> reported issue with xfs/433 test did not reproduce on 5.17-rc1 with this
>> patch.
>> 
>> Signed-off-by: Oliver Glitta <glittao@xxxxxxxxx>
>> Signed-off-by: Vlastimil Babka <vbabka@xxxxxxx>
>> Cc: David Rientjes <rientjes@xxxxxxxxxx>
>> Cc: Christoph Lameter <cl@xxxxxxxxx>
>> Cc: Pekka Enberg <penberg@xxxxxxxxxx>
>> Cc: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>

...

>> @@ -314,9 +315,13 @@ kmem_cache_create_usercopy(const char *name,
>>  	 * If no slub_debug was enabled globally, the static key is not yet
>>  	 * enabled by setup_slub_debug(). Enable it if the cache is being
>>  	 * created with any of the debugging flags passed explicitly.
>> +	 * It's also possible that this is the first cache created with
>> +	 * SLAB_STORE_USER and we should init stack_depot for it.
>>  	 */
>>  	if (flags & SLAB_DEBUG_FLAGS)
>>  		static_branch_enable(&slub_debug_enabled);
>> +	if (flags & SLAB_STORE_USER && IS_ENABLED(CONFIG_STACKDEPOT))
>> +		stack_depot_init();
>>  #endif
> 
> Is this comment and code still valid in v3?

The comment is still valid, as there can be a kmem_cache_create() call with
SLAB_STORE_USER (in fact there's one in kernel/rcu/rcutorture.c) that's not
covered by the slub_debug parsing.
The code in v3 is just without IS_ENABLED(CONFIG_STACKDEPOT).

>>  	mutex_lock(&slab_mutex);
>> diff --git a/mm/slub.c b/mm/slub.c
>> index 1fc451f4fe62..42cb79af70a0 100644
>> --- a/mm/slub.c
>> +++ b/mm/slub.c
>> @@ -26,6 +26,7 @@
>>  #include <linux/cpuset.h>
>>  #include <linux/mempolicy.h>
>>  #include <linux/ctype.h>
>> +#include <linux/stackdepot.h>
>>  #include <linux/debugobjects.h>
>>  #include <linux/kallsyms.h>
>>  #include <linux/kfence.h>
>> @@ -264,8 +265,8 @@ static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
>>  #define TRACK_ADDRS_COUNT 16
>>  struct track {
>>  	unsigned long addr;	/* Called from address */
>> -#ifdef CONFIG_STACKTRACE
>> -	unsigned long addrs[TRACK_ADDRS_COUNT];	/* Called from address */
>> +#ifdef CONFIG_STACKDEPOT
>> +	depot_stack_handle_t handle;
>>  #endif
>>  	int cpu;		/* Was running on cpu */
>>  	int pid;		/* Pid context */
>> @@ -724,22 +725,19 @@ static struct track *get_track(struct kmem_cache *s, void *object,
>>  	return kasan_reset_tag(p + alloc);
>>  }
>>  
>> -static void set_track(struct kmem_cache *s, void *object,
>> +static void noinline set_track(struct kmem_cache *s, void *object,
>>  			enum track_item alloc, unsigned long addr)
>>  {
> 
> noinline for debugging purpose?
> I think it's okay. just a question.

These noinlines make sure that the amount of stack entries are stable and
not subject to inline decisions of compiler...

>>  	struct track *p = get_track(s, object, alloc);
>>  
>> -#ifdef CONFIG_STACKTRACE
>> +#ifdef CONFIG_STACKDEPOT
>> +	unsigned long entries[TRACK_ADDRS_COUNT];
>>  	unsigned int nr_entries;
>>  
>> -	metadata_access_enable();
>> -	nr_entries = stack_trace_save(kasan_reset_tag(p->addrs),
>> -				      TRACK_ADDRS_COUNT, 3);
>> -	metadata_access_disable();
>> -
>> -	if (nr_entries < TRACK_ADDRS_COUNT)
>> -		p->addrs[nr_entries] = 0;
>> +	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);

... so that here '3' removes the correct count of 'internal' stack trace
entries that are not interesting for us.

>> +	p->handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
>>  #endif
>> +
>>  	p->addr = addr;
>>  	p->cpu = smp_processor_id();
>>  	p->pid = current->pid;
>> @@ -759,20 +757,19 @@ static void init_tracking(struct kmem_cache *s, void *object)
>>  
>>  static void print_track(const char *s, struct track *t, unsigned long pr_time)
>>  {
>> +	depot_stack_handle_t handle __maybe_unused;
>> +
>>  	if (!t->addr)
>>  		return;
>>  
>>  	pr_err("%s in %pS age=%lu cpu=%u pid=%d\n",
>>  	       s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
>> -#ifdef CONFIG_STACKTRACE
>> -	{
>> -		int i;
>> -		for (i = 0; i < TRACK_ADDRS_COUNT; i++)
>> -			if (t->addrs[i])
>> -				pr_err("\t%pS\n", (void *)t->addrs[i]);
>> -			else
>> -				break;
>> -	}
>> +#ifdef CONFIG_STACKDEPOT
>> +	handle = READ_ONCE(t->handle);
>> +	if (handle)
>> +		stack_depot_print(handle);
>> +	else
>> +		pr_err("object allocation/free stack trace missing\n");
>>  #endif
>>  }
>>  
>> @@ -1532,6 +1529,8 @@ static int __init setup_slub_debug(char *str)
>>  			global_slub_debug_changed = true;
>>  		} else {
>>  			slab_list_specified = true;
>> +			if (flags & SLAB_STORE_USER)
>> +				stack_depot_want_early_init = true;
> 
> This is updated to stack_depot_want_early_init() in v3.

Yes.

>>  		}
>>  	}
>>  
>> @@ -1549,6 +1548,8 @@ static int __init setup_slub_debug(char *str)
>>  	}
>>  out:
>>  	slub_debug = global_flags;
>> +	if (slub_debug & SLAB_STORE_USER)
>> +		stack_depot_want_early_init = true;
> 
> This too.

Yes.

>>  	if (slub_debug != 0 || slub_debug_string)
>>  		static_branch_enable(&slub_debug_enabled);
>>  	else
>> @@ -4352,18 +4353,26 @@ void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
>>  	objp = fixup_red_left(s, objp);
>>  	trackp = get_track(s, objp, TRACK_ALLOC);
>>  	kpp->kp_ret = (void *)trackp->addr;
>> -#ifdef CONFIG_STACKTRACE
>> -	for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) {
>> -		kpp->kp_stack[i] = (void *)trackp->addrs[i];
>> -		if (!kpp->kp_stack[i])
>> -			break;
>> -	}
>> +#ifdef CONFIG_STACKDEPOT
>> +	{
>> +		depot_stack_handle_t handle;
>> +		unsigned long *entries;
>> +		unsigned int nr_entries;
>> +
>> +		handle = READ_ONCE(trackp->handle);
>> +		if (handle) {
>> +			nr_entries = stack_depot_fetch(handle, &entries);
>> +			for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
>> +				kpp->kp_stack[i] = (void *)entries[i];
>> +		}
>>  
>> -	trackp = get_track(s, objp, TRACK_FREE);
>> -	for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) {
>> -		kpp->kp_free_stack[i] = (void *)trackp->addrs[i];
>> -		if (!kpp->kp_free_stack[i])
>> -			break;
>> +		trackp = get_track(s, objp, TRACK_FREE);
>> +		handle = READ_ONCE(trackp->handle);
>> +		if (handle) {
>> +			nr_entries = stack_depot_fetch(handle, &entries);
>> +			for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
>> +				kpp->kp_free_stack[i] = (void *)entries[i];
>> +		}
>>  	}
>>  #endif
>>  #endif
>> -- 
>> 2.35.1
> 
> Otherwise looks good.
> Reviewed-by: Hyeonggon Yoo <42.hyeyoo@xxxxxxxxx>
> 
> kmem_dump_obj() and slab error report functionality works fine.
> Tested-by: Hyeonggon Yoo <42.hyeyoo@xxxxxxxxx>

Thanks!