Re: [PATCH tip/core/rcu 1/2] rcu: Support kfree_bulk() interface in kfree_rcu()

[Boqun Feng]

Hi,

On Fri, Feb 14, 2020 at 04:00:52PM -0800, paulmck@xxxxxxxxxx wrote:
> From: "Uladzislau Rezki (Sony)" <urezki@xxxxxxxxx>
> 
> The kfree_rcu() logic can be improved further by using kfree_bulk()
> interface along with "basic batching support" introduced earlier.
> 
> The are at least two advantages of using "bulk" interface:
> - in case of large number of kfree_rcu() requests kfree_bulk()
>   reduces the per-object overhead caused by calling kfree()
>   per-object.
> 
> - reduces the number of cache-misses due to "pointer chasing"
>   between objects which can be far spread between each other.
> 
> This approach defines a new kfree_rcu_bulk_data structure that
> stores pointers in an array with a specific size. Number of entries
> in that array depends on PAGE_SIZE making kfree_rcu_bulk_data
> structure to be exactly one page.
> 
> Since it deals with "block-chain" technique there is an extra
> need in dynamic allocation when a new block is required. Memory
> is allocated with GFP_NOWAIT | __GFP_NOWARN flags, i.e. that
> allows to skip direct reclaim under low memory condition to
> prevent stalling and fails silently under high memory pressure.
> 
> The "emergency path" gets maintained when a system is run out of
> memory. In that case objects are linked into regular list.
> 
> The "rcuperf" was run to analyze this change in terms of memory
> consumption and kfree_bulk() throughput.
> 
> 1) Testing on the Intel(R) Xeon(R) W-2135 CPU @ 3.70GHz, 12xCPUs
> with following parameters:
> 
> kfree_loops=200000 kfree_alloc_num=1000 kfree_rcu_test=1 kfree_vary_obj_size=1
> dev.2020.01.10a branch
> 
> Default / CONFIG_SLAB
> 53607352517 ns, loops: 200000, batches: 1885, memory footprint: 1248MB
> 53529637912 ns, loops: 200000, batches: 1921, memory footprint: 1193MB
> 53570175705 ns, loops: 200000, batches: 1929, memory footprint: 1250MB
> 
> Patch / CONFIG_SLAB
> 23981587315 ns, loops: 200000, batches: 810, memory footprint: 1219MB
> 23879375281 ns, loops: 200000, batches: 822, memory footprint: 1190MB
> 24086841707 ns, loops: 200000, batches: 794, memory footprint: 1380MB
> 
> Default / CONFIG_SLUB
> 51291025022 ns, loops: 200000, batches: 1713, memory footprint: 741MB
> 51278911477 ns, loops: 200000, batches: 1671, memory footprint: 719MB
> 51256183045 ns, loops: 200000, batches: 1719, memory footprint: 647MB
> 
> Patch / CONFIG_SLUB
> 50709919132 ns, loops: 200000, batches: 1618, memory footprint: 456MB
> 50736297452 ns, loops: 200000, batches: 1633, memory footprint: 507MB
> 50660403893 ns, loops: 200000, batches: 1628, memory footprint: 429MB
> 
> in case of CONFIG_SLAB there is double increase in performance and
> slightly higher memory usage. As for CONFIG_SLUB, the performance
> figures are better together with lower memory usage.
> 
> 2) Testing on the HiKey-960, arm64, 8xCPUs with below parameters:
> 
> CONFIG_SLAB=y
> kfree_loops=200000 kfree_alloc_num=1000 kfree_rcu_test=1
> 
> 102898760401 ns, loops: 200000, batches: 5822, memory footprint: 158MB
> 89947009882  ns, loops: 200000, batches: 6715, memory footprint: 115MB
> 
> rcuperf shows approximately ~12% better throughput in case of
> using "bulk" interface. The "drain logic" or its RCU callback
> does the work faster that leads to better throughput.
> 
> Signed-off-by: Uladzislau Rezki (Sony) <urezki@xxxxxxxxx>
> Tested-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> ---
>  kernel/rcu/tree.c | 204 ++++++++++++++++++++++++++++++++++++++++++++----------
>  1 file changed, 169 insertions(+), 35 deletions(-)
> 
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index d91c915..51a3aa8 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -2689,22 +2689,47 @@ EXPORT_SYMBOL_GPL(call_rcu);
>  #define KFREE_DRAIN_JIFFIES (HZ / 50)
>  #define KFREE_N_BATCHES 2
>  
> +/*
> + * This macro defines how many entries the "records" array
> + * will contain. It is based on the fact that the size of
> + * kfree_rcu_bulk_data structure becomes exactly one page.
> + */
> +#define KFREE_BULK_MAX_ENTR ((PAGE_SIZE / sizeof(void *)) - 3)
> +

Why use the "magic" number "3" here? Could we just define struct
kfree_rcu_bulk_data as:

struct kfree_rcu_bulk_data {
	struct kfree_rcu_bulk_data *next;
	struct rcu_head *head_free_debug;
	unsigned long nr_records;
	void *records[];
}

?

And the the above macro becomes:

#define KFREE_BULK_MAX_ENTR ((PAGE_SIZE - sizeof(struct kfree_rcu_bulk_data)) / sizeof(void *))

Regards,
Boqun

> +/**
> + * struct kfree_rcu_bulk_data - single block to store kfree_rcu() pointers
> + * @nr_records: Number of active pointers in the array
> + * @records: Array of the kfree_rcu() pointers
> + * @next: Next bulk object in the block chain
> + * @head_free_debug: For debug, when CONFIG_DEBUG_OBJECTS_RCU_HEAD is set
> + */
> +struct kfree_rcu_bulk_data {
> +	unsigned long nr_records;
> +	void *records[KFREE_BULK_MAX_ENTR];
> +	struct kfree_rcu_bulk_data *next;
> +	struct rcu_head *head_free_debug;
> +};
> +
>  /**
>   * struct kfree_rcu_cpu_work - single batch of kfree_rcu() requests
>   * @rcu_work: Let queue_rcu_work() invoke workqueue handler after grace period
>   * @head_free: List of kfree_rcu() objects waiting for a grace period
> + * @bhead_free: Bulk-List of kfree_rcu() objects waiting for a grace period
>   * @krcp: Pointer to @kfree_rcu_cpu structure
>   */
>  
>  struct kfree_rcu_cpu_work {
>  	struct rcu_work rcu_work;
>  	struct rcu_head *head_free;
> +	struct kfree_rcu_bulk_data *bhead_free;
>  	struct kfree_rcu_cpu *krcp;
>  };
>  
>  /**
>   * struct kfree_rcu_cpu - batch up kfree_rcu() requests for RCU grace period
>   * @head: List of kfree_rcu() objects not yet waiting for a grace period
> + * @bhead: Bulk-List of kfree_rcu() objects not yet waiting for a grace period
> + * @bcached: Keeps at most one object for later reuse when build chain blocks
>   * @krw_arr: Array of batches of kfree_rcu() objects waiting for a grace period
>   * @lock: Synchronize access to this structure
>   * @monitor_work: Promote @head to @head_free after KFREE_DRAIN_JIFFIES
> @@ -2718,6 +2743,8 @@ struct kfree_rcu_cpu_work {
>   */
>  struct kfree_rcu_cpu {
>  	struct rcu_head *head;
> +	struct kfree_rcu_bulk_data *bhead;
> +	struct kfree_rcu_bulk_data *bcached;
>  	struct kfree_rcu_cpu_work krw_arr[KFREE_N_BATCHES];
>  	spinlock_t lock;
>  	struct delayed_work monitor_work;
> @@ -2727,14 +2754,24 @@ struct kfree_rcu_cpu {
>  
>  static DEFINE_PER_CPU(struct kfree_rcu_cpu, krc);
>  
> +static __always_inline void
> +debug_rcu_head_unqueue_bulk(struct rcu_head *head)
> +{
> +#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
> +	for (; head; head = head->next)
> +		debug_rcu_head_unqueue(head);
> +#endif
> +}
> +
>  /*
>   * This function is invoked in workqueue context after a grace period.
> - * It frees all the objects queued on ->head_free.
> + * It frees all the objects queued on ->bhead_free or ->head_free.
>   */
>  static void kfree_rcu_work(struct work_struct *work)
>  {
>  	unsigned long flags;
>  	struct rcu_head *head, *next;
> +	struct kfree_rcu_bulk_data *bhead, *bnext;
>  	struct kfree_rcu_cpu *krcp;
>  	struct kfree_rcu_cpu_work *krwp;
>  
> @@ -2744,22 +2781,41 @@ static void kfree_rcu_work(struct work_struct *work)
>  	spin_lock_irqsave(&krcp->lock, flags);
>  	head = krwp->head_free;
>  	krwp->head_free = NULL;
> +	bhead = krwp->bhead_free;
> +	krwp->bhead_free = NULL;
>  	spin_unlock_irqrestore(&krcp->lock, flags);
>  
> -	// List "head" is now private, so traverse locklessly.
> +	/* "bhead" is now private, so traverse locklessly. */
> +	for (; bhead; bhead = bnext) {
> +		bnext = bhead->next;
> +
> +		debug_rcu_head_unqueue_bulk(bhead->head_free_debug);
> +
> +		rcu_lock_acquire(&rcu_callback_map);
> +		kfree_bulk(bhead->nr_records, bhead->records);
> +		rcu_lock_release(&rcu_callback_map);
> +
> +		if (cmpxchg(&krcp->bcached, NULL, bhead))
> +			free_page((unsigned long) bhead);
> +
> +		cond_resched_tasks_rcu_qs();
> +	}
> +
> +	/*
> +	 * Emergency case only. It can happen under low memory
> +	 * condition when an allocation gets failed, so the "bulk"
> +	 * path can not be temporary maintained.
> +	 */
>  	for (; head; head = next) {
>  		unsigned long offset = (unsigned long)head->func;
>  
>  		next = head->next;
> -		// Potentially optimize with kfree_bulk in future.
>  		debug_rcu_head_unqueue(head);
>  		rcu_lock_acquire(&rcu_callback_map);
>  		trace_rcu_invoke_kfree_callback(rcu_state.name, head, offset);
>  
> -		if (!WARN_ON_ONCE(!__is_kfree_rcu_offset(offset))) {
> -			/* Could be optimized with kfree_bulk() in future. */
> +		if (!WARN_ON_ONCE(!__is_kfree_rcu_offset(offset)))
>  			kfree((void *)head - offset);
> -		}
>  
>  		rcu_lock_release(&rcu_callback_map);
>  		cond_resched_tasks_rcu_qs();
> @@ -2774,26 +2830,48 @@ static void kfree_rcu_work(struct work_struct *work)
>   */
>  static inline bool queue_kfree_rcu_work(struct kfree_rcu_cpu *krcp)
>  {
> +	struct kfree_rcu_cpu_work *krwp;
> +	bool queued = false;
>  	int i;
> -	struct kfree_rcu_cpu_work *krwp = NULL;
>  
>  	lockdep_assert_held(&krcp->lock);
> -	for (i = 0; i < KFREE_N_BATCHES; i++)
> -		if (!krcp->krw_arr[i].head_free) {
> -			krwp = &(krcp->krw_arr[i]);
> -			break;
> -		}
>  
> -	// If a previous RCU batch is in progress, we cannot immediately
> -	// queue another one, so return false to tell caller to retry.
> -	if (!krwp)
> -		return false;
> +	for (i = 0; i < KFREE_N_BATCHES; i++) {
> +		krwp = &(krcp->krw_arr[i]);
>  
> -	krwp->head_free = krcp->head;
> -	krcp->head = NULL;
> -	INIT_RCU_WORK(&krwp->rcu_work, kfree_rcu_work);
> -	queue_rcu_work(system_wq, &krwp->rcu_work);
> -	return true;
> +		/*
> +		 * Try to detach bhead or head and attach it over any
> +		 * available corresponding free channel. It can be that
> +		 * a previous RCU batch is in progress, it means that
> +		 * immediately to queue another one is not possible so
> +		 * return false to tell caller to retry.
> +		 */
> +		if ((krcp->bhead && !krwp->bhead_free) ||
> +				(krcp->head && !krwp->head_free)) {
> +			/* Channel 1. */
> +			if (!krwp->bhead_free) {
> +				krwp->bhead_free = krcp->bhead;
> +				krcp->bhead = NULL;
> +			}
> +
> +			/* Channel 2. */
> +			if (!krwp->head_free) {
> +				krwp->head_free = krcp->head;
> +				krcp->head = NULL;
> +			}
> +
> +			/*
> +			 * One work is per one batch, so there are two "free channels",
> +			 * "bhead_free" and "head_free" the batch can handle. It can be
> +			 * that the work is in the pending state when two channels have
> +			 * been detached following each other, one by one.
> +			 */
> +			queue_rcu_work(system_wq, &krwp->rcu_work);
> +			queued = true;
> +		}
> +	}
> +
> +	return queued;
>  }
>  
>  static inline void kfree_rcu_drain_unlock(struct kfree_rcu_cpu *krcp,
> @@ -2830,19 +2908,65 @@ static void kfree_rcu_monitor(struct work_struct *work)
>  		spin_unlock_irqrestore(&krcp->lock, flags);
>  }
>  
> +static inline bool
> +kfree_call_rcu_add_ptr_to_bulk(struct kfree_rcu_cpu *krcp,
> +	struct rcu_head *head, rcu_callback_t func)
> +{
> +	struct kfree_rcu_bulk_data *bnode;
> +
> +	if (unlikely(!krcp->initialized))
> +		return false;
> +
> +	lockdep_assert_held(&krcp->lock);
> +
> +	/* Check if a new block is required. */
> +	if (!krcp->bhead ||
> +			krcp->bhead->nr_records == KFREE_BULK_MAX_ENTR) {
> +		bnode = xchg(&krcp->bcached, NULL);
> +		if (!bnode) {
> +			WARN_ON_ONCE(sizeof(struct kfree_rcu_bulk_data) > PAGE_SIZE);
> +
> +			bnode = (struct kfree_rcu_bulk_data *)
> +				__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
> +		}
> +
> +		/* Switch to emergency path. */
> +		if (unlikely(!bnode))
> +			return false;
> +
> +		/* Initialize the new block. */
> +		bnode->nr_records = 0;
> +		bnode->next = krcp->bhead;
> +		bnode->head_free_debug = NULL;
> +
> +		/* Attach it to the head. */
> +		krcp->bhead = bnode;
> +	}
> +
> +#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
> +	head->func = func;
> +	head->next = krcp->bhead->head_free_debug;
> +	krcp->bhead->head_free_debug = head;
> +#endif
> +
> +	/* Finally insert. */
> +	krcp->bhead->records[krcp->bhead->nr_records++] =
> +		(void *) head - (unsigned long) func;
> +
> +	return true;
> +}
> +
>  /*
> - * Queue a request for lazy invocation of kfree() after a grace period.
> + * Queue a request for lazy invocation of kfree_bulk()/kfree() after a grace
> + * period. Please note there are two paths are maintained, one is the main one
> + * that uses kfree_bulk() interface and second one is emergency one, that is
> + * used only when the main path can not be maintained temporary, due to memory
> + * pressure.
>   *
>   * Each kfree_call_rcu() request is added to a batch. The batch will be drained
> - * every KFREE_DRAIN_JIFFIES number of jiffies. All the objects in the batch
> - * will be kfree'd in workqueue context. This allows us to:
> - *
> - * 1.	Batch requests together to reduce the number of grace periods during
> - *	heavy kfree_rcu() load.
> - *
> - * 2.	It makes it possible to use kfree_bulk() on a large number of
> - *	kfree_rcu() requests thus reducing cache misses and the per-object
> - *	overhead of kfree().
> + * every KFREE_DRAIN_JIFFIES number of jiffies. All the objects in the batch will
> + * be free'd in workqueue context. This allows us to: batch requests together to
> + * reduce the number of grace periods during heavy kfree_rcu() load.
>   */
>  void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func)
>  {
> @@ -2861,9 +2985,16 @@ void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func)
>  			  __func__, head);
>  		goto unlock_return;
>  	}
> -	head->func = func;
> -	head->next = krcp->head;
> -	krcp->head = head;
> +
> +	/*
> +	 * Under high memory pressure GFP_NOWAIT can fail,
> +	 * in that case the emergency path is maintained.
> +	 */
> +	if (unlikely(!kfree_call_rcu_add_ptr_to_bulk(krcp, head, func))) {
> +		head->func = func;
> +		head->next = krcp->head;
> +		krcp->head = head;
> +	}
>  
>  	// Set timer to drain after KFREE_DRAIN_JIFFIES.
>  	if (rcu_scheduler_active == RCU_SCHEDULER_RUNNING &&
> @@ -3769,8 +3900,11 @@ static void __init kfree_rcu_batch_init(void)
>  		struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
>  
>  		spin_lock_init(&krcp->lock);
> -		for (i = 0; i < KFREE_N_BATCHES; i++)
> +		for (i = 0; i < KFREE_N_BATCHES; i++) {
> +			INIT_RCU_WORK(&krcp->krw_arr[i].rcu_work, kfree_rcu_work);
>  			krcp->krw_arr[i].krcp = krcp;
> +		}
> +
>  		INIT_DELAYED_WORK(&krcp->monitor_work, kfree_rcu_monitor);
>  		krcp->initialized = true;
>  	}
> -- 
> 2.9.5
>