Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 16 Mar 2020 11:45:39 -0400
From:   Joel Fernandes <joel@joelfernandes.org>
To:     "Uladzislau Rezki (Sony)" <urezki@gmail.com>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        "Paul E . McKenney" <paulmck@kernel.org>,
        RCU <rcu@vger.kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Steven Rostedt <rostedt@goodmis.org>,
        Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>
Subject: Re: [PATCH v1 2/6] rcu: introduce kvfree_rcu() interface
Message-ID: <20200316154539.GE190951@google.com>
References: <20200315181840.6966-1-urezki@gmail.com>
 <20200315181840.6966-3-urezki@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200315181840.6966-3-urezki@gmail.com>
User-Agent: Mutt/1.12.2 (2019-09-21)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sun, Mar 15, 2020 at 07:18:36PM +0100, Uladzislau Rezki (Sony) wrote:
> kvfree_rcu() can deal with an allocated memory that is obtained
> via kvmalloc(). It can return two types of allocated memory or
> "pointers", one can belong to regular SLAB allocator and another
> one can be vmalloc one. It depends on requested size and memory
> pressure.
> 
> Based on that, two streams are split, thus if a pointer belongs
> to vmalloc allocator it is queued to the list, otherwise SLAB
> one is queued into "bulk array" for further processing.
> 
> The main reason of such splitting is:
>     a) to distinguish kmalloc()/vmalloc() ptrs;
>     b) there is no vmalloc_bulk() interface.
> 
> As of now we have list_lru.c user that needs such interface,
> also there will be new comers. Apart of that it is preparation
> to have a head-less variant later.
> 
> Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
> ---
>  include/linux/rcupdate.h |  9 +++++++++
>  kernel/rcu/tiny.c        |  3 ++-
>  kernel/rcu/tree.c        | 17 ++++++++++++-----
>  3 files changed, 23 insertions(+), 6 deletions(-)
> 
> diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
> index 2be97a83f266..bb270221dbdc 100644
> --- a/include/linux/rcupdate.h
> +++ b/include/linux/rcupdate.h
> @@ -845,6 +845,15 @@ do {									\
>  		__kfree_rcu(&((___p)->rhf), offsetof(typeof(*(ptr)), rhf)); \
>  } while (0)
>  
> +/**
> + * kvfree_rcu() - kvfree an object after a grace period.
> + * @ptr:	pointer to kvfree
> + * @rhf:	the name of the struct rcu_head within the type of @ptr.
> + *
> + * Same as kfree_rcu(), just simple alias.
> + */
> +#define kvfree_rcu(ptr, rhf) kfree_rcu(ptr, rhf)
> +
>  /*
>   * Place this after a lock-acquisition primitive to guarantee that
>   * an UNLOCK+LOCK pair acts as a full barrier.  This guarantee applies
> diff --git a/kernel/rcu/tiny.c b/kernel/rcu/tiny.c
> index dd572ce7c747..4b99f7b88bee 100644
> --- a/kernel/rcu/tiny.c
> +++ b/kernel/rcu/tiny.c
> @@ -23,6 +23,7 @@
>  #include <linux/cpu.h>
>  #include <linux/prefetch.h>
>  #include <linux/slab.h>
> +#include <linux/mm.h>
>  
>  #include "rcu.h"
>  
> @@ -86,7 +87,7 @@ static inline bool rcu_reclaim_tiny(struct rcu_head *head)
>  	rcu_lock_acquire(&rcu_callback_map);
>  	if (__is_kfree_rcu_offset(offset)) {
>  		trace_rcu_invoke_kfree_callback("", head, offset);
> -		kfree((void *)head - offset);
> +		kvfree((void *)head - offset);
>  		rcu_lock_release(&rcu_callback_map);
>  		return true;
>  	}
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index 2f4c91a3713a..1c0a73616872 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -2899,9 +2899,9 @@ static void kfree_rcu_work(struct work_struct *work)
>  	}
>  
>  	/*
> -	 * Emergency case only. It can happen under low memory
> -	 * condition when an allocation gets failed, so the "bulk"
> -	 * path can not be temporary maintained.
> +	 * vmalloc() pointers end up here also emergency case. It can

Suggest rephrase for clarity:

nit: We can end up here either with 1) vmalloc() pointers or 2) low on memory
and could not allocate a bulk array.

thanks,

 - Joel


> +	 * 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;
> @@ -2912,7 +2912,7 @@ static void kfree_rcu_work(struct work_struct *work)
>  		trace_rcu_invoke_kfree_callback(rcu_state.name, head, offset);
>  
>  		if (!WARN_ON_ONCE(!__is_kfree_rcu_offset(offset)))
> -			kfree((void *)head - offset);
> +			kvfree((void *)head - offset);
>  
>  		rcu_lock_release(&rcu_callback_map);
>  		cond_resched_tasks_rcu_qs();
> @@ -3084,10 +3084,17 @@ void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func)
>  	}
>  
>  	/*
> +	 * We do not queue vmalloc pointers into array,
> +	 * instead they are just queued to the list. We
> +	 * do it because of:
> +	 *    a) to distinguish kmalloc()/vmalloc() ptrs;
> +	 *    b) there is no vmalloc_bulk() interface.
> +	 *
>  	 * 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))) {
> +	if (is_vmalloc_addr((void *) head - (unsigned long) func) ||
> +			!kfree_call_rcu_add_ptr_to_bulk(krcp, head, func)) {
>  		head->func = func;
>  		head->next = krcp->head;
>  		krcp->head = head;
> -- 
> 2.20.1
>