Message-ID: <b04b7d5d-582f-1b45-efa3-6e951dfc3cbf@debian.org>
Date:   Tue, 21 Feb 2023 18:38:03 +0000
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:102.0)
 Gecko/20100101 Thunderbird/102.7.2
Cc:     leit@meta.com, linux-kernel@vger.kernel.org, gustavold@meta.com
Subject: Re: [PATCH 1/2] io_uring: Move from hlist to io_wq_work_node
To:     Pavel Begunkov <asml.silence@gmail.com>, axboe@kernel.dk,
        io-uring@vger.kernel.org
References: <20230221135721.3230763-1-leitao@debian.org>
 <782b4b43-790c-6e89-ea74-aac1fd4ff1e2@gmail.com>
From:   Breno Leitao <leitao@debian.org>
In-Reply-To: <782b4b43-790c-6e89-ea74-aac1fd4ff1e2@gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 21/02/2023 17:45, Pavel Begunkov wrote:
> On 2/21/23 13:57, Breno Leitao wrote:
>> Having cache entries linked using the hlist format brings no benefit, and
>> also requires an unnecessary extra pointer address per cache entry.
>>
>> Use the internal io_wq_work_node single-linked list for the internal
>> alloc caches (async_msghdr and async_poll)
>>
>> This is required to be able to use KASAN on cache entries, since we do
>> not need to touch unused (and poisoned) cache entries when adding more
>> entries to the list.
> 
> Looks good, a few nits
> 
>>
>> Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
>> Signed-off-by: Breno Leitao <leitao@debian.org>
>> ---
>>   include/linux/io_uring_types.h |  2 +-
>>   io_uring/alloc_cache.h         | 27 +++++++++++++++------------
>>   2 files changed, 16 insertions(+), 13 deletions(-)
>>
>> diff --git a/include/linux/io_uring_types.h
>> b/include/linux/io_uring_types.h
>> index 0efe4d784358..efa66b6c32c9 100644
>> --- a/include/linux/io_uring_types.h
>> +++ b/include/linux/io_uring_types.h
>> @@ -188,7 +188,7 @@ struct io_ev_fd {
>>   };
>>   
> [...]
>> -    if (!hlist_empty(&cache->list)) {
>> -        struct hlist_node *node = cache->list.first;
>> -
>> -        hlist_del(node);
>> -        return container_of(node, struct io_cache_entry, node);
>> +    struct io_wq_work_node *node;
>> +    struct io_cache_entry *entry;
>> +
>> +    if (cache->list.next) {
>> +        node = cache->list.next;
>> +        entry = container_of(node, struct io_cache_entry, node);
> 
> I'd prefer to get rid of the node var, it'd be a bit cleaner
> than keeping two pointers to the same chunk.
> 
> entry = container_of(node, struct io_cache_entry,
>                      cache->list.next);
> 
>> +        cache->list.next = node->next;
>> +        return entry;
>>       }
>>         return NULL;
>> @@ -35,19 +38,19 @@ static inline struct io_cache_entry
>> *io_alloc_cache_get(struct io_alloc_cache *c
>>     static inline void io_alloc_cache_init(struct io_alloc_cache *cache)
>>   {
>> -    INIT_HLIST_HEAD(&cache->list);
>> +    cache->list.next = NULL;
>>       cache->nr_cached = 0;
>>   }
>>     static inline void io_alloc_cache_free(struct io_alloc_cache *cache,
>>                       void (*free)(struct io_cache_entry *))
>>   {
>> -    while (!hlist_empty(&cache->list)) {
>> -        struct hlist_node *node = cache->list.first;
>> +    struct io_cache_entry *entry;
>>   -        hlist_del(node);
>> -        free(container_of(node, struct io_cache_entry, node));
>> +    while ((entry = io_alloc_cache_get(cache))) {
>> +        free(entry);
> 
> We don't need brackets here.

The extra brackets are required if we are assignments in if, otherwise
the compiler raises a warning (bugprone-assignment-in-if-condition)

> Personally, I don't have anything
> against assignments in if, but it's probably better to avoid them

Sure. I will remove the assignents in "if" part and maybe replicate what
we have
in io_alloc_cache_get(). Something as:
       if (cache->list.next) {
               node = cache->list.next;

Thanks for the review!