Re: [PATCH net-next] tcp: add tcp_tx_skb_cache_key checking in sk_stream_alloc_skb()

[Yunsheng Lin <linyunsheng@huawei.com>]

On 2021/9/2 9:13, Eric Dumazet wrote:
> On Wed, Sep 1, 2021 at 5:47 PM Yunsheng Lin <linyunsheng@huawei.com> wrote:
>>
>> On 2021/9/1 18:39, Yunsheng Lin wrote:
>>> Since tcp_tx_skb_cache is disabled by default in:
>>> commit 0b7d7f6b2208 ("tcp: add tcp_tx_skb_cache sysctl")
>>>
>>> Add tcp_tx_skb_cache_key checking in sk_stream_alloc_skb() to
>>> avoid possible branch-misses.
>>>
>>> Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
>>> ---
>>> Also, the sk->sk_tx_skb_cache may be both changed by allocation
>>> and freeing side, I assume there may be some implicit protection
>>> here too, such as the NAPI protection for rx?
>>
>> Hi, Eric
>>    Is there any implicit protection for sk->sk_tx_skb_cache?
>> As my understanding, sk_stream_alloc_skb() seems to be protected
>> by lock_sock(), and the sk_wmem_free_skb() seems to be mostly
>> happening in NAPI polling for TCP(when ack packet is received)
>> without lock_sock(), so it seems there is no protection here?
>>
> 
> Please look again.
> This is protected by socket lock of course.
> Otherwise sk_mem_uncharge() would be very broken, sk->sk_forward_alloc
> is not an atomic field.

Thanks for clarifying.
I have been looking for a point to implement the socket'pp_alloc_cache for
page pool, and sk_wmem_free_skb() seems like the place to avoid the
scalablity problem of ptr_ring in page pool.

The protection for sk_wmem_free_skb() is in tcp_v4_rcv(), right?
https://elixir.bootlin.com/linux/latest/source/net/ipv4/tcp_ipv4.c#L2081

> 
> TCP stack has no direct relation  with NAPI.
> It can run over loopback interface, no NAPI there.
> .
>