Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <58241c427684e6da0ab454d344421c2fb29a0465.camel@redhat.com>
Subject: Re: [PATCH v3 4/5] sock: Consider memcg pressure when raising
 sockmem
From:   Paolo Abeni <pabeni@redhat.com>
To:     Abel Wu <wuyun.abel@bytedance.com>,
        "David S . Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>
Cc:     Simon Horman <simon.horman@corigine.com>, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Date:   Tue, 23 May 2023 12:26:15 +0200
In-Reply-To: <20230523094652.49411-5-wuyun.abel@bytedance.com>
References: <20230523094652.49411-1-wuyun.abel@bytedance.com>
         <20230523094652.49411-5-wuyun.abel@bytedance.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Tue, 2023-05-23 at 17:46 +0800, Abel Wu wrote:
> For now __sk_mem_raise_allocated() mainly considers global socket
> memory pressure and allows to raise if no global pressure observed,
> including the sockets whose memcgs are in pressure, which might
> result in longer memcg memstall.
>=20
> So take net-memcg's pressure into consideration when allocating
> socket memory to alleviate long tail latencies.
>=20
> Signed-off-by: Abel Wu <wuyun.abel@bytedance.com>
> ---
>  net/core/sock.c | 23 ++++++++++++++++-------
>  1 file changed, 16 insertions(+), 7 deletions(-)
>=20
> diff --git a/net/core/sock.c b/net/core/sock.c
> index 801df091e37a..b899e0b9feda 100644
> --- a/net/core/sock.c
> +++ b/net/core/sock.c
> @@ -2976,22 +2976,31 @@ EXPORT_SYMBOL(sk_wait_data);
>  int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kin=
d)
>  {
>  	bool memcg_charge =3D mem_cgroup_sockets_enabled && sk->sk_memcg;
> +	bool charged =3D true, pressured =3D false;
>  	struct proto *prot =3D sk->sk_prot;
> -	bool charged =3D true;
>  	long allocated;
> =20
>  	sk_memory_allocated_add(sk, amt);
>  	allocated =3D sk_memory_allocated(sk);
> -	if (memcg_charge &&
> -	    !(charged =3D mem_cgroup_charge_skmem(sk->sk_memcg, amt,
> -						gfp_memcg_charge())))
> -		goto suppress_allocation;
> +
> +	if (memcg_charge) {
> +		charged =3D mem_cgroup_charge_skmem(sk->sk_memcg, amt,
> +						  gfp_memcg_charge());
> +		if (!charged)
> +			goto suppress_allocation;
> +		if (mem_cgroup_under_socket_pressure(sk->sk_memcg))
> +			pressured =3D true;
> +	}
> =20
>  	/* Under limit. */
> -	if (allocated <=3D sk_prot_mem_limits(sk, 0)) {
> +	if (allocated <=3D sk_prot_mem_limits(sk, 0))
>  		sk_leave_memory_pressure(sk);
> +	else
> +		pressured =3D true;

The above looks not correct to me.=20

	allocated > sk_prot_mem_limits(sk, 0)=20

does not mean the protocol has memory pressure. Such condition is
checked later with:

	if (allocated > sk_prot_mem_limits(sk, 1))

Here an allocation could fail even if memcg charge is successful and
the protocol is not under pressure, which in turn sounds quite (too
much?) conservative.

Cheers,

Paolo