Message-ID: <aedf06ce-15c1-b06a-daa0-507e288e4e29@sberdevices.ru>
Date:   Tue, 21 Mar 2023 13:36:55 +0300
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.7.1
Subject: Re: [RFC PATCH v3] virtio/vsock: allocate multiple skbuffs on tx
Content-Language: en-US
To:     Stefano Garzarella <sgarzare@redhat.com>
CC:     Stefan Hajnoczi <stefanha@redhat.com>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>,
        Bobby Eshleman <bobby.eshleman@bytedance.com>,
        <kvm@vger.kernel.org>, <virtualization@lists.linux-foundation.org>,
        <netdev@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
        <kernel@sberdevices.ru>, <oxffffaa@gmail.com>
References: <f33ef593-982e-2b3f-0986-6d537a3aaf08@sberdevices.ru>
 <20230321084002.5anjcr3ikw3ynbse@sgarzare-redhat>
From:   Arseniy Krasnov <avkrasnov@sberdevices.ru>
In-Reply-To: <20230321084002.5anjcr3ikw3ynbse@sgarzare-redhat>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 21.03.2023 11:40, Stefano Garzarella wrote:
> On Tue, Mar 21, 2023 at 12:31:48AM +0300, Arseniy Krasnov wrote:
>> This adds small optimization for tx path: instead of allocating single
>> skbuff on every call to transport, allocate multiple skbuff's until
>> credit space allows, thus trying to send as much as possible data without
>> return to af_vsock.c.
>>
>> Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
>> ---
>> Link to v1:
>> https://lore.kernel.org/netdev/2c52aa26-8181-d37a-bccd-a86bd3cbc6e1@sberdevices.ru/
>> Link to v2:
>> https://lore.kernel.org/netdev/ea5725eb-6cb5-cf15-2938-34e335a442fa@sberdevices.ru/
>>
>> Changelog:
>> v1 -> v2:
>> - If sent something, return number of bytes sent (even in
>>   case of error). Return error only if failed to sent first
>>   skbuff.
>>
>> v2 -> v3:
>> - Handle case when transport callback returns unexpected value which
>>   is not equal to 'skb->len'. Break loop.
>> - Don't check for zero value of 'rest_len' before calling
>>   'virtio_transport_put_credit()'. Decided to add this check directly
>>   to 'virtio_transport_put_credit()' in separate patch.
>>
>> net/vmw_vsock/virtio_transport_common.c | 59 +++++++++++++++++++------
>> 1 file changed, 45 insertions(+), 14 deletions(-)
>>
>> diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
>> index 6564192e7f20..e0b2c6ecbe22 100644
>> --- a/net/vmw_vsock/virtio_transport_common.c
>> +++ b/net/vmw_vsock/virtio_transport_common.c
>> @@ -196,7 +196,8 @@ static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
>>     const struct virtio_transport *t_ops;
>>     struct virtio_vsock_sock *vvs;
>>     u32 pkt_len = info->pkt_len;
>> -    struct sk_buff *skb;
>> +    u32 rest_len;
>> +    int ret;
>>
>>     info->type = virtio_transport_get_type(sk_vsock(vsk));
>>
>> @@ -216,10 +217,6 @@ static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
>>
>>     vvs = vsk->trans;
>>
>> -    /* we can send less than pkt_len bytes */
>> -    if (pkt_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
>> -        pkt_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
>> -
>>     /* virtio_transport_get_credit might return less than pkt_len credit */
>>     pkt_len = virtio_transport_get_credit(vvs, pkt_len);
>>
>> @@ -227,17 +224,51 @@ static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
>>     if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
>>         return pkt_len;
>>
>> -    skb = virtio_transport_alloc_skb(info, pkt_len,
>> -                     src_cid, src_port,
>> -                     dst_cid, dst_port);
>> -    if (!skb) {
>> -        virtio_transport_put_credit(vvs, pkt_len);
>> -        return -ENOMEM;
>> -    }
>> +    ret = 0;
>> +    rest_len = pkt_len;
>> +
>> +    do {
>> +        struct sk_buff *skb;
>> +        size_t skb_len;
>> +
>> +        skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, rest_len);
>> +
>> +        skb = virtio_transport_alloc_skb(info, skb_len,
>> +                         src_cid, src_port,
>> +                         dst_cid, dst_port);
>> +        if (!skb) {
>> +            ret = -ENOMEM;
>> +            break;
>> +        }
>> +
>> +        virtio_transport_inc_tx_pkt(vvs, skb);
>>
>> -    virtio_transport_inc_tx_pkt(vvs, skb);
>> +        ret = t_ops->send_pkt(skb);
>>
>> -    return t_ops->send_pkt(skb);
>> +        if (ret < 0)
>> +            break;
>> +
>> +        /* Both virtio and vhost 'send_pkt()' returns 'skb_len',
>> +         * but for reliability use 'ret' instead of 'skb_len'.
>> +         * Also if partial send happens (e.g. 'ret' != 'skb_len')
>> +         * somehow, we break this loop, but account such returned
>> +         * value in 'virtio_transport_put_credit()'.
>> +         */
>> +        rest_len -= ret;
>> +
>> +        if (ret != skb_len) {
>> +            ret = -EFAULT;
> 
> Okay, but `ret` will be overwritten by the check we have before the
> return ...
> 
Yes, you're right, this assignment has no effect, since 'rest_len' is already
changed and ret will be changed. I'll fix it.

Thanks, Arseniy
>> +            break;
>> +        }
>> +    } while (rest_len);
>> +
>> +    virtio_transport_put_credit(vvs, rest_len);
>> +
>> +    /* Return number of bytes, if any data has been sent. */
>> +    if (rest_len != pkt_len)
>> +        ret = pkt_len - rest_len;
> 
> ... here.
> 
> Since we don't expect this condition for now, perhaps we can avoid
> setting ret with -EFAULT, but we can add a WARN_ONCE (interrupting the
> loop as you did here).
> 
> This way we return the partial length as we did before.
> 
> Thanks,
> Stefano
> 
>> +
>> +    return ret;
>> }
>>
>> static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
>> -- 
>> 2.25.1
>>
>