kvm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [RFC PATCH v3] virtio/vsock: allocate multiple skbuffs on tx
@ 2023-03-20 21:31 Arseniy Krasnov
  2023-03-21  8:40 ` Stefano Garzarella
  0 siblings, 1 reply; 3+ messages in thread
From: Arseniy Krasnov @ 2023-03-20 21:31 UTC (permalink / raw)
  To: Stefan Hajnoczi, Stefano Garzarella, David S. Miller,
	Eric Dumazet, Jakub Kicinski, Paolo Abeni, Bobby Eshleman
  Cc: kvm, virtualization, netdev, linux-kernel, kernel, oxffffaa, avkrasnov

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;
+			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;
+
+	return ret;
 }
 
 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
-- 
2.25.1

^ permalink raw reply related	[flat|nested] 3+ messages in thread
* Re: [RFC PATCH v3] virtio/vsock: allocate multiple skbuffs on tx
  2023-03-20 21:31 [RFC PATCH v3] virtio/vsock: allocate multiple skbuffs on tx Arseniy Krasnov
@ 2023-03-21  8:40 ` Stefano Garzarella
  2023-03-21 10:36   ` Arseniy Krasnov
  0 siblings, 1 reply; 3+ messages in thread
From: Stefano Garzarella @ 2023-03-21  8:40 UTC (permalink / raw)
  To: Arseniy Krasnov
  Cc: Stefan Hajnoczi, David S. Miller, Eric Dumazet, Jakub Kicinski,
	Paolo Abeni, Bobby Eshleman, kvm, virtualization, netdev,
	linux-kernel, kernel, oxffffaa

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 ...

>+			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
>


^ permalink raw reply	[flat|nested] 3+ messages in thread
* Re: [RFC PATCH v3] virtio/vsock: allocate multiple skbuffs on tx
  2023-03-21  8:40 ` Stefano Garzarella
@ 2023-03-21 10:36   ` Arseniy Krasnov
  0 siblings, 0 replies; 3+ messages in thread
From: Arseniy Krasnov @ 2023-03-21 10:36 UTC (permalink / raw)
  To: Stefano Garzarella
  Cc: Stefan Hajnoczi, David S. Miller, Eric Dumazet, Jakub Kicinski,
	Paolo Abeni, Bobby Eshleman, kvm, virtualization, netdev,
	linux-kernel, kernel, oxffffaa



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
>>
> 

^ permalink raw reply	[flat|nested] 3+ messages in thread
end of thread, other threads:[~2023-03-21 10:40 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-03-20 21:31 [RFC PATCH v3] virtio/vsock: allocate multiple skbuffs on tx Arseniy Krasnov
2023-03-21  8:40 ` Stefano Garzarella
2023-03-21 10:36   ` Arseniy Krasnov

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).