From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from webmail.webked.de (webmail.webked.de [159.69.203.94]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 6915C1953AD; Fri, 4 Apr 2025 08:52:14 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=159.69.203.94 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1743756742; cv=none; b=SGQN+MlZ5vrNSc5C1WSLdC4Q6YMeVx0kEcYFkOOGWs4buRRcKJtiRW6nuXgY8SXs33pHdxUwVQk4DaWs5A9kmHsuR6kQuv9wRbn7eLOyqMEe/hTQoorFyZYzt+/6+rf/aHHoeOYdDuXFUITqmUHi4VtAidhhFAV0B8bHYNzTOrs= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1743756742; c=relaxed/simple; bh=Y/iH1wk4bgH4xibIUWVmQXY/sVGMZAhzE5PSW0nKRek=; h=Message-ID:Subject:From:To:Cc:Date:In-Reply-To:References: Content-Type:MIME-Version; b=tHfKh7OkUHK7n2FZYxXAjVvUAemTCDXpibLFuIYZ5NL1Nn5OKuP/YbH74wLmGkF4TsdS3hGmvcRkK0cVsBfZs3RFadq7zS7yQEy2I2bTiZMRnUz5dTofelqIOm0Jf+AZrEtsIitV8i+y+GCFepJe9l2nEW/geLQ7w6+KgxNoz7g= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=webked.de; spf=pass smtp.mailfrom=webked.de; arc=none smtp.client-ip=159.69.203.94 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=webked.de Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=webked.de Received: from [127.0.0.1] (localhost [127.0.0.1]) by localhost (Mailerdaemon) with ESMTPSA id C106E62BAA; Fri, 4 Apr 2025 10:52:09 +0200 (CEST) Message-ID: Subject: Re: [REGRESSION] Massive virtio-net throughput drop in guest VM with Linux 6.8+ From: Markus Fohrer To: "Michael S. Tsirkin" Cc: virtualization@lists.linux-foundation.org, jasowang@redhat.com, davem@davemloft.net, edumazet@google.com, netdev@vger.kernel.org, linux-kernel@vger.kernel.org Date: Fri, 04 Apr 2025 10:52:09 +0200 In-Reply-To: <20250404042711-mutt-send-email-mst@kernel.org> References: <1d388413ab9cfd765cd2c5e05b5e69cdb2ec5a10.camel@webked.de> <20250403090001-mutt-send-email-mst@kernel.org> <11c5cb52d024a5158c5b8c5e69e2e4639a055a31.camel@webked.de> <20250404042711-mutt-send-email-mst@kernel.org> Organization: WEBKED IT Markus Fohrer Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable User-Agent: Evolution 3.52.3-0ubuntu1 Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-Last-TLS-Session-Version: TLSv1.3 Am Freitag, dem 04.04.2025 um 04:29 -0400 schrieb Michael S. Tsirkin: > On Fri, Apr 04, 2025 at 10:16:55AM +0200, Markus Fohrer wrote: > > Am Donnerstag, dem 03.04.2025 um 09:04 -0400 schrieb Michael S. > > Tsirkin: > > > On Wed, Apr 02, 2025 at 11:12:07PM +0200, Markus Fohrer wrote: > > > > Hi, > > > >=20 > > > > I'm observing a significant performance regression in KVM guest > > > > VMs > > > > using virtio-net with recent Linux kernels (6.8.1+ and 6.14). > > > >=20 > > > > When running on a host system equipped with a Broadcom > > > > NetXtreme-E > > > > (bnxt_en) NIC and AMD EPYC CPUs, the network throughput in the > > > > guest drops to 100=E2=80=93200 KB/s. The same guest configuration > > > > performs > > > > normally (~100 MB/s) when using kernel 6.8.0 or when the VM is > > > > moved to a host with Intel NICs. > > > >=20 > > > > Test environment: > > > > - Host: QEMU/KVM, Linux 6.8.1 and 6.14.0 > > > > - Guest: Linux with virtio-net interface > > > > - NIC: Broadcom BCM57416 (bnxt_en driver, no issues at host > > > > level) > > > > - CPU: AMD EPYC > > > > - Storage: virtio-scsi > > > > - VM network: virtio-net, virtio-scsi (no CPU or IO > > > > bottlenecks) > > > > - Traffic test: iperf3, scp, wget consistently slow in guest > > > >=20 > > > > This issue is not present: > > > > - On 6.8.0=20 > > > > - On hosts with Intel NICs (same VM config) > > > >=20 > > > > I have bisected the issue to the following upstream commit: > > > >=20 > > > > =C2=A0 49d14b54a527 ("virtio-net: Suppress tx timeout warning for > > > > small > > > > tx") > > > > =C2=A0 https://git.kernel.org/linus/49d14b54a527 > > >=20 > > > Thanks a lot for the info! > > >=20 > > >=20 > > > both the link and commit point at: > > >=20 > > > commit 49d14b54a527289d09a9480f214b8c586322310a > > > Author: Eric Dumazet > > > Date:=C2=A0=C2=A0 Thu Sep 26 16:58:36 2024 +0000 > > >=20 > > > =C2=A0=C2=A0=C2=A0 net: test for not too small csum_start in > > > virtio_net_hdr_to_skb() > > > =C2=A0=C2=A0=C2=A0=20 > > >=20 > > > is this what you mean? > > >=20 > > > I don't know which commit is "virtio-net: Suppress tx timeout > > > warning > > > for small tx" > > >=20 > > >=20 > > >=20 > > > > Reverting this commit restores normal network performance in > > > > affected guest VMs. > > > >=20 > > > > I=E2=80=99m happy to provide more data or assist with testing a > > > > potential > > > > fix. > > > >=20 > > > > Thanks, > > > > Markus Fohrer > > >=20 > > >=20 > > > Thanks! First I think it's worth checking what is the setup, e.g. > > > which offloads are enabled. > > > Besides that, I'd start by seeing what's doing on. Assuming I'm > > > right > > > about > > > Eric's patch: > > >=20 > > > diff --git a/include/linux/virtio_net.h > > > b/include/linux/virtio_net.h > > > index 276ca543ef44d8..02a9f4dc594d02 100644 > > > --- a/include/linux/virtio_net.h > > > +++ b/include/linux/virtio_net.h > > > @@ -103,8 +103,10 @@ static inline int > > > virtio_net_hdr_to_skb(struct > > > sk_buff *skb, > > > =C2=A0 > > > =C2=A0 if (!skb_partial_csum_set(skb, start, off)) > > > =C2=A0 return -EINVAL; > > > + if (skb_transport_offset(skb) < nh_min_len) > > > + return -EINVAL; > > > =C2=A0 > > > - nh_min_len =3D max_t(u32, nh_min_len, > > > skb_transport_offset(skb)); > > > + nh_min_len =3D skb_transport_offset(skb); > > > =C2=A0 p_off =3D nh_min_len + thlen; > > > =C2=A0 if (!pskb_may_pull(skb, p_off)) > > > =C2=A0 return -EINVAL; > > >=20 > > >=20 > > > sticking a printk before return -EINVAL to show the offset and > > > nh_min_len > > > would be a good 1st step. Thanks! > > >=20 > >=20 > > I added the following printk inside virtio_net_hdr_to_skb(): > >=20 > > =C2=A0=C2=A0=C2=A0 if (skb_transport_offset(skb) < nh_min_len){ > > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 printk(KERN_INFO "virtio_net= : 3 drop, transport_offset=3D%u, > > nh_min_len=3D%u\n", > > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2= =A0=C2=A0=C2=A0 skb_transport_offset(skb), nh_min_len); > > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 return -EINVAL; > > =C2=A0=C2=A0=C2=A0 } > >=20 > > Built and installed the kernel, then triggered a large download > > via: > >=20 > > =C2=A0=C2=A0=C2=A0 wget http://speedtest.belwue.net/10G > >=20 > > Relevant output from `dmesg -w`: > >=20 > > [=C2=A0=C2=A0 57.327943] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.428942] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.428962] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.553068] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.553088] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.576678] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.618438] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.618453] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.703077] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.823072] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.891982] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 57.946190] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 > > [=C2=A0=C2=A0 58.218686] virtio_net: 3 drop, transport_offset=3D34, > > nh_min_len=3D40=C2=A0=20 >=20 > Hmm indeed. And what about these values? > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0 u32 start =3D __virtio16_to_cpu(little_endian, hdr- > >csum_start); > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0 u32 off =3D __virtio16_to_cpu(little_endian, hdr- > >csum_offset); > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0 u32 needed =3D start + max_t(u32, thlen, off + > sizeof(__sum16)); > print them too? >=20 >=20 >=20 > > I would now do the test with commit > > 49d14b54a527289d09a9480f214b8c586322310a and commit > > 49d14b54a527289d09a9480f214b8c586322310a~1 > >=20 >=20 > Worth checking though it seems likely now the hypervisor is doing > weird > things. what kind of backend is it? qemu? tun? vhost-user? vhost-net? >=20 Backend: QEMU/KVM hypervisor (Proxmox) printk output: [ 58.641906] virtio_net: drop, transport_offset=3D34 start=3D34, off=3D1= 6, needed=3D54, nh_min_len=3D40 [ 58.678048] virtio_net: drop, transport_offset=3D34 start=3D34, off=3D1= 6, needed=3D54, nh_min_len=3D40 [ 58.952871] virtio_net: drop, transport_offset=3D34 start=3D34, off=3D1= 6, needed=3D54, nh_min_len=3D40 [ 58.962157] virtio_net: drop, transport_offset=3D34 start=3D34, off=3D1= 6, needed=3D54, nh_min_len=3D40 [ 59.071645] virtio_net: drop, transport_offset=3D34 start=3D34, off=3D1= 6, needed=3D54, nh_min_len=3D40