[Qemu-devel] [ RFC Patch v5 0/2] Support Receive-Segment-Offload(RSC) for WHQL

[Qemu-devel] [ RFC Patch v5 0/2] Support Receive-Segment-Offload(RSC) for WHQL

[wexu]

From: Wei Xu <wexu@redhat.com>

Changes in V5:
- Passed all IPv4/6 test cases
- Add new fields in 'virtio_net_hdr'
- Set 'gso_type' & 'coalesced packets' in new field.
- Bypass all 'tcp option' packet
- Bypass all 'pure ack' packet
- Bypass all 'duplicate ack' packet
- Change 'guest_rsc' feature bit to 'false' by default
- Feedbacks from v4, typo, etc.

Note:
There is still a few pending issues about the feature bit, and need to be 
discussed with windows driver maintainer, so linux guests with this patch
won't work at current, haven't figure it out yet, but i'm guessing it's
caused by the 'gso_type' is set to 'VIRTIO_NET_HDR_GSO_TCPV4/6',
will fix it after get the final solution, the below test steps and
performance data is based on v4.

Another suggestion from Jason is to adjust part of the code to make it
more readable, since there maybe still few change about the flowchart
in the future, such as timestamp, duplicate ack, so i'd like to delay it
temporarily.

Changes in V4:
- Add new host feature bit
- Replace using fixed header lenght with dynamic header lenght in VirtIONet 
- Change ip/ip6 header union in NetRscUnit to void* pointer
- Add macro prefix, adjust code indent, etc.

Changes in V3:
- Removed big param list, replace it with 'NetRscUnit' 
- Different virtio header size
- Modify callback function to direct call.
- Needn't check the failure of g_malloc()
- Other code format adjustment, macro naming, etc 

Changes in V2:
- Add detailed commit log

This patch is to support WHQL test for Windows guest, while this feature also
benifits other guest works as a kernel 'gro' like feature with userspace 
implementation.
Feature information:
  http://msdn.microsoft.com/en-us/library/windows/hardware/jj853324

Both IPv4 and IPv6 are supported, though performance with userspace virtio
is slow than vhost-net, there is about 1.5x to 2x performance improvement to
userspace virtio, this is done by turning this feature on and disable
'tso/gso/gro' on corresponding tap interface and guest interface, while get
less improment with all these feature on.

Linux guest performance data(Netperf):
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.2.101 () port 0 AF_INET : nodelay
Size   Size    Size     Time     Throughput  
bytes  bytes   bytes    secs.    10^6bits/sec  

 87380  16384     64    6.00     1221.20   
 87380  16384     64    6.00     1260.30   

 87380  16384    128    6.00     1978.51   
 87380  16384    128    6.00     2286.05   

 87380  16384    256    6.00     2677.94   
 87380  16384    256    6.00     4615.42   

 87380  16384    512    6.00     2956.54   
 87380  16384    512    6.00     5356.39   

 87380  16384   1024    6.00     2798.17   
 87380  16384   1024    6.00     4943.30   

 87380  16384   2048    6.00     2681.09   
 87380  16384   2048    6.00     4835.81   

 87380  16384   4096    6.00     3390.14   
 87380  16384   4096    6.00     5391.54   

 87380  16384   8092    6.00     3008.27   
 87380  16384   8092    6.00     5381.68   

 87380  16384  10240    6.00     2999.89   
 87380  16384  10240    6.00     5393.11 

Test steps:
Although this feature is mainly used for window guest, i used linux guest to 
help test the feature, to make things simple, i used 3 steps to test the patch
as i moved on.

1. With a tcp socket client/server pair running on 2 linux guest, thus i can 
control
the traffic and debugging the code as i want.
2. Netperf on linux guest test the throughput.
3. WHQL test with 2 Windows guests.

Wei Xu (2):
  virtio-net rsc: support coalescing ipv4 tcp traffic
  virtio-net rsc: support coalescing ipv6 tcp traffic

 hw/net/virtio-net.c                         | 623 +++++++++++++++++++++++++++-
 include/hw/virtio/virtio-net.h              |   2 +
 include/hw/virtio/virtio.h                  |  75 ++++
 include/standard-headers/linux/virtio_net.h |   2 +
 4 files changed, 699 insertions(+), 3 deletions(-)

-- 
2.7.1

[Qemu-devel] [ RFC Patch v5 1/2] virtio-net rsc: support coalescing ipv4 tcp traffic

[wexu]

From: Wei Xu <wexu@redhat.com>

All the data packets in a tcp connection will be cached to a big buffer
in every receive interval, and will be sent out via a timer, the
'virtio_net_rsc_timeout' controls the interval, the value will influent the
performance and response of tcp connection extremely, 50000(50us) is a
experience value to gain a performance improvement, since the whql test
sends packets every 100us, so '300000(300us)' can pass the test case,
this is also the default value, it's tunable via the command line
parameter 'rsc_interval' with 'virtio-net-pci' device, for example, below
parameter is to launch a guest with interval set as '500000'.

'virtio-net-pci,netdev=hostnet1,bus=pci.0,id=net1,mac=00,rsc_interval=500000'
will

The timer will only be triggered if the packets pool is not empty,
and it'll drain off all the cached packets.

'NetRscChain' is used to save the segments of different protocols in a
VirtIONet device.

The main handler of TCP includes TCP window update, duplicated ACK check
and the real data coalescing if the new segment passed sanity check
and is identified as an 'wanted' one.

An 'wanted' segment means:
1. Segment is within current window and the sequence is the expected one.
2. 'ACK' of the segment is in the valid window.

Sanity check includes:
1. Incorrect version in IP header
2. IP options & IP fragment
3. Not a TCP packets
4. Sanity size check to prevent buffer overflow attack.

There maybe more cases should be considered such as ip identification other
flags, while it broke the test because windows set it to the same even it's
not a fragment.

Normally it includes 2 typical ways to handle a TCP control flag, 'bypass'
and 'finalize', 'bypass' means should be sent out directly, and 'finalize'
means the packets should also be bypassed, and this should be done
after searching for the same connection packets in the pool and sending
all of them out, this is to avoid out of data.

All the 'SYN' packets will be bypassed since this always begin a new'
connection, other flags such 'FIN/RST' will trigger a finalization, because
this normally happens upon a connection is going to be closed, an 'URG' packet
also finalize current coalescing unit.

Statistics can be used to monitor the basic coalescing status, the 'out of order'
and 'out of window' means how many retransmitting packets, thus describe the
performance intuitively.

Signed-off-by: Wei Xu <wexu@redhat.com>
---
 hw/net/virtio-net.c                         | 487 +++++++++++++++++++++++++++-
 include/hw/virtio/virtio-net.h              |   2 +
 include/hw/virtio/virtio.h                  |  75 +++++
 include/standard-headers/linux/virtio_net.h |   2 +
 4 files changed, 563 insertions(+), 3 deletions(-)

diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index bd91a4b..1ff0135 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -15,10 +15,12 @@
 #include "qemu/iov.h"
 #include "hw/virtio/virtio.h"
 #include "net/net.h"
+#include "net/eth.h"
 #include "net/checksum.h"
 #include "net/tap.h"
 #include "qemu/error-report.h"
 #include "qemu/timer.h"
+#include "qemu/sockets.h"
 #include "hw/virtio/virtio-net.h"
 #include "net/vhost_net.h"
 #include "hw/virtio/virtio-bus.h"
@@ -38,6 +40,25 @@
 #define endof(container, field) \
     (offsetof(container, field) + sizeof(((container *)0)->field))
 
+#define VIRTIO_NET_IP4_ADDR_SIZE   8        /* ipv4 saddr + daddr */
+#define VIRTIO_NET_TCP_PORT_SIZE   4        /* sport + dport */
+
+#define VIRTIO_NET_TCP_FLAG         0x3F
+#define VIRTIO_NET_TCP_HDR_LENGTH   0xF000
+
+/* IPv4 max payload, 16 bits in the header */
+#define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
+#define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
+
+/* header length value in ip header without option */
+#define VIRTIO_NET_IP4_HEADER_LENGTH 5
+
+/* Purge coalesced packets timer interval, This value affects the performance
+   a lot, and should be tuned carefully, '300000'(300us) is the recommended
+   value to pass the WHQL test, '50000' can gain 2x netperf throughput with
+   tso/gso/gro 'off'. */
+#define VIRTIO_NET_RSC_INTERVAL  300000
+
 typedef struct VirtIOFeature {
     uint32_t flags;
     size_t end;
@@ -1688,20 +1709,476 @@ static int virtio_net_load_device(VirtIODevice *vdev, QEMUFile *f,
     return 0;
 }
 
+static void virtio_net_rsc_extract_unit4(NetRscChain *chain,
+                                         const uint8_t *buf, NetRscUnit* unit)
+{
+    uint16_t hdr_len;
+    uint16_t ip_hdrlen;
+    struct ip_header *ip;
+
+    hdr_len = chain->n->guest_hdr_len;
+    ip = (struct ip_header *)(buf + hdr_len + sizeof(struct eth_header));
+    unit->ip = (void *)ip;
+    ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
+    unit->ip_plen = &ip->ip_len;
+    unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
+    unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
+    unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
+}
+
+static void virtio_net_rsc_ipv4_checksum(struct virtio_net_hdr* vhdr,
+                                         struct ip_header *ip)
+{
+    uint32_t sum;
+
+    ip->ip_sum = 0;
+    sum = net_checksum_add_cont(sizeof(struct ip_header), (uint8_t *)ip, 0);
+    ip->ip_sum = cpu_to_be16(net_checksum_finish(sum));
+    vhdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
+    vhdr->flags |= VIRTIO_NET_HDR_F_DATA_VALID;
+}
+
+static size_t virtio_net_rsc_drain_seg(NetRscChain *chain, NetRscSeg *seg)
+{
+    int ret;
+    struct virtio_net_hdr *h;
+
+    h = (struct virtio_net_hdr *)seg->buf; 
+    virtio_net_rsc_ipv4_checksum(h, seg->unit.ip);
+    h->coalesced = seg->packets; 
+    h->dup_ack = seg->dup_ack_count; 
+    h->gso_type = chain->gso_type;
+    h->gso_size = chain->max_payload; 
+    ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
+    QTAILQ_REMOVE(&chain->buffers, seg, next);
+    g_free(seg->buf);
+    g_free(seg);
+
+    return ret;
+}
+
+static void virtio_net_rsc_purge(void *opq)
+{
+    NetRscSeg *seg, *rn;
+    NetRscChain *chain = (NetRscChain *)opq;
+
+    QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
+        if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
+            chain->stat.purge_failed++;
+            continue;
+        }
+    }
+
+    chain->stat.timer++;
+    if (!QTAILQ_EMPTY(&chain->buffers)) {
+        timer_mod(chain->drain_timer,
+              qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + chain->n->rsc_timeout);
+    }
+}
+
+static void virtio_net_rsc_cleanup(VirtIONet *n)
+{
+    NetRscChain *chain, *rn_chain;
+    NetRscSeg *seg, *rn_seg;
+
+    QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
+        QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
+            QTAILQ_REMOVE(&chain->buffers, seg, next);
+            g_free(seg->buf);
+            g_free(seg);
+        }
+
+        timer_del(chain->drain_timer);
+        timer_free(chain->drain_timer);
+        QTAILQ_REMOVE(&n->rsc_chains, chain, next);
+        g_free(chain);
+    }
+}
+
+static void virtio_net_rsc_cache_buf(NetRscChain *chain, NetClientState *nc,
+                                     const uint8_t *buf, size_t size)
+{
+    uint16_t hdr_len;
+    NetRscSeg *seg;
+
+    hdr_len = chain->n->guest_hdr_len;
+    seg = g_malloc(sizeof(NetRscSeg));
+    seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)\
+                   + VIRTIO_NET_MAX_TCP_PAYLOAD);
+    memcpy(seg->buf, buf, size);
+    seg->size = size;
+    seg->packets = 1;
+    seg->dup_ack_count = 0;
+    seg->is_coalesced = 0;
+    seg->nc = nc;
+
+    QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
+    chain->stat.cache++;
+
+    virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
+}
+
+static int32_t virtio_net_rsc_handle_ack(NetRscChain *chain,
+                                         NetRscSeg *seg, const uint8_t *buf,
+                                         struct tcp_header *n_tcp,
+                                         struct tcp_header *o_tcp)
+{
+    uint32_t nack, oack;
+    uint16_t nwin, owin;
+
+    nack = htonl(n_tcp->th_ack);
+    nwin = htons(n_tcp->th_win);
+    oack = htonl(o_tcp->th_ack);
+    owin = htons(o_tcp->th_win);
+
+    if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
+        chain->stat.ack_out_of_win++;
+        return RSC_FINAL;
+    } else if (nack == oack) {
+        /* duplicated ack or window probe */
+        if (nwin == owin) {
+            /* duplicated ack, add dup ack count due to whql test up to 1 */
+            chain->stat.dup_ack++;
+            return RSC_FINAL;
+        } else {
+            /* Coalesce window update */
+            o_tcp->th_win = n_tcp->th_win;
+            chain->stat.win_update++;
+            return RSC_COALESCE;
+        }
+    } else {
+        /* pure ack, go to 'C', finalize*/
+        chain->stat.pure_ack++;
+        return RSC_FINAL;
+    }
+}
+
+static int32_t virtio_net_rsc_coalesce_data(NetRscChain *chain,
+                                            NetRscSeg *seg, const uint8_t *buf,
+                                            NetRscUnit *n_unit)
+{
+    void *data;
+    uint16_t o_ip_len;
+    uint32_t nseq, oseq;
+    NetRscUnit *o_unit;
+
+    o_unit = &seg->unit;
+    o_ip_len = htons(*o_unit->ip_plen);
+    nseq = htonl(n_unit->tcp->th_seq);
+    oseq = htonl(o_unit->tcp->th_seq);
+
+    /* out of order or retransmitted. */
+    if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
+        chain->stat.data_out_of_win++;
+        return RSC_FINAL;
+    }
+
+    data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
+    if (nseq == oseq) {
+        if ((o_unit->payload == 0) && n_unit->payload) {
+            /* From no payload to payload, normal case, not a dup ack or etc */
+            chain->stat.data_after_pure_ack++;
+            goto coalesce;
+        } else {
+            return virtio_net_rsc_handle_ack(chain, seg, buf,
+                                             n_unit->tcp, o_unit->tcp);
+        }
+    } else if ((nseq - oseq) != o_unit->payload) {
+        /* Not a consistent packet, out of order */
+        chain->stat.data_out_of_order++;
+        return RSC_FINAL;
+    } else {
+coalesce:
+        if ((o_ip_len + n_unit->payload) > chain->max_payload) {
+            chain->stat.over_size++;
+            return RSC_FINAL;
+        }
+
+        /* Here comes the right data, the payload length in v4/v6 is different,
+           so use the field value to update and record the new data len */
+        o_unit->payload += n_unit->payload; /* update new data len */
+
+        /* update field in ip header */
+        *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
+
+        /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
+           for windows guest, while this may change the behavior for linux
+           guest. */
+        o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
+
+        o_unit->tcp->th_ack = n_unit->tcp->th_ack;
+        o_unit->tcp->th_win = n_unit->tcp->th_win;
+
+        memmove(seg->buf + seg->size, data, n_unit->payload);
+        seg->size += n_unit->payload;
+        seg->packets++;
+        chain->stat.coalesced++;
+        return RSC_COALESCE;
+    }
+}
+
+static int32_t virtio_net_rsc_coalesce4(NetRscChain *chain, NetRscSeg *seg,
+                                        const uint8_t *buf, size_t size,
+                                        NetRscUnit *unit)
+{
+    struct ip_header *ip1, *ip2;
+
+    ip1 = (struct ip_header *)(unit->ip);
+    ip2 = (struct ip_header *)(seg->unit.ip);
+    if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
+        || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
+        || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
+        chain->stat.no_match++;
+        return RSC_NO_MATCH;
+    }
+
+    return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
+}
+
+/* Pakcets with 'SYN' should bypass, other flag should be sent after drain
+ * to prevent out of order */
+static int virtio_net_rsc_tcp_ctrl_check(NetRscChain *chain,
+                                         struct tcp_header *tcp)
+{
+    uint16_t tcp_hdr;
+    uint16_t tcp_flag;
+
+    tcp_flag = htons(tcp->th_offset_flags);
+    tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
+    tcp_flag &= VIRTIO_NET_TCP_FLAG;
+    tcp_flag = htons(tcp->th_offset_flags) & 0x3F;
+    if (tcp_flag & TH_SYN) {
+        chain->stat.tcp_syn++;
+        return RSC_BYPASS;
+    }
+
+    if (tcp_flag & (TH_FIN | TH_URG | TH_RST)) {
+        chain->stat.tcp_ctrl_drain++;
+        return RSC_FINAL;
+    }
+
+    if (tcp_hdr > sizeof(struct tcp_header)) {
+        chain->stat.tcp_all_opt++; 
+        return RSC_FINAL;
+    }
+
+    return RSC_WANT;
+}
+
+static bool virtio_net_rsc_empty_cache(NetRscChain *chain, NetClientState *nc,
+                                       const uint8_t *buf, size_t size)
+{
+    if (QTAILQ_EMPTY(&chain->buffers)) {
+        chain->stat.empty_cache++;
+        virtio_net_rsc_cache_buf(chain, nc, buf, size);
+        timer_mod(chain->drain_timer,
+              qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + chain->n->rsc_timeout);
+        return 1;
+    }
+
+    return 0;
+}
+
+static size_t virtio_net_rsc_do_coalesce(NetRscChain *chain, NetClientState *nc,
+                                         const uint8_t *buf, size_t size,
+                                         NetRscUnit *unit)
+{
+    int ret;
+    NetRscSeg *seg, *nseg;
+
+    QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
+        ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
+
+        if (ret == RSC_FINAL) {
+            if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
+                /* Send failed */
+                chain->stat.final_failed++;
+                return 0;
+            }
+
+            /* Send current packet */
+            return virtio_net_do_receive(nc, buf, size);
+        } else if (ret == RSC_NO_MATCH) {
+            continue;
+        } else {
+            /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
+            seg->is_coalesced = 1;
+            return size;
+        }
+    }
+
+    chain->stat.no_match_cache++;
+    virtio_net_rsc_cache_buf(chain, nc, buf, size);
+    return size;
+}
+
+/* Drain a connection data, this is to avoid out of order segments */
+static size_t virtio_net_rsc_drain_flow(NetRscChain *chain, NetClientState *nc,
+                                        const uint8_t *buf, size_t size,
+                                        uint16_t ip_start, uint16_t ip_size,
+                                        uint16_t tcp_port, uint16_t port_size)
+{
+    NetRscSeg *seg, *nseg;
+
+    QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
+        if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
+            || memcmp(buf + tcp_port, seg->buf + tcp_port, port_size)) {
+            continue;
+        }
+        if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
+            chain->stat.drain_failed++;
+        }
+
+        break;
+    }
+
+    return virtio_net_do_receive(nc, buf, size);
+}
+
+static int32_t virtio_net_rsc_sanity_check4(NetRscChain *chain,
+                                            struct ip_header *ip,
+                                            const uint8_t *buf, size_t size)
+{
+    uint16_t ip_len;
+    uint16_t hdr_len;
+
+    hdr_len = chain->n->guest_hdr_len;
+    if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
+        + sizeof(struct tcp_header))) {
+        return RSC_BYPASS;
+    }
+
+    /* Not an ipv4 one */
+    if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
+        chain->stat.ip_option++;
+        return RSC_BYPASS;
+    }
+
+    /* Don't handle packets with ip option */
+    if (VIRTIO_NET_IP4_HEADER_LENGTH != (ip->ip_ver_len & 0xF)) {
+        chain->stat.ip_option++;
+        return RSC_BYPASS;
+    }
+
+    if (ip->ip_p != IPPROTO_TCP) {
+        chain->stat.bypass_not_tcp++;
+        return RSC_BYPASS;
+    }
+
+    /* Don't handle packets with ip fragment */
+    if (!(htons(ip->ip_off) & IP_DF)) {
+        chain->stat.ip_frag++;
+        return RSC_BYPASS;
+    }
+
+    ip_len = htons(ip->ip_len);
+    if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
+        || ip_len > (size - hdr_len - sizeof(struct eth_header))) {
+        chain->stat.ip_hacked++;
+        return RSC_BYPASS;
+    }
+
+    return RSC_WANT;
+}
+
+static size_t virtio_net_rsc_receive4(NetRscChain *chain, NetClientState* nc,
+                                      const uint8_t *buf, size_t size)
+{
+    int32_t ret;
+    uint16_t hdr_len;
+    NetRscUnit unit;
+
+    hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
+    virtio_net_rsc_extract_unit4(chain, buf, &unit);
+    if (RSC_WANT != virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)) {
+        return virtio_net_do_receive(nc, buf, size);
+    }
+
+    ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
+    if (ret == RSC_BYPASS) {
+        return virtio_net_do_receive(nc, buf, size);
+    } else if (ret == RSC_FINAL) {
+        return virtio_net_rsc_drain_flow(chain, nc, buf, size,
+                ((hdr_len + sizeof(struct eth_header)) + 12),
+                VIRTIO_NET_IP4_ADDR_SIZE,
+                hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header),
+                VIRTIO_NET_TCP_PORT_SIZE);
+    }
+
+    if (virtio_net_rsc_empty_cache(chain, nc, buf, size)) {
+        return size;
+    }
+
+    return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
+}
+
+static NetRscChain *virtio_net_rsc_lookup_chain(VirtIONet * n,
+                                                NetClientState *nc,
+                                                uint16_t proto)
+{
+    NetRscChain *chain;
+
+    if (proto != (uint16_t)ETH_P_IP) {
+        return NULL;
+    }
+
+    QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
+        if (chain->proto == proto) {
+            return chain;
+        }
+    }
+
+    chain = g_malloc(sizeof(*chain));
+    chain->n = n;
+    chain->proto = proto;
+    chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
+    chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+    chain->drain_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
+                                      virtio_net_rsc_purge, chain);
+    memset(&chain->stat, 0, sizeof(chain->stat));
+
+    QTAILQ_INIT(&chain->buffers);
+    QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
+
+    return chain;
+}
 
 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
                                       const uint8_t *buf, size_t size)
 {
-    return virtio_net_do_receive(nc, buf, size);
+    uint16_t proto;
+    NetRscChain *chain;
+    struct eth_header *eth;
+    VirtIONet *n;
+
+    n = qemu_get_nic_opaque(nc);
+    if (size < (n->guest_hdr_len + sizeof(struct eth_header))) {
+        return virtio_net_do_receive(nc, buf, size);
+    }
+
+    eth = (struct eth_header *)(buf + n->guest_hdr_len);
+    proto = htons(eth->h_proto);
+
+    chain = virtio_net_rsc_lookup_chain(n, nc, proto);
+    if (!chain) {
+        return virtio_net_do_receive(nc, buf, size);
+    } else {
+        chain->stat.received++;
+        return virtio_net_rsc_receive4(chain, nc, buf, size);
+    }
 }
 
 static ssize_t virtio_net_receive(NetClientState *nc,
                                   const uint8_t *buf, size_t size)
 {
     VirtIONet *n;
+    struct virtio_net_hdr *h;
 
     n = qemu_get_nic_opaque(nc);
-    if (n->curr_guest_offloads & VIRTIO_NET_F_GUEST_RSC) {
+    if (n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_RSC)) {
+        h = (struct virtio_net_hdr *)buf;
+        h->coalesced = 0;
+        h->dup_ack = 0;
         return virtio_net_rsc_receive(nc, buf, size);
     } else {
         return virtio_net_do_receive(nc, buf, size);
@@ -1837,6 +2314,7 @@ static void virtio_net_device_realize(DeviceState *dev, Error **errp)
     nc = qemu_get_queue(n->nic);
     nc->rxfilter_notify_enabled = 1;
 
+    QTAILQ_INIT(&n->rsc_chains);
     n->qdev = dev;
     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
                     virtio_net_save, virtio_net_load, n);
@@ -1871,6 +2349,7 @@ static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
     g_free(n->vqs);
     qemu_del_nic(n->nic);
     virtio_cleanup(vdev);
+    virtio_net_rsc_cleanup(n);
 }
 
 static void virtio_net_instance_init(Object *obj)
@@ -1933,7 +2412,9 @@ static Property virtio_net_properties[] = {
     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
     DEFINE_PROP_BIT("guest_rsc", VirtIONet, host_features,
-                    VIRTIO_NET_F_GUEST_RSC, true),
+                    VIRTIO_NET_F_GUEST_RSC, false),
+    DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
+                      VIRTIO_NET_RSC_INTERVAL),
     DEFINE_PROP_END_OF_LIST(),
 };
 
diff --git a/include/hw/virtio/virtio-net.h b/include/hw/virtio/virtio-net.h
index 0cabdb6..e995190 100644
--- a/include/hw/virtio/virtio-net.h
+++ b/include/hw/virtio/virtio-net.h
@@ -59,6 +59,8 @@ typedef struct VirtIONet {
     VirtIONetQueue *vqs;
     VirtQueue *ctrl_vq;
     NICState *nic;
+    QTAILQ_HEAD(, NetRscChain) rsc_chains;
+    uint32_t rsc_timeout;
     uint32_t tx_timeout;
     int32_t tx_burst;
     uint32_t has_vnet_hdr;
diff --git a/include/hw/virtio/virtio.h b/include/hw/virtio/virtio.h
index 2b5b248..02f7d81 100644
--- a/include/hw/virtio/virtio.h
+++ b/include/hw/virtio/virtio.h
@@ -30,6 +30,8 @@
                                 (0x1ULL << VIRTIO_F_ANY_LAYOUT))
 
 struct VirtQueue;
+struct VirtIONet;
+typedef struct VirtIONet VirtIONet;
 
 static inline hwaddr vring_align(hwaddr addr,
                                              unsigned long align)
@@ -128,6 +130,79 @@ typedef struct VirtioDeviceClass {
     int (*load)(VirtIODevice *vdev, QEMUFile *f, int version_id);
 } VirtioDeviceClass;
 
+/* Coalesced packets type & status */
+typedef enum {
+    RSC_COALESCE,           /* Data been coalesced */
+    RSC_FINAL,              /* Will terminate current connection */
+    RSC_NO_MATCH,           /* No matched in the buffer pool */
+    RSC_BYPASS,             /* Packet to be bypass, not tcp, tcp ctrl, etc */
+    RSC_WANT                /* Data want to be coalesced */
+} COALESCE_STATUS;
+
+typedef struct NetRscStat {
+    uint32_t received;
+    uint32_t coalesced;
+    uint32_t over_size;
+    uint32_t cache;
+    uint32_t empty_cache;
+    uint32_t no_match_cache;
+    uint32_t win_update;
+    uint32_t no_match;
+    uint32_t tcp_syn;
+    uint32_t tcp_ctrl_drain;
+    uint32_t dup_ack;
+    uint32_t dup_ack1;
+    uint32_t dup_ack2;
+    uint32_t pure_ack;
+    uint32_t ack_out_of_win;
+    uint32_t data_out_of_win;
+    uint32_t data_out_of_order;
+    uint32_t data_after_pure_ack;
+    uint32_t bypass_not_tcp;
+    uint32_t tcp_option;
+    uint32_t tcp_all_opt;
+    uint32_t ip_frag;
+    uint32_t ip_hacked;
+    uint32_t ip_option;
+    uint32_t purge_failed;
+    uint32_t drain_failed;
+    uint32_t final_failed;
+    int64_t  timer;
+} NetRscStat;
+
+/* Rsc unit general info used to checking if can coalescing */
+typedef struct NetRscUnit {
+    void *ip;   /* ip header */
+    uint16_t *ip_plen;      /* data len pointer in ip header field */
+    struct tcp_header *tcp; /* tcp header */
+    uint16_t tcp_hdrlen;    /* tcp header len */
+    uint16_t payload;       /* pure payload without virtio/eth/ip/tcp */
+} NetRscUnit;
+
+/* Coalesced segmant */
+typedef struct NetRscSeg {
+    QTAILQ_ENTRY(NetRscSeg) next;
+    void *buf;
+    size_t size;
+    uint16_t packets;
+    uint16_t dup_ack_count;
+    bool is_coalesced;      /* need recal ipv4 header checksum, mark here */
+    NetRscUnit unit;
+    NetClientState *nc;
+} NetRscSeg;
+
+/* Chain is divided by protocol(ipv4/v6) and NetClientInfo */
+typedef struct NetRscChain {
+    QTAILQ_ENTRY(NetRscChain) next;
+    VirtIONet *n;                            /* VirtIONet */
+    uint16_t proto;
+    uint8_t  gso_type;
+    uint16_t max_payload;
+    QEMUTimer *drain_timer;
+    QTAILQ_HEAD(, NetRscSeg) buffers;
+    NetRscStat stat;
+} NetRscChain;
+
 void virtio_instance_init_common(Object *proxy_obj, void *data,
                                  size_t vdev_size, const char *vdev_name);
 
diff --git a/include/standard-headers/linux/virtio_net.h b/include/standard-headers/linux/virtio_net.h
index 5b95762..23f28aa 100644
--- a/include/standard-headers/linux/virtio_net.h
+++ b/include/standard-headers/linux/virtio_net.h
@@ -114,6 +114,8 @@ struct virtio_net_hdr {
 	__virtio16 gso_size;		/* Bytes to append to hdr_len per frame */
 	__virtio16 csum_start;	/* Position to start checksumming from */
 	__virtio16 csum_offset;	/* Offset after that to place checksum */
+	__virtio16 coalesced;
+	__virtio16 dup_ack;
 };
 
 /* This is the version of the header to use when the MRG_RXBUF
-- 
2.7.1

[Qemu-devel] [ RFC Patch v5 2/2] virtio-net rsc: support coalescing ipv6 tcp traffic

[wexu]

From: Wei Xu <wexu@redhat.com>

    Most stuffs are like ipv4 2 differences between ipv4 and ipv6.

    1. Fragment length in ipv4 header includes itself, while it's not
    included for ipv6, thus means ipv6 can carry a real '65535' payload.

    2. IPv6 header does not need calculate header checksum.

Signed-off-by: Wei Xu <wexu@redhat.com>
---
 hw/net/virtio-net.c | 152 +++++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 144 insertions(+), 8 deletions(-)

diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index 1ff0135..0586045 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -53,6 +53,10 @@
 /* header length value in ip header without option */
 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
 
+#define ETH_IP6_HDR_SZ (ETH_HDR_SZ + IP6_HDR_SZ)
+#define VIRTIO_NET_IP6_ADDR_SIZE   32      /* ipv6 saddr + daddr */
+#define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
+
 /* Purge coalesced packets timer interval, This value affects the performance
    a lot, and should be tuned carefully, '300000'(300us) is the recommended
    value to pass the WHQL test, '50000' can gain 2x netperf throughput with
@@ -1726,6 +1730,25 @@ static void virtio_net_rsc_extract_unit4(NetRscChain *chain,
     unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
 }
 
+static void virtio_net_rsc_extract_unit6(NetRscChain *chain,
+                                         const uint8_t *buf, NetRscUnit* unit)
+{
+    uint16_t hdr_len;
+    struct ip6_header *ip6;
+
+    hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
+    ip6 = (struct ip6_header *)(buf + hdr_len + sizeof(struct eth_header));
+    unit->ip = ip6;
+    unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
+    unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)\
+                                        + sizeof(struct ip6_header));
+    unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
+
+    /* There is a difference between payload lenght in ipv4 and v6,
+       ip header is excluded in ipv6 */
+    unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
+}
+
 static void virtio_net_rsc_ipv4_checksum(struct virtio_net_hdr* vhdr,
                                          struct ip_header *ip)
 {
@@ -1744,7 +1767,9 @@ static size_t virtio_net_rsc_drain_seg(NetRscChain *chain, NetRscSeg *seg)
     struct virtio_net_hdr *h;
 
     h = (struct virtio_net_hdr *)seg->buf; 
-    virtio_net_rsc_ipv4_checksum(h, seg->unit.ip);
+    if ((chain->proto == ETH_P_IP) && seg->is_coalesced) {
+        virtio_net_rsc_ipv4_checksum(h, seg->unit.ip);
+    }
     h->coalesced = seg->packets; 
     h->dup_ack = seg->dup_ack_count; 
     h->gso_type = chain->gso_type;
@@ -1804,7 +1829,7 @@ static void virtio_net_rsc_cache_buf(NetRscChain *chain, NetClientState *nc,
     hdr_len = chain->n->guest_hdr_len;
     seg = g_malloc(sizeof(NetRscSeg));
     seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)\
-                   + VIRTIO_NET_MAX_TCP_PAYLOAD);
+                   + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
     memcpy(seg->buf, buf, size);
     seg->size = size;
     seg->packets = 1;
@@ -1815,7 +1840,18 @@ static void virtio_net_rsc_cache_buf(NetRscChain *chain, NetClientState *nc,
     QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
     chain->stat.cache++;
 
-    virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
+    switch (chain->proto) {
+    case ETH_P_IP:
+        virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
+        break;
+
+    case ETH_P_IPV6:
+        virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
+        break;
+
+    default:
+        g_assert_not_reached();
+    }
 }
 
 static int32_t virtio_net_rsc_handle_ack(NetRscChain *chain,
@@ -1935,6 +1971,24 @@ static int32_t virtio_net_rsc_coalesce4(NetRscChain *chain, NetRscSeg *seg,
     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
 }
 
+static int32_t virtio_net_rsc_coalesce6(NetRscChain *chain, NetRscSeg *seg,
+                        const uint8_t *buf, size_t size, NetRscUnit *unit)
+{
+    struct ip6_header *ip1, *ip2;
+
+    ip1 = (struct ip6_header *)(unit->ip);
+    ip2 = (struct ip6_header *)(seg->unit.ip);
+    if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
+        || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
+        || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
+        || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
+            chain->stat.no_match++;
+            return RSC_NO_MATCH;
+    }
+
+    return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
+}
+
 /* Pakcets with 'SYN' should bypass, other flag should be sent after drain
  * to prevent out of order */
 static int virtio_net_rsc_tcp_ctrl_check(NetRscChain *chain,
@@ -1987,7 +2041,11 @@ static size_t virtio_net_rsc_do_coalesce(NetRscChain *chain, NetClientState *nc,
     NetRscSeg *seg, *nseg;
 
     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
-        ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
+        if (chain->proto == ETH_P_IP) {
+            ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
+        } else {
+            ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
+        }
 
         if (ret == RSC_FINAL) {
             if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
@@ -2112,13 +2170,82 @@ static size_t virtio_net_rsc_receive4(NetRscChain *chain, NetClientState* nc,
     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
 }
 
+static int32_t virtio_net_rsc_sanity_check6(NetRscChain *chain,
+                                            struct ip6_header *ip6,
+                                            const uint8_t *buf, size_t size)
+{
+    uint16_t ip_len;
+    uint16_t hdr_len;
+
+    hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
+    if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
+        + sizeof(tcp_header))) {
+        return RSC_BYPASS;
+    }
+
+    if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
+        != IP_HEADER_VERSION_6) {
+        return RSC_BYPASS;
+    }
+
+    /* Both option and protocol is checked in this */
+    if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
+        chain->stat.bypass_not_tcp++;
+        return RSC_BYPASS;
+    }
+
+    ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
+    if (ip_len < sizeof(struct tcp_header)
+        || ip_len > (size - hdr_len - sizeof(struct eth_header)
+                     - sizeof(struct ip6_header))) {
+        chain->stat.ip_hacked++;
+        return RSC_BYPASS;
+    }
+
+    return RSC_WANT;
+}
+
+static size_t virtio_net_rsc_receive6(void *opq, NetClientState* nc,
+                                      const uint8_t *buf, size_t size)
+{
+    int32_t ret;
+    uint16_t hdr_len;
+    NetRscChain *chain;
+    NetRscUnit unit;
+
+    chain = (NetRscChain *)opq;
+    hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
+    virtio_net_rsc_extract_unit6(chain, buf, &unit);
+    if (RSC_WANT != virtio_net_rsc_sanity_check6(chain,
+                                                 unit.ip, buf, size)) {
+        return virtio_net_do_receive(nc, buf, size);
+    }
+
+    ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
+    if (ret == RSC_BYPASS) {
+        return virtio_net_do_receive(nc, buf, size);
+    } else if (ret == RSC_FINAL) {
+        return virtio_net_rsc_drain_flow(chain, nc, buf, size,
+                ((hdr_len + sizeof(struct eth_header)) + 8),
+                VIRTIO_NET_IP6_ADDR_SIZE,
+                hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header),
+                VIRTIO_NET_TCP_PORT_SIZE);
+    }
+
+    if (virtio_net_rsc_empty_cache(chain, nc, buf, size)) {
+        return size;
+    }
+
+    return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
+}
+
 static NetRscChain *virtio_net_rsc_lookup_chain(VirtIONet * n,
                                                 NetClientState *nc,
                                                 uint16_t proto)
 {
     NetRscChain *chain;
 
-    if (proto != (uint16_t)ETH_P_IP) {
+    if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
         return NULL;
     }
 
@@ -2131,8 +2258,13 @@ static NetRscChain *virtio_net_rsc_lookup_chain(VirtIONet * n,
     chain = g_malloc(sizeof(*chain));
     chain->n = n;
     chain->proto = proto;
-    chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
-    chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+    if (proto == (uint16_t)ETH_P_IP) {
+        chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
+        chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+    } else {
+        chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
+        chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
+    }
     chain->drain_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
                                       virtio_net_rsc_purge, chain);
     memset(&chain->stat, 0, sizeof(chain->stat));
@@ -2164,7 +2296,11 @@ static ssize_t virtio_net_rsc_receive(NetClientState *nc,
         return virtio_net_do_receive(nc, buf, size);
     } else {
         chain->stat.received++;
-        return virtio_net_rsc_receive4(chain, nc, buf, size);
+        if (proto == (uint16_t)ETH_P_IP) {
+            return virtio_net_rsc_receive4(chain, nc, buf, size);
+        } else  {
+            return virtio_net_rsc_receive6(chain, nc, buf, size);
+        }
     }
 }
 
-- 
2.7.1

Re: [Qemu-devel] [ RFC Patch v5 0/2] Support Receive-Segment-Offload(RSC) for WHQL

[Jason Wang]

On 2016年05月24日 04:14, wexu@redhat.com wrote:
> From: Wei Xu <wexu@redhat.com>
>
> Changes in V5:
> - Passed all IPv4/6 test cases
> - Add new fields in 'virtio_net_hdr'
> - Set 'gso_type' & 'coalesced packets' in new field.
> - Bypass all 'tcp option' packet
> - Bypass all 'pure ack' packet
> - Bypass all 'duplicate ack' packet
> - Change 'guest_rsc' feature bit to 'false' by default
> - Feedbacks from v4, typo, etc.

Patch does not apply on master ...

>
> Note:
> There is still a few pending issues about the feature bit, and need to be
> discussed with windows driver maintainer, so linux guests with this patch
> won't work at current, haven't figure it out yet, but i'm guessing it's
> caused by the 'gso_type' is set to 'VIRTIO_NET_HDR_GSO_TCPV4/6',
> will fix it after get the final solution, the below test steps and
> performance data is based on v4.

Can we split the patches into smaller ones to make review or merging 
easier? E.g can we send the patches without any feature negotiation and 
vnet header extension?

We can focus on the coalescing (maybe ipv4) without any guest 
involvement in this series. In this way, the issues were limited and can 
be converged soon. After this has been merged, we can add patches that 
co-operate with guests on top (since it needs agreement on virtio 
specs). Does this sounds a good plan?

>
> Another suggestion from Jason is to adjust part of the code to make it
> more readable, since there maybe still few change about the flowchart
> in the future, such as timestamp, duplicate ack, so i'd like to delay it
> temporarily.
>
> Changes in V4:
> - Add new host feature bit
> - Replace using fixed header lenght with dynamic header lenght in VirtIONet
> - Change ip/ip6 header union in NetRscUnit to void* pointer
> - Add macro prefix, adjust code indent, etc.
>
> Changes in V3:
> - Removed big param list, replace it with 'NetRscUnit'
> - Different virtio header size
> - Modify callback function to direct call.
> - Needn't check the failure of g_malloc()
> - Other code format adjustment, macro naming, etc
>
> Changes in V2:
> - Add detailed commit log
>
> This patch is to support WHQL test for Windows guest, while this feature also
> benifits other guest works as a kernel 'gro' like feature with userspace
> implementation.
> Feature information:
>    http://msdn.microsoft.com/en-us/library/windows/hardware/jj853324
>
> Both IPv4 and IPv6 are supported, though performance with userspace virtio
> is slow than vhost-net, there is about 1.5x to 2x performance improvement to
> userspace virtio, this is done by turning this feature on and disable
> 'tso/gso/gro' on corresponding tap interface and guest interface, while get
> less improment with all these feature on.
>
> Linux guest performance data(Netperf):
> MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.2.101 () port 0 AF_INET : nodelay
> Size   Size    Size     Time     Throughput
> bytes  bytes   bytes    secs.    10^6bits/sec
>
>   87380  16384     64    6.00     1221.20
>   87380  16384     64    6.00     1260.30
>
>   87380  16384    128    6.00     1978.51
>   87380  16384    128    6.00     2286.05
>
>   87380  16384    256    6.00     2677.94
>   87380  16384    256    6.00     4615.42
>
>   87380  16384    512    6.00     2956.54
>   87380  16384    512    6.00     5356.39
>
>   87380  16384   1024    6.00     2798.17
>   87380  16384   1024    6.00     4943.30
>
>   87380  16384   2048    6.00     2681.09
>   87380  16384   2048    6.00     4835.81
>
>   87380  16384   4096    6.00     3390.14
>   87380  16384   4096    6.00     5391.54
>
>   87380  16384   8092    6.00     3008.27
>   87380  16384   8092    6.00     5381.68
>
>   87380  16384  10240    6.00     2999.89
>   87380  16384  10240    6.00     5393.11
>
> Test steps:
> Although this feature is mainly used for window guest, i used linux guest to
> help test the feature, to make things simple, i used 3 steps to test the patch
> as i moved on.
>
> 1. With a tcp socket client/server pair running on 2 linux guest, thus i can
> control
> the traffic and debugging the code as i want.
> 2. Netperf on linux guest test the throughput.
> 3. WHQL test with 2 Windows guests.
>
> Wei Xu (2):
>    virtio-net rsc: support coalescing ipv4 tcp traffic
>    virtio-net rsc: support coalescing ipv6 tcp traffic
>
>   hw/net/virtio-net.c                         | 623 +++++++++++++++++++++++++++-
>   include/hw/virtio/virtio-net.h              |   2 +
>   include/hw/virtio/virtio.h                  |  75 ++++
>   include/standard-headers/linux/virtio_net.h |   2 +
>   4 files changed, 699 insertions(+), 3 deletions(-)
>

Re: [Qemu-devel] [ RFC Patch v5 0/2] Support Receive-Segment-Offload(RSC) for WHQL

[Michael S. Tsirkin]

On Tue, May 24, 2016 at 04:03:04PM +0800, Jason Wang wrote:
> 
> 
> On 2016年05月24日 04:14, wexu@redhat.com wrote:
> >From: Wei Xu <wexu@redhat.com>
> >
> >Changes in V5:
> >- Passed all IPv4/6 test cases
> >- Add new fields in 'virtio_net_hdr'
> >- Set 'gso_type' & 'coalesced packets' in new field.
> >- Bypass all 'tcp option' packet
> >- Bypass all 'pure ack' packet
> >- Bypass all 'duplicate ack' packet
> >- Change 'guest_rsc' feature bit to 'false' by default
> >- Feedbacks from v4, typo, etc.
> 
> Patch does not apply on master ...
> 
> >
> >Note:
> >There is still a few pending issues about the feature bit, and need to be
> >discussed with windows driver maintainer, so linux guests with this patch
> >won't work at current, haven't figure it out yet, but i'm guessing it's
> >caused by the 'gso_type' is set to 'VIRTIO_NET_HDR_GSO_TCPV4/6',
> >will fix it after get the final solution, the below test steps and
> >performance data is based on v4.
> 
> Can we split the patches into smaller ones to make review or merging easier?
> E.g can we send the patches without any feature negotiation and vnet header
> extension?
> 
> We can focus on the coalescing (maybe ipv4) without any guest involvement in
> this series. In this way, the issues were limited and can be converged soon.
> After this has been merged, we can add patches that co-operate with guests
> on top (since it needs agreement on virtio specs). Does this sounds a good
> plan?

True but disabling everything when feature is not negotiated
reduces the risk somewhat.

> >
> >Another suggestion from Jason is to adjust part of the code to make it
> >more readable, since there maybe still few change about the flowchart
> >in the future, such as timestamp, duplicate ack, so i'd like to delay it
> >temporarily.
> >
> >Changes in V4:
> >- Add new host feature bit
> >- Replace using fixed header lenght with dynamic header lenght in VirtIONet
> >- Change ip/ip6 header union in NetRscUnit to void* pointer
> >- Add macro prefix, adjust code indent, etc.
> >
> >Changes in V3:
> >- Removed big param list, replace it with 'NetRscUnit'
> >- Different virtio header size
> >- Modify callback function to direct call.
> >- Needn't check the failure of g_malloc()
> >- Other code format adjustment, macro naming, etc
> >
> >Changes in V2:
> >- Add detailed commit log
> >
> >This patch is to support WHQL test for Windows guest, while this feature also
> >benifits other guest works as a kernel 'gro' like feature with userspace
> >implementation.
> >Feature information:
> >   http://msdn.microsoft.com/en-us/library/windows/hardware/jj853324
> >
> >Both IPv4 and IPv6 are supported, though performance with userspace virtio
> >is slow than vhost-net, there is about 1.5x to 2x performance improvement to
> >userspace virtio, this is done by turning this feature on and disable
> >'tso/gso/gro' on corresponding tap interface and guest interface, while get
> >less improment with all these feature on.
> >
> >Linux guest performance data(Netperf):
> >MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.2.101 () port 0 AF_INET : nodelay
> >Size   Size    Size     Time     Throughput
> >bytes  bytes   bytes    secs.    10^6bits/sec
> >
> >  87380  16384     64    6.00     1221.20
> >  87380  16384     64    6.00     1260.30
> >
> >  87380  16384    128    6.00     1978.51
> >  87380  16384    128    6.00     2286.05
> >
> >  87380  16384    256    6.00     2677.94
> >  87380  16384    256    6.00     4615.42
> >
> >  87380  16384    512    6.00     2956.54
> >  87380  16384    512    6.00     5356.39
> >
> >  87380  16384   1024    6.00     2798.17
> >  87380  16384   1024    6.00     4943.30
> >
> >  87380  16384   2048    6.00     2681.09
> >  87380  16384   2048    6.00     4835.81
> >
> >  87380  16384   4096    6.00     3390.14
> >  87380  16384   4096    6.00     5391.54
> >
> >  87380  16384   8092    6.00     3008.27
> >  87380  16384   8092    6.00     5381.68
> >
> >  87380  16384  10240    6.00     2999.89
> >  87380  16384  10240    6.00     5393.11
> >
> >Test steps:
> >Although this feature is mainly used for window guest, i used linux guest to
> >help test the feature, to make things simple, i used 3 steps to test the patch
> >as i moved on.
> >
> >1. With a tcp socket client/server pair running on 2 linux guest, thus i can
> >control
> >the traffic and debugging the code as i want.
> >2. Netperf on linux guest test the throughput.
> >3. WHQL test with 2 Windows guests.
> >
> >Wei Xu (2):
> >   virtio-net rsc: support coalescing ipv4 tcp traffic
> >   virtio-net rsc: support coalescing ipv6 tcp traffic
> >
> >  hw/net/virtio-net.c                         | 623 +++++++++++++++++++++++++++-
> >  include/hw/virtio/virtio-net.h              |   2 +
> >  include/hw/virtio/virtio.h                  |  75 ++++
> >  include/standard-headers/linux/virtio_net.h |   2 +
> >  4 files changed, 699 insertions(+), 3 deletions(-)
> >

Re: [Qemu-devel] [ RFC Patch v5 1/2] virtio-net rsc: support coalescing ipv4 tcp traffic/

[Michael S. Tsirkin]

On Tue, May 24, 2016 at 04:14:44AM +0800, wexu@redhat.com wrote:
> diff --git a/include/standard-headers/linux/virtio_net.h b/include/standard-headers/linux/virtio_net.h
> index 5b95762..23f28aa 100644
> --- a/include/standard-headers/linux/virtio_net.h
> +++ b/include/standard-headers/linux/virtio_net.h
> @@ -114,6 +114,8 @@ struct virtio_net_hdr {
>  	__virtio16 gso_size;		/* Bytes to append to hdr_len per frame */
>  	__virtio16 csum_start;	/* Position to start checksumming from */
>  	__virtio16 csum_offset;	/* Offset after that to place checksum */
> +	__virtio16 coalesced;
> +	__virtio16 dup_ack;
>  };
>  
>  /* This is the version of the header to use when the MRG_RXBUF

OK so please post a proposal documenting feature bit
and new fields to virtio TC mailing list.
No need to wait to get it upstream.

> -- 
> 2.7.1

Re: [Qemu-devel] [ RFC Patch v5 0/2] Support Receive-Segment-Offload(RSC) for WHQL

[Wei Xu]

On 2016年05月24日 16:03, Jason Wang wrote:
>
>
> On 2016年05月24日 04:14, wexu@redhat.com wrote:
>> From: Wei Xu <wexu@redhat.com>
>>
>> Changes in V5:
>> - Passed all IPv4/6 test cases
>> - Add new fields in 'virtio_net_hdr'
>> - Set 'gso_type' & 'coalesced packets' in new field.
>> - Bypass all 'tcp option' packet
>> - Bypass all 'pure ack' packet
>> - Bypass all 'duplicate ack' packet
>> - Change 'guest_rsc' feature bit to 'false' by default
>> - Feedbacks from v4, typo, etc.
>
> Patch does not apply on master ...
>
>>
>> Note:
>> There is still a few pending issues about the feature bit, and need to be
>> discussed with windows driver maintainer, so linux guests with this patch
>> won't work at current, haven't figure it out yet, but i'm guessing it's
>> caused by the 'gso_type' is set to 'VIRTIO_NET_HDR_GSO_TCPV4/6',
>> will fix it after get the final solution, the below test steps and
>> performance data is based on v4.
>
> Can we split the patches into smaller ones to make review or merging
> easier? E.g can we send the patches without any feature negotiation and
> vnet header extension?
OK.
>
> We can focus on the coalescing (maybe ipv4) without any guest
> involvement in this series. In this way, the issues were limited and can
> be converged soon. After this has been merged, we can add patches that
> co-operate with guests on top (since it needs agreement on virtio
> specs). Does this sounds a good plan?
Exactly, both ipv4/6 passed whql test actually, so maybe leave the 
feature bit and header issue alone.
>
>>
>> Another suggestion from Jason is to adjust part of the code to make it
>> more readable, since there maybe still few change about the flowchart
>> in the future, such as timestamp, duplicate ack, so i'd like to delay it
>> temporarily.
>>
>> Changes in V4:
>> - Add new host feature bit
>> - Replace using fixed header lenght with dynamic header lenght in
>> VirtIONet
>> - Change ip/ip6 header union in NetRscUnit to void* pointer
>> - Add macro prefix, adjust code indent, etc.
>>
>> Changes in V3:
>> - Removed big param list, replace it with 'NetRscUnit'
>> - Different virtio header size
>> - Modify callback function to direct call.
>> - Needn't check the failure of g_malloc()
>> - Other code format adjustment, macro naming, etc
>>
>> Changes in V2:
>> - Add detailed commit log
>>
>> This patch is to support WHQL test for Windows guest, while this
>> feature also
>> benifits other guest works as a kernel 'gro' like feature with userspace
>> implementation.
>> Feature information:
>>    http://msdn.microsoft.com/en-us/library/windows/hardware/jj853324
>>
>> Both IPv4 and IPv6 are supported, though performance with userspace
>> virtio
>> is slow than vhost-net, there is about 1.5x to 2x performance
>> improvement to
>> userspace virtio, this is done by turning this feature on and disable
>> 'tso/gso/gro' on corresponding tap interface and guest interface,
>> while get
>> less improment with all these feature on.
>>
>> Linux guest performance data(Netperf):
>> MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
>> 192.168.2.101 () port 0 AF_INET : nodelay
>> Size   Size    Size     Time     Throughput
>> bytes  bytes   bytes    secs.    10^6bits/sec
>>
>>   87380  16384     64    6.00     1221.20
>>   87380  16384     64    6.00     1260.30
>>
>>   87380  16384    128    6.00     1978.51
>>   87380  16384    128    6.00     2286.05
>>
>>   87380  16384    256    6.00     2677.94
>>   87380  16384    256    6.00     4615.42
>>
>>   87380  16384    512    6.00     2956.54
>>   87380  16384    512    6.00     5356.39
>>
>>   87380  16384   1024    6.00     2798.17
>>   87380  16384   1024    6.00     4943.30
>>
>>   87380  16384   2048    6.00     2681.09
>>   87380  16384   2048    6.00     4835.81
>>
>>   87380  16384   4096    6.00     3390.14
>>   87380  16384   4096    6.00     5391.54
>>
>>   87380  16384   8092    6.00     3008.27
>>   87380  16384   8092    6.00     5381.68
>>
>>   87380  16384  10240    6.00     2999.89
>>   87380  16384  10240    6.00     5393.11
>>
>> Test steps:
>> Although this feature is mainly used for window guest, i used linux
>> guest to
>> help test the feature, to make things simple, i used 3 steps to test
>> the patch
>> as i moved on.
>>
>> 1. With a tcp socket client/server pair running on 2 linux guest, thus
>> i can
>> control
>> the traffic and debugging the code as i want.
>> 2. Netperf on linux guest test the throughput.
>> 3. WHQL test with 2 Windows guests.
>>
>> Wei Xu (2):
>>    virtio-net rsc: support coalescing ipv4 tcp traffic
>>    virtio-net rsc: support coalescing ipv6 tcp traffic
>>
>>   hw/net/virtio-net.c                         | 623
>> +++++++++++++++++++++++++++-
>>   include/hw/virtio/virtio-net.h              |   2 +
>>   include/hw/virtio/virtio.h                  |  75 ++++
>>   include/standard-headers/linux/virtio_net.h |   2 +
>>   4 files changed, 699 insertions(+), 3 deletions(-)
>>
>
>

Re: [Qemu-devel] [ RFC Patch v5 0/2] Support Receive-Segment-Offload(RSC) for WHQL

[Wei Xu]

On 2016年05月24日 16:26, Michael S. Tsirkin wrote:
> On Tue, May 24, 2016 at 04:03:04PM +0800, Jason Wang wrote:
>>
>>
>> On 2016年05月24日 04:14, wexu@redhat.com wrote:
>>> From: Wei Xu <wexu@redhat.com>
>>>
>>> Changes in V5:
>>> - Passed all IPv4/6 test cases
>>> - Add new fields in 'virtio_net_hdr'
>>> - Set 'gso_type' & 'coalesced packets' in new field.
>>> - Bypass all 'tcp option' packet
>>> - Bypass all 'pure ack' packet
>>> - Bypass all 'duplicate ack' packet
>>> - Change 'guest_rsc' feature bit to 'false' by default
>>> - Feedbacks from v4, typo, etc.
>>
>> Patch does not apply on master ...
>>
>>>
>>> Note:
>>> There is still a few pending issues about the feature bit, and need to be
>>> discussed with windows driver maintainer, so linux guests with this patch
>>> won't work at current, haven't figure it out yet, but i'm guessing it's
>>> caused by the 'gso_type' is set to 'VIRTIO_NET_HDR_GSO_TCPV4/6',
>>> will fix it after get the final solution, the below test steps and
>>> performance data is based on v4.
>>
>> Can we split the patches into smaller ones to make review or merging easier?
>> E.g can we send the patches without any feature negotiation and vnet header
>> extension?
>>
>> We can focus on the coalescing (maybe ipv4) without any guest involvement in
>> this series. In this way, the issues were limited and can be converged soon.
>> After this has been merged, we can add patches that co-operate with guests
>> on top (since it needs agreement on virtio specs). Does this sounds a good
>> plan?
>
> True but disabling everything when feature is not negotiated
> reduces the risk somewhat.
Sure.
>
>>>
>>> Another suggestion from Jason is to adjust part of the code to make it
>>> more readable, since there maybe still few change about the flowchart
>>> in the future, such as timestamp, duplicate ack, so i'd like to delay it
>>> temporarily.
>>>
>>> Changes in V4:
>>> - Add new host feature bit
>>> - Replace using fixed header lenght with dynamic header lenght in VirtIONet
>>> - Change ip/ip6 header union in NetRscUnit to void* pointer
>>> - Add macro prefix, adjust code indent, etc.
>>>
>>> Changes in V3:
>>> - Removed big param list, replace it with 'NetRscUnit'
>>> - Different virtio header size
>>> - Modify callback function to direct call.
>>> - Needn't check the failure of g_malloc()
>>> - Other code format adjustment, macro naming, etc
>>>
>>> Changes in V2:
>>> - Add detailed commit log
>>>
>>> This patch is to support WHQL test for Windows guest, while this feature also
>>> benifits other guest works as a kernel 'gro' like feature with userspace
>>> implementation.
>>> Feature information:
>>>    http://msdn.microsoft.com/en-us/library/windows/hardware/jj853324
>>>
>>> Both IPv4 and IPv6 are supported, though performance with userspace virtio
>>> is slow than vhost-net, there is about 1.5x to 2x performance improvement to
>>> userspace virtio, this is done by turning this feature on and disable
>>> 'tso/gso/gro' on corresponding tap interface and guest interface, while get
>>> less improment with all these feature on.
>>>
>>> Linux guest performance data(Netperf):
>>> MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.2.101 () port 0 AF_INET : nodelay
>>> Size   Size    Size     Time     Throughput
>>> bytes  bytes   bytes    secs.    10^6bits/sec
>>>
>>>   87380  16384     64    6.00     1221.20
>>>   87380  16384     64    6.00     1260.30
>>>
>>>   87380  16384    128    6.00     1978.51
>>>   87380  16384    128    6.00     2286.05
>>>
>>>   87380  16384    256    6.00     2677.94
>>>   87380  16384    256    6.00     4615.42
>>>
>>>   87380  16384    512    6.00     2956.54
>>>   87380  16384    512    6.00     5356.39
>>>
>>>   87380  16384   1024    6.00     2798.17
>>>   87380  16384   1024    6.00     4943.30
>>>
>>>   87380  16384   2048    6.00     2681.09
>>>   87380  16384   2048    6.00     4835.81
>>>
>>>   87380  16384   4096    6.00     3390.14
>>>   87380  16384   4096    6.00     5391.54
>>>
>>>   87380  16384   8092    6.00     3008.27
>>>   87380  16384   8092    6.00     5381.68
>>>
>>>   87380  16384  10240    6.00     2999.89
>>>   87380  16384  10240    6.00     5393.11
>>>
>>> Test steps:
>>> Although this feature is mainly used for window guest, i used linux guest to
>>> help test the feature, to make things simple, i used 3 steps to test the patch
>>> as i moved on.
>>>
>>> 1. With a tcp socket client/server pair running on 2 linux guest, thus i can
>>> control
>>> the traffic and debugging the code as i want.
>>> 2. Netperf on linux guest test the throughput.
>>> 3. WHQL test with 2 Windows guests.
>>>
>>> Wei Xu (2):
>>>    virtio-net rsc: support coalescing ipv4 tcp traffic
>>>    virtio-net rsc: support coalescing ipv6 tcp traffic
>>>
>>>   hw/net/virtio-net.c                         | 623 +++++++++++++++++++++++++++-
>>>   include/hw/virtio/virtio-net.h              |   2 +
>>>   include/hw/virtio/virtio.h                  |  75 ++++
>>>   include/standard-headers/linux/virtio_net.h |   2 +
>>>   4 files changed, 699 insertions(+), 3 deletions(-)
>>>

Re: [Qemu-devel] [ RFC Patch v5 0/2] Support Receive-Segment-Offload(RSC) for WHQL

[Jason Wang]

On 2016年05月24日 16:26, Michael S. Tsirkin wrote:
> On Tue, May 24, 2016 at 04:03:04PM +0800, Jason Wang wrote:
>> >
>> >
>> >On 2016年05月24日 04:14,wexu@redhat.com  wrote:
>>> > >From: Wei Xu<wexu@redhat.com>
>>> > >
>>> > >Changes in V5:
>>> > >- Passed all IPv4/6 test cases
>>> > >- Add new fields in 'virtio_net_hdr'
>>> > >- Set 'gso_type' & 'coalesced packets' in new field.
>>> > >- Bypass all 'tcp option' packet
>>> > >- Bypass all 'pure ack' packet
>>> > >- Bypass all 'duplicate ack' packet
>>> > >- Change 'guest_rsc' feature bit to 'false' by default
>>> > >- Feedbacks from v4, typo, etc.
>> >
>> >Patch does not apply on master ...
>> >
>>> > >
>>> > >Note:
>>> > >There is still a few pending issues about the feature bit, and need to be
>>> > >discussed with windows driver maintainer, so linux guests with this patch
>>> > >won't work at current, haven't figure it out yet, but i'm guessing it's
>>> > >caused by the 'gso_type' is set to 'VIRTIO_NET_HDR_GSO_TCPV4/6',
>>> > >will fix it after get the final solution, the below test steps and
>>> > >performance data is based on v4.
>> >
>> >Can we split the patches into smaller ones to make review or merging easier?
>> >E.g can we send the patches without any feature negotiation and vnet header
>> >extension?
>> >
>> >We can focus on the coalescing (maybe ipv4) without any guest involvement in
>> >this series. In this way, the issues were limited and can be converged soon.
>> >After this has been merged, we can add patches that co-operate with guests
>> >on top (since it needs agreement on virtio specs). Does this sounds a good
>> >plan?
> True but disabling everything when feature is not negotiated
> reduces the risk somewhat.
>

Yes, but I believe we can only merge the patch with new virtio features 
after they were accepted by spec?

Re: [Qemu-devel] [ RFC Patch v5 0/2] Support Receive-Segment-Offload(RSC) for WHQL

[Michael S. Tsirkin]

On Wed, May 25, 2016 at 04:08:50PM +0800, Jason Wang wrote:
> 
> 
> On 2016年05月24日 16:26, Michael S. Tsirkin wrote:
> >On Tue, May 24, 2016 at 04:03:04PM +0800, Jason Wang wrote:
> >>>
> >>>
> >>>On 2016年05月24日 04:14,wexu@redhat.com  wrote:
> >>>> >From: Wei Xu<wexu@redhat.com>
> >>>> >
> >>>> >Changes in V5:
> >>>> >- Passed all IPv4/6 test cases
> >>>> >- Add new fields in 'virtio_net_hdr'
> >>>> >- Set 'gso_type' & 'coalesced packets' in new field.
> >>>> >- Bypass all 'tcp option' packet
> >>>> >- Bypass all 'pure ack' packet
> >>>> >- Bypass all 'duplicate ack' packet
> >>>> >- Change 'guest_rsc' feature bit to 'false' by default
> >>>> >- Feedbacks from v4, typo, etc.
> >>>
> >>>Patch does not apply on master ...
> >>>
> >>>> >
> >>>> >Note:
> >>>> >There is still a few pending issues about the feature bit, and need to be
> >>>> >discussed with windows driver maintainer, so linux guests with this patch
> >>>> >won't work at current, haven't figure it out yet, but i'm guessing it's
> >>>> >caused by the 'gso_type' is set to 'VIRTIO_NET_HDR_GSO_TCPV4/6',
> >>>> >will fix it after get the final solution, the below test steps and
> >>>> >performance data is based on v4.
> >>>
> >>>Can we split the patches into smaller ones to make review or merging easier?
> >>>E.g can we send the patches without any feature negotiation and vnet header
> >>>extension?
> >>>
> >>>We can focus on the coalescing (maybe ipv4) without any guest involvement in
> >>>this series. In this way, the issues were limited and can be converged soon.
> >>>After this has been merged, we can add patches that co-operate with guests
> >>>on top (since it needs agreement on virtio specs). Does this sounds a good
> >>>plan?
> >True but disabling everything when feature is not negotiated
> >reduces the risk somewhat.
> >
> 
> Yes, but I believe we can only merge the patch with new virtio features
> after they were accepted by spec?

Spec acceptance takes a lot of time. I think it's required to send
proposal to virtio tc and to give a week or two for people
to comment, but if there are no comments I don't think
we must hold up code.

I am interested in hearing whether new fields should be added to
all packets, or rather to packets with a specific bit set in
flags or gso_type field.

-- 
MST