[PATCH v8 0/7] reference implementation of RSS and hash report

[PATCH v8 0/7] reference implementation of RSS and hash report

[Yuri Benditovich]

Support for VIRTIO_NET_F_RSS and VIRTIO_NET_F_HASH_REPORT
features in QEMU for reference purpose.
Implements Toeplitz hash calculation for incoming
packets according to configuration provided by driver.
Uses calculated hash for decision on receive virtqueue
and/or reports the hash in the virtio header

Changes from v7:
Patch 7.1: removed (RSS and hash report definitions)
Patch 7.2: delete configuration struct with RSS definitions
Patch 7.4: delete duplicated packet structure
Added patch 7 - adapt RSC definitions to updated header

Yuri Benditovich (7):
  virtio-net: implement RSS configuration command
  virtio-net: implement RX RSS processing
  tap: allow extended virtio header with hash info
  virtio-net: reference implementation of hash report
  vmstate.h: provide VMSTATE_VARRAY_UINT16_ALLOC macro
  virtio-net: add migration support for RSS and hash report
  virtio-net: align RSC fields with updated virtio-net header

 hw/net/trace-events            |   3 +
 hw/net/virtio-net.c            | 387 +++++++++++++++++++++++++++++----
 include/hw/virtio/virtio-net.h |  16 ++
 include/migration/vmstate.h    |  10 +
 net/tap.c                      |   3 +-
 5 files changed, 379 insertions(+), 40 deletions(-)

-- 
2.17.1

[PATCH v8 1/7] virtio-net: implement RSS configuration command

[Yuri Benditovich]

Optionally report RSS feature.
Handle RSS configuration command and keep RSS parameters
in virtio-net device context.

Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com>
---
 hw/net/trace-events            |   3 +
 hw/net/virtio-net.c            | 167 +++++++++++++++++++++++++++++++--
 include/hw/virtio/virtio-net.h |  13 +++
 3 files changed, 174 insertions(+), 9 deletions(-)

diff --git a/hw/net/trace-events b/hw/net/trace-events
index e18f883cfd..a00ef6dc1c 100644
--- a/hw/net/trace-events
+++ b/hw/net/trace-events
@@ -381,6 +381,9 @@ virtio_net_announce_notify(void) ""
 virtio_net_announce_timer(int round) "%d"
 virtio_net_handle_announce(int round) "%d"
 virtio_net_post_load_device(void)
+virtio_net_rss_disable(void)
+virtio_net_rss_error(const char *msg, uint32_t value) "%s, value 0x%08x"
+virtio_net_rss_enable(uint32_t p1, uint16_t p2, uint8_t p3) "hashes 0x%x, table of %d, key of %d"
 
 # tulip.c
 tulip_reg_write(uint64_t addr, const char *name, int size, uint64_t val) "addr 0x%02"PRIx64" (%s) size %d value 0x%08"PRIx64
diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index 3301869d4f..75e9769f79 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -77,6 +77,16 @@
    tso/gso/gro 'off'. */
 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
 
+#define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
+                                         VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
+                                         VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
+                                         VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
+                                         VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
+                                         VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
+                                         VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
+                                         VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
+                                         VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
+
 /* temporary until standard header include it */
 #if !defined(VIRTIO_NET_HDR_F_RSC_INFO)
 
@@ -108,6 +118,8 @@ static VirtIOFeature feature_sizes[] = {
      .end = endof(struct virtio_net_config, mtu)},
     {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
      .end = endof(struct virtio_net_config, duplex)},
+    {.flags = 1ULL << VIRTIO_NET_F_RSS,
+     .end = endof(struct virtio_net_config, supported_hash_types)},
     {}
 };
 
@@ -138,6 +150,11 @@ static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
     memcpy(netcfg.mac, n->mac, ETH_ALEN);
     virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
     netcfg.duplex = n->net_conf.duplex;
+    netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
+    virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
+                 VIRTIO_NET_RSS_MAX_TABLE_LEN);
+    virtio_stl_p(vdev, &netcfg.supported_hash_types,
+                 VIRTIO_NET_RSS_SUPPORTED_HASHES);
     memcpy(config, &netcfg, n->config_size);
 }
 
@@ -701,6 +718,7 @@ static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
         return features;
     }
 
+    virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
     features = vhost_net_get_features(get_vhost_net(nc->peer), features);
     vdev->backend_features = features;
 
@@ -860,6 +878,7 @@ static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
     }
 
     virtio_net_set_multiqueue(n,
+                              virtio_has_feature(features, VIRTIO_NET_F_RSS) ||
                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
 
     virtio_net_set_mrg_rx_bufs(n,
@@ -1136,25 +1155,152 @@ static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
     }
 }
 
+static void virtio_net_disable_rss(VirtIONet *n)
+{
+    if (n->rss_data.enabled) {
+        trace_virtio_net_rss_disable();
+    }
+    n->rss_data.enabled = false;
+}
+
+static uint16_t virtio_net_handle_rss(VirtIONet *n,
+                                      struct iovec *iov, unsigned int iov_cnt)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(n);
+    struct virtio_net_rss_config cfg;
+    size_t s, offset = 0, size_get;
+    uint16_t queues, i;
+    struct {
+        uint16_t us;
+        uint8_t b;
+    } QEMU_PACKED temp;
+    const char *err_msg = "";
+    uint32_t err_value = 0;
+
+    if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
+        err_msg = "RSS is not negotiated";
+        goto error;
+    }
+    size_get = offsetof(struct virtio_net_rss_config, indirection_table);
+    s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
+    if (s != size_get) {
+        err_msg = "Short command buffer";
+        err_value = (uint32_t)s;
+        goto error;
+    }
+    n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types);
+    n->rss_data.indirections_len =
+        virtio_lduw_p(vdev, &cfg.indirection_table_mask);
+    n->rss_data.indirections_len++;
+    if (!is_power_of_2(n->rss_data.indirections_len)) {
+        err_msg = "Invalid size of indirection table";
+        err_value = n->rss_data.indirections_len;
+        goto error;
+    }
+    if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) {
+        err_msg = "Too large indirection table";
+        err_value = n->rss_data.indirections_len;
+        goto error;
+    }
+    n->rss_data.default_queue =
+        virtio_lduw_p(vdev, &cfg.unclassified_queue);
+    if (n->rss_data.default_queue >= n->max_queues) {
+        err_msg = "Invalid default queue";
+        err_value = n->rss_data.default_queue;
+        goto error;
+    }
+    offset += size_get;
+    size_get = sizeof(uint16_t) * n->rss_data.indirections_len;
+    g_free(n->rss_data.indirections_table);
+    n->rss_data.indirections_table = g_malloc(size_get);
+    if (!n->rss_data.indirections_table) {
+        err_msg = "Can't allocate indirections table";
+        err_value = n->rss_data.indirections_len;
+        goto error;
+    }
+    s = iov_to_buf(iov, iov_cnt, offset,
+                   n->rss_data.indirections_table, size_get);
+    if (s != size_get) {
+        err_msg = "Short indirection table buffer";
+        err_value = (uint32_t)s;
+        goto error;
+    }
+    for (i = 0; i < n->rss_data.indirections_len; ++i) {
+        uint16_t val = n->rss_data.indirections_table[i];
+        n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val);
+    }
+    offset += size_get;
+    size_get = sizeof(temp);
+    s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get);
+    if (s != size_get) {
+        err_msg = "Can't get queues";
+        err_value = (uint32_t)s;
+        goto error;
+    }
+    queues = virtio_lduw_p(vdev, &temp.us);
+    if (queues == 0 || queues > n->max_queues) {
+        err_msg = "Invalid number of queues";
+        err_value = queues;
+        goto error;
+    }
+    if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
+        err_msg = "Invalid key size";
+        err_value = temp.b;
+        goto error;
+    }
+    if (!temp.b && n->rss_data.hash_types) {
+        err_msg = "No key provided";
+        err_value = 0;
+        goto error;
+    }
+    if (!temp.b && !n->rss_data.hash_types) {
+        virtio_net_disable_rss(n);
+        return queues;
+    }
+    offset += size_get;
+    size_get = temp.b;
+    s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get);
+    if (s != size_get) {
+        err_msg = "Can get key buffer";
+        err_value = (uint32_t)s;
+        goto error;
+    }
+    n->rss_data.enabled = true;
+    trace_virtio_net_rss_enable(n->rss_data.hash_types,
+                                n->rss_data.indirections_len,
+                                temp.b);
+    return queues;
+error:
+    trace_virtio_net_rss_error(err_msg, err_value);
+    virtio_net_disable_rss(n);
+    return 0;
+}
+
 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
                                 struct iovec *iov, unsigned int iov_cnt)
 {
     VirtIODevice *vdev = VIRTIO_DEVICE(n);
-    struct virtio_net_ctrl_mq mq;
-    size_t s;
     uint16_t queues;
 
-    s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
-    if (s != sizeof(mq)) {
-        return VIRTIO_NET_ERR;
-    }
+    virtio_net_disable_rss(n);
+    if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
+        queues = virtio_net_handle_rss(n, iov, iov_cnt);
+    } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
+        struct virtio_net_ctrl_mq mq;
+        size_t s;
+        if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) {
+            return VIRTIO_NET_ERR;
+        }
+        s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
+        if (s != sizeof(mq)) {
+            return VIRTIO_NET_ERR;
+        }
+        queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
 
-    if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
+    } else {
         return VIRTIO_NET_ERR;
     }
 
-    queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
-
     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
         queues > n->max_queues ||
@@ -3111,6 +3257,7 @@ static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
     g_free(n->vqs);
     qemu_del_nic(n->nic);
     virtio_net_rsc_cleanup(n);
+    g_free(n->rss_data.indirections_table);
     virtio_cleanup(vdev);
 }
 
@@ -3212,6 +3359,8 @@ static Property virtio_net_properties[] = {
     DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
     DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
+    DEFINE_PROP_BIT64("rss", VirtIONet, host_features,
+                    VIRTIO_NET_F_RSS, false),
     DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
                     VIRTIO_NET_F_RSC_EXT, false),
     DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
diff --git a/include/hw/virtio/virtio-net.h b/include/hw/virtio/virtio-net.h
index 96c68d4a92..d3fad7c8f3 100644
--- a/include/hw/virtio/virtio-net.h
+++ b/include/hw/virtio/virtio-net.h
@@ -126,6 +126,18 @@ typedef struct VirtioNetRscChain {
 /* Maximum packet size we can receive from tap device: header + 64k */
 #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 * KiB))
 
+#define VIRTIO_NET_RSS_MAX_KEY_SIZE     40
+#define VIRTIO_NET_RSS_MAX_TABLE_LEN    128
+
+typedef struct VirtioNetRssData {
+    bool    enabled;
+    uint32_t hash_types;
+    uint8_t key[VIRTIO_NET_RSS_MAX_KEY_SIZE];
+    uint16_t indirections_len;
+    uint16_t *indirections_table;
+    uint16_t default_queue;
+} VirtioNetRssData;
+
 typedef struct VirtIONetQueue {
     VirtQueue *rx_vq;
     VirtQueue *tx_vq;
@@ -199,6 +211,7 @@ struct VirtIONet {
     bool failover;
     DeviceListener primary_listener;
     Notifier migration_state;
+    VirtioNetRssData rss_data;
 };
 
 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
-- 
2.17.1

[PATCH v8 2/7] virtio-net: implement RX RSS processing

[Yuri Benditovich]

If VIRTIO_NET_F_RSS negotiated and RSS is enabled, process
incoming packets, calculate packet's hash and place the
packet into respective RX virtqueue.

Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com>
---
 hw/net/virtio-net.c            | 88 +++++++++++++++++++++++++++++++++-
 include/hw/virtio/virtio-net.h |  1 +
 2 files changed, 87 insertions(+), 2 deletions(-)

diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index 75e9769f79..3511ea6c13 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -42,6 +42,7 @@
 #include "trace.h"
 #include "monitor/qdev.h"
 #include "hw/pci/pci.h"
+#include "net_rx_pkt.h"
 
 #define VIRTIO_NET_VM_VERSION    11
 
@@ -1533,8 +1534,80 @@ static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
     return 0;
 }
 
+static uint8_t virtio_net_get_hash_type(bool isip4,
+                                        bool isip6,
+                                        bool isudp,
+                                        bool istcp,
+                                        uint32_t types)
+{
+    if (isip4) {
+        if (istcp && (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4)) {
+            return NetPktRssIpV4Tcp;
+        }
+        if (isudp && (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4)) {
+            return NetPktRssIpV4Udp;
+        }
+        if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
+            return NetPktRssIpV4;
+        }
+    } else if (isip6) {
+        uint32_t mask = VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
+                        VIRTIO_NET_RSS_HASH_TYPE_TCPv6;
+
+        if (istcp && (types & mask)) {
+            return (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) ?
+                NetPktRssIpV6TcpEx : NetPktRssIpV6Tcp;
+        }
+        mask = VIRTIO_NET_RSS_HASH_TYPE_UDP_EX | VIRTIO_NET_RSS_HASH_TYPE_UDPv6;
+        if (isudp && (types & mask)) {
+            return (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) ?
+                NetPktRssIpV6UdpEx : NetPktRssIpV6Udp;
+        }
+        mask = VIRTIO_NET_RSS_HASH_TYPE_IP_EX | VIRTIO_NET_RSS_HASH_TYPE_IPv6;
+        if (types & mask) {
+            return (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) ?
+                NetPktRssIpV6Ex : NetPktRssIpV6;
+        }
+    }
+    return 0xff;
+}
+
+static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
+                                  size_t size)
+{
+    VirtIONet *n = qemu_get_nic_opaque(nc);
+    unsigned int index = nc->queue_index, new_index;
+    struct NetRxPkt *pkt = n->rx_pkt;
+    uint8_t net_hash_type;
+    uint32_t hash;
+    bool isip4, isip6, isudp, istcp;
+
+    net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len,
+                             size - n->host_hdr_len);
+    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
+    if (isip4 && (net_rx_pkt_get_ip4_info(pkt)->fragment)) {
+        istcp = isudp = false;
+    }
+    if (isip6 && (net_rx_pkt_get_ip6_info(pkt)->fragment)) {
+        istcp = isudp = false;
+    }
+    net_hash_type = virtio_net_get_hash_type(isip4, isip6, isudp, istcp,
+                                             n->rss_data.hash_types);
+    if (net_hash_type > NetPktRssIpV6UdpEx) {
+        return n->rss_data.default_queue;
+    }
+
+    hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
+    new_index = hash & (n->rss_data.indirections_len - 1);
+    new_index = n->rss_data.indirections_table[new_index];
+    if (index == new_index) {
+        return -1;
+    }
+    return new_index;
+}
+
 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
-                                      size_t size)
+                                      size_t size, bool no_rss)
 {
     VirtIONet *n = qemu_get_nic_opaque(nc);
     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
@@ -1548,6 +1621,14 @@ static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
         return -1;
     }
 
+    if (!no_rss && n->rss_data.enabled) {
+        int index = virtio_net_process_rss(nc, buf, size);
+        if (index >= 0) {
+            NetClientState *nc2 = qemu_get_subqueue(n->nic, index);
+            return virtio_net_receive_rcu(nc2, buf, size, true);
+        }
+    }
+
     /* hdr_len refers to the header we supply to the guest */
     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
         return 0;
@@ -1642,7 +1723,7 @@ static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
 {
     RCU_READ_LOCK_GUARD();
 
-    return virtio_net_receive_rcu(nc, buf, size);
+    return virtio_net_receive_rcu(nc, buf, size, false);
 }
 
 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
@@ -3221,6 +3302,8 @@ static void virtio_net_device_realize(DeviceState *dev, Error **errp)
 
     QTAILQ_INIT(&n->rsc_chains);
     n->qdev = dev;
+
+    net_rx_pkt_init(&n->rx_pkt, false);
 }
 
 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
@@ -3258,6 +3341,7 @@ static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
     qemu_del_nic(n->nic);
     virtio_net_rsc_cleanup(n);
     g_free(n->rss_data.indirections_table);
+    net_rx_pkt_uninit(n->rx_pkt);
     virtio_cleanup(vdev);
 }
 
diff --git a/include/hw/virtio/virtio-net.h b/include/hw/virtio/virtio-net.h
index d3fad7c8f3..5081f3c52a 100644
--- a/include/hw/virtio/virtio-net.h
+++ b/include/hw/virtio/virtio-net.h
@@ -212,6 +212,7 @@ struct VirtIONet {
     DeviceListener primary_listener;
     Notifier migration_state;
     VirtioNetRssData rss_data;
+    struct NetRxPkt *rx_pkt;
 };
 
 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
-- 
2.17.1

[PATCH v8 3/7] tap: allow extended virtio header with hash info

[Yuri Benditovich]

Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com>
---
 net/tap.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/tap.c b/net/tap.c
index 6207f61f84..ca48f2a285 100644
--- a/net/tap.c
+++ b/net/tap.c
@@ -254,7 +254,8 @@ static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 
     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
-           len == sizeof(struct virtio_net_hdr));
+           len == sizeof(struct virtio_net_hdr) ||
+           len == sizeof(struct virtio_net_hdr_v1_hash));
 
     tap_fd_set_vnet_hdr_len(s->fd, len);
     s->host_vnet_hdr_len = len;
-- 
2.17.1

[PATCH v8 4/7] virtio-net: reference implementation of hash report

[Yuri Benditovich]

Suggest VIRTIO_NET_F_HASH_REPORT if specified in device
parameters.
If the VIRTIO_NET_F_HASH_REPORT is set,
the device extends configuration space. If the feature
is negotiated, the packet layout is extended to
accomodate the hash information. In this case deliver
packet's hash value and report type in virtio header
extension.
Use for configuration the same procedure as already
used for RSS. We add two fields in rss_data that
controls what the device does with the calculated hash
if rss_data.enabled is set. If field 'populate' is set
the hash is set in the packet, if field 'redirect' is
set the hash is used to decide the queue to place the
packet to.

Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com>
---
 hw/net/virtio-net.c            | 99 +++++++++++++++++++++++++++-------
 include/hw/virtio/virtio-net.h |  2 +
 2 files changed, 81 insertions(+), 20 deletions(-)

diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index 3511ea6c13..5facd333ce 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -119,7 +119,7 @@ static VirtIOFeature feature_sizes[] = {
      .end = endof(struct virtio_net_config, mtu)},
     {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
      .end = endof(struct virtio_net_config, duplex)},
-    {.flags = 1ULL << VIRTIO_NET_F_RSS,
+    {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT),
      .end = endof(struct virtio_net_config, supported_hash_types)},
     {}
 };
@@ -153,7 +153,8 @@ static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
     netcfg.duplex = n->net_conf.duplex;
     netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
     virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
-                 VIRTIO_NET_RSS_MAX_TABLE_LEN);
+                 virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ?
+                 VIRTIO_NET_RSS_MAX_TABLE_LEN : 1);
     virtio_stl_p(vdev, &netcfg.supported_hash_types,
                  VIRTIO_NET_RSS_SUPPORTED_HASHES);
     memcpy(config, &netcfg, n->config_size);
@@ -579,7 +580,7 @@ static int peer_has_ufo(VirtIONet *n)
 }
 
 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
-                                       int version_1)
+                                       int version_1, int hash_report)
 {
     int i;
     NetClientState *nc;
@@ -587,7 +588,10 @@ static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
     n->mergeable_rx_bufs = mergeable_rx_bufs;
 
     if (version_1) {
-        n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
+        n->guest_hdr_len = hash_report ?
+            sizeof(struct virtio_net_hdr_v1_hash) :
+            sizeof(struct virtio_net_hdr_mrg_rxbuf);
+        n->rss_data.populate_hash = !!hash_report;
     } else {
         n->guest_hdr_len = n->mergeable_rx_bufs ?
             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
@@ -708,6 +712,8 @@ static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
+
+        virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
     }
 
     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
@@ -720,6 +726,7 @@ static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
     }
 
     virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
+    virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
     features = vhost_net_get_features(get_vhost_net(nc->peer), features);
     vdev->backend_features = features;
 
@@ -886,12 +893,15 @@ static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
                                virtio_has_feature(features,
                                                   VIRTIO_NET_F_MRG_RXBUF),
                                virtio_has_feature(features,
-                                                  VIRTIO_F_VERSION_1));
+                                                  VIRTIO_F_VERSION_1),
+                               virtio_has_feature(features,
+                                                  VIRTIO_NET_F_HASH_REPORT));
 
     n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
     n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
+    n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS);
 
     if (n->has_vnet_hdr) {
         n->curr_guest_offloads =
@@ -1165,7 +1175,9 @@ static void virtio_net_disable_rss(VirtIONet *n)
 }
 
 static uint16_t virtio_net_handle_rss(VirtIONet *n,
-                                      struct iovec *iov, unsigned int iov_cnt)
+                                      struct iovec *iov,
+                                      unsigned int iov_cnt,
+                                      bool do_rss)
 {
     VirtIODevice *vdev = VIRTIO_DEVICE(n);
     struct virtio_net_rss_config cfg;
@@ -1178,10 +1190,14 @@ static uint16_t virtio_net_handle_rss(VirtIONet *n,
     const char *err_msg = "";
     uint32_t err_value = 0;
 
-    if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
+    if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
         err_msg = "RSS is not negotiated";
         goto error;
     }
+    if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) {
+        err_msg = "Hash report is not negotiated";
+        goto error;
+    }
     size_get = offsetof(struct virtio_net_rss_config, indirection_table);
     s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
     if (s != size_get) {
@@ -1193,6 +1209,9 @@ static uint16_t virtio_net_handle_rss(VirtIONet *n,
     n->rss_data.indirections_len =
         virtio_lduw_p(vdev, &cfg.indirection_table_mask);
     n->rss_data.indirections_len++;
+    if (!do_rss) {
+        n->rss_data.indirections_len = 1;
+    }
     if (!is_power_of_2(n->rss_data.indirections_len)) {
         err_msg = "Invalid size of indirection table";
         err_value = n->rss_data.indirections_len;
@@ -1203,8 +1222,8 @@ static uint16_t virtio_net_handle_rss(VirtIONet *n,
         err_value = n->rss_data.indirections_len;
         goto error;
     }
-    n->rss_data.default_queue =
-        virtio_lduw_p(vdev, &cfg.unclassified_queue);
+    n->rss_data.default_queue = do_rss ?
+        virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0;
     if (n->rss_data.default_queue >= n->max_queues) {
         err_msg = "Invalid default queue";
         err_value = n->rss_data.default_queue;
@@ -1238,7 +1257,7 @@ static uint16_t virtio_net_handle_rss(VirtIONet *n,
         err_value = (uint32_t)s;
         goto error;
     }
-    queues = virtio_lduw_p(vdev, &temp.us);
+    queues = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queues;
     if (queues == 0 || queues > n->max_queues) {
         err_msg = "Invalid number of queues";
         err_value = queues;
@@ -1284,8 +1303,12 @@ static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
     uint16_t queues;
 
     virtio_net_disable_rss(n);
+    if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) {
+        queues = virtio_net_handle_rss(n, iov, iov_cnt, false);
+        return queues ? VIRTIO_NET_OK : VIRTIO_NET_ERR;
+    }
     if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
-        queues = virtio_net_handle_rss(n, iov, iov_cnt);
+        queues = virtio_net_handle_rss(n, iov, iov_cnt, true);
     } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
         struct virtio_net_ctrl_mq mq;
         size_t s;
@@ -1572,15 +1595,34 @@ static uint8_t virtio_net_get_hash_type(bool isip4,
     return 0xff;
 }
 
+static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report,
+                                   uint32_t hash)
+{
+    struct virtio_net_hdr_v1_hash *hdr = (void *)buf;
+    hdr->hash_value = hash;
+    hdr->hash_report = report;
+}
+
 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
                                   size_t size)
 {
     VirtIONet *n = qemu_get_nic_opaque(nc);
-    unsigned int index = nc->queue_index, new_index;
+    unsigned int index = nc->queue_index, new_index = index;
     struct NetRxPkt *pkt = n->rx_pkt;
     uint8_t net_hash_type;
     uint32_t hash;
     bool isip4, isip6, isudp, istcp;
+    static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = {
+        VIRTIO_NET_HASH_REPORT_IPv4,
+        VIRTIO_NET_HASH_REPORT_TCPv4,
+        VIRTIO_NET_HASH_REPORT_TCPv6,
+        VIRTIO_NET_HASH_REPORT_IPv6,
+        VIRTIO_NET_HASH_REPORT_IPv6_EX,
+        VIRTIO_NET_HASH_REPORT_TCPv6_EX,
+        VIRTIO_NET_HASH_REPORT_UDPv4,
+        VIRTIO_NET_HASH_REPORT_UDPv6,
+        VIRTIO_NET_HASH_REPORT_UDPv6_EX
+    };
 
     net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len,
                              size - n->host_hdr_len);
@@ -1594,16 +1636,24 @@ static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
     net_hash_type = virtio_net_get_hash_type(isip4, isip6, isudp, istcp,
                                              n->rss_data.hash_types);
     if (net_hash_type > NetPktRssIpV6UdpEx) {
-        return n->rss_data.default_queue;
+        if (n->rss_data.populate_hash) {
+            virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0);
+        }
+        return n->rss_data.redirect ? n->rss_data.default_queue : -1;
     }
 
     hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
-    new_index = hash & (n->rss_data.indirections_len - 1);
-    new_index = n->rss_data.indirections_table[new_index];
-    if (index == new_index) {
-        return -1;
+
+    if (n->rss_data.populate_hash) {
+        virtio_set_packet_hash(buf, reports[net_hash_type], hash);
     }
-    return new_index;
+
+    if (n->rss_data.redirect) {
+        new_index = hash & (n->rss_data.indirections_len - 1);
+        new_index = n->rss_data.indirections_table[new_index];
+    }
+
+    return (index == new_index) ? -1 : new_index;
 }
 
 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
@@ -1679,6 +1729,11 @@ static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
             }
 
             receive_header(n, sg, elem->in_num, buf, size);
+            if (n->rss_data.populate_hash) {
+                offset = sizeof(mhdr);
+                iov_from_buf(sg, elem->in_num, offset,
+                             buf + offset, n->host_hdr_len - sizeof(mhdr));
+            }
             offset = n->host_hdr_len;
             total += n->guest_hdr_len;
             guest_offset = n->guest_hdr_len;
@@ -2671,7 +2726,9 @@ static int virtio_net_post_load_device(void *opaque, int version_id)
     trace_virtio_net_post_load_device();
     virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
                                virtio_vdev_has_feature(vdev,
-                                                       VIRTIO_F_VERSION_1));
+                                                       VIRTIO_F_VERSION_1),
+                               virtio_vdev_has_feature(vdev,
+                                                       VIRTIO_NET_F_HASH_REPORT));
 
     /* MAC_TABLE_ENTRIES may be different from the saved image */
     if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
@@ -3290,7 +3347,7 @@ static void virtio_net_device_realize(DeviceState *dev, Error **errp)
 
     n->vqs[0].tx_waiting = 0;
     n->tx_burst = n->net_conf.txburst;
-    virtio_net_set_mrg_rx_bufs(n, 0, 0);
+    virtio_net_set_mrg_rx_bufs(n, 0, 0, 0);
     n->promisc = 1; /* for compatibility */
 
     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
@@ -3445,6 +3502,8 @@ static Property virtio_net_properties[] = {
     DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
     DEFINE_PROP_BIT64("rss", VirtIONet, host_features,
                     VIRTIO_NET_F_RSS, false),
+    DEFINE_PROP_BIT64("hash", VirtIONet, host_features,
+                    VIRTIO_NET_F_HASH_REPORT, false),
     DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
                     VIRTIO_NET_F_RSC_EXT, false),
     DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
diff --git a/include/hw/virtio/virtio-net.h b/include/hw/virtio/virtio-net.h
index 5081f3c52a..a45ef8278e 100644
--- a/include/hw/virtio/virtio-net.h
+++ b/include/hw/virtio/virtio-net.h
@@ -131,6 +131,8 @@ typedef struct VirtioNetRscChain {
 
 typedef struct VirtioNetRssData {
     bool    enabled;
+    bool    redirect;
+    bool    populate_hash;
     uint32_t hash_types;
     uint8_t key[VIRTIO_NET_RSS_MAX_KEY_SIZE];
     uint16_t indirections_len;
-- 
2.17.1

[PATCH v8 5/7] vmstate.h: provide VMSTATE_VARRAY_UINT16_ALLOC macro

[Yuri Benditovich]

Similar to VMSTATE_VARRAY_UINT32_ALLOC, but the size is
16-bit field.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com>
---
 include/migration/vmstate.h | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/include/migration/vmstate.h b/include/migration/vmstate.h
index 30667631bc..baaefb6b9b 100644
--- a/include/migration/vmstate.h
+++ b/include/migration/vmstate.h
@@ -432,6 +432,16 @@ extern const VMStateInfo vmstate_info_qlist;
     .offset     = vmstate_offset_pointer(_state, _field, _type),     \
 }
 
+#define VMSTATE_VARRAY_UINT16_ALLOC(_field, _state, _field_num, _version, _info, _type) {\
+    .name       = (stringify(_field)),                               \
+    .version_id = (_version),                                        \
+    .num_offset = vmstate_offset_value(_state, _field_num, uint16_t),\
+    .info       = &(_info),                                          \
+    .size       = sizeof(_type),                                     \
+    .flags      = VMS_VARRAY_UINT16 | VMS_POINTER | VMS_ALLOC,       \
+    .offset     = vmstate_offset_pointer(_state, _field, _type),     \
+}
+
 #define VMSTATE_VARRAY_UINT16_UNSAFE(_field, _state, _field_num, _version, _info, _type) {\
     .name       = (stringify(_field)),                               \
     .version_id = (_version),                                        \
-- 
2.17.1

[PATCH v8 6/7] virtio-net: add migration support for RSS and hash report

[Yuri Benditovich]

Save and restore RSS/hash report configuration.

Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com>
---
 hw/net/virtio-net.c | 37 +++++++++++++++++++++++++++++++++++++
 1 file changed, 37 insertions(+)

diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index 5facd333ce..c263b1511d 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -2777,6 +2777,13 @@ static int virtio_net_post_load_device(void *opaque, int version_id)
         }
     }
 
+    if (n->rss_data.enabled) {
+        trace_virtio_net_rss_enable(n->rss_data.hash_types,
+                                    n->rss_data.indirections_len,
+                                    sizeof(n->rss_data.key));
+    } else {
+        trace_virtio_net_rss_disable();
+    }
     return 0;
 }
 
@@ -2954,6 +2961,32 @@ static const VMStateDescription vmstate_virtio_net_has_vnet = {
     },
 };
 
+static bool virtio_net_rss_needed(void *opaque)
+{
+    return VIRTIO_NET(opaque)->rss_data.enabled;
+}
+
+static const VMStateDescription vmstate_virtio_net_rss = {
+    .name      = "virtio-net-device/rss",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = virtio_net_rss_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_BOOL(rss_data.enabled, VirtIONet),
+        VMSTATE_BOOL(rss_data.redirect, VirtIONet),
+        VMSTATE_BOOL(rss_data.populate_hash, VirtIONet),
+        VMSTATE_UINT32(rss_data.hash_types, VirtIONet),
+        VMSTATE_UINT16(rss_data.indirections_len, VirtIONet),
+        VMSTATE_UINT16(rss_data.default_queue, VirtIONet),
+        VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet,
+                            VIRTIO_NET_RSS_MAX_KEY_SIZE),
+        VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet,
+                                    rss_data.indirections_len, 0,
+                                    vmstate_info_uint16, uint16_t),
+        VMSTATE_END_OF_LIST()
+    },
+};
+
 static const VMStateDescription vmstate_virtio_net_device = {
     .name = "virtio-net-device",
     .version_id = VIRTIO_NET_VM_VERSION,
@@ -3004,6 +3037,10 @@ static const VMStateDescription vmstate_virtio_net_device = {
                             has_ctrl_guest_offloads),
         VMSTATE_END_OF_LIST()
    },
+    .subsections = (const VMStateDescription * []) {
+        &vmstate_virtio_net_rss,
+        NULL
+    }
 };
 
 static NetClientInfo net_virtio_info = {
-- 
2.17.1

[PATCH v8 7/7] virtio-net: align RSC fields with updated virtio-net header

[Yuri Benditovich]

Removal of duplicated RSC definitions. Changing names of the
fields to ones defined in the Linux header.

Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com>
---
 hw/net/virtio-net.c | 28 ++++------------------------
 1 file changed, 4 insertions(+), 24 deletions(-)

diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index c263b1511d..10b4ba50ad 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -88,26 +88,6 @@
                                          VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
                                          VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
 
-/* temporary until standard header include it */
-#if !defined(VIRTIO_NET_HDR_F_RSC_INFO)
-
-#define VIRTIO_NET_HDR_F_RSC_INFO  4 /* rsc_ext data in csum_ fields */
-#define VIRTIO_NET_F_RSC_EXT       61
-
-#endif
-
-static inline __virtio16 *virtio_net_rsc_ext_num_packets(
-    struct virtio_net_hdr *hdr)
-{
-    return &hdr->csum_start;
-}
-
-static inline __virtio16 *virtio_net_rsc_ext_num_dupacks(
-    struct virtio_net_hdr *hdr)
-{
-    return &hdr->csum_offset;
-}
-
 static VirtIOFeature feature_sizes[] = {
     {.flags = 1ULL << VIRTIO_NET_F_MAC,
      .end = endof(struct virtio_net_config, mac)},
@@ -1821,15 +1801,15 @@ static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
                                        VirtioNetRscSeg *seg)
 {
     int ret;
-    struct virtio_net_hdr *h;
+    struct virtio_net_hdr_v1 *h;
 
-    h = (struct virtio_net_hdr *)seg->buf;
+    h = (struct virtio_net_hdr_v1 *)seg->buf;
     h->flags = 0;
     h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 
     if (seg->is_coalesced) {
-        *virtio_net_rsc_ext_num_packets(h) = seg->packets;
-        *virtio_net_rsc_ext_num_dupacks(h) = seg->dup_ack;
+        h->rsc.segments = seg->packets;
+        h->rsc.dup_acks = seg->dup_ack;
         h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
         if (chain->proto == ETH_P_IP) {
             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
-- 
2.17.1

Re: [PATCH v8 0/7] reference implementation of RSS and hash report

[Jason Wang]

On 2020/5/8 下午8:59, Yuri Benditovich wrote:
> Support for VIRTIO_NET_F_RSS and VIRTIO_NET_F_HASH_REPORT
> features in QEMU for reference purpose.
> Implements Toeplitz hash calculation for incoming
> packets according to configuration provided by driver.
> Uses calculated hash for decision on receive virtqueue
> and/or reports the hash in the virtio header
>
> Changes from v7:
> Patch 7.1: removed (RSS and hash report definitions)
> Patch 7.2: delete configuration struct with RSS definitions
> Patch 7.4: delete duplicated packet structure
> Added patch 7 - adapt RSC definitions to updated header
>
> Yuri Benditovich (7):
>    virtio-net: implement RSS configuration command
>    virtio-net: implement RX RSS processing
>    tap: allow extended virtio header with hash info
>    virtio-net: reference implementation of hash report
>    vmstate.h: provide VMSTATE_VARRAY_UINT16_ALLOC macro
>    virtio-net: add migration support for RSS and hash report
>    virtio-net: align RSC fields with updated virtio-net header
>
>   hw/net/trace-events            |   3 +
>   hw/net/virtio-net.c            | 387 +++++++++++++++++++++++++++++----
>   include/hw/virtio/virtio-net.h |  16 ++
>   include/migration/vmstate.h    |  10 +
>   net/tap.c                      |   3 +-
>   5 files changed, 379 insertions(+), 40 deletions(-)
>

Applied.

Thanks