nagai/kernel
3
0
Fork 0
forked from mirrors/linux
Code Pull requests Activity

netvsc: negotiate checksum and segmentation parameters

Browse source 
Redo how Hyper-V network driver negotiates offload features. Query the
host to determine offload settings, and use the result.

Also:
  * disable IPv4 header checksum offload (not used by Linux)
  * enable TSO only if host supports
  * enable UDP checksum offload if supported
  * don't advertise support for checksumming of non-IP protocols
  * adjust GSO maximum segment size
  * enable HIGHDMA

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
stephen hemminger 2017-01-24 13:05:59 -08:00 committed by David S. Miller
parent 0b307ebd68
commit 23312a3be9
3 changed files with 305 additions and 76 deletions
repo.diff.show_diff_stats Download patch file Download diff file Expand all files Collapse all files

149
drivers/net/hyperv/hyperv_net.h
View file

@ -34,6 +34,7 @@
#define NDIS_OBJECT_TYPE_RSS_CAPABILITIES 0x88
#define NDIS_OBJECT_TYPE_RSS_PARAMETERS 0x89
#define NDIS_OBJECT_TYPE_OFFLOAD 0xa7
#define NDIS_RECEIVE_SCALE_CAPABILITIES_REVISION_2 2
#define NDIS_RECEIVE_SCALE_PARAMETERS_REVISION_2 2
@ -685,6 +686,7 @@ struct net_device_context {
struct work_struct work;
u32 msg_enable; /* debug level */
u32 tx_checksum_mask;
struct netvsc_stats __percpu *tx_stats;
struct netvsc_stats __percpu *rx_stats;
@ -934,7 +936,7 @@ struct ndis_pkt_8021q_info {
};
};
struct ndis_oject_header {
struct ndis_object_header {
u8 type;
u8 revision;
u16 size;
@ -942,6 +944,9 @@ struct ndis_oject_header {
#define NDIS_OBJECT_TYPE_DEFAULT 0x80
#define NDIS_OFFLOAD_PARAMETERS_REVISION_3 3
#define NDIS_OFFLOAD_PARAMETERS_REVISION_2 2
#define NDIS_OFFLOAD_PARAMETERS_REVISION_1 1
#define NDIS_OFFLOAD_PARAMETERS_NO_CHANGE 0
#define NDIS_OFFLOAD_PARAMETERS_LSOV2_DISABLED 1
#define NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED 2
@ -968,8 +973,135 @@ struct ndis_oject_header {
#define OID_TCP_CONNECTION_OFFLOAD_HARDWARE_CAPABILITIES 0xFC01020F /* query */
#define OID_OFFLOAD_ENCAPSULATION 0x0101010A /* set/query */
/*
* OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES
* ndis_type: NDIS_OBJTYPE_OFFLOAD
*/
#define NDIS_OFFLOAD_ENCAP_NONE 0x0000
#define NDIS_OFFLOAD_ENCAP_NULL 0x0001
#define NDIS_OFFLOAD_ENCAP_8023 0x0002
#define NDIS_OFFLOAD_ENCAP_8023PQ 0x0004
#define NDIS_OFFLOAD_ENCAP_8023PQ_OOB 0x0008
#define NDIS_OFFLOAD_ENCAP_RFC1483 0x0010
struct ndis_csum_offload {
u32 ip4_txenc;
u32 ip4_txcsum;
#define NDIS_TXCSUM_CAP_IP4OPT 0x001
#define NDIS_TXCSUM_CAP_TCP4OPT 0x004
#define NDIS_TXCSUM_CAP_TCP4 0x010
#define NDIS_TXCSUM_CAP_UDP4 0x040
#define NDIS_TXCSUM_CAP_IP4 0x100
#define NDIS_TXCSUM_ALL_TCP4 (NDIS_TXCSUM_CAP_TCP4 | NDIS_TXCSUM_CAP_TCP4OPT)
u32 ip4_rxenc;
u32 ip4_rxcsum;
#define NDIS_RXCSUM_CAP_IP4OPT 0x001
#define NDIS_RXCSUM_CAP_TCP4OPT 0x004
#define NDIS_RXCSUM_CAP_TCP4 0x010
#define NDIS_RXCSUM_CAP_UDP4 0x040
#define NDIS_RXCSUM_CAP_IP4 0x100
u32 ip6_txenc;
u32 ip6_txcsum;
#define NDIS_TXCSUM_CAP_IP6EXT 0x001
#define NDIS_TXCSUM_CAP_TCP6OPT 0x004
#define NDIS_TXCSUM_CAP_TCP6 0x010
#define NDIS_TXCSUM_CAP_UDP6 0x040
u32 ip6_rxenc;
u32 ip6_rxcsum;
#define NDIS_RXCSUM_CAP_IP6EXT 0x001
#define NDIS_RXCSUM_CAP_TCP6OPT 0x004
#define NDIS_RXCSUM_CAP_TCP6 0x010
#define NDIS_RXCSUM_CAP_UDP6 0x040
#define NDIS_TXCSUM_ALL_TCP6 (NDIS_TXCSUM_CAP_TCP6 | \
NDIS_TXCSUM_CAP_TCP6OPT | \
NDIS_TXCSUM_CAP_IP6EXT)
};
struct ndis_lsov1_offload {
u32 encap;
u32 maxsize;
u32 minsegs;
u32 opts;
};
struct ndis_ipsecv1_offload {
u32 encap;
u32 ah_esp;
u32 xport_tun;
u32 ip4_opts;
u32 flags;
u32 ip4_ah;
u32 ip4_esp;
};
struct ndis_lsov2_offload {
u32 ip4_encap;
u32 ip4_maxsz;
u32 ip4_minsg;
u32 ip6_encap;
u32 ip6_maxsz;
u32 ip6_minsg;
u32 ip6_opts;
#define NDIS_LSOV2_CAP_IP6EXT 0x001
#define NDIS_LSOV2_CAP_TCP6OPT 0x004
#define NDIS_LSOV2_CAP_IP6 (NDIS_LSOV2_CAP_IP6EXT | \
NDIS_LSOV2_CAP_TCP6OPT)
};
struct ndis_ipsecv2_offload {
u32 encap;
u16 ip6;
u16 ip4opt;
u16 ip6ext;
u16 ah;
u16 esp;
u16 ah_esp;
u16 xport;
u16 tun;
u16 xport_tun;
u16 lso;
u16 extseq;
u32 udp_esp;
u32 auth;
u32 crypto;
u32 sa_caps;
};
struct ndis_rsc_offload {
u16 ip4;
u16 ip6;
};
struct ndis_encap_offload {
u32 flags;
u32 maxhdr;
};
struct ndis_offload {
struct ndis_object_header header;
struct ndis_csum_offload csum;
struct ndis_lsov1_offload lsov1;
struct ndis_ipsecv1_offload ipsecv1;
struct ndis_lsov2_offload lsov2;
u32 flags;
/* NDIS >= 6.1 */
struct ndis_ipsecv2_offload ipsecv2;
/* NDIS >= 6.30 */
struct ndis_rsc_offload rsc;
struct ndis_encap_offload encap_gre;
};
#define NDIS_OFFLOAD_SIZE sizeof(struct ndis_offload)
#define NDIS_OFFLOAD_SIZE_6_0 offsetof(struct ndis_offload, ipsecv2)
#define NDIS_OFFLOAD_SIZE_6_1 offsetof(struct ndis_offload, rsc)
struct ndis_offload_params {
struct ndis_oject_header header;
struct ndis_object_header header;
u8 ip_v4_csum;
u8 tcp_ip_v4_csum;
u8 udp_ip_v4_csum;
@ -1296,15 +1428,10 @@ struct rndis_message {
#define NDIS_PACKET_TYPE_FUNCTIONAL 0x00000400
#define NDIS_PACKET_TYPE_MAC_FRAME 0x00000800
#define INFO_IPV4 2
#define INFO_IPV6 4
#define INFO_TCP 2
#define INFO_UDP 4
#define TRANSPORT_INFO_NOT_IP 0
#define TRANSPORT_INFO_IPV4_TCP ((INFO_IPV4 << 16) | INFO_TCP)
#define TRANSPORT_INFO_IPV4_UDP ((INFO_IPV4 << 16) | INFO_UDP)
#define TRANSPORT_INFO_IPV6_TCP ((INFO_IPV6 << 16) | INFO_TCP)
#define TRANSPORT_INFO_IPV6_UDP ((INFO_IPV6 << 16) | INFO_UDP)
#define TRANSPORT_INFO_IPV4_TCP 0x01
#define TRANSPORT_INFO_IPV4_UDP 0x02
#define TRANSPORT_INFO_IPV6_TCP 0x10
#define TRANSPORT_INFO_IPV6_UDP 0x20
#endif /* _HYPERV_NET_H */

92
drivers/net/hyperv/netvsc_drv.c
View file

@ -42,14 +42,6 @@
#define RING_SIZE_MIN 64
#define LINKCHANGE_INT (2 * HZ)
#define NETVSC_HW_FEATURES (NETIF_F_RXCSUM | \
NETIF_F_SG | \
NETIF_F_TSO | \
NETIF_F_TSO6 | \
NETIF_F_HW_CSUM)
/* Restrict GSO size to account for NVGRE */
#define NETVSC_GSO_MAX_SIZE 62768
static int ring_size = 128;
module_param(ring_size, int, S_IRUGO);
@ -323,33 +315,25 @@ static int netvsc_get_slots(struct sk_buff *skb)
return slots + frag_slots;
}
static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
static u32 net_checksum_info(struct sk_buff *skb)
{
u32 ret_val = TRANSPORT_INFO_NOT_IP;
if (skb->protocol == htons(ETH_P_IP)) {
struct iphdr *ip = ip_hdr(skb);
if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
(eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
goto not_ip;
}
*trans_off = skb_transport_offset(skb);
if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
struct iphdr *iphdr = ip_hdr(skb);
if (iphdr->protocol == IPPROTO_TCP)
ret_val = TRANSPORT_INFO_IPV4_TCP;
else if (iphdr->protocol == IPPROTO_UDP)
ret_val = TRANSPORT_INFO_IPV4_UDP;
if (ip->protocol == IPPROTO_TCP)
return TRANSPORT_INFO_IPV4_TCP;
else if (ip->protocol == IPPROTO_UDP)
return TRANSPORT_INFO_IPV4_UDP;
} else {
if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
ret_val = TRANSPORT_INFO_IPV6_TCP;
struct ipv6hdr *ip6 = ipv6_hdr(skb);
if (ip6->nexthdr == IPPROTO_TCP)
return TRANSPORT_INFO_IPV6_TCP;
else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
ret_val = TRANSPORT_INFO_IPV6_UDP;
return TRANSPORT_INFO_IPV6_UDP;
}
not_ip:
return ret_val;
return TRANSPORT_INFO_NOT_IP;
}
static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
@ -362,9 +346,6 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
struct rndis_packet *rndis_pkt;
u32 rndis_msg_size;
struct rndis_per_packet_info *ppi;
struct ndis_tcp_ip_checksum_info *csum_info;
int hdr_offset;
u32 net_trans_info;
u32 hash;
u32 skb_length;
struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
@ -445,13 +426,7 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
VLAN_PRIO_SHIFT;
}
net_trans_info = get_net_transport_info(skb, &hdr_offset);
/*
* Setup the sendside checksum offload only if this is not a
* GSO packet.
*/
if ((net_trans_info & (INFO_TCP | INFO_UDP)) && skb_is_gso(skb)) {
if (skb_is_gso(skb)) {
struct ndis_tcp_lso_info *lso_info;
rndis_msg_size += NDIS_LSO_PPI_SIZE;
@ -462,7 +437,7 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
ppi->ppi_offset);
lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
if (net_trans_info & (INFO_IPV4 << 16)) {
if (skb->protocol == htons(ETH_P_IP)) {
lso_info->lso_v2_transmit.ip_version =
NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
ip_hdr(skb)->tot_len = 0;
@ -478,10 +453,12 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
}
lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
if (net_trans_info & INFO_TCP) {
if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
struct ndis_tcp_ip_checksum_info *csum_info;
rndis_msg_size += NDIS_CSUM_PPI_SIZE;
ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
TCPIP_CHKSUM_PKTINFO);
@ -489,15 +466,25 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
ppi->ppi_offset);
if (net_trans_info & (INFO_IPV4 << 16))
csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
if (skb->protocol == htons(ETH_P_IP)) {
csum_info->transmit.is_ipv4 = 1;
else
if (ip_hdr(skb)->protocol == IPPROTO_TCP)
csum_info->transmit.tcp_checksum = 1;
else
csum_info->transmit.udp_checksum = 1;
} else {
csum_info->transmit.is_ipv6 = 1;
csum_info->transmit.tcp_checksum = 1;
csum_info->transmit.tcp_header_offset = hdr_offset;
if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
csum_info->transmit.tcp_checksum = 1;
else
csum_info->transmit.udp_checksum = 1;
}
} else {
/* UDP checksum (and other) offload is not supported. */
/* Can't do offload of this type of checksum */
if (skb_checksum_help(skb))
goto drop;
}
@ -1372,10 +1359,6 @@ static int netvsc_probe(struct hv_device *dev,
INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
net->netdev_ops = &device_ops;
net->hw_features = NETVSC_HW_FEATURES;
net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
net->ethtool_ops = &ethtool_ops;
SET_NETDEV_DEV(net, &dev->device);
@ -1395,10 +1378,15 @@ static int netvsc_probe(struct hv_device *dev,
}
memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
/* hw_features computed in rndis_filter_device_add */
net->features = net->hw_features |
NETIF_F_HIGHDMA | NETIF_F_SG |
NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
net->vlan_features = net->features;
nvdev = net_device_ctx->nvdev;
netif_set_real_num_tx_queues(net, nvdev->num_chn);
netif_set_real_num_rx_queues(net, nvdev->num_chn);
netif_set_gso_max_size(net, NETVSC_GSO_MAX_SIZE);
/* MTU range: 68 - 1500 or 65521 */
net->min_mtu = NETVSC_MTU_MIN;

140
drivers/net/hyperv/rndis_filter.c
View file

@ -485,7 +485,35 @@ static int rndis_filter_query_device(struct rndis_device *dev, u32 oid,
query->info_buflen = 0;
query->dev_vc_handle = 0;
if (oid == OID_GEN_RECEIVE_SCALE_CAPABILITIES) {
if (oid == OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES) {
struct net_device_context *ndevctx = netdev_priv(dev->ndev);
struct netvsc_device *nvdev = ndevctx->nvdev;
struct ndis_offload *hwcaps;
u32 nvsp_version = nvdev->nvsp_version;
u8 ndis_rev;
size_t size;
if (nvsp_version >= NVSP_PROTOCOL_VERSION_5) {
ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
size = NDIS_OFFLOAD_SIZE;
} else if (nvsp_version >= NVSP_PROTOCOL_VERSION_4) {
ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_2;
size = NDIS_OFFLOAD_SIZE_6_1;
} else {
ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_1;
size = NDIS_OFFLOAD_SIZE_6_0;
}
request->request_msg.msg_len += size;
query->info_buflen = size;
hwcaps = (struct ndis_offload *)
((unsigned long)query + query->info_buf_offset);
hwcaps->header.type = NDIS_OBJECT_TYPE_OFFLOAD;
hwcaps->header.revision = ndis_rev;
hwcaps->header.size = size;
} else if (oid == OID_GEN_RECEIVE_SCALE_CAPABILITIES) {
struct ndis_recv_scale_cap *cap;
request->request_msg.msg_len +=
@ -526,6 +554,44 @@ static int rndis_filter_query_device(struct rndis_device *dev, u32 oid,
return ret;
}
/* Get the hardware offload capabilities */
static int
rndis_query_hwcaps(struct rndis_device *dev, struct ndis_offload *caps)
{
u32 caps_len = sizeof(*caps);
int ret;
memset(caps, 0, sizeof(*caps));
ret = rndis_filter_query_device(dev,
OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES,
caps, &caps_len);
if (ret)
return ret;
if (caps->header.type != NDIS_OBJECT_TYPE_OFFLOAD) {
netdev_warn(dev->ndev, "invalid NDIS objtype %#x\n",
caps->header.type);
return -EINVAL;
}
if (caps->header.revision < NDIS_OFFLOAD_PARAMETERS_REVISION_1) {
netdev_warn(dev->ndev, "invalid NDIS objrev %x\n",
caps->header.revision);
return -EINVAL;
}
if (caps->header.size > caps_len ||
caps->header.size < NDIS_OFFLOAD_SIZE_6_0) {
netdev_warn(dev->ndev,
"invalid NDIS objsize %u, data size %u\n",
caps->header.size, caps_len);
return -EINVAL;
}
return 0;
}
static int rndis_filter_query_device_mac(struct rndis_device *dev)
{
u32 size = ETH_ALEN;
@ -974,10 +1040,12 @@ int rndis_filter_device_add(struct hv_device *dev,
struct netvsc_device *net_device;
struct rndis_device *rndis_device;
struct netvsc_device_info *device_info = additional_info;
struct ndis_offload hwcaps;
struct ndis_offload_params offloads;
struct nvsp_message *init_packet;
struct ndis_recv_scale_cap rsscap;
u32 rsscap_size = sizeof(struct ndis_recv_scale_cap);
unsigned int gso_max_size = GSO_MAX_SIZE;
u32 mtu, size;
u32 num_rss_qs;
u32 sc_delta;
@ -1034,19 +1102,65 @@ int rndis_filter_device_add(struct hv_device *dev,
memcpy(device_info->mac_adr, rndis_device->hw_mac_adr, ETH_ALEN);
/* Turn on the offloads; the host supports all of the relevant
* offloads.
*/
/* Find HW offload capabilities */
ret = rndis_query_hwcaps(rndis_device, &hwcaps);
if (ret != 0) {
rndis_filter_device_remove(dev);
return ret;
}
/* A value of zero means "no change"; now turn on what we want. */
memset(&offloads, 0, sizeof(struct ndis_offload_params));
/* A value of zero means "no change"; now turn on what we
* want.
*/
offloads.ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
offloads.tcp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
offloads.udp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
offloads.tcp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
offloads.udp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
offloads.lso_v2_ipv4 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
/* Linux does not care about IP checksum, always does in kernel */
offloads.ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_DISABLED;
/* Compute tx offload settings based on hw capabilities */
net->hw_features = NETIF_F_RXCSUM;
if ((hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_ALL_TCP4) == NDIS_TXCSUM_ALL_TCP4) {
/* Can checksum TCP */
net->hw_features |= NETIF_F_IP_CSUM;
net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_TCP;
offloads.tcp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
if (hwcaps.lsov2.ip4_encap & NDIS_OFFLOAD_ENCAP_8023) {
offloads.lso_v2_ipv4 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
net->hw_features |= NETIF_F_TSO;
if (hwcaps.lsov2.ip4_maxsz < gso_max_size)
gso_max_size = hwcaps.lsov2.ip4_maxsz;
}
if (hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_CAP_UDP4) {
offloads.udp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_UDP;
}
}
if ((hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_ALL_TCP6) == NDIS_TXCSUM_ALL_TCP6) {
net->hw_features |= NETIF_F_IPV6_CSUM;
offloads.tcp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_TCP;
if ((hwcaps.lsov2.ip6_encap & NDIS_OFFLOAD_ENCAP_8023) &&
(hwcaps.lsov2.ip6_opts & NDIS_LSOV2_CAP_IP6) == NDIS_LSOV2_CAP_IP6) {
offloads.lso_v2_ipv6 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
net->hw_features |= NETIF_F_TSO6;
if (hwcaps.lsov2.ip6_maxsz < gso_max_size)
gso_max_size = hwcaps.lsov2.ip6_maxsz;
}
if (hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_CAP_UDP6) {
offloads.udp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_UDP;
}
}
netif_set_gso_max_size(net, gso_max_size);
ret = rndis_filter_set_offload_params(net, &offloads);
if (ret)