Module Name: src
Committed By: jdolecek
Date: Mon Mar 16 20:49:22 UTC 2020
Modified Files:
src/sys/arch/xen/xen: if_xennet_xenbus.c xennet_checksum.c
xennetback_xenbus.c
Log Message:
drop the disabled M_EXT_ROMAP mbuf code, and convert xennet_checksum_fill()
to use in_undefer_cksum() instead of custom code to compute the checksum
for incoming packets
make it possible to defer/skip the checksum computation by appropriate Rx
flag similarily as we do for Tx; with
To generate a diff of this commit:
cvs rdiff -u -r1.88 -r1.89 src/sys/arch/xen/xen/if_xennet_xenbus.c
cvs rdiff -u -r1.4 -r1.5 src/sys/arch/xen/xen/xennet_checksum.c
cvs rdiff -u -r1.76 -r1.77 src/sys/arch/xen/xen/xennetback_xenbus.c
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/arch/xen/xen/if_xennet_xenbus.c
diff -u src/sys/arch/xen/xen/if_xennet_xenbus.c:1.88 src/sys/arch/xen/xen/if_xennet_xenbus.c:1.89
--- src/sys/arch/xen/xen/if_xennet_xenbus.c:1.88 Wed Jan 29 05:41:48 2020
+++ src/sys/arch/xen/xen/if_xennet_xenbus.c Mon Mar 16 20:49:22 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: if_xennet_xenbus.c,v 1.88 2020/01/29 05:41:48 thorpej Exp $ */
+/* $NetBSD: if_xennet_xenbus.c,v 1.89 2020/03/16 20:49:22 jdolecek Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -84,7 +84,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: if_xennet_xenbus.c,v 1.88 2020/01/29 05:41:48 thorpej Exp $");
+__KERNEL_RCSID(0, "$NetBSD: if_xennet_xenbus.c,v 1.89 2020/03/16 20:49:22 jdolecek Exp $");
#include "opt_xen.h"
#include "opt_nfs_boot.h"
@@ -386,7 +386,9 @@ xennet_xenbus_attach(device_t parent, de
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_timer = 0;
ifp->if_snd.ifq_maxlen = uimax(ifqmaxlen, NET_TX_RING_SIZE * 2);
- ifp->if_capabilities = IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_UDPv4_Tx;
+ ifp->if_capabilities =
+ IFCAP_CSUM_UDPv4_Rx | IFCAP_CSUM_UDPv4_Tx
+ | IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_TCPv4_Tx;
IFQ_SET_READY(&ifp->if_snd);
if_attach(ifp);
ether_ifattach(ifp, sc->sc_enaddr);
@@ -1119,12 +1121,7 @@ again:
m->m_flags |= M_EXT_RW; /* we own the buffer */
}
if ((rx->flags & NETRXF_csum_blank) != 0) {
- xennet_checksum_fill(&m);
- if (m == NULL) {
- if_statinc(ifp, if_ierrors);
- xennet_rx_free_req(req);
- continue;
- }
+ xennet_checksum_fill(ifp, m);
}
/* free req may overwrite *rx, better doing it late */
xennet_rx_free_req(req);
Index: src/sys/arch/xen/xen/xennet_checksum.c
diff -u src/sys/arch/xen/xen/xennet_checksum.c:1.4 src/sys/arch/xen/xen/xennet_checksum.c:1.5
--- src/sys/arch/xen/xen/xennet_checksum.c:1.4 Sat Mar 14 11:52:12 2020
+++ src/sys/arch/xen/xen/xennet_checksum.c Mon Mar 16 20:49:22 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: xennet_checksum.c,v 1.4 2020/03/14 11:52:12 jdolecek Exp $ */
+/* $NetBSD: xennet_checksum.c,v 1.5 2020/03/16 20:49:22 jdolecek Exp $ */
/*-
* Copyright (c)2006 YAMAMOTO Takashi,
@@ -27,7 +27,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: xennet_checksum.c,v 1.4 2020/03/14 11:52:12 jdolecek Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xennet_checksum.c,v 1.5 2020/03/16 20:49:22 jdolecek Exp $");
#include <sys/types.h>
#include <sys/param.h>
@@ -39,73 +39,51 @@ __KERNEL_RCSID(0, "$NetBSD: xennet_check
#include <netinet/in.h>
#include <netinet/in_systm.h>
+#include <netinet/in_offload.h>
#include <netinet/ip.h>
#include <netinet/tcp.h>
#include <netinet/udp.h>
#include <xen/xennet_checksum.h>
-static const void *m_extract(struct mbuf *, int, int, void *);
-static void *m_extract_write(struct mbuf *, int, int, void *);
-
-static void *m_extract1(struct mbuf *, int, int, void *, int);
-#define MBUF_EXTRACT_WRITE 1
+/* ratecheck(9) for checksum validation failures */
+static const struct timeval xn_cksum_errintvl = { 600, 0 }; /* 10 min, each */
static void *
-m_extract1(struct mbuf *m, int off, int len, void *buf, int flags)
-{
- void *result;
-
- KASSERT((m->m_flags & M_PKTHDR) != 0);
-
- if (m->m_pkthdr.len < off + len) {
- result = NULL;
- } else if (m->m_len >= off + len &&
- ((flags & MBUF_EXTRACT_WRITE) != 0 || !M_READONLY(m))) {
- result = mtod(m, char *) + off;
- } else {
- m_copydata(m, off, len, buf);
- result = buf;
- }
-
- return result;
-}
-
-static const void *
-m_extract(struct mbuf *m, int off, int len, void *buf)
+m_extract(struct mbuf *m, int off, int len)
{
+ KASSERT(m->m_pkthdr.len >= off + len);
+ KASSERT(m->m_len >= off + len);
- return m_extract1(m, off, len, buf, 0);
-}
-
-static void *
-m_extract_write(struct mbuf *m, int off, int len, void *buf)
-{
-
- return m_extract1(m, off, len, buf, MBUF_EXTRACT_WRITE);
+ if (m->m_pkthdr.len >= off + len)
+ return mtod(m, char *) + off;
+ else
+ return NULL;
}
/*
- * xennet_checksum_fill: fill TCP/UDP checksum
+ * xennet_checksum_fill: fill TCP/UDP checksum, or arrange
+ * for hw offload to do it
*/
-
int
-xennet_checksum_fill(struct mbuf **mp)
+xennet_checksum_fill(struct ifnet *ifp, struct mbuf *m)
{
- struct mbuf *m = *mp;
- struct ether_header eh_store;
const struct ether_header *eh;
- struct ip iph_store;
- const struct ip *iph;
+ struct ip *iph;
int ehlen;
int iphlen;
int iplen;
uint16_t etype;
uint8_t nxt;
int error = 0;
+ int sw_csum;
+
+ KASSERT(!M_READONLY(m));
+ KASSERT((m->m_flags & M_PKTHDR) != 0);
- eh = m_extract(m, 0, sizeof(*eh), &eh_store);
+ eh = m_extract(m, 0, sizeof(*eh));
if (eh == NULL) {
+ /* Too short, packet will be dropped by upper layer */
return EINVAL;
}
etype = eh->ether_type;
@@ -114,69 +92,58 @@ xennet_checksum_fill(struct mbuf **mp)
} else if (etype == htobe16(ETHERTYPE_IP)) {
ehlen = ETHER_HDR_LEN;
} else {
+ static struct timeval lasttime;
+ if (ratecheck(&lasttime, &xn_cksum_errintvl))
+ printf("%s: unknown etype %#x passed no checksum\n",
+ ifp->if_xname, ntohs(etype));
return EINVAL;
}
- iph = m_extract(m, ehlen, sizeof(*iph), &iph_store);
+ iph = m_extract(m, ehlen, sizeof(*iph));
if (iph == NULL) {
+ /* Too short, packet will be dropped by upper layer */
return EINVAL;
}
nxt = iph->ip_p;
- iphlen = iph->ip_hl * 4;
+ iphlen = iph->ip_hl << 2;
iplen = ntohs(iph->ip_len);
if (ehlen + iplen > m->m_pkthdr.len) {
+ /* Too short, packet will be dropped by upper layer */
return EINVAL;
}
- if (nxt == IPPROTO_UDP) {
- struct udphdr uh_store;
- struct udphdr *uh;
- int ulen;
-
- uh = m_extract_write(m, ehlen + iphlen, sizeof(*uh), &uh_store);
- ulen = ntohs(uh->uh_ulen);
- if (ehlen + iphlen + ulen > m->m_pkthdr.len) {
- return EINVAL;
- }
- m->m_len -= ehlen;
- m->m_data += ehlen;
- uh->uh_sum = 0;
- uh->uh_sum = in4_cksum(m, nxt, iphlen, iplen - iphlen);
- m->m_len += ehlen;
- m->m_data -= ehlen;
- if (uh != &uh_store) {
- m = m_copyback_cow(m, ehlen + iphlen, sizeof(*uh), uh,
- M_DONTWAIT);
- if (m == NULL) {
- error = ENOMEM;
- }
- }
- } else if (nxt == IPPROTO_TCP) {
- struct tcphdr th_store;
- struct tcphdr *th;
- int thlen;
-
- th = m_extract_write(m, ehlen + iphlen, sizeof(*th), &th_store);
- thlen = th->th_off * 4;
- if (ehlen + iphlen + thlen > m->m_pkthdr.len) {
- return EINVAL;
- }
- m->m_len -= ehlen;
- m->m_data += ehlen;
- th->th_sum = 0;
- th->th_sum = in4_cksum(m, nxt, iphlen, iplen - iphlen);
- m->m_len += ehlen;
- m->m_data -= ehlen;
- if (th != &th_store) {
- m = m_copyback_cow(m, ehlen + iphlen, sizeof(*th), th,
- M_DONTWAIT);
- if (m == NULL) {
- error = ENOMEM;
- }
- }
- } else {
+
+ switch (nxt) {
+ case IPPROTO_UDP:
+ m->m_pkthdr.csum_flags = M_CSUM_UDPv4;
+ m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
+ m->m_pkthdr.csum_data |= iphlen << 16;
+ break;
+ case IPPROTO_TCP:
+ m->m_pkthdr.csum_flags = M_CSUM_TCPv4;
+ m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
+ m->m_pkthdr.csum_data |= iphlen << 16;
+ break;
+ default:
+ {
+ static struct timeval lasttime;
+ if (ratecheck(&lasttime, &xn_cksum_errintvl))
+ printf("%s: unknown proto %d passed no checksum\n",
+ ifp->if_xname, nxt);
error = EINVAL;
+ goto out;
+ }
+ }
+
+ /*
+ * Only compute the checksum if impossible to defer.
+ */
+ sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_rx;
+
+ if (sw_csum & (M_CSUM_UDPv4|M_CSUM_TCPv4)) {
+ in_undefer_cksum(m, ehlen,
+ sw_csum & (M_CSUM_UDPv4|M_CSUM_TCPv4));
}
- *mp = m;
+ out:
return error;
}
Index: src/sys/arch/xen/xen/xennetback_xenbus.c
diff -u src/sys/arch/xen/xen/xennetback_xenbus.c:1.76 src/sys/arch/xen/xen/xennetback_xenbus.c:1.77
--- src/sys/arch/xen/xen/xennetback_xenbus.c:1.76 Wed Jan 29 05:41:48 2020
+++ src/sys/arch/xen/xen/xennetback_xenbus.c Mon Mar 16 20:49:22 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: xennetback_xenbus.c,v 1.76 2020/01/29 05:41:48 thorpej Exp $ */
+/* $NetBSD: xennetback_xenbus.c,v 1.77 2020/03/16 20:49:22 jdolecek Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -25,7 +25,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: xennetback_xenbus.c,v 1.76 2020/01/29 05:41:48 thorpej Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xennetback_xenbus.c,v 1.77 2020/03/16 20:49:22 jdolecek Exp $");
#include "opt_xen.h"
@@ -90,7 +90,7 @@ struct xni_pkt {
/* pools for xni_pkt */
struct pool xni_pkt_pool;
/* ratecheck(9) for pool allocation failures */
-struct timeval xni_pool_errintvl = { 30, 0 }; /* 30s, each */
+static const struct timeval xni_pool_errintvl = { 30, 0 }; /* 30s, each */
/* state of a xnetback instance */
typedef enum {
@@ -301,7 +301,9 @@ xennetback_xenbus_create(struct xenbus_d
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_snd.ifq_maxlen =
uimax(ifqmaxlen, NET_TX_RING_SIZE * 2);
- ifp->if_capabilities = IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_UDPv4_Tx;
+ ifp->if_capabilities =
+ IFCAP_CSUM_UDPv4_Rx | IFCAP_CSUM_UDPv4_Tx
+ | IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_TCPv4_Tx;
ifp->if_ioctl = xennetback_ifioctl;
ifp->if_start = xennetback_ifstart;
ifp->if_watchdog = xennetback_ifwatchdog;
@@ -857,52 +859,28 @@ xennetback_evthandler(void *arg)
}
}
-#ifdef notyet
/*
- * A lot of work is needed in the tcp stack to handle read-only
- * ext storage so always copy for now.
+ * This is the last TX buffer. Copy the data and
+ * ack it. Delaying it until the mbuf is
+ * freed will stall transmit.
*/
- if (((req_cons + 1) & (NET_TX_RING_SIZE - 1)) ==
- (xneti->xni_txring.rsp_prod_pvt & (NET_TX_RING_SIZE - 1)))
-#else
- if (1)
-#endif /* notyet */
- {
- /*
- * This is the last TX buffer. Copy the data and
- * ack it. Delaying it until the mbuf is
- * freed will stall transmit.
- */
- m->m_len = uimin(MHLEN, txreq.size);
- m->m_pkthdr.len = 0;
- m_copyback(m, 0, txreq.size,
- (void *)(pkt_va + txreq.offset));
- xni_pkt_unmap(pkt, pkt_va);
- if (m->m_pkthdr.len < txreq.size) {
- if_statinc(ifp, if_ierrors);
- m_freem(m);
- xennetback_tx_response(xneti, txreq.id,
- NETIF_RSP_DROPPED);
- continue;
- }
+ m->m_len = uimin(MHLEN, txreq.size);
+ m->m_pkthdr.len = 0;
+ m_copyback(m, 0, txreq.size,
+ (void *)(pkt_va + txreq.offset));
+ xni_pkt_unmap(pkt, pkt_va);
+ if (m->m_pkthdr.len < txreq.size) {
+ if_statinc(ifp, if_ierrors);
+ m_freem(m);
xennetback_tx_response(xneti, txreq.id,
- NETIF_RSP_OKAY);
- } else {
-
- pkt->pkt_id = txreq.id;
- pkt->pkt_xneti = xneti;
-
- MEXTADD(m, pkt_va + txreq.offset,
- txreq.size, M_DEVBUF, xennetback_tx_free, pkt);
- m->m_pkthdr.len = m->m_len = txreq.size;
- m->m_flags |= M_EXT_ROMAP;
+ NETIF_RSP_DROPPED);
+ continue;
}
+ xennetback_tx_response(xneti, txreq.id,
+ NETIF_RSP_OKAY);
+
if ((txreq.flags & NETTXF_csum_blank) != 0) {
- xennet_checksum_fill(&m);
- if (m == NULL) {
- if_statinc(ifp, if_ierrors);
- continue;
- }
+ xennet_checksum_fill(ifp, m);
}
m_set_rcvif(m, ifp);
