From: jan <jan@openbsd.org>
Date: Mon, 27 Jun 2022 15:11:23 +0000 (+0000)
Subject: Introduce Large Receive Offloading of TCP segment offloading for ix(4).  It is
X-Git-Url: http://artulab.com/gitweb/?a=commitdiff_plain;h=bde39c359d9e72963516d6eecdca33fbe33904e4;p=openbsd

Introduce Large Receive Offloading of TCP segment offloading for ix(4).  It is
disabled by default.  Also add a tso option to ifconfig(8) to enable and
disable this feature.

ok deraadt
---

diff --git a/sbin/ifconfig/ifconfig.8 b/sbin/ifconfig/ifconfig.8
index cc5a6995cdd..4daccdb1d69 100644
--- a/sbin/ifconfig/ifconfig.8
+++ b/sbin/ifconfig/ifconfig.8
@@ -1,4 +1,4 @@
-.\"	$OpenBSD: ifconfig.8,v 1.382 2022/05/14 07:22:10 denis Exp $
+.\"	$OpenBSD: ifconfig.8,v 1.383 2022/06/27 15:11:23 jan Exp $
 .\"	$NetBSD: ifconfig.8,v 1.11 1996/01/04 21:27:29 pk Exp $
 .\"     $FreeBSD: ifconfig.8,v 1.16 1998/02/01 07:03:29 steve Exp $
 .\"
@@ -31,7 +31,7 @@
 .\"
 .\"     @(#)ifconfig.8	8.4 (Berkeley) 6/1/94
 .\"
-.Dd $Mdocdate: May 14 2022 $
+.Dd $Mdocdate: June 27 2022 $
 .Dt IFCONFIG 8
 .Os
 .Sh NAME
@@ -279,6 +279,8 @@ tag.
 As CSUM_TCPv4, but supports IPv6 datagrams.
 .It Sy CSUM_UDPv6
 As above, for UDP.
+.It Sy TSO
+The device supports TCP segment offloading (TSO).
 .It Sy WOL
 The device supports Wake on LAN (WoL).
 .It Sy hardmtu
@@ -486,6 +488,22 @@ Query and display information and diagnostics from GBIC and SFP
 modules installed in an interface.
 It is only supported by drivers implementing the necessary functionality
 on hardware which supports it.
+.It Cm tso
+Enable TCP segmentation offloading (TSO) if its supported by the hardware; see
+.Cm hwfeatures .
+TSO enabled NICs modify received TCP/IP packets.
+This will also affect traffic of upper layer interfaces
+.Po
+e.g.,
+.Xr vlan 4 ,
+.Xr aggr 4 ,
+.Xr carp 4 ,
+etc.
+.Pc .
+Changing this option will re-initialize the network interface.
+.It Cm -tso
+Disable TSO.
+TSO is disabled by default.
 .It Cm up
 Mark an interface
 .Dq up .
diff --git a/sbin/ifconfig/ifconfig.c b/sbin/ifconfig/ifconfig.c
index ed05d8f2659..ea3b29dbe2f 100644
--- a/sbin/ifconfig/ifconfig.c
+++ b/sbin/ifconfig/ifconfig.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ifconfig.c,v 1.454 2022/04/07 16:41:13 naddy Exp $	*/
+/*	$OpenBSD: ifconfig.c,v 1.455 2022/06/27 15:11:23 jan Exp $	*/
 /*	$NetBSD: ifconfig.c,v 1.40 1997/10/01 02:19:43 enami Exp $	*/
 
 /*
@@ -126,7 +126,7 @@
 #define HWFEATURESBITS							\
 	"\024\1CSUM_IPv4\2CSUM_TCPv4\3CSUM_UDPv4"			\
 	"\5VLAN_MTU\6VLAN_HWTAGGING\10CSUM_TCPv6"			\
-	"\11CSUM_UDPv6\20WOL"
+	"\11CSUM_UDPv6\17TSO\20WOL"
 
 struct ifencap {
 	unsigned int	 ife_flags;
@@ -470,6 +470,8 @@ const struct	cmd {
 	{ "-soii",	IFXF_INET6_NOSOII,	0,	setifxflags },
 	{ "monitor",	IFXF_MONITOR,	0,		setifxflags },
 	{ "-monitor",	-IFXF_MONITOR,	0,		setifxflags },
+	{ "tso",	IFXF_TSO,	0,		setifxflags },
+	{ "-tso",	-IFXF_TSO,	0,		setifxflags },
 #ifndef SMALL
 	{ "hwfeatures", NEXTARG0,	0,		printifhwfeatures },
 	{ "metric",	NEXTARG,	0,		setifmetric },
@@ -673,7 +675,7 @@ const struct	cmd {
 	"\7RUNNING\10NOARP\11PROMISC\12ALLMULTI\13OACTIVE\14SIMPLEX"	\
 	"\15LINK0\16LINK1\17LINK2\20MULTICAST"				\
 	"\23AUTOCONF6TEMP\24MPLS\25WOL\26AUTOCONF6\27INET6_NOSOII"	\
-	"\30AUTOCONF4" "\31MONITOR"
+	"\30AUTOCONF4" "\31MONITOR" "\32TSO"
 
 int	getinfo(struct ifreq *, int);
 void	getsock(int);
diff --git a/sys/dev/pci/if_ix.c b/sys/dev/pci/if_ix.c
index 924b605c742..cb233034d23 100644
--- a/sys/dev/pci/if_ix.c
+++ b/sys/dev/pci/if_ix.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_ix.c,v 1.185 2022/03/15 11:22:10 jan Exp $	*/
+/*	$OpenBSD: if_ix.c,v 1.186 2022/06/27 15:11:23 jan Exp $	*/
 
 /******************************************************************************
 
@@ -160,8 +160,6 @@ void	ixgbe_dma_free(struct ix_softc *, struct ixgbe_dma_alloc *);
 static int
 	ixgbe_tx_ctx_setup(struct tx_ring *, struct mbuf *, uint32_t *,
 	    uint32_t *);
-int	ixgbe_tso_setup(struct tx_ring *, struct mbuf *, uint32_t *,
-	    uint32_t *);
 void	ixgbe_set_ivar(struct ix_softc *, uint8_t, uint8_t, int8_t);
 void	ixgbe_configure_ivars(struct ix_softc *);
 uint8_t	*ixgbe_mc_array_itr(struct ixgbe_hw *, uint8_t **, uint32_t *);
@@ -1924,6 +1922,9 @@ ixgbe_setup_interface(struct ix_softc *sc)
 	ifp->if_capabilities |= IFCAP_CSUM_TCPv6 | IFCAP_CSUM_UDPv6;
 	ifp->if_capabilities |= IFCAP_CSUM_IPv4;
 
+	if (sc->hw.mac.type != ixgbe_mac_82598EB)
+		ifp->if_capabilities |= IFCAP_TSO;
+
 	/*
 	 * Specify the media types supported by this sc and register
 	 * callbacks to update media and link information
@@ -2868,9 +2869,10 @@ fail:
 void
 ixgbe_initialize_receive_units(struct ix_softc *sc)
 {
+	struct ifnet	*ifp = &sc->arpcom.ac_if;
 	struct rx_ring	*rxr = sc->rx_rings;
 	struct ixgbe_hw	*hw = &sc->hw;
-	uint32_t	bufsz, fctrl, srrctl, rxcsum;
+	uint32_t	bufsz, fctrl, srrctl, rxcsum, rdrxctl;
 	uint32_t	hlreg;
 	int		i;
 
@@ -2894,6 +2896,19 @@ ixgbe_initialize_receive_units(struct ix_softc *sc)
 	hlreg |= IXGBE_HLREG0_JUMBOEN;
 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg);
 
+	if (ISSET(ifp->if_xflags, IFXF_TSO)) {
+		rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
+
+		/* This field has to be set to zero. */
+		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
+
+		/* Enable TSO Receive Offloading */
+		rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
+		rdrxctl |= IXGBE_RDRXCTL_FCOE_WRFIX;
+
+		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
+	}
+
 	bufsz = (sc->rx_mbuf_sz - ETHER_ALIGN) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
 
 	for (i = 0; i < sc->num_queues; i++, rxr++) {
@@ -2910,6 +2925,16 @@ ixgbe_initialize_receive_units(struct ix_softc *sc)
 		srrctl = bufsz | IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
 		IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(i), srrctl);
 
+		if (ISSET(ifp->if_xflags, IFXF_TSO)) {
+			rdrxctl = IXGBE_READ_REG(&sc->hw, IXGBE_RSCCTL(i));
+
+			/* Enable TSO Receive Side Coalescing */
+			rdrxctl |= IXGBE_RSCCTL_RSCEN;
+			rdrxctl |= IXGBE_RSCCTL_MAXDESC_16;
+
+			IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(i), rdrxctl);
+		}
+
 		/* Setup the HW Rx Head and Tail Descriptor Pointers */
 		IXGBE_WRITE_REG(hw, IXGBE_RDH(i), 0);
 		IXGBE_WRITE_REG(hw, IXGBE_RDT(i), 0);
@@ -3175,7 +3200,7 @@ ixgbe_rxeof(struct rx_ring *rxr)
 			sendmp = mp;
 			sendmp->m_pkthdr.len = mp->m_len;
 #if NVLAN > 0
-			if (staterr & IXGBE_RXD_STAT_VP) {
+			if (sc->vlan_stripping && staterr & IXGBE_RXD_STAT_VP) {
 				sendmp->m_pkthdr.ether_vtag = vtag;
 				sendmp->m_flags |= M_VLANTAG;
 			}
@@ -3248,8 +3273,20 @@ ixgbe_rx_checksum(uint32_t staterr, struct mbuf * mp, uint32_t ptype)
 void
 ixgbe_setup_vlan_hw_support(struct ix_softc *sc)
 {
-	uint32_t	ctrl;
-	int		i;
+	struct ifnet	*ifp = &sc->arpcom.ac_if;
+	uint32_t	 ctrl;
+	int		 i;
+
+	/*
+	 * We have to disable VLAN striping when using TCP offloading, due to a
+	 * firmware bug.
+	 */
+	if (ISSET(ifp->if_xflags, IFXF_TSO)) {
+		sc->vlan_stripping = 0;
+		return;
+	}
+
+	sc->vlan_stripping = 1;
 
 	/*
 	 * A soft reset zero's out the VFTA, so
diff --git a/sys/dev/pci/if_ix.h b/sys/dev/pci/if_ix.h
index fdae408fdea..41d756110c7 100644
--- a/sys/dev/pci/if_ix.h
+++ b/sys/dev/pci/if_ix.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_ix.h,v 1.44 2022/01/09 05:42:52 jsg Exp $	*/
+/*	$OpenBSD: if_ix.h,v 1.45 2022/06/27 15:11:23 jan Exp $	*/
 
 /******************************************************************************
 
@@ -225,6 +225,7 @@ struct ix_softc {
 	struct ifmedia		media;
 	struct intrmap		*sc_intrmap;
 	int			if_flags;
+	int			vlan_stripping;
 
 	uint16_t		num_vlans;
 	uint16_t		num_queues;
diff --git a/sys/net/if.c b/sys/net/if.c
index 58af5730cde..c12943d6af9 100644
--- a/sys/net/if.c
+++ b/sys/net/if.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if.c,v 1.653 2022/06/07 22:18:34 sashan Exp $	*/
+/*	$OpenBSD: if.c,v 1.654 2022/06/27 15:11:23 jan Exp $	*/
 /*	$NetBSD: if.c,v 1.35 1996/05/07 05:26:04 thorpej Exp $	*/
 
 /*
@@ -2019,6 +2019,38 @@ ifioctl(struct socket *so, u_long cmd, caddr_t data, struct proc *p)
 			ifr->ifr_flags &= ~IFXF_WOL;
 			error = ENOTSUP;
 		}
+
+		if (ISSET(ifp->if_capabilities, IFCAP_TSO) &&
+		    ISSET(ifr->ifr_flags, IFXF_TSO) !=
+		    ISSET(ifp->if_xflags, IFXF_TSO)) {
+			struct ifreq ifrq;
+
+			s = splnet();
+
+			if (ISSET(ifr->ifr_flags, IFXF_TSO))
+				ifp->if_xflags |= IFXF_TSO;
+			else
+				ifp->if_xflags &= ~IFXF_TSO;
+
+			NET_ASSERT_LOCKED();	/* for ioctl */
+			KERNEL_ASSERT_LOCKED();	/* for if_flags */
+
+			if (ISSET(ifp->if_flags, IFF_UP)) {
+				/* go down for a moment... */
+				ifp->if_flags &= ~IFF_UP;
+				ifrq.ifr_flags = ifp->if_flags;
+				(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS,
+				    (caddr_t)&ifrq);
+
+				/* ... and up again */
+				ifp->if_flags |= IFF_UP;
+				ifrq.ifr_flags = ifp->if_flags;
+				(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS,
+				    (caddr_t)&ifrq);
+			}
+
+			splx(s);
+		}
 #endif
 
 		if (error == 0)
diff --git a/sys/net/if.h b/sys/net/if.h
index 42d0328480e..278679be2ef 100644
--- a/sys/net/if.h
+++ b/sys/net/if.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if.h,v 1.208 2021/03/11 19:53:40 florian Exp $	*/
+/*	$OpenBSD: if.h,v 1.209 2022/06/27 15:11:23 jan Exp $	*/
 /*	$NetBSD: if.h,v 1.23 1996/05/07 02:40:27 thorpej Exp $	*/
 
 /*
@@ -231,6 +231,7 @@ struct if_status_description {
 #define IFXF_INET6_NOSOII	0x40	/* [N] don't do RFC 7217 */
 #define	IFXF_AUTOCONF4		0x80	/* [N] v4 autoconf (aka dhcp) enabled */
 #define	IFXF_MONITOR		0x100	/* [N] only used for bpf */
+#define	IFXF_TSO		0x200	/* [N] TCP segment offloading */
 
 #define	IFXF_CANTCHANGE \
 	(IFXF_MPSAFE|IFXF_CLONED)
@@ -250,6 +251,7 @@ struct if_status_description {
 #define	IFCAP_VLAN_HWTAGGING	0x00000020	/* hardware VLAN tag support */
 #define	IFCAP_CSUM_TCPv6	0x00000080	/* can do IPv6/TCP checksums */
 #define	IFCAP_CSUM_UDPv6	0x00000100	/* can do IPv6/UDP checksums */
+#define	IFCAP_TSO		0x00004000	/* TCP segment offloading */
 #define	IFCAP_WOL		0x00008000	/* can do wake on lan */
 
 #define IFCAP_CSUM_MASK		(IFCAP_CSUM_IPv4 | IFCAP_CSUM_TCPv4 | \