1. 30 Oct, 2014 3 commits
  2. 28 Oct, 2014 1 commit
  3. 27 Oct, 2014 17 commits
  4. 25 Oct, 2014 4 commits
    • David S. Miller's avatar
      Merge branch 'sunvnet-napi' · 9286ae01
      David S. Miller authored
      Sowmini Varadhan says:
      
      ====================
      sunvnet: NAPIfy sunvnet
      
      This patchset converts the sunvnet driver to use the NAPI framework.
      Changes since v4 to Patch1:
        vnet_event accumulates LDC_EVENT_* bits into rx_event.
        vnet_event_napi() unrolls send_events() logic to process all rx_event bits.
      Changes since v5:
        Patch 1: use net_device.h definition for NAPI_POLL_WEIGHT.
        Drop sparclinux changes (patch3) per David Miller feedback
      
      Patch 1 in the series addresses the packet-receive path- all
      the vnet_event() processing is moved into NAPI context.
      This patch is dependant on the sparc-next commit:
        "sparc64: Add vio_set_intr() to enable/disable Rx interrupts"
        (sparc commit id ca605b7d)
      
      Patch 2 uses RCU to fix race conditions between vnet_port_remove and
      paths that access/modify port-related state, such as vnet_start_xmit.
      
      Patch 3 leverages from the NAPIfied Rx path,
      dropping superfluous usage of the irqsave/irqrestores on the vio.lock
      where possible.
      ====================
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      9286ae01
    • Sowmini Varadhan's avatar
      sunvnet: Remove irqsave/irqrestore on vio.lock · 13b13dd9
      Sowmini Varadhan authored
      After the  NAPIfication of sunvnet, we no longer need to
      synchronize by doing irqsave/restore on vio.lock in the
      I/O fastpath.
      
      NAPI ->poll() is non-reentrant, so all RX processing occurs
      strictly in a serialized environment. TX reclaim is done in NAPI
      context, so the netif_tx_lock can be used to serialize
      critical sections between Tx and Rx paths.
      Signed-off-by: default avatarSowmini Varadhan <sowmini.varadhan@oracle.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      13b13dd9
    • Sowmini Varadhan's avatar
      sunvnet: Use RCU to synchronize port usage with vnet_port_remove() · 2a968dd8
      Sowmini Varadhan authored
      A vnet_port_remove could be triggered as a result of an ldm-unbind
      operation by the peer, module unload, or other changes to the
      inter-vnet-link configuration.  When this is concurrent with
      vnet_start_xmit(), there are several race sequences possible,
      such as
      
      thread 1                                    thread 2
      vnet_start_xmit
      -> tx_port_find
         spin_lock_irqsave(&vp->lock..)
         ret = __tx_port_find(..)
         spin_lock_irqrestore(&vp->lock..)
                                                 vio_remove -> ..
                                                     ->vnet_port_remove
                                                 spin_lock_irqsave(&vp->lock..)
                                                 cleanup
                                                 spin_lock_irqrestore(&vp->lock..)
                                                 kfree(port)
      /* attempt to use ret will bomb */
      
      This patch adds RCU locking for port access so that vnet_port_remove
      will correctly clean up port-related state.
      Signed-off-by: default avatarSowmini Varadhan <sowmini.varadhan@oracle.com>
      Acked-by: default avatarDwight Engen <dwight.engen@oracle.com>
      Acked-by: default avatarBob Picco <bob.picco@oracle.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      2a968dd8
    • Sowmini Varadhan's avatar
      sunvnet: NAPIfy sunvnet · 69088822
      Sowmini Varadhan authored
      Move Rx packet procssing to the NAPI poll callback.
      Disable VIO interrupt and unconditioanlly go into NAPI
      context from vnet_event.
      
      Note that we want to minimize the number of LDC
      STOP/START messages sent. Specifically, do not send a STOP
      message if vnet_walk_rx does not read all the available descriptors
      because of the NAPI budget limitation. Instead, note the end index
      as part of port state, and resume from this index when the
      next poll callback is triggered.
      Signed-off-by: default avatarSowmini Varadhan <sowmini.varadhan@oracle.com>
      Acked-by: default avatarRaghuram Kothakota <raghuram.kothakota@oracle.com>
      Acked-by: default avatarDwight Engen <dwight.engen@oracle.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      69088822
  5. 24 Oct, 2014 15 commits
    • David S. Miller's avatar
      Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net-next · 132fb579
      David S. Miller authored
      Jeff Kirsher says:
      
      ====================
      Intel Wired LAN Driver Updates 2014-10-23
      
      This series contains updates to i40e and i40evf.
      
      Jesse modifies the i40e driver to only notify the firmware on link up/down
      and qualified module events.  Also simplified the job of managing link
      state by using the admin queue receive event for link events as a signal
      to tell the driver to update link state.
      
      Jeff (me) cleans up the inconsistent use of tabs for indentation in the admin
      queue command header file.
      
      Neerav converts the use of udelay() to usleep_range().
      
      Anjali fixes a bug where receive would stop after some stress by adding
      a sleep and restart as well as moving the setting of flow control because
      it should be done at a PF level and not a VSI level.
      
      Mitch adds code to handle link events when updating the PF switch, which
      allows link information to be properly provided to VFS in all cases.
      
      Catherine adds driver support for 10GBaseT and bumps driver version.
      ====================
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      132fb579
    • Fabian Frederick's avatar
    • Fabian Frederick's avatar
      lapb: move EXPORT_SYMBOL after functions. · 75da1469
      Fabian Frederick authored
      See Documentation/CodingStyle Chapter 6
      Signed-off-by: default avatarFabian Frederick <fabf@skynet.be>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      75da1469
    • David S. Miller's avatar
      Merge branch 'berlin_ethernet' · 5f3619f2
      David S. Miller authored
      Sebastian Hesselbarth says:
      
      ====================
      Marvell PXA168 libphy handling and Berlin Ethernet
      
      This patch series deals with a removing a IP feature that can be found
      on all currently supported Marvell Ethernet IP (pxa168_eth, mv643xx_eth,
      mvneta). The MAC IP allows to automatically perform PHY auto-negotiation
      without software interaction.
      
      However, this feature (a) fundamentally clashes with the way libphy works
      and (b) is unable to deal with quirky PHYs that require special treatment.
      In this series, pxa168_eth driver is rewritten to completely disable that
      feature and properly deal with libphy provided PHYs.
      
      As usual, a branch on top of v3.18-rc1 can be found at
      
      git://git.infradead.org/users/hesselba/linux-berlin.git devel/bg2-bg2cd-eth-v2
      
      Patches 1-5 should go through David's net tree, I'll pick up the DT patches
      6-9.
      
      There have been some changes,
      compared to the RFT
      - added phy-connection-type property to BG2Q PHY DT node
      - bail out from pxa168_eth_adjust_link when there is no change in
        PHY parameters. Also, add a call to phy_print_status.
      compared to v1
      - move phy-connection-type to ethernet node instead of PHY node
      
      Patch 1 adds support for Marvell 88E3016 FastEthernet PHY that is also
      integrated in Marvell Berlin BG2/BG2CD SoCs.
      
      Patch 2 allows to pass phy_interface_t on pxa168_eth platform_data that
      is only used by mach-mmp/gplug. From the board setup, I guessed gplug's
      PHY is connected via RMII. The patch still isn't even compile tested.
      
      Patches 3-5 prepare proper libphy handling and finally remove all in-driver
      PHY mangling related to the feature explained above.
      
      Patches 6-9 add corresponding ethernet DT nodes to BG2, BG2CD, add a
      phy-connection-type property to BG2Q and enable ethernet on BG2-based Sony
      NSZ-GS7. I have tested all this on GS7 successfully with ip=dhcp on 100M FD.
      ====================
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      5f3619f2
    • Sebastian Hesselbarth's avatar
      net: pxa168_eth: Remove in-driver PHY mangling · 9ff32fe1
      Sebastian Hesselbarth authored
      With properly using libphy PHYs now, remove the in-driver PHY
      mangling.
      Tested-by: default avatarAntoine Ténart <antoine.tenart@free-electrons.com>
      Reviewed-by: default avatarFlorian Fainelli <f.fainelli@gmail.com>
      Signed-off-by: default avatarSebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      9ff32fe1
    • Sebastian Hesselbarth's avatar
      net: pxa168_eth: Remove HW auto-negotiaion · 1a149132
      Sebastian Hesselbarth authored
      Marvell Ethernet IP supports PHY negotiation driven by HW. This
      fundamentally clashes with libphy (software) driven negotiation and
      also cannot cope with quirky PHYs. Therefore, always disable any HW
      negotiation features and properly use libphy's phy_device.
      Tested-by: default avatarAntoine Ténart <antoine.tenart@free-electrons.com>
      Reviewed-by: default avatarFlorian Fainelli <f.fainelli@gmail.com>
      Signed-off-by: default avatarSebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      1a149132
    • Sebastian Hesselbarth's avatar
      net: pxa168_eth: Prepare proper libphy handling · 9d8ea73d
      Sebastian Hesselbarth authored
      Current libphy handling in pxa168_eth lacks proper phy_connect. Prepare
      to fix this by first moving phy properties from platform_data to private
      driver data.
      Tested-by: default avatarAntoine Ténart <antoine.tenart@free-electrons.com>
      Reviewed-by: default avatarFlorian Fainelli <f.fainelli@gmail.com>
      Signed-off-by: default avatarSebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      9d8ea73d
    • Sebastian Hesselbarth's avatar
      net: pxa168_eth: Provide phy_interface mode on platform_data · e7de17ab
      Sebastian Hesselbarth authored
      The PXA168 Ethernet IP support MII and RMII connection to its PHY.
      Currently, pxa168 platform_data does not provide a way to pass that
      and there is one user of pxa168 platform_data (mach-mmp/gplug).
      Given the pinctrl settings of gplug it uses RMII, so add and pass
      a corresponding phy_interface_t.
      Tested-by: default avatarAntoine Ténart <antoine.tenart@free-electrons.com>
      Reviewed-by: default avatarFlorian Fainelli <f.fainelli@gmail.com>
      Signed-off-by: default avatarSebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      e7de17ab
    • Sebastian Hesselbarth's avatar
      phy: marvell: Add support for 88E3016 FastEthernet PHY · 6b358aed
      Sebastian Hesselbarth authored
      Marvell 88E3016 is a FastEthernet PHY that also can be found in Marvell
      Berlin SoCs as integrated PHY.
      Tested-by: default avatarAntoine Ténart <antoine.tenart@free-electrons.com>
      Reviewed-by: default avatarFlorian Fainelli <f.fainelli@gmail.com>
      Signed-off-by: default avatarSebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      6b358aed
    • Geert Uytterhoeven's avatar
      natsemi/macsonic: Remove superfluous interrupt disable/restore · d4c3363e
      Geert Uytterhoeven authored
      As of commit e4dc601b ("m68k: Disable/restore interrupts in
      hwreg_present()/hwreg_write()"), this is no longer needed.
      Signed-off-by: default avatarGeert Uytterhoeven <geert@linux-m68k.org>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      d4c3363e
    • Geert Uytterhoeven's avatar
      cirrus/mac89x0: Remove superfluous interrupt disable/restore · 7f30b742
      Geert Uytterhoeven authored
      As of commit e4dc601b ("m68k: Disable/restore interrupts in
      hwreg_present()/hwreg_write()"), this is no longer needed.
      Signed-off-by: default avatarGeert Uytterhoeven <geert@linux-m68k.org>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      7f30b742
    • Rasmus Villemoes's avatar
      net: typhoon: Remove redundant casts · 00fd5d94
      Rasmus Villemoes authored
      Both image_data and typhoon_fw->data are const u8*, so the cast to u8*
      is unnecessary and confusing.
      Signed-off-by: default avatarRasmus Villemoes <linux@rasmusvillemoes.dk>
      Acked-by: default avatarDavid Dillow <dave@thedillows.org>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      00fd5d94
    • Sébastien Barré's avatar
      Removed unused function sctp_addr_is_valid() · 16704b12
      Sébastien Barré authored
      sctp_addr_is_valid() only appeared in its definition.
      Acked-by: default avatarNeil Horman <nhorman@tuxdriver.com>
      Signed-off-by: default avatarSébastien Barré <sebastien.barre@uclouvain.be>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      16704b12
    • David S. Miller's avatar
      Merge branch 'ipv6_route' · fad71e4a
      David S. Miller authored
      Martin KaFai Lau says:
      
      ====================
      ipv6: Reduce the number of fib6_lookup() calls from ip6_pol_route()
      
      This patch set is trying to reduce the number of fib6_lookup()
      calls from ip6_pol_route().
      
      I have adapted davem's udpflooda and kbench_mod test
      (https://git.kernel.org/pub/scm/linux/kernel/git/davem/net_test_tools.git) to
      support IPv6 and here is the result:
      
      Before:
      [root]# for i in $(seq 1 3); do time ./udpflood -l 20000000 -c 250 2401:face:face:face::2; done
      
      real    0m34.190s
      user    0m3.047s
      sys     0m31.108s
      
      real    0m34.635s
      user    0m3.125s
      sys     0m31.475s
      
      real    0m34.517s
      user    0m3.034s
      sys     0m31.449s
      
      [root]# insmod ip6_route_kbench.ko oif=2 src=2401:face:face:face::1 dst=2401:face:face:face::2
      [  660.160976] ip6_route_kbench: ip6_route_output tdiff: 933
      [  660.207261] ip6_route_kbench: ip6_route_output tdiff: 988
      [  660.253492] ip6_route_kbench: ip6_route_output tdiff: 896
      [  660.298862] ip6_route_kbench: ip6_route_output tdiff: 898
      
      After:
      [root]# for i in $(seq 1 3); do time ./udpflood -l 20000000 -c 250 2401:face:face:face::2; done
      
      real    0m32.695s
      user    0m2.925s
      sys     0m29.737s
      
      real    0m32.636s
      user    0m3.007s
      sys     0m29.596s
      
      real    0m32.797s
      user    0m2.866s
      sys     0m29.898s
      
      [root]# insmod ip6_route_kbench.ko oif=2 src=2401:face:face:face::1 dst=2401:face:face:face::2
      [  881.220793] ip6_route_kbench: ip6_route_output tdiff: 684
      [  881.253477] ip6_route_kbench: ip6_route_output tdiff: 640
      [  881.286867] ip6_route_kbench: ip6_route_output tdiff: 630
      [  881.320749] ip6_route_kbench: ip6_route_output tdiff: 653
      
      /****************************** udpflood.c ******************************/
      /* It is an adaptation of the Eric Dumazet's and David Miller's
       * udpflood tool, by adding IPv6 support.
       */
      
      typedef uint32_t u32;
      
      static int debug =3D 0;
      
      /* Allow -fstrict-aliasing */
      typedef union sa_u {
      	struct sockaddr_storage a46;
      	struct sockaddr_in a4;
      	struct sockaddr_in6 a6;
      } sa_u;
      
      static int usage(void)
      {
      	printf("usage: udpflood [ -l count ] [ -m message_size ] [ -c num_ip_addrs=
       ] IP_ADDRESS\n");
      	return -1;
      }
      
      static u32 get_last32h(const sa_u *sa)
      {
      	if (sa->a46.ss_family =3D=3D PF_INET)
      		return ntohl(sa->a4.sin_addr.s_addr);
      	else
      		return ntohl(sa->a6.sin6_addr.s6_addr32[3]);
      }
      
      static void set_last32h(sa_u *sa, u32 last32h)
      {
      	if (sa->a46.ss_family =3D=3D PF_INET)
      		sa->a4.sin_addr.s_addr =3D htonl(last32h);
      	else
      		sa->a6.sin6_addr.s6_addr32[3] =3D htonl(last32h);
      }
      
      static void print_saddr(const sa_u *sa, const char *msg)
      {
      	char buf[64];
      
      	if (!debug)
      		return;
      
      	switch (sa->a46.ss_family) {
      	case PF_INET:
      		inet_ntop(PF_INET, &(sa->a4.sin_addr.s_addr), buf,
      			  sizeof(buf));
      		break;
      	case PF_INET6:
      		inet_ntop(PF_INET6, &(sa->a6.sin6_addr), buf, sizeof(buf));
      		break;
      	}
      
      	printf("%s: %s\n", msg, buf);
      }
      
      static int send_packets(const sa_u *sa, size_t num_addrs, int count, int ms=
      g_sz)
      {
      	char *msg =3D malloc(msg_sz);
      	sa_u saddr;
      	u32 start_addr32h, end_addr32h, cur_addr32h;
      	int fd, i, err;
      
      	if (!msg)
      		return -ENOMEM;
      
      	memset(msg, 0, msg_sz);
      
      	memcpy(&saddr, sa, sizeof(saddr));
      	cur_addr32h =3D start_addr32h =3D get_last32h(&saddr);
      	end_addr32h =3D start_addr32h + num_addrs;
      
      	fd =3D socket(saddr.a46.ss_family, SOCK_DGRAM, 0);
      	if (fd < 0) {
      		perror("socket");
      		err =3D fd;
      		goto out_nofd;
      	}
      
      	/* connect to avoid the kernel spending time in figuring
      	 * out the source address (i.e pin the src address)
      	 */
      	err =3D connect(fd, (struct sockaddr *) &saddr, sizeof(saddr));
      	if (err < 0) {
      		perror("connect");
      		goto out;
      	}
      
      	print_saddr(&saddr, "start_addr");
      	for (i =3D 0; i < count; i++) {
      		print_saddr(&saddr, "sendto");
      		err =3D sendto(fd, msg, msg_sz, 0, (struct sockaddr *)&saddr,
      			     sizeof(saddr));
      		if (err < 0) {
      			perror("sendto");
      			goto out;
      		}
      
      		if (++cur_addr32h >=3D end_addr32h)
      			cur_addr32h =3D start_addr32h;
      		set_last32h(&saddr, cur_addr32h);
      	}
      
      	err =3D 0;
      out:
      	close(fd);
      out_nofd:
      	free(msg);
      	return err;
      }
      
      int main(int argc, char **argv, char **envp)
      {
      	int port, msg_sz, count, num_addrs, ret;
      
      	sa_u start_addr;
      
      	port =3D 6000;
      	msg_sz =3D 32;
      	count =3D 10000000;
      	num_addrs =3D 1;
      
      	while ((ret =3D getopt(argc, argv, "dl:s:p:c:")) >=3D 0) {
      		switch (ret) {
      		case 'l':
      			sscanf(optarg, "%d", &count);
      			break;
      		case 's':
      			sscanf(optarg, "%d", &msg_sz);
      			break;
      		case 'p':
      			sscanf(optarg, "%d", &port);
      			break;
      		case 'c':
      			sscanf(optarg, "%d", &num_addrs);
      			break;
      		case 'd':
      			debug =3D 1;
      			break;
      		case '?':
      			return usage();
      		}
      	}
      
      	if (num_addrs < 1)
      		return usage();
      
      	if (!argv[optind])
      		return usage();
      
      	start_addr.a4.sin_port =3D htons(port);
      	if (inet_pton(PF_INET, argv[optind], &start_addr.a4.sin_addr))
      		start_addr.a46.ss_family =3D PF_INET;
      	else if (inet_pton(PF_INET6, argv[optind], &start_addr.a6.sin6_addr.s6_add=
      r))
      		start_addr.a46.ss_family =3D PF_INET6;
      	else
      		return usage();
      
      	return send_packets(&start_addr, num_addrs, count, msg_sz);
      }
      
      /****************** ip6_route_kbench_mod.c ******************/
      
      /* We can't just use "get_cycles()" as on some platforms, such
       * as sparc64, that gives system cycles rather than cpu clock
       * cycles.
       */
      
      static inline unsigned long long get_tick(void)
      {
      	unsigned long long t;
      
      	__asm__ __volatile__("rd %%tick, %0" : "=r" (t));
      	return t;
      }
      static inline unsigned long long get_tick(void)
      {
      	unsigned long long t;
      
      	rdtscll(t);
      
      	return t;
      }
      static inline unsigned long long get_tick(void)
      {
      	return get_cycles();
      }
      
      static int flow_oif = DEFAULT_OIF;
      static int flow_iif = DEFAULT_IIF;
      static u32 flow_mark = DEFAULT_MARK;
      static struct in6_addr flow_dst_ip_addr;
      static struct in6_addr flow_src_ip_addr;
      static int flow_tos = DEFAULT_TOS;
      
      static char dst_string[64];
      static char src_string[64];
      
      module_param_string(dst, dst_string, sizeof(dst_string), 0);
      module_param_string(src, src_string, sizeof(src_string), 0);
      
      static int __init flow_setup(void)
      {
      	if (dst_string[0] &&
      	    !in6_pton(dst_string, -1, &flow_dst_ip_addr.s6_addr[0], -1, NULL)) {
      		pr_info("cannot parse \"%s\"\n", dst_string);
      		return -1;
      	}
      
      	if (src_string[0] &&
      	    !in6_pton(src_string, -1, &flow_src_ip_addr.s6_addr[0], -1, NULL)) {
      		pr_info("cannot parse \"%s\"\n", dst_string);
      		return -1;
      	}
      
      	return 0;
      }
      
      module_param_named(oif, flow_oif, int, 0);
      module_param_named(iif, flow_iif, int, 0);
      module_param_named(mark, flow_mark, uint, 0);
      module_param_named(tos, flow_tos, int, 0);
      
      static int warmup_count = DEFAULT_WARMUP_COUNT;
      module_param_named(count, warmup_count, int, 0);
      
      static void flow_init(struct flowi6 *fl6)
      {
      	memset(fl6, 0, sizeof(*fl6));
      	fl6->flowi6_proto = IPPROTO_ICMPV6;
      	fl6->flowi6_oif = flow_oif;
      	fl6->flowi6_iif = flow_iif;
      	fl6->flowi6_mark = flow_mark;
      	fl6->flowi6_tos = flow_tos;
      	fl6->daddr = flow_dst_ip_addr;
      	fl6->saddr = flow_src_ip_addr;
      }
      
      static struct sk_buff * fake_skb_get(void)
      {
      	struct ipv6hdr *hdr;
      	struct sk_buff *skb;
      
      	skb = alloc_skb(4096, GFP_KERNEL);
      	if (!skb) {
      		pr_info("Cannot alloc SKB for test\n");
      		return NULL;
      	}
      	skb->dev = __dev_get_by_index(&init_net, flow_iif);
      	if (skb->dev == NULL) {
      		pr_info("Input device (%d) does not exist\n", flow_iif);
      		goto err;
      	}
      
      	skb_reset_mac_header(skb);
      	skb_reset_network_header(skb);
      	skb_reserve(skb, MAX_HEADER + sizeof(struct ipv6hdr));
      	hdr = ipv6_hdr(skb);
      
      	hdr->priority = 0;
      	hdr->version = 6;
      	memset(hdr->flow_lbl, 0, sizeof(hdr->flow_lbl));
      	hdr->payload_len = htons(sizeof(struct icmp6hdr));
      	hdr->nexthdr = IPPROTO_ICMPV6;
      	hdr->saddr = flow_src_ip_addr;
      	hdr->daddr = flow_dst_ip_addr;
      	skb->protocol = htons(ETH_P_IPV6);
      	skb->mark = flow_mark;
      
      	return skb;
      err:
      	kfree_skb(skb);
      	return NULL;
      }
      
      static void do_full_output_lookup_bench(void)
      {
      	unsigned long long t1, t2, tdiff;
      	struct rt6_info *rt;
      	struct flowi6 fl6;
      	int i;
      
      	rt = NULL;
      
      	for (i = 0; i < warmup_count; i++) {
      		flow_init(&fl6);
      
      		rt = (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl6);
      		if (IS_ERR(rt))
      			break;
      		ip6_rt_put(rt);
      	}
      	if (IS_ERR(rt)) {
      		pr_info("ip_route_output_key: err=%ld\n", PTR_ERR(rt));
      		return;
      	}
      
      	flow_init(&fl6);
      
      	t1 = get_tick();
      	rt = (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl6);
      	t2 = get_tick();
      	if (!IS_ERR(rt))
      		ip6_rt_put(rt);
      
      	tdiff = t2 - t1;
      	pr_info("ip6_route_output tdiff: %llu\n", tdiff);
      }
      
      static void do_full_input_lookup_bench(void)
      {
      	unsigned long long t1, t2, tdiff;
      	struct sk_buff *skb;
      	struct rt6_info *rt;
      	int err, i;
      
      	skb = fake_skb_get();
      	if (skb == NULL)
      		goto out_free;
      
      	err = 0;
      	local_bh_disable();
      	for (i = 0; i < warmup_count; i++) {
      		ip6_route_input(skb);
      		rt = (struct rt6_info *)skb_dst(skb);
      		err = (!rt || rt == init_net.ipv6.ip6_null_entry);
      		skb_dst_drop(skb);
      		if (err)
      			break;
      	}
      	local_bh_enable();
      
      	if (err) {
      		pr_info("Input route lookup fails\n");
      		goto out_free;
      	}
      
      	local_bh_disable();
      	t1 = get_tick();
      	ip6_route_input(skb);
      	t2 = get_tick();
      	local_bh_enable();
      
      	rt = (struct rt6_info *)skb_dst(skb);
      	err = (!rt || rt == init_net.ipv6.ip6_null_entry);
      	skb_dst_drop(skb);
      	if (err) {
      		pr_info("Input route lookup fails\n");
      		goto out_free;
      	}
      
      	tdiff = t2 - t1;
      	pr_info("ip6_route_input tdiff: %llu\n", tdiff);
      
      out_free:
      	kfree_skb(skb);
      }
      
      static void do_full_lookup_bench(void)
      {
      	if (!flow_iif)
      		do_full_output_lookup_bench();
      	else
      		do_full_input_lookup_bench();
      }
      
      static void do_bench(void)
      {
      	do_full_lookup_bench();
      	do_full_lookup_bench();
      	do_full_lookup_bench();
      	do_full_lookup_bench();
      }
      
      static int __init kbench_init(void)
      {
      	if (flow_setup())
      		return -EINVAL;
      
      	pr_info("flow [IIF(%d),OIF(%d),MARK(0x%08x),D("IP6_FMT"),"
      		"S("IP6_FMT"),TOS(0x%02x)]\n",
      		flow_iif, flow_oif, flow_mark,
      		IP6_PRT(flow_dst_ip_addr),
      		IP6_PRT(flow_src_ip_addr),
      		flow_tos);
      
      	if (!cpu_has_tsc) {
      		pr_err("X86 TSC is required, but is unavailable.\n");
      		return -EINVAL;
      	}
      
      	pr_info("sizeof(struct rt6_info)==%zu\n", sizeof(struct rt6_info));
      
      	do_bench();
      
      	return -ENODEV;
      }
      
      static void __exit kbench_exit(void)
      {
      }
      
      module_init(kbench_init);
      module_exit(kbench_exit);
      MODULE_LICENSE("GPL");
      ====================
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      fad71e4a
    • Martin KaFai Lau's avatar
      ipv6: Avoid redoing fib6_lookup() with reachable = 0 by saving fn · 367efcb9
      Martin KaFai Lau authored
      This patch save the fn before doing rt6_backtrack.
      Hence, without redo-ing the fib6_lookup(), saved_fn can be used
      to redo rt6_select() with RT6_LOOKUP_F_REACHABLE off.
      
      Some minor changes I think make sense to review as a single patch:
      * Remove the 'out:' goto label.
      * Remove the 'reachable' variable. Only use the 'strict' variable instead.
      
      After this patch, "failing ip6_ins_rt()" should be the only case that
      requires a redo of fib6_lookup().
      
      Cc: David Miller <davem@davemloft.net>
      Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
      Signed-off-by: default avatarMartin KaFai Lau <kafai@fb.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      367efcb9