1. 23 May, 2018 34 commits
  2. 22 May, 2018 6 commits
    • David S. Miller's avatar
      Merge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queue · 9c803cfd
      David S. Miller authored
      Jeff Kirsher says:
      
      ====================
      40GbE Intel Wired LAN Driver Updates 2018-05-22
      
      This series contains updates to i40e only.
      
      Jake provides all the changes in this series starting with making it
      consistent in how we approach the bit lock.  Fixed the reporting of the
      VEB statistics and the queue statistics to always return every queue
      even if it is not currently in use.  Use WARN_ONCE() so that the first
      time we end up with an incorrect size we will dump a stack trace and a
      message to help highlight the issue early in testing.  Folded the fixed
      string prefix into the stat string definition.  Instead of using a
      separate char *p pointer when copying strings, use the data pointer
      directly.  Added code comments for several of the statistic functions to
      better explain the number and ordering of statistics.
      ====================
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      9c803cfd
    • David S. Miller's avatar
      Merge branch 'tcp-ECN-quickack' · 119768c9
      David S. Miller authored
      Eric Dumazet says:
      
      ====================
      tcp: reduce quickack pressure for ECN
      
      Small patch series changing TCP behavior vs quickack and ECN
      
      First patch is a refactoring, adding parameter to tcp_incr_quickack()
      and tcp_enter_quickack_mode() helpers.
      
      Second patch implements the change, lowering number of ACK packets
      sent after an ECN event.
      ====================
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      119768c9
    • Eric Dumazet's avatar
      tcp: do not aggressively quick ack after ECN events · 522040ea
      Eric Dumazet authored
      ECN signals currently forces TCP to enter quickack mode for
      up to 16 (TCP_MAX_QUICKACKS) following incoming packets.
      
      We believe this is not needed, and only sending one immediate ack
      for the current packet should be enough.
      
      This should reduce the extra load noticed in DCTCP environments,
      after congestion events.
      
      This is part 2 of our effort to reduce pure ACK packets.
      Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
      Acked-by: default avatarSoheil Hassas Yeganeh <soheil@google.com>
      Acked-by: default avatarYuchung Cheng <ycheng@google.com>
      Acked-by: default avatarNeal Cardwell <ncardwell@google.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      522040ea
    • Eric Dumazet's avatar
      tcp: add max_quickacks param to tcp_incr_quickack and tcp_enter_quickack_mode · 9a9c9b51
      Eric Dumazet authored
      We want to add finer control of the number of ACK packets sent after
      ECN events.
      
      This patch is not changing current behavior, it only enables following
      change.
      Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
      Acked-by: default avatarSoheil Hassas Yeganeh <soheil@google.com>
      Acked-by: default avatarNeal Cardwell <ncardwell@google.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      9a9c9b51
    • Vlad Buslov's avatar
      net: sched: don't disable bh when accessing action idr · 290aa0ad
      Vlad Buslov authored
      Initial net_device implementation used ingress_lock spinlock to synchronize
      ingress path of device. This lock was used in both process and bh context.
      In some code paths action map lock was obtained while holding ingress_lock.
      Commit e1e992e5 ("[NET_SCHED] protect action config/dump from irqs")
      modified actions to always disable bh, while using action map lock, in
      order to prevent deadlock on ingress_lock in softirq. This lock was removed
      from net_device, so disabling bh, while accessing action map, is no longer
      necessary.
      
      Replace all action idr spinlock usage with regular calls that do not
      disable bh.
      Signed-off-by: default avatarVlad Buslov <vladbu@mellanox.com>
      Acked-by: default avatarJamal Hadi Salim <jhs@mojatatu.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      290aa0ad
    • David S. Miller's avatar
      Merge branch 'net-ipv6-Fix-route-append-and-replace-use-cases' · 73bf1fc5
      David S. Miller authored
      David Ahern says:
      
      ====================
      net/ipv6: Fix route append and replace use cases
      
      This patch set fixes a few append and replace uses cases for IPv6 and
      adds test cases that codifies the expectations of how append and replace
      are expected to work. In paricular it allows a multipath route to have
      a dev-only nexthop, something Thomas tried to accomplish with commit
      edd7ceb7 ("ipv6: Allow non-gateway ECMP for IPv6") which had to be
      reverted because of breakage, and to replace an existing FIB entry
      with a reject route.
      
      There are a number of inconsistent and surprising aspects to the Linux
      API for adding, deleting, replacing and changing FIB entries. For example,
      with IPv4 NLM_F_APPEND means insert the route after any existing entries
      with the same key (prefix + priority + TOS for IPv4) and NLM_F_CREATE
      without the append flag inserts the new route before any existing entries.
      
      IPv6 on the other hand attempts to guess whether a new route should be
      appended to an existing one, possibly creating a multipath route, or to
      add a new entry after any existing ones. This applies to both the 'append'
      (NLM_F_CREATE + NLM_F_APPEND) and 'prepend' (NLM_F_CREATE only) cases
      meaning for IPv6 the NLM_F_APPEND is basically ignored. This guessing
      whether the route should be added to a multipath route (gateway routes)
      or inserted after existing entries (non-gateway based routes) means a
      multipath route can not have a dev only nexthop (potentially required in
      some cases - tunnels or VRF route leaking for example) and route 'replace'
      is a bit adhoc treating gateway based routes and dev-only / reject routes
      differently.
      
      This has led to frustration with developers working on routing suites
      such as FRR where workarounds such as delete and add are used instead of
      replace.
      
      After this patch set there are 2 differences between IPv4 and IPv6:
      1. 'ip ro prepend' = NLM_F_CREATE only
          IPv4 adds the new route before any existing ones
          IPv6 adds new route after any existing ones
      
      2. 'ip ro append' = NLM_F_CREATE|NLM_F_APPEND
         IPv4 adds the new route after any existing ones
         IPv6 adds the nexthop to existing routes converting to multipath
      
      For the former, there are cases where we want same prefix routes added
      after existing ones (e.g., multicast, prefix routes for macvlan when used
      for virtual router redundancy). Requiring the APPEND flag to add a new
      route to an existing one helps here but is a slight change in behavior
      since prepend with gateway routes now create a separate entry.
      
      For the latter IPv6 behavior is preferred - appending a route for the same
      prefix and metric to make a multipath route, so really IPv4 not allowing an
      existing route to be updated is the limiter. This will be fixed when
      nexthops become separate objects - a future patch set.
      
      Thank you to Thomas and Ido for testing earlier versions of this set, and
      to Ido for providing an update to the mlxsw driver.
      
      Changes since RFC
      - cleanup wording in test script; add comments about expected failures
        and why
      ====================
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      73bf1fc5