1. 14 Jan, 2013 2 commits
  2. 13 Jan, 2013 9 commits
  3. 12 Jan, 2013 15 commits
  4. 11 Jan, 2013 6 commits
  5. 10 Jan, 2013 8 commits
    • Alexander Duyck's avatar
      net: Add support for XPS without sysfs being defined · 024e9679
      Alexander Duyck authored
      
      
      This patch makes it so that we can support transmit packet steering without
      sysfs needing to be enabled.  The reason for making this change is to make
      it so that a driver can make use of the XPS even while the sysfs portion of
      the interface is not present.
      Signed-off-by: default avatarAlexander Duyck <alexander.h.duyck@intel.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      024e9679
    • Alexander Duyck's avatar
      net: Rewrite netif_set_xps_queues to address several issues · 01c5f864
      Alexander Duyck authored
      
      
      This change is meant to address several issues I found within the
      netif_set_xps_queues function.
      
      If the allocation of one of the maps to be assigned to new_dev_maps failed
      we could end up with the device map in an inconsistent state since we had
      already worked through a number of CPUs and removed or added the queue.  To
      address that I split the process into several steps.  The first of which is
      just the allocation of updated maps for CPUs that will need larger maps to
      store the queue.  By doing this we can fail gracefully without actually
      altering the contents of the current device map.
      
      The second issue I found was the fact that we were always allocating a new
      device map even if we were not adding any queues.  I have updated the code
      so that we only allocate a new device map if we are adding queues,
      otherwise if we are not adding any queues to CPUs we just skip to the
      removal process.
      
      The last change I made was to reuse the code from remove_xps_queue to remove
      the queue from the CPU.  By making this change we can be consistent in how
      we go about adding and removing the queues from the CPUs.
      Signed-off-by: default avatarAlexander Duyck <alexander.h.duyck@intel.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      01c5f864
    • Alexander Duyck's avatar
      net: Rewrite netif_reset_xps_queue to allow for better code reuse · 10cdc3f3
      Alexander Duyck authored
      
      
      This patch does a minor refactor on netif_reset_xps_queue to address a few
      items I noticed.
      
      First is the fact that we are doing removal of queues in both
      netif_reset_xps_queue and netif_set_xps_queue.  Since there is no need to
      have the code in two places I am pushing it out into a separate function
      and will come back in another patch and reuse the code in
      netif_set_xps_queue.
      
      The second item this change addresses is the fact that the Tx queues were
      not getting their numa_node value cleared as a part of the XPS queue reset.
      This patch resolves that by resetting the numa_node value if the dev_maps
      value is set.
      Signed-off-by: default avatarAlexander Duyck <alexander.h.duyck@intel.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      10cdc3f3
    • Alexander Duyck's avatar
      net: Add functions netif_reset_xps_queue and netif_set_xps_queue · 537c00de
      Alexander Duyck authored
      
      
      This patch adds two functions, netif_reset_xps_queue and
      netif_set_xps_queue.  The main idea behind these two functions is to
      provide a mechanism through which drivers can update their defaults in
      regards to XPS.
      
      Currently no such mechanism exists and as a result we cannot use XPS for
      things such as ATR which would require a basic configuration to start in
      which the Tx queues are mapped to CPUs via a 1:1 mapping.  With this change
      I am making it possible for drivers such as ixgbe to be able to use the XPS
      feature by controlling the default configuration.
      Signed-off-by: default avatarAlexander Duyck <alexander.h.duyck@intel.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      537c00de
    • Alexander Duyck's avatar
      net: Split core bits of netdev_pick_tx into __netdev_pick_tx · 416186fb
      Alexander Duyck authored
      
      
      This change splits the core bits of netdev_pick_tx into a separate function.
      The main idea behind this is to make this code accessible to select queue
      functions when they decide to process the standard path instead of their
      own custom path in their select queue routine.
      Signed-off-by: default avatarAlexander Duyck <alexander.h.duyck@intel.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      416186fb
    • Eric Dumazet's avatar
      softirq: reduce latencies · c10d7367
      Eric Dumazet authored
      
      
      In various network workloads, __do_softirq() latencies can be up
      to 20 ms if HZ=1000, and 200 ms if HZ=100.
      
      This is because we iterate 10 times in the softirq dispatcher,
      and some actions can consume a lot of cycles.
      
      This patch changes the fallback to ksoftirqd condition to :
      
      - A time limit of 2 ms.
      - need_resched() being set on current task
      
      When one of this condition is met, we wakeup ksoftirqd for further
      softirq processing if we still have pending softirqs.
      
      Using need_resched() as the only condition can trigger RCU stalls,
      as we can keep BH disabled for too long.
      
      I ran several benchmarks and got no significant difference in
      throughput, but a very significant reduction of latencies (one order
      of magnitude) :
      
      In following bench, 200 antagonist "netperf -t TCP_RR" are started in
      background, using all available cpus.
      
      Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC
      IRQ (hard+soft)
      
      Before patch :
      
      # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k
      RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY
      MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET
      to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind
      RT_LATENCY=550110.424
      MIN_LATENCY=146858
      MAX_LATENCY=997109
      P50_LATENCY=305000
      P90_LATENCY=550000
      P99_LATENCY=710000
      MEAN_LATENCY=376989.12
      STDDEV_LATENCY=184046.92
      
      After patch :
      
      # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k
      RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY
      MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET
      to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind
      RT_LATENCY=40545.492
      MIN_LATENCY=9834
      MAX_LATENCY=78366
      P50_LATENCY=33583
      P90_LATENCY=59000
      P99_LATENCY=69000
      MEAN_LATENCY=38364.67
      STDDEV_LATENCY=12865.26
      Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
      Cc: David Miller <davem@davemloft.net>
      Cc: Tom Herbert <therbert@google.com>
      Cc: Ben Hutchings <bhutchings@solarflare.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      c10d7367
    • Eric Dumazet's avatar
      net_sched: more precise pkt_len computation · 1def9238
      Eric Dumazet authored
      One long standing problem with TSO/GSO/GRO packets is that skb->len
      doesn't represent a precise amount of bytes on wire.
      
      Headers are only accounted for the first segment.
      For TCP, thats typically 66 bytes per 1448 bytes segment missing,
      an error of 4.5 % for normal MSS value.
      
      As consequences :
      
      1) TBF/CBQ/HTB/NETEM/... can send more bytes than the assigned limits.
      2) Device stats are slightly under estimated as well.
      
      Fix this by taking account of headers in qdisc_skb_cb(skb)->pkt_len
      computation.
      
      Packet schedulers should use qdisc pkt_len instead of skb->len for their
      bandwidth limitations, and TSO enabled devices drivers could use pkt_len
      if their statistics are not hardware assisted, and if they don't scratch
      skb->cb[] first word.
      
      Both egress and ingress paths work, thanks to commit fda55eca
      
      
      (net: introduce skb_transport_header_was_set()) : If GRO built
      a GSO packet, it also set the transport header for us.
      Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
      Cc: Jamal Hadi Salim <jhs@mojatatu.com>
      Cc: Stephen Hemminger <shemminger@vyatta.com>
      Cc: Paolo Valente <paolo.valente@unimore.it>
      Cc: Herbert Xu <herbert@gondor.apana.org.au>
      Cc: Patrick McHardy <kaber@trash.net>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      1def9238
    • Vijay Subramanian's avatar
      doc: Clarify behavior when sysctl tcp_ecn = 1 · 3d55b323
      Vijay Subramanian authored
      Recent commit (commit 7e3a2dc5
      
       doc: make the description of how tcp_ecn
      works more explicit and clear ) clarified the behavior of tcp_ecn sysctl
      variable but description is inconsistent. When requested by incoming conections,
      ECN is enabled with not just tcp_ecn = 2 but also with tcp_ecn = 1.
      
      This patch makes it clear that with tcp_ecn = 1, ECN is enabled when requested
      by incoming connections.
      
      Also fix spelling of 'incoming'.
      Signed-off-by: default avatarVijay Subramanian <subramanian.vijay@gmail.com>
      Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      3d55b323