htb_sched structures are big, and source of false sharing on SMP.
Every time a packet is queued or dequeue, many cache lines must be
touched because structures are not lay out properly.
By carefully splitting htb_sched in two parts, and define sub structures
to increase data locality, we can improve performance dramatically on
SMP.
New htb_prio structure can also be used in htb_class to increase data
locality.
I got 26 % performance increase on a 24 threads machine, with 200
concurrent netperf in TCP_RR mode, using a HTB hierarchy of 4 classes.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In previous discussions, I tried to find some reasonable heuristics
for delayed ACK, however this seems not possible, according to Eric:
"ACKS might also be delayed because of bidirectional
traffic, and is more controlled by the application
response time. TCP stack can not easily estimate it."
"ACK can be incredibly useful to recover from losses in
a short time.
The vast majority of TCP sessions are small lived, and we
send one ACK per received segment anyway at beginning or
retransmits to let the sender smoothly increase its cwnd,
so an auto-tuning facility wont help them that much."
and according to David:
"ACKs are the only information we have to detect loss.
And, for the same reasons that TCP VEGAS is fundamentally
broken, we cannot measure the pipe or some other
receiver-side-visible piece of information to determine
when it's "safe" to stretch ACK.
And even if it's "safe", we should not do it so that losses are
accurately detected and we don't spuriously retransmit.
The only way to know when the bandwidth increases is to
"test" it, by sending more and more packets until drops happen.
That's why all successful congestion control algorithms must
operate on explicited tested pieces of information.
Similarly, it's not really possible to universally know if
it's safe to stretch ACK or not."
It still makes sense to enable or disable quick ack mode like
what TCP_QUICK_ACK does.
Similar to TCP_QUICK_ACK option, but for people who can't
modify the source code and still wants to control
TCP delayed ACK behavior. As David suggested, this should belong
to per-path scope, since different pathes may want different
behaviors.
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Rick Jones <rick.jones2@hp.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Graf <tgraf@suug.ch>
CC: David Laight <David.Laight@ACULAB.COM>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pci core has been saved pm cap register offset by pdev->pm_cap in pci_pm_init()
in init path. So we can use pdev->pm_cap instead of using
pci_find_capability(pdev, PCI_CAP_ID_PM) for better performance and simplified code.
Signed-off-by: Yijing Wang <wangyijing@huawei.com>
Cc: Michael Chan <mchan@broadcom.com>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Pci core has been saved pm cap register offset by pdev->pm_cap in pci_pm_init()
in init path. So we can use pdev->pm_cap instead of using
pci_find_capability(pdev, PCI_CAP_ID_PM) for better performance and simplified code.
Signed-off-by: Yijing Wang <wangyijing@huawei.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Bill Pemberton <wfp5p@virginia.edu>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: netdev@vger.kernel.org (open list:NETWORKING DRIVERS)
Signed-off-by: David S. Miller <davem@davemloft.net>
Pci_enable_device() will set device power state to D0,
so it's no need to do it again in bnx2x_init_dev().
Also remove redundant PM Cap find code, because pci core
has been saved the pci device pm cap value.
Signed-off-by: Yijing Wang <wangyijing@huawei.com>
Cc: Eilon Greenstein <eilong@broadcom.com>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Acked-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ETRAX_ETHERNET selects ETHERNET and MII, which depend on NETDEVICES.
I don't think anything should select NETDEVICES, so make it a
dependency. It also doesn't need to select or depend on ETHERNET,
which has nothing to do with the Ethernet library functions.
BPCTL selects MII, which depends on NETDEVICES. But everything in the
drivers/staging/silicom directory is related to net devices, so make
NET_VENDOR_SILICOM depend on NETDEVICES and remove the now-redundant
dependencies on NET.
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
This has no dependency on any of the drivers under NET_CORE.
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
All drivers that select MII also need to select NET_CORE because MII
depends on it. This is a bit ridiculous because NET_CORE is just a
menu option that doesn't enable any code by itself.
There is also no need for it to be a visible option, since its users
all select it.
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Also, cleanup bond_alb_handle_active_change() from 2 identical ifs.
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
be_find_vfs() is no longer needed as the common PCI calls provide the same
functionality.
Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The use of this attribute has been added in 32b8a8e59c (sit: add IPv4 over
IPv4 support). It is optional, by default proto is IPPROTO_IPV6.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current situation is that SOCK_MIN_RCVBUF is 2048 + sizeof(struct sk_buff))
while SOCK_MIN_SNDBUF is 2048. Since in both cases, skb->truesize is used for
sk_{r,w}mem_alloc accounting, we should have both sizes adjusted via defining a
TCP_SKB_MIN_TRUESIZE.
Further, as Eric Dumazet points out, the minimal skb truesize in transmit path is
SKB_TRUESIZE(2048) after commit f07d960df3 ("tcp: avoid frag allocation for
small frames"), and tcp_sendmsg() tries to limit skb size to half the congestion
window, meaning we try to build two skbs at minimum. Thus, having SOCK_MIN_SNDBUF
as 2048 can hit a small regression for some applications setting to low
SO_SNDBUF / SO_RCVBUF. Note that we define a TCP_SKB_MIN_TRUESIZE, because
SKB_TRUESIZE(2048) adds SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), but in
case of TCP skbs, the skb_shared_info is part of the 2048 bytes allocation for
skb->head.
The minor adaption in sk_stream_moderate_sndbuf() is to silence a warning by
using a typed max macro, as similarly done in SOCK_MIN_RCVBUF occurences, that
would appear otherwise.
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
thresh and interval are global resources,
only init net can change them.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Though we don't export the /proc/sys/net/ipv[4,6]/neigh/default/
directory to the un-init_net, but we can still use cmd such as
"ip ntable change name arp_cache locktime 129" to change the locktime
of default neigh_parms.
This patch disallows the un-init_net to find out the neigh_table.parms.
So the un-init_net will failed to influence the init_net.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
neigh_table.parms always exist and is initialized,kmemdup
can use it to create new neigh_parms, actually lookup_neigh_parms
here will return neigh_table.parms too.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Check next packet availability by validating that HW has finished CQE
placement. This saves latency of another dma transaction performed to update
SB indexes.
Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adds ndo_ll_poll method and locking for FPs between LL and the napi.
When receiving a packet we use skb_mark_ll to record the napi it came from.
Add each napi to the napi_hash right after netif_napi_add().
Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add basic support for LLS.
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pravin B Shelar says:
====================
Following patch series adds support for gre tunneling.
First six patches extend kernel gre and ip_tunnel modules
api so that there is more code sharing between gre modules
and ovs. Rest of patches adds ovs tunneling infrastructre
and gre protocol vport.
V2 fixes two patches according to comments from Jesse.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add gre vport implementation. Most of gre protocol processing
is pushed to gre module. It make use of gre demultiplexer
therefore it can co-exist with linux device based gre tunnels.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Following patch adds start offset for sw_flow-key, so that we can
skip tunneling information in key for non-tunnel flows.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
MAX_ACTIONS_BUFSIZE limits action list size, set tunnel action
needs extra space on action list, for now increase max actions list limit.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add ovs tunnel interface for set tunnel action for userspace.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rather than validating actions and then copying all actiaons
in one block, following patch does same operation in single pass.
This validate and copy action one by one. This is required for
ovs tunneling patch.
This patch does not change any functionality.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Process skb tunnel header before sending packet to protocol handler.
this allows code sharing between gre and ovs gre modules.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Refactor various ip tunnels xmit functions and extend iptunnel_xmit()
so that there is more code sharing.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently there is only one user is allowed to register for gre
protocol. Following patch adds de-multiplexer. So that multiple
modules can listen on gre protocol e.g. kernel gre devices and ovs.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use cmpxchg() for atomic protocol registration which saves
code and data space.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The only R8A7740 specific #ifdef hindering ARM multiplatform build is left in
sh_eth_rx(): it covers the code shifting Rx buffer descriptor word 0 by 16. Get
rid of the #ifdef by adding 'shift_rd0' field to the 'struct sh_eth_cpu_data',
making the shift dependent on it, and setting it to 1 for the R8A7740 case...
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the comment to 'enum TD_STS_BIT', reformat the values, and add a couple of
values missing before (though unused by the driver).
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For SH7724 'eesipr_value' field initializer includes DMAC_M_RFRMER & DMAC_M_ECI
bits which are already contained in 0x01ff009f -- remove them.
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The 'tx_error_check' field of 'struct sh_eth_cpu_data' is write-only, so remove
it along with the DEFAULT_TX_ERROR_CHECK macro.
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver hasn't used NAPI so far; implement its support at last...
The patch was tested on Renesas R8A77781 BOCK-W board.
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sh_eth_interrupt() uses the same Rx interrupt mask twice to check the interrupt
status register -- #define EESR_RX_CHECK and use it instead.
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/wireless/ath/ath9k/Kconfig
drivers/net/xen-netback/netback.c
net/batman-adv/bat_iv_ogm.c
net/wireless/nl80211.c
The ath9k Kconfig conflict was a change of a Kconfig option name right
next to the deletion of another option.
The xen-netback conflict was overlapping changes involving the
handling of the notify list in xen_netbk_rx_action().
Batman conflict resolution provided by Antonio Quartulli, basically
keep everything in both conflict hunks.
The nl80211 conflict is a little more involved. In 'net' we added a
dynamic memory allocation to nl80211_dump_wiphy() to fix a race that
Linus reported. Meanwhile in 'net-next' the handlers were converted
to use pre and post doit handlers which use a flag to determine
whether to hold the RTNL mutex around the operation.
However, the dump handlers to not use this logic. Instead they have
to explicitly do the locking. There were apparent bugs in the
conversion of nl80211_dump_wiphy() in that we were not dropping the
RTNL mutex in all the return paths, and it seems we very much should
be doing so. So I fixed that whilst handling the overlapping changes.
To simplify the initial returns, I take the RTNL mutex after we try
to allocate 'tb'.
Signed-off-by: David S. Miller <davem@davemloft.net>
Since my commit 3713b4e364 ("nl80211: allow splitting wiphy
information in dumps"), nl80211_dump_wiphy() uses the global
nl80211_fam.attrbuf for parsing the incoming data. This wouldn't
be a problem if it only did so on the first dump iteration which
is locked against other commands in generic netlink, but due to
space constraints in cb->args (the needed state doesn't fit) I
decided to always parse the original message. That's racy though
since nl80211_fam.attrbuf could be used by some other parsing in
generic netlink concurrently.
For now, fix this by allocating a separate parse buffer (it's a
bit too big for the stack, currently 1448 bytes on 64-bit). For
-next, I'll change the code to parse into the global buffer in
the first round only and then allocate a smaller buffer to keep
the data in cb->args.
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
The check introduced by:
commit 26a41ae604
Author: stephen hemminger <stephen@networkplumber.org>
Date: Mon Jun 17 12:09:58 2013 -0700
vxlan: only migrate dynamic FDB entries
was not correct because it is checking flag about type of FDB
entry, rather than the state (dynamic versus static). The confusion
arises because vxlan is reusing values from bridge, and bridge is
reusing values from neighbour table, and easy to get lost in translation.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case of error, the function devm_ioremap_resource() returns ERR_PTR()
and never returns NULL. The NULL test in the return value check should
be replaced with IS_ERR().
Introduce by commit 0ae99b5fed
(bcm63xx_enet: split DMA channel register accesses)
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
The usb_8dev hardware has problems on some xhci USB hosts. The driver fails to
read the firmware revision in the probe function. This leads to the following
Oops:
[ 3356.635912] kernel BUG at net/core/dev.c:5701!
The driver tries to free the netdev, which has already been registered, without
unregistering it.
This patch fixes the problem by unregistering the netdev in the error path.
Reported-by: Michael Olbrich <m.olbrich@pengutronix.de>
Reviewed-by: Bernd Krumboeck <krumboeck@universalnet.at>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Since some refactoring in 5f5a011, ndisc_send_redirect called
ndisc_fill_redirect_hdr_option on the wrong skb, leading to data corruption or
in the worst case a panic when the skb_put failed.
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
Reviewed-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The change set of 7a884dc "[SK_BUFF]: Convert skb->tail to sk_buff_data_t"
converted skb->tail from pointer into sk_buff_data_t.
Thus skb->tail is not always pointer, the area pointed by skb->tail
should be accessed via skb_tail_pointer().
Found by inspection. Compile tested only.
Cc: Simon Horman <horms@verge.net.au>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: devel@driverdev.osuosl.org
Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
Reviewed-by: Simon Horman <horms@verge.net.au>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The change set of 4305b541, "[SK_BUFF]: Convert skb->end to sk_buff_data_t"
converted skb->end from pointer type to sk_buff_data_t.
The pointed value should be accessed via skb_end_pointer().
Since arm arch doesn't define NET_SKBUFF_DATA_USES_OFFSET,
skb->end is effectively pointer. So it doesn't cause a real problem.
But this patch is good for consistency.
Found by inspection. Compile tested only.
Cc: Simon Horman <horms@verge.net.au>
Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
Reviewed-by: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
