Various patches have ended up changing the style of the symbolic macros/register
defines to different style.
As a result, the current kernel.org files are a mix of different macro styles.
Since this macro/register defines is used by different drivers a
few patch series have ended up adding duplicate macro/register define entries
with different styles. This makes these register define/macro files a complete
mess and we want to make them clean and consistent. This patch cleans up a part
of it.
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Various patches have ended up changing the style of the symbolic macros/register
to different style.
As a result, the current kernel.org files are a mix of different macro styles.
Since this macro/register defines is used by different drivers a
few patch series have ended up adding duplicate macro/register define entries
with different styles. This makes these register define/macro files a complete
mess and we want to make them clean and consistent. This patch cleans up a part
of it.
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
To enable gro_flush_timeout, a driver has to use napi_complete_done()
instead of napi_complete().
Tested:
Ran 200 netperf TCP_STREAM from A to B (10Gbe mlx4 link, 8 RX queues)
Without this feature, we send back about 305,000 ACK per second.
GRO aggregation ratio is low (811/305 = 2.65 segments per GRO packet)
Setting a timer of 2000 nsec is enough to increase GRO packet sizes
and reduce number of ACK packets. (811/19.2 = 42)
Receiver performs less calls to upper stacks, less wakes up.
This also reduces cpu usage on the sender, as it receives less ACK
packets.
Note that reducing number of wakes up increases cpu efficiency, but can
decrease QPS, as applications wont have the chance to warmup cpu caches
doing a partial read of RPC requests/answers if they fit in one skb.
B:~# sar -n DEV 1 10 | grep eth0 | tail -1
Average: eth0 811269.80 305732.30 1199462.57 19705.72 0.00
0.00 0.50
B:~# echo 2000 >/sys/class/net/eth0/gro_flush_timeout
B:~# sar -n DEV 1 10 | grep eth0 | tail -1
Average: eth0 811577.30 19230.80 1199916.51 1239.80 0.00
0.00 0.50
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tuning coalescing parameters on NIC can be really hard.
Servers can handle both bulk and RPC like traffic, with conflicting
goals : bulk flows want as big GRO packets as possible, RPC want minimal
latencies.
To reach big GRO packets on 10Gbe NIC, one can use :
ethtool -C eth0 rx-usecs 4 rx-frames 44
But this penalizes rpc sessions, with an increase of latencies, up to
50% in some cases, as NICs generally do not force an interrupt when
a packet with TCP Push flag is received.
Some NICs do not have an absolute timer, only a timer rearmed for every
incoming packet.
This patch uses a different strategy : Let GRO stack decides what do do,
based on traffic pattern.
Packets with Push flag wont be delayed.
Packets without Push flag might be held in GRO engine, if we keep
receiving data.
This new mechanism is off by default, and shall be enabled by setting
/sys/class/net/ethX/gro_flush_timeout to a value in nanosecond.
To fully enable this mechanism, drivers should use napi_complete_done()
instead of napi_complete().
Tested:
Ran 200 netperf TCP_STREAM from A to B (10Gbe mlx4 link, 8 RX queues)
Without this feature, we send back about 305,000 ACK per second.
GRO aggregation ratio is low (811/305 = 2.65 segments per GRO packet)
Setting a timer of 2000 nsec is enough to increase GRO packet sizes
and reduce number of ACK packets. (811/19.2 = 42)
Receiver performs less calls to upper stacks, less wakes up.
This also reduces cpu usage on the sender, as it receives less ACK
packets.
Note that reducing number of wakes up increases cpu efficiency, but can
decrease QPS, as applications wont have the chance to warmup cpu caches
doing a partial read of RPC requests/answers if they fit in one skb.
B:~# sar -n DEV 1 10 | grep eth0 | tail -1
Average: eth0 811269.80 305732.30 1199462.57 19705.72 0.00
0.00 0.50
B:~# echo 2000 >/sys/class/net/eth0/gro_flush_timeout
B:~# sar -n DEV 1 10 | grep eth0 | tail -1
Average: eth0 811577.30 19230.80 1199916.51 1239.80 0.00
0.00 0.50
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Babel uses rt_proto 42. Add to userspace visible header file.
Signed-off-by: Dave Taht <dave.taht@bufferbloat.net>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the dependency on the "warning" sysctl (net_msg_warn)
which is only used by the LIMIT_NETDEBUG macro.
Convert the LIMIT_NETDEBUG use in DCCP_WARN to the more
common net_warn_ratelimited mechanism.
This still ratelimits based on the net_ratelimit()
function, but removes the check for the sysctl.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As NIC multicast filtering isn't perfect, and some platforms are
quite content to spew broadcasts, we should not trigger an event
for skb:kfree_skb when we do not have a match for such an incoming
datagram. We do though want to avoid sweeping the matter under the
rug entirely, so increment a suitable statistic.
This incorporates feedback from David L. Stevens, Karl Neiss and Eric
Dumazet.
V3 - use bool per David Miller
Signed-off-by: Rick Jones <rick.jones2@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Olivier having laid the groundwork this patch transmits the
multicast flag to the device to save some bus traffic.
Signed-off-by: Oliver Neukum <oneukum@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The entries in the Kbuild files are incorrectly sorted.
Matters for aesthetics only.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes the following sparse warnings. One is fixed by casting return
value to a return type of the function. The others by creating a specific
stmmac_platform.h which provides the bits related to the platform driver.
drivers/net/ethernet/stmicro/stmmac/dwmac-meson.c:59:29: warning: incorrect type in return expression (different address spaces)
drivers/net/ethernet/stmicro/stmmac/dwmac-meson.c:59:29: expected void *
drivers/net/ethernet/stmicro/stmmac/dwmac-meson.c:59:29: got void [noderef] <asn:2>*reg
drivers/net/ethernet/stmicro/stmmac/dwmac-meson.c:64:29: warning: symbol 'meson6_dwmac_data' was not declared. Should it be static?
drivers/net/ethernet/stmicro/stmmac/dwmac-sti.c:354:29: warning: symbol 'stih4xx_dwmac_data' was not declared. Should it be static?
drivers/net/ethernet/stmicro/stmmac/dwmac-sti.c:361:29: warning: symbol 'stid127_dwmac_data' was not declared. Should it be static?
drivers/net/ethernet/stmicro/stmmac/dwmac-sunxi.c:133:29: warning: symbol 'sun7i_gmac_data' was not declared. Should it be static?
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is a kernel helper to dump buffers in a hexdecimal format. This patch
substitutes the open coded function by calling that helper.
The output is slightly changed:
- no lead space
- ASCII part will be printed along with the dump
- offset is longer than 3 characters (now 8)
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Herbert Xu says:
====================
Replace skb_copy_datagram_const_iovec with iterator version
This patch series adds the helper skb_copy_datagram_iter, which
is meant to replace both skb_copy_datagram_iovec and its evil
twin skb_copy_datagram_const_iovec.
It then converts tun and macvtap over to the new helper and finally
removes skb_copy_datagram_const_iovec which is only used by tun
and macvtap.
The copy_to_iter return value issue pointed out by Al has now been
fixed.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that both macvtap and tun are using skb_copy_datagram_iter, we
can kill the abomination that is skb_copy_datagram_const_iovec.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes the use of skb_copy_datagram_const_iovec in
favour of the iovec iterator-based skb_copy_datagram_iter.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes the use of skb_copy_datagram_const_iovec in
favour of the iovec iterator-based skb_copy_datagram_iter.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds skb_copy_datagram_iter, which is identical to
skb_copy_datagram_iovec except that it operates on iov_iter
instead of iovec.
Eventually all users of skb_copy_datagram_iovec should switch
over to iov_iter and then we can remove skb_copy_datagram_iovec.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add definition to vxlan nla_policy for UDP checksum. This is necessary
to enable UDP checksums on VXLAN.
In some instances, enabling UDP checksums can improve performance on
receive for devices that return legacy checksum-unnecessary for UDP/IP.
Also, UDP checksum provides some protection against VNI corruption.
Testing:
Ran 200 instances of TCP_STREAM and TCP_RR on bnx2x.
TCP_STREAM
IPv4, without UDP checksums
14.41% TX CPU utilization
25.71% RX CPU utilization
9083.4 Mbps
IPv4, with UDP checksums
13.99% TX CPU utilization
13.40% RX CPU utilization
9095.65 Mbps
TCP_RR
IPv4, without UDP checksums
94.08% TX CPU utilization
156/248/462 90/95/99% latencies
1.12743e+06 tps
IPv4, with UDP checksums
94.43% TX CPU utilization
158/250/462 90/95/99% latencies
1.13345e+06 tps
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tom Lendacky says:
====================
amd-xgbe: AMD XGBE driver updates 2014-11-06
The following series of patches fixes a couple of bugs that slipped
through my last series.
- Free channel structure after freeing the per channel interrupts
- If an skb error allocation occurs during receive processing check
whether more descriptors are associated with the packet or whether
to start on a new packet
This patch series is based on net-next.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
If the skb allocation fails during receive processing, the driver would
continue reading descriptors without first determining if there were
any more descriptors for the current packet. Update the code to check
whether more descriptors are associated with the current packet or
whether to move on to the next descriptor as a new packet.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The channel structure is freed before freeing the per channel
interrupts resulting in a kernel oops. Move the call to free
the channel structure to after the freeing of the per channel
interrupts.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o Poll for the link events only if firmware doesn't have capability
to notify the driver for the link events.
Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eli Cohen says:
====================
mlx5_core fixes for 3.18
the following two patches fix races to could lead to kernel panic in some cases.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
When events arrive at driver load, the event handler gets called even before
the spinlock and list are initialized. Fix this by moving the initialization
before EQs creation.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After the EQ is created, it can possibly generate interrupts and the interrupt
handler is referencing eq->dev. It is therefore required to set eq->dev before
calling request_irq() so if an event is generated before request_irq() returns,
we will have a valid eq->dev field.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pravin B Shelar says:
====================
Open vSwitch
First two patches are related to OVS MPLS support. Rest of patches
are mostly refactoring and minor improvements to openvswitch.
v1-v2:
- Fix conflicts due to "gue: Remote checksum offload"
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Sowmini Varadhan says:
====================
sunvnet: bug fixes
This patch series has a coding-style fix and a bug fix.
The coding style fix (patch 1) is the extra indentation flagged by
Ben Hutchings:
http://marc.info/?l=linux-netdev&m=141529243409594&w=2
The bugfix (patch 2) is the following:
when vnet_event_napi() is called as part of napi_resume
(i.e., continuation of a previous NAPI read that was truncated
due to budget constraints), and then finds no more packets to read,
the code was trying to avoid an additional trip through ldc_rx
as an optimization. However, when this corner case happens, we would
need to reset a number of dring state bits such as rcv_nxt carefully,
which quickly becomes complex and hacky. The cleaner solution
is to just roll back to vnet_poll, re-enable interrupts and set up
dring state as was done in the pre-NAPI version of the driver.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
vnet_event_napi() may be called as part of the NAPI ->poll,
to resume reading descriptor rings. When no data is available,
descriptor ring state (e.g., rcv_nxt) needs to be reset
carefully to stay in lock-step with ldc_read(). In the interest
of simplicity, the best way to do this is to return from
vnet_event_napi() when there are no more packets to read.
The next trip through ldc_rx will correctly set up the dring state.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Tested-by: David Stevens <david.stevens@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
remove redundant tab.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Reported-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Hayes Wang says:
====================
r8152: rtl_ops_init modify
Initialize the ops through tp->version. This could skip checking
each VID/PID.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The PIDs are only used in the id table, so the definitions are
unnacessary. Remove them wouldn't have confusion.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace using VID/PID with using tp->version to initialize the ops.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move r8152b_get_version() to the location before rtl_ops_init().
Then, the rtl_ops_init() could use tp->version.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 64ce207306 ("[NET]: Make NETDEBUG pure printk wrappers")
originally had these NETDEBUG printks as always emitting.
Commit a2a316fd06 ("[NET]: Replace CONFIG_NET_DEBUG with sysctl")
added a net_msg_warn sysctl to these NETDEBUG uses.
Convert these NETDEBUG uses to normal pr_info calls.
This changes the output prefix from "ESP: " to include
"IPSec: " for the ipv4 case and "IPv6: " for the ipv6 case.
These output lines are now like the other messages in the files.
Other miscellanea:
Neaten the arithmetic spacing to be consistent with other
arithmetic spacing in the files.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
These messages aren't useful as there's a generic dump_stack()
on OOM.
Neaten the comment and if test above the OOM by separating the
assign in if into an allocation then if test.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The mv88e6172 is very similar to the mv88e6171. So extend the
mv88e6171 driver to support it.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the ports phys are connected to the switches internal MDIO bus,
we need to connect the phy to the slave netdev, otherwise
auto-negotiation etc, does not work.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes: 4bba3925 ("[PKT_SCHED]: Prefix tc actions with act_")
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
when cpsw is build as modulea and simple insert and removal of module
creates a deadlock, due to delete timer. the timer is created and destroyed
in cpsw_ale_start and cpsw_ale_stop which are from device open and close.
root@am437x-evm:~# modprobe -r ti_cpsw
[ 158.505333] INFO: trying to register non-static key.
[ 158.510623] the code is fine but needs lockdep annotation.
[ 158.516448] turning off the locking correctness validator.
[ 158.522282] CPU: 0 PID: 1339 Comm: modprobe Not tainted 3.14.23-00445-gd41c88f #44
[ 158.530359] [<c0015380>] (unwind_backtrace) from [<c0012088>] (show_stack+0x10/0x14)
[ 158.538603] [<c0012088>] (show_stack) from [<c054ad70>] (dump_stack+0x78/0x94)
[ 158.546295] [<c054ad70>] (dump_stack) from [<c0088008>] (__lock_acquire+0x176c/0x1b74)
[ 158.554711] [<c0088008>] (__lock_acquire) from [<c0088944>] (lock_acquire+0x9c/0x104)
[ 158.563043] [<c0088944>] (lock_acquire) from [<c004e520>] (del_timer_sync+0x44/0xd8)
[ 158.571289] [<c004e520>] (del_timer_sync) from [<bf2eac1c>] (cpsw_ale_destroy+0x10/0x3c [ti_cpsw])
[ 158.580821] [<bf2eac1c>] (cpsw_ale_destroy [ti_cpsw]) from [<bf2eb268>] (cpsw_remove+0x30/0xa0 [ti_cpsw])
[ 158.591000] [<bf2eb268>] (cpsw_remove [ti_cpsw]) from [<c035ef44>] (platform_drv_remove+0x18/0x1c)
[ 158.600527] [<c035ef44>] (platform_drv_remove) from [<c035d8bc>] (__device_release_driver+0x70/0xc8)
[ 158.610236] [<c035d8bc>] (__device_release_driver) from [<c035e0d4>] (driver_detach+0xb4/0xb8)
[ 158.619386] [<c035e0d4>] (driver_detach) from [<c035d6e4>] (bus_remove_driver+0x4c/0x90)
[ 158.627988] [<c035d6e4>] (bus_remove_driver) from [<c00af2a8>] (SyS_delete_module+0x10c/0x198)
[ 158.637144] [<c00af2a8>] (SyS_delete_module) from [<c000e580>] (ret_fast_syscall+0x0/0x48)
[ 179.524727] INFO: rcu_sched detected stalls on CPUs/tasks: {} (detected by 0, t=2102 jiffies, g=1487, c=1486, q=6)
[ 179.535741] INFO: Stall ended before state dump start
Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ATM, txq_reclaim will dequeue and free an skb for each tx desc released
by the hw that has TX_LAST_DESC set. However, in case of TSO, each
hw desc embedding the last part of a segment has TX_LAST_DESC set,
losing the one-to-one 'last skb frag'/'TX_LAST_DESC set' correspondance,
which causes data corruption.
Fix this by checking TX_ENABLE_INTERRUPT instead of TX_LAST_DESC, and
warn when trying to dequeue from an empty txq (which can be symptomatic
of releasing skbs prematurely).
Fixes: 3ae8f4e0b9 ('net: mv643xx_eth: Implement software TSO')
Reported-by: Slawomir Gajzner <slawomir.gajzner@gmail.com>
Reported-by: Julien D'Ascenzio <jdascenzio@yahoo.fr>
Signed-off-by: Karl Beldan <karl.beldan@rivierawaves.com>
Cc: Ian Campbell <ijc@hellion.org.uk>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Ezequiel Garcia <ezequiel.garcia@free-electrons.com>
Cc: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Shradha Shah says:
====================
sfc: Clean up Siena SR-IOV support in preparation for EF10 SR-IOV support
This patch series provides a base and clean up for the upcoming
EF10 SRIOV patches.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Also add dummy functions where required to avoid NULL pointer dereference.
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Patch in preparation for the upcoming EF10 sriov support.
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch series provides a base and cleanup for the
upcoming EF10 SRIOV support.
This patch moves the VF state into siena_nic_data as a basis to
save the VF state based on nic type.
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Unconditionally pulling 128 bytes into the linear area is not required
for:
- security: Every protocol demux starts with pskb_may_pull() to pull
frag data into the linear area, if necessary, before looking at
headers.
- performance: Netback has already grant copied up-to 128 bytes from
the first slot of a packet into the linear area. The first slot
normally contain all the IPv4/IPv6 and TCP/UDP headers.
The unconditional pull would often copy frag data unnecessarily. This
is a performance problem when running on a version of Xen where grant
unmap avoids TLB flushes for pages which are not accessed. TLB
flushes can now be avoided for > 99% of unmaps (it was 0% before).
Grant unmap TLB flush avoidance will be available in a future version
of Xen (probably 4.6).
Signed-off-by: Malcolm Crossley <malcolm.crossley@citrix.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
