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>
Andy Shevchenko says:
====================
stmmac: pci: various cleanups and fixes
There are few cleanups and fixes regarding to stmmac PCI driver.
This has been tested on Intel Galileo board with recent net-next tree.
Since v2:
- drop patch 5/5 since it will be part of a big change across entire subsystem
Since v1:
- remove already applied patch
- append patch 1/5
- rework patch 3/5 to be functional compatible with original code
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of pr_* macros let's use dev_* macros which provide device name.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Migrate pci driver to managed resources to reduce boilerplate error handling
code.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Convert system PM callbacks to use dev_pm_ops. In addition remove the PCI calls
related to a power state since the bus code cares about this already.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The last standard PCI resource is defined as PCI_STD_RESOURCE_END. Thus, we
could use it instead of plain integer.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes the following sparse warnings.
drivers/net/ethernet/stmicro/stmmac/enh_desc.c:381:30: warning: symbol 'enh_desc_ops' was not declared. Should it be static?
drivers/net/ethernet/stmicro/stmmac/norm_desc.c:253:30: warning: symbol 'ndesc_ops' was not declared. Should it be static?
drivers/net/ethernet/stmicro/stmmac/stmmac_hwtstamp.c:141:33: warning: symbol 'stmmac_ptp' was not declared. Should it be static?
There is no functional change.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Giuseppe CAVALLARO <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for tunnels with local or
remote wildcard endpoints. With this we get a
NBMA tunnel mode like we have it for ipv4 and
sit tunnels.
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently we need the IP6_TNL_F_CAP_XMIT capabiltiy to transmit
packets through an ipv6 tunnel. This capability is set when the
tunnel gets configured, based on the tunnel endpoint addresses.
On tunnels with wildcard tunnel endpoints, we need to do the
capabiltiy checking on a per packet basis like it is done in
the receive path.
This patch extends ip6_tnl_xmit_ctl() to take local and remote
addresses as parameters to allow for per packet capabiltiy
checking.
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
OVS does mask validation even if it does not need to convert
netlink mask attributes to mask structure. ovs_nla_get_match()
caller can pass NULL mask structure pointer if the caller does
not need mask. Therefore NULL check is required in SW_FLOW_KEY*
macros. Following patch does not convert mask netlink attributes
if mask pointer is NULL, so we do not need these checks in
SW_FLOW_KEY* macro.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Daniele Di Proietto <ddiproietto@vmware.com>
Acked-by: Andy Zhou <azhou@nicira.com>
There are two separate API to allocate and copy actions list. Anytime
OVS needs to copy action list, it needs to call both functions.
Following patch moves action allocation to copy function to avoid
code duplication.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
The 'flow' memeber was chosen for removal because it's only used
in ovs_execute_actions() we can pass it as argument to this
function.
Signed-off-by: Lorand Jakab <lojakab@cisco.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Should be the same as other IPv6 address fields.
Current master produces sparse warnings without this change.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
If the internal device is not up, it should drop received
packets. Sometimes it receive the broadcast or multicast
packets, and the ip protocol stack will casue more cpu
usage wasted.
Signed-off-by: Chunhe Li <lichunhe@huawei.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Avoid recursive read_rcu_lock() by using the lighter weight
get_dp_rcu() API. Add proper locking assertions to get_dp().
Signed-off-by: Andy Zhou <azhou@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Split up ovs_flow_cmd_fill_info() to make it easier to cache parts of a
dump reply. This will be used to streamline flow_dump in a future patch.
Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Thomas Graf <tgraf@noironetworks.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
skb_clone() NULL check is implemented in do_output(), as past of the
common (fast) path. Refactoring so that NULL check is done in the
slow path, immediately after skb_clone() is called.
Besides optimization, this change also improves code readability by
making the skb_clone() NULL check consistent within OVS datapath
module.
Signed-off-by: Andy Zhou <azhou@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
There are many possible ways that a flow can be invalid so we've
added logging for most of them. This adds logs for the remaining
possible cases so there isn't any ambiguity while debugging.
CC: Federico Iezzi <fiezzi@enter.it>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Thomas Graf <tgraf@noironetworks.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
These two cases used to be treated differently for IPv4/IPv6,
but they are now identical.
Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Ths simplifies flow-table-destroy API. No need to pass explicit
parameter about context.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@redhat.com>
Allow datapath to recognize and extract MPLS labels into flow keys
and execute actions which push, pop, and set labels on packets.
Based heavily on work by Leo Alterman, Ravi K, Isaku Yamahata and Joe Stringer.
Cc: Ravi K <rkerur@gmail.com>
Cc: Leo Alterman <lalterman@nicira.com>
Cc: Isaku Yamahata <yamahata@valinux.co.jp>
Cc: Joe Stringer <joe@wand.net.nz>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Device can export MPLS GSO support in dev->mpls_features same way
it export vlan features in dev->vlan_features. So it is safe to
remove NETIF_F_GSO_MPLS redundant flag.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
When filling netlink info, dport is being returned as flags. Fix
instances to return correct value.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Let the tasklet only be enabled after open(), and be disabled for
the other situation. The tasklet is only necessary after open() for
tx/rx, so it could be disabled by default.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It has been reported that generating an MLD listener report on
devices with large MTUs (e.g. 9000) and a high number of IPv6
addresses can trigger a skb_over_panic():
skbuff: skb_over_panic: text:ffffffff80612a5d len:3776 put:20
head:ffff88046d751000 data:ffff88046d751010 tail:0xed0 end:0xec0
dev:port1
------------[ cut here ]------------
kernel BUG at net/core/skbuff.c:100!
invalid opcode: 0000 [#1] SMP
Modules linked in: ixgbe(O)
CPU: 3 PID: 0 Comm: swapper/3 Tainted: G O 3.14.23+ #4
[...]
Call Trace:
<IRQ>
[<ffffffff80578226>] ? skb_put+0x3a/0x3b
[<ffffffff80612a5d>] ? add_grhead+0x45/0x8e
[<ffffffff80612e3a>] ? add_grec+0x394/0x3d4
[<ffffffff80613222>] ? mld_ifc_timer_expire+0x195/0x20d
[<ffffffff8061308d>] ? mld_dad_timer_expire+0x45/0x45
[<ffffffff80255b5d>] ? call_timer_fn.isra.29+0x12/0x68
[<ffffffff80255d16>] ? run_timer_softirq+0x163/0x182
[<ffffffff80250e6f>] ? __do_softirq+0xe0/0x21d
[<ffffffff8025112b>] ? irq_exit+0x4e/0xd3
[<ffffffff802214bb>] ? smp_apic_timer_interrupt+0x3b/0x46
[<ffffffff8063f10a>] ? apic_timer_interrupt+0x6a/0x70
mld_newpack() skb allocations are usually requested with dev->mtu
in size, since commit 72e09ad107 ("ipv6: avoid high order allocations")
we have changed the limit in order to be less likely to fail.
However, in MLD/IGMP code, we have some rather ugly AVAILABLE(skb)
macros, which determine if we may end up doing an skb_put() for
adding another record. To avoid possible fragmentation, we check
the skb's tailroom as skb->dev->mtu - skb->len, which is a wrong
assumption as the actual max allocation size can be much smaller.
The IGMP case doesn't have this issue as commit 57e1ab6ead
("igmp: refine skb allocations") stores the allocation size in
the cb[].
Set a reserved_tailroom to make it fit into the MTU and use
skb_availroom() helper instead. This also allows to get rid of
igmp_skb_size().
Reported-by: Wei Liu <lw1a2.jing@gmail.com>
Fixes: 72e09ad107 ("ipv6: avoid high order allocations")
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: David L Stevens <david.stevens@oracle.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Using a single fixed string is smaller code size than using
a format and many string arguments.
Reduces overall code size a little.
$ size net/ipv4/igmp.o* net/ipv6/mcast.o* net/ipv6/ip6_flowlabel.o*
text data bss dec hex filename
34269 7012 14824 56105 db29 net/ipv4/igmp.o.new
34315 7012 14824 56151 db57 net/ipv4/igmp.o.old
30078 7869 13200 51147 c7cb net/ipv6/mcast.o.new
30105 7869 13200 51174 c7e6 net/ipv6/mcast.o.old
11434 3748 8580 23762 5cd2 net/ipv6/ip6_flowlabel.o.new
11491 3748 8580 23819 5d0b net/ipv6/ip6_flowlabel.o.old
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
By default the arch_fast_hash hashing function pointers are initialized
to jhash(2). If during boot-up a CPU with SSE4.2 is detected they get
updated to the CRC32 ones. This dispatching scheme incurs a function
pointer lookup and indirect call for every hashing operation.
rhashtable as a user of arch_fast_hash e.g. stores pointers to hashing
functions in its structure, too, causing two indirect branches per
hashing operation.
Using alternative_call we can get away with one of those indirect branches.
Acked-by: Daniel Borkmann <dborkman@redhat.com>
Cc: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tom Lendacky says:
====================
amd-xgbe: AMD XGBE driver updates 2014-11-04
The following series of patches includes functional updates to the
driver as well as some trivial changes for function renaming and
spelling fixes.
- Move channel and ring structure allocation into the device open path
- Rename the pre_xmit function to dev_xmit
- Explicitly use the u32 data type for the device descriptors
- Use page allocation for the receive buffers
- Add support for split header/payload receive
- Add support for per DMA channel interrupts
- Add support for receive side scaling (RSS)
- Add support for ethtool receive side scaling commands
- Fix the spelling of descriptors
- After a PCS reset, sync the PCS and PHY modes
- Add dependency on HAS_IOMEM to both the amd-xgbe and amd-xgbe-phy
drivers
This patch series is based on net-next.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The amd-xgbe-phy driver needs to perform ioremap calls, so add HAS_IOMEM
to its build dependency.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The amd-xgbe driver needs to perform ioremap calls, so add HAS_IOMEM
to its build dependency.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support to sync the states of the PCS and the PHY
after a reset is performed. If the PCS and the PHY are not in the
same state after reset an extra mode change would be performed. This
extra mode change might not be needed if the PCS and the PHY are
synced up after reset.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes the spelling of the word "descriptor" in a couple
of locations.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for ethtool receive side scaling (RSS) commands.
Support is added to get/set the RSS hash key and the RSS lookup table.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch provides support for receive side scaling (RSS). RSS allows
for spreading incoming network packets across the Rx queues. When used
in conjunction with the per DMA channel interrupt support, this allows
the receive processing to be spread across multiple processors.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch provides support for interrupts that are generated by the
Tx/Rx DMA channel pairs of the device. This allows for Tx and Rx
processing to run across multiple processsors.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Provide support for splitting IP packets so that the header and
payload can be sent to different DMA addresses. This will allow
the IP header to be put into the linear part of the skb while the
payload can be added as frags.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use page allocations for Rx buffers instead of pre-allocating skbs
of a set size.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Tx and Rx descriptors are unsigned 32 bit values. Use the u32
type, rather than unsigned int, to map these descriptors.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The pre_xmit function name implies that it performs operations prior
to transmitting the packet when in fact it is responsible for setting
up the descriptors and initiating the transmit. Rename this to
function from pre_xmit to dev_xmit, which is consistent with the name
used during receive processing - dev_read.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the channel and ring tracking structures allocation to device
open. This will allow for future support to vary the number of Tx/Rx
queues without unloading the module.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
if_bridge.h uses struct in6_addr ip6, but wasn't including the in6.h
header. Thomas Backlund originally sent a patch to do this, but this
revealed a redefinition issue: https://lkml.org/lkml/2013/1/13/116
The redefinition issue should have been fixed by the following Linux
commits:
ee262ad827 inet: defines IPPROTO_* needed for module alias generation
cfd280c912 net: sync some IP headers with glibc
and the following glibc commit:
6c82a2f8d7c8e21e39237225c819f182ae438db3 Coordinate IPv6 definitions for Linux and glibc
so actually include the header now.
Reported-by: Colin Guthrie <colin@mageia.org>
Reported-by: Christiaan Welvaart <cjw@daneel.dyndns.org>
Reported-by: Thomas Backlund <tmb@mageia.org>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Gregory Fong <gregory.0xf0@gmail.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ueki Kohei reported that when we are using NewReno with connections that
have a very low traffic, we may timeout the connection too early if a
second loss occurs after the first one was successfully acked but no
data was transfered later. Below is his description of it:
When SACK is disabled, and a socket suffers multiple separate TCP
retransmissions, that socket's ETIMEDOUT value is calculated from the
time of the *first* retransmission instead of the *latest*
retransmission.
This happens because the tcp_sock's retrans_stamp is set once then never
cleared.
Take the following connection:
Linux remote-machine
| |
send#1---->(*1)|--------> data#1 --------->|
| | |
RTO : :
| | |
---(*2)|----> data#1(retrans) ---->|
| (*3)|<---------- ACK <----------|
| | |
| : :
| : :
| : :
16 minutes (or more) :
| : :
| : :
| : :
| | |
send#2---->(*4)|--------> data#2 --------->|
| | |
RTO : :
| | |
---(*5)|----> data#2(retrans) ---->|
| | |
| | |
RTO*2 : :
| | |
| | |
ETIMEDOUT<----(*6)| |
(*1) One data packet sent.
(*2) Because no ACK packet is received, the packet is retransmitted.
(*3) The ACK packet is received. The transmitted packet is acknowledged.
At this point the first "retransmission event" has passed and been
recovered from. Any future retransmission is a completely new "event".
(*4) After 16 minutes (to correspond with retries2=15), a new data
packet is sent. Note: No data is transmitted between (*3) and (*4).
The socket's timeout SHOULD be calculated from this point in time, but
instead it's calculated from the prior "event" 16 minutes ago.
(*5) Because no ACK packet is received, the packet is retransmitted.
(*6) At the time of the 2nd retransmission, the socket returns
ETIMEDOUT.
Therefore, now we clear retrans_stamp as soon as all data during the
loss window is fully acked.
Reported-by: Ueki Kohei
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Marcelo Ricardo Leitner <mleitner@redhat.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is only used in net/ipv6/inet6_hashtables.c.
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This encapsulates all of the skb_copy_datagram_iovec() callers
with call argument signature "skb, offset, msghdr->msg_iov, length".
When we move to iov_iters in the networking, the iov_iter object will
sit in the msghdr.
Having a helper like this means there will be less places to touch
during that transformation.
Based upon descriptions and patch from Al Viro.
Signed-off-by: David S. Miller <davem@davemloft.net>
Tom Herbert says:
====================
gue: Remote checksum offload
This patch set implements remote checksum offload for
GUE, which is a mechanism that provides checksum offload of
encapsulated packets using rudimentary offload capabilities found in
most Network Interface Card (NIC) devices. The outer header checksum
for UDP is enabled in packets and, with some additional meta
information in the GUE header, a receiver is able to deduce the
checksum to be set for an inner encapsulated packet. Effectively this
offloads the computation of the inner checksum. Enabling the outer
checksum in encapsulation has the additional advantage that it covers
more of the packet than the inner checksum including the encapsulation
headers.
Remote checksum offload is described in:
http://tools.ietf.org/html/draft-herbert-remotecsumoffload-01
The GUE transmit and receive paths are modified to support the
remote checksum offload option. The option contains a checksum
offset and checksum start which are directly derived from values
set in stack when doing CHECKSUM_PARTIAL. On receipt of the option, the
operation is to calculate the packet checksum from "start" to end of
the packet (normally derived for checksum complete), and then set
the resultant value at checksum "offset" (the checksum field has
already been primed with the pseudo header). This emulates a NIC
that implements NETIF_F_HW_CSUM.
The primary purpose of this feature is to eliminate cost of performing
checksum calculation over a packet when encpasulating.
In this patch set:
- Move fou_build_header into fou.c and split it into a couple of
functions
- Enable offloading of outer UDP checksum in encapsulation
- Change udp_offload to support remote checksum offload, includes
new GSO type and ensuring encapsulated layers (TCP) doesn't try to
set a checksum covered by RCO
- TX support for RCO with GUE. This is configured through ip_tunnel
and set the option on transmit when packet being encapsulated is
CHECKSUM_PARTIAL
- RX support for RCO with GUE for normal and GRO paths. Includes
resolving the offloaded checksum
v2:
Address comments from davem: Move accounting for private option
field in gue_encap_hlen to patch in which we add the remote checksum
offload option.
Testing:
I ran performance numbers using netperf TCP_STREAM and TCP_RR with 200
streams, comparing GUE with and without remote checksum offload (doing
checksum-unnecessary to complete conversion in both cases). These
were run on mlnx4 and bnx2x. Some mlnx4 results are below.
GRE/GUE
TCP_STREAM
IPv4, with remote checksum offload
9.71% TX CPU utilization
7.42% RX CPU utilization
36380 Mbps
IPv4, without remote checksum offload
12.40% TX CPU utilization
7.36% RX CPU utilization
36591 Mbps
TCP_RR
IPv4, with remote checksum offload
77.79% CPU utilization
91/144/216 90/95/99% latencies
1.95127e+06 tps
IPv4, without remote checksum offload
78.70% CPU utilization
89/152/297 90/95/99% latencies
1.95458e+06 tps
IPIP/GUE
TCP_STREAM
With remote checksum offload
10.30% TX CPU utilization
7.43% RX CPU utilization
36486 Mbps
Without remote checksum offload
12.47% TX CPU utilization
7.49% RX CPU utilization
36694 Mbps
TCP_RR
With remote checksum offload
77.80% CPU utilization
87/153/270 90/95/99% latencies
1.98735e+06 tps
Without remote checksum offload
77.98% CPU utilization
87/150/287 90/95/99% latencies
1.98737e+06 tps
SIT/GUE
TCP_STREAM
With remote checksum offload
9.68% TX CPU utilization
7.36% RX CPU utilization
35971 Mbps
Without remote checksum offload
12.95% TX CPU utilization
8.04% RX CPU utilization
36177 Mbps
TCP_RR
With remote checksum offload
79.32% CPU utilization
94/158/295 90/95/99% latencies
1.88842e+06 tps
Without remote checksum offload
80.23% CPU utilization
94/149/226 90/95/99% latencies
1.90338e+06 tps
VXLAN
TCP_STREAM
35.03% TX CPU utilization
20.85% RX CPU utilization
36230 Mbps
TCP_RR
77.36% CPU utilization
84/146/270 90/95/99% latencies
2.08063e+06 tps
We can also look at CPU time in csum_partial using perf (with bnx2x
setup). For GRE with TCP_STREAM I see:
With remote checksum offload
0.33% TX
1.81% RX
Without remote checksum offload
6.00% TX
0.51% RX
I suspect the fact that time in csum_partial noticably increases
with remote checksum offload for RX is due to taking the cache miss on
the encapsulated header in that function. By similar reasoning, if on
the TX side the packet were not in cache (say we did a splice from a
file whose data was never touched by the CPU) the CPU savings for TX
would probably be more pronounced.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add processing of the remote checksum offload option in both the normal
path as well as the GRO path. The implements patching the affected
checksum to derive the offloaded checksum.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add if_tunnel flag TUNNEL_ENCAP_FLAG_REMCSUM to configure
remote checksum offload on an IP tunnel. Add logic in gue_build_header
to insert remote checksum offload option.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define a private flag for remote checksun offload as well as a length
for the option.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a new GSO type, SKB_GSO_TUNNEL_REMCSUM, which indicates remote
checksum offload being done (in this case inner checksum must not
be offloaded to the NIC).
Added logic in __skb_udp_tunnel_segment to handle remote checksum
offload case.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Malcolm Crossley