Introduce new classify() method that uses AVX2 instructions.
>From my measurements:
On HSW boards when processing >= 16 packets per call,
AVX2 method outperforms it's SSE counterpart by 10-25%,
(depending on the ruleset).
When build with the compilers that don't support AVX2 instructions,
make rte_acl_classify_avx2() do nothing and return an error.
At runtime, if librte_acl was build with the compiler that supports AVX2,
this method is selected as default one on HW that supports AVX2.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
New data type to manipulate 256 bit AVX values.
Rename field in the rte_xmm to keep common naming across SSE/AVX fields.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Move common check for input parameters up into rte_acl_classify_alg().
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Make classify_scalar to behave in the same way as it's vector counterpart:
move match check out of the inner loop, etc.
That makes scalar and vector code look more identical.
Plus it improves scalar code performance.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Right now we allocate indexes for all types of nodes, except MATCH,
at 'gen final RT table' stage.
For MATCH type nodes we are doing it at building temporary tree stage.
This is totally unnecessary and makes code more complex and error prone.
Rework the code and make MATCH indexes being allocated at the same stage
as all others.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Introduced division of whole 256 child transition enties
into 4 sub-groups (64 kids per group).
So 2 groups within the same node with identical children,
can use one set of transition entries.
That allows to compact some DFA nodes and get space savings in the RT table,
without any negative performance impact.
>From what I've seen an average space savings: ~20%.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
There was a bug at build phase that can cause matches beeing overwritten.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Current rule-wildness based heuristics can cause unnecessary splits of
the ruleset.
That might have negative performance effect:
more tries to traverse, bigger RT tables.
After removing it, on some test-cases with big rulesets (~10K)
observed ~50% speedup.
No difference for smaller rulesets.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Make data_indexes long enough to survive idle transitions.
That allows to simplify match processing code.
Also fix incorrect size calculations for data indexes.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Fix compilation with RTE_LIBRTE_ACL_STANDALONE=y
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
For ixgbe, when queue start fails, for example, mbuf allocate
failure, the device will still start successfully, which could be
an issue.
Add return status check of queue start to avoid this issue.
Signed-off-by: Michael Qiu <michael.qiu@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
To follow up the comments from Pawel Wodkowski, remove this unnecessary check,
as check_mq_mode has already check the queue number in device configure stage,
if the queue number of vf is not correct, it will return error code and exit,
so it doesn't need check again here in device start stage (note: pf_host_configure
is called in device start stage).
Fixes: 42d2f78abc ("configure VF RSS")
Suggested-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
The link_status variable is not set when device is initialized.
This can lead to problems with link never being reported as up
if using some SFP modules where the link is instantly on.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
The rte_eth_stats_get is the only API that should call the device
statistics function directly, and it already does a memset of the
resulting structure since commit 02331c16ec. Therefore doing
memset() in the driver is redundant and should be removed.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
[David: remove also in igbvf and pcap PMDs]
Acked-By: David Marchand <david.marchand@6wind.com>
Add missing extern 'C' decls in rte_ip_frag.h.
Fixes: 601e279df0 ("move fragmentation/reassembly headers into a library")
Signed-off-by: Marc Sune <marc.sune@bisdn.de>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
rte_power_freq_min function did not include "extern" keyword,
causing linking errors.
Fixes: 445c6528b5 ("power: common interface for guest and host")
Reported-by: Ildar Mustafin <imustafin@bk.ru>
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Ethernet device's data should contain the virtual device name for pcap port.
This name is correctly set by rte_eth_dev_allocate() at initialization time,
but it is directly lost.
Fixes: 83b4113693 ("ethdev: add unique name to devices")
Signed-off-by: Remi Pommarel <repk@triplefau.lt>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
When using core list argument to define which core to enable (ie -l) the
core_num field of the rte configuration is not updated the same way as using
coremask. This causes rte_lcore_num() to yield different value from the one
using coremask.
Fixes: d888cb8b96 ("add core list input format")
Signed-off-by: Remi Pommarel <repk@triplefau.lt>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Structure rte_ethertype_filter is removed.
Following APIs are removed:
- rte_eth_dev_add_ethertype_filter
- rte_eth_dev_remove_ethertype_filter
- rte_eth_dev_get_ethertype_filter
It is replaced by filter_ctrl API.
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
This patch removes old functions which deal with ethertype filter in ixgbe driver.
It also defines ixgbe_dev_filter_ctrl which is binding to filter_ctrl API,
and ethertype filter can be dealt with through this new entrance.
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
This patch removes old functions which deal with ethertype filter in igb driver.
It also defines eth_igb_filter_ctrl which is binding to filter_ctrl API,
and ethertype filter can be dealt with through this new entrance.
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
In commit 2fc8d6d the behaviour of function rte_is_power_of_2 was
changed to not return true for 0. memzone_reserve_aligned_thread_unsafe
and rte_malloc_socket both make the assumption that for align = 0
!rte_is_power_of_2(align) will return false. This patch adds a check
that align parameter is non-zero before doing the power of 2 check.
Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
[Thomas: use && operator instead of ternary ?: and fix precedence with parens]
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
It needs config RSS and IXGBE_MRQC and IXGBE_VFPSRTYPE to enable VF RSS.
The psrtype will determine how many queues the received packets will distribute to,
and the value of psrtype should depends on both facet: max VF rxq number which
has been negotiated with PF, and the number of rxq specified in config on guest.
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Reviewed-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Check mq mode for VMDq RSS, handle it correctly instead of returning an error;
Also remove the limitation of per pool queue number has max value of 1, because
the per pool queue number could be 2 or 4 if it is VMDq RSS mode;
The number of rxq specified in config will determine the mq mode for VMDq RSS.
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Reviewed-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Get the available Rx and Tx queue number when receiving IXGBE_VF_GET_QUEUES
message from VF.
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Reviewed-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Negotiate API version with VF when receiving the IXGBE_VF_API_NEGOTIATE message.
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Reviewed-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Put global register configuring out of loop for queue; also fix typo and indent.
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Reviewed-by: Vlad Zolotarov <vladz@cloudius-systems.com>
The lack of result checking of fscanf function, breaks compilation
for default "-Werror=unused-result" flag.
Signed-off-by: Tomasz Kulasek <tomaszx.kulasek@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Replace d_thread_t with struct thread in nic_uio.
Ref: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=196691
Quote:
"The d_thread_t typedef is a compat shim to support FreeBSD 4.x.
I'm planning to remove this shim from 11 and dpdk is very unlikely
to ever be ported to 4.x.
If it does it will need far more changes than just d_thread_t"
Reported-by: John Baldwin <jhb@freebsd.org>
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
This patch contains a fix for link bonding handling of vlan tagged packets in mode 3 and 5.
Currently xmit_slave_hash function misinterprets the PKT_RX_VLAN_PKT flag to mean that
there is a vlan tag within the packet when in actually means that there is a valid entry
in the vlan_tci field in the mbuf.
- Fixed VLAN tag support in hashing functions.
- Adds support for TCP in layer 4 header hashing.
- Splits transmit hashing function into separate functions for each policy to
reduce branching and to make the code clearer.
- Fixed incorrect flag set in test application packet generator.
Test report: http://dpdk.org/ml/archives/dev/2015-January/010792.html
Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Tested-by: SunX Jiajia <sunx.jiajia@intel.com>
When vfio module is not loaded when kernel support vfio feature,
the routine still try to open the container to get file
description.
This action is not safe, and of course got error messages:
EAL: Detected 40 lcore(s)
EAL: unsupported IOMMU type!
EAL: VFIO support could not be initialized
EAL: Setting up memory...
This may make user confuse, this patch make it reasonable
and much more smooth to user.
Signed-off-by: Michael Qiu <michael.qiu@intel.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
The read/seek/close stub functions are unnecessary on the
log stream. Per glibc fopencookie man page:
cookie_read_function_t *read
If *read is a null pointer, then reads from the custom stream
always return end of file.
cookie_seek_function_t *seek
If *seek is a null pointer, then it is not possible to perform
seek operations on the stream.
cookie_close_function_t *close
If *close is NULL, then no special action is performed when the
stream is closed.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
When scanning the hugetlbfs maps search only for the DPDK maps.
This will allow the application create its own hugetlbfs mappings
and use the DPDK facilities on the same hugetlbfs mount point.
Signed-off-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
rte_is_power_of_2 returns true for 0 and 0 is not power_of_2.
Fix by checking for n.
Signed-off-by: Ravi Kerur <rkerur@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
This patch fixes checking the link state of a virtual function. If the
state has already been checked, it does not need to be checked
again. Previously, get_link_status in the ixgbe_hw struct was used to
track if the information had already been retrieved, but this field
was always set to false (signifying that the information was
up-to-date). The problem was introduced by commit 8ef32003 which was
part of a patch set to update the ixgbe portion of the PMD. This patch
does not break consistency with the ixgbevf driver. Instead, it fixes
the problem at the level of DPDK.
Applications that rely on the reported link speed could fail without
this patch. The qos_sched example application provided with DPDK did
not run when virtual functions were used. The output for this example
application is shown below:
EAL: Error - exiting with code: 1
Cause: Unable to config sched subport 0, err=-2
The problem and the effect of the patch can been seen by running the
l2fwd example application using the following command:
sudo ./build/l2fwd -c 0x3 -n 4 -- -p 0x3 -T 0
Before the patch has been applied (with both links up):
...
Checking link statusdone
Port 0 Link Up - speed 100 Mbps - half-duplex
Port 1 Link Up - speed 100 Mbps - half-duplex
L2FWD: entering main loop on lcore 1
...
After the patch has been applied (with both links up):
...
Checking link statusdone
Port 0 Link Up - speed 10000 Mbps - full-duplex
Port 1 Link Up - speed 10000 Mbps - full-duplex
L2FWD: entering main loop on lcore 1
...
Before the patch has been applied (with link 0 down, link 1 up):
...
Checking link statusdone
Port 0 Link Up - speed 100 Mbps - half-duplex
Port 1 Link Up - speed 100 Mbps - half-duplex
L2FWD: entering main loop on lcore 1
...
After the patch has been applied (with link 0 down, link 1 up):
...
Checking link status............................................................
..............................done
Port 0 Link Down
Port 1 Link Up - speed 10000 Mbps - full-duplex
...
Signed-off-by: Balazs Nemeth <balazs.nemeth@intel.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
EAL: probe driver: 8086:10fb rte_ixgbe_pmd
EAL: PCI memory mapped at 0x7f18c2a00000
EAL: PCI memory mapped at 0x7f18c2a80000
Segmentation fault (core dumped)
This is introduced by commit: 46bc9d75
ixgbe: fix multi-process support
When start primary process with command line:
./app/test/test -n 1 -c ffff -m 64
then start the second one:
./app/test/test -n 1 --proc-type=secondary --file-prefix=rte
This segment-fault will occur.
Root cause is test app on primary process only starts device, but
the queue need initialized by manually command line.
So the tx queue is still NULL when secondary process startup.
Reported-by: Yong Liu <yong.liu@intel.com>
Signed-off-by: Michael Qiu <michael.qiu@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
This patch removes the interrupt registration code which was under the flag
VFIO_PRESENT and relies on the rte_lib code for the same.
This also ignores the initial trigger of ISR from the lib.
Signed-off-by: Sujith Sankar <ssujith@cisco.com>
In rte_pmd_init_internals, we are mapping memory but not released
if error occurs it could produce memory leak.
Add unmmap function to release memory.
Signed-off-by: Daniel Mrzyglod <danielx.t.mrzyglod@intel.com>
Acked-by: Michal Jastrzebski <michalx.k.jastrzebski@intel.com>
Acked-by: John W. Linville <linville@tuxdriver.com>
In rte_eth_af_packet.c we are we are missing NULL pointer
checks after calls to allocate memory for queues.
Add checking NULL pointer and error handling.
Signed-off-by: Daniel Mrzyglod <danielx.t.mrzyglod@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Back in commit aaa662e75c ("cmdline: fix overflow on bsd"),
the author failed to fixup a call to cmdline_parse_etheraddr in xenvirt.
This patch makes the needed correction to avoid a build break.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
This patch fixes the issue whereby when using userspace vhost ports
in the context of vSwitching, the name provided to the hypervisor/QEMU
of the vhost tap device needs to be exposed in the library, in order
for the vSwitch to be able to direct packets to the correct device.
This patch introduces an 'ifname' member to the virtio-net structure
which is populated with the tap device name when QEMU is brought up
with a vhost device.
Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
Signed-off-by: Anthony Fee <anthonyx.fee@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
From CentOS 6.6, function skb_set_hash is introduced, this breaks
the previous assumption. So modify RHEL_RELEASE_VERSION from 7.0
to 6.6 to fix build for rte_kni.ko.
Related mail from Barak Enat:
http://dpdk.org/ml/archives/dev/2014-December/010124.html
building error likes:
CC [M] lib/librte_eal/linuxapp/kni/e1000_82575.o
In file included from lib/librte_eal/linuxapp/kni/ethtool/igb/e1000_osdep.h:41,
from lib/librte_eal/linuxapp/kni/ethtool/igb/e1000_hw.h:31,
from lib/librte_eal/linuxapp/kni/ethtool/igb/e1000_api.h:31,
from lib/librte_eal/linuxapp/kni/e1000_82575.c:38:
lib/librte_eal/linuxapp/kni/ethtool/igb/kcompat.h:3870: error: conflicting types for ‘skb_set_hash’
include/linux/skbuff.h:620: note: previous definition of ‘skb_set_hash’ was here
Reported-by: Barak Enat <barak@saguna.net>
Signed-off-by: Jincheng Miao <jincheng.miao@gmail.com>
Compile warning which is treated as error occurs on Oracle Linux
(kernel 2.6.39, gcc 4.4.7) as below, or RHEL, CentOS. Aliasing
'struct i40e_aqc_debug_reg_read_write' should be avoided. Use the
elements inside that structure directly can fix the issue.
lib/librte_pmd_i40e/i40e_ethdev.c: In function 'eth_i40e_dev_init':
lib/librte_pmd_i40e/i40e_ethdev.c:5318: error: dereferencing pointer
'cmd' does break strict-aliasing rules
lib/librte_pmd_i40e/i40e_ethdev.c:5314: note: initialized from here
Signed-off-by: Helin Zhang <helin.zhang@intel.com>