Demonstrates the use of the ACL library in the DPDK application to
implement packet classification and L3 forwarding.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
[Thomas: some code-style changes]
Usage example and main test application for the ACL library.
Provides IPv4/IPv6 5-tuple classification.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
[Thomas: some code-style changes]
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
The ACL library is used to perform an N-tuple search over a set of rules with
multiple categories and find the best match for each category.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
[Thomas: some code-style changes]
This fixes style problems reported by checkpatch including:
* extra whitespace
* spaces before tabs
* strings broken across lines
* excessively long lines
* missing spaces after keywords
* unnecessary paren's in return statements
Signed-off-by: Stephen Hemminger <shemming@brocade.com>
Acked-by: Changchun Ouyang <changchun.ouyang@intel.com>
Add a set of unit tests and some basic performance test for the
distributor library. These tests cover all the major functionality of
the library on both distributor and worker sides.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
This adds the code for a new Intel DPDK library for packet distribution.
The distributor is a component which is designed to pass packets
one-at-a-time to workers, with dynamic load balancing. Using the RSS
field in the mbuf as a tag, the distributor tracks what packet tag is
being processed by what worker and then ensures that no two packets with
the same tag are in-flight simultaneously. Once a tag is not in-flight,
then the next packet with that tag will be sent to the next available
core.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
[Thomas: add doxygen @file comment]
With latest HW and optimised RX/TX path there is a huge gap between
tespmd iofwd and l3fwd performance results.
So there is an attempt to optimise l3fwd LPM code path and reduce the gap:
- Instead of processing each input packet up to completion -
divide packet processing into several stages and perform
stage by stage for the whole burst.
- Unroll things by the factor of 4 whenever possible.
- Use SSE instincts for some operations (bswap, replace MAC addresses, etc).
- Avoid TX packet buffering whenever possible.
- Move some checks from RX/TX into setup phase.
Note that new(optimized) code path can be switched on/off by setting
ENABLE_MULTI_BUFFER_OPTIMIZE macro to 1/0.
Some performance data:
SUT: dual-socket board IVB 2.8GHz, 2x1GB pages.
4 ports on 4 NICs (all at socket 0) connected to the traffic generator.
kernel: 3.11.3-201.fc19.x86_64, gcc: 4.8.2.
64B packets, using the packet flooding method.
All 4 ports are managed by one logical core:
Optimised scalar PMD RX/TX was used.
DIFF % (NEW-OLD)
IPV4-CONT-BURST: +23%
IPV6-CONT-BURST : +13%
IPV4/IPV6-CONT-BURST: +8%
IPV4-4STREAMSX8: +7%
IPV4-4STREAMSX1: -2%
Test cases description:
IPV4-CONT-BURST - IPV4 packets all packets from the one input port
are destined for the same output port.
IPV6-CONT-BURST - IPV6 packets all packets from the one input port
are destined for the same output port.
IPV4/IPV6-CONT-BURST - mix of the first 2 with interleave=1
(e.g: IPV4,IPV6,IPV4,IPV6, ...)
IPV4-4STREAMSX1 - 4 streams of IPV4 packets, where all packets
from same stream are destined for the same output port
(e.g: IPV4_DST_P0, IPV4_DST_P1, IPV4_DST_P2, IPV4_DST_P3, IPV4_DST_P0, ...)
IPV4-4STREAMSX8 - same as above but packets for each stream
are coming in groups of 8
(e.g: IPV4_DST_P0 X 8, IPV4_DST_P1 X 8, IPV4_DST_P2 X 8, IPV4_DST_P3 X 8,
IPV4_DST_P0 X 8, ...)
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
Allows to lookup four IP addresses in an LPM table.
Uses SSE instrincts.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
Add API to support setting TX rate for a queue and a VF.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Acked-by: Jijiang Liu <jijiang.liu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
This patch adds commands to test the functionality of setting link up and down.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked by: Ivan Boule <ivan.boule@6wind.com>
It is implemented by enabling or disabling TX laser.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked by: Ivan Boule <ivan.boule@6wind.com>
This patch adds API to support the functionality of setting link up and down.
It can be used to repeatedly stop and restart RX/TX of a port without
re-allocating resources for the port and re-configuring the port.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked by: Ivan Boule <ivan.boule@6wind.com>
icc 12.1 complains about RTE_LOG() format:
"argument is incompatible with corresponding format string conversion"
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
If igb_alloc_rx_queue_mbufs() would fail to allocate an mbuf for RX queue,
it calls igb_rx_queue_release(rxq).
That causes rxq to be silently freed, without updating
dev->data->rx_queues[].
So any further reference to it will trigger the SIGSEGV.
Same thing in em PMD too.
To fix: igb_alloc_rx_queue_mbufs() should just return an error to the
caller and let upper layer to deal with the probem.
That's what ixgbe PMD is doing right now.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This commit removes trailing whitespace from lines in files. Almost all
files are affected, as the BSD license copyright header had trailing
whitespace on 4 lines in it [hence the number of files reporting 8 lines
changed in the diffstat].
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
[Thomas: remove spaces before tabs in libs]
[Thomas: remove more trailing spaces in non-C files]
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Add __rte_unused to
pci_unbind_kernel_driver(struct rte_pci_device *dev)
Signed-off-by: Alan Carew <alan.carew@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Recent change to rte_dump_tailq (commit 591a9d7985),
which now uses a FILE parameter causes compilation to fail under FreeBSD
and sourced to a missing include of stdio.h.
Errors:
rte_tailq.h: unknown type name 'FILE' void rte_dump_tailq(FILE *f);
rte_memory.h: unknown type name 'FILE' void rte_dump_physmem_layout(FILE *f);
Signed-off-by: Alan Carew <alan.carew@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Error message for missing template is factorized in notemplate rule.
RTE_OUTPUT directory is marked as order-only prerequisite.
RTE_OUTPUT is always created after having been cleaned for rte_config.h.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by Olivier Matz <olivier.matz@6wind.com>
When running "make config", an additional config.orig file is also
generated, which is intended to hold the original, clean configuration
from the template.
When running make install, we first check if there is no existing
.config file, and run make config if not. If there is a file, we then
check if it's unmodified, in which case we regenerate a new .config to
take account of any possible updates to the template. Finally, in the
case where there is an existing .config file, and it HAS been modified,
we then do a check to see if the template has had further updates, and
throw an error if so. If no updates, we continue with the build using
the existing, user-modified config.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Tested-by: Bruce Richardson <bruce.richardson@intel.com>
Some linker options were not prefixed by -Wl, when using CC:
-z muldefs
-melf_i386 (CPU_LDFLAGS in 32-bit config)
I didn't see any error with -z muldefs but it isn't documented in gcc
manual. So it's safer to explicitly pass it to the linker.
Also building 32-bit shared library raises this error:
gcc: error: unrecognized command line option ‘-melf_i386’
Using macro linkerprefix fixes it.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Reviewed-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
If pcap_sendpacket() fails, then eth_pcap_tx shouldn't silently free that
mbuf and continue.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
The unit of allocated_size is MB, so the change below is made.
Otherwise, it will fail to free memory when available memory is not enough.
Signed-off-by: Jijiang Liu <jijiang.liu@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
Tested-by: Heng Ding <hengx.ding@intel.com>
Since Linux kernel version 3.13.0,
the xen_create/destroy_contiguous_region() API has been changed,
and the first parameter is physical address in the API.
Signed-off-by: Jijiang Liu <jijiang.liu@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
Tested-by: Heng Ding <hengx.ding@intel.com>
The patch changes the way of reserving memory in Dom0 driver.
It will reserve memory at installing rte_dom0_mm.ko kernel module
instead of requesting memory dynamically during DPDK application startup.
Meanwhile, now driver requests memory size of 4M once first,
if it failed, and request memory size of 2M once.
The main reasons for these changes are as follows:
First, to reduce the impact of increasing in memory fragment
after system run a long time.
Second, to reduce number of memory segment.
Signed-off-by: Jijiang Liu <jijiang.liu@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This patch supports multiple queues feature in DPDK based virtio-net frontend.
It firstly gets max queue number of virtio-net from virtio PCI configuration and
then send command to negotiate the queue number with backend; When receiving and
transmitting packets, it negotiates multiple virtio-net queues which serve RX/TX;
To utilize this feature, the backend also need support multiple queues feature
and enable it.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This patch cleanups some coding style issue, and fixes some errors and warnings
reported by checkpatch.pl.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This patch supports user space vhost zero copy. It removes packets copying between host and guest in RX/TX.
It introduces an extra ring to store the detached mbufs. At initialization stage all mbufs will put into
this ring; when one guest starts, vhost gets the available buffer address allocated by guest for RX and
translates them into host space addresses, then attaches them to mbufs and puts the attached mbufs into
mempool.
Queue starting and DMA refilling will get mbufs from mempool and use them to set the DMA addresses.
For TX, it gets the buffer addresses of available packets to be transmitted from guest and translates
them to host space addresses, then attaches them to mbufs and puts them to TX queues.
After TX finishes, it pulls mbufs out from mempool, detaches them and puts them back into the extra ring.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This patch implements queue start and stop functionality in IXGBE PMD;
it also enable hardware loopback for VMDQ mode in IXGBE PMD.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This patch adds API to support queue start and stop functionality for RX/TX.
It allows RX and TX queue is started or stopped one by one, instead of starting
and stopping all of them at the same time.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Add the command "port config X rss-hash-key key" in the 'testpmd'
application to configure the RSS hash key used to compute the RSS
hash of input [IP] packets received on port X.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
1) Add a new function "rss_hash_conf_get" in the PMD API to retrieve the
current configuration of the RSS functions and/or of the RSS key used
by a NIC to compute the RSS hash of input packets.
The new function uses the existing data structure "rte_eth_rss_conf" for
returning the RSS hash configuration.
2) Add the ixgbe-specific function "ixgbe_dev_rss_hash_conf_get" and the
igb-specific function "eth_igb_rss_hash_conf_get" to retrieve the RSS
hash configuration of ixgbe and igb controllers respectively.
3) Add the command "show port X rss-hash [key]" in the testpmd application
to display the RSS hash configuration of port X.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
The function cmd_config_rss_parsed() associated with the command
"port config rss all" required to first stop all ports, in order to
then entirely re-configure all ports with the new RSS hash computation
parameters.
Use now the new function rte_eth_dev_rss_hash_conf_update() that dynamically
only changes the RSS hash computation parameters of a port, without needing
to previously stop the port.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
1) Add a new function "rss_hash_update" in the PMD API to dynamically
update the RSS flags and/or the RSS key used by a NIC to compute the RSS
hash of input packets.
The new function uses the existing data structure "rte_eth_rss_conf" for
the argument that contains the new hash flags and/or the new hash key to
use.
2) Add the ixgbe-specific function "ixgbe_dev_rss_hash_update" and the
igb-specific function "eth_igb_rss_hash_update" to update the RSS
hash configuration of ixgbe and igb controllers respectively.
Before changing anything, these 2 functions check that the update RSS
operation does not attempt to disable RSS, if RSS was enabled at port
initialization time, or does not attempt to enable RSS, if RSS was
disabled at port initialization time.
Note:
Configuring the RSS hash flags and the RSS key used by a NIC consists in
updating appropriate PCI registers of the NIC.
These operations have been manually tested with the interactive commands
"write reg" and "write regbit" of the testpmd application.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Each entry of the RSS redirection table (RETA) of igb and ixgbe ports
contains a 4-bit RX queue index, thus imposing RSS RX queue indices to
be strictly lower than 16.
In addition, if a RETA entry is configured with a RX queue index that is
strictly lower than 16, but is greater or equal to the number of RX queues
of the port, then all input packets whose RSS hash value indexes that RETA
entry are silently dropped by the NIC.
Make the function rte_eth_dev_rss_reta_update() check that RX queue indices
that are supplied in the reta_conf argument are strictly lower than
ETH_RSS_RETA_MAX_QUEUE (16) and are strictly lower than the number of
RX queues of the port.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
e1000_vfadapt type corresponds to 82576 VF devices,
check e1000_set_mac_type() for more details.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
When initializing a VF with no initial MAC address assigned by
the underlying Host PF driver, assign a default MAC address.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
The VF_RESET message of the 82599 PF/VF communication protocol issued by a
a Guest VF driver may include an optional permanent MAC address assigned to
the VF by the Guest OS, in order to make it recorded into the 82599 RAR
registers by the Host PF driver.
To indicate the absence of this optional MAC address, the VF_RESET command
assumes that a NULL MAC address is sent, instead of using a dedicated bit
for this purpose. However, when sending a VF_RESET command with no permanent
MAC address, the function ixgbe_reset_hw_vf() of the 82599 VF driver
directly invokes the function ixgbe_write_mbx_vf() with a message that does
not include a NULL MAC address, wrongly assuming that this function fills in
with zero all unused mailbox data registers.
More globally, it is safer to explicitely reset to zero all remaining mailbox
data registers that are not used to store the content of a message, in order
to reset the data sent in a previous VF/PF exchange (in either side),
including the last exchange performed by another Guest OS to which that VF
was previously assigned.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
On a 82599 VF, the deletion of a dynamically added MAC address consists in
first flushing all added MAC addresses, then in adding again all remaining MAC
addresses.
For this purpose, the function ixgbevf_remove_mac_addr() parses the pool
of MAC addresses associated with a VF, and must skip the VF permanent MAC
address that is stored into it, as well as all NULL MAC addresses.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
During the initialization of a VF device, the rte_eth_dev_start() function
indirectly invokes the PMD "mac_addr_add" function with the permanent MAC
address assigned to the device.
In the case of 82599 VFs, this operation leads to exhausting the very
limited set of PF resources used to store VF MAC addresses.
To address this issue, do nothing in the function ixgbevf_add_mac_addr()
if the added MAC address is equal to the permanent MAC address of the VF.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Add missing PMD functions in the ixgbevf driver to add (respectively remove)
a MAC address to/from a 82599 VF.
For this purpose, these 2 functions use the VF/PF mailbox-based protocol.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
When latest Linux ixgbe PF is used, and DPDK VF is used in DPDK application,
jumbo frames are not received.
Also - if Linux ixgbe PF has MTU set to 1500 (default),
then normal sized packets can be received by DPDK VF.
However, if Linux PF has MTU > 1500, then DPDK VF receives no packets
(normal or jumbo).
With ixgbe_mbox_api_10 ixgbe simply didn't allow set VF MTU > 1514 for 82599.
With ixgbe_mbox_ajpi_11 it does, though now, if PF uses jumbo frames,
it simply disables RX for all VFs.
So to work with PF ithat using jumbo frames, at startup each VF has to:
1. negotiate with PF mbox_api_11.
2. Send to PF SET_LPE message with desired MTU.
Note, that if PF already uses MTU bigger then asked by the VF,
then PF wouldn't take any action.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Bug: When a timer is running
- if rte_timer_stop is called, the pending decrement is
skipped (decremented only if the timer is pending) and due
to the update flag the future processing is skipped so the
timer is counted as pending while it is stopped. - the same
applies when rte_timer_reset is called but then the pending
statistics is additionally incremented so the timer is
counted pending twice.
Solution: decrement the pending
statistics after returning from the callback. If
rte_timer_stop was called, it skipped decrementing the
pending statistics. If rte_time_reset was called, the
pending statistics was incremented. If neither was called
and the timer is periodic, the pending statistics is
incremented when it is reloaded
Signed-off-by: Vadim Suraev <vadim.suraev@gmail.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
