Reduce host CPU overhead of Tx packet processing:
* Use local variables inside per-packet loop instead of fields in structs.
* Factor book keeping and conditionals out of the per-packet loop where
possible.
* Post buffers to the nic at a maximum of every 64 packets
Signed-off-by: Nelson Escobar <neescoba@cisco.com>
Signed-off-by: John Daley <johndale@cisco.com>
Mbufs were returned to the pool one at a time. Use rte_mempool_put_bulk
instead. There were multiple function calls for each buffer returned.
Refactor this code into just 2 functions.
Signed-off-by: John Daley <johndale@cisco.com>
The NIC can either DMA a separate completion message for each completed
send or periodically just DMA the index of the last completed send.
Switch to the latter method which improves cache locality and performance.
Signed-off-by: John Daley <johndale@cisco.com>
The list of mbufs held by the driver on Tx was allocated in chunks
(a hold-over from the enic kernel mode driver). The structure used
next pointers across chunks which led to cache misses.
Allocate the array used to hold mbufs in flight on Tx with
rte_zmalloc_socket(). Remove unnecessary fields from the structure
and use head and tail pointers instead of next pointers.
Signed-off-by: John Daley <johndale@cisco.com>
Functions existed which were never called. Removed them. Also
rename the 'pmd' from the name of the Tx function to improve clarity.
Signed-off-by: John Daley <johndale@cisco.com>
The Tx functions were in enic_ethdev.c and enic_main.c - files in which
they did not logically belong. To make things consistent with most
other drivers, we therefore extract them and place them with the equivalent
Rx functions into a file called enic_rxtx.c.
Signed-off-by: John Daley <johndale@cisco.com>
Truncated packets occur on enic if an mbuf is not big enough to
receive it or there aren't enough mbufs if rx scatter is in use.
They show up as error packets but unlike other error packets (like
packets bad FCS) there are no nic drop counts incremented for them.
Truncated packets are calculated by subtracting hardware errors from
software errors. Note: this causes transient inaccuracies in the
ipackets count. Also, the length of truncated packets are counted
in ibytes even though truncated packets are dropped which can make
ibytes be slightly higher than it should be.
Signed-off-by: Nelson Escobar <neescoba@cisco.com>
Signed-off-by: John Daley <johndale@cisco.com>
Following the discussions from:
http://dpdk.org/ml/archives/dev/2015-July/021721.htmlhttp://dpdk.org/ml/archives/dev/2016-April/038143.html
Remove the unused flag from enic driver. Also, the enic driver is
now modified to drop bad packets instead of using a non-existent
flag to try and identify them as bad.
Fixes: 947d860c82 ("enic: improve Rx performance")
Fixes: 5776c30293 ("enic: fix error packets handling")
Fixes: 50765c820e ("enic: remove packet error conditional")
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Signed-off-by: John Daley <johndale@cisco.com>
rx_no_bufs is a hardware counter of packets dropped on the
interface due to no host buffers and should be used to update
r_stats->imissed counter instead of rx_nombuf.
Include rx_drop in ierrors. rx_drop is incremented if packets
arrive when the receive queue is disabled.
Add a structure and functions for initializing and clearing
software counters. Add count of Rx mbuf allocation failures
(rx_nombuf) as the first counter.
Fixes: fefed3d1e6 ("enic: new driver")
Signed-off-by: John Daley <johndale@cisco.com>
GCC_VERSION is empty in case of clang:
/bin/sh: line 0: test: -ge: unary operator expected
It is the same issue as http://dpdk.org/dev/patchwork/patch/5994/
Fixes: 366113dbfb ("e1000: suppress misleading indentation warning")
Signed-off-by: Hiroyuki Mikita <h.mikita89@gmail.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: John W. Linville <linville@tuxdriver.com>
Suspicious implicit sign extension: pf->fdir.match_counter_index
with type unsigned short (16 bits, unsigned) is promoted in
"pf->fdir.match_counter_index << 20" to type int (32 bits, signed),
then sign-extended to type unsigned long (64 bits, unsigned).
If "pf->fdir.match_counter_index << 20" is greater than 0x7FFFFFFF,
the upper bits of the result will all be 1.
To fix the issue explicitly cast pf->fdir.match_counter_index to uint32_t.
Coverity issue: 13315
Fixes: 05999aab4c ("i40e: add or delete flow director")
Signed-off-by: Slawomir Mrozowicz <slawomirx.mrozowicz@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
This patch enables configuring MTU for i40e.
Since changing MTU needs to reconfigure queue, the port must be
stopped before configuring MTU.
Signed-off-by: Beilei Xing <beilei.xing@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
When setting up the flexible paylaod selection rules, the value
NONUSE_FLX_PIT_DEST_OFF (== 63) is meant to disable the rule.
However, since the MK_FLX_PIT macro always added on an additional
offset of I40E_FLX_OFFSET_IN_FIELD_VECTOR (== 50) to the value passed
the functionality to disable the rule was broken.
This patch fixes this by checking for the disable value and not adding
the offset in that case.
Fixes: d8b90c4eab ("i40e: take flow director flexible payload configuration")
Reported-by: Michael Habibi <mikehabibi@gmail.com>
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Zhe Tao <zhe.tao@intel.com>
Previously, there was a known issue "On Intel® 40G Ethernet
Controller stopping the port does not really down the port link."
There were two reasons why the port was always kept up.
1. Old firmware versions had issues when "Set PHY config command"
was used on 40G NICs.
2. The kernel i40e driver didn't call "Set PHY config command" when
ifconfig up/down was used, it assumes the link is always up. But
in DPDK, ports are forced down when an applications quits. So if
the port is then switched to being controlled by kernel the driver,
the port can not be brought up through "ifconfig <ethx> up".
This patch fixes this issue by adding in "Set PHY config command"
into our driver. This is now possible because with newer firmware
there is no longer a problem using this command.
With this fix, after DPDK quit, if the port is switched to being used
by the kernel driver, "ethtool -s <ethx> autoneg on" can be used to
turn on the auto negotiation, and then port can be brought up through
"ifconfig <ethx> up".
NOTE: requires kernel i40e driver version >= 1.4.X
Fixes: 2f1e228174 ("i40e: skip link control as firmware workaround")
Fixes: 16c979f9ad ("i40e: disable setting of PHY configuration")
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Change the Tx routine to ring the doorbell once per burst
and not on every Tx packet. This driver-level optimization
is necessary to achieve line rates for larger frame
sizes (1k or more).
Signed-off-by: Rasesh Mody <rasesh.mody@qlogic.com>
Signed-off-by: Harish Patil <harish.patil@qlogic.com>
- Process Tx completions based on configured Tx free threshold and
determine how much TX BDs are required before invoking bnx2x_tx_encap()
- Change bnx2x_tx_encap() to void function as it can now never fail
Signed-off-by: Rasesh Mody <rasesh.mody@qlogic.com>
Signed-off-by: Harish Patil <harish.patil@qlogic.com>
Under certain scenarios, management firmware (MFW) periodically polls
the driver for LAN statistics. This patch implements the osal hook to
fill in the stats.
Fixes: ec94dbc573 ("qede: add base driver")
Signed-off-by: Harish Patil <harish.patil@qlogic.com>
Program the PCIe completion timeout to 4 sec to give enough time
to allow completions to be received successfully in some older systems.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
To be consistent with the naming for ARM NEON implementation,
ixgbe_rxtx_vec.c is renamed to ixgbe_rxtx_vec_sse.c.
Signed-off-by: Jianbo Liu <jianbo.liu@linaro.org>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Previously, if an adminq message is sent successfully, but no response is
received, function "i40evf_execute_vf_cmd" will return without error.
The root cause is value "err" is overwritten. This patch fixes this by
ensuring the value of err is set appropriately for each cmd.
Fixes: ae19955e7c ("i40evf: support reporting PF reset")
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Use ARM NEON intrinsic to implement ixgbe vPMD
Signed-off-by: Jianbo Liu <jianbo.liu@linaro.org>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
[style fixes as highlighted by checkpatch.pl]
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
move scalar code which does not use x86 intrinsic functions to new file
"ixgbe_rxtx_vec_common.h", while keeping x86 code in ixgbe_rxtx_vec.c.
This allows the scalar code to to be shared among vector drivers for
different platforms.
Suggested-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Jianbo Liu <jianbo.liu@linaro.org>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
The VLAN tag information should be stored in the first mbuf of a chain
of buffers, not in the last one.
Fixes: 9fd5e98b62 ("vmxnet3: support RSS and refactor Rx offload")
Signed-off-by: John Guzik <john@shieldxnetworks.com>
Acked-by: Yong Wang <yongwang@vmware.com>
When linking drivers as shared libraries, the dependencies need
to be marked as DT_NEEDED entries.
The crypto dependencies (libsso and libIPSec) are static libraries.
To make them linked in the shared PMDs, the code must relocatable:
- libIPSec_MB.a must be built with -fPIC
- libsso_kasumi.a must be built with KASUMI_CFLAGS=-DKASUMI_C
Fixes: 924e84f873 ("aesni_mb: add driver for multi buffer based crypto")
Fixes: eec136f3c5 ("aesni_gcm: add driver for AES-GCM crypto operations")
Fixes: 3aafc423cf ("snow3g: add driver for SNOW 3G library")
Fixes: 2773c86d06 ("crypto/kasumi: add driver for KASUMI library")
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Some libraries were missing their dependency on eal, mbuf, mempool,
ring and kvargs.
It is revealed by the linker option "-z defs".
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
There is no need to have this parsing inlined in the header.
It brings kvargs dependency to every crypto drivers.
The functions are moved into rte_cryptodev.c.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
The compilation for 32-bit fails when CONFIG_RTE_VIRTIO_USER is enabled:
drivers/net/virtio/virtio_user_ethdev.c:84:47:
error: format ‘%llu’ expects argument of type ‘long long unsigned int’,
but argument 5 has type ‘size_t {aka unsigned int}’
Fixes: e9efa4d938 ("net/virtio-user: add new virtual PCI driver")
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
NSH packet can be recognized by Intel X710/XL710 series.
This patch enables the new packet type.
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Tested-by: Yulong Pei <yulong.pei@intel.com>
Acked-by: Zhe Tao <zhe.tao@intel.com>
In the following loop:
while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
...
}
There is no external function call or any explict memory barrier
in the loop, the re-read of used->idx might be optimized and only
be retrieved once.
Use of voaltile normally should be prohibited, and access_once
is Linux kernel's style to handle this issue; Once we have that
macro in DPDK, we could change to that style.
virtio_recv_mergable_pkts might also have the same issue, so fix
it as well.
Fixes: 823ad64795 ("virtio: support multiple queues")
Fixes: 13ce5e7eb9 ("virtio: mergeable buffers")
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Trying to access xstats_names after "if (xstats_names == NULL)" is
obviously wrong, which would result to a crash while running "show
port xstats 0" in testpmd with virtio PMD.
The fix is straightforward; just reverse the check.
Fixes: baf91c395b ("net/virtio: fetch extended statistics with integer ids")
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
In virtio-user driver, when notify ctrl-queue, invoke API of
virtio-user device emulation to handle ctrl-q command.
Besides, multi-queue requires ctrl-queue and ctrl-queue will be
enabled automatically when multi-queue is specified.
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
The main purpose of this patch is to enable multi-queue. But
multi-queue requires ctrl-queue so that driver can send how many
queues will be enabled through ctrl-queue messages.
So we partially implement ctrl-queue to handle control command
with class of VIRTIO_NET_CTRL_MQ and with cmd of
VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET to handle mq support. This patch
provides a function, virtio_user_handle_cq(), for driver to handle
ctrl-queue messages.
Besides, multi-queue requires VIRTIO_NET_F_MQ and VIRTIO_NET_F_CTRL_VQ
are enabled when we do feature negotiation.
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
This patch mainly adds method in vhost user adapter to communicate
enable/disable queues messages with vhost user backend, aka,
VHOST_USER_SET_VRING_ENABLE.
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Add a new virtual device named virtio-user, which can be used just like
eth_ring, eth_null, etc. To reuse the code of original virtio, we do
some adjustment in virtio_ethdev.c, such as remove key _static_ of
eth_virtio_dev_init() so that it can be reused in virtual device; and
we add some check to make sure it will not crash.
Configured parameters include:
- queues (optional, 1 by default), number of queue pairs, multi-queue
not supported for now.
- cq (optional, 0 by default), not supported for now.
- mac (optional), random value will be given if not specified.
- queue_size (optional, 256 by default), size of virtqueues.
- path (madatory), path of vhost user.
When enable CONFIG_RTE_VIRTIO_USER (enabled by default), the compiled
library can be used in both VM and container environment.
Examples:
path_vhost=<path_to_vhost_user> # use vhost-user as a backend
sudo ./examples/l2fwd/build/l2fwd -c 0x100000 -n 4 \
--socket-mem 0,1024 --no-pci --file-prefix=l2fwd \
--vdev=virtio-user0,mac=00:01:02:03:04:05,path=$path_vhost -- -p 0x1
Known issues:
- Control queue and multi-queue are not supported yet.
- Cannot work with --huge-unlink.
- Cannot work with no-huge.
- Cannot work when there are more than VHOST_MEMORY_MAX_NREGIONS(8)
hugepages.
- Root privilege is a must (mainly becase of sorting hugepages according
to physical address).
- Applications should not use file name like HUGEFILE_FMT ("%smap_%d").
- Cannot work with vhost-net backend.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
This patch is related to how to calculate relative address for vhost
backend.
The principle is that: based on one or multiple shared memory regions,
vhost maintains a reference system with the frontend start address,
backend start address, and length for each segment, so that each
frontend address (GPA, Guest Physical Address) can be translated into
vhost-recognizable backend address. To make the address translation
efficient, we need to maintain as few regions as possible. In the case
of VM, GPA is always locally continuous. But for some other case, like
virtio-user, GPA continuous is not guaranteed, therefore, we use virtual
address here.
It basically means:
a. when set_base_addr, VA address is used;
b. when preparing RX's descriptors, VA address is used;
c. when transmitting packets, VA is filled in TX's descriptors;
d. in TX and CQ's header, VA is used.
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
This patch moves phys addr check from virtio_dev_queue_setup
to pci ops. To make that happen, make sure virtio_ops.setup_queue
return the result if we pass through the check.
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
We skip kernel managed virtio devices, if it isn't whitelisted.
Before checking if the virtio device is whitelisted, check if devargs
is specified.
Fixes: ac5e1d838d ("virtio: skip error when probing kernel managed device")
Reported-by: Vincent Li <vincent.mc.li@gmail.com>
Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Add a new paramter (flags) to rte_vhost_driver_register(). DPDK
vhost-user acts as client mode when RTE_VHOST_USER_CLIENT flag
is set. The flags would also allow future extensions without
breaking the API (again).
The rest is straingfoward then: allocate a unix socket, and
bind/listen for server, connect for client.
This extension is for vhost-user only, therefore we simply quit
and report error when any flags are given for vhost-cuse.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
