This patch fixes the situation when there is already connected pair of
memif interfaces and another slave tries to initiate the connection with
(already occupied) master. Expected behavior is that the second slave
is refused and gets disconnect message with reason: "Already connected",
while old connection remains functional.
Fixes: 09c7e63a71 ("net/memif: introduce memory interface PMD")
Cc: stable@dpdk.org
Signed-off-by: Július Milan <jmilan.dev@gmail.com>
Reviewed-by: Jakub Grajciar <jgrajcia@cisco.com>
With this patch it is possible to connect 2 DPDK memifs into loopback,
i.e. when they have the same id and different roles, as for example:
"--vdev=net_memif0,role=master,id=0"
"--vdev=net_memif1,role=slave,id=0"
Signed-off-by: Július Milan <jmilan.dev@gmail.com>
Reviewed-by: Jakub Grajciar <jgrajcia@cisco.com>
When setting the CONFIG_RTE_BUILD_SHARED_LIB to y,
and build the mlx5 pmd, there is a building error.
To fix it, add RTE_IBVERBS_LINK_DLOPEN to include
relative codes.
mlx5_common.o: In function `mlx5_glue_init':
drivers/common/mlx5/mlx5_common.c:324: undefined reference to `dlclose'
Fixes: 7b4f1e6bd3 ("common/mlx5: introduce common library")
Cc: stable@dpdk.org
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Ali Alnubani <alialnu@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
This patch adds to MLX5 PMD the support of matching on IPv4 TTL and
IPv6 hoplimit.
Support is valid when using DV flow engine (Verbs flow engine doesn't
support matching on this field).
Signed-off-by: Eli Britstein <elibr@mellanox.com>
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
The nic_mask struct in function mlx5_flow_validate_item_ipv6()
includes hop_limits field.
This is redundant since matching on this item is not supported using
Verbs flow engine (dv_flow_en=0).
(Separate patch will be sent to handle this issue for DV flow engine.)
Setting a rule including match on this field will fail:
testpmd> flow create 0 ingress pattern eth / ipv6 hop is 1 / end actions
drop / end
port_flow_complain(): Caught PMD error type 1 (cause unspecified):
hardware refuses to create flow: Operation not supported
This patch removes the redundant field from nic_mask, to ensure that
such flow rules will be rejected by PMD validation:
testpmd> flow create 0 ingress pattern eth / ipv6 hop is 1 / end actions
drop / end
port_flow_complain(): Caught PMD error type 13 (specific pattern item):
cause: 0x7fffffffa2c8, mask enables non supported bits: Operation not
supported
The related lines in function flow_verbs_translate_item_ipv6() are
removed as well, since they are redundant.
Fixes: 23c1d42c71 ("net/mlx5: split flow validation to dedicated function")
Fixes: 84c406e745 ("net/mlx5: add flow translate function")
Cc: stable@dpdk.org
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Remove CONFIG_RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC with below
consideration:
1. A default Rx path can always be selected by setting a proper
rx_free_thresh value at runtime, see
ice_check_rx_burst_bulk_alloc_preconditions.
2. Its not a big deal to always reserve more space for desc ring.
"ring_size = (uint16_t)(rxq->nb_rx_desc + ICE_RX_MAX_BURST);"
3. Fixes a potential invalid memory access in ice_reset_rx_queue.
If CONFIG_RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC is turned on while
ice_check_rx_burst_bulk_alloc_preconditions return fail.
Below code will have problem.
for (i = 0; i < ICE_RX_MAX_BURST; ++i)
rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
Fixes: 50370662b7 ("net/ice: support device and queue ops")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
This is needed to reuse pattern processing engine for MAE.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
As a preparation step, generalise flow start and stop path using
callbacks.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Generalise flow attribute parsing function in regard to transfer attribute.
Add a method table and factor out VNIC-specific parsing code as a callback.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Riverhead boards maintain support for MAE, a low-level Match-Action Engine.
The upcoming patches will bring support for MAE to RTE flow implementation.
A follow-up patch will introduce MAE-specific specification cleanup method.
In order to prepare for the patch, introduce flow allocation and free path.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Add the concept of a flow specification type.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
A backend which a driver employs to handle flow rules of a given
type depends on the underlying NIC flow engine. The driver in
question in its current state is tailored to support the only
flow engine, VNIC filtering. As the need arises to add support
for transfer rules, the driver has to be reworked so that it
becomes possible to introduce yet another backend.
As a preparation step, make the flow list shared
between different engines.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
The flow specification structure will be transformed to a generic one, and
its current contents will be fenced off to form a VNIC-specific parameters.
Flow RSS details do not belong to the said specification currently, albeit
being VNIC-specific. This patch addresses this issue as a preparation step.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Implement eth_dev_ops uc_hash_table_set and uc_all_hash_table_set
Signed-off-by: Chandu Babu N <chandu@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
Supports adding MAC addresses to enable whitelist filtering to
accept packets
implement eth_dev_ops mac_addr_set, mac_addr_add, mac_addr_remove and
set_mc_addr_list
Signed-off-by: Chandu Babu N <chandu@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
Stopping LLDP is necessary for DPDK, but it will cause DCB init failure.
For kernel shared code, the prerequisite for successful initialization
of DCB is that LLDP is enabled. This patch starts LLDP before DCB init
and stops it when init finishes.
Fixes: a8e84b22bd ("net/i40e/base: support persistent LLDP")
Cc: stable@dpdk.org
Signed-off-by: Shougang Wang <shougangx.wang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
NIC deinitialization should not happen after a failed NIC
initialization.
Fixes: 91831d4068 ("net/sfc: estimate available resources")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
When calling to setup RSS on v4 API, ESX will expect
IPv4/6 TCP RSS to be set/requested mandatory.
This patch will:
- Set IPv4/6 TCP RSS when these have not been set. A warning
message is thrown to make sure we warn the application we are
setting IPv4/6 TCP RSS when not set.
- An additional check has been added to dodge RSS configuration
altogether unless MQ_RSS has been requested, similar to v3.
The alternative (returning error) was considered, the intent
is to ease the task of setting up and running vmxnet3 in situations
where it's supposed to be most straightforward (testpmd, pktgen).
Bugzilla ID: 400
Fixes: 643fba7707 ("net/vmxnet3: add v4 boot and guest UDP RSS config")
Cc: stable@dpdk.org
Signed-off-by: Eduard Serra <eserra@vmware.com>
Acked-by: Yong Wang <yongwang@vmware.com>
Requested promiscuous or all-multicast mode may be dropped on port
start if FW denies to enable it (e.g. because of no permission
in the case of VF). Return applied value on get.
Fixes: f5258439ee ("net/sfc: avoid failure on port start if Rx mode is rejected")
Cc: stable@dpdk.org
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Remove unnecessary variable "meta" in ice_flow_create and
ice_flow_validate, it should be defined when really needed:
its ice_parse_engine_create and ice_parse_engine_validate.
A meta data created by parser->parse_pattern_action should be
freed inside parser->engine->create.
During a validate operation, since parser->engine_>create will not
be invoked, it is unnecessary to create a meta during parser, so
NULL will passed to engine->parse_pattern_action and all parser's
parse_pattern_action need to be modified to handle meta = NULL
properly.
With above implementation, the patch also fixes a potential memory
leak in ice_parse_engine_validate, since meta may not be freed.
BTW, an engine without a create op should be regarded as a bug. So
use RTE_ASSERT to replace runtime engine->create == NULL check in
ice_parse_engine_create.
Fixes: 4e27d3ed02 ("net/ice: fix flow API framework")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
This was left to its default value. With the patch transactions are:
- coherent,
- do not allocate in downstream cache (there is none on LS1028a),
- merge surrounding data for BD writes,
- overwrite surrounding data for frame data writes.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
LS1028A does not have platform cache so any reads following a hardware
write will go directly to DDR. Latency of such a read is in excess of
100 core cycles, so try to prefetch more in advance to mitigate this.
How much is worth prefetching really depends on traffic conditions.
With congested Rx this could go up to 4 cache lines or so. But if
software keeps up with hardware and follows behind Rx PI by a cache line
then it's harmful in terms of performance to cache more. We would only
prefetch data that's yet to be written by ENETC, which will be evicted
again anyway.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
Use rte_pktmbuf_free_bulk to release all mbufs at once. This is flagged
as obsolete/not yet stable in DPDK but seems to be functional.
Don't count the released frames, it's no longer needed in the caller.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
Since we know in advance that we're going to fill in multiple
descriptors it's convenient to allocate the buffers in batches.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
Reorder the members of the structure so that the ones used on datapath
fit in a single cache line, to slightly reduce pressure on cache and
miss rate.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
Move from doing batch refill of Rx ring from bundles of 8 to once per
enetc_clean_rx_ring call. One benefit is that we're cleaning up all the
BDs that we just processed, which should still be cached. The other is
that hardware Rx index stays a little back and doesn't cause contention
on the BDs processed in the Rx loop.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
The default value in hardware for the Rx MAC FIFO (@) is higher than it
should be and can lead to Rx lock-up under traffic.
Set it to the value recommended by hardware team, 1.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Gagandeep Singh <g.singh@nxp.com>
Each call to enetc_clean_tx_ring will cost at least 150-200 CPU cycles
even if no clean-up is done, due to the CI register read.
We're only calling it once at the end of the function, on the assumption
that software is slower than hardware and hardware completed sending
older frames out by now.
We're also cleaning up the ring before kicking off Tx for the new batch
to minimize chances of contention on the Tx ring.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
We don't need barriers here since this read doesn't have to be strictly
serialized in relation to other surrounding memory/register accesses.
We only want a reasonably recent value out of hardware so we know how
much we can clean.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
Don't read the hardware CI register in a loop, read it once, clean up
and exit.
The issue with reading the register in a loop is that we're stalling
here trying to catch up with hardware which keeps sending traffic as
long as it has traffic to send, so in effect we could be waiting here
for the Tx ring to be drained by hardware, instead of us doing Rx in
that meantime.
At the time we return the function there may be new BDs in the ring that
could be cleaned, we're just leaving those there for the next time.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
Driver destroys the VNIC when the port is brought down.
Port HW filter setting such as default MAC address and
unicast MAC filters will be applied when port is started.
Fixed to return success silently for these callbacks
when port is stopped.
Fixes: 39b88344e3 ("net/bnxt: fix enable/disable VLAN filtering")
Cc: stable@dpdk.org
Reported-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
There is a race condition when port is closed while error recovery
is happening in driver. Fixed this by canceling the recovery threads
during port close.
Fixes: df6cd7c1f7 ("net/bnxt: handle reset notify async event from FW")
Cc: stable@dpdk.org
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Santoshkumar Karanappa Rastapur <santosh.rastapur@broadcom.com>
Commit 'dc36bd5dfdeb' mistakenly initializes ret as -rte_errno, then
return ret if ice_search_pattern_match_item fails, if rte_errno happens
to be 0 in this case, it would cause an unrecognized flow rule to be
marked as recognition success. Later, when the code tries to parse the
flow rule, a segmentation fault will occur due to null pointer
reference.
Fixes: dc36bd5dfd ("net/ice: fix flow FDIR/switch memory leak")
Cc: stable@dpdk.org
Signed-off-by: Tao Zhu <taox.zhu@intel.com>
Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
Lack of GL_SWR_PM_UP_THR value for mac type 4x10G X722 caused it can't
reach line rate in one queue one core forwarding case. It's about 1%
performance drop. This patch adds value as:
Device-ID Value Comments
0x37D0 0x03030303 4x10G FPK
This value is hardware specific.
Fixes: 3320d4a240 ("net/i40e: workaround performance degradation")
Cc: stable@dpdk.org
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
Since qpairs is part of the vdev arguments, there is no need to
limit it to 16. The queue arrays can be dynamically sized based
on the requested parameters.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: John W. Linville <linville@tuxdriver.com>
The VLAN 16 bits tci field contains both values of PCP and VID. When
extracting any one of them - it is required not to affect the other one.
Previous to this commit in routine flow_dev_get_vlan_info_from_items()
we calculated the PCP as follows:
(1) vlan->vlan_tci &= MLX5DV_FLOW_VLAN_PCP_MASK;
(2) vlan->vlan_tci |= <3 bits value of PCP>
In line (1) we should have used the negated mask ('~' operator) such
that only the PCP bits will be nullified before ORing them with the
updated PCP value.
Fixes: 9aee7a8418 ("net/mlx5: support push flow action on VLAN header")
Cc: stable@dpdk.org
Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
If configuring VF promiscuous mode is not supported, return '-ENOTSUP'
error code in '.promiscuous_enable/disable' dev_ops.
This is to fix the port start during config restore, where if
'.promiscuous_enable/disable' dev_ops exists and return any value other
than '-ENOTSUP', start will fail.
Same is done for '.allmulticast_enable/disable' dev_ops.
Fixes: ddc7cb0d94 ("net/i40e: re-program promiscuous mode on VF interface")
Cc: stable@dpdk.org
Signed-off-by: Xiao Zhang <xiao.zhang@intel.com>
Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
Add support for .set_link_up/down() eth ops to bring
link up and down.
Signed-off-by: Harman Kalra <hkalra@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Loopback devices are exposed as ethdev device in octeontx2.
This patch fixes the link information for loopback ports.
Fixes: 38f566280a ("net/octeontx2: add link stats operations")
Cc: stable@dpdk.org
Signed-off-by: Ashish Gupta <ashishg@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Since packet is talking to a network device (probably tap)
it is possible for the device driver to change the hardware address
by making an ioctl to kernel.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
The devices of the family ConnectX may have two letters as suffix.
Such suffix is preceded with a space and the second x is lowercase:
- ConnectX-4 Lx
- ConnectX-5 Ex
- ConnectX-6 Dx
Uppercase of the device family name BlueField is also fixed.
The lists of supported devices are fixed.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
The routine sending packets with Multi-Packet Write method assigns
the wqe_last variable with transmit descriptor (WQE - work queue entry)
being built. If send queue is close to full state, the WQE has no data
yet (trying to put the first packet) and there is no enough space
in descriptor for the next packet the WQE is discarded and the stored
wqe_last value becomes invalid - points to the discarded WQE.
The mlx5_tx_burst_request_completion() routine might set the completion
request flags in the WQE pointed by wqe_last, it is safe, but the next
mlx5_tx_burst call uses the WQE as the first free one and request
completion flags might be overwritten and completion request will be
lost causing the transmit datapath malfunction.
Fixes: 8b581c690a ("net/mlx5: move Tx complete request routine")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
In order to move the mbuf metadata from the WQE to the FDB steering
domain, the PMD add for each NIC TX flow a new action to copy the
metadata register REG_A to REG_C_0.
This copy action is considered as modify header action from HW
perspective.
The HW doesn't support to do modify header action after ant
encapsulation action.
The split metadata function wrongly added the copy action in the end of
the original actions list, hence, NIC egress flow with encapsulation
action failed when the PMD worked with dv_xmeta_en mode.
Move the copy action to be before and back to back with the
encapsulation action for the aforementioned case.
Fixes: 71e254bc02 ("net/mlx5: split Rx flows to provide metadata copy")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
When running mlx5_dev_start in mlx5_ethdev the function calls
mlx5_dev_configure_rss_reta in order to configure the rxq's.
Before mlx5_dev_configure_rss_reta there isn't a check whether
there are rxq's and if rxq's are 0 the function fails.
For example, this command:
/build/app/test-pmd/testpmd -n 4 -w 0000:08:00.0,rx_vec_en=0
-- --burst=64 --mbcache=512 -i --nb-cores=27 --txd=2048 --rxd=2048
--vxlan-gpe-port=6081 --mp-alloc=xbuf --rxq 0 --forward-mode=txonly
would fail.
In order to fix this issue, we should call mlx5_dev_configure_rss_reta
only if we have rxq's.
Fixes: 63bd16292c ("net/mlx5: support RSS on hairpin")
Cc: stable@dpdk.org
Reported-by: Sylvain Rodon <srn@nexatech.fr>
Signed-off-by: Shiri Kuzin <shirik@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
