Replacing memcpy() with rte_memcpy() improved 64 byte packet
performance by 33% on a POWER9 system and by 10% on an x86_64
system.
Signed-off-by: David Christensen <drc@linux.vnet.ibm.com>
Acked-by: Tyler Retzlaff <roretzla@linux.microsoft.com>
In a multi-TC scenario, if the length of packets destined for different
TCs is different, for example, 64B and 1500B packets destined for TC0 and
TC1 respectively. There is a problem that the bandwidth of the TC to which
large packets are sent is preempted by the TC to which small packets are
sent on the Kunpeng 920 network engine. As a result, the TC bandwidth
accuracy is inaccurate.
To solve this problem, this patch made the following adjustments:
1/ During initialization, firmware reports the capability bit indicating
whether the TM function is supported.
2/ The command word for configuring TC and port rate limiting is added,
instead of reusing the existing command word. And firmware configured
to the correct module.
3/ When the PF driver is loaded, firmware completes the default
initialization of the TC and port.
Fixes: c09c7847d8 ("net/hns3: support traffic management")
Cc: stable@dpdk.org
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
This patch creates a new function for handling PF probing of a secondary
process. A CPP handle is obtained for the CPP bridge service and the
service itself is also registered during secondary process
initialization. DPDK services aren't shared between processes so it is
not enough to only have the primary register the service if it is also
needed in a secondary process. This implies that both the primary and
secondary will have their own copy of the bridge service.
Fixes: 5e15e799d6 ("net/nfp: create separate entity for PF device")
Cc: stable@dpdk.org
Signed-off-by: Heinrich Kuhn <heinrich.kuhn@netronome.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
When using rte_eth_dev_pci_generic_probe() during probing a
rte_eth_dev will be created with the name field corresponding to the PCI
address of the device. NFP4000/6000 devices only have a single PF (but
potentially multiple physical ports). This means that in a simple two
port example the rte_eth_devices[] array will be populated with two
devices: 0000:02:00.0 and 0000:02:00.0_port1. This is inconsistent and
not ideal. It will also cause issues when a secondary process tries to
attach to these ports.
This patch removes the use of rte_eth_dev_pci_generic_probe() and
allocates eth_dev's for each physical port during PF initialization,
giving them more consistent names.
Fixes: 5e15e799d6 ("net/nfp: create separate entity for PF device")
Cc: stable@dpdk.org
Signed-off-by: Heinrich Kuhn <heinrich.kuhn@netronome.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Since commit d5df2ae042 ("net: fix unneeded replacement of TCP
checksum 0"), the functions rte_ipv4_udptcp_cksum() and
rte_ipv6_udptcp_cksum() can return either 0x0000 or 0xffff when used to
verify a packet containing a valid checksum.
Since these functions should be used to calculate the checksum to set in
a packet, introduce 2 new helpers for checksum verification. They return
0 if the checksum is valid in the packet.
Use this new helper in net/tap driver.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Since commit d5df2ae042 ("net: fix unneeded replacement of TCP
checksum 0"), the functions rte_ipv4_udptcp_cksum() or
rte_ipv6_udptcp_cksum() can return either 0x0000 or 0xffff when used to
verify a packet containing a valid checksum.
This new behavior broke the checksum verification in tap driver for TCP
packets: these packets are marked with PKT_RX_L4_CKSUM_BAD.
Fix this by checking the 2 possible values. A next commit will introduce
a checksum verification helper to simplify this a bit.
Fixes: d5df2ae042 ("net: fix unneeded replacement of TCP checksum 0")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
When packet type is IPV4_EXT, the checksum is always marked as good in
the mbuf offload flags.
Since we know the header lengths, we can easily call
rte_ipv4_udptcp_cksum() in this case too.
Fixes: 8ae3023387 ("net/tap: add Rx/Tx checksum offload support")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Chelsio T6 ASIC doesn't track Rx promisc traffic dropped due to lack
of Rx buffers and hence the imissed counter doesn't increment. Add
support for RAW MAC filter to insert a wildcard matchall rule at
the end of MPS TCAM to make MPS track the promisc traffic. This
rule will only be added/removed when promisc mode is turned on/off
on the interface.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Do not return "GOOD" flag if the status was bad.
Fixes: 9e6e7f479a ("net/sfc: support Rx checksum offload for EF100")
Cc: stable@dpdk.org
Signed-off-by: Thierry Herbelot <thierry.herbelot@6wind.com>
Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
ethdev library does the check of dev before calling the link update.
Fixes: acd4818ea2 ("net/pfe: add link status update")
Cc: stable@dpdk.org
Signed-off-by: Thierry Herbelot <thierry.herbelot@6wind.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
Kunpeng 930 support Tx push mode which could improve performance.
It works like below:
1. Add PCIe bar45 which support driver direct write the Tx descriptor
or tail reg to it.
2. Support three operations: a) direct write one Tx descriptor, b)
direct write two Tx descriptors, c) direct write tail reg.
3. The original tail reg located at bar23, the above bar45 tail reg
could provide better bandwidth from the hardware perspective.
The hns3 driver only support direct write tail reg (also have the name
of quick doorbell), the detail:
Considering compatibility, firmware will report Tx push capa if the
hardware support it.
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
Delete redundant blank lines to make:
1.Return value judgment follow the function call.
2.No blank lines at the end of a code block defined by braces.
Signed-off-by: Hongbo Zheng <zhenghongbo3@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
The "cfg_max_queues" maintains configured max queue numbers from user,
and is equal to the maximum of "nb_rx_queues" and "nb_tx_queues" in
"dev->data".
From the ethdev layer framework, "nb_rx/tx_queues" in "dev->data" were set
to zero in rte_eth_dev_configure() if ops.dev_configure in PMD fails to be
executed, In addition, if ops.dev_configure in HNS3 PMD failed, the fake
queues are also cleared on a device that does not support independent Rx/Tx
queues.
Therefore, the "cfg_max_queues" should be also set to zero when
dev_configure fails.
Fixes: fa29fe45a7 ("net/hns3: support queue start and stop")
Cc: stable@dpdk.org
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
When the current VLAN stripping is set, the log print always prompts
that the enabling fails, bug if may actually be the disabling failure.
Fixes: 411d23b9ea ("net/hns3: support VLAN")
Cc: stable@dpdk.org
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
When the device supports independent Rx/Tx queues, fake queues do not need
to be created in unequal Rx/Tx queues case. However, dev_configure fails
to be executed on the device supported independent Rx/Tx queues, the
current rollback code logic contains the fake queue. As a result, the fake
queue is created. When dev_configure is successfully called again, these
fake queues still exists and are configured to the hardware.
Fixes: fa29fe45a7 ("net/hns3: support queue start and stop")
Cc: stable@dpdk.org
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
When the primary process executes dev_stop or is being reset, the packet
sending and receiving functions is changed. In this moment, the primary
process requests secondary processes to change their Rx/Tx functions, and
delays a period of time in case of crashes when queues are still in use.
The delay time depends on the number of queues actually used, instead of
the maximum number of queues supported by the device.
Fixes: 23d4b61fee ("net/hns3: support multiple process")
Cc: stable@dpdk.org
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
When the device is very busy, VF reset may have to be
retried many times to succeed, leading to the current
max reset fail retry count not enough.
Modify max reset fail retry count to 30 to enhance
the reliability of reset function.
Fixes: 2790c64647 ("net/hns3: support device reset")
Cc: stable@dpdk.org
Signed-off-by: Hongbo Zheng <zhenghongbo3@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
Introduce an internal firmware loading helper to remove code duplication
in our drivers and handle xz compressed firmware by calling libarchive.
This helper tries to look for .xz suffixes so that drivers are not aware
the firmware has been compressed.
libarchive is set as an optional dependency: without libarchive, a
runtime warning is emitted so that users know there is a compressed
firmware.
Windows implementation is left as an empty stub.
Signed-off-by: David Marchand <david.marchand@redhat.com>
Reviewed-by: Igor Russkikh <irusskikh@marvell.com>
Acked-by: Aaron Conole <aconole@redhat.com>
Tested-by: Haiyue Wang <haiyue.wang@intel.com>
Both "normal" and "dcf" inits have their copy of some firmware loading
code.
The DSN query is moved in specific parts for the "normal" and "dcf" init.
A common helper ice_load_pkg is then introduced and takes an adapter
pointer as its main input.
This helper takes care of finding the right firmware file and loading
it.
The adapter active_pkg_type field is set by this helper.
The ice_access macro is removed from the osdep.h header: osdep.h should
only hosts wrappers for base driver code.
Signed-off-by: David Marchand <david.marchand@redhat.com>
Acked-by: Haiyue Wang <haiyue.wang@intel.com>
Caught by code review.
Using a random name for memzone allocations can result in init failures
in the unlikely case that a name collision occurs.
Use a simple sequential generator on 64 bits.
Fixes: 3f50f072ff ("i40e: fix memzone freeing")
Fixes: 22b123a36d ("net/avf: initialize PMD")
Fixes: 5f0978e962 ("net/ice/base: add OS specific implementation")
Fixes: 737f30e1c3 ("net/hns3: support command interface with firmware")
Cc: stable@dpdk.org
Signed-off-by: David Marchand <david.marchand@redhat.com>
Acked-by: Min Hu (Connor) <humin29@huawei.com>
Acked-by: Haiyue Wang <haiyue.wang@intel.com>
This patch fixed the unmatched integer type issue in the comparison,
which causing compilation failure on RHEL.
drivers/net/ice/ice_dcf_sched.c:353:15: error:
comparison between signed and unsigned integer expressions
if (node_id > tc_nb * hw->num_vfs) {
^
CI reported the failure:
http://mails.dpdk.org/archives/test-report/2021-July/201905.html
Fixes: 3a6bfc37ea ("net/ice: support QoS config VF bandwidth in DCF")
Signed-off-by: Ting Xu <ting.xu@intel.com>
Tested-by: Ali Alnubani <alialnu@nvidia.com>
The current code doesn't use match on invalid outer rule ID
in action rules of non-encap. flows. This is wrong and must
be fixed. Do that and explain correct behaviour in the code.
Fixes: dadff13793 ("net/sfc: support encap flow items in transfer rules")
Cc: stable@dpdk.org
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Reviewed-by: Andy Moreton <amoreton@xilinx.com>
The container in the action configuration is U32,
but the ID is U16, and overflow check is missing.
Fixes: 1fb65e4dae ("net/sfc: support flow action port ID in transfer rules")
Cc: stable@dpdk.org
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Take into account VLAN presence fields in items ETH and VLAN.
Provided that the item ETH does not match on the EtherType,
the pattern behaviour will be as follows:
- ETH (mask->has_vlan = 0) | IPv4 = match both tagged and untagged;
- ETH (mask->has_vlan = 1) | IPv4 = match as per spec->has_vlan;
- ETH (mask->has_vlan = 0) | VLAN | IPv4 = match only tagged.
Similar logic applies to double tagging.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Introduce necessary infrastructure for these fields to
be set, validated and compared during class comparison.
Enumeration and mappings envisaged are MCDI-compatible.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
In poll mode driver of octeontx2 the RQ is connected to a CQ and it is
responsible for asserting backpressure to the CGX channel.
When event eth Rx adapter is configured, the RQ is connected to a event
queue, to enable backpressure we need to configure AURA assigned to a
given RQ to backpressure CGX channel.
Event device expects unique AURA to be configured per ethernet device.
If multiple RQ from different ethernet devices use the same AURA,
the backpressure will be disabled, application can override this
using devargs:
-a 0002:0e:00.0,force_rx_bp=1
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Shift is used to generate an 8-bit saturate value from the current
aura used count. The shift value should be derived from the log2 of
block count if it is greater than 256 else the shift should be 0.
Fixes: 7bcc47cbe2 ("mempool/octeontx2: add mempool alloc op")
Cc: stable@dpdk.org
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Set the appropriate capability flags for the RX, crypto and timer
eventdev adapters to use.
Signed-off-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
Tested-by: Heng Wang <heng.wang@ericsson.com>
clang-10 build issue log:
drivers/event/cnxk/cnxk_tim_worker.h:372:23:
warning: value size does not match register size
specified by the constraint and modifier [-Wasm-operand-widths]
: [rem] "=&r"(rem)
^
cnxk/cnxk_tim_worker.h:365:17: note: use constraint modifier "w"
"ldxr %[rem], [%[crem]] \n"
^~~~~~
%w[rem]
Changed variable type to match register size, which placates clang.
Fixes: 300b796262 ("event/cnxk: add timer arm routine")
Cc: stable@dpdk.org
Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Packet was corrupted when TSO requested without CSUM update.
Enables CSUM automatically if only TSO requested.
Fixes: 2aa8444b00 ("vdpa/mlx5: support stateless offloads")
Cc: stable@dpdk.org
Signed-off-by: Xueming Li <xuemingl@nvidia.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
This patch adds default RSS support for IPv4 and IPv6 fragment packet.
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
The rte_eth_devices array is not in share memory, it should not be
referenced by i40e_adapter which is shared by primary and secondary.
Any process set i40e_adapter->eth_dev will corrupt another process's
context.
The patch removed the field "eth_dev" from i40e_adapter.
Now, when the data paths try to access the rte_eth_dev_data instance,
they should replace adapter->eth_dev->data with adapter->pf.dev_data.
Fixes: 4861cde461 ("i40e: new poll mode driver")
Cc: stable@dpdk.org
Signed-off-by: Dapeng Yu <dapengx.yu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
For i40e vector Tx path, if tx_offload is set as FAST_FREE_MBUF mode,
no mbuf fast free operations are executed. To fix this, add mbuf fast
free mode for vector Tx path.
Furthermore, for i40e vector Tx path, if implement FAST_FREE_MBUF mode,
it means per-queue all mbufs come from the same mempool and have
refcnt = 1. Thus we can use bulk free of the buffers when mbuf fast free
mode is enabled.
For vector path in arm platform:
In n1sdp, performance is improved by 18.4%;
In thunderx2, performance is improved by 23%.
For vector path in x86 platform:
No performance changes.
Suggested-by: Ruifeng Wang <ruifeng.wang@arm.com>
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
For i40e scalar Tx path, if implement FAST_FREE_MBUF mode, it means
per-queue all mbufs come from the same mempool and have refcnt = 1.
Thus we can use bulk free of the buffers when mbuf fast free mode is
enabled.
Following are the test results with this patch:
MRR L3FWD Test:
two ports & bi-directional flows & one core
RX API: i40e_recv_pkts_bulk_alloc
TX API: i40e_xmit_pkts_simple
ring_descs_size = 1024;
Ring_I40E_TX_MAX_FREE_SZ = 64;
tx_rs_thresh = I40E_DEFAULT_TX_RSBIT_THRESH = 32;
tx_free_thresh = I40E_DEFAULT_TX_FREE_THRESH = 32;
For scalar path in arm platform with default 'tx_rs_thresh':
In n1sdp, performance is improved by 7.9%;
In thunderx2, performance is improved by 7.6%.
For scalar path in x86 platform with default 'tx_rs_thresh':
performance is improved by 4.7%.
Suggested-by: Ruifeng Wang <ruifeng.wang@arm.com>
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
The HW debug mask is always zero, so user can't enable the related debug
function like ICE_DBG_XXX etc, add the devarg 'hw_debug_mask' to set the
debug mask log output at runtime.
Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Currently, there'll be conflict error when running
the following commands:
1. flow create 0 ingress
pattern eth / ipv4 / udp src is 32 / end
actions queue index 2 / end
2. flow destroy 0 rule 0
3. flow create 0 ingress
pattern eth / ipv4 / udp dst is 32 / end
actions queue index 2 / end
This patch fixes the input set conflict issue.
Fixes: 42044b69c6 ("net/i40e: support input set selection for FDIR")
Fixes: 4a072ad434 ("net/i40e: fix flow director config after flow validate")
Cc: stable@dpdk.org
Signed-off-by: Beilei Xing <beilei.xing@intel.com>
Tested-by: Lingli Chen <linglix.chen@intel.com>
Add check in the Tx packet preparation function, to guarantee that the
packet with specific user priority is distributed to the correct Tx
queue according to the configured Tx queue TC mapping.
Signed-off-by: Ting Xu <ting.xu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
This patch added the support for VF to config the ETS-based Tx QoS,
including querying current QoS configuration from PF and config queue TC
mapping. PF QoS is configured in advance and the queried info is
provided to the user for future usage. VF queues are mapped to different
TCs in PF through virtchnl.
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Signed-off-by: Ting Xu <ting.xu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
This patch supports the ETS-based QoS configuration. It enables the DCF
to configure bandwidth limits for each VF VSI of different TCs. A
hierarchy scheduler tree is built with port, TC and VSI nodes.
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Signed-off-by: Ting Xu <ting.xu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
When link status changes, DCF will receive virtchnl PF event message.
Add support to handle this event, change link status and update link
info.
Signed-off-by: Ting Xu <ting.xu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
In the adminq command query port ETS function, the root node teid is
needed. However, for DCF, the root node is not initialized, which will
cause error when we refer to the variable. In this patch, we will check
whether the root node is available or not first.
Signed-off-by: Ting Xu <ting.xu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
This patch adds new virtchnl opcodes and structures for QoS
configuration, which includes:
1. VIRTCHNL_VF_OFFLOAD_TC, to negotiate the capability supporting QoS
configuration. If VF and PF both have this flag, then the ETS-based QoS
offload function is supported.
2. VIRTCHNL_OP_DCF_CONFIG_BW, DCF is supposed to configure min and max
bandwidth for each VF per enabled TCs. To make the VSI node bandwidth
configuration work, DCF also needs to configure TC node bandwidth
directly.
3. VIRTCHNL_OP_GET_QOS_CAPS, VF queries current QoS configuration, such
as enabled TCs, arbiter type, up2tc and bandwidth of VSI node. The
configuration is previously set by DCB and DCF, and now is the potential
QoS capability of VF. VF can take it as reference to configure queue TC
mapping.
4. VIRTCHNL_OP_CONFIG_TC_MAP, set VF queues to TC mapping for all Tx and
Rx queues. Queues mapping to one TC should be continuous and all
allocated queues should be mapped.
Signed-off-by: Ting Xu <ting.xu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Add a specific path for RX AVX2.
In this path, support the HW offload features, like,
checksum, VLAN stripping, RSS hash.
This path is chosen automatically according to the
configuration.
'inline' is used, then the duplicate code is generated
by the compiler.
Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Acked-by: Leyi Rong <leyi.rong@intel.com>
Add a specific path for TX AVX2.
In this path, support the HW offload features, like,
checksum insertion, VLAN insertion.
This path is chosen automatically according to the
configuration.
'inline' is used, then the duplicate code is generated
by the compiler.
Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Acked-by: Leyi Rong <leyi.rong@intel.com>
Since each PF does not share the same structure space, the first
mask value should start at 0 instead of hw->pf_id * per_pf to avoid
address overflow. Otherwise, address space will overlap when
masks.first + masks.count > ICE_PROF_MASK_COUNT, and it may lead to
unexpected variable assignment, which causes segmentation fault.
Fixes: 9467486f17 ("net/ice/base: enable masking for RSS and FD field vectors")
Cc: stable@dpdk.org
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
i40e can support following rss hash function types: default/toeplitz,
symmetric toeplitz, and simple_xor. However, when filter engine parses
pattern action, it only supports symmetric toeplitz & default.
Add simple_xor and toeplitz hash functions support when parsing pattern
action.
Fixes: ef4c16fd91 ("net/i40e: refactor RSS flow")
Cc: stable@dpdk.org
Signed-off-by: Steve Yang <stevex.yang@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
