A VF representor allocates queues from PF's pool of queues and use
them for its Tx and Rx. It supports 1 Tx queue and 1 Rx queue.
Implicit packet forwarding between representor queues and VF does not
yet exist. It will be enabled in subsequent commits using flowman API.
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
Enable the minimal VF representor without Tx/Rx and flow API support.
1. Enable the standard devarg 'representor'
When the devarg is specified, create VF representor ports.
2. Initialize flowman early during PF probe
Representors require the flowman API from the firmware. Initialize it
before creating VF representors, so probe can detect the flowman
support and fail if not available.
3. Add enic_fm_allocate_switch_domain() to allocate switch domain ID
PFs and VFs on the same VIC adapter can forward packets to each other,
so the switch domain is the physical adapter.
4. Create a vnic_dev lock to serialize concurrent devcmd calls
PF and VF representor ports may invoke devcmd (e.g. dump stats)
simultaneously. As they all share a single PF devcmd instance in the
firmware, use a lock to serialize devcmd calls.
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
VF representors need to proxy devcmd through the PF vnic_dev
instance. Extend vnic_dev to accommodate them as follows.
1. Add vnic_vf_rep_register()
A VF representor creates its own vnic_dev instance via this function
and saves VF ID. When performing devcmd, vnic_dev uses the saved VF ID
to proxy devcmd through the PF vnic_dev instance.
2. Add vnic_register_lock()
As PF and VF representors appear as independent ports to the
application, its threads may invoke APIs on them simultaneously,
leading to race conditions on the PF vnic_dev. For example, thread A
can query stats on PF port, while thread B queries stats on a VF
representor.
The PF port invokes this function to provide a lock to vnic_dev. This
lock is used to serialize devcmd calls from PF and VF representors.
3. Add utility functions to assist VF representor settings
vnic_dev_mtu() and vnic_dev_uif() retrieve vnic MTU and UIF number
(uplink index), respectively.
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
Add a field named rx_buf_size in rte_eth_rxq_info to indicate the buffer
size used in receiving packets for HW.
In this way, upper-layer users can get this information by calling
rte_eth_rx_queue_info_get.
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
The address should be calculated before type cast, not after.
Fixes: cc02518132 ("net/netvsc: split send buffers from Tx descriptors")
Cc: stable@dpdk.org
Reported-by: Souvik Dey <sodey@rbbn.com>
Signed-off-by: Long Li <longli@microsoft.com>
Using 'rte_mb' to synchronize the shared ring head/tail between producer
and consumer will stall the pipeline and damage performance on the weak
memory model platforms, such like aarch64.
Relax the expensive barrier with c11 atomic with explicit memory
ordering can improve 3.6% performance on throughput.
Signed-off-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Jakub Grajciar <jgrajcia@cisco.com>
Currently, there is a possibility that segment faults occur when sending
packets whose payloads are stored in multiple buffers based on hns3
network engine. The related core dump information as follows:
Program terminated with signal 11, Segmentation fault.
0 hns3_reassemble_tx_pkts
2512 temp = temp->next;
Missing separate debuginfos, use:
(gdb) bt
0 hns3_reassemble_tx_pkts
1 0x0000000000969c60 in hns3_check_non_tso_pkt
2 0x000000000096adbc in hns3_xmit_pkts
3 0x000000000050d4d0 in rte_eth_tx_burst
4 0x000000000050fca4 in pkt_burst_transmit
5 0x00000000004ca6b8 in run_pkt_fwd_on_lcore
6 0x00000000004ca7fc in start_pkt_forward_on_core
7 0x00000000006975a4 in eal_thread_loop
8 0x0000ffffa6f7fc48 in start_thread
9 0x0000ffffa6ed1600 in thread_start
The root cause is that hns3 PMD driver invokes the rte_pktmbuf_free_seg
API function to release the same rte_mbuf multiple times. The rte_mbuf
pointer is not set to NULL in the internal function
hns3_rx_queue_release_mbufs which is invoked during queue setup, stop
and close. As a result the rte_mbuf in Rx queues will be repeatedly
released when the user application setup queues or stop/start the dev
for multiple times.
Probably for performance reasons, DPDK mempool lib does not check for
the repeated rte_mbuf releases. The Address of released rte_mbuf are
directly stored into the per lcore cache of the mempool. This makes the
rte_mbufs obtained from mempool by calling rte_mempool_get_bulk API
function repetitively. ultimately, it causes to access to a NULL
pointer in PMD driver.
This patch fixes this problem by setting released mbuf pointer to NULL
in the internal function named hns3_rx_queue_release_mbuf. And the other
internal function named hns3_reassemble_tx_pkts is optimized to avoid a
similar problem.
Fixes: bba6366983 ("net/hns3: support Rx/Tx and related operations")
Cc: stable@dpdk.org
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
When Rx of scattered packets is off, we have some possibility of using
vector Rx process function or simple Rx functions in hns3 PMD driver.
If the input MTU is increased and the maximum length of received packets
is greater than the length of a buffer for Rx packets, the hardware
network engine needs to use multiple BDs and buffers to store these
packets. This will cause problems when still using vector Rx process
function or simple Rx function to receiving packets. So, when Rx of
scattered packets is off and device is started, it is not permitted to
increase MTU so that the maximum length of Rx packets is greater than Rx
buffer length.
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
This patch adds simple Tx process function. When multiple segment
packets are not needed, Which means that DEV_TX_OFFLOAD_MBUF_FAST_FREE
offload is not set, we can simple Tx process.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
This patch adds simple Rx process function and support chose Rx function
by real Rx offloads capability.
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Huisong Li <lihuisong@huawei.com>
This patch adds the internal function named hns3_write_reg_opt to avoid
performance loss from address calculation during register access in the
'.rx_pkt_burst' ops implementation function named hns3_recv_pkts.
In addition, because hardware always access register in little-endian
mode based on hns3 network engine, so driver should also call
rte_cpu_to_le_32 to convert data in little-endian mode before writing
register and call rte_le_to_cpu_32 to convert data after reading from
register. Here the driver encapsulates the data conversion operation in
the register read/write operation function as below:
hns3_write_reg
hns3_write_reg_opt
hns3_read_reg
Therefore, when calling these functions, conversion is not required
again.
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
This patch reports .rx_free_thresh value in the .dev_infos_get ops
implementation function named hns3_dev_infos_get and
hns3vf_dev_infos_get.
In addition, the name of the member variable of struct hns3_rx_queue is
modified and comments are added to improve code readability.
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Add usage of rte_eth_link_to_str function to example
applications.
Signed-off-by: Ivan Dyukov <i.dyukov@samsung.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Add usage of rte_eth_link_to_str function to applications and docs.
Signed-off-by: Ivan Dyukov <i.dyukov@samsung.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
There is new link_speed value introduced. It's INT_MAX value which
means that speed is unknown. To simplify processing of the value
in application, new function is added which convert link_speed to
string. Also dpdk examples have many duplicated code which format
entire link status structure to text.
This commit adds two functions:
* rte_eth_link_speed_to_str - format link_speed to string
* rte_eth_link_to_str - convert link status structure to string
Signed-off-by: Ivan Dyukov <i.dyukov@samsung.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
The kernel expects the start address of the UMEM to be page size
aligned.
Since the mempool is not guaranteed to have such alignment, we have been
aligning the address to the start of the page the mempool is on. However
when passing the 'size' of the UMEM during it's creation we did not take
this into account.
This commit adds the amount by which the address was aligned to the size
of the UMEM.
Bugzilla ID: 532
Fixes: d8a210774e ("net/af_xdp: support unaligned umem chunks")
Cc: stable@dpdk.org
Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
Inside Macro ASSIGN_UNLESS_EQUAL(var, val), assignment to var is always
failing as assignment done using var_ having local scope only.
This leads to TX packets not going out and found broken due to cleanup
malfunctioning. This patch fixes the wrong variable assignment.
Fixes: 57f90f8945 ("net/virtio: reuse packed ring functions")
Cc: stable@dpdk.org
Signed-off-by: Vipul Ashri <vipul.ashri@oracle.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The CQ polling is done in order to notify the guest about new traffic
bursts and to release FW resources for the next bursts management.
When HW is faster than SW, it may be that all the FW resources are busy
in SW due to late polling.
In this case, due to wrong WQE counter masking, the fullness
calculation of the completions number is 0 while the queue is full.
Change the WQE counter masking to 16-bit wideness instead of the CQ
size mask as defined by the CQE format.
Fixes: c5f714e50b ("vdpa/mlx5: optimize completion queue poll")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@nvidia.com>
Acked-by: Xueming Li <xuemingl@nvidia.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The CQ configuration enables the collapse feature in HW what cause HW to
write all the completions in the first CQE.
When this feature is enabled the HW doesn't switch the owner bit when it
starts a new cycle of the CQ, not like working without the collapse
feature.
The current SW CQ polling wrongly added an assertion to validate the
owner bit switch what causes a panic in debug mode.
Remove the aforementioned assertion.
Fixes: c5f714e50b ("vdpa/mlx5: optimize completion queue poll")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@nvidia.com>
Acked-by: Xueming Li <xuemingl@nvidia.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Control thread (which handles iotlb msg) and forwarding thread
both use iotlb to translate address. The former may modify the
same entry of mempool and may cause a loop in iotlb_pending_entries
list.
Bugzilla ID: 523
Fixes: d012d1f293 ("vhost: add IOTLB helper functions")
Cc: stable@dpdk.org
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
During vDPA device setup, if some error happens, event channel
release stucks at polling event channel.
Event channel fd is set to non-blocking in cqe setup, so if any
error happens before this function and after event channel created,
the pooling before releasing resources will stuck.
This patch moves event channel to non-blocking mode right after
creation.
Fixes: 8395927cdf ("vdpa/mlx5: prepare HW queues")
Cc: stable@dpdk.org
Signed-off-by: Xueming Li <xuemingl@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
vhost lib now does not have definition of reset status. This patch
adds the reset status definition and changes related log.
Signed-off-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
virtio 1.0 introduced a mechanism for the driver to verify that the
feature bits it sets are accepted by the device. This mechanism consists
in setting the VIRTIO_STATUS_FEATURE_OK status bit and re-reading it,
which gives a chance for the device to clear it if the features
were not accepted.
This is currently being done only in modern virtio-pci devices but since
the appropriate vhost-user messages have been added, it can also be done
in virtio-user (vhost-user only).
This patch activates this mechanism on virtio-user.
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
This patch adds support for VHOST_USER_GET_STATUS request.
Only vhost-user backed is supported for now
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
This patch adds support for VHOST_USER_SET_STATUS
request. It is used to make the backend aware of
Virtio devices status update.
It is useful for the backend to know when the Virtio
driver is done with the Virtio device configuration.
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
For the sake of completeness, add the definition of the missing status
bit in accordance with the virtio spec
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
Chelsio T6 NIC can support up to 8 priority channels to manage
congestion. So, increase to 8 congestion channels for T6. Also,
add Rxq state to avoid unnecessarily ringing doorbell and polling
the hardware for more traffic when the Rxq is stopped.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Firmware returns the max queues that can be allocated on the entire
PF. The driver evenly distributes them across all the ports belonging
to the PF. However, some ports may need more queues than others and
this equal distribution scheme prevents accessing these other ports
unused queues. So, remove the equal distribution scheme and allow the
ports to allocate as many queues as they need.
Also remove the hardcoded 64 max limit on queue allocation. Instead,
use the max limit given by firmware.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Enable RTE_ETH_DEV_CLOSE_REMOVE during PCI probe for all ports
enumerated under the PF. Free up the underlying port Virtual
Identifier (VI) and associated resources during port close.
Once all the ports under the PF are closed, free up the PF-wide
shared resources. Invoke port close function of all ports under
the PF, in PCI remove too.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Free up the DMA memzones properly for all the port's queues during
port close. So, rework DMA ring allocation/free logic to use
rte_eth_dma_zone_reserve()/rte_eth_dma_zone_free() helper functions
for allocating/freeing the memzones.
The firmware event queue doesn't have an associated freelist queue.
So, remove check that tries to give memzone name for a non-existent
freelist queue.
Also, add a missing free for the control queue mempools.
Fixes: 0462d11544 ("cxgbe: add device related operations")
Cc: stable@dpdk.org
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Tunnel configuration may fail because of insufficient access rights
on a virtual function. Ignore the failure if a tunnel configuration
with empty UDP ports is requested.
Fixes: 17551f6dff ("net/sfc/base: add API to control UDP tunnel ports")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Adding support to set RSS level from ethdev config.
level-default will requests the default behavior.
level-outer will requests RSS to be performed on the outermost packet
encapsulation level.
level-inner will request RSS to be performed on the specified inner
packet encapsulation level, from outermost to innermost.
Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
This patch reserves 2 bits as input selection to select inner and outer
encapsulation level for RSS computation. It is combined with existing
ETH_RSS_* to choose inner or outer layers.
This functionality already exists in rte_flow through level parameter in
RSS action configuration rte_flow_action_rss.
Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Enlarge the L3 and tunnel header length from 8-bit to 16-bit to handle
the bigger headers. And reorder the fields to avoid creating a structure
hole.
Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
The Agilio CX family of smartNIC's generally have a 1:many mapping of PF
to physical ports. Elaborate on this mapping in the PF multiport section
of the NFP PMD documentation.
Fixes: d625beafc8 ("doc: update NFP with PF support information")
Cc: stable@dpdk.org
Signed-off-by: Heinrich Kuhn <heinrich.kuhn@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Ensure packets are available before accessing the mbuf list in Tx
burst function. Otherwise, just reclaim completed Tx descriptors and
exit.
Fixes: b1df19e43e ("net/cxgbe: fix prefetch for non-coalesced Tx packets")
Cc: stable@dpdk.org
Reported-by: Brian Poole <brian90013@gmail.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Fragment offset of IPv4 header is measured in units of
8 bytes. Fragment offset of UDP fragments will be wrong
after GSO if pyld_unit_size isn't multiple of 8. Say
pyld_unit_size is 1500, fragment offset of the second
UDP fragment will be 187 (i.e. 1500 / 8), which means 1496,
and it will result in 4-byte data loss (1500 - 1496 = 4).
So UDP GRO will reassemble out a wrong packet.
Fixes: b166d4f30b ("gso: support UDP/IPv4 fragmentation")
Cc: stable@dpdk.org
Signed-off-by: Yi Yang <yangyi01@inspur.com>
Acked-by: Jiayu Hu <jiayu.hu@intel.com>
Added support for below new devices:
ICE_DEV_ID_E822L_BACKPLANE 0x1897
ICE_DEV_ID_E822L_SFP 0x1898
ICE_DEV_ID_E822L_10G_BASE_T 0x1899
ICE_DEV_ID_E822L_SGMII 0x189A
The patch also reordered items in pci_id_ice_map to align with
ice_devids.h
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Qiming Yang <qiming.yang@intel.com>
For RSS attribute don't have an associated RSS type, we need
to reject it.
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Junfeng Guo <junfeng.guo@intel.com>
def_cp_lock was added to sync race between dev_configure and
int_handler. It should not be used to synchronize scheduling of FW
health check between dev_start and async event handler as well,
use a separate mutex for the same.
Fixes: a73b8e939f ("net/bnxt: fix race between start and interrupt handler")
Cc: stable@dpdk.org
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Add a couple of NULL pointer checks in bnxt_free_all_filters()
and bnxt_free_vnics() respectively to guard against certain error
injection/recovery scenarios where it was found that the application
was crashing with the bp->vnic_info pointer being NULL.
Fixes: 51fafb89a9 ("net/bnxt: get rid of ff pools and use VNIC info array")
Cc: stable@dpdk.org
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Added support for mutex protection for the flow database to prevent
simultaneous access to flow database and protect flow creation and
deletion.
Signed-off-by: Kishore Padmanabha <kishore.padmanabha@broadcom.com>
Reviewed-by: Mike Baucom <michael.baucom@broadcom.com>
Reviewed-by: Shahaji Bhosle <sbhosle@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
1.Representor Rx ring producer index was not getting reset in
the ring full case. Fix it by incrementing only in
success case.
2.Instead of calling the mbuf specific routine to free the mbuf when
representor ring is full rte_free was being called leading to
'invalid memory' errors being logged.
3. Do not account the pkt meant for the representor in the parent
Rx ring's array that is returned to the application.
Fixes: 6dc83230b4 ("net/bnxt: support port representor data path")
Cc: stable@dpdk.org
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Reviewed-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Some applications need switch_info of the device to be returned
as a part of eth_dev_info_get(). The offload logic in such
applications could use this info. Pass this info to the when VF
representors are configured.
Fixes: 322bd6e702 ("net/bnxt: add port representor infrastructure")
Cc: stable@dpdk.org
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
