Jiawei Wang [Mon, 26 Jul 2021 06:22:33 +0000 (09:22 +0300)]
net/mlx5: fix mirror flow split with L3 encapsulation
Due to hardware limitations, the decap action (such as
VXLAN/NVGRE/RAW decap) can't follow the sample action in the
same flow, to keep the original action order of sample and decap
actions the flow was internally split into two subflows by PMD,
the sample action was moved into prefix subflow in the original table,
and decap action was moved into suffix subflow in the new table.
There is a specific combination of raw decap and raw encap actions
to specify "L3 encapsulation" packet transformation - raw decap action
to remove L2 header and raw encap to add the tunnel header.
This specific L3 encapsulation is encoded as a single packet reformat
hardware transaction and is supported by hardware after sample
action (no hardware limitations for packet reformat).
The "L3 encapsulation" with mirror actions in the same flow was not handled
correctly in the previous commit.
The patch checks whether the decap action is part of "L3 encapsulation"
and does not move the decap action into suffix subflow for the case.
Fixes: cafd87f62a06 ("net/mlx5: fix VLAN push/pop and decap actions with mirror") Cc: stable@dpdk.org Signed-off-by: Jiawei Wang <jiaweiw@nvidia.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
net/mlx5: fix queue leaking in hairpin auto bind check
During the start up stage, the hairpin auto bind was executed for
each port. All the Tx and Rx queues configured for this port should
be checked to confirm if the auto bind of hairpin is needed.
1. The queue is hairpin queue.
2. The peer port is the same one and the peer queue should also be
with hairpin type.
3. The manual bind attribute is not set for this queue.
If the queue is not a hairpin queue or it doesn't need to be bound
automatically, the reference count should be decreased by 1 since
the count was increased when calling the mlx5_*xq_get().
When the peer port is not the same, it means that no auto bind is
supported and the mlx5_*xq_release() was missed in the current
implementation.
By calling the release function before continue, the count is
correct when calling the device close.
In mlx5 PMD the PCI device interrupt vector was used by Uplink
representor exclusively and other VF representors did not support
interrupt mode.
All the VFs and Uplink representors are separate ethernet devices
and must have dedicated interrupt vectors.
The fix provides each representor with a dedicated interrupt
vector.
Kernel PF may not respond to virtual channel commands
VIRTCHNL_OP_GET_RSS_HENA_CAPS and VIRTCHNL_OP_SET_RSS_HENA, which
will cause VF to fail to start.
RSS offload type configuration is not a necessary feature for VF,
so in order to improve VF compatibility, in this patch the PMD will
ignore the error result of above two commands and will print warnings
instead.
Xiaoyun Li [Thu, 22 Jul 2021 07:56:20 +0000 (15:56 +0800)]
net/iavf: fix Tx threshold check
Function check_tx_thresh is called with wrong parameter. If the
check fails, tx_queue_setup should return error not keep going.
This patch fixes above issues.
Fixes: 69dd4c3d0898 ("net/avf: enable queue and device") Cc: stable@dpdk.org Signed-off-by: Xiaoyun Li <xiaoyun.li@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
When virtio front-end initializes, the duplex mode should be set
unknown before reading any duplex mode information from configuration
space. This patch fixes the issue that duplex mode is by default set
to zero, which equals ETH_LINK_HALF_DUPLEX. This will lead to duplex
mode being half duplex when front-end does not have the feature
named VIRTIO_NET_F_SPEED_DUPLEX.
Fixes: 1357b4b36246 ("net/virtio: support Virtio link speed feature") Cc: stable@dpdk.org Signed-off-by: Chenbo Xia <chenbo.xia@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Gaoxiang Liu [Mon, 26 Jul 2021 14:42:05 +0000 (22:42 +0800)]
net/virtio: fix interrupt handle leak
Free memory of interrupt handle in virtio_user_dev_uninit() to
avoid memory leak.
when virtio user dev closes, memory of interrupt handle is not freed
that is allocated in virtio_user_fill_intr_handle().
Fixes: 3d4fb6fd2505 ("net/virtio-user: support Rx interrupt") Cc: stable@dpdk.org Signed-off-by: Gaoxiang Liu <liugaoxiang@huawei.com> Reviewed-by: Chenbo Xia <chenbo.xia@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Maxime Coquelin [Mon, 26 Jul 2021 07:58:14 +0000 (09:58 +0200)]
vhost: fix crash on reconnect
When the vhost-user frontend like Virtio-user tries to
reconnect to the restarted Vhost backend, the Vhost backend
segfaults when multiqueue is enabled.
This is caused by VHOST_USER_GET_VRING_BASE being called for
a virtqueue that has not been created before, causing a NULL
pointer dereferencing.
This patch adds the VHOST_USER_GET_VRING_BASE requests to
the list of requests that trigger queue pair allocations.
Fixes: 160cbc815b41 ("vhost: remove a hack on queue allocation") Cc: stable@dpdk.org Reported-by: Yinan Wang <yinan.wang@intel.com> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com> Tested-by: Yinan Wang <yinan.wang@intel.com> Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Ivan Ilchenko [Wed, 21 Jul 2021 09:22:25 +0000 (12:22 +0300)]
net/virtio: report maximum MTU in device info
Fix the driver to report maximum MTU obtained from config if
VIRTIO_NET_F_MTU is supported or calculated based on maximum
Rx packet length.
Fixes: ad97ceece12c ("ethdev: add min/max MTU to device info") Cc: stable@dpdk.org Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Andrew Rybchenko [Thu, 17 Jun 2021 14:20:25 +0000 (17:20 +0300)]
app/testpmd: send failure logs to stderr
Running with stdout suppressed or redirected for further processing
is very confusing in the case of errors. Fix it by logging errors and
warnings to stderr.
Since lines with log messages are touched anyway concatenate split
format strings to make it easier to search using grep.
Fix indent of format string arguments.
Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Acked-by: Xiaoyun Li <xiaoyun.li@intel.com>
app/testpmd: remove most uses of internal ethdev array
This patch removes most uses of the global variable rte_eth_devices
from testpmd. This was done to avoid using the object directly which
applications should not do.
Most uses have been replaced with standard function calls, however
the use of it in the show_macs function could not be replaced as no
function call exists to get all mac addresses of a given port.
MAC address of each port in global variable ports hasn't been updated
after resetting. It was the initial one after resetting VF MAC address.
This patch gets correct port MAC address when starting port.
Fixes: a5279d25616d ("app/testpmd: check status of getting MAC address") Cc: stable@dpdk.org Signed-off-by: Yuying Zhang <yuying.zhang@intel.com> Acked-by: Aman Deep Singh <aman.deep.singh@intel.com> Acked-by: Xiaoyun Li <xiaoyun.li@intel.com>
Huisong Li [Fri, 23 Apr 2021 11:01:11 +0000 (19:01 +0800)]
sched: fix profile allocation failure handling
This patch fixes return value judgment when allocate memory to store the
subport profile, and releases memory of 'rte_sched_port' if code fails to
apply for this memory.
Fixes: 0ea4c6afcaf1 ("sched: add subport profile table") Cc: stable@dpdk.org Signed-off-by: Huisong Li <lihuisong@huawei.com> Signed-off-by: Min Hu (Connor) <humin29@huawei.com> Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
power: check frequencies count before filling array
The freqs array size is RTE_MAX_LCORE_FREQS. Before filling the
array with num_freqs elements, restrict the total num to
RTE_MAX_LCORE_FREQS. This fix aims to fix the coverity scan issue
like:
Overrunning array "pi->freqs" of 256 bytes by passing it to a
function which accesses it at byte offset 464.
Coverity issue: 371913 Fixes: ef1cc88f1837 ("power: support cppc_cpufreq driver") Cc: stable@dpdk.org Signed-off-by: Richael Zhuang <richael.zhuang@arm.com> Acked-by: David Hunt <david.hunt@intel.com>
The first argument to rte_bsf32_safe was incorrectly declared as
a 64 bit value. The code only works on 32 bit values and the underlying
function rte_bsf32 only accepts 32 bit values. This was a mistake
introduced when the safe version was added and probably cause
by copy/paste from the 64 bit version.
The bug passed silently under the radar until some other code was
built with -Wall and -Wextra in C++ and C++ complains about the
missing cast.
Yes, this is a API signature change, but the original code was wrong.
It is an inline so not an ABI change.
Ivan Malov [Wed, 21 Jul 2021 04:15:05 +0000 (07:15 +0300)]
net/sfc: improve logging in MAE backend of flow API
Errors detected during parsing of pattern items and actions
are reflected by setting RTE error, but the name of the bad
element is not disclosed, thus leaving the user to join the
dots themselves. Adjust the code to log missing information.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru> Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Ivan Malov [Wed, 21 Jul 2021 04:15:04 +0000 (07:15 +0300)]
net/sfc: extend logging in MAE backend of flow API
Extra log statements will provide more details to the user
in the case of errors discovered in the pattern or actions.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru> Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Ivan Ilchenko [Fri, 23 Jul 2021 13:15:14 +0000 (16:15 +0300)]
net/sfc: add xstats for Rx/Tx doorbells
Rx/Tx doorbells statistics are collected in software and
available per queue. These stats are useful for performance
investigation.
Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Ivan Ilchenko [Fri, 23 Jul 2021 13:15:13 +0000 (16:15 +0300)]
net/sfc: prepare to add more xstats
Move getting MAC stats code that involves locking to separate functions
to simplify addition of new xstats.
Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Ivan Ilchenko [Fri, 23 Jul 2021 13:15:12 +0000 (16:15 +0300)]
net/sfc: simplify getting xstats count
There is no point to recalculate number of available xstats on
each request. The number is calculated once on device start
and may be returned on subsequent calls.
Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Ivan Ilchenko [Fri, 23 Jul 2021 13:15:10 +0000 (16:15 +0300)]
net/sfc: fix xstats query by unsorted list of IDs
Device may support only some MAC stats. Add mapping from ids to subset
of supported MAC stats for each port.
Fixes: 73280c1e4ff ("net/sfc: support xstats retrieval by ID") Cc: stable@dpdk.org Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Ivan Ilchenko [Fri, 23 Jul 2021 13:15:09 +0000 (16:15 +0300)]
net/sfc: fix xstats query by ID according to ethdev
Fix xstats by ID callbacks according to ethdev usage.
Handle combinations of input arguments that are required by ethdev
and sanity check and reject other combinations on callback entry.
Fixes: 73280c1e4ff ("net/sfc: support xstats retrieval by ID") Cc: stable@dpdk.org Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Ivan Ilchenko [Fri, 23 Jul 2021 13:15:06 +0000 (16:15 +0300)]
net/sfc: fix reading adapter state without locking
Update MAC stats function reads adapter state with MAC stats locking
but without adapter locking. Add adapter locking before calling this
function and remove MAC stats locking since there's no point to have
it together with adapter locking. The second place MAC stats locking
is used is MAC stats reset function. It's called with adapter being
already locked so there's no point to use MAC stats locking anymore.
Fixes: 1caab2f1e68 ("net/sfc: add basic statistics") Cc: stable@dpdk.org Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Ivan Ilchenko [Fri, 23 Jul 2021 13:15:05 +0000 (16:15 +0300)]
net/sfc: fix MAC stats lock in xstats query by ID
Add MAC stats lock in xstats_get_by_id() callback before reading
number of supported MAC stats.
Fixes: 73280c1e4ff ("net/sfc: support xstats retrieval by ID") Cc: stable@dpdk.org Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
examples/vhost: handle memory hotplug for async vhost
When the guest memory is hotplugged, the vhost application which
enables DMA acceleration must stop DMA transfers before the vhost
re-maps the guest memory.
To accomplish that, we need to do these changes in the vhost sample:
1. add inflight packets count.
2. add vring_state_changed() callback.
3. add inflight packets clear process in destroy_device() and
vring_state_changed().
Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Jiayu Hu [Fri, 23 Jul 2021 08:09:36 +0000 (08:09 +0000)]
vhost: handle memory hotplug for async vhost
When the guest memory is hotplugged, the vhost application which
enables DMA acceleration must stop DMA transfers before the vhost
re-maps the guest memory.
This patch is to notify the vhost application of stopping DMA
transfers.
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Applications need to stop DMA transfers and finish all the inflight
packets when in VM memory hot-plug case and async vhost is used. This
patch is to provide an unsafe API to clear inflight packets which
are submitted to DMA engine in vhost async data path. Update the
program guide and release notes for virtqueue inflight packets clear
API in vhost lib.
Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The async vhost callback ops should return negative value when there
are something wrong in the callback, so the return type should be
changed into int32_t. The issue in vhost example is also fixed.
Fixes: cd6760da1076 ("vhost: introduce async enqueue for split ring") Fixes: 819a71685826 ("vhost: fix async callback return type") Fixes: 6b3c81db8bb7 ("vhost: simplify async copy completion") Fixes: abec60e7115d ("examples/vhost: support vhost async data path") Fixes: 6e9a9d2a02ae ("examples/vhost: fix ioat dependency") Fixes: 873e8dad6f49 ("vhost: support packed ring in async datapath") Cc: stable@dpdk.org Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
This patch adds validation of the port id for
rte_pmd_dpaa2_set_custom_hash API to check if the
port is a valid DPAA2 port. Also handles some
edge cases in the rte_pmd_dpaa2_mux_flow_create API.
DPAA2 hardware require a hardware portal context.
If a thread doing DPAA2 i/o do not have portal, it will
allocate it on run-time. This may cause a delay in the
datapath at run-time. To avoid it, it is better to allocate
a hw context portal at the start of thread expected to do
i/o with DPAA2 hardware.
This patch makes necessary changes for the same and creates
a pmd API to allocate a hw context portal for a thread.
Michal Krawczyk [Fri, 23 Jul 2021 10:24:54 +0000 (12:24 +0200)]
net/ena: update version to 2.4.0
This version update contains:
* Rx interrupts feature,
* Support for the RSS hash function reconfiguration,
* Small rework of the works,
* Reset trigger on Tx path fix.
Michal Krawczyk [Fri, 23 Jul 2021 10:24:53 +0000 (12:24 +0200)]
net/ena: rework RSS configuration
Allow user to specify his own hash key and hash ctrl if the
device is supporting that. HW interprets the key in reverse byte order,
so the PMD reorders the key before passing it to the ena_com layer.
Default key is being set in random matter each time the device is being
initialized.
Moreover, make minor adjustments for reta size setting in terms
of returning error values.
RSS code was moved to ena_rss.c file to improve readability.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Shai Brandes <shaibran@amazon.com> Reviewed-by: Shay Agroskin <shayagr@amazon.com> Reviewed-by: Amit Bernstein <amitbern@amazon.com>
Michal Krawczyk [Fri, 23 Jul 2021 10:24:52 +0000 (12:24 +0200)]
net/ena: support Rx interrupt
In order to support asynchronous Rx in the applications, the driver has
to configure the event file descriptors and configure the HW.
This patch configures appropriate data structures for the rte_ethdev
layer, adds .rx_queue_intr_enable and .rx_queue_intr_disable API
handlers, and configures IO queues to work in the interrupt mode, if it
was requested by the application.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Artur Rojek <ar@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Shai Brandes <shaibran@amazon.com> Reviewed-by: Shay Agroskin <shayagr@amazon.com>
Michal Krawczyk [Fri, 23 Jul 2021 10:24:51 +0000 (12:24 +0200)]
net/ena: trigger reset on Tx prepare failure
If the prepare function failed, then it means the descriptors are in the
invalid state.
This condition now triggers the reset, which should be further handled
by the application.
To notify the application about prepare function failure, the error log
was added. In general, it should never fail in normal conditions, as the
Tx function checks for the available space in the Tx ring before the
preparation even starts.
Michal Krawczyk [Fri, 23 Jul 2021 10:24:50 +0000 (12:24 +0200)]
net/ena: use common debug options
ENA defined its own logger flags for Tx and Rx, but they weren't
technically used anywhere. Those data path loggers weren't used anywhere
after the definition.
This commit uses the generic RTE_ETHDEV_DEBUG_RX and RTE_ETHDEV_DEBUG_TX
flags to define PMD_TX_LOG and PMD_RX_LOG which are now being used on
the data path. The PMD_TX_FREE_LOG was removed, as it has no usage in
the current version of the driver.
RTE_ETH_DEBUG_[TR]X now wraps extra checks for the driver state in the
IO path - this saves extra conditionals on the hot path.
ena_com logger is no longer optional (previously it had to be explicitly
enabled by defining this flag: RTE_LIBRTE_ENA_COM_DEBUG). Having this
logger optional makes tracing of ena_com errors much harder.
Due to ena_com design, it's impossible to separate IO path logs
from the management path logs, so for now they will be always enabled.
Default levels for the affected loggers were modified. Hot path loggers
are initialized with the default level of DEBUG instead of NOTICE, as
they have to be explicitly enabled. ena_com logging level was reduced
from NOTICE to WARNING - as it's no longer optional, the driver should
report just a warnings in the ena_com layer.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Shai Brandes <shaibran@amazon.com>
Michal Krawczyk [Fri, 23 Jul 2021 10:24:49 +0000 (12:24 +0200)]
net/ena: adjust logs
ENA logs were not consistent regarding the new line character. Few of
them were relying on the new line character added by the PMD_*_LOG
macros, but most were adding the new line character by themselves. It
was causing ENA logs to add extra empty line after almost each log.
To unify this behavior, the missing new line characters were added to
the driver logs, and they were removed from the logging macros. After
this patch, every ENA log message should add '\n' at the end.
Moreover, the logging messages were adjusted in terms of wording
(removed unnecessary abbreviations), capitalizing of the words (start
sentences with capital letters, and use 'Tx/Rx' instead of 'tx/TX' etc.
Some of the logs were rephrased to make them more clear for the reader.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Shai Brandes <shaibran@amazon.com>
The configuration of MTU is inconsistent in the driver and
firmware when the port is stopped, started and reconfigured.
Before, HINIC_MAX_JUMBO_FRAME_SIZE include VLAN tag, but when
frame and pktlen are converted to each other do not include
VLAN tag. And port_mtu_set function will use HINIC_MAX_JUMBO_FRAME_SIZE
to calculate eth_overhead, so MTU will be inconsistent in the driver and
firmware.
Fixes: e542ab51ab27 ("net/hinic: fix jumbo frame flag condition for MTU set") Cc: stable@dpdk.org Signed-off-by: Guoyang Zhou <zhouguoyang@huawei.com>
The Rx queue must config as ceq disables, and must set MSI-X
state disabled. Otherwise when LRO is enables, there will be
problems with packet aggregation because of firmware.
Huisong Li [Sat, 17 Jul 2021 01:04:19 +0000 (09:04 +0800)]
net/hns3: disable PFC if not configured
If "dcb_capability_en" in "data->dev_conf" delivered from the dev_configure
does not have the ETH_DCB_PFC_SUPPORT flag, the user wants to disable PFC,
and only enable ETS. Therefore, this patch supports the function of
disabling PFC by the field. In addition, this patch updates
"current_fc_status" of the driver based on the flow control mode requested
by user so as to enable the flow control mode in multi-TC scenarios.
Signed-off-by: Huisong Li <lihuisong@huawei.com> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
Huisong Li [Sat, 17 Jul 2021 02:02:56 +0000 (10:02 +0800)]
net/hns3: fix Tx prepare after stop
In some special scenarios, such as TSO scenarios, the user layer may need
to call the tx_pkt_prepare(), and then call tx_pkt_burst() to send packets.
If the return value of tx_pkt_parepare() isn't equal to the numbers of
packets requested to send, warning message may be printed at the user
layer. Currently, tx_pkt_prepare() is assigned to dummy function when
dev_stop() is called in hns3 PMD. At this moment, if user layer continues
to send packets, the warning message will always be printed. So this patch
modifies the address to NULL.
Fixes: 2790c6464725 ("net/hns3: support device reset") Cc: stable@dpdk.org Signed-off-by: Huisong Li <lihuisong@huawei.com> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
Currently, hns3 driver saves rte_flow list into the
rte_eth_dev.process_private field, it may cause following problem:
The FDIR/RSS rules cannot be managed in a unified manner because
the management structure is not visible between processes.
This patch fixes it by moving rte_flow list to struct hns3_hw which is
visible between processes.
Fixes: fcba820d9b9e ("net/hns3: support flow director") Fixes: c37ca66f2b27 ("net/hns3: support RSS") Cc: stable@dpdk.org Signed-off-by: Chengwen Feng <fengchengwen@huawei.com> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
Huisong Li [Sat, 17 Jul 2021 02:02:54 +0000 (10:02 +0800)]
net/hns3: move speed auto-negotiation warning
PF driver prints a warning on device that does not support auto-negotiation
when user does not configure "link_speeds" (default 0), which means
auto-negotiation. Currently, this warning information is printed in
dev_configure stage and a success is returned. Perhaps the user may call
dev_configure multiple times before dev_start for some reason or purpose.
In this case, this message may be printed multiple times. So this patch
moves it to dev_start stage.
Fixes: cfc9fe48c4d4 ("net/hns3: move link speeds check to configure") Cc: stable@dpdk.org Signed-off-by: Huisong Li <lihuisong@huawei.com> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
Currently, the PF/VF does not clear the interrupt source immediately
after receiving the interrupt. As a result, if the second interrupt
task is triggered when processing the first interrupt task, clearing
the interrupt source before exiting will clear the interrupt sources
of the two tasks at the same time. As a result, no interrupt is
triggered for the second task.
Clearing interrupt source immediately after checking event cause
ensures that:
1. Even if two interrupt tasks are triggered at the same time, they can
be processed.
2. If the second task is triggered during the processing of the first
task and the interrupt source is not cleared, the interrupt is reported
after vector0 is enabled.
Fixes: a5475d61fa34 ("net/hns3: support VF") Fixes: 3988ab0eee52 ("net/hns3: add abnormal interrupt process") Cc: stable@dpdk.org Signed-off-by: Hongbo Zheng <zhenghongbo3@huawei.com> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
Currently, even if we fail to remove the origin MAC address from the HW,
the set_default_mac will go on, and add the new MAC address to the HW.
Eventually cause the original MAC address entry to remain in the HW, and
users may receive unexpected packets.
This patch make set_default_mac return directly to failure if deleting
the original MAC address fails, simplifying the behavior of the driver
and solving the problem of residual MAC address entry.
Fixes: 7d7f9f80bbfb ("net/hns3: support MAC address related operations") Cc: stable@dpdk.org Signed-off-by: Chengchang Tang <tangchengchang@huawei.com> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
net/af_packet: run on kernel without qdisc bypass support
Some older kernels do not support the PACKET_QDISC_BYPASS socket
option. Such an example is the CentOS 7 kernel (3.10).
If we only check for the definition of PACKET_QDISC_BYPASS, it might mean
that we will not be able to compile the PMD driver on a newer platform,
and run in on a machine with an older kernel.
Setting the socket option only if it is specifically requested from
the EAL arguments, allows us to have a way to run the PMD compiled
against newer kernel headers, on platforms having older kernels.
Signed-off-by: Tudor Cornea <tudor.cornea@keysight.com> Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
In function pmd_parse_args(), firmware path is duplicated from device
arguments as character string, but is never freed, which cause memory
leak.
This patch changes the type of firmware member of struct pmd_params to
character array, to make memory resource release unnecessary, and
changes the type of name member to character array, to keep the
consistency of character string handling in struct pmd_params.
Tomasz Duszynski [Thu, 15 Jul 2021 13:53:27 +0000 (08:53 -0500)]
common/cnxk: support reading BPHY CGX/RPM FEC
Before setting FEC for specific LMAC one needs to know which type is
actually supported because it generally differs between modes
LMAC operates in (SGMII, SFI, etc.).
Signed-off-by: Tomasz Duszynski <tduszynski@marvell.com>
Jie Zhou [Wed, 7 Jul 2021 20:25:38 +0000 (13:25 -0700)]
eal/windows: check callback parameter of alarm functions
EAL functions rte_eal_alarm_set() and rte_eal_alarm_cancel()
did not for invalid parameters in Windows implementation,
which is caught by the unit test alarm_autotest.
Enforce parameter check to fail fast for invalid parameters.
Andrew Rybchenko [Thu, 22 Jul 2021 07:49:05 +0000 (10:49 +0300)]
net/sfc: fix build with clang 3.4.2
Old clang requires libatomic as well as gcc. Avoid compiler name and
version based checks. Add custom test for 16-byte atomic operations
to find out if libatomic is required to build.
Bugzilla ID: 760 Fixes: 96fd2bd69b58 ("net/sfc: support flow action count in transfer rules") Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Acked-by: David Marchand <david.marchand@redhat.com>
Currently in scale mode, multi-queue initialization will attempt to
initialize and de-initialize the per-lcore power library structures
multiple times. Fix it to only do this whenever we either enabling
first queue or disabling last queue.
Fixes: 5dff9a72b0ef ("power: support callbacks for multiple Rx queues") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: David Hunt <david.hunt@intel.com>
Kalesh AP [Sun, 18 Jul 2021 05:30:59 +0000 (11:00 +0530)]
net/bnxt: remove workaround for default VNIC
On older Wh+ firmware versions, HWRM_FUNC_QCFG returns zero
for the parent default vnic. Commit "3fb93bc7c349" added a
temporary Wh+-specific workaround in the PMD.
This has been fixed in latest firmware and hence removing
the workaround.
Since the header type of IPv6 fragment is wrong, the L3 dst/src RSS hash
fields cannot work properly. This patch changed the header type from any
to outer.
Fixes: f1ea76eb6394 ("net/ice: support RSS hash for IP fragment") Cc: stable@dpdk.org Signed-off-by: Ting Xu <ting.xu@intel.com> Acked-by: Qi Zhang <qi.z.zhang@intel.com>
When closing DCF, the bandwidth limit configured for VFs by DCF is not
cleared correctly. The configuration will still take effect when DCF starts
again, if VFs are not re-allocated. This patch cleared VFs bandwidth limit
when DCF closes, and DCF needs to re-configure bandwidth for VFs when it
starts next time.
Fixes: 3a6bfc37eaf4 ("net/ice: support QoS config VF bandwidth in DCF") Signed-off-by: Ting Xu <ting.xu@intel.com> Acked-by: Qi Zhang <qi.z.zhang@intel.com>
The RoCE LAG bond device requires neither E-Switch nor SR-IOV
configurations. It means the RoCE LAG bond device might be
presented as a single port Infiniband device.
The mlx5 PMD wrongly recognized standalone RoCE LAG bond device
as E-Switch configuration, this triggered the calls of E-Switch
ports related API and the latter failed (over the new OFED kernel
driver, starting since 5.4.1), causing the overall device probe
failure.
If there is a single port Infiniband bond device found the
E-Switch related flags must be cleared indicating standalone
configuration.
Also, it is not true anymore the bond device can exist
over E-Switch configurations only (as it was claimed for VF LAG
bond devices). The related checks are not relevant anymore
and removed.
Fixes: 790164ce1d2d ("net/mlx5: check kernel support for VF LAG bonding") Cc: stable@dpdk.org Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com> Acked-by: Matan Azrad <matan@nvidia.com>
The Mark action is a two-stage process in the Mellanox driver.
First, a hardware register is filled with the required value,
then this value is registered in the software resource table.
The MODIFY_FIELD action can instruct a Mellanox NIC to copy
some value from an arbitrary packet header field into the
hardware register, associated with the Mark item. But there
is no way NIC can modify the software resource table as well.
Due to these driver limitations the copying of arbitrary value
to the MARK can not be supported and should be rejected in the
MODIFY_FIELD action.
Signed-off-by: Alexander Kozyrev <akozyrev@nvidia.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
net/mlx5: fix meta register conversion for extensive mode
Register C is used in the extensive metadata mode number 1 and its
width can vary from 0 to 32 bits depending on the kernel usage of it.
There are several issues associated with this mode (dv_xmeta_en=1):
1. The metadata setting assumes that the width is always 16 bits,
which is the most common case in this mode. Use the proper mask.
2. The same is true for the modify_field Flow API. 16-bits width
is hardcoded for dv_xmeta_en=1. Switch to the register C mask width.
3. Metadata is stored in the most significant bits in CQE in this
mode because the registers copy code was not updated during the
metadata conversion to the big-endian format. Update this code to
avoid shifting the metadata in the datapath.
Fixes: b57e414b48 ("net/mlx5: convert meta register to big-endian") Cc: stable@dpdk.org Signed-off-by: Alexander Kozyrev <akozyrev@nvidia.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
mlx5_ind_table_obj_modify() first references queues from the new list,
then applies the new list to HW. In case of apply failure the function
dereferenced queues from the old list, while it should be the new list.
When device configuration was interrupted by a signal,
mlx5_rxq/txq_release() could access yet unitinialized array
and crash the application. Add checks whether queue array
is initialized.
For ConnectX-6 in FDB domain, pop and push VLAN
on both ingress and egress directions are supported.
For ConnectX-6 in NIC domain, and ConnectX-5 in both FWD and NIC domain,
pop VLAN is only supported on ingress direction,
push VLAN is only supported on egress direction.
Michael Baum [Mon, 12 Jul 2021 07:06:44 +0000 (10:06 +0300)]
regex/mlx5: fix redundancy in device removal
In the removal function, PMD releases all driver resources and
cancels the regexdev registry.
However, regexdev registration is accidentally canceled twice.
Remove one of them.
Fixes: b34d816363b5 ("regex/mlx5: support rules import") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Ori Kam <orika@nvidia.com>
Michael Baum [Mon, 12 Jul 2021 07:06:43 +0000 (10:06 +0300)]
regex/mlx5: fix leak on device removal
In the removal function, PMD releases all driver resources allocated
in the probe function.
The MR btree memory is allocated in the probe function, but it is not
freed in remove function what caused a memory leak.
Release it.
Fixes: cda883bbb655 ("regex/mlx5: add dynamic memory registration to datapath") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Ori Kam <orika@nvidia.com>
Michael Baum [Mon, 12 Jul 2021 07:06:42 +0000 (10:06 +0300)]
regex/mlx5: fix memory region unregistration
The issue can cause illegal physical address access while a huge-page A
is released and huge-page B is allocated on the same virtual address.
The old MR can be matched using the virtual address of huge-page B but
the HW will access the physical address of huge-page A which is no more
part of the DPDK process.
Register a driver callback for memory event in order to free out all the
MRs of memory that is going to be freed from the DPDK process.
Fixes: cda883bbb655 ("regex/mlx5: add dynamic memory registration to datapath") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Ori Kam <orika@nvidia.com>
Michael Baum [Thu, 1 Jul 2021 06:39:16 +0000 (09:39 +0300)]
net/mlx5: fix overflow in mempool argument
The mlx5_mprq_alloc_mp function makes shifting to the numeric constant
1, for sending it as a parameter to rte_mempool_create function.
The rte_mempool_create function expects to get void pointer (uintptr_t,
might be 64-bit) and instead gets a 32-bit variable, because the
numeric constant size is a 32-bit.
In case the shift is greater than 32 the variable might lose its value
even though the function might get 64-bit argument.
Change the size of the numeric constant 1 to uintptr_t.
Michael Baum [Thu, 1 Jul 2021 06:39:15 +0000 (09:39 +0300)]
vdpa/mlx5: fix overflow in queue attribute
The mlx5_vdpa_event_qp_create function makes shifting to the numeric
constant 1, then multiplies it by another constant and finally assigns
it into a uint64_t variable.
The numeric constant type is an int with a 32-bit sign. if after
shifting , its MSB (bit of sign) will change, the uint64 variable will
get into it a different value than what the function intended it to get.
Set the numeric constant 1 to be uint64_t in the first place.
Michael Baum [Thu, 1 Jul 2021 06:39:14 +0000 (09:39 +0300)]
compress/mlx5: fix overflow in queue size
The mlx5_compress_qp_setup function makes shifting to the numeric
constant 1, then sends it as a parameter to rte_calloc function.
The rte_calloc function expects to get size_t (might be 64 bit) and
instead gets a 32-bit variable, because the numeric constant size is a
32-bit.
In case the shift is greater than 32 bit and it 64-system, the variable
will lose its value even though the function can get 64-bit argument.
Change the size of the numeric constant 1 to size_t.
Fixes: 8619fcd5161b ("compress/mlx5: support queue pair operations") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Matan Azrad <matan@nvidia.com>
Michael Baum [Thu, 1 Jul 2021 06:39:13 +0000 (09:39 +0300)]
regex/mlx5: fix size of setup constants
The constant representing the size of the metadata is defined as an
unsigned int variable with 32-bit.
Similarly the constant representing the maximal output is also defined
as an unsigned int variable with 32-bit.
There is potentially overflowing expression when those constants are
evaluated using 32-bit arithmetic, and then used in a context that
expects an expression of type size_t that might be 64-bit.
Change the size of the above constants to size_t.
Fixes: 30d604bb1504 ("regex/mlx5: fix type of setup constants") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Matan Azrad <matan@nvidia.com>
net/mlx5: check consistency of meter policy and profile
In the previous implementation, only green color policy was
supported in mlx5 PMD. Since yellow color policy is supported now,
the consistency of meter policy and profile should be checked.
1. If the profile supports yellow but the policy doesn't, an error
should be returned when creating the meter. Or else, there is
no explicit steering action for the packets marked with yellow.
2. If the policy supports yellow but the profile doesn't, it will
be considered as a valid case. Even if no packet will be
handled with the yellow steering action, it is just like that
only the green policy presents.
Usually the green color is supported by default, but when it is
disabled intentionally with setting the CBS to a small value like
zero in the profile, the similar checking on green policy and
profile should also be done.
net/mlx5: support yellow in meter policy validation
In the previous implementation, the policy for yellow color was not
supported. The action validation for yellow was skipped.
Since the yellow color policy needs to be supported, the validation
should also be done for the yellow color. In the meanwhile, due to
the fact that color policies of one meter should be used for the
same flow(s), the domains supported of both colors should be the
same. If both of the colors have RSS as the termination actions,
except the queues, all other parameters of RSS should be the same.
If the fate action is either RSS or Queue of a meter policy, the
action will only be created in the flow splitting stage. With queue
as the fate action, only one sub-policy is needed. And RSS will
have more than one sub-policies if there is an expansion.
Since the RSS parameters are the same for both green and yellow
colors except the queues, the expansion result will be unique.
Even if only one color has the RSS action, the checking and possible
expansion will be done then. For each sub-policy, the action rules
need to be created separately on its own policy table.
When creating a meter policy, both / either of the action rules for
green and yellow colors may be provided. After validation, usually
the actions are created before the meter is using by a flow rule.
If there is action specified for the yellow color, the action rules
should be created together with green color in the same time. The
action of green / yellow color can be empty, then the default
behavior is the jump action of the rule, just the same as that of
the default policy.
If the fate action of either one color is queue / RSS, all the
actions rules will be created on the flow splitting stage instead of
the policy adding stage.
net/mlx5: enable meter bucket overflow for yellow color
To support the meter policy for yellow action, the prerequisite is
that the hardware needs to support the EBS, as defined in the
RFC2697.
https://datatracker.ietf.org/doc/html/rfc2697
Then some of the packets can be marked as yellow if the tokens of C
bucket is not enough but enough in E bucket. The color could be used
for the further steering of the packets.
In the current implementation EBS and overflow were ignored when
creating a meter profile. With this commit, if EBS is set by the
application, the generation of yellow color will be enabled in the
hardware for flow rules steering of packets.
net/mlx5: handle yellow case in default meter policy
In order to support the yellow color for the default meter policy,
the default policy action for yellow should be created together
with the green policy.
The default policy action for yellow action is the same as that for
green. In the same table, the same matcher will be reused for yellow
and the destination group will be the same.
git.droids-corp.org / dpdk.git / log