dpdk.git
4 years agovhost: prefix vDPA enum value for PCI address type
Maxime Coquelin [Thu, 26 Mar 2020 14:14:22 +0000 (15:14 +0100)]
vhost: prefix vDPA enum value for PCI address type

In order to avoid potential conflicts, rename the PCI_ADDR
enum value to VDPA_ADDR_PCI in vdpa_addr_type_enum.

All symbols referencing this enum are experimental, so it
does not break API policy.

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agovhost: make IOTLB cache name unique among processes
Itsuro Oda [Wed, 11 Mar 2020 23:19:18 +0000 (08:19 +0900)]
vhost: make IOTLB cache name unique among processes

Currently, iotlb cache name is comprised of vid and virtqueue
index. For example, "iotlb_cache_0_0". Because vid is assigned
per process, iotlb cache name is not unique among multi processes.
For example a secondary process uses a vhost
(ex. eth_vhost0,iface=/tmp/sock0) and another secondary process
uses a vhost (ex. eth_vhost1,iface=/tmp/sock1), iotlb cache
name of both vhost ("iotlb_cache_0_0") are same and as a result
iotlb cache is broken.

This patch makes iotlb cache name unique among milti processes
by adding process id to the iotlb cache name.

The prefix of the name is shortened to "iotlb_" since the maximum
length of pool name is 25 bytes (RTE_MEMPOOL_NAMESIZE is 26).
Note that it is just 25 characters in maximum at the moment.
Here,
* pid_t == int: max 10 digits.
* vid < MAX_VHOST_DECICE(1024): max 4 digits.
* vq_index < VHOST_MAX_VRING(256): max 3 digits.

Fixes: d012d1f293f4 ("vhost: add IOTLB helper functions")
Cc: stable@dpdk.org
Signed-off-by: Itsuro Oda <oda@valinux.co.jp>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agovhost: remove unused variable
Xiaolong Ye [Sat, 7 Mar 2020 13:22:35 +0000 (21:22 +0800)]
vhost: remove unused variable

VHOST_FEATURES has been removed in previous refactoring.

Fixes: 0917f9d1f059 ("vhost: use new APIs to handle features")
Cc: stable@dpdk.org
Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agonet/virtio: fix outdated comment
Xiaolong Ye [Sat, 7 Mar 2020 13:22:34 +0000 (21:22 +0800)]
net/virtio: fix outdated comment

Fix comment that is no more correct as the code evolved.

Fixes: 9470427c88e1 ("net/virtio: do not store PCI device pointer at shared memory")
Cc: stable@dpdk.org
Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agonet/vhost: fix potential memory leak on close
Itsuro Oda [Thu, 5 Mar 2020 02:54:50 +0000 (11:54 +0900)]
net/vhost: fix potential memory leak on close

If a vhost device is closed before eth_dev_configure is done
to the device, internal resources allocated to the device
would not be freed. This patch fixes it.

Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
Cc: stable@dpdk.org
Signed-off-by: Itsuro Oda <oda@valinux.co.jp>
Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/vhost: enable promiscuous and multicast by default
Xiaolong Ye [Wed, 26 Feb 2020 13:45:34 +0000 (21:45 +0800)]
net/vhost: enable promiscuous and multicast by default

With this patch, the promiscuous and multicast fields are initialized as
enabled for vhost PMD by default, this allows the devices to be used
when running applications that attempt to enable promiscuous or
multicast mode.
Similar things have done for other virtual PMDs by commit f165210321c4
("drivers/net: enable promiscuous and multicast by default")

Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agonet/vhost: add options for linear and external buffer
Sivaprasad Tummala [Wed, 26 Feb 2020 10:00:34 +0000 (10:00 +0000)]
net/vhost: add options for linear and external buffer

Added vHost PMD arguments 'linear-buffer' and 'ext-buffer'
to configure  'RTE_VHOST_USER_LINEARBUF_SUPPORT' and
'RTE_VHOST_USER_EXTBUF_SUPPORT' flags in the vhost library

Signed-off-by: Sivaprasad Tummala <sivaprasad.tummala@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agovhost: fix packed ring zero-copy
Marvin Liu [Mon, 24 Feb 2020 15:14:19 +0000 (23:14 +0800)]
vhost: fix packed ring zero-copy

Available buffer ID should be stored in the zmbuf in the packed-ring
dequeue path. There's no guarantee that local queue avail index is
equal to buffer ID.

Fixes: d1eafb532268 ("vhost: add packed ring zcopy batch and single dequeue")
Cc: stable@dpdk.org
Signed-off-by: Marvin Liu <yong.liu@intel.com>
Reported-by: Yinan Wang <yinan.wang@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agoexamples/vhost_blk: use common macro for minimum
Thomas Monjalon [Wed, 19 Feb 2020 10:39:22 +0000 (11:39 +0100)]
examples/vhost_blk: use common macro for minimum

The macro RTE_MIN can be used in vhost-blk example.

This change implies fixing the sign of used_len as size_t
as defined in vhost_strcpy_pad().

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agovhost/crypto: add missing user protocol flag
Fan Zhang [Wed, 29 Jan 2020 10:19:37 +0000 (10:19 +0000)]
vhost/crypto: add missing user protocol flag

This patch fixes the vhost crypto missed
"VHOST_USER_PROTOCOL_F_CONFIG" flag problem during initialization.
Newer Qemu version requires this feature enabled.

Fixes: 939066d96563 ("vhost/crypto: add public function implementation")
Cc: stable@dpdk.org
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agonet/hinic: adds Tx queue xstats members
Xiaoyun Wang [Fri, 10 Apr 2020 09:21:47 +0000 (17:21 +0800)]
net/hinic: adds Tx queue xstats members

Because some apps may pass illegal parameters, driver increases
checks on illegal parameters and DFX statistics, which includes
sge_len0 and mbuf_null txq xstats member.

Signed-off-by: Xiaoyun Wang <cloud.wangxiaoyun@huawei.com>
4 years agonet/hinic/base: optimize log style
Xiaoyun Wang [Fri, 10 Apr 2020 09:21:46 +0000 (17:21 +0800)]
net/hinic/base: optimize log style

The patch adds space between descriptors and variables in log files.

Signed-off-by: Xiaoyun Wang <cloud.wangxiaoyun@huawei.com>
4 years agonet/hinic/base: fix PF firmware hot-active problem
Xiaoyun Wang [Fri, 10 Apr 2020 09:21:45 +0000 (17:21 +0800)]
net/hinic/base: fix PF firmware hot-active problem

When FW is hotactive which means updating the FW but not needs
to reboot OS, FW returns HINIC_DEV_BUSY_ACTIVE_FW for pf driver
because firmware is being reinitialized, at which point the cmdq
initialization that relies on the fw channel will fail, so driver
should reinit the cmdq when port start.

Fixes: 0194313b2df6 ("net/hinic/base: fix port start during FW hot update")
Cc: stable@dpdk.org
Signed-off-by: Xiaoyun Wang <cloud.wangxiaoyun@huawei.com>
4 years agonet/null: add argument for no Rx
Ferruh Yigit [Mon, 2 Mar 2020 17:36:45 +0000 (17:36 +0000)]
net/null: add argument for no Rx

Add an new device argument 'no-rx', which will prevent PMD receiving
packets.

This is useful for testing when a PMD is needed only to send packets to.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/null: group device arguments
Ferruh Yigit [Mon, 2 Mar 2020 17:36:44 +0000 (17:36 +0000)]
net/null: group device arguments

Group device argument to the struct, to increase code readability.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/null: prefer unsigned int
Ferruh Yigit [Mon, 2 Mar 2020 17:36:43 +0000 (17:36 +0000)]
net/null: prefer unsigned int

Prefer 'unsigned int' storage type keyword against 'unsigned', this also
silence the checkpatch warnings.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/null: remove redundant check
Ferruh Yigit [Mon, 2 Mar 2020 17:36:42 +0000 (17:36 +0000)]
net/null: remove redundant check

There is no need to check if the argument exist or not,
`rte_kvargs_process` returns success if the argument is not provided at
all.

Fixes: c743e50c475f ("null: new poll mode driver")
Cc: stable@dpdk.org
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agodoc: add net null PMD guide
Ferruh Yigit [Mon, 2 Mar 2020 17:36:41 +0000 (17:36 +0000)]
doc: add net null PMD guide

Net null PMD was missing documentation, adding it.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/null: fix secondary burst function selection
Ferruh Yigit [Mon, 2 Mar 2020 17:36:40 +0000 (17:36 +0000)]
net/null: fix secondary burst function selection

Secondary process uses the primary process device and while setting the
Rx/Tx functions it uses the device arguments from the secondary process
instead of the primary ones.

This may cause primary and secondary process use different Rx/Tx
functions unintentionally.

Fixes: bccc77a6a74a ("net/null: fix multi-process Rx and Tx")
Cc: stable@dpdk.org
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/i40e/base: update version
Jiaqi Min [Wed, 8 Apr 2020 10:05:23 +0000 (10:05 +0000)]
net/i40e/base: update version

Update base code release version in readme.

Signed-off-by: Jiaqi Min <jiaqix.min@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
4 years agonet/i40e/base: add constants for PTP pins
Jiaqi Min [Wed, 8 Apr 2020 10:05:22 +0000 (10:05 +0000)]
net/i40e/base: add constants for PTP pins

Introduce constants for handling PTP pins used for external
clock source.

Signed-off-by: Piotr Kwapulinski <piotr.kwapulinski@intel.com>
Signed-off-by: Jiaqi Min <jiaqix.min@intel.com>
Acked-by: Piotr Kwapulinski <piotr.kwapulinski@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
4 years agonet/i40e/base: introduce device ID for V710-TL 5G
Jiaqi Min [Wed, 8 Apr 2020 10:05:21 +0000 (10:05 +0000)]
net/i40e/base: introduce device ID for V710-TL 5G

This change is adding new device ID and handling it in the same way as
X710-T*L head of family. A new device ID is for new V710-T*L adapter
supporting speeds up to 5G.

Signed-off-by: Zalfresso-Jundzillo <marekx.zalfresso-jundzillo@intel.com>
Signed-off-by: Jiaqi Min <jiaqix.min@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
4 years agonet/i40e/base: update X722/X710 FW API version to 1.10
Jiaqi Min [Wed, 8 Apr 2020 10:05:20 +0000 (10:05 +0000)]
net/i40e/base: update X722/X710 FW API version to 1.10

update X722/X710 FW API version to 1.10.

Signed-off-by: Piotr Azarewicz <piotr.azarewicz@intel.com>
Signed-off-by: Jiaqi Min <jiaqix.min@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
4 years agonet/iavf: support more flow patterns
Xiao Zhang [Fri, 3 Apr 2020 05:42:42 +0000 (13:42 +0800)]
net/iavf: support more flow patterns

Add patterns support for AH/ESP/L2TPV3OIP/PFCP.
Added patterns are as follows:
/* GTPU */
    eth/ipv4/udp/gtpu
    eth/ipv4/udp/gtpu/gtp_psc
/* ESP */
    eth/ipv4/esp  eth/ipv4/udp/esp
    eth/ipv6/esp  eth/ipv6/udp/esp
/* AH */
    eth/ipv4/ah  eth/ipv6/ah
/* L2TPV3 */
    eth/ipv4/l2tpv3oip  eth/ipv6/l2tpv3oip
/* PFCP */
    eth/ipv4/udp/pfcp  eth/ipv6/udp/pfcp

Signed-off-by: Xiao Zhang <xiao.zhang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/iavf: support generic flow API
Qiming Yang [Fri, 3 Apr 2020 05:42:41 +0000 (13:42 +0800)]
net/iavf: support generic flow API

This patch added iavf_flow_create, iavf_flow_destroy,
iavf_flow_flush and iavf_flow_validate support,
these are used to handle all the generic filters.

This patch supported basic L2, L3, L4 and GTPU patterns.

Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/pfe: fix double free of MAC address
Yunjian Wang [Thu, 9 Apr 2020 01:59:00 +0000 (09:59 +0800)]
net/pfe: fix double free of MAC address

The 'mac_addrs' freeing has been moved to rte_eth_dev_release_port(),
so freeing 'mac_addrs' like this in pfe_eth_exit() is unnecessary and
will cause double free.

Fixes: 67fc3ff97c39 ("net/pfe: introduce basic functions")
Cc: stable@dpdk.org
Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agoapp/testpmd: fix PPPoE flow command
Xiao Zhang [Tue, 31 Mar 2020 13:29:40 +0000 (21:29 +0800)]
app/testpmd: fix PPPoE flow command

The command line to create RTE flow for specific proto_id of PPPOES can
not work.

It was:
testpmd> flow create 0 ingress pattern proto_id
 proto_id [TOKEN]: match PPPoE session protocol identifier
testpmd> flow create 0 ingress pattern proto_id proto_id
 proto_id [TOKEN]: match PPPoE session protocol identifier
testpmd> flow create 0 ingress pattern proto_id proto_id proto_id
 proto_id [TOKEN]: match PPPoE session protocol identifier

The proto_id can not be set with previous implementation.

This patch is to fix this issue, and change the command line to:
testpmd> flow create 0 pattern pppoe_proto_id is xxxx

Fixes: 226c6e60c35b ("ethdev: add PPPoE to flow API")
Cc: stable@dpdk.org
Signed-off-by: Xiao Zhang <xiao.zhang@intel.com>
Acked-by: Ori Kam <orika@mellanox.com>
4 years agonet/mlx5: reorganize fallback counter management
Suanming Mou [Tue, 7 Apr 2020 03:59:47 +0000 (11:59 +0800)]
net/mlx5: reorganize fallback counter management

Currently, the fallback counter is also allocated from the pool, the
specify fallback function code becomes a bit duplicate.

Reorganize the fallback counter code to make it reuse from the normal
counter code.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/mlx5: split flow counter struct
Suanming Mou [Tue, 7 Apr 2020 03:59:46 +0000 (11:59 +0800)]
net/mlx5: split flow counter struct

Currently, the counter struct saves both the members used by batch
counters and none batch counters. The members which are only used
by none batch counters cost 16 bytes extra memory for batch counters.
As normally there will be limited none batch counters, mix the none
batch counter and batch counter members becomes quite expensive for
batch counter. If 1 million batch counters are created, it means 16 MB
memory which will not be used by the batch counters are allocated.

Split the mlx5_flow_counter struct for batch and none batch counters
helps save the memory.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/mlx5: optimize flow counter handle type
Suanming Mou [Tue, 7 Apr 2020 03:59:45 +0000 (11:59 +0800)]
net/mlx5: optimize flow counter handle type

Currently, DV and verbs counters are both changed to indexed. It means
while creating the flow with counter, flow can save the indexed value to
address the counter.

Save the 4 bytes indexed value in the rte_flow instead of 8 bytes
pointer helps to save memory with millions of flows.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/mlx5: change Direct Verbs counter to indexed
Suanming Mou [Tue, 7 Apr 2020 03:59:44 +0000 (11:59 +0800)]
net/mlx5: change Direct Verbs counter to indexed

This part of the counter optimize change the DV counter to indexed as
what have already done in verbs. In this case, all the mlx5 flow counter
can be addressed by index.

The counter index is composed of pool index and the counter offset in
the pool counter array. The batch and none batch counter dcs ID offset
0x800000 is used to avoid the mix up for the index. As batch counter dcs
ID starts from 0x800000 and none batch counter dcs starts from 0, the
0x800000 offset is added to the batch counter index to indicate the
index of batch counter.

The counter pointer in rte_flow struct will be aligned to index instead
of pointer. It will save 4 bytes memory for every rte_flow. With
millions of rte_flow, it will save MBytes memory.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agocommon/mlx5: add batch counter id offset
Suanming Mou [Tue, 7 Apr 2020 03:59:43 +0000 (11:59 +0800)]
common/mlx5: add batch counter id offset

This commit is a part for the DV counter optimization.

The batch counter dcs id starts from 0x800000 and none batch counter
starts from 0. As currently, the counter is changed to be indexed by
pool index and the offset of the counter in the pool counters_raw array.
It means now the counter index is same for batch and none batch counter.
Add the 0x800000 batch counter offset to the batch counter index helps
indicate the counter index is from batch or none batch container pool.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/mlx5: change verbs counter allocator to indexed
Suanming Mou [Tue, 7 Apr 2020 03:59:42 +0000 (11:59 +0800)]
net/mlx5: change verbs counter allocator to indexed

This is part of the counter optimize which will save the indexed counter
id instead of the counter pointer in the rte_flow.

Place the verbs counter into the container pool helps the counter to be
indexed correctly independent with the raw counter.

The counter pointer in rte_flow will be changed to indexed value after
the DV counter is also changed to indexed.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/mlx5: optimize counter release query generation
Suanming Mou [Tue, 7 Apr 2020 03:59:41 +0000 (11:59 +0800)]
net/mlx5: optimize counter release query generation

Query generation was introduced to avoid counter to be reallocated
before the counter statistics be fully updated. Since the counters
be released between query trigger and query handler may miss the
packets arrived in the trigger and handler gap period. In this case,
user can only allocate the counter while pool query_gen is greater
than the counter query_gen + 1 which indicates a new round of query
finished, the statistic is fully updated.

Split the pool query_gen to start_query_gen and end_query_gen helps
to have a better identify for the counter released in the gap period.
And it helps the counter released before query trigger or after query
handler can be reallocated more efficiently.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/mlx5: fix counter container usage
Suanming Mou [Tue, 7 Apr 2020 03:59:40 +0000 (11:59 +0800)]
net/mlx5: fix counter container usage

As none-batch counter pool allocates only one counter every time, after
the new allocated counter pop out, the pool will be empty and moved to
the end of the container list in the container.

Currently, the new non-batch counter allocation maybe happened with new
counter pool allocated, it means the new counter comes from a new pool.
While new pool is allocated, the container resize and switch happens.
In this case, after the pool becomes empty, it should be added to the
new container pool list as the pool belongs.

Update the container pointer accordingly with pool allocation to avoid
add the pool to the incorrect container.

Fixes: 5382d28c2110 ("net/mlx5: accelerate DV flow counter transactions")
Cc: stable@dpdk.org
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/ena: update driver version to v2.1.0
Michal Krawczyk [Wed, 8 Apr 2020 08:29:21 +0000 (10:29 +0200)]
net/ena: update driver version to v2.1.0

The v2.1.0 is refactoring Tx and Rx paths, including few bug fixes and
is also adding a new features which are going to be available with the
newest hardware.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agodoc: add notes on ENA usage on metal instances
Michal Krawczyk [Wed, 8 Apr 2020 08:29:20 +0000 (10:29 +0200)]
doc: add notes on ENA usage on metal instances

As AWS metal instances are supporting IOMMU, the usage of igb_uio or
vfio-pci can lead to a problems (when to use which module), especially
that the vfio-pci isn't supporting SMMU on arm64.

To clear up the problem of using those modules in various setup
conditions (with or without IOMMU) on metal instances, more detailed
explanation was added.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: reuse zero length Rx descriptor
Michal Krawczyk [Wed, 8 Apr 2020 08:29:19 +0000 (10:29 +0200)]
net/ena: reuse zero length Rx descriptor

Some ENA devices can pass to the driver descriptor with length 0. To
avoid extra allocation, the descriptor can be reused by simply putting
it back to the device.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: refactor Tx
Michal Krawczyk [Wed, 8 Apr 2020 08:29:18 +0000 (10:29 +0200)]
net/ena: refactor Tx

The original Tx function was very long and was containing both cleanup
and the sending sections. Because of that it was having a lot of local
variables, big indentation and was hard to read.

This function was split into 2 sections:
  * Sending - which is responsible for preparing the mbuf, mapping it
    to the device descriptors and finally, sending packet to the HW
  * Cleanup - which is releasing packets sent by the HW. Loop which was
    releasing packets was reworked a bit, to make intention more visible
    and aligned with other parts of the driver.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: use macros for ring index operations
Michal Krawczyk [Wed, 8 Apr 2020 08:29:17 +0000 (10:29 +0200)]
net/ena: use macros for ring index operations

To improve code readability, abstraction was added for operating on IO
rings indexes.

Driver was defining local variable for ring mask in each function that
needed to operate on the ring indexes. Now it is being stored in the
ring as this value won't change unless size of the ring will change and
macros for advancing indexes using the mask has been added.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: limit refill threshold by fixed value
Michal Krawczyk [Wed, 8 Apr 2020 08:29:16 +0000 (10:29 +0200)]
net/ena: limit refill threshold by fixed value

Divider used for both Tx and Rx cleanup/refill threshold can cause too
big delay in case of the really big rings - for example if the 8k Rx
ring will be used, the refill won't trigger unless 1024 threshold will
be reached. It will also cause driver to try to allocate that much
descriptors.

Limiting it by fixed value - 256 in that case, would limit maximum
time spent in repopulate function.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: rework getting number of available descriptors
Michal Krawczyk [Wed, 8 Apr 2020 08:29:15 +0000 (10:29 +0200)]
net/ena: rework getting number of available descriptors

ena_com API should be preferred for getting number of used/available
descriptors unless extra calculation needs to be performed.

Some helper variables were added for storing values that are later
reused. Moreover, for limiting the value of sent/received packets to
the number of available descriptors, the RTE_MIN is used instead of
if function, which was doing similar thing but was less descriptive.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: refactor Rx
Michal Krawczyk [Wed, 8 Apr 2020 08:29:14 +0000 (10:29 +0200)]
net/ena: refactor Rx

* Split main Rx function into multiple ones - the body of the main
  was very big and further there were 2 nested loops, which were
  making the code hard to read
* Rework how the Rx mbuf chains are being created - Instead of having
  while loop which has conditional check if it's first segment, handle
  this segment outside the loop and if more fragments are existing,
  process them inside.
* Initialize Rx mbuf using simple function - it's the common thing for
  the 1st and next segments.
* Create structure for Rx buffer to align it with Tx path, other ENA
  drivers and to make the variable name more descriptive - on DPDK, Rx
  buffer must hold only mbuf, so initially array of mbufs was used as
  the buffers. However, it was misleading, as it was named
  "rx_buffer_info". To make it more clear, the structure holding mbuf
  pointer was added and now there is possibility to expand it in the
  future without reworking the driver.
* Remove redundant variables and conditional checks.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: disable meta caching
Michal Krawczyk [Wed, 8 Apr 2020 08:29:13 +0000 (10:29 +0200)]
net/ena: disable meta caching

In the LLQ (Low-latency queue) mode, the device can indicate that meta
data descriptor caching is disabled. In that case the driver should send
valid meta descriptor on every Tx packet.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: add Tx drops statistic
Michal Krawczyk [Wed, 8 Apr 2020 08:29:12 +0000 (10:29 +0200)]
net/ena: add Tx drops statistic

ENA device can report in the AENQ handler amount of Tx packets that were
dropped and not sent.

This statistic is showing global value for the device and because
rte_eth_stats is missing field that could indicate this value (it
isn't the Tx error), it is being presented as a extended statistic.

As the current design of extended statistics prevents tx_drops from
being an atomic variable and both tx_drops and rx_drops are only updated
from the AENQ handler, both were set as non-atomic for the alignment.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: remove memory barriers before doorbells
Michal Krawczyk [Wed, 8 Apr 2020 08:29:11 +0000 (10:29 +0200)]
net/ena: remove memory barriers before doorbells

The doorbell code is already issuing the doorbell by using rte_write.
Because of that, there is no need to do that before calling the
function.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: support large LLQ headers
Michal Krawczyk [Wed, 8 Apr 2020 08:29:10 +0000 (10:29 +0200)]
net/ena: support large LLQ headers

Default LLQ (Low-latency queue) maximum header size is 96 bytes and can
be too small for some types of packets - like IPv6 packets with multiple
extension. This can be fixed, by using large LLQ headers.

If the device supports larger LLQ headers, the user can activate them by
using device argument 'large_llq_hdr' with value '1'.

If the device isn't supporting this feature, the default value (96B)
will be used.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: refactor getting IO queues capabilities
Michal Krawczyk [Wed, 8 Apr 2020 08:29:09 +0000 (10:29 +0200)]
net/ena: refactor getting IO queues capabilities

Reading values from the device is about the maximum capabilities of the
device. Because of that, the names of the fields storing those values,
functions and temporary variables, should be more descriptive in order
to improve self documentation of the code.

In connection with this, the way of getting maximum queue size could be
simplified - no hardcoded values are needed, as the device is going to
send it's capabilities anyway.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: set IO ring size to valid value
Michal Krawczyk [Wed, 8 Apr 2020 08:29:08 +0000 (10:29 +0200)]
net/ena: set IO ring size to valid value

IO rings were configured with the maximum allowed size for the Tx/Rx
rings. However, the application could decide to create smaller rings.

This patch is using value stored in the ring instead of the value from
the adapter which is indicating the maximum allowed value.

Fixes: df238f84c0a2 ("net/ena: recreate HW IO rings on start and stop")
Cc: stable@dpdk.org
Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: update generation date and commit
Michal Krawczyk [Wed, 8 Apr 2020 08:29:07 +0000 (10:29 +0200)]
net/ena/base: update generation date and commit

The current ena_com version was generated on 25.09.2019.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: fix indentation of multiple defines
Michal Krawczyk [Wed, 8 Apr 2020 08:29:06 +0000 (10:29 +0200)]
net/ena/base: fix indentation of multiple defines

As the alignment of the defines wasn't valid, it was removed at all, so
instead of using multiple spaces or tabs, the single space after define
name is being used.

Fixes: 99ecfbf845b3 ("ena: import communication layer")

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: fix types for printing timestamps
Michal Krawczyk [Wed, 8 Apr 2020 08:29:05 +0000 (10:29 +0200)]
net/ena/base: fix types for printing timestamps

Because ena_com is being used by multiple platforms which are using
different C versions, PRIu64 cannot be used directly and must be defined
in the platform file.

Fixes: b2b02edeb0d6 ("net/ena/base: upgrade HAL for new HW features")

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: use 48-bit memory addresses
Michal Krawczyk [Wed, 8 Apr 2020 08:29:04 +0000 (10:29 +0200)]
net/ena/base: use 48-bit memory addresses

ENA device is using 48-bit memory for IO. Because of that, the upper
limit had to be updated.

From the driver perspective, it's just a cosmetic change to make
definition of the structure 'ena_common_mem_addr' more descriptive and
the address value was verified anyway for the valid range in the
function 'ena_com_mem_addr_set()'.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: add error logs when preparing Tx
Michal Krawczyk [Wed, 8 Apr 2020 08:29:03 +0000 (10:29 +0200)]
net/ena/base: add error logs when preparing Tx

To make the debugging easier, the error logs were added in the Tx path.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: fix indentation in CQ polling
Michal Krawczyk [Wed, 8 Apr 2020 08:29:02 +0000 (10:29 +0200)]
net/ena/base: fix indentation in CQ polling

The spaces instead of tabs were used for the indent.

Fixes: 3adcba9a8987 ("net/ena: update HAL to the newer version")

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: fix documentation of functions
Michal Krawczyk [Wed, 8 Apr 2020 08:29:01 +0000 (10:29 +0200)]
net/ena/base: fix documentation of functions

The documentation format was aligned and few typos were fixed.

Fixes: 99ecfbf845b3 ("ena: import communication layer")

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: add accelerated LLQ mode
Michal Krawczyk [Wed, 8 Apr 2020 08:29:00 +0000 (10:29 +0200)]
net/ena/base: add accelerated LLQ mode

In order to use the accelerated LLQ (Low-lateny queue) mode, the driver
must limit the Tx burst and be aware that the device has the meta
caching disabled. In that situation, the meta descriptor must be valid
on each Tx packet.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: remove extra properties strings
Michal Krawczyk [Wed, 8 Apr 2020 08:28:59 +0000 (10:28 +0200)]
net/ena/base: remove extra properties strings

This buffer was never used by the ENA PMD. It could be used for
debugging, but it's presence is redundant now.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: rework interrupt moderation
Michal Krawczyk [Wed, 8 Apr 2020 08:28:58 +0000 (10:28 +0200)]
net/ena/base: rework interrupt moderation

This feature allows for adaptive interrupt moderation. It's not used by
the DPDK PMD, but is a part of the newest HAL version.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: remove conversion of indirection table
Michal Krawczyk [Wed, 8 Apr 2020 08:28:57 +0000 (10:28 +0200)]
net/ena/base: remove conversion of indirection table

After the indirection table is being saved in the device, there is no
need to convert it back, as it's already saved in host_rss_ind_tbl
array.

As a result, the call to the ena_com_ind_tbl_convert_from_device() is
not needed.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
4 years agonet/ena/base: fix testing for supported hash function
Michal Krawczyk [Wed, 8 Apr 2020 08:28:56 +0000 (10:28 +0200)]
net/ena/base: fix testing for supported hash function

There was a bug in ena_com_fill_hash_function(), which was causing bit to
be shifted left one bit too much.

To fix that, the ENA_FFS macro is being used (returning the location of
the first bit set), hash_function value is being subtracted by 1 if any
hash function is supported by the device and BIT macro is used for
shifting for better verbosity.

Fixes: 99ecfbf845b3 ("ena: import communication layer")
Cc: stable@dpdk.org
Signed-off-by: Michal Krawczyk <mk@semihalf.com>
4 years agonet/ena/base: generate default random RSS hash key
Michal Krawczyk [Wed, 8 Apr 2020 08:28:55 +0000 (10:28 +0200)]
net/ena/base: generate default random RSS hash key

Although the RSS key still cannot be set, it is now being generated
every time the driver is being initialized.

Multiple devices can still have the same key if they're used by the same
driver.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: prevent allocation of zero sized memory
Igor Chauskin [Wed, 8 Apr 2020 08:28:54 +0000 (10:28 +0200)]
net/ena/base: prevent allocation of zero sized memory

rte_memzone_reserve() will reserve the biggest contiguous memzone
available if received 0 as size param.

Fixes: 9ba7981ec992 ("ena: add communication layer for DPDK")
Cc: stable@dpdk.org
Signed-off-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena/base: make allocation macros thread-safe
Igor Chauskin [Wed, 8 Apr 2020 08:28:53 +0000 (10:28 +0200)]
net/ena/base: make allocation macros thread-safe

Memory allocation region id could possibly be non-unique
due to non-atomic increment, causing allocation failure.

Fixes: 9ba7981ec992 ("ena: add communication layer for DPDK")
Cc: stable@dpdk.org
Signed-off-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agonet/ena: ensure Rx buffer size is at least 1400B
Michal Krawczyk [Wed, 8 Apr 2020 08:28:52 +0000 (10:28 +0200)]
net/ena: ensure Rx buffer size is at least 1400B

Some of the ENA devices can't handle buffers which are smaller than a
1400B. Because of this limitation, size of the buffer is being checked
and limited during the Rx queue setup.

If it's below the allowed value, PMD won't finish it's configuration
successfully..

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
4 years agodoc: update ice guide driver/firmware/package version
Zhaoyan Chen [Fri, 20 Mar 2020 01:58:17 +0000 (09:58 +0800)]
doc: update ice guide driver/firmware/package version

Update driver/firmware/package version for DPDK20.02

Signed-off-by: Zhaoyan Chen <zhaoyan.chen@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/ice/base: remove unused code in switch rule
Qi Zhang [Tue, 7 Apr 2020 00:25:23 +0000 (08:25 +0800)]
net/ice/base: remove unused code in switch rule

Update a switch rule' action from "to VSI" to "to VSI List"
should only happen when the same rule has been programmed with
a different fwd destination. This is already handled by below
code block:

m_entry = ice_find_adv_rule_entry(...)
if (m_entry) {
...
ice_adv_add_update_vsi_list(...)
}

The following ice_update_pkt_fwd_rule is unnecessary and should be
removed due to:
1) If a switch rule's action is still to VSI, which means, it is
   the first time be issued,  we don't need to update it "to VSI
   List."
2) Actually the implementation does not match the comment, it still
   update the rule with "to VSI" action.

Fixes: fed0c5ca5f19 ("net/ice/base: support programming a new switch recipe")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/iavf: enable port reset
Lunyuan Cui [Wed, 1 Apr 2020 07:12:14 +0000 (07:12 +0000)]
net/iavf: enable port reset

This patch is intended to add iavf_dev_reset ops, enable iavf to support
"port reset all".

Signed-off-by: Lunyuan Cui <lunyuanx.cui@intel.com>
Tested-by: Zhaoyan Chen <zhaoyan.chen@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/i40e: enable MAC address as flow director input set
Lunyuan Cui [Thu, 2 Apr 2020 07:58:47 +0000 (07:58 +0000)]
net/i40e: enable MAC address as flow director input set

Enable source MAC address and destination MAC address as FDIR's
input set for ipv4-other, ipv4-udp and ipv4-tcp. When OVS-DPDK is
working as a pure L2 switch, enable MAC address as FDIR input set
with Mark+RSS action would help the performance speed up. And FVL
FDIR supports to change input set with MAC address.

Signed-off-by: Lunyuan Cui <lunyuanx.cui@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
4 years agonet/ice: fix RSS advanced rule
Junyu Jiang [Tue, 7 Apr 2020 01:48:17 +0000 (01:48 +0000)]
net/ice: fix RSS advanced rule

This patch moved the RSS initialization from dev start to dev configure
to fix RSS advanced rule invalid issue after running port stop and port
start.

Fixes: 5ad3db8d4bdd ("net/ice: enable advanced RSS")
Cc: stable@dpdk.org
Signed-off-by: Junyu Jiang <junyux.jiang@intel.com>
Tested-by: Zhiwei He <zhiwei.he@intel.com>
Acked-by: Qiming Yang <qiming.yang@intel.com>
4 years agonet/nfp: fix dangling pointer on probe failure
Yunjian Wang [Tue, 7 Apr 2020 11:37:27 +0000 (19:37 +0800)]
net/nfp: fix dangling pointer on probe failure

When nfp_pf_create_dev() is cleaning up, it does not correctly set
the dev_private variable to NULL, which will lead to a double free.

Fixes: ef28aa96e53b ("net/nfp: support multiprocess")
Cc: stable@dpdk.org
Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
Acked-by: Heinrich Kuhn <heinrich.kuhn@netronome.com>
4 years agonet/nfp: fix log format specifiers
Ferruh Yigit [Tue, 3 Mar 2020 13:36:46 +0000 (13:36 +0000)]
net/nfp: fix log format specifiers

Changing format specifier for the 'size_t' as '%z' and for 'off_t' as
'%jd'.

Also this fix enables compiling PMD for 32bit architecture.

Fixes: 29a62d1476b6 ("net/nfp: add CPP bridge as service")
Cc: stable@dpdk.org
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Heinrich Kuhn <heinrich.kuhn@netronome.com>
4 years agoexamples/vmdq: fix RSS configuration
Junyu Jiang [Wed, 25 Mar 2020 06:32:48 +0000 (06:32 +0000)]
examples/vmdq: fix RSS configuration

In order that all queues of pools can receive packets,
add enable-rss argument to change RSS configuration.

Fixes: 6bb97df521aa ("examples/vmdq: new app")
Cc: stable@dpdk.org
Signed-off-by: Junyu Jiang <junyux.jiang@intel.com>
Acked-by: Xiaoyun Li <xiaoyun.li@intel.com>
Tested-by: Yingya Han <yingyax.han@intel.com>
4 years agodoc: add user guide for VMDq example
Junyu Jiang [Wed, 25 Mar 2020 06:32:47 +0000 (06:32 +0000)]
doc: add user guide for VMDq example

Currently, there is no documentation for VMDq example,
this path added the user guide for VMDq.

Signed-off-by: Junyu Jiang <junyux.jiang@intel.com>
Reviewed-by: Jingjing Wu <jingjing.wu@intel.com>
4 years agoethdev: fix build when vtune profiling is on
Eugeny Parshutin [Mon, 6 Apr 2020 18:27:05 +0000 (21:27 +0300)]
ethdev: fix build when vtune profiling is on

Add the previous prototype for the 'profile_hook_rx_burst_cb' function
to fix the compiler warning when the DPDK is built with
'RTE_ETHDEV_PROFILE_WITH_VTUNE' config option enabled:
/home/dpdk/lib/librte_ethdev/ethdev_profile.c:17:1: warning:
no previous prototype for profile_hook_rx_burst_cb [-Wmissing-prototypes]

Fixes: 2c1bbab7f09d ("ethdev: change vtune profiling approach")
Cc: stable@dpdk.org
Signed-off-by: Eugeny Parshutin <eugeny.parshutin@linux.intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/memif: fix resource leak
Vadim Podovinnikov [Sat, 4 Apr 2020 10:15:57 +0000 (13:15 +0300)]
net/memif: fix resource leak

Fixes: c41a04958b09 ("net/memif: support multi-process")
Cc: stable@dpdk.org
Signed-off-by: Vadim Podovinnikov <podovinnikov@protei.ru>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agoapp/testpmd: support dumping socket memory
Xueming Li [Sun, 5 Apr 2020 02:49:22 +0000 (02:49 +0000)]
app/testpmd: support dumping socket memory

Introduce new command to dump memory statistics of each socket,
summary, also show changes since last call.

Usage:
    dump_socket_mem

Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/netvsc: avoid possible live lock
Stephen Hemminger [Tue, 31 Mar 2020 17:14:04 +0000 (10:14 -0700)]
net/netvsc: avoid possible live lock

Since the ring buffer with host is shared for both transmit
completions and receive packets, it is possible that transmitter
could get starved if receive ring gets full.

Better to process all outstanding events which frees up transmit
buffer slots, even if means dropping some packets.

Fixes: 7e6c82430702 ("net/netvsc: avoid over filling Rx descriptor ring")
Cc: stable@dpdk.org
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
4 years agobus/vmbus: simplify arguments to need signal function
Stephen Hemminger [Tue, 31 Mar 2020 17:14:03 +0000 (10:14 -0700)]
bus/vmbus: simplify arguments to need signal function

The transmit need signal function can avoid an unnecessary
dereference by passing the right pointer. This also makes
code better match FreeBSD driver.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
4 years agonet/netvsc: handle Tx completions based on burst size
Stephen Hemminger [Tue, 31 Mar 2020 17:14:02 +0000 (10:14 -0700)]
net/netvsc: handle Tx completions based on burst size

If tx_free_thresh is quite low, it is possible that we need to
cleanup based on burst size.

Fixes: fc30efe3a22e ("net/netvsc: change Rx descriptor setup and sizing")
Cc: stable@dpdk.org
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
4 years agonet/netvsc: remove process event optimization
Stephen Hemminger [Tue, 31 Mar 2020 17:14:01 +0000 (10:14 -0700)]
net/netvsc: remove process event optimization

Remove unlocked check for data in receive ring.
This check is not safe because of missing barriers etc.

Fixes: 4e9c73e96e83 ("net/netvsc: add Hyper-V network device")
Cc: stable@dpdk.org
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
4 years agonet/netvsc: fix memory free on device close
Stephen Hemminger [Tue, 31 Mar 2020 17:14:00 +0000 (10:14 -0700)]
net/netvsc: fix memory free on device close

The netvsc PMD was putting the mac address in private data but the
core rte_ethdev doesn't allow that it. It has to be in rte_malloc'd
memory or a message will be printed on shutdown/close.
 EAL: Invalid memory

Fixes: f8279f47dd89 ("net/netvsc: fix crash in secondary process")
Cc: stable@dpdk.org
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
4 years agonet/netvsc: split send buffers from Tx descriptors
Stephen Hemminger [Tue, 31 Mar 2020 17:13:59 +0000 (10:13 -0700)]
net/netvsc: split send buffers from Tx descriptors

The VMBus has reserved transmit area (per device) and transmit
descriptors (per queue). The previous code was always having a 1:1
mapping between send buffers and descriptors.
This can lead to one queue starving another and also buffer bloat.

Change to working more like FreeBSD where there is a pool of transmit
descriptors per queue. If send buffer is not available then no
aggregation happens but the queue can still drain.

Fixes: 4e9c73e96e83 ("net/netvsc: add Hyper-V network device")
Cc: stable@dpdk.org
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
4 years agonet/netvsc: handle Rx packets during multi-channel setup
Stephen Hemminger [Tue, 31 Mar 2020 17:13:58 +0000 (10:13 -0700)]
net/netvsc: handle Rx packets during multi-channel setup

It is possible for a packet to arrive during the configuration
process when setting up multiple queue mode. This would cause
configure to fail; fix by just ignoring receive packets while
waiting for control commands.

Use the receive ring lock to avoid possible races between
oddly behaved applications doing rx_burst and control operations
concurrently.

Fixes: 4e9c73e96e83 ("net/netvsc: add Hyper-V network device")
Cc: stable@dpdk.org
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
4 years agonet/netvsc: propagate descriptor limits from VF
Stephen Hemminger [Tue, 31 Mar 2020 17:13:57 +0000 (10:13 -0700)]
net/netvsc: propagate descriptor limits from VF

If application cares about descriptor limits, the netvsc device
values should reflect those of the VF as well.

Fixes: dc7680e8597c ("net/netvsc: support integrated VF")
Cc: stable@dpdk.org
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
4 years agoapp/testpmd: enhance GTP support
Dekel Peled [Wed, 25 Mar 2020 08:12:31 +0000 (10:12 +0200)]
app/testpmd: enhance GTP support

This patch adds CLI option to enter the v_pt_rsv_flags value for GTP
flow pattern item.
It also adds GTP as valid item in raw_encap and raw_decap setting.

Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Ori Kam <orika@mellanox.com>
4 years agonet/ice: redirect switch rule to new VSI
Beilei Xing [Fri, 3 Apr 2020 04:46:09 +0000 (12:46 +0800)]
net/ice: redirect switch rule to new VSI

After VF reset, VF's VSI number may be changed,
the switch rule which forwards packet to the old
VSI number should be redirected to the new VSI
number.

Signed-off-by: Beilei Xing <beilei.xing@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: enable flow redirect on switch
Beilei Xing [Fri, 3 Apr 2020 04:46:08 +0000 (12:46 +0800)]
net/ice: enable flow redirect on switch

Enable flow redirect on switch, currently only
support VSI redirect.

Signed-off-by: Beilei Xing <beilei.xing@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: fix input set of VLAN item
Wei Zhao [Fri, 3 Apr 2020 04:46:07 +0000 (12:46 +0800)]
net/ice: fix input set of VLAN item

The input set for inner type of vlan item should
be ICE_INSET_ETHERTYPE, not ICE_INSET_VLAN_OUTER.
This mac vlan filter is also part of DCF switch filter.

Fixes: 47d460d63233 ("net/ice: rework switch filter")
Cc: stable@dpdk.org
Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: add more flow support for permission stage
Wei Zhao [Fri, 3 Apr 2020 04:46:06 +0000 (12:46 +0800)]
net/ice: add more flow support for permission stage

This patch add switch filter permission stage support
for more flow pattern in pf only pipeline mode.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: support IPv6 NAT-T
Wei Zhao [Fri, 3 Apr 2020 04:46:05 +0000 (12:46 +0800)]
net/ice: support IPv6 NAT-T

This patch add switch filter support for IPv6 NAT-T packets,
it enable switch filter to direct IPv6 packets with
NAT-T payload to specific action.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: support PFCP
Wei Zhao [Fri, 3 Apr 2020 04:46:04 +0000 (12:46 +0800)]
net/ice: support PFCP

This patch add switch filter support for PFCP packets,
it enable switch filter to direct IPv4/IPv6 packets with
PFCP session or node payload to specific action.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: support ESP/AH/L2TP
Wei Zhao [Fri, 3 Apr 2020 04:46:03 +0000 (12:46 +0800)]
net/ice: support ESP/AH/L2TP

This patch add support for ESP/AH/L2TP packets,
it enable switch filter to direct IPv6 packets with
ESP/AH/L2TP payload to specific action.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: add action number check for switch
Wei Zhao [Fri, 3 Apr 2020 04:46:02 +0000 (12:46 +0800)]
net/ice: add action number check for switch

The action number can only be one for DCF or PF
switch filter, not support multiple actions.

Fixes: 47d460d63233 ("net/ice: rework switch filter")
Cc: stable@dpdk.org
Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: change default tunnel type
Wei Zhao [Fri, 3 Apr 2020 04:46:01 +0000 (12:46 +0800)]
net/ice: change default tunnel type

The default tunnel type for switch filter change to new
definition of ICE_SW_TUN_AND_NON_TUN in order that the rule
will be apply to more packet type.

Fixes: 47d460d63233 ("net/ice: rework switch filter")
Cc: stable@dpdk.org
Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: support MAC VLAN rule
Wei Zhao [Fri, 3 Apr 2020 04:46:00 +0000 (12:46 +0800)]
net/ice: support MAC VLAN rule

This patch add support for MAC VLAN rule,
it enable switch filter to direct packet base on
mac address and VLAN id.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: change switch parser to support flexible mask
Wei Zhao [Fri, 3 Apr 2020 04:45:59 +0000 (12:45 +0800)]
net/ice: change switch parser to support flexible mask

DCF need to make configuration of flexible mask, that is to say
some input set mask may be not 0xFFFF type all one bit. In order
to direct L2/IP multicast packets, the mask for source IP maybe
0xF0000000, this patch enable switch filter parser for it.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: support more PPPoE input set
Wei Zhao [Fri, 3 Apr 2020 04:45:58 +0000 (12:45 +0800)]
net/ice: support more PPPoE input set

This patch add more support for PPPoE packet,
it enable switch filter to direct PPPoE packet base on
session id and PPP protocol type.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/ice: enable switch flow on DCF
Wei Zhao [Fri, 3 Apr 2020 04:45:57 +0000 (12:45 +0800)]
net/ice: enable switch flow on DCF

DCF on CVL is a control plane VF which take the responsibility to
configure all the PF/global resources, this patch add support DCF
on to program forward rule to direct packets to VFs.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
4 years agonet/octeontx2: support custom L2 header
Satheesh Paul [Mon, 6 Apr 2020 06:43:03 +0000 (12:13 +0530)]
net/octeontx2: support custom L2 header

This patch adds SDP packet parsing support with custom L2 header,
adds support to include a field from custom header for flow tag
generation.

Signed-off-by: Satheesh Paul <psatheesh@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>