When set the GRE item, GRE key should follow after GRE header, or the
header gre_item pointer used by the key will be invalid.
Currently in the mlx5_flow_validate_item_gre_key() function, the header
gre_item pointer is access before checking if the key is after the
header or not. Once the key item is before the header, invalid gre_item
pointer access happens.
Move the gre_item pointer access after the GRE header check to avoid the
crash issue.
Fixes: a7a0365565a4 ("net/mlx5: match GRE key and present bits")
Cc: stable@dpdk.org
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Ori Kam <orika@mellanox.com>
GENEVE is available in tunnel offloads. Add it as the default support
option.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Ori Kam <orika@mellanox.com>
The header mlx5_glue.h includes rte_byteorder.h, which isn't
pulled by global_inc. This patch adds the missing include directory
and fixes the build error:
In file included from drivers/net/mlx5/mlx5_glue.c:31:
drivers/net/mlx5/mlx5_glue.h:12:10: fatal error:
rte_byteorder.h: No such file or directory
The build failure can be triggered with:
meson -Dibverbs_link=dlopen build && ninja -C build
Fixes: 400d985eb586 ("net/mlx5: add VLAN push/pop DR commands to glue")
Signed-off-by: Ali Alnubani <alialnu@mellanox.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
ConnectX-4LX supports multiple packets within the single Tx
descriptor. This feature is named as "Legacy Multi-Packet Write"
and imposes a lot of limitations:
- no ACLs, it means no NIC Tx Flows are supported and Tx metadata
become meaningless
- the required minimal inline data must be zero
- no SR-IOV, it means no support in E-Switch configurations,
- no priority and dscp forcing
- no VLAN insertion
- no TSO
- all packets within MPW session must have the same size
This legacy MPW feature is mainly intended for test purposes.
To explicitly engage the feature on ConnectX-4LX the devargs
should be specified:
- txq_mpw_en=1
This feature was dropped in 19.08, this patch reverts it back.
Fixes: 18a1c20044c0 ("net/mlx5: implement Tx burst template")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Commit in fixes line sets the DV (Direct Verbs) flow engine as default.
Newer versions of DV flow engine use the DR (Direct Rules) features.
DR is supported from RDMA Core library version rdma-core-24.0.
This cause failure to start port when using older rdma-core version,
without DR support.
This patch selects DV flow engine if rdma-core version is v24.0 or
higher. Verbs flow engine is selected otherwise.
Fixes: cd4569d2bf3c ("net/mlx5: change default flow engine to DV")
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Ori Kam <orika@mellanox.com>
The metadata register used for META item/action within E-Switch
steering domain should be altered depending on dv_xmeta_en
devarg value.
Fixes: 3e8edd0ef848 ("net/mlx5: update metadata register ID query")
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
For tunnel mode, there may be two L3/L4 layer match pattern items, one
for inner and one for outer layers. The L3 TTL and L4 port header
modify actions should handle the outermost layer items.
Currently flow_dv_attr_init() doesn't distinguish between outer and
inner layers, since inner layer comes later than the outer layer, this
may cause inner layer values also set to the flow attributes and may
lead actions to use inner L3/L4 pattern items.
Adding a check in flow_dv_attr_init() to prevent inner layer to set the
flow attribute if the previous L3/L4 outer pattern exist.
Fixes: 4bb14c83df95 ("net/mlx5: support modify header using Direct Verbs")
Cc: stable@dpdk.org
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
The current raw_encap function selects if to decap or encap
based on the egress/ingress attribute. This concepts doesn't work
in case of FDB since all flows are considered ingress.
To solve this issue we moved to check the encap size.
if the encap size is larger then eth + ipv4 it means we are
trying to encap.
Fixes: 8ba9eee4ce32 ("net/mlx5: add raw data encap/decap to Direct Verbs")
Cc: stable@dpdk.org
Signed-off-by: Ori Kam <orika@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
When DR is not supported and DV is supported, tag action still can be
used by the metadata feature.
Wrongly, the tag hash list was not created what caused failure in
metadata action creation.
Create the tag hash list for each DV case.
Fixes: 860897d2895a ("net/mlx5: reorganize flow tables with hash list")
Signed-off-by: Matan Azrad <matan@mellanox.com>
Merge all versions in linker version script files to DPDK_20.0.
This commit was generated by running the following command:
:~/DPDK$ buildtools/update-abi.sh 20.0
Signed-off-by: Pawel Modrak <pawelx.modrak@intel.com>
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Since the library versioning for both stable and experimental ABI's is
now managed globally, the LIBABIVER and version variables no longer
serve any useful purpose, and can be removed.
The replacement in Makefiles was done using the following regex:
^(#.*\n)?LIBABIVER\s*:=\s*\d+\n(\s*\n)?
(LIBABIVER := numbers, optionally preceded by a comment and optionally
succeeded by an empty line)
The replacement for meson files was done using the following regex:
^(#.*\n)?version\s*=\s*\d+\n(\s*\n)?
(version = numbers, optionally preceded by a comment and optionally
succeeded by an empty line)
[David]: those variables are manually removed for the files:
- drivers/common/qat/Makefile
- lib/librte_eal/meson.build
[David]: the LIBABIVER is restored for the external ethtool example
library.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Signed-off-by: David Marchand <david.marchand@redhat.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
In debug mode, there is assertion to validate the CQ object before the
release.
Wrongly, the assertion is done for any type of RX queue even if it
doesn't use CQ at all, for example in hairpin Rx queue.
Ignore CQ assertion when hairpin queue is released.
Fixes: e79c9be91515 ("net/mlx5: support Rx hairpin queues")
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Ori Kam <orika@mellanox.com>
In order to configure L3 encapsulation\decapsulation flow to mlx5
devices, 2 actions should be added to the flow actions list:
RTE_FLOW_ACTION_TYPE_RAW_DECAP and RTE_FLOW_ACTION_TYPE_RAW_ENCAP.
One of the validations for this scenario is to check that modify actions
is not done before the L3 decapsulation, because it doesn't make sense
to decapsulate a modified data.
Wrongly, this check was done for the case of L3 encapsulation what
causes a validation failure in modify + L3 encapsulation flow.
Ignore this check in case of L3 encapsulation.
Fixes: 4bb14c83df95 ("net/mlx5: support modify header using Direct Verbs")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@mellanox.com>
When creating a flow, the flow is checked if it should be split into
2 flows based on the queue/rss acton.
If the RSS action with given without any queues, it will result in crash
due to the fact that the function checks the queue type.
This commit fixes this issue by checking if the rss action is not empty,
and at least one queue.
Fixes: d85c7b5ea59f ("net/mlx5: split hairpin flows")
Signed-off-by: Ori Kam <orika@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Assert condition is fixed to not alert for the case
when multi-packet write is not supported/engaged at all.
Fixes: b53cd86965a1 ("net/mlx5: adjust inline setting for large Tx queue sizes")
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
To support LAG configurations the mlx5dv_dr_action_create_dest_ib_port()
should be called instead of mlx5dv_dr_action_create_dest_vport().
Fixes: f07341e7aed3 ("net/mlx5: update source and destination vport translations")
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Ori Kam <orika@mellanox.com>
As the result of testing it was found that some hosts have
the performance penalty imposed by required write memory barrier
after doorbell writing. Before 19.08 release there was some
heuristics to decide whether write memory barrier should be
performed. For the bursts of recommended size (or multiple)
it was supposed there were some extra ongoing packets in the
next burst and write memory barrier may be skipped (supposed
to be performed in the next burst, at least after descriptor
writing).
This patch restores that behaviour, the devargs tx_db_nc=2
must be specified to engage this performance tuning feature.
Fixes: 8409a28573d3 ("net/mlx5: control transmit doorbell register mapping")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
The default flow engine is Verbs flow engine, for legacy reasons.
This patch changes the default to DV flow engine (dv_flow_en = 1).
Documentation is updated accordingly.
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
The rdma_core routine mlx5dv_dr_create_flow_action_dest_vport()
requires the vport id parameter to create port action.
The register c[0] value was used to deduce the port id value
and it fails in bonding configuration. The correct way is
to apply vport_num value queried from the rdma_core library.
Fixes: f07341e7aed3 ("net/mlx5: update source and destination vport translations")
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
For the case when DR is not supported and DV is supported:
multi-tables feature is off.
In this case, only table 0 is supported.
Table 0 structure wrongly was not created what prevented any
matcher object to be created and even caused crashes.
Create the table hash list in DV case too.
Create table zero empty structure for each domain when DR is not
supported.
Allow NULL DR internal table object to be used.
Fixes: 860897d2895a ("net/mlx5: reorganize flow tables with hash list")
Signed-off-by: Matan Azrad <matan@mellanox.com>
By default RSS hash delivery (offload) is bound to RSS mode and
it is incorrect to advertise it as enabled if Rx multi-queue mode
has no RSS.
Fixes: 8b945a7f7dcb ("drivers/net: update Rx RSS hash offload capabilities")
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
The kernel driver (starting from OFED 4.7.3.1.3) uses the upper half
of metadata register C0 to provide VHCA and vport id's. The mlx5 PMD
should check the actual mask of register C0 and shift the vport id
value for port actions appropriately.
Fixes: f07341e7aed3 ("net/mlx5: update source and destination vport translations")
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
The state of environment variable MLX5_BF_SHUT_UP was not
recovered correctly if there was no tx_db_nc devarg specified.
Fixes: 8409a28573d3 ("net/mlx5: control transmit doorbell register mapping")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
The layer bits should be unique otherwise layer info will be
interpreted wrongly.
Fixes: 70d84dc797b7 ("net/mlx5: add internal tag item and action")
Fixes: 55deee1715f0 ("net/mlx5: extend flow mark support")
Signed-off-by: Xiaoyu Min <jackmin@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Function flow_verbs_counter_query() was recently modified.
The new 'if' condition uses a pointer to flow counter-set.
This pointer is valid only if flow contains a count action.
This patch adds check to verify the pointer is valid.
Fixes: d85c7b5ea59f ("net/mlx5: split hairpin flows")
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Function mlx5_rxq_get_type() uses the input queue index, without
checking it, as index to the Rx queues array.
If this value is too high, it will result in pointer to memory out
of Rx queues array bounds.
This patch adds check of the input queue index, to verify it is valid.
Fixes: d85c7b5ea59f ("net/mlx5: split hairpin flows")
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
The variable cleanup was erroneously done before pointer usage.
The cleanup is moved to appropriate place, comment is clarified.
Fixes: 8d72fa668964 ("net/mlx5: share tag between meter and metadata")
Reported-by: Eli Britstein <elibr@mellanox.com>
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
This reverts commit 304ffe576f239e5405228c0feec04b6138d525b7.
It tried to optimize the amount of the default flow
rules and created it only once on top of the PF representor.
For each FDB rule, the default port ID to match on is like of the port
that triggers the flow.
Hence, the single default rule will not be matched on the VF traffic.
As a result, all the traffic of the VFs will not match to the jump flow
in the root table and will always be forwarded to the representor port.
Revert the commit.
Fixes: 304ffe576f23 ("net/mlx5: fix condition to create default rule")
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
RSS action validation function checks the queues included in RSS
to make sure they are valid.
A Queue is considered valid if the pointer to the queue (item at
location queue-index of RxQ array) is not a null value.
The queue indices are not checked. If a large value is entered as
queue index, using it as an index in RxQ array will result in a
pointer to memory out of array bounds. If this memory contains a
value which is not null, this queue will be wrongly considered valid.
This patch updates function mlx5_flow_validate_action_rss() with
check of the input queue indices, as done in function
mlx5_flow_validate_action_queue().
Fixes: 23c1d42c7138 ("net/mlx5: split flow validation to dedicated function")
Cc: stable@dpdk.org
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Host adapter attributes are queried from kernel via multiple
DevX calls. The retrieved data were erroneously overwritten,
the order of querying is fixed.
Fixes: 6bc327b94fe8 ("net/mlx5: fill meter capabilities using DevX")
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
This patch implements use of the API for LRO aggregated packet
max size.
Rx queue create is updated to use the relevant configuration.
Documentation is updated accordingly.
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Allow to configure the default MAC address of a VF
via its representor port in the host.
An API was proposed to specify explicitly the VF as a
target: https://patches.dpdk.org/patch/62176/
It has been rejected by the technical board in order to
keep compatibility with behavior in Intel PMDs.
http://mails.dpdk.org/archives/dev/2019-November/150588.html
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Signed-off-by: Raslan Darawsheh <rasland@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Add DEV_RX_OFFLOAD_RSS_HASH flag for all PMDs that support RSS hash
delivery.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Tag action for flow mark/flag could be reused by different flows.
When creating a new flow with mark, the existing tag resources will
be traversed in order to confirm if the action is already created.
If only one linked list is used, the searching rate will drop
significantly with the number of tag actions increasing.
By using a hash lists table, it will speed up the searching process
and in the meanwhile, the memory consumption won't be large if only
a small number tag action resources are created(compared to other
hash table implementations). The list heads array size could be
optimized with some extendable hash table in the future.
Signed-off-by: Bing Zhao <bingz@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Matchers are created on the specific table. If a single linked list
is used to store these, then the finding process might be the
bottleneck when there are a lot of different flow matchers on a
huge amount of tables. The matchers could be move into the table
data resource structure in order to reduce the comparison times
when finding.
Signed-off-by: Bing Zhao <bingz@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Jump object is associated with table object, so there is no need to
use a single linked list to store it. All the jump objects could be
put together with related flow tables.
Signed-off-by: Bing Zhao <bingz@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
In the current flow tables organization, arrays are used. This is
fast for searching, creating related object that will be used in
flow creation. But it introduces some limitation to the table index.
Then we can reorganize the flow tables information with hash list.
When using hash list, there is no need to maintain three arrays for
NIC TX, RX and FDB tables object information.
This attribute could be used together with the table ID to generate
a 64-bits key that is unique for the hash list insertion, lookup and
deletion.
Signed-off-by: Bing Zhao <bingz@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
The rdma core library can map doorbell register in two ways,
depending on the environment variable "MLX5_SHUT_UP_BF":
- as regular cached memory, the variable is either missing or
set to zero. This type of mapping may cause the significant
doorbell register writing latency and requires explicit
memory write barrier to mitigate this issue and prevent
write combining.
- as non-cached memory, the variable is present and set to
not "0" value. This type of mapping may cause performance
impact under heavy loading conditions but the explicit write
memory barrier is not required and it may improve core
performance.
The new devarg is introduced "tx_db_nc", if this parameter is
set to zero, the doorbell register is forced to be mapped to
cached memory and requires explicit memory barrier after
writing to. If "tx_db_nc" is set to non-zero value the doorbell
will be mapped as non-cached memory, not requiring the memory
barrier. If "tx_db_nc" is missing the behaviour will be defined
by presence of "MLX5_SHUT_UP_BF" in environment. If variable
is missed the default value zero will be set for ARM64 hosts
and one for others.
In run time the code checks the mapping type and provides the
memory barrier after writing to tx doorbell register if it is
needed. The mapping type is extracted directly from the
uar_mmap_offset field in the queue properties.
Fixes: 18a1c20044c0 ("net/mlx5: implement Tx burst template")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
To support multi-segment packet sending the Tx datapath
DEV_TX_OFFLOAD_MULTI_SEGS offload must be configured.
In debug configuration the asserts arise if the multi-segment
packet is being sent and no DEV_TX_OFFLOAD_MULTI_SEGS is set.
It was found that in some execution paths asserts were missed,
this patch adds ones.
Fixes: 18a1c20044c0 ("net/mlx5: implement Tx burst template")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
When the port is closed or program exits ungraceful, the meter rulers
should be flushed after the flow destroyed.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
In the meter flow split, metadata flow will be as the sub flow of meter
suffix flow. In meter suffix flow, there is already a unique id tag
exist as for the meter prefix and suffix flow match.
Make metadata feature and meter both share the unique id tag for match.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
The flow with meter action will be divided to three sub flows, the
prefix flow, the meter flow and the suffix flow.
For these three sub flows, as the prefix flow and meter flow have the
meter color match. In order the make the packet from prefix flow to
match with the suffix flow, an extra metadata register is allocated.
The prefix flow will add a unique id to the register and the suffix
flow matches on that unique id.
As RSS will also divided the flow to several sub flows, flow with meter
will be divided as the sub flow of the RSS sub flows if have as below:
Original flow ->
RSS sub flow 1 ->
Meter sub flow 1 (Contain three sub flows.)
RSS sub flow 2 ->
Meter sub flow 2 (Contain three sub flows.)
......
RSS sub flow n ->
Meter sub flow n (Contain three sub flows.)
The metadata feature flow will be split as the sub flow of the meter
suffix flow.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Add meter flow action support in flow validate and translate.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Add the meter attach and detach for the flow create.
When create the flow with meter, first try to find any created meter
action matches the flow meter id. If the meter action is already
created, just attach to it and increase the ref_cnt. If not, create
one.
For the dettach, decrease the ref_cnt, destroy the meter action while
the ref_cnt decreased to zero.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
This commit add the meter statistics read and update to check the meter
statistics.
New internal functions in rte_mtr_ops callback:
1. stats_update()
2. stats_read()
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Add count action to meter for metering packet statistics. All the
packets be colored and dropped will be recorded.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Expose the flow counter management mechanism for other components to
use.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
This commit add the meter profile update support.
New internal function in rte_mtr_ops callback:
1. meter_profile_update()
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
This commit add meter enable and disable supoort.
New internal functions in rte_mtr_ops callback:
1. meter_enable()
2. meter_disable()
The meter_enable() enables the meter action and the meter_disable()
disables the meter action.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
This commit add the meter action creation to the glue code.
Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
