dpdk.git
4 years agokvargs: fix buffer overflow when parsing list
Yunjian Wang [Fri, 27 Mar 2020 08:09:55 +0000 (09:09 +0100)]
kvargs: fix buffer overflow when parsing list

When the input string is "key=[", the ending '\0' is replaced
by a ',', leading to a heap buffer overflow.

Check the content of ctx1 to avoid this problem.

Fixes: cc0579f2339a ("kvargs: support list value")
Cc: stable@dpdk.org
Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
4 years agotest/kvargs: fix invalid cases check
Olivier Matz [Fri, 27 Mar 2020 08:09:54 +0000 (09:09 +0100)]
test/kvargs: fix invalid cases check

The return was not properly placed, and only the first test case
was validated.

Fixes: e495f5435524 ("kvargs: add test case in app/test")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
4 years agotest/kvargs: fix to consider empty elements as valid
Olivier Matz [Fri, 27 Mar 2020 08:09:53 +0000 (09:09 +0100)]
test/kvargs: fix to consider empty elements as valid

Empty elements passed to the kvargs parser are silently
ignored. Examples of valid strings:
  ""
  ","
  ",,,,,,key=val,,,,"

Fix the unit tests to conform to this behavior.

Note: the test_invalid_kvargs() function is currently broken, which
explain why the tests were not failing. It is fixed in the next commit.

Fixes: e495f5435524 ("kvargs: add test case in app/test")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
4 years agolog: add API to check if a logtype can log in a given level
Ferruh Yigit [Fri, 13 Mar 2020 14:51:59 +0000 (14:51 +0000)]
log: add API to check if a logtype can log in a given level

This is a helper function in case components would like to do more work
than just logging a message based on log level, like for example
collecting some stats if the log type is DEBUG etc..

A few existing relevant usage converted to this new API.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Hyong Youb Kim <hyonkim@cisco.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Reviewed-by: Andrzej Ostruszka <aostruszka@marvell.com>
Acked-by: David Marchand <david.marchand@redhat.com>
4 years agomem: preallocate VA space in no-huge mode
Anatoly Burakov [Fri, 7 Feb 2020 11:11:14 +0000 (11:11 +0000)]
mem: preallocate VA space in no-huge mode

When --no-huge mode is used, the memory is currently allocated with
mmap(NULL, ...). This is fine in most cases, but can fail in cases
where DPDK is run on a machine with an IOMMU that is of more limited
address width than that of a VA, because we're not specifying the
address hint for mmap() call.

Fix it by preallocating VA space before mapping it.

Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: David Marchand <david.marchand@redhat.com>
Tested-by: Jun W Zhou <junx.w.zhou@intel.com>
4 years agovfio: map contiguous areas in one go
Anatoly Burakov [Tue, 25 Feb 2020 13:24:48 +0000 (13:24 +0000)]
vfio: map contiguous areas in one go

Currently, when we are creating DMA mappings for memory that's
either external or is backed by hugepages in IOVA as PA mode, we
assume that each page is necessarily discontiguous. This may not
actually be the case, especially for external memory, where the
user is able to create their own IOVA table and make it
contiguous. This is a problem because VFIO has a limited number
of DMA mappings, and it does not appear to concatenate them and
treats each mapping as separate, even when they cover adjacent
areas.

Fix this so that we always map contiguous memory in a single
chunk, as opposed to mapping each segment separately.

Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Ray Kinsella <ray.kinsella@intel.com>
4 years agoeal/freebsd: fix queuing duplicate alarm callbacks
Mit Matelske [Wed, 20 Nov 2019 20:10:56 +0000 (14:10 -0600)]
eal/freebsd: fix queuing duplicate alarm callbacks

The source callback list grows infinitely when more than alarm
is queued.

This fix recognizes that an alarm interrupt in FreeBSD should never
have more than one callback on its list, so if
rte_intr_callback_register() is called with an interrupt handle type
of RTE_INTR_HANDLE_ALARM, so if such an interrupt type already has a
non-empty list, then a new callback is not created, but the kevent
timer is restarted properly.

Fixes: 23150bd8d8a8 ("eal/bsd: add interrupt thread")
Cc: stable@dpdk.org
Signed-off-by: Mit Matelske <mit@pt.net>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agoeal: store control thread CPU affinity in TLS
Jerin Jacob [Wed, 5 Feb 2020 10:24:15 +0000 (15:54 +0530)]
eal: store control thread CPU affinity in TLS

_cpuset TLS variable stores the CPU affinity of eal thread.
Populate the _cpuset TLS variable for control thread to

1) Make rte_thread_get_affinity() and eal_thread_dump_affinity
functional with control thread.
2) Quick access to cpu affinity.

Signed-off-by: Jerin Jacob <jerinj@marvell.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
4 years agoci: enable unit test for aarch64
Ruifeng Wang [Mon, 23 Mar 2020 19:59:26 +0000 (20:59 +0100)]
ci: enable unit test for aarch64

Add Travis CI jobs to run unit tests on aarch64 platform.

Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Acked-by: Aaron Conole <aconole@redhat.com>
4 years agoci: run tests even without hugepage
Ruifeng Wang [Mon, 23 Mar 2020 19:59:25 +0000 (20:59 +0100)]
ci: run tests even without hugepage

As fast-tests suite generated with only applicable cases included,
hugepage is not a mandatory to run the test.
Ignore the result of hugepage set up, so that validation in environment
without hugepage can proceed.

Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Acked-by: Aaron Conole <aconole@redhat.com>
4 years agotest: allow no-huge mode for fast-tests
Ruifeng Wang [Mon, 23 Mar 2020 19:59:24 +0000 (20:59 +0100)]
test: allow no-huge mode for fast-tests

In environments where hugepage are not available, such as
containers, many cases in fast-tests suite should also run
if no-huge EAL option is used.

Flag is appended to each case in fast-tests suite to indicate
whether it lives with no-huge mode.
With the flag, fast-tests suite can be generated based on
detected hugepage availability of building environment.
All cases will be valid if hugepage is available, whereas
only applicable cases will be added if environment has no
hugepage support.

Suggested-by: Aaron Conole <aconole@redhat.com>
Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Acked-by: Aaron Conole <aconole@redhat.com>
Signed-off-by: David Marchand <david.marchand@redhat.com>
4 years agotest: skip some subtests in no-huge mode
Ruifeng Wang [Mon, 23 Mar 2020 19:59:23 +0000 (20:59 +0100)]
test: skip some subtests in no-huge mode

When running with '--no-huge' flag, tests failed with messages as:
    ACL context creation with invalid NUMA should have failed!
    fbk hash creation should have failed
    test_table_pipeline: Check pipeline invalid params failed.

These cases test against invalid socket ID as input parameter, and
expect error return. But function calls return success because
invalid sock ID is overwritten to SOCKET_ID_ANY when in no-huge mode.

The tests against invalid socket ID are skipped in no-huge mode.

Fixes: 5640171c528a ("malloc: fix external heap allocation in no-huge mode")
Cc: stable@dpdk.org
Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Acked-by: Aaron Conole <aconole@redhat.com>
Acked-by: Yipeng Wang <yipeng1.wang@intel.com>
4 years agotest: load drivers when required
David Marchand [Mon, 23 Mar 2020 19:59:22 +0000 (20:59 +0100)]
test: load drivers when required

Concatenating to test_args if event_eth_tx_adapter_autotest is executed
makes all subsequent tests inherit from the drivers loading while this
is unneeded.

Fixes: 207b1c813f39 ("test: fix build without ring PMD")
Cc: stable@dpdk.org
Signed-off-by: David Marchand <david.marchand@redhat.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Aaron Conole <aconole@redhat.com>
4 years agodevtools: silence meson install
David Marchand [Tue, 17 Mar 2020 16:25:21 +0000 (17:25 +0100)]
devtools: silence meson install

Installing with ninja is quite verbose by default, hide ninja output under
TEST_MESON_BUILD_VERBOSE and TEST_MESON_BUILD_VERY_VERBOSE options.

Signed-off-by: David Marchand <david.marchand@redhat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
4 years agodevtools: fix symbol map change check
Nithin Dabilpuram [Mon, 23 Mar 2020 11:56:13 +0000 (17:26 +0530)]
devtools: fix symbol map change check

Fix check symbol change script to detect new diff file when
it is in between  "--- /dev/null" to "b/lib/...".
Current awk line expects line to start with "a/..."
which is not always true for all diffs.
As a result if in_map was '1' earlier, it will not be changed
to '0' and we get check patch errors which are not true.

Fixes: 4bec48184e33 ("devtools: add checks for ABI symbol addition")
Cc: stable@dpdk.org
Signed-off-by: David Marchand <david.marchand@redhat.com>
Signed-off-by: Nithin Dabilpuram <ndabilpuram@marvell.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Tested-by: Jerin Jacob <jerinj@marvell.com>
4 years agomaintainers: update for ring library
Olivier Matz [Thu, 5 Mar 2020 17:00:08 +0000 (18:00 +0100)]
maintainers: update for ring library

Honnappa and Konstantin contributed actively to the ring library
and volunteered to replace myself as maintainers.

Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
4 years agoring: fix unaligned memory access on aarch32
Phil Yang [Mon, 9 Mar 2020 17:19:55 +0000 (01:19 +0800)]
ring: fix unaligned memory access on aarch32

The 32-bit arm machine doesn't support unaligned memory access. It
will cause a bus error on aarch32 with the custom element size ring.

Thread 1 "test" received signal SIGBUS, Bus error.
__rte_ring_enqueue_elems_64 (n=1, obj_table=0xf5edfe41, prod_head=0, \
r=0xf5edfb80) at /build/dpdk/build/include/rte_ring_elem.h:177
177                             ring[idx++] = obj[i++];

Fixes: cc4b218790f6 ("ring: support configurable element size")

Signed-off-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
4 years agocontigmem: cleanup properly when load fails
Jim Harris [Mon, 9 Mar 2020 10:00:25 +0000 (03:00 -0700)]
contigmem: cleanup properly when load fails

If contigmem is not able to allocate all of the
requested buffers, it frees whatever buffers were
able to be allocated up until that point.

But the pointers are not set to NULL in that case.
After the load fails, the FreeBSD kernel will
immediately call the contigmem unload handler, which
tries to free the buffers again since the pointers
were not set to NULL.

It's not clear that we should just rely on the unload
handler getting called after load failure. So let's
keep the existing cleanup code in the load handler,
but explicitly set the pointers to NULL after freeing
them.

Fixes: 5f51eca22489 ("contigmem: free allocated memory on error")
Cc: stable@dpdk.org
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agonet/mlx5: reduce Tx completion index memory loads
Alexander Kozyrev [Mon, 16 Mar 2020 15:34:59 +0000 (15:34 +0000)]
net/mlx5: reduce Tx completion index memory loads

There is a non-optimal check if doorbell is needed present in the
mlx5_tx_handle_completion() function. Advancing a copy of the txq
consumer index and checking this copy with initial value causes
unnecessary memory loads and hurts the performance. It is better to
have a simple small boolean variable for this purpose. That allows
to eliminate all the excessive memory operations with the txq consumer
index and restore the performance of the tx completions.

Fixes: 1fd9af05e44e ("net/mlx5: update Tx error handling routine")
Cc: stable@dpdk.org
Signed-off-by: Alexander Kozyrev <akozyrev@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
4 years agonet/mlx5: fix CVLAN tag set in IP item translation
Dekel Peled [Mon, 16 Mar 2020 08:57:39 +0000 (10:57 +0200)]
net/mlx5: fix CVLAN tag set in IP item translation

Previous fix added, at the end of functions
flow_dv_translate_item_ipv4() and flow_dv_translate_item_ipv6(), the
setting of cvlan_tag mask.
In the case of unspecified item (item->spec == null) these functions
return, and the new code section is not reached.

This patch moves the setting of cvlan_tag mask to be done before the
check of item->spec, to make sure it is always executed.

Fixes: 797329d6c4a1 ("net/mlx5: fix match on ethertype and CVLAN tag")
Cc: stable@dpdk.org
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/hinic: allocate IO memory with socket id
Xiaoyun Wang [Sat, 14 Mar 2020 10:02:06 +0000 (18:02 +0800)]
net/hinic: allocate IO memory with socket id

This patch makes IO memory allocation with socketid,
the txq or rxq descriptor and IO resource can be allocated with
socketid that passed by queue setup ops, which can improve
performance for cross-numa scene.

Cc: stable@dpdk.org
Signed-off-by: Xiaoyun Wang <cloud.wangxiaoyun@huawei.com>
4 years agonet/axgbe: support descriptor status
Amaranath Somalapuram [Mon, 2 Mar 2020 08:16:55 +0000 (13:46 +0530)]
net/axgbe: support descriptor status

Adding API axgbe_dev_rx_descriptor_status, axgbe_dev_tx_descriptor_status

Signed-off-by: Amaranath Somalapuram <asomalap@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
4 years agonet/axgbe: advertise supported packet types
Amaranath Somalapuram [Mon, 2 Mar 2020 08:16:54 +0000 (13:46 +0530)]
net/axgbe: advertise supported packet types

Adding api for eth_dev_supported_ptypes_get_t

Signed-off-by: Amaranath Somalapuram <asomalap@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
4 years agonet/axgbe: support queue query
Amaranath Somalapuram [Mon, 2 Mar 2020 08:16:53 +0000 (13:46 +0530)]
net/axgbe: support queue query

Adding API for axgbe_rxq_info_get, axgbe_txq_info_get

Signed-off-by: Amaranath Somalapuram <asomalap@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
4 years agonet/axgbe: support priority flow control
Amaranath Somalapuram [Tue, 3 Mar 2020 08:19:58 +0000 (13:49 +0530)]
net/axgbe: support priority flow control

Adding API for priority_flow_ctrl_set.
Priority flow control to set water high and low,
pause_time and priority.

Signed-off-by: Amaranath Somalapuram <asomalap@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
4 years agonet/axgbe: support flow control
Amaranath Somalapuram [Tue, 3 Mar 2020 08:19:57 +0000 (13:49 +0530)]
net/axgbe: support flow control

Adding api for flow_ctrl_get and flow_ctrl_set.
By default axgbe driver flow control is disabled.
Adding dpdk flow control to set water high and low.

Signed-off-by: Amaranath Somalapuram <asomalap@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
4 years agonet/mlx5: fix zero metadata action
Wisam Jaddo [Thu, 12 Mar 2020 09:24:40 +0000 (11:24 +0200)]
net/mlx5: fix zero metadata action

There is no reason to prevent the user from using zero metadata,
it indeed means no metadata, but sometimes the user may need to
report to datapath that there is no metadata anymore, which is
valid case.

Fixes: fcc8d2f716fd ("net/mlx5: extend flow metadata support")
Cc: stable@dpdk.org
Signed-off-by: Wisam Jaddo <wisamm@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
4 years agonet/ixgbe: fix setting VF MAC address
Guinan Sun [Wed, 11 Mar 2020 09:06:51 +0000 (09:06 +0000)]
net/ixgbe: fix setting VF MAC address

The reason why PF cannot receive data normally is that vf performed
the clear_rar operation through dev close without adding a mac address. 
This will cause the association between the index and rx address set
by VMDq to be cancelled,thus affecting the data reception of PF. The
correction method is to add a check action, and do not perform the
set_rar operation without adding a mac address to prevent affecting the
reception of data.

Fixes: 3c4270187518 ("net/ixgbe: support VF MAC address add/remove")
Cc: stable@dpdk.org
Signed-off-by: Guinan Sun <guinanx.sun@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/cxgbe: add devargs to control filtermode and filtermask
Karra Satwik [Wed, 11 Mar 2020 09:05:51 +0000 (14:35 +0530)]
net/cxgbe: add devargs to control filtermode and filtermask

Apart from the 4-tuple (IP src/dst addresses and TCP/UDP src/dst
port addresses), there are only 40-bits available to match other
fields in packet headers. Not all combinations of packet header
fields can fit in the 40-bit tuple.

Currently, the combination of packet header fields to match are
configured via filterMode for LETCAM filters and filterMask for
HASH filters in firmware config files (t5/t6-config.txt). So, add
devargs to allow User to dynamically select the filterMode and
filterMask combination during runtime, without having to modify the
firmware config files and reflashing them onto the adapter. A table
of supported combinations is maintained by the driver to internally
translate the User specified devargs combination to hardware's internal
format before writing the requested combination to hardware

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: use firmware API for validating filter spec
Karra Satwik [Wed, 11 Mar 2020 09:05:50 +0000 (14:35 +0530)]
net/cxgbe: use firmware API for validating filter spec

Add new firmware API FW_PARAM_DEV_FILTER_MODE_MASK to fetch
the filtermode and filtermask values configured in hardware,
which are used to validate the match combinations in the filter
spec before offloading the filter rules to hardware. For older
firmware that doesn't support the new API, fallback to older way
of directly reading from indirect registers

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: support flow API for source MAC rewrite
Karra Satwik [Wed, 11 Mar 2020 09:05:49 +0000 (14:35 +0530)]
net/cxgbe: support flow API for source MAC rewrite

Add support to rewrite Source MAC addresses. The new Source
MAC address is written into a free entry in the SMT table
and the corresponding SMT index is used by hardware to
rewrite the Source MAC address of the packets hitting the
flow.

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: support Source MAC Table
Karra Satwik [Wed, 11 Mar 2020 09:05:48 +0000 (14:35 +0530)]
net/cxgbe: support Source MAC Table

Source MAC Table (SMT) is used for storing Source MAC
addresses to be written in packets transmitted on the
wire. Hence, the SMT table can be used for overwriting
Source MAC addresses in packets, hitting corresponding
filter rules inserted by the rte_flow API.

Query firmware for SMT start and size information available
to the underlying PF. Allocate and maintain the corresponding
driver's copy of the hardware SMT table, with appropriate
refcount mechanism. If SMT information is not available, then
use the entire hardware SMT table.

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: support flow API for destination MAC rewrite
Karra Satwik [Wed, 11 Mar 2020 09:05:47 +0000 (14:35 +0530)]
net/cxgbe: support flow API for destination MAC rewrite

Add support for overwriting destination MAC addresses.
The new MAC address is written into a free entry in the
L2T table and the corresponding L2T index is used by
hardware to overwrite the destination MAC address of the
packets hitting the flow

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: support flow API for matching all packets on VF
Karra Satwik [Wed, 11 Mar 2020 09:05:46 +0000 (14:35 +0530)]
net/cxgbe: support flow API for matching all packets on VF

Add support to match all packets received on the underlying VF.

Use new firmware API to fetch the Virtual Interface Number (VIN)
allocated to each VF by the firmware. The VIN is required to
write filter rules to match all packets on VFs, whose identifier
is beyond max 7-bit value (i.e. 127) in VIID.

If firmware doesn't support fetching the VIN information, then
fallback to manually retrieving the VIN value from the 7-bit field
in the VIID, which only supports in range of 0..127. In this case,
packets belonging to VFs, whose identifier is beyond 127 can't be
matched.

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: support flow API for matching all packets on PF
Karra Satwik [Wed, 11 Mar 2020 09:05:45 +0000 (14:35 +0530)]
net/cxgbe: support flow API for matching all packets on PF

Add support to match all packets received on the underlying PF

Note that the same 17-bit hardware tuple is shared between QinQ
and PF match. Hence, match on either QinQ or PF only can be done
at a time. Both QinQ and PF match can't be enabled at the same time.

Also, remove check to reject rules without spec because
RTE_FLOW_ITEM_TYPE_PF doesn't require a spec. Due to this check
removal, RTE_FLOW_ITEM_TYPE_PHY_PORT item needs to be updated to
handle NULL spec

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: support flow API for matching IP TOS
Karra Satwik [Wed, 11 Mar 2020 09:05:44 +0000 (14:35 +0530)]
net/cxgbe: support flow API for matching IP TOS

Add support to match Type of Service (TOS) field in
IPv4/IPv6 header

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: support flow API for matching QinQ VLAN
Karra Satwik [Wed, 11 Mar 2020 09:05:43 +0000 (14:35 +0530)]
net/cxgbe: support flow API for matching QinQ VLAN

Add support to match fields in 802.1ad Q-in-Q VLAN packets.
Relax check for repeated pattern items for RTE_FLOW_ITEM_TYPE_VLAN
item, since the same item is used to represent both QinQ and VLAN
packets.

When QinQ match is enabled, the ethertype field in the hardware
spec must contain the innermost VLAN header's ethertype field,
and not the Ethernet header's ethertype field. The hardware
automatically searches for ethertype 0x88A8/0x8100 in Ethernet
header, when parsing incoming packet against QinQ/VLAN pattern,
respectively.

Signed-off-by: Karra Satwik <kaara.satwik@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/sfc/base: handle manual and auto filter clashes in EF10
Igor Romanov [Tue, 10 Mar 2020 09:48:39 +0000 (09:48 +0000)]
net/sfc/base: handle manual and auto filter clashes in EF10

Make user filters a priority in EF10 datapath. When a manual
filter with a specification that is equal to an existing auto
filter is inserted, the manual filter:
- replaces auto filter if the specification is exclusive;
- is inserted along existing auto filter otherwise;

In the first case the auto filter that was replaced is saved.
This saved filter can be updated on filter reconfiguration and
is restored on the manual filter removal.

Fixes: e7cd430c864f ("net/sfc/base: import SFN7xxx family support")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc/base: refactor filter lookup loop in EF10
Igor Romanov [Tue, 10 Mar 2020 09:48:38 +0000 (09:48 +0000)]
net/sfc/base: refactor filter lookup loop in EF10

The code is refactored to make it more clear and allow for
more convenient further changes.

Fixes: e7cd430c864f ("net/sfc/base: import SFN7xxx family support")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc/base: reject automatic filter creation by users
Igor Romanov [Tue, 10 Mar 2020 09:48:37 +0000 (09:48 +0000)]
net/sfc/base: reject automatic filter creation by users

The automatic filter priority may only be used by filter implementation,
so reject the attempts to create such filters by client drivers.

Fixes: f9565517ff4f ("net/sfc/base: import filters support")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc/base: reduce filter priorities to implemented only
Igor Romanov [Tue, 10 Mar 2020 09:48:36 +0000 (09:48 +0000)]
net/sfc/base: reduce filter priorities to implemented only

Only implemented filter priorities are manual and auto. Remove
unused priorities since exposing unimplemented functionality
is of no benefit.

Tx filter specification default priority is changed to manual,
since required priority does not differ from it in the implementation.

Fixes: f9565517ff4f ("net/sfc/base: import filters support")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: set priority of created filters to manual
Igor Romanov [Tue, 10 Mar 2020 09:48:35 +0000 (09:48 +0000)]
net/sfc: set priority of created filters to manual

The priority should be explicitly set to manual for correct
usage of libefx filters.

Fixes: a9825ccf5bb8 ("net/sfc: support flow API filters")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: fix promiscuous and allmulticast toggles errors
Igor Romanov [Tue, 10 Mar 2020 09:42:09 +0000 (09:42 +0000)]
net/sfc: fix promiscuous and allmulticast toggles errors

Returned errors of ethdev callbacks are negative. Internal
sfc functions return positive errors, so convert them to
negative value.

Fixes: 9039c8125730 ("ethdev: change promiscuous callbacks to return status")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agoethdev: fix spelling
Stephen Hemminger [Tue, 10 Mar 2020 16:24:05 +0000 (09:24 -0700)]
ethdev: fix spelling

Minor spelling errors found by aspell and codespell

Fixes: 1daa33805824 ("ethdev: validate offloads set by PMD")
Fixes: 81f9db8ecc2c ("ethdev: add vlan offload support")
Fixes: c8231c63ddcb ("ethdev: insert Rx callback as head of list")
Cc: stable@dpdk.org
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/ionic: use standard boolean type
Sunil Kumar Kori [Mon, 9 Mar 2020 09:09:27 +0000 (14:39 +0530)]
net/ionic: use standard boolean type

Any DPDK public header file which includes stdbool.h may conflict with
local definition of bool, if any, which further results in compilation
error. To avoid, used standard stdbool.h instead of defining bool
internally.

I observed this issue during a development where I included rte_uuid.h
into rte_ethdev.h. As rte_ethdev.h is included to PMD driver, it started
throwing error as given below:

  CC ionic_rxtx.o
In file included from .../dpdk/build/include/rte_uuid.h:17:0,
                 from .../dpdk/build/include/rte_ethdev.h:161,
                 from .../dpdk/build/include/rte_ethdev_driver.h:18,
                 from .../dpdk/drivers/net/ionic/ionic_rxtx.c:34:
.../dpdk/drivers/net/ionic/ionic_osdep.h:48:17:
    error: two or more data types in declaration specifiers
typedef uint8_t bool;
                 ^
In file included from .../dpdk/drivers/net/ionic/ionic_dev.h:8:0,
                 from .../dpdk/drivers/net/ionic/ionic.h:13,
                 from .../dpdk/drivers/net/ionic/ionic_mac_api.h:8,
                 from .../dpdk/drivers/net/ionic/ionic_rxtx.c:45:
.../dpdk/drivers/net/ionic/ionic_osdep.h:48:1:
    warning: useless type name in empty declaration
typedef uint8_t bool;
^~~~~~~
cc1: warning: unrecognized command line option
‘-Wno-address-of-packed-member’
.../dpdk/mk/internal/rte.compile-pre.mk:114:
recipe for target 'ionic_rxtx.o' failed

Fixes: 5ef518098ec6 ("net/ionic: register and initialize adapter")
Cc: stable@dpdk.org
Signed-off-by: Sunil Kumar Kori <skori@marvell.com>
Acked-by: Alfredo Cardigliano <cardigliano@ntop.org>
4 years agonet/hns3: fix promiscuous mode for PF
Chengchang Tang [Mon, 9 Mar 2020 09:32:43 +0000 (17:32 +0800)]
net/hns3: fix promiscuous mode for PF

Currently, when promiscuous mode is enabled, it is just allowed to
accept all the unicast and broadcast packets in hns3 PF PMD driver. It
should also be able to receive multicast packets.

Fixes: 19a3ca4c99cf ("net/hns3: add start/stop and configure operations")
Cc: stable@dpdk.org
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agonet/hns3: support promiscuous and allmulticast mode for VF
Chengchang Tang [Mon, 9 Mar 2020 09:32:42 +0000 (17:32 +0800)]
net/hns3: support promiscuous and allmulticast mode for VF

Currently, we only support VF device is bound to vfio_pci or igb_uio and
then driven by DPDK driver when PF is driven by kernel mode hns3 ethdev
driver, VF is not supported when PF is driven by hns3 DPDK driver.

This patch adds promiscuous and allmulticast mode support for hns3 VF
PMD driver.
1) The promiscuous/allmulticast mode can be configured successfully only
   based on the trusted VF device. If based on the non trusted VF
   device, configuring promiscuous/allmulticast mode will fail. The hns3
   VF device can be configured as trusted device by hns3 PF kernel
   ethdev driver on the host by "ip link set <eth num> vf <vf id> turst
   on" command.
2) After the promiscuous mode is configured successfully, hns3 VF PMD
   driver can receive the ingress and outgoing traffic. In the words,
   all the ingress packets, all the packets sent from the PF and other
   VFs on the same physical port.
3) Note: Because of the hardware constraints, By default vlan filter is
   enabled and couldn't be turned off based on VF device, so vlan filter
   is still effective even in promiscuous mode. If upper applications
   don't call rte_eth_dev_vlan_filter API function to set vlan based on
   VF device, hns3 VF PMD driver will can't receive the packets with
   vlan tag in promiscuous mode.

Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agonet/hns3: remove restriction on setting VF MTU
Chengchang Tang [Mon, 9 Mar 2020 09:32:41 +0000 (17:32 +0800)]
net/hns3: remove restriction on setting VF MTU

The hns3 PF/VF devices on the same port share the hardware MTU
configuration. Currently, we send mailbox to inform hns3 PF kernel
ethdev driver to finish hardware MTU configuration in hns3 VF PMD
driver, there is no need to stop the port for hns3 VF device, and
the MTU value issued by hns3 VF PMD driver must be less than or
equal to PF's MTU.

Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agonet/hns3: fix Tx interrupt when enabling Rx interrupt
Wei Hu (Xavier) [Mon, 9 Mar 2020 09:32:40 +0000 (17:32 +0800)]
net/hns3: fix Tx interrupt when enabling Rx interrupt

Currently, when receiving and transmitting packets based on hns3 network
engine there are probably unexpected and redundant Tx interrupts if Rx
interrupt is enabled.

The root cause as below:
Tx and Rx queues with the same number share the interrupt vector in hns3
network engine, and in this case there are the residual hardware mapping
relationship configuration between queue and interrupt vector configured
in hns3 kernel ethdev driver.

We should clear the all hardware mapping relationship configurations in
the initialization. Because of the hardware constraints, we have to
implement clearing the relationship by binding all queues to the last
interrupt vector and reserving the last interrupt vector, this method
results in a decrease of the maximum queues when upper applications call
the rte_eth_dev_configure API function to enable Rx interrupt.

Fixes: 02a7b55657b2 ("net/hns3: support Rx interrupt")
Cc: stable@dpdk.org
Signed-off-by: Hao Chen <chenhao164@huawei.com>
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agonet/hns3: support TSO
Hongbo Zheng [Mon, 9 Mar 2020 09:32:39 +0000 (17:32 +0800)]
net/hns3: support TSO

This patch adds TCP segment offload support for hns3 PMD driver.

Signed-off-by: Hongbo Zheng <zhenghongbo3@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agoethdev: add PFCP header to flow API
Xiao Zhang [Fri, 6 Mar 2020 06:39:26 +0000 (14:39 +0800)]
ethdev: add PFCP header to flow API

This patch adds the new flow item RTE_FLOW_ITEM_TYPE_PFCP to flow API to
match a PFCP header.
Add sample PFCP rules for testpmd guide. Since Session Endpoint
Identifier (SEID) only will be present in PFCP Session header and PFCP
Session headers shall be identified when the S field is equal to 1, when
create rules for PFCP Session header with certain SEID the S field need
be set 1.

Signed-off-by: Xiao Zhang <xiao.zhang@intel.com>
Acked-by: Ori Kam <orika@mellanox.com>
4 years agonet/ice: remove redundant functions
Qi Zhang [Mon, 9 Mar 2020 12:12:14 +0000 (20:12 +0800)]
net/ice: remove redundant functions

Remove function ice_clear_queues, since all equivalent code
has already been executed during ice_rx|tx_queue_stop.

Also function ice_rx|tx_queue_release_mbufs simply wrapped a
function pointer call which is not necessary, remove them.

Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/hinic: fix repeating cable log and length check
Kevin Traynor [Fri, 6 Mar 2020 12:06:52 +0000 (12:06 +0000)]
net/hinic: fix repeating cable log and length check

gcc 10.0.1 reports:

../drivers/net/hinic/base/hinic_pmd_hwdev.c: In function ‘print_cable_info’:
../drivers/net/hinic/base/hinic_pmd_hwdev.c:1398:3:
warning:
‘snprintf’ argument 4 may overlap destination object ‘tmp_str’
[-Wrestrict]
 1398 |   snprintf(tmp_str + strlen(tmp_str), (sizeof(tmp_str) - 1),
      |   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1399 |     "%s, Temperature: %u", tmp_str,
      |     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1400 |     info->cable_temp);
      |     ~~~~~~~~~~~~~~~~~

The warning is that tmp_str is in both src and dest.

Anyway, the current code is incorrect and because of the +strlen
the existing string will be repeated twice and max length
does not limit to the end of the string.

Fix by removing tmp_str from the src of snprintf and adding the
correct max length.

Fixes: d9ce1917941c ("net/hinic/base: add hardware operation")
Cc: stable@dpdk.org
Signed-off-by: Kevin Traynor <ktraynor@redhat.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/hinic: fix snprintf length of cable info
Kevin Traynor [Fri, 6 Mar 2020 12:06:51 +0000 (12:06 +0000)]
net/hinic: fix snprintf length of cable info

Correct length and remove unnecessary brackets.

Fixes: d9ce1917941c ("net/hinic/base: add hardware operation")
Cc: stable@dpdk.org
Signed-off-by: Kevin Traynor <ktraynor@redhat.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/memif: fix init when already connected
Július Milan [Wed, 19 Feb 2020 08:19:16 +0000 (09:19 +0100)]
net/memif: fix init when already connected

This patch fixes the situation when there is already connected pair of
memif interfaces and another slave tries to initiate the connection with
(already occupied) master. Expected behavior is that the second slave
is refused and gets disconnect message with reason: "Already connected",
while old connection remains functional.

Fixes: 09c7e63a71f9 ("net/memif: introduce memory interface PMD")
Cc: stable@dpdk.org
Signed-off-by: Július Milan <jmilan.dev@gmail.com>
Reviewed-by: Jakub Grajciar <jgrajcia@cisco.com>
4 years agonet/memif: enable loopback
Július Milan [Mon, 9 Mar 2020 14:22:08 +0000 (15:22 +0100)]
net/memif: enable loopback

With this patch it is possible to connect 2 DPDK memifs into loopback,
i.e. when they have the same id and different roles, as for example:
  "--vdev=net_memif0,role=master,id=0"
  "--vdev=net_memif1,role=slave,id=0"

Signed-off-by: Július Milan <jmilan.dev@gmail.com>
Reviewed-by: Jakub Grajciar <jgrajcia@cisco.com>
4 years agocommon/mlx5: fix build without dlopen option
Tonghao Zhang [Fri, 6 Mar 2020 14:27:19 +0000 (22:27 +0800)]
common/mlx5: fix build without dlopen option

When setting the CONFIG_RTE_BUILD_SHARED_LIB to y,
and build the mlx5 pmd, there is a building error.
To fix it, add RTE_IBVERBS_LINK_DLOPEN to include
relative codes.

    mlx5_common.o: In function `mlx5_glue_init':
    drivers/common/mlx5/mlx5_common.c:324: undefined reference to `dlclose'

Fixes: 7b4f1e6bd367 ("common/mlx5: introduce common library")
Cc: stable@dpdk.org
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Ali Alnubani <alialnu@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/mlx5: support match on IPv4 TTL and IPv6 HLIM
Dekel Peled [Tue, 3 Mar 2020 12:37:31 +0000 (14:37 +0200)]
net/mlx5: support match on IPv4 TTL and IPv6 HLIM

This patch adds to MLX5 PMD the support of matching on IPv4 TTL and
IPv6 hoplimit.
Support is valid when using DV flow engine (Verbs flow engine doesn't
support matching on this field).

Signed-off-by: Eli Britstein <elibr@mellanox.com>
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/mlx5: fix mask used for IPv6 item validation
Dekel Peled [Tue, 3 Mar 2020 12:37:30 +0000 (14:37 +0200)]
net/mlx5: fix mask used for IPv6 item validation

The nic_mask struct in function mlx5_flow_validate_item_ipv6()
includes hop_limits field.
This is redundant since matching on this item is not supported using
Verbs flow engine (dv_flow_en=0).

(Separate patch will be sent to handle this issue for DV flow engine.)

Setting a rule including match on this field will fail:
testpmd> flow create 0 ingress pattern eth / ipv6 hop is 1 / end actions
drop / end
port_flow_complain(): Caught PMD error type 1 (cause unspecified):
hardware refuses to create flow: Operation not supported

This patch removes the redundant field from nic_mask, to ensure that
such flow rules will be rejected by PMD validation:
testpmd> flow create 0 ingress pattern eth / ipv6 hop is 1 / end actions
drop / end
port_flow_complain(): Caught PMD error type 13 (specific pattern item):
cause: 0x7fffffffa2c8, mask enables non supported bits: Operation not
supported

The related lines in function flow_verbs_translate_item_ipv6() are
removed as well, since they are redundant.

Fixes: 23c1d42c7138 ("net/mlx5: split flow validation to dedicated function")
Fixes: 84c406e74524 ("net/mlx5: add flow translate function")
Cc: stable@dpdk.org
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
4 years agonet/ice: remove bulk alloc option
Qi Zhang [Wed, 4 Mar 2020 11:40:38 +0000 (19:40 +0800)]
net/ice: remove bulk alloc option

Remove CONFIG_RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC with below
consideration:

1. A default Rx path can always be selected by setting a proper
   rx_free_thresh value at runtime, see
   ice_check_rx_burst_bulk_alloc_preconditions.

2. Its not a big deal to always reserve more space for desc ring.
   "ring_size = (uint16_t)(rxq->nb_rx_desc + ICE_RX_MAX_BURST);"

3. Fixes a potential invalid memory access in ice_reset_rx_queue.
   If CONFIG_RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC is turned on while
   ice_check_rx_burst_bulk_alloc_preconditions return fail.
   Below code will have problem.

   for (i = 0; i < ICE_RX_MAX_BURST; ++i)
    rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;

Fixes: 50370662b727 ("net/ice: support device and queue ops")
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/sfc: fix Rx queue start failure path
Igor Romanov [Thu, 5 Mar 2020 15:48:55 +0000 (15:48 +0000)]
net/sfc: fix Rx queue start failure path

Call correct cleanup operations on failure in Rx queue start function.

Fixes: df1bfde4ff0d ("net/sfc: factor out libefx-based Rx datapath")
Fixes: 28944ac098aa ("net/sfc: implement Rx queue start and stop operations")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: generalise flow pattern item processing
Ivan Malov [Thu, 5 Mar 2020 10:47:53 +0000 (10:47 +0000)]
net/sfc: generalise flow pattern item processing

This is needed to reuse pattern processing engine for MAE.

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: generalise flow start and stop path
Ivan Malov [Thu, 5 Mar 2020 10:47:52 +0000 (10:47 +0000)]
net/sfc: generalise flow start and stop path

As a preparation step, generalise flow start and stop path using
callbacks.

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: generalise flow parsing
Ivan Malov [Thu, 5 Mar 2020 10:47:51 +0000 (10:47 +0000)]
net/sfc: generalise flow parsing

Generalise flow attribute parsing function in regard to transfer attribute.
Add a method table and factor out VNIC-specific parsing code as a callback.

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: introduce flow allocation and free path
Ivan Malov [Thu, 5 Mar 2020 10:47:50 +0000 (10:47 +0000)]
net/sfc: introduce flow allocation and free path

Riverhead boards maintain support for MAE, a low-level Match-Action Engine.
The upcoming patches will bring support for MAE to RTE flow implementation.

A follow-up patch will introduce MAE-specific specification cleanup method.
In order to prepare for the patch, introduce flow allocation and free path.

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: generalise flow specification structure
Ivan Malov [Thu, 5 Mar 2020 10:47:49 +0000 (10:47 +0000)]
net/sfc: generalise flow specification structure

Add the concept of a flow specification type.

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: make flow list engine-agnostic
Ivan Malov [Thu, 5 Mar 2020 10:47:48 +0000 (10:47 +0000)]
net/sfc: make flow list engine-agnostic

A backend which a driver employs to handle flow rules of a given
type depends on the underlying NIC flow engine. The driver in
question in its current state is tailored to support the only
flow engine, VNIC filtering. As the need arises to add support
for transfer rules, the driver has to be reworked so that it
becomes possible to introduce yet another backend.

As a preparation step, make the flow list shared
between different engines.

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/sfc: make flow RSS details VNIC-specific
Ivan Malov [Thu, 5 Mar 2020 10:47:47 +0000 (10:47 +0000)]
net/sfc: make flow RSS details VNIC-specific

The flow specification structure will be transformed to a generic one, and
its current contents will be fenced off to form a VNIC-specific parameters.
Flow RSS details do not belong to the said specification currently, albeit
being VNIC-specific. This patch addresses this issue as a preparation step.

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/axgbe: support unicast hash table for MAC address
Chandu Babu N [Fri, 28 Feb 2020 14:20:53 +0000 (19:50 +0530)]
net/axgbe: support unicast hash table for MAC address

Implement eth_dev_ops uc_hash_table_set and uc_all_hash_table_set

Signed-off-by: Chandu Babu N <chandu@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
4 years agonet/axgbe: support setting MAC address
Chandu Babu N [Fri, 28 Feb 2020 14:20:52 +0000 (19:50 +0530)]
net/axgbe: support setting MAC address

Supports adding MAC addresses to enable whitelist filtering to
accept packets
implement eth_dev_ops mac_addr_set, mac_addr_add, mac_addr_remove and
set_mc_addr_list

Signed-off-by: Chandu Babu N <chandu@amd.com>
Acked-by: Ravi Kumar <ravi1.kumar@amd.com>
4 years agonet/i40e: fix DCB initialization
Shougang Wang [Thu, 5 Mar 2020 10:42:35 +0000 (10:42 +0000)]
net/i40e: fix DCB initialization

Stopping LLDP is necessary for DPDK, but it will cause DCB init failure.
For kernel shared code, the prerequisite for successful initialization
of DCB is that LLDP is enabled. This patch starts LLDP before DCB init
and stops it when init finishes.

Fixes: a8e84b22bd55 ("net/i40e/base: support persistent LLDP")
Cc: stable@dpdk.org
Signed-off-by: Shougang Wang <shougangx.wang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/sfc: fix initialization error path
Igor Romanov [Thu, 5 Mar 2020 10:38:36 +0000 (10:38 +0000)]
net/sfc: fix initialization error path

NIC deinitialization should not happen after a failed NIC
initialization.

Fixes: 91831d4068c8 ("net/sfc: estimate available resources")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/vmxnet3: fix RSS setting on v4
Eduard Serra [Wed, 4 Mar 2020 02:35:56 +0000 (02:35 +0000)]
net/vmxnet3: fix RSS setting on v4

When calling to setup RSS on v4 API, ESX will expect
IPv4/6 TCP RSS to be set/requested mandatory.

This patch will:
- Set IPv4/6 TCP RSS when these have not been set. A warning
message is thrown to make sure we warn the application we are
setting IPv4/6 TCP RSS when not set.
- An additional check has been added to dodge RSS configuration
altogether unless MQ_RSS has been requested, similar to v3.

The alternative (returning error) was considered, the intent
is to ease the task of setting up and running vmxnet3 in situations
where it's supposed to be most straightforward (testpmd, pktgen).

Bugzilla ID: 400
Fixes: 643fba770705 ("net/vmxnet3: add v4 boot and guest UDP RSS config")
Cc: stable@dpdk.org
Signed-off-by: Eduard Serra <eserra@vmware.com>
Acked-by: Yong Wang <yongwang@vmware.com>
4 years agonet/axgbe: support scattered Rx
Selwin Sebastian [Wed, 4 Mar 2020 13:44:14 +0000 (19:14 +0530)]
net/axgbe: support scattered Rx

Enable scattered rx support and add jumbo packet receive capability

Signed-off-by: Selwin Sebastian <selwin.sebastian@amd.com>
4 years agonet/sfc: fix reported promiscuous/multicast mode
Andrew Rybchenko [Wed, 4 Mar 2020 15:01:04 +0000 (15:01 +0000)]
net/sfc: fix reported promiscuous/multicast mode

Requested promiscuous or all-multicast mode may be dropped on port
start if FW denies to enable it (e.g. because of no permission
in the case of VF). Return applied value on get.

Fixes: f5258439ee5d ("net/sfc: avoid failure on port start if Rx mode is rejected")
Cc: stable@dpdk.org
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agodoc: fix number of failsafe sub-devices
Gaetan Rivet [Sat, 22 Feb 2020 00:14:40 +0000 (01:14 +0100)]
doc: fix number of failsafe sub-devices

Supporting more than 2 sub-device is a remnant from earlier fail-safe
design, that was never actually published.

Fixes: a46f8d584eb8 ("net/failsafe: add fail-safe PMD")
Cc: stable@dpdk.org
Signed-off-by: Gaetan Rivet <grive@u256.net>
4 years agodoc: improve failsafe guide
Gaetan Rivet [Sat, 22 Feb 2020 00:14:39 +0000 (01:14 +0100)]
doc: improve failsafe guide

Reading the fail-safe doc with a few years added, a few phrasing
choices are ambiguous or confusing.

Signed-off-by: Gaetan Rivet <grive@u256.net>
Acked-by: Marko Kovacevic <marko.kovacevic@intel.com>
4 years agonet/ice: remove unnecessary variable
Qi Zhang [Wed, 4 Mar 2020 02:05:49 +0000 (10:05 +0800)]
net/ice: remove unnecessary variable

Remove unnecessary variable "meta" in ice_flow_create and
ice_flow_validate, it should be defined when really needed:
its ice_parse_engine_create and ice_parse_engine_validate.

A meta data created by parser->parse_pattern_action should be
freed inside parser->engine->create.

During a validate operation, since parser->engine_>create will not
be invoked, it is unnecessary to create a meta during parser, so
NULL will passed to engine->parse_pattern_action and all parser's
parse_pattern_action need to be modified to handle meta = NULL
properly.

With above implementation, the patch also fixes a potential memory
leak in ice_parse_engine_validate, since meta may not be freed.

BTW, an engine without a create op should be regarded as a bug. So
use RTE_ASSERT to replace runtime engine->create == NULL check in
ice_parse_engine_create.

Fixes: 4e27d3ed02bd ("net/ice: fix flow API framework")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/enetc: init SI transactions attribute register
Alex Marginean [Mon, 2 Mar 2020 14:32:09 +0000 (20:02 +0530)]
net/enetc: init SI transactions attribute register

This was left to its default value.  With the patch transactions are:
- coherent,
- do not allocate in downstream cache (there is none on LS1028a),
- merge surrounding data for BD writes,
- overwrite surrounding data for frame data writes.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: improve prefetch in Rx ring clean
Alex Marginean [Mon, 2 Mar 2020 14:32:08 +0000 (20:02 +0530)]
net/enetc: improve prefetch in Rx ring clean

LS1028A does not have platform cache so any reads following a hardware
write will go directly to DDR.  Latency of such a read is in excess of
100 core cycles, so try to prefetch more in advance to mitigate this.
How much is worth prefetching really depends on traffic conditions.
With congested Rx this could go up to 4 cache lines or so.  But if
software keeps up with hardware and follows behind Rx PI by a cache line
then it's harmful in terms of performance to cache more.  We would only
prefetch data that's yet to be written by ENETC, which will be evicted
again anyway.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: use bulk free in Tx clean
Alex Marginean [Mon, 2 Mar 2020 14:32:07 +0000 (20:02 +0530)]
net/enetc: use bulk free in Tx clean

Use rte_pktmbuf_free_bulk to release all mbufs at once.  This is flagged
as obsolete/not yet stable in DPDK but seems to be functional.
Don't count the released frames, it's no longer needed in the caller.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: use bulk alloc in Rx refill ring
Alex Marginean [Mon, 2 Mar 2020 14:32:06 +0000 (20:02 +0530)]
net/enetc: use bulk alloc in Rx refill ring

Since we know in advance that we're going to fill in multiple
descriptors it's convenient to allocate the buffers in batches.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: reorder BDR structure
Alex Marginean [Mon, 2 Mar 2020 14:32:05 +0000 (20:02 +0530)]
net/enetc: reorder BDR structure

Reorder the members of the structure so that the ones used on datapath
fit in a single cache line, to slightly reduce pressure on cache and
miss rate.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: improve batching Rx ring refill
Alex Marginean [Mon, 2 Mar 2020 14:32:04 +0000 (20:02 +0530)]
net/enetc: improve batching Rx ring refill

Move from doing batch refill of Rx ring from bundles of 8 to once per
enetc_clean_rx_ring call.  One benefit is that we're cleaning up all the
BDs that we just processed, which should still be cached.  The other is
that hardware Rx index stays a little back and doesn't cause contention
on the BDs processed in the Rx loop.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: fix Rx lock-up
Alex Marginean [Mon, 2 Mar 2020 14:32:03 +0000 (20:02 +0530)]
net/enetc: fix Rx lock-up

The default value in hardware for the Rx MAC FIFO (@) is higher than it
should be and can lead to Rx lock-up under traffic.
Set it to the value recommended by hardware team, 1.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: batch process clean Tx ring calls
Alex Marginean [Mon, 2 Mar 2020 14:32:02 +0000 (20:02 +0530)]
net/enetc: batch process clean Tx ring calls

Each call to enetc_clean_tx_ring will cost at least 150-200 CPU cycles
even if no clean-up is done, due to the CI register read.
We're only calling it once at the end of the function, on the assumption
that software is slower than hardware and hardware completed sending
older frames out by now.
We're also cleaning up the ring before kicking off Tx for the new batch
to minimize chances of contention on the Tx ring.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: relax read for Tx CI in clean Tx
Alex Marginean [Mon, 2 Mar 2020 14:32:01 +0000 (20:02 +0530)]
net/enetc: relax read for Tx CI in clean Tx

We don't need barriers here since this read doesn't have to be strictly
serialized in relation to other surrounding memory/register accesses.
We only want a reasonably recent value out of hardware so we know how
much we can clean.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/enetc: do not stall in clean Tx ring
Alex Marginean [Mon, 2 Mar 2020 14:32:00 +0000 (20:02 +0530)]
net/enetc: do not stall in clean Tx ring

Don't read the hardware CI register in a loop, read it once, clean up
and exit.
The issue with reading the register in a loop is that we're stalling
here trying to catch up with hardware which keeps sending traffic as
long as it has traffic to send, so in effect we could be waiting here
for the Tx ring to be drained by hardware, instead of us doing Rx in
that meantime.
At the time we return the function there may be new BDs in the ring that
could be cleaned, we're just leaving those there for the next time.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Acked-by: Gagandeep Singh <g.singh@nxp.com>
4 years agonet/bnxt: fix MAC address setting when port is stopped
Kalesh AP [Thu, 27 Feb 2020 08:18:51 +0000 (13:48 +0530)]
net/bnxt: fix MAC address setting when port is stopped

Driver destroys the VNIC when the port is brought down.
Port HW filter setting such as default MAC address and
unicast MAC filters will be applied when port is started.

Fixed to return success silently for these callbacks
when port is stopped.

Fixes: 39b88344e364 ("net/bnxt: fix enable/disable VLAN filtering")
Cc: stable@dpdk.org
Reported-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
4 years agonet/bnxt: fix recovery alarm race condition in port close
Kalesh AP [Fri, 28 Feb 2020 10:21:31 +0000 (15:51 +0530)]
net/bnxt: fix recovery alarm race condition in port close

There is a race condition when port is closed while error recovery
is happening in driver. Fixed this by canceling the recovery threads
during port close.

Fixes: df6cd7c1f73a ("net/bnxt: handle reset notify async event from FW")
Cc: stable@dpdk.org
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Santoshkumar Karanappa Rastapur <santosh.rastapur@broadcom.com>
4 years agonet/ice: fix hash flow crash
Tao Zhu [Tue, 3 Mar 2020 05:38:46 +0000 (13:38 +0800)]
net/ice: fix hash flow crash

Commit 'dc36bd5dfdeb' mistakenly initializes ret as -rte_errno, then
return ret if ice_search_pattern_match_item fails, if rte_errno happens
to be 0 in this case, it would cause an unrecognized flow rule to be
marked as recognition success. Later, when the code tries to parse the
flow rule, a segmentation fault will occur due to null pointer
reference.

Fixes: dc36bd5dfdeb ("net/ice: fix flow FDIR/switch memory leak")
Cc: stable@dpdk.org
Signed-off-by: Tao Zhu <taox.zhu@intel.com>
Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/i40e: fix X722 performance
Qiming Yang [Tue, 3 Mar 2020 05:45:00 +0000 (13:45 +0800)]
net/i40e: fix X722 performance

Lack of GL_SWR_PM_UP_THR value for mac type 4x10G X722 caused it can't
reach line rate in one queue one core forwarding case. It's about 1%
performance drop.  This patch adds value as:

 Device-ID  Value        Comments
 0x37D0    0x03030303   4x10G FPK

This value is hardware specific.

Fixes: 3320d4a240da ("net/i40e: workaround performance degradation")
Cc: stable@dpdk.org
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/af_packet: remove queue number limitation
Stephen Hemminger [Thu, 27 Feb 2020 20:00:03 +0000 (12:00 -0800)]
net/af_packet: remove queue number limitation

Since qpairs is part of the vdev arguments, there is no need to
limit it to 16. The queue arrays can be dynamically sized based
on the requested parameters.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: John W. Linville <linville@tuxdriver.com>
4 years agoapp/testpmd: parse flow command line for AH
Xiao Zhang [Fri, 14 Feb 2020 00:52:44 +0000 (08:52 +0800)]
app/testpmd: parse flow command line for AH

Add AH and AH SPI to testpmd rte flow command line.
Add note and sample AH rules in testpmd guide.

Signed-off-by: Xiao Zhang <xiao.zhang@intel.com>
Acked-by: Ori Kam <orika@mellanox.com>
4 years agonet/mlx5: fix VLAN PCP item calculation
Ophir Munk [Wed, 26 Feb 2020 08:27:48 +0000 (08:27 +0000)]
net/mlx5: fix VLAN PCP item calculation

The VLAN 16 bits tci field contains both values of PCP and VID. When
extracting any one of them - it is required not to affect the other one.
Previous to this commit in routine flow_dev_get_vlan_info_from_items()
we calculated the PCP as follows:
    (1) vlan->vlan_tci &= MLX5DV_FLOW_VLAN_PCP_MASK;
    (2) vlan->vlan_tci |= <3 bits value of PCP>
In line (1) we should have used the negated mask ('~' operator) such
that only the PCP bits will be nullified before ORing them with the
updated PCP value.

Fixes: 9aee7a8418d4 ("net/mlx5: support push flow action on VLAN header")
Cc: stable@dpdk.org
Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
4 years agonet/i40e: relax barrier in Tx
Gavin Hu [Wed, 12 Feb 2020 05:56:21 +0000 (13:56 +0800)]
net/i40e: relax barrier in Tx

To keep ordering of mixed accesses, rte_cio is sufficient.
The rte_io barrier inside the I40E_PCI_REG_WRITE is overkill.[1]

[1] http://inbox.dpdk.org/dev/CALBAE1M-ezVWCjqCZDBw+MMDEC4O9
qf0Kpn89EMdGDajepKoZQ@mail.gmail.com

Fixes: 4861cde46116 ("i40e: new poll mode driver")
Cc: stable@dpdk.org
Signed-off-by: Gavin Hu <gavin.hu@arm.com>
Reviewed-by: Jerin Jacob <jerinj@marvell.com>
4 years agonet/i40e: fix port start during config restore
Xiao Zhang [Fri, 28 Feb 2020 03:22:04 +0000 (11:22 +0800)]
net/i40e: fix port start during config restore

If configuring VF promiscuous mode is not supported, return '-ENOTSUP'
error code in '.promiscuous_enable/disable' dev_ops.

This is to fix the port start during config restore, where if
'.promiscuous_enable/disable' dev_ops exists and return any value other
than '-ENOTSUP', start will fail.

Same is done for '.allmulticast_enable/disable' dev_ops.

Fixes: ddc7cb0d9453 ("net/i40e: re-program promiscuous mode on VF interface")
Cc: stable@dpdk.org
Signed-off-by: Xiao Zhang <xiao.zhang@intel.com>
Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
4 years agonet/thunderx: support link up and down
Harman Kalra [Wed, 5 Feb 2020 09:26:33 +0000 (14:56 +0530)]
net/thunderx: support link up and down

Add support for .set_link_up/down() eth ops to bring
link up and down.

Signed-off-by: Harman Kalra <hkalra@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
4 years agonet/octeontx2: fix link information for loopback port
Ashish Gupta [Thu, 6 Feb 2020 12:30:58 +0000 (18:00 +0530)]
net/octeontx2: fix link information for loopback port

Loopback devices are exposed as ethdev device in octeontx2.
This patch fixes the link information for loopback ports.

Fixes: 38f566280abb ("net/octeontx2: add link stats operations")
Cc: stable@dpdk.org
Signed-off-by: Ashish Gupta <ashishg@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
4 years agonet/af_packet: support default MAC address change
Stephen Hemminger [Sat, 15 Feb 2020 19:20:12 +0000 (11:20 -0800)]
net/af_packet: support default MAC address change

Since packet is talking to a network device (probably tap)
it is possible for the device driver to change the hardware address
by making an ioctl to kernel.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>