The mlx5 PMD uses Netlink socket to communicate with Infiniband
devices kernel drivers to perform some control and setup operations.
The kernel drivers send the information back to the user mode
with Netlink messages which are processed in libnl callback routine.
This routine perform reply message (or set of messages) processing
and returned the processing result in ibindex field of provided
context structure (of mlx5_nl_ifindex_data type). The zero ibindex
value meant an error of reply message processing. It was found in
some configurations the zero is valid value for ibindex and error
was wrongly raised. To avoid this the new flags field is provided
in context structure, attribute processing flags are introduced
and these flags are used to decide whether no error occurred and
valid queried values are returned.
Fixes: e505508a3858 ("net/mlx5: modify get ifindex routine for multiport IB") Cc: stable@dpdk.org Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
This patch implements ethdev operations get_module_info and
get_module_eeprom, to support ethtool commands ETHTOOL_GMODULEINFO
and ETHTOOL_GMODULEEEPROM.
New functions mlx5_get_module_info() and mlx5_get_module_eeprom()
added in mlx5_ethdev.c.
net/mlx5: support modify VLAN ID on new VLAN header
This commit adds support for modifying the VLAN ID (VID) field
in an about-to-be-pushed VLAN header.
This feature can only modify the VID field of a new VLAN header yet
to be pushed. It does not support modifying an existing or already
pushed VLAN headers.
net/mlx5: support modifying VLAN priority on VLAN header
This commit adds support for modifying the VLAN priority (PCP) field
in about-to-be-pushed VLAN header.
This feature can only modify the PCP field of a new VLAN header yet
to be pushed. It does not support modifying an existing or already
pushed VLAN headers.
This commit adds support for RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN using
direct verbs flow rules.
If present in the flow, The VLAN default values are taken from the
VLAN item configuration.
In this commit only the VLAN TPID value can be set since VLAN
modification actions are not supported yet.
If VLAN tag insertion transmit offload is engaged
(DEV_TX_OFFLOAD_VLAN_INSERT in tx queue configuration is set)
the transmit descriptor may be built with wrong format, due to
packet length is not adjusted. Also, the ring buffer wrap up
is not handled correctly.
Akhil Goyal [Fri, 4 Oct 2019 11:26:59 +0000 (16:56 +0530)]
event/dpaa: support crypto adapter
event dpaa device support both ethernet as well as
crypto queues to be attached to it. eth_rx_adapter
provide infrastructure to attach ethernet queues and
crypto_adapter provide support for crypto queues.
This patch add support for dpaa_eventdev to attach
dpaa_sec queues.
Akhil Goyal [Fri, 4 Oct 2019 11:26:58 +0000 (16:56 +0530)]
crypto/dpaa_sec: support event crypto adapter
dpaa_sec hw queues can be attached to a hw dpaa event
device and the application can configure the event
crypto adapter to access the dpaa_sec packets using
hardware events.
This patch defines APIs which can be used by the
dpaa event device to attach/detach dpaa_sec queues.
Anoob Joseph [Fri, 4 Oct 2019 09:44:35 +0000 (15:14 +0530)]
crypto/octeontx: fix global log variable definition
'cpt_logtype' & 'otx_cryptodev_driver_id' global variables are defined
in a header file which was causing multiple definitions of the
variables. Fixed it by moving the required vars to the .c file and
introducing a new macro so the CPT_LOG macros in common/cpt would use
the associated PMD log var.
Issue has been detected by '-fno-common' gcc flag.
Xiao Wang [Wed, 4 Sep 2019 13:21:51 +0000 (21:21 +0800)]
crypto/virtio: allocate private data by NUMA affinity
It's better to allocate device private data on the same NUMA node with
device, rather than with the main thread. This helps avoid cross-NUMA
access for worker thread.
Signed-off-by: Xiao Wang <xiao.w.wang@intel.com> Reviewed-by: Jay Zhou <jianjay.zhou@huawei.com>
To allow shared library builds of e.g. test-bbdev app, we need to export
the configure function. Since this needs to be exported as experimental by
default, we update the header file to add the experimental tag there too.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Now that all driver names follow a consistent pattern, remove the override
of the name in each driver which adds the prefix. Instead we can just add
the prefix at a higher level.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
baseband/fpga_lte_fec: align naming to other bbdevs
The fpga_lte_fec is the only bbdev driver that does not use bbdev in the
name, so modify it to keep consistency with the other bbdev drivers. This
will then allow later simplification due to all drivers using the same
basic naming format.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
For baseband drivers, the macros used to indicate the presence of a
particular driver were subtly different from that used in make. The make
values hand "PMD" before the individual driver name, while in meson it came
afterwards. Update meson to put the "PMD" part first.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Adam Dybkowski [Tue, 8 Oct 2019 12:44:33 +0000 (14:44 +0200)]
crypto/qat: handle Single Pass Crypto Requests on GEN3
This patch improves the performance of AES GCM by using
the Single Pass Crypto Request functionality when running
on GEN3 QAT. Falls back to the classic 2-pass mode on older
hardware.
Signed-off-by: Adam Dybkowski <adamx.dybkowski@intel.com> Acked-by: Fiona Trahe <fiona.trahe@intel.com>
Adam Dybkowski [Tue, 8 Oct 2019 12:44:31 +0000 (14:44 +0200)]
test/crypto: add more AES-GCM cases for QAT
This patch adds 256-bit AES GCM tests for QAT PMD
(which already existed for AESNI and OpenSSL) and also adds
a number of negative unit tests for AES GCM for QAT PMD, in order
to verify authenticated encryption and decryption with modified data.
Signed-off-by: Adam Dybkowski <adamx.dybkowski@intel.com> Acked-by: Fiona Trahe <fiona.trahe@intel.com>
Hemant Agrawal [Tue, 1 Oct 2019 11:41:18 +0000 (17:11 +0530)]
test/crypto: change failure condition check
In some of the cases, the test is looking for a specific
failure returned from the CryptoDev. Not all cryptodev
support returning specific errors.
This patch changes such checks to NOT-SUCCESS
examples/ipsec-secgw: update required Scapy version
Update Scapy version requirement from 2.4.3rc1 to 2.4.3, which has been
used because 2.4.2 had a bug which made this version unable to install.
Accept future versions of Scapy too.
Signed-off-by: Marcin Smoczynski <marcinx.smoczynski@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Test cases for PDCP with scattered input and output
buffers are added for both inplace and out of place scenarios.
1. input SG - output non SG
2. input and output both SG and inplace buffers
3. input and output both SG with different segment sizes
4. input SG and output non-SG
This patch add test cases for user plane PDCP
5/7/12/15/18 bit SN test cases for decapsulation of
cipher only and 12/18 bit for cipher + integrity
for all algo combinations.
Test vectors are reused from the encap cases
by swapping the input and output data.
The case are added to DPAA_SEC and DPAA2_SEC
testsuite.
test/crypto: add PDCP U-Plane encap with integrity cases
PDCP User/Data plane can optionally support integrity
along with confidentiality.
This patch add test cases and test vectors for all
supported algos combos for cipher+integrity for 12
and 18bit SN.
The test cases are added in DPAA_SEC and DPAA2_SEC
test suite as a reference.
PDCP User/Data plane can support 5/7/12/15/18 bit
sequence number. This patch add test cases and test
vectors for all supported algos for cipher only.
(NULL, AES, SNOW, ZUC ciphers)
The test cases are added in DPAA_SEC and DPAA2_SEC
test suite as a reference.
PDCP control plane can support 5 bit and 12 bit
sequence number. This patch add test cases and test
vectors for all supported algo combinations of cipher+integrity
(NULL+NULL, NULL+AES, NULL+SNOW, NULL+ZUC,
AES+NULL, AES+SNOW, AES+AES, AES+ZUC,
SNOW+NULL, SNOW+AES, SNOW+SNOW, SNOW+ZUC,
ZUC+NULL, ZUC+AES, ZUC+SNOW, ZUC+ZUC).
The test cases are added in DPAA_SEC and DPAA2_SEC
test suite as a reference.
One issue caught by Coverity 343408
*deref_parm: Directly dereferencing parameter val->val.
In writeback_tdes_hex_str(), tmp_val is initialised to null.
tmp_val.val is updated only if keys are found.
If keys are not found,it doesn't fails but continues
to invoke writeback_hex_str(),where val->val is accessed
without null check.
The fix is to return the error,
if keys are not found in writeback_tdes_hex_str().
In cases where single cryptodev is used by multiple cores
using multiple queues, there will be contention for mempool
resources and may eventually get exhausted.
Basically, mempool should be defined per core.
Now since qp is used per core, mempools are defined in qp setup.
This patch add support for ZUC Encryption and ZUC Authentication.
Before passing to CAAM, the 16-byte ZUCA IV is converted to 8-byte
format which consists of 38-bits of count||bearer|direction.
PDCP descriptors in some cases internally use commands which overwrite
memory with extra '0s' if write-safe is kept enabled. This breaks
correct functional behavior of PDCP apis and they in many cases give
incorrect crypto output. There we disable 'write-safe' bit in FLC for
PDCP cases. If there is a performance drop, then write-safe would be
enabled selectively through a separate patch.
crypto/dpaa2_sec/hw: update 12-bit SN desc for NULL auth
For sec era 8, NULL auth using protocol command does not add
4 bytes of null MAC-I and treat NULL integrity as no integrity which
is not correct.
Hence converting this particular case of null integrity on 12b SN
on SEC ERA 8 from protocol offload to non-protocol offload case.
Adding minimal support for CAAM HW era 10 (used in LX2)
Primary changes are:
1. increased shard desc length form 6 bit to 7 bits
2. support for several PDCP operations as PROTOCOL offload.
Per packet HFN override is supported in NXP PMDs
(dpaa2_sec and dpaa_sec). DPOVRD register can be
updated with the per packet value if it is enabled
in session configuration. The value is read from
the IV offset.
HFN can be given as a per packet value also.
As we do not have IV in case of PDCP, and HFN is
used to generate IV. IV field can be used to get the
per packet HFN while enq/deq
If hfn_ovrd field in pdcp_xform is set,
application is expected to set the per packet HFN
in place of IV. Driver will extract the HFN and perform
operations accordingly.
Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Added support for 12-bit c-plane. We implement it using 'u-plane for RN'
protocol descriptors. This is because 'c-plane' protocol descriptors
assume 5-bit sequence numbers. Since the crypto processing remains same
irrespective of c-plane or u-plane, we choose 'u-plane for RN' protocol
descriptors to implement 12-bit c-plane. 'U-plane for RN' protocol
descriptors support both confidentiality and integrity (required for
c-plane) for 7/12/15 bit sequence numbers.
For little endian platforms, incorrect IV is generated if MOVE command
is used in pdcp non-proto descriptors. This is because MOVE command
treats data as word. We changed MOVE to MOVEB since we require data to
be treated as byte array. The change works on both ls1046, ls2088.
Adam Dybkowski [Tue, 24 Sep 2019 14:10:33 +0000 (16:10 +0200)]
app/compress-perf: add --external-mbufs option
This patch adds new performance measurement option --external-mbufs
that allocates and uses memzones as external buffers instead of
putting the data directly inside mbufs.
Signed-off-by: Adam Dybkowski <adamx.dybkowski@intel.com> Acked-by: Fiona Trahe <fiona.trahe@intel.com>
When sending an encrypted packet which size after encapsulation exceeds
MTU, ipsec-secgw application tries to fragment it. If --reassemble
option has not been set it results with a segmantation fault, because
fragmentation buckets have not been initialized.
Fix crashing by adding extra check if --ressemble option has not been
set and packet exceeds MTU after encapsulation - drop it.
Fixes: b01d1cd213 ("examples/ipsec-secgw: support fragmentation and reassembly") Cc: stable@dpdk.org Signed-off-by: Marcin Smoczynski <marcinx.smoczynski@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
One issue caught by Coverity 344984
*overrun-local: Overrunning array cperf_test_type_strs of 2 8-byte elements
The array cperf_test_type_strs defined in app/test-compress-perf conflicts
with the same name being defined in app/test-crypto-perf. Due to which
coverity reports error.
The fix is to rename "cperf_test_type_strs" into "comp_perf_test_type_strs"
in app/test-compress-perf to avoid name confusion.
examples/ipsec-secgw: support IPv6 tunnel for lookaside proto
IPv6 tunnels are already supported in case of inline and
lookaside none cases. In case of protocol offload, the details
for IPv6 header need to be added in session configuration
for security session create.
Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
crypto/dpaa2_sec: support IPv6 tunnel for protocol offload
outer IP header is formed at the time of session initialization
using the ipsec xform. This outer IP header will be appended by
hardware for each packet.
crypto/dpaa_sec: support IPv6 tunnel for protocol offload
outer IP header is formed at the time of session initialization
using the ipsec xform. This outer IP header will be appended by
hardware for each packet.
Like for Ethernet ports, the OCTEON TX crypto engines must
first be unbound from their kernel module, then rebound to
vfio-pci, before being used in DPDK.
As this capability is detected at runtime by dpdk-pmdinfo,
add the info in the PMD registering directives.
Then an external script can be used for bind and unbind.
dpaa_sec needs translations between physical and virtual addresses.
V to P translation is relatively fast, as memory is managed in
contiguous segments.
The result of each V to P translation is used to update the DPAA iova
table, which should be updated by a Mem event callback, but is not.
Then the DPAA iova table has entries for all needed memory ranges.
With this patch, dpaa_mem_ptov will always use dpaax_iova_table_get_va,
which ensures optimal performance.
Pablo de Lara [Thu, 5 Sep 2019 14:45:06 +0000 (15:45 +0100)]
crypto/aesni_mb: update chain order for AES-CCM
Up to version 0.52 of the IPSec Multi buffer library,
the chain order for AES-CCM was CIPHER_HASH when encrypting.
However, after this version, the order has been reversed in the library
since, when encrypting, hashing is done first and then ciphering.
Therefore, order is changed to be compatible with newer versions
of the library.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Ruifeng Wang [Thu, 5 Sep 2019 08:10:44 +0000 (16:10 +0800)]
test/crypto: fix armv8 virtual PMD autotest
armv8 cryptodev virtual PMD autotest failed with output:
CRYPTODEV: [CRYPTODEV_NAME_ARMV8_CRYPTO_PMD]
armv8_crypto_pmd_qp_create_processed_ops_ring() line 210:
Unable to reuse existing ring armv8_crypto_pmd_0_qp_0 for processed ops
CRYPTODEV: rte_cryptodev_stop() line 942:
Device with dev_id=0 already stopped
+ TestCase [ 0] : test_AES_chain_armv8_all failed
This is due to the ring size used in ut_setup is bigger than that used
in testsuite_setup.
Fix this issue by enlarge ring size used in testsuite_setup accordingly.
Fixes: 24054e3640a2 ("test/crypto: use separate session mempools") Cc: stable@dpdk.org Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Adam Dybkowski [Tue, 6 Aug 2019 10:37:38 +0000 (12:37 +0200)]
app/compress-perf: fix floating point exception
This patch fixes the floating point exception that happened
when the number of cores to be used during the benchmark
was zero. After the fix such situation is detected, the error
message is printed and the benchmark application exits.
Fixes: 424dd6c8c1a8 ("app/compress-perf: add weak functions for multicore test") Cc: stable@dpdk.org Signed-off-by: Adam Dybkowski <adamx.dybkowski@intel.com> Acked-by: Artur Trybula <arturx.trybula@intel.com>
Adam Dybkowski [Tue, 6 Aug 2019 09:40:53 +0000 (11:40 +0200)]
app/compress-perf: fix memory deallocation
This patch fixes the memory deallocation issue which happened
after unsuccessful allocation (e.g. due to the out of memory)
and produced the segmentation fault.
Fixes: 424dd6c8c1 ("app/compress-perf: add weak functions for multicore test") Cc: stable@dpdk.org Signed-off-by: Adam Dybkowski <adamx.dybkowski@intel.com> Acked-by: Artur Trybula <arturx.trybula@intel.com>
Just open the sysfs file and handle failure, rather than using access().
This eliminates Coverity warnings about TOCTOU
"time of check versus time of use"; although for this sysfs file that is
not really an issue anyway.
Coverity issue: 347276 Fixes: 54a328f552ff ("bus/pci: forbid IOVA mode if IOMMU address width too small") Cc: stable@dpdk.org Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Reviewed-by: David Marchand <david.marchand@redhat.com>
git.droids-corp.org / dpdk.git / log