Pablo de Lara [Tue, 5 Sep 2017 02:20:02 +0000 (03:20 +0100)]
crypto/openssl: do not append digest
When performing an authentication verification,
the PMD was using memory at the end of the input buffer,
to store temporarily the digest.
This operation requires the buffer to have enough
tailroom unnecessarily.
Instead, memory is allocated for each queue pair, to store
temporarily the digest generated by the driver, so it can
be compared with the one provided in the crypto operation,
without needing to touch the input buffer.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Fan Zhang <roy.fan.zhang@intel.com>
Pablo de Lara [Tue, 5 Sep 2017 02:20:01 +0000 (03:20 +0100)]
crypto/armv8: do not append digest
When performing an authentication verification,
the PMD was using memory at the end of the input buffer,
to store temporarily the digest.
This operation requires the buffer to have enough
tailroom unnecessarily.
Instead, memory is allocated for each queue pair, to store
temporarily the digest generated by the driver, so it can
be compared with the one provided in the crypto operation,
without needing to touch the input buffer.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com> Acked-by: Fan Zhang <roy.fan.zhang@intel.com>
Pablo de Lara [Tue, 5 Sep 2017 02:20:00 +0000 (03:20 +0100)]
crypto/aesni_gcm: do not append digest
When performing an authentication verification,
the PMD was using memory at the end of the input buffer,
to store temporarily the digest.
This operation requires the buffer to have enough
tailroom unnecessarily.
Instead, memory is allocated for each queue pair, to store
temporarily the digest generated by the driver, so it can
be compared with the one provided in the crypto operation,
without needing to touch the input buffer.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Fan Zhang <roy.fan.zhang@intel.com>
Pablo de Lara [Wed, 16 Aug 2017 02:41:51 +0000 (03:41 +0100)]
cryptodev: allocate driver structure statically
When register a crypto driver, a cryptodev driver
structure was being allocated, using malloc.
Since this call may fail, it is safer to allocate
this memory statically in each PMD, so driver registration
will never fail.
Coverity issue: 158645 Fixes: 7a364faef185 ("cryptodev: remove crypto device type enumeration") Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Reviewed-by: Kirill Rybalchenko <kirill.rybalchenko@intel.com>
Pablo de Lara [Mon, 14 Aug 2017 00:45:21 +0000 (01:45 +0100)]
crypto/aesni_gcm: fix zero data operation
When data length passed to the PMD was zero,
the PMD was trying to get more data from a non-existent
next segment.
Fixes: 9c2a5775c028 ("crypto/aesni_gcm: migrate from MB library to ISA-L") Cc: stable@dpdk.org Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Reviewed-by: Kirill Rybalchenko <kirill.rybalchenko@intel.com>
crypto/openssl: update key and algo during session init
Key and algo are added in the openssl ctx during
session initialization instead of adding it for
each packet, since it remains constant for that session,
improving the performance.
Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com> Reviewed-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Original commit assumes there is 1:1 mapping between physical device and
ethdev port, so that device name can be used per port instead of ethdev
name field.
But one physical device may have multiple ethdev ports and each port
needs its own unique name.
One issue reported here:
http://dpdk.org/ml/archives/users/2017-September/002484.html
So reverting back the commit to continue using ethdev name field per
port.
Fixes: a1e7c17555e8 ("ethdev: use device name from device structure") Cc: stable@dpdk.org Reported-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com> Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
Wei Zhao [Wed, 11 Oct 2017 08:55:33 +0000 (16:55 +0800)]
app/testpmd: add commands for RSS queue region
This patch add a API configuration of queue region in rss.
It can parse the parameters of region index, queue number,
queue start index, user priority, traffic classes and so on.
According to commands from command line, it will call i40e
private API and start the process of set or flush queue region
configure. As this feature is specific for i40e, so private API
will be used. Aslo add a document for these new commands.
Queue region only support PF by now, so this document is
only for configuration of queue region on PF port.
Wei Zhao [Wed, 11 Oct 2017 08:55:32 +0000 (16:55 +0800)]
net/i40e: support queue region set and flush
This feature enable queue regions configuration for RSS in PF,
so that different traffic classes or different packet
classification types can be separated to different queues in
different queue regions.This patch can set queue region range,
it include queue number in a region and the index of first queue.
This patch enable mapping between different priorities (UP) and
different traffic classes.It also enable mapping between a region
index and a sepcific flowtype(PCTYPE).It also provide the solution
of flush all configuration about queue region the above described.
Wei Dai [Thu, 28 Sep 2017 02:28:33 +0000 (10:28 +0800)]
net/ixgbe: fix VFIO interrupt mapping in VF
When a VF port is bound to VFIO-PIC, only miscellaneous interrupt
is mapped to VFIO vector 0 in eth_ixgbevf_dev_init( ).
In ixgbevf_dev_start(), if previous VFIO interrupt mapping set in
eth_ixgbevf_dev_init( ) is not cleard, it will fail when calling
rte_intr_enable( ) tries to map Rx queue interrupt to other VFIO
vectors. This patch clears the VFIO interrupt mappings before
setting both miscellaneous and Rx queue interrupt mappings again
to avoid failure.
Fixes: 77234603fba0 ("net/ixgbe: support VF mailbox interrupt for link up/down") Cc: stable@dpdk.org Signed-off-by: Wei Dai <wei.dai@intel.com> Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com> Tested-by: Jianwei Ma <jianwei.ma@intel.com>
Wei Dai [Wed, 20 Sep 2017 10:18:13 +0000 (18:18 +0800)]
net/ixgbe: fix Rx queue interrupt mapping in VF
When a VF port is bound to VFIO-PCI, miscellaneous interrupt is
mapped to MSI-X vector 0 and Rx queues interrupt are mapped to
other vectors in vfio_enable_msix( ). To simplify implementation,
all VFIO-PCI bound ixgbe VF Rx queue interrupts can be mapped in
vector 1. And as current igb_uio only support only one vector,
ixgbe VF PMD should use vector 0 for igb_uio and vector 1 for
VFIO-PCI. Without this patch, VF Rx queue interrupt is mapped
to vector 0 in register settings and mapped to VFIO vector 1
in vfio_enable_msix( ), and then all Rx queue interrupts will
be missed.
Fixes: b13bfab4cdbe ("eal: reserve VFIO vector zero for misc interrupt") Cc: stable@dpdk.org Signed-off-by: Wei Dai <wei.dai@intel.com> Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com> Tested-by: Jianwei Ma <jianwei.ma@intel.com>
Matan Azrad [Tue, 10 Oct 2017 20:20:18 +0000 (20:20 +0000)]
ethdev: add return value to stats get dev op
The stats_get dev op API doesn't include return value, so PMD cannot
return an error in case of failure at stats getting process time.
Since PCI devices can be removed and there is a time between the
physical removal to the RMV interrupt, the user may get invalid stats
without any indication.
This patch changes the stats_get API return value to be int instead of
void.
All the net PMDs stats_get dev ops are adjusted by this patch.
Ajit Khaparde [Tue, 10 Oct 2017 14:23:00 +0000 (09:23 -0500)]
net/bnxt: fix cleanup if a filter allocation fails
We are not checking if a filter allocation succeeded.
And we end up accessing a null pointer after that.
Also invalidate the fw_l2_filter_id to prevent unnecessary
HW access and hence HWRM command failures during exit.
Yongseok Koh [Tue, 10 Oct 2017 14:04:02 +0000 (07:04 -0700)]
net/mlx5: fix deadlock due to buffered slots in Rx SW ring
When replenishing Rx ring, there're always buffered slots reserved
between consumed entries and HW owned entries. These have to be filled
with fake mbufs to protect from possible overflow rather than
optimistically expecting successful replenishment which can cause
deadlock with small-sized queue.
Fixes: fc048bd52cb7 ("net/mlx5: fix overflow of Rx SW ring") Cc: stable@dpdk.org Reported-by: Martin Weiser <martin.weiser@allegro-packets.com> Signed-off-by: Yongseok Koh <yskoh@mellanox.com> Tested-by: Martin Weiser <martin.weiser@allegro-packets.com>
As it stands, the existing assignment to mbuf has no effect outside of
the function. Prior to this change, the mbuf argument would contain
an invalid address, but it would not be null. After this change, the
caller gets a null mbuf back.
This commit extends the testpmd application with new forwarding engine
that demonstrates the use of ethdev traffic management APIs and softnic
PMD for QoS traffic management.
In this mode, 5-level hierarchical tree of the QoS scheduler is built
with the help of ethdev TM APIs such as shaper profile add/delete,
shared shaper add/update, node add/delete, hierarchy commit, etc.
The hierarchical tree has following nodes; root node(x1, level 0),
subport node(x1, level 1), pipe node(x4096, level 2),
tc node(x16348, level 3), queue node(x65536, level 4).
During runtime, each received packet is first classified by mapping the
packet fields information to 5-tuples (HQoS subport, pipe, traffic class,
queue within traffic class, and color) and storing it in the packet mbuf
sched field. After classification, each packet is sent to softnic port
which prioritizes the transmission of the received packets, and
accordingly sends them on to the output interface.
To enable traffic management mode, following testpmd command is used;
Yongseok Koh [Mon, 9 Oct 2017 18:46:59 +0000 (11:46 -0700)]
net/mlx5: fix configuration of Rx CQE compression
With the upstream rdma-core, to enable Rx CQE compression,
mlx5dv_create_cq() in Direct Verbs has to be used instead of regular
Verbs call (ibv_create_cq()). And if the size of CQE is 128 bytes,
compression is supported only by certain devices. Thus, it has to be
decided by checking the capability bits.
Yongseok Koh [Mon, 9 Oct 2017 18:46:58 +0000 (11:46 -0700)]
net/mlx5: match Rx completion entry size to cacheline
The size of Rx completion entry should match the size of a cacheline.
This is already reflected in struct mlx5_cqe by adding 64bytes padding
if a cacheline is 128bytes. Some ARM CPUs have 128bytes cacheline.
Yongseok Koh [Mon, 9 Oct 2017 18:46:57 +0000 (11:46 -0700)]
net/mlx5: separate shareable vector functions
Considering more architecture (e.g. ARM and PowerPC) will be added for
vectorized Rx/Tx burst, all the shareable functions which don't use any
vector intrinsics need to be separated from architecture-dependent
functions. All the vector functions for x86 SSE are moved to a new
header file - mlx5_rxtx_vec_sse.h. And shareable common functions are
now in mlx5_rxtx_vec.c.
Yongseok Koh [Mon, 9 Oct 2017 18:46:54 +0000 (11:46 -0700)]
net/mlx5: cleanup memory barriers
Updating a consumer index to HW doesn't require a memory barrier in case
that there's no updated data to be posted to HW, but a compiler barrier
is sufficient. rte_wmb() is replaced with rte_io_wmb() when it makes
changes visible to HW, not other core.
net/mlx5: handle RSS hash configuration in RSS flow
Add RSS support according to the RSS configuration.
A special case is handled, when the pattern does not cover the RSS hash
configuration request such as:
flow create 0 ingress pattern eth / end actions rss queues 0 1 end / end
In such situation with the default configuration of testpmd RSS i.e. IP,
it should be converted to 3 Verbs flow to handle correctly the request:
1. IPv4 flow, an extra IPv4 wildcard specification needs to be added in
the conversion.
2. IPv6 flow, same as for IPv4.
3. Ethernet followed by any other protocol on which no RSS can be
performed and thus the traffic will be redirected to the first queue
of the user request.
The same kind of issue is handled if the RSS is performed only on UDPv4
or UDPv6 or TCPv*.
This does not handle a priority conflict which can occurs if the user
adds several colliding flow rules. Currently in the example above, the
request is already consuming 2 priorities (1 for IPv4/IPV6 matching
rule priority and one for Ethernet matching rule priority + 1).
Moves ibv_attr containing the specification of the flow from Verbs point
of view also with the verbs flow itself near the related verbs objects
making the flow.
This is also a preparation to handle correctly the RSS hash
configuration provided by the user, has multiple Verbs flows will be
necessary for a single generic flow.
struct mlx5_flow_parse was commonly used with the name "flow" confusing
sometimes the development. The variable name is replaced by parser to
reflect its use.
In case the pattern contains an RSS actions, the RSS configuration to
use is the one provided by the user. To make the correct conversion
from DPDK RSS hash fields to Verbs ones according to the users requests
the actions must be processed first.
net/mlx5: fully convert a flow to verbs in validate
Validation of flows is only making few verifications on the pattern, in
some situation the validate action could end by with success whereas the
pattern could not be converted correctly.
This brings this conversion verification part also to the validate.
drivers/net/mlx5/mlx5_rxq.c:606:6: error: comparison of constant 4
with expression of type 'enum hash_rxq_flow_type' is always true
[-Werror,-Wtautological-constant-out-of-range-compare]
i != (int)RTE_DIM((*priv->hash_rxqs)[0].special_flow);
~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Clang expects to have an index going upto special_flow size which is
defined by MLX5_MAX_SPECIAL_FLOWS and value is 4. Comparing to an
unrelated enum where index my be lower cause this compilation issue.
Hash Rx queue is an high level queue providing the RSS hash algorithm,
key and indirection table to spread the packets. Those objects can be
easily shared between several Verbs flows. This commit bring this
capability to the PMD.
Indirection table in verbs side resides in a list of final work queues
to spread the packets according to an higher level queue. This
indirection table can be shared among the hash Rx queues which points
to them.
Use the same design for DPDK queue as for Verbs queue for symmetry, this
also helps in fixing some issues like the DPDK release queue API which
is not expected to fail. With such design, the queue is released when
the reference counters reaches 0.
Use the same design for DPDK queue as for Verbs queue for symmetry, this
also helps in fixing some issues like the DPDK release queue API which
is not expected to fail. With such design, the queue is released when
the reference counters reaches 0.
net/mlx5: separate DPDK from verbs Tx queue objects
Move verbs object to their own functions to allocate/release them
independently from the DPDK queue. At the same time a reference counter
is added to help in issues detections when the queue is being release
but still in use somewhere else (flows for instance).
net/mlx5: separate DPDK from verbs Rx queue objects
Move verbs object to their own functions to allocate/release them
independently from the DPDK queue. At the same time a reference counter
is added to help in issues detections when the queue is being release
but still in use somewhere else (flows for instance).
This patch introduce the Memory region as a shared object where users
should get a reference to it by calling the priv_mr_get() or
priv_mr_new() to create the memory region. This last one will
register the memory pool in the kernel driver and retrieve the
associated memory region.
This should help to reduce the memory consumption cause by registering
multiple times the same memory pool.
The number of queues in DPDK does not means that the array of queue will be
totally filled, those information are uncorrelated. The number of queues
is provided in the port configuration whereas the array is filled by
calling tx/rx_queue_setup(). As this number of queue is not increased or
decrease according to tx/rx_queue_setup() or tx/rx_queue_release(), PMD
must consider a queue may not be initialised in some position of the array.
Reta update needs to stop/start the port but stopping the port does not
disable the polling functions which may end in a segfault if a core is
polling the queue while the control thread is modifying it.
This patch changes the sequences to an order where such situation cannot
happen.
Precompiler instructions #ifdef RTE_LIBRTE_I40E_PMD ... #endif
were not placed correctly, which caused number of
compilation errors if I40E PMD is disabled.
Mark Kavanagh [Mon, 9 Oct 2017 13:59:30 +0000 (14:59 +0100)]
net/bnxt: fix build
For gcc (GCC) 4.8.3 20140911 (Red Hat 4.8.3-7),
As of 5ef3b79fdfe6f, compilation of DPDK fails with the following
ERROR MESSAGE:
"bnxt_filter.c:960:117: error: ‘vnic’ may be used uninitialized in this
function [-Werror=maybe-uninitialized]".
Resolve this by initializing 'vnic' to NULL;
Fixes: 5ef3b79fdfe6 ("net/bnxt: support flow filter ops") Signed-off-by: Mark Kavanagh <mark.b.kavanagh@intel.com>
Phil Yang [Fri, 22 Sep 2017 06:21:32 +0000 (14:21 +0800)]
app/testpmd: fix quitting in container
In container, the process cannot be terminated by SIGINT/SIGTERM when
execute with 'stats-period' option.
Fixed by adding a flag to exit stats period loop after received either
SIGINT or SIGTERM.
Fixes: cfea1f3048d1 ("app/testpmd: print statistics periodically") Cc: stable@dpdk.org Signed-off-by: Phil Yang <phil.yang@arm.com> Acked-by: Jianbo Liu <jianbo.liu@linaro.org>
git.droids-corp.org / dpdk.git / log