Fan Zhang [Wed, 29 Mar 2017 16:31:32 +0000 (17:31 +0100)]
crypto/scheduler: add fail-over scheduling mode
Fail-over mode works with 2 slaves, primary slave and secondary slave.
In this mode, the scheduler will enqueue the incoming crypto op burst
to the primary slave. When one or more crypto ops are failed to be
enqueued, they then will be enqueued to the secondary slave.
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Fan Zhang [Thu, 30 Mar 2017 10:47:14 +0000 (11:47 +0100)]
crypto/scheduler: add packet size based mode
Packet-size based distribution mode is a scheduling mode works with 2
slaves, primary slave and secondary slave, and distribute the enqueued
crypto ops to them based on their data lengths. A crypto op will be
distributed to the primary slave if its data length equals or bigger
than the designated threshold, otherwise it will be handled by the
secondary slave.
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Pablo de Lara [Mon, 27 Mar 2017 11:26:03 +0000 (12:26 +0100)]
app/crypto-perf: display results in test runner
Instead of printing the test results in the destructor
function, print them just after each run.
This will be useful for after the commit that allows
a range of buffer and burst sizes.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
Akhil Goyal [Fri, 24 Mar 2017 09:29:21 +0000 (14:59 +0530)]
examples/ipsec-secgw: attach session-qp
adding support for attaching session to queue pairs.
This is required as underlying crypto driver may only
support limited number of sessions per queue pair
if max_nb_sessions_per_qp > 0, session should be
attached to a particular qp.
Akhil Goyal [Fri, 24 Mar 2017 09:29:20 +0000 (14:59 +0530)]
cryptodev: add API to associate session with queue pair
HW based crypto drivers may only support limited number of
sessions per queue pair. This requires support for attaching
sessions to specific queue pair. New APIs are introduced to
attach/detach a session with/from a particular queue pair.
These are optional APIs.
Application can call attach API after creating a session
and can call detach API before deleting a session.
Application needs to check if max_nb_sessions_per_qp > 0,
then it should call the attach API.
max_nb_sessions_per_qp = 0 means infinite sessions per qp
Pablo de Lara [Fri, 3 Mar 2017 16:10:49 +0000 (16:10 +0000)]
examples/dpdk-qat: remove app
Remove DPDK QAT sample app, in favour of the newer applications
that use the cryptodev library: ipsec-gw and l2fwd-crypto,
which has support for Intel QuickAssist devices.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Fan Zhang [Fri, 17 Feb 2017 12:01:01 +0000 (12:01 +0000)]
crypto/scheduler: improve slave configuration
Since the new device configuration API is updated, we can make use of
this feature to the crypto scheduler PMD to configure its slaves
automatically with the same configurations it got. As originally the
slaves have to be manually configured one by one, this patch should
help reducing the coding complexity.
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Fan Zhang [Thu, 2 Mar 2017 14:18:34 +0000 (14:18 +0000)]
crypto/scheduler: optimize crypto op ordering
This patch optimizes the crypto op ordering by replacing the
ordering method from using rte_reorder library to using rte_ring
to avoid unnecessary crypto op storing and recovering cost.
Fan Zhang [Thu, 2 Mar 2017 11:12:11 +0000 (11:12 +0000)]
crypto/scheduler: change enqueue and dequeue functions
This patch changes the enqueue and dequeue methods to cryptodev
scheduler PMD. Originally a 2-layer function call is carried out
upon enqueuing or dequeuing a burst of crypto ops. This patch
removes one layer to improve the performance.
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Tomasz Kulasek [Fri, 24 Feb 2017 15:26:00 +0000 (16:26 +0100)]
app/crypto-perf: fix uninitialized values for null ops
Some values are uninitialized for "cipher null" and "auth null"
operations. It may cause unpredictable results for some crypto pmd
drivers, or even segmentation fault.
This patch sets values for null operations to zero.
Fixes: f8be1786b1b8 ("app/crypto-perf: introduce performance test application") Signed-off-by: Tomasz Kulasek <tomaszx.kulasek@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Daniel Mrzyglod [Wed, 8 Mar 2017 17:03:25 +0000 (18:03 +0100)]
app/crypto-perf: avoid wrong operation type for AEAD algos
When somebody use bad --optype with aead algorithms
segmentation fault could happen.
Fixes: f8be1786b1b8 ("app/crypto-perf: introduce performance test application") Signed-off-by: Daniel Mrzyglod <danielx.t.mrzyglod@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Fixes invalid latency result when using the performance application and
hardware QAT PMD. It occurred when the number of processed packets was
higher then the size of the internal QAT PMD ring buffer and the buffer
was overflowed.
Fixed by correcting the registration of the enqueued packets and freeing
memory space for not enqueued packets.
Fixes: f8be1786b1b8 ("app/crypto-perf: introduce performance test application") Signed-off-by: Slawomir Mrozowicz <slawomirx.mrozowicz@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Pablo de Lara [Thu, 23 Feb 2017 14:02:55 +0000 (14:02 +0000)]
cryptodev: enable BPI for Cablelabs DOCSIS security spec
Extend the DPDK cryptodev API to enable processing of packets according
to the Baseline Privacy Interface Plus (BPI+) Specification described in
the security specification of the Cablelabs Data-over-Cable Service
Interface Specification (DOCSIS).
Brief summary of BPI+ symmetric cryptography requirements:
BPI+ cryptography uses a block cipher (AES-CBC/DES-CBC) to encrypt/decrypt
all the whole blocks in the packet. However the data length is not always
a block-multiple, so where there is a final block less than the full block
size this residual block requires special handling using AES-CFB/DES-CFB
mode. Similar special handling is specified where there is only one block,
smaller than the block size for the cipher. See spec for further details.
https://apps.cablelabs.com/specification/docsis-3-1-security-specification/
Two new elements are added to the enum rte_crypto_cipher_algorithm.
Note elements of this enum are actually a combination of an algorithm (AES,
3DES, etc) and mode (CBC, CTR, etc). The new DOCSISBPI mode is used to
convey to the PMD that the mode applied should be the specific combination
of CBC and CFB required by the DOCSIS Baseline Privacy Plus Spec.
Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Deepak Kumar Jain <deepak.k.jain@intel.com>
Pablo de Lara [Mon, 27 Feb 2017 14:38:46 +0000 (14:38 +0000)]
examples/l2fwd-crypto: use cryptodev algorithm parser
L2fwd-crypto app was creating an array of strings for the
supported algorithms, which was different from the strings
that are now in cryptodev.
Use the new API in cryptodev to parse the string from the user,
to get the algorithm enum, instead, so it is not necessary to add
a new supported algorithm in the cryptodev library and this app.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Fiona Trahe <fiona.trahe@intel.com> Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Pablo de Lara [Mon, 27 Feb 2017 14:38:45 +0000 (14:38 +0000)]
app/crypto-perf: use cryptodev algorithm parser
Instead of going through the array of supported algorithms
in the app, to get the algorithm enum, use the new API in
cryptodev to parse this string, so it is not necessary to add
a new supported algorithm in the cryptodev library and this app.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Fiona Trahe <fiona.trahe@intel.com> Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Pablo de Lara [Mon, 27 Feb 2017 14:38:44 +0000 (14:38 +0000)]
cryptodev: add algorithm string parsers
Adds functions to get the cipher/authentication
algorithm enums, given a string. This is useful for applications
which gets the algorithm required from the user, to have a common
string-enum mapping.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Fiona Trahe <fiona.trahe@intel.com> Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
There is a compilation error in the following case:
make config T=x86_64-native-linuxapp-gcc
sed -i 's,CONFIG_RTE_LIBRTE_CRYPTODEV=y,CONFIG_RTE_LIBRTE_CRYPTODEV=n,' \
build/.config
sed -i 's,CONFIG_RTE_APP_CRYPTO_PERF=y,CONFIG_RTE_APP_CRYPTO_PERF=n,' \
build/.config
make -j4
make -j4 test-build
test_eventdev.c:39:27: fatal error:
rte_cryptodev.h: No such file or directory
#include <rte_cryptodev.h>
^
Removing the include induces a new error, (missing declaration of
rte_eal_vdev_init()), which is fixed by including rte_dev.h.
Fixes: f8f9d233ea0e ("test/eventdev: add unit tests") Signed-off-by: Olivier Matz <olivier.matz@6wind.com> Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Ed Czeck [Thu, 6 Apr 2017 13:59:47 +0000 (09:59 -0400)]
ring: fix C++ cast error
build error:
include/rte_ring.h:459:22: error: invalid conversion from ‘void*’
to ‘void**’ [-fpermissive]
ENQUEUE_PTRS(r, &r[1], prod_head, obj_table, n, void *);
Implicit casts of void* to void** are considered warnings in some
compilers. E.g. g++ version 5.8. Cast directly to object types
Fixes: a6619414 ("ring: make struct and macros type agnostic") Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
I get the following error when linking the test application:
build/lib/librte_pmd_thunderx_nicvf.a(nicvf_hw.o):
In function `nicvf_qsize_regbit':
drivers/net/thunderx/base/nicvf_hw.c:451: undefined reference to `log2'
build/lib/librte_pmd_thunderx_nicvf.a(nicvf_hw.o):
In function `nicvf_rss_reta_update':
drivers/net/thunderx/base/nicvf_hw.c:804: undefined reference to `log2'
build/lib/librte_pmd_thunderx_nicvf.a(nicvf_hw.o):
In function `nicvf_rss_reta_query':
drivers/net/thunderx/base/nicvf_hw.c:825: undefined reference to `log2'
While I don't know why it does not happen for a default build, the error
can be explained. The link command line is:
gcc -o test ... *.o ... -Wl,-lm ... -Wl,-lrte_pmd_thunderx_nicvf ...
rte_pmd_thunderx_nicvf needs the math library, and it should be
added after. This is not the case because the test application also
adds the math library.
The makefile already filters the libraries, but it keeps the first
occurrence of the lib. Instead, the last one should be kept.
Fixes: edf4d331dcdb ("mk: eliminate duplicates from libraries list") Cc: stable@dpdk.org Signed-off-by: Olivier Matz <olivier.matz@6wind.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
On my system, the version of the compiler is not properly retrieved,
resulting in strange logs when building the dpdk:
/bin/sh: line 0: test: too many arguments
This happens when mk/toolchain/clang/rte.toolchain-compat.mk is included
from a directory that use gcc to build (ex: kernel modules). In that
case, the CLANG_VERSION variable contains spaces that breaks some shell
calls to the test program.
The error is because the output of "gcc -v" on my system contains 2 lines
that matches the "version" string:
Thomas Monjalon [Thu, 6 Apr 2017 10:44:33 +0000 (12:44 +0200)]
mk: build examples list before each doxygen run
The file examples.dox was not re-generated when a file
is added or removed from examples/.
It is now removed on clean operation.
The ordering of operations (clean before generation) is also
better defined.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
The macro RTE_PMD_TAP_MAX_QUEUES was defined twice.
On machines with kernel < 3.8, IFF_MULTI_QUEUE didn't exist, and thus
both definitions used different values.
Fixes: cf5643661161 ("net/tap: move private elements to external header") Signed-off-by: Pascal Mazon <pascal.mazon@6wind.com>
Since patch "mbuf: structure reorganization" the compiler complains
sometimes (in some conditions):
.../drivers/net/mlx5/mlx5_rxtx.c: In function ‘mlx5_rx_burst’:
.../drivers/net/mlx5/mlx5_rxtx.c:2082:17: error: ‘len’ may be used
uninitialized in this function [-Werror=maybe-uninitialized]
len is not initialised as it will be at the first segment of a received
packet, but it remains hard for the compiler to determine it.
APIs for selecting the architecure specific implementation and computing
the crc (16-bit and 32-bit CRCs) are added. For CRCs calculation, scalar
as well as x86 intrinsic(sse4.2) versions are implemented.
The scalar version is based on generic Look-Up Table(LUT) algorithm,
while x86 intrinsic version uses carry-less multiplication for
fast CRC computation.
Signed-off-by: Jasvinder Singh <jasvinder.singh@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Gaetan Rivet [Fri, 31 Mar 2017 12:04:37 +0000 (14:04 +0200)]
ethdev: expose device states
The hotplug API introduced multiple states for a device with possible
values defined internally, while the related field in struct rte_eth_dev
was made public.
Exposing those states improves consistency because applications have to
deal with the device list directly.
"DEV_DETACHED" is renamed "RTE_ETH_DEV_UNUSED" to better reflect that
the emptiness of a slot is not necessarily the result of detaching a
device.
build error:
.../drivers/event/octeontx/ssovf_worker.c(212):
error #592: variable "get_work0" is used before its value is set
RTE_SET_USED(get_work0);
^
.../drivers/event/octeontx/ssovf_worker.c(213):
error #592: variable "get_work1" is used before its value is set
RTE_SET_USED(get_work1);
^
For x86 these variables set but not used, move macros below
where values assigned.
build error:
In file included from .../lib/librte_ring/rte_ring.c(90):
.../lib/librte_ring/rte_ring.h(162):
error #1366: a reduction in alignment without the "packed" attribute
is ignored
} __rte_cache_aligned;
^
Alignment attribute moved to first element of the struct
Fixes: a6619414e0a9 ("ring: make struct and macros type agnostic") Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Reshma Pattan [Thu, 30 Mar 2017 21:01:01 +0000 (22:01 +0100)]
latency: added new library for latency stats
Add a library designed to calculate latency statistics and report them
to the application when queried. The library measures minimum, average and
maximum latencies, and jitter in nano seconds. The current implementation
supports global latency stats, i.e. per application stats.
Signed-off-by: Reshma Pattan <reshma.pattan@intel.com> Signed-off-by: Remy Horton <remy.horton@intel.com> Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Remy Horton [Thu, 30 Mar 2017 21:00:59 +0000 (22:00 +0100)]
bitrate: add bitrate statistics library
This patch adds a library that calculates peak and average data-rate
statistics. For ethernet devices. These statistics are reported using
the metrics library.
Remy Horton [Thu, 30 Mar 2017 21:00:57 +0000 (22:00 +0100)]
metrics: add information metrics library
This patch adds a new information metrics library. This Metrics
library implements a mechanism by which producers can publish
numeric information for later querying by consumers. Metrics
themselves are statistics that are not generated by PMDs, and
hence are not reported via ethdev extended statistics.
Metric information is populated using a push model, where
producers update the values contained within the metric
library by calling an update function on the relevant metrics.
Consumers receive metric information by querying the central
metric data, which is held in shared memory.
Thomas Monjalon [Wed, 5 Apr 2017 12:18:01 +0000 (14:18 +0200)]
doc: remove old deprecation notices
The ring and distributor reworks are done.
Fixes: a6619414e0a9 ("ring: make struct and macros type agnostic") Fixes: 775003ad2f96 ("distributor: add new burst-capable library") Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Deprecate the following functions:
- rte_set_log_level(), replaced by rte_log_set_global_level()
- rte_get_log_level(), replaced by rte_log_get_global_level()
- rte_set_log_type(), replaced by rte_log_set_level()
- rte_get_log_type(), replaced by rte_log_get_level()
The new functions provide a better control of the per-type log level,
and have a better name prefix (rte_log_).
Change the size of m->port and m->nb_segs to 16 bits. It is now possible
to reference a port identifier larger than 256 and have a mbuf chain
larger than 256 segments.
Jerin Jacob [Tue, 4 Apr 2017 16:28:04 +0000 (18:28 +0200)]
mbuf: make rearm data address naturally aligned
To avoid multiple stores on fast path, Ethernet drivers
aggregate the writes to data_off, refcnt, nb_segs and port
to an uint64_t data and write the data in one shot
with uint64_t* at &mbuf->rearm_data address.
Some of the non-IA platforms have store operation overhead
if the store address is not naturally aligned.This patch
fixes the performance issue on those targets.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com> Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
drivers/net: do not touch mbuf next or nb segs on Rx
Now that the m->next pointer and m->nb_segs is expected to be set (to
NULL and 1 respectively) after a mempool_get(), we can avoid to write them
in the Rx functions of drivers.
Only some drivers are patched, it's not an exhaustive patch. It gives
the idea to do the same in other drivers.
Set the value of m->refcnt to 1, m->nb_segs to 1 and m->next
to NULL when the mbuf is stored inside the mempool (unused).
This is done in rte_pktmbuf_prefree_seg(), before freeing or
recycling a mbuf.
Before this patch, the value of m->refcnt was expected to be 0
while in pool.
The objectives are:
- to avoid drivers to set m->next to NULL in the early Rx path, since
this field is in the second 64B of the mbuf and its access could
trigger a cache miss
- rationalize the behavior of raw_alloc/raw_free: one is now the
symmetric of the other, and refcnt is never changed in these functions.
To optimize the freeing of the segments, we try try to only update
m->refcnt, m->next, and m->nb_segs when it's required (idea from
Konstantin Ananyev <konstantin.ananyev@intel.com>).
Jerin Jacob [Fri, 3 Mar 2017 17:28:21 +0000 (22:58 +0530)]
doc: add OCTEONTX ssovf details
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com> Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Acked-by: Harry van Haaren <harry.van.haaren@intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
Jerin Jacob [Fri, 3 Mar 2017 17:28:20 +0000 (22:58 +0530)]
test/eventdev: add remaining tests based on existing helpers
Add the following tests based existing helper functions
- Queue based producer-consumer ingress order test
- Run existing queue and flow based ordering test in dequeue timeout
mode
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com> Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Jerin Jacob [Fri, 3 Mar 2017 17:28:01 +0000 (22:58 +0530)]
event/octeontx: support worker dequeue
If device is configured with RTE_EVENT_DEV_CFG_PER_DEQUEUE_TIMEOUT
configuration then use different fast path dequeue handler to wait till
requested amount of nanosecond if the event is not available.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com> Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Jerin Jacob [Fri, 3 Mar 2017 17:27:57 +0000 (22:57 +0530)]
event/octeontx: support dequeue timeout tick conversion
SSO co-processor runs at a different frequency than core clock.
Request PF to convert the ns to SSO get_work timeout period.
On dequeue, If device is configured with
RTE_EVENT_DEV_CFG_PER_DEQUEUE_TIMEOUT configuration then
use different fast path dequeue handler to wait till requested
amount of nanosecond if the event is not available.
Jerin Jacob [Fri, 3 Mar 2017 17:27:55 +0000 (22:57 +0530)]
event/octeontx: support event ports
Add in the data-structures for the ports used by workers
to sent events to/from the HW scheduler. Also add a
function to release the resource allocated in setup
git.droids-corp.org / dpdk.git / log