Optimize service loop so that the starting point is the lowest-indexed
service mapped to the lcore in question, and terminate the loop at the
highest-indexed service.
While the worst case latency remains the same, this patch
significantly reduces the service framework overhead for the average
case. In particular, scenarios where an lcore only runs a single
service, or multiple services which id values are close (e.g., three
services with ids 17, 18 and 22), show significant improvements.
The worse case is a where the lcore two services mapped to it; one
with service id 0 and the other with id 63.
On a service lcore serving a single service, the service loop overhead
is reduced from ~190 core clock cycles to ~46, on an Intel Cascade
Lake generation Xeon. On weakly ordered CPUs, the gain is larger,
since the loop included load-acquire atomic operations.
Signed-off-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Introduce a per-lcore counter for the total time spent on processing
services on that core.
This counter is useful when measuring individual lcore load.
Signed-off-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Move the statistics from the service data structure to the per-lcore
struct. This eliminates contention for the counter cache lines, which
decreases the producer-side statistics overhead for services deployed
across many lcores.
Prior to this patch, enabling statistics for a service with a
per-service function call latency of 1000 clock cycles deployed across
16 cores on a Intel Xeon 6230N @ 2,3 GHz would incur a cost of ~10000
core clock cycles per service call. After this patch, the statistics
overhead is reduce to 22 clock cycles per call.
Signed-off-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
This commit fixes a potential racey-add that could occur if
multiple service-lcores were executing the same MT-safe service
at the same time, with service statistics collection enabled.
Because multiple threads can run and execute the service, the
stats values can have multiple writer threads, resulting in the
requirement of using atomic addition for correctness.
Note that when a MT unsafe service is executed, a spinlock is
held, so the stats increments are protected. This fact is used
to avoid executing atomic add instructions when not required.
Regular reads and increments are used, and only the store is
specified as atomic, reducing perf impact on e.g. x86 arch.
This patch causes a 1.25x increase in cycle-cost for polling a
MT safe service when statistics are enabled. No change was seen
for MT unsafe services, or when statistics are disabled.
Fixes: 21698354c8 ("service: introduce service cores concept")
Reported-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
Suggested-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Suggested-by: Morten Brørup <mb@smartsharesystems.com>
Suggested-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
This commit improves the performance reporting of the service
cores polling loop to show both with and without statistics
collection modes. Collecting cycle statistics is costly, due
to calls to rte_rdtsc() per service iteration.
Reported-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
Suggested-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Suggested-by: Morten Brørup <mb@smartsharesystems.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
There is a possibility of deadlock in this API,
as same spinlock is tried to be acquired in nested manner.
If the lcore that is stopping the timer is different from the lcore
that owns the timer, the timer list lock is acquired in timer_del(),
even if local_is_locked is true. Because the same lock was already
acquired in rte_timer_stop_all(), the thread will hang.
This patch removes the acquisition of nested lock.
Fixes: 821c51267b ("timer: add function to stop all timers in a list")
Cc: stable@dpdk.org
Signed-off-by: Naga Harish K S V <s.v.naga.harish.k@intel.com>
Acked-by: Erik Gabriel Carrillo <erik.g.carrillo@intel.com>
When more than two packets are merged in a flow, and if we receive
a 3rd packet which is matching the sequence of the 2nd packet the
prev_idx will be 1 and not 2, hence resulting in packet re-ordering
Signed-off-by: Kumara Parameshwaran <kumaraparamesh92@gmail.com>
Acked-by: Jiayu Hu <jiayu.hu@intel.com>
Currently, when vm_power_manager exits, we are using a LIST_FOREACH
macro to iterate over VM info structures while freeing them. This
leads to use-after-free error. To address this, replace all usages of
LIST_* with TAILQ_* macros, and use the RTE_TAILQ_FOREACH_SAFE macro
to iterate and delete VM info structures.
Fixes: e8ae9b6625 ("examples/vm_power: channel manager and monitor in host")
Cc: stable@dpdk.org
Signed-off-by: Hamza Khan <hamza.khan@intel.com>
Signed-off-by: Reshma Pattan <reshma.pattan@intel.com>
Acked-by: David Hunt <david.hunt@intel.com>
The function return type is changed to fixed width uint32_t
to be consistent with what appears to be the original authors intent.
It doesn't make much sense to return signed integers for these functions.
Signed-off-by: Tyler Retzlaff <roretzla@linux.microsoft.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Structure rte_security_session is moved to internal
headers which are not visible to applications.
The only field which should be used by app is opaque_data.
This field can now be accessed via set/get APIs added in this
patch.
Subsequent changes in app and lib are made to compile the code.
Signed-off-by: Akhil Goyal <gakhil@marvell.com>
Tested-by: Gagandeep Singh <g.singh@nxp.com>
Tested-by: David Coyle <david.coyle@intel.com>
Tested-by: Kevin O'Sullivan <kevin.osullivan@intel.com>
Added the support for rte_security_op.session_get_size()
in all the PMDs which support rte_security sessions and the
op was not supported.
Signed-off-by: Akhil Goyal <gakhil@marvell.com>
Acked-by: Kai Ji <kai.ji@intel.com>
Tested-by: Gagandeep Singh <g.singh@nxp.com>
Tested-by: David Coyle <david.coyle@intel.com>
Tested-by: Kevin O'Sullivan <kevin.osullivan@intel.com>
As per current design, rte_security_session_create()
unnecessarily use 2 mempool objects for a single session.
To address this, the API will now take only 1 mempool
object instead of 2. With this change, the library layer
will get the object from mempool and session priv data is
stored contiguously in the same mempool object.
User need to ensure that the mempool created in application
is big enough for session private data as well. This can be
ensured if the pool is created after getting size of session
priv data using API rte_security_session_get_size().
Since set and get pkt metadata for security sessions are now
made inline for Inline crypto/proto mode, a new member fast_mdata
is added to the rte_security_session.
To access opaque data and fast_mdata will be accessed via inline
APIs which can do pointer manipulations inside library from
session_private_data pointer coming from application.
Signed-off-by: Akhil Goyal <gakhil@marvell.com>
Tested-by: Gagandeep Singh <g.singh@nxp.com>
Tested-by: David Coyle <david.coyle@intel.com>
Tested-by: Kevin O'Sullivan <kevin.osullivan@intel.com>
Structure rte_cryptodev_sym_session is moved to internal
headers which are not visible to applications.
The only field which should be used by app is opaque_data.
This field can now be accessed via set/get APIs added in this
patch.
Subsequent changes in app and lib are made to compile the code.
Signed-off-by: Akhil Goyal <gakhil@marvell.com>
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Kai Ji <kai.ji@intel.com>
Tested-by: Gagandeep Singh <g.singh@nxp.com>
Tested-by: David Coyle <david.coyle@intel.com>
Tested-by: Kevin O'Sullivan <kevin.osullivan@intel.com>
This patch updates the scheduler PMD to use unified session
data structure. Previously thanks to the private session
array in cryptodev sym session there are no necessary
change needed for scheduler PMD other than the way ops
are enqueued/dequeued. The patch inherits the same design
in the original session data structure to the scheduler PMD
so the cryptodev sym session can be as a linear buffer for
both session header and driver private data.
With the change there are inevitable extra cost on both memory
(64 bytes per session per driver type) and cycle count (set
the correct session for each cop based on the worker before
enqueue, and retrieve the original session after dequeue).
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Signed-off-by: Akhil Goyal <gakhil@marvell.com>
Acked-by: Kai Ji <kai.ji@intel.com>
Tested-by: Gagandeep Singh <g.singh@nxp.com>
Tested-by: David Coyle <david.coyle@intel.com>
Tested-by: Kevin O'Sullivan <kevin.osullivan@intel.com>
As per current design, rte_cryptodev_sym_session_create() and
rte_cryptodev_sym_session_init() use separate mempool objects
for a single session.
And structure rte_cryptodev_sym_session is not directly used
by the application, it may cause ABI breakage if the structure
is modified in future.
To address these two issues, the rte_cryptodev_sym_session_create
will take one mempool object that the session and session private
data are virtually/physically contiguous, and initializes both
fields. The API rte_cryptodev_sym_session_init is removed.
rte_cryptodev_sym_session_create will now return an opaque session
pointer which will be used by the app and other APIs.
In data path, opaque session pointer is attached to rte_crypto_op
and the PMD can call an internal library API to get the session
private data pointer based on the driver id.
Note: currently single session may be used by different device
drivers, given it is initialized by them. After the change the
session created by one device driver cannot be used or
reinitialized by another driver.
Signed-off-by: Akhil Goyal <gakhil@marvell.com>
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Kai Ji <kai.ji@intel.com>
Tested-by: Gagandeep Singh <g.singh@nxp.com>
Tested-by: David Coyle <david.coyle@intel.com>
Tested-by: Kevin O'Sullivan <kevin.osullivan@intel.com>
During EAL init, all buses are probed and the devices found are
initialized. On eal_cleanup(), the inverse does not happen, meaning any
allocated memory and other configuration will not be cleaned up
appropriately on exit.
Currently, in order for device cleanup to take place, applications must
call the driver-relevant functions to ensure proper cleanup is done before
the application exits. Since initialization occurs for all devices on the
bus, not just the devices used by an application, it requires a)
application awareness of all bus devices that could have been probed on the
system, and b) code duplication across applications to ensure cleanup is
performed. An example of this is rte_eth_dev_close() which is commonly used
across the example applications.
This patch proposes adding bus cleanup to the eal_cleanup() to make EAL's
init/exit more symmetrical, ensuring all bus devices are cleaned up
appropriately without the application needing to be aware of all bus types
that may have been probed during initialization.
Contained in this patch are the changes required to perform cleanup for
devices on the PCI bus and VDEV bus during eal_cleanup(). There would be an
ask for bus maintainers to add the relevant cleanup for their buses since
they have the domain expertise.
Signed-off-by: Kevin Laatz <kevin.laatz@intel.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Reviewed-by: Bruce Richardson <bruce.richardson@intel.com>
Since 10 years, memzone allocation is allowed on secondary
processes. Now it's time to update the documentation accordingly.
At the same time, fix mempool, mbuf and ring documentation which rely on
memzones internally.
Bugzilla ID: 1074
Fixes: 916e4f4f4e ("memory: fix for multi process support")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Driver allocates "bp->rep_info" inside bnxt_init_rep_info() which is
invoked from bnxt_rep_port_probe(). But the memory is freed inside
bnxt_uninit_resources(), which is wrong. As a result, after error
recovery bp->rep_info will be NULL. The memory should have freed inside
bnxt_drv_uninit() to maintain symmetry of calls.
Fixes: 6dc83230b4 ("net/bnxt: support port representor data path")
Cc: stable@dpdk.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>
We are not invoking rte_eth_switch_domain_free currently owing to
an unnecessary check. This patch fixes that.
Fixes: e2895305a5 ("net/bnxt: fix resource cleanup")
Cc: stable@dpdk.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>
When the BNXT PMD was made to disallow MTU changes on active ports, the
error code chosen for the case in bnxt_set_mtu_op() was -EPERM.
The doc comment for rte_eth_dev_set_mtu() in lib/ethdev/rte_ethdev.h
lists -EBUSY as the value to be used if the port must be stopped before
applying an MTU change and does not list -EPERM as a possible return
value.
This patch makes bnxt_set_mtu_op() return -EBUSY instead of -EPERM so
that rte_eth_dev_set_mtu() behaves as expected.
Fixes: a42ab1eb33 ("net/bnxt: disallow MTU change when device is started")
Cc: stable@dpdk.org
Signed-off-by: Benjamin Le Berre <benjamin.le_berre@6wind.com>
Acked-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
This function was never implemented and has been deprecated for a long
time. We can remove it.
Signed-off-by: David Marchand <david.marchand@redhat.com>
Added support for CN10K SoC DMA engine to dmadev.
Signed-off-by: Radha Mohan Chintakuntla <radhac@marvell.com>
Reviewed-by: Jerin Jacob <jerinj@marvell.com>
Currently the example using DMA in asynchronous mode, which are:
nb_rx = rte_eth_rx_burst();
if (nb_rx == 0)
continue;
...
dma_enqueue(); // enqueue the received packets copy request
nb_cpl = dma_dequeue(); // get copy completed packets
...
There are no waiting inside dma_dequeue(), and this is why it's called
asynchronus. If there are no packet received, it won't call
dma_dequeue(), but some packets may still in the DMA queue which
enqueued in last cycle. As a result, when the traffic is stopped, the
sent packets and received packets are unbalanced from the perspective
of the traffic generator.
The patch supports DMA dequeue when no packet received, it helps to
judge the test result by comparing the sent packets with the received
packets on traffic generator sides.
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Kevin Laatz <kevin.laatz@intel.com>
As part of the agreed process for deprecating KNI in DPDK, the example
app is scheduled for removal as part of the 22.11 release.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
NVIDIA acquired Mellanox Technologies in 2020.
The DPDK documentation and code might still include instances
of or references to Mellanox trademarks (like BlueField and ConnectX)
that are now NVIDIA trademarks.
The PCI IDs and copyrights are unchanged.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: Gal Cohen <galco@nvidia.com>
Add support for vector aggregation of crypto operations for cn10k.
Crypto operations will be grouped by sub event type, flow id, scheduler
type and queue id fields from rte_event_crypto_metadata::response_info.
Signed-off-by: Volodymyr Fialko <vfialko@marvell.com>
Introduce ability to aggregate crypto operations processed by event
crypto adapter into single event containing rte_event_vector whose event
type is RTE_EVENT_TYPE_CRYPTODEV_VECTOR.
Application should set RTE_EVENT_CRYPTO_ADAPTER_EVENT_VECTOR in
rte_event_crypto_adapter_queue_conf::flag and provide vector configuration
with respect of rte_event_crypto_adapter_vector_limits, which could be
obtained by calling rte_event_crypto_adapter_vector_limits_get, to enable
vectorization.
The event crypto adapter would be responsible for vectorizing the crypto
operations based on provided response information in
rte_event_crypto_metadata::response_info.
Updated drivers and tests accordingly to new API.
Signed-off-by: Volodymyr Fialko <vfialko@marvell.com>
Acked-by: Akhil Goyal <gakhil@marvell.com>
The CDEV_LOG_* macros already add a '\n' at the end of
the line. Remove it from format strings to avoid duplicated
newlines.
Fixes: 9e6edea418 ("cryptodev: add APIs to assist PMD initialisation")
Fixes: e764cd72a9 ("cryptodev: update symmetric session structure")
Fixes: 1d6f89885e ("cryptodev: add sym session mempool create")
Fixes: 1f1e4b7cba ("cryptodev: use single mempool for asymmetric session")
Fixes: 757f40e28e ("cryptodev: modify return value for asym session create")
Fixes: cea66374dc ("cryptodev: support asymmetric operations")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
Acked-by: Akhil Goyal <gakhil@marvell.com>
Add trace points for cryptodev functions.
Some of the APIs are restructured to add traces and return
appropriately as needed.
Signed-off-by: Amit Prakash Shukla <amitprakashs@marvell.com>
Reviewed-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Akhil Goyal <gakhil@marvell.com>
Add enumeration in EC xform for FPM (fixed point multiplication).
Crypto driver would need this to xform point multiplication based
on given type of EC curve.
Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
Signed-off-by: Gowrishankar Muthukrishnan <gmuthukrishn@marvell.com>
Acked-by: Kai Ji <kai.ji@intel.com>
ShangMi 3 (SM3) is a cryptographic hash function used in
the Chinese National Standard.
- Added SM3 algorithm
Signed-off-by: Arek Kusztal <arkadiuszx.kusztal@intel.com>
Acked-by: Kai Ji <kai.ji@intel.com>
Acked-by: Anoob Joseph <anoobj@marvell.com>
Acked-by: Akhil Goyal <gakhil@marvell.com>
ShangMi 4 (SM4) is a block cipher used in the
Chinese National Standard for Wireless LAN WAPI and also
used with Transport Layer Security.
Added SM4 encryption algorithm in ECB, CBC and CTR modes.
Signed-off-by: Arek Kusztal <arkadiuszx.kusztal@intel.com>
Acked-by: Kai Ji <kai.ji@intel.com>
Acked-by: Anoob Joseph <anoobj@marvell.com>
Acked-by: Akhil Goyal <gakhil@marvell.com>
The API rte_security_get_userdata() was being unused by most of
the drivers and it was retrieving userdata from mbuf dynamic field.
Hence, the API was removed and the application can directly get the
userdata from dynamic field. This helps in removing extra checks
in datapath.
Signed-off-by: Srujana Challa <schalla@marvell.com>
Acked-by: Akhil Goyal <gakhil@marvell.com>
Modify reader/writer lock to avoid starvation of writer. The previous
implementation would cause a writer to get starved if readers kept
acquiring the lock. The new version uses an additional bit to indicate
that a writer is waiting and which keeps readers from starving the
writer.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
DPDK libraries should never call rte_exit on failure, so change the
function return type of rte_metrics_init to "int" to allow returning an
error code to the application rather than exiting the whole app on init
failure.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Library functions should not cause the app to exit or panic. Replace the
existing panic call in the EAL remote launch functions with an error
code return instead.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
Deprecation notice targeted for v22.11 of event vector has been
merged in the following commits, remove deprecation notices.
Fixes: 0fbb55efa5 ("eventdev: add element offset to event vector")
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Replace *u64s with u64s in rte_event_vector structure as
the *ptrs already serves the purpose of holding pointers
and the intention of u64s is to hold array of uint64_t
values.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Add `rte` prefix to stop flush callback function pointer
declaration to avoid conflicts with application functions,
``eventdev_stop_flush_t`` is renamed to
``rte_eventdev_stop_flush_t``.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
