After dequeuing Rx packets and enqueueing them to the
temporary buffer towards eventdev, the packet Rx loop exits
if the temporary buffer is full but the current WRR position
is not saved.
Save away the current value of the WRR position, so packets
are dequeued from the correct Rx queue in the next invocation.
Fixes: 9c38b704d280 ("eventdev: add eth Rx adapter implementation")
Cc: stable@dpdk.org
Suggested-by: Gage Eads <gage.eads@intel.com>
Signed-off-by: Nikhil Rao <nikhil.rao@intel.com>
The dev_id parameter to fill_event_buffer() should be 16 bit,
also rename to to eth_dev_id to avoid confusion with event device
id elsewhere in the file.
Fixes: c2189c907dd1 ("eventdev: make ethdev port identifiers 16-bit")
Cc: stable@dpdk.org
Signed-off-by: Nikhil Rao <nikhil.rao@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
vhost_vring_call() used rte_mb(), which translates into
mfence instruction on x86.
This patch changes to use rte_smp_mb(), which changed recently
to translate into a locked ADD instruction for performance
reason.
The measured gain is up to 3% with the testpmd benchmarks.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>
Introduce an new common helper to avoid redundancy.
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
When a vDPA device is attached, vhost user will try to
register host notifiers to QEMU to allow notifications
to be delivered between the driver in the guest and the
vDPA device in the host directly.
Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Make sure find avalid device id before allocating
virtio_net, if not, return directly. It may avoid
allocating and freeing virtio_net when there is
not valid device id.
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
PMDs should provide supported RSS hash functions via
dev_info.flow_type_rss_offloads variable.
There is a check in ethdev if requested RSS hash function is supported
by PMD or not.
This check has been relaxed in previous release to not return an error
when a non supported has function requested [1], this has been done to
not break the applications.
Adding the error return back.
PMDs need to provide correct list of supported hash functions and
applications need to take care this information before configuring
the RSS otherwise they will get an error from APIs:
rte_eth_dev_rss_hash_update()
rte_eth_dev_configure()
[1] commit af7551e2bfce ("ethdev: remove error return on RSS hash check")
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
The error path was disabled in previous release to let apps to be more
flexible.
But this release they are enabled, applications have to obey offload API
rules otherwise they will get errors from following APIs:
rte_eth_dev_configure
rte_eth_rx_queue_setup
rte_eth_tx_queue_setup
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Instead of copying batch_copy_nb_elems into the stack,
this patch uses it directly.
Small performance gain of 3% is seen when running PVP
benchmark.
Acked-by: Zhihong Wang <zhihong.wang@intel.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
This patch reworks the vhost enqueue path so that a single
code path is used for both Rx mergeable or non-mergeable cases.
Acked-by: Zhihong Wang <zhihong.wang@intel.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The *_callback_* functions are not implemented,
so they are removed from the export map file.
Fixes: ed7dd94f7f66 ("compressdev: add basic device management")
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
The functions
- vfio_get_container_fd
- vfio_get_group_fd
- vfio_get_group_no
have been renamed to
- rte_vfio_get_container_fd
- rte_vfio_get_group_fd
- rte_vfio_get_group_num
The old names are removed from the map file.
Fixes: 964b2f3bfb07 ("vfio: export some internal functions")
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Deprecate rte_eal_mbuf_default_mempool_ops(), it shall be replaced by
rte_mbuf_best_mempool_ops().
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Currently, memzone allocation with length set to 0 that are also
IOVA-contiguous is not supported. Document this limitation.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Mem event and validator callbacks may not be supported under all
circumstances (such as when running in legacy memory mode, or on
FreeBSD), and this case needs to be handled by any code that will
use these callbacks. Spell this out more clearly, because it's not
immediately obvious that this is an expected use case.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
The rte_tm_get_number_of_leaf_nodes() API function was added in DPDK
17.08. However, it was added to the .map file with the wrong function
name (rte_tm_get_leaf_nodes), which was subsequently removed in commit
3e8ea3d ('lib: remove unused map symbols').
Add it back under the 17.08 release with the correct function name.
Fixes: 5d109deffa87 ("ethdev: add traffic management API")
Cc: stable@dpdk.org
Signed-off-by: Ben Shelton <benjamin.h.shelton@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Change adds extra information on name parameter for API
rte_eth_dev_get_name_by_port and rte_eth_dev_get_port_by_name.
Signed-off-by: Vipin Varghese <vipin.varghese@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Like original commit mentioned below, this fix synchronizes flow rule copy
function with testpmd's own implementation following "app/testpmd: fix copy
of raw flow item (revisited)".
It addresses a crash that occurs when feeding a RAW pattern item to
rte_flow_copy(). Besides external applications, two PMDs (bonding and
failsafe) rely on this function internally.
Note the scope of this patch is limited to the RAW pattern item and has no
impact on all others.
Fixes: 972bf3610611 ("ethdev: fix shallow copy of flow API RSS action")
Cc: stable@dpdk.org
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
eth_dev_last_created_port is used to store port id type and should
be extended to 16bits corresponding to ethdev port id range.
Fixes: f8244c6399d9 ("ethdev: increase port id range")
Cc: stable@dpdk.org
Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
rte_event_eth_rx_adapter_create allocates eth_devices for
currently available eth devices. For newly created eth
devices a new instance for rx adapter has to be created.
Signed-off-by: Vipin Varghese <vipin.varghese@intel.com>
Acked-by: Nikhil Rao <nikhil.rao@intel.com>
Fix the call to rte_timer_reset_sync() in sw_event_timer_cb(). The
second parameter is the number of ticks, the third is the timer type.
Fixes: 6750b21bd6af ("eventdev: add default software timer adapter")
Signed-off-by: Dan Gora <dg@adax.com>
Acked-by: Erik Gabriel Carrillo <erik.g.carrillo@intel.com>
rte_cryptodev_get_header_session_size() and
rte_cryptodev_get_private_session_size() functions are
targeting symmetric sessions.
With the future addition of asymmetric operations,
these functions need to be renamed from *cryptodev_*_session_*
to *cryptodev_sym_*_session_* to be symmetric specific.
The two original functions are marked as deprecated
and will be removed in 18.08, so applications can still
use the functions in 18.05.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Shally Verma <shally.verma@caviumnetworks.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Deepak Kumar Jain <deepak.k.jain@intel.com>
Functions rte_cryptodev_queue_pair_start/stop
are not really used in any of the crypto drivers
(they all just return 0 or -ENOTSUP).
Therefore, this API can be deprecated from 18.05
and removed in 18.08.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Deepak Kumar Jain <deepak.k.jain@intel.com>
Functions rte_cryptodev_queue_pair_attach_sym_session
and rte_cryptodev_queue_pair_detach_sym_sessions
are not really used in any of the crypto drivers
(only one driver implements it and it just return 0).
Therefore, this API can be deprecated from 18.05
and removed in 18.08.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Deepak Kumar Jain <deepak.k.jain@intel.com>
Add extra clarification about offset in source and
destination mbuf used in compressdev, when they
are a chain of mbufs.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Reviewed-by: Shally Verma <shally.verma@caviumnetworks.com>
Session data can only be cleared once there are no
inflight operations using the session. It is the application's
responsability to make sure of this.
Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
As the private_xform data can be shared by many operations
and across queue_pairs, it would be performance impacting
for PMDs to track inflights associated with one. It makes
more sense to push the responsibility to the application to
keep track of its usage and only delete the private_xform when
there are no more ops using it.
Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Driver id field is not set/used anywhere,
so it should be removed from rte_compressdev structure.
Fixes: ed7dd94f7f66 ("compressdev: add basic device management")
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
Security protocol flag string was not added
when the actual flag was added.
Fixes: eadb4fa1e1fe ("cryptodev: support security APIs")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Vipin Varghese <vipin.varghese@intel.com>
Extend the description of cryptodev feature flags,
adding extra information.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Vipin Varghese <vipin.varghese@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
Crypto capability structure contains supported
sizes for key, IV, digest, etc. on different algorithms.
These sizes can be expressed as a single value or
a range of values.
The check was broken when a size was checked against
a range with multiple values.
Also, for more clarity, the param_range_check macro
has been converted into a function.
Fixes: 38227c0e3ad2 ("cryptodev: retrieve device info")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
GCC 8.1 warned:
In function 'rte_eth_rx_burst':
rte_ethdev.h:3836:18: warning: conversion to 'int16_t'
{aka 'short int'} from 'uint16_t' {aka 'short unsigned int'}
may change the sign of the result [-Wsign-conversion]
int16_t nb_rx = (*dev->rx_pkt_burst)(dev->data->rx_queues[queue_id],
^
rte_ethdev.h:3844:50: warning: conversion to 'uint16_t'
{aka 'short unsigned int'} from 'int16_t' {aka 'short int'}
may change the sign of the result [-Wsign-conversion]
nb_rx = cb->fn.rx(port_id, queue_id, rx_pkts, nb_rx,
^~~~~
rte_ethdev.h:3844:12: warning: conversion to 'int16_t'
{aka 'short int'} from 'uint16_t' {aka 'short unsigned int'}
may change the sign of the result [-Wsign-conversion]
nb_rx = cb->fn.rx(port_id, queue_id, rx_pkts, nb_rx,
^~
rte_ethdev.h:3851:9: warning: conversion to 'uint16_t'
{aka 'short unsigned int'} from 'int16_t' {aka 'short int'}
may change the sign of the result [-Wsign-conversion]
return nb_rx;
^~~~~
The second part of the patch is solved by its own basic
block because it is inside a preprocessor conditional.
Bringing the declaration of the var to the top of the
function would require that also being given its own
preprocessor conditional, or a (void)var to avoid an
unused var warning. The basic block is no worse than
those imho.
Fixes: 467465d86df1 ("ethdev: add packet count parameter to Rx callback")
Fixes: 4dc294158cac ("ethdev: support optional Rx and Tx callbacks")
Cc: stable@dpdk.org
Signed-off-by: Andy Green <andy@warmcat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
GCC 8.1 warned:
In function 'rte_pktmbuf_prepend':
rte_mbuf.h:1908:17: warning: conversion from 'int' to 'uint16_t'
{aka 'short unsigned int'} may change value [-Wconversion]
m->data_off -= len;
^~~
m->data_off is a uint16_t
uint16_t data_off;
len (a uint16_t) is promoted to an int using -=. Do the
subtraction explicitly and cast the result to uint16_t.
The below += or -= changes are solving the same thing.
In function 'rte_pktmbuf_adj':
rte_mbuf.h:1969:17: warning: conversion from 'int' to 'uint16_t'
{aka 'short unsigned int'} may change value [-Wconversion]
m->data_off += len;
^~~
In function 'rte_pktmbuf_chain':
rte_mbuf.h:2082:19: warning: conversion from 'int' to 'uint16_t'
{aka 'short unsigned int'} may change value [-Wconversion]
head->nb_segs += tail->nb_segs;
^~~~
Also uint16_t
uint16_t nb_segs; /**< Number of segments. */
Fixes: 08b563ffb19d ("mbuf: replace data pointer by an offset")
Fixes: 1a60a0daa6e4 ("mbuf: fix segments number type increase")
Cc: stable@dpdk.org
Signed-off-by: Andy Green <andy@warmcat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
GCC 8.1 warned:
In function 'rte_validate_tx_offload':
rte_mbuf.h:2112:19: warning: conversion to 'uint64_t'
{aka 'long unsigned int'} from 'int' may change the
sign of the result [-Wsign-conversion]
inner_l3_offset += m->outer_l2_len + m->outer_l3_len;
^~
uint64_t inner_l3_offset...
/* fields for TX offloading of tunnels */
uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
We want to do the arithmetic entirely in uint64_t
space, but with the +=, the rhs type becomes int since the
bitfields will fit in int.
Elaborate the artithmetic to be u64 = u64 + int + int, so
the type of the result is correct to be stored in the u64.
Fixes: 4fb7e803eb ("ethdev: add Tx preparation")
Cc: stable@dpdk.org
Signed-off-by: Andy Green <andy@warmcat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
GCC 8.1 warned:
In function 'rte_pktmbuf_linearize':
rte_mbuf.h:1873:32: warning: conversion to 'int' from 'uint32_t'
{aka 'unsigned int'} may change the sign of the result [-Wsign-conversion]
rte_mbuf.h:2166:13: note: in expansion of macro 'rte_pktmbuf_pkt_len'
copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
rte_mbuf.h:2180:51: warning: conversion to 'size_t'
{aka 'long unsigned int'} from 'int' may change the
sign of the result [-Wsign-conversion]
rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
^~~~~~~
The temp is consumed as a size_t. So let's make it
a size_t in the first place.
Fixes: 1feda4d8fc ("mbuf: add a function to linearize a packet")
Cc: stable@dpdk.org
Signed-off-by: Andy Green <andy@warmcat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
GCC 8.1 warned:
In function 'rte_pktmbuf_detach':
rte_mbuf.h:1583:17: warning: conversion from 'uint32_t'
{aka 'unsigned int'} to 'uint16_t' {aka 'short unsigned int'}
may change value [-Wconversion]
m->priv_size = priv_size;
^~~~~~~~~
The temp priv_size is declared as a uint32_t. But it
only deals in uint16_t. m->priv_size is a uint16_t.
Change it to a uint16_t.
Fixes: 355e6735b3 ("mbuf: fix cloning with private mbuf data")
Cc: stable@dpdk.org
Signed-off-by: Andy Green <andy@warmcat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
GCC 8.1 warned:
In function 'rte_ipv4_udptcp_cksum':
rte_byteorder.h:51:24: warning: conversion from 'long unsigned int' to
'uint32_t' {aka 'unsigned int'} may change value [-Wconversion]
#define rte_bswap16(x) ((uint16_t) (__builtin_constant_p(x) ? \
^
rte_byteorder.h:85:29: note: in expansion of macro 'rte_bswap16'
#define rte_be_to_cpu_16(x) rte_bswap16(x)
^~~~~~~~~~~
rte_ip.h:321:11: note: in expansion of macro 'rte_be_to_cpu_16'
l4_len = rte_be_to_cpu_16(ipv4_hdr->total_length) -
^~~~~~~~~~~~~~~~
Also with this one, it is a cast that always occurred
and is just being done explicitly, with no changes to
the generated code.
The warning stack is misleading, it points to the last
element in the macro that produced the lhs of the subtraction
above. But the only "unsigned long int" in the expression is
the result of the sizeof() on the rhs, it promotes the
subtraction result to unsigned long. So the error actually
relates to the result of the outer subtraction.
The actual error is "you are trying to put an unsigned long
into a uint32_t". We always did so, the fix is just to inform
the compiler it is intentional with an explicit cast.
Fixes: 6006818cfb ("net: new checksum functions")
Cc: stable@dpdk.org
Signed-off-by: Andy Green <andy@warmcat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
GCC 8.1 warned:
In function 'rte_rwlock_read_lock':
rte_rwlock.h:74:12: warning: conversion to 'uint32_t'
{aka 'unsigned int'} from 'int32_t' {aka 'int'} may
change the sign of the result [-Wsign-conversion]
x, x + 1);
^
rte_rwlock.h:74:17: warning: conversion to 'uint32_t'
{aka 'unsigned int'} from 'int' may change the sign
of the result [-Wsign-conversion]
x, x + 1);
~~^~~
In function 'rte_rwlock_write_lock':
rte_rwlock.h:110:15: warning: unsigned conversion
from 'int' to 'uint32_t' {aka 'unsigned int'}
changes value from '-1' to '4294967295' [-Wsign-conversion]
0, -1);
^~
Again in this case we are making explicit the exact cast
that was always happening implicitly. The patch does not
change the generated code.
The int32_t temp "x" is required to be signed to detect
a < 0 error condition from the lock status. Afterwards,
it has always been implicitly cast to uint32_t when it
is used in the arguments to rte_atomic32_cmpset()...
gcc8.1 objects to the implicit cast now and requires us
to cast it explicitly.
Fixes: af75078fec ("first public release")
Cc: stable@dpdk.org
Signed-off-by: Andy Green <andy@warmcat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
GCC 8.1 warned:
rte_memcpy.h:793:2: note: in expansion of macro 'MOVEUNALIGNED_LEFT47'
MOVEUNALIGNED_LEFT47(dst, src, n, srcofs);
^~~~~~~~~~~~~~~~~~~~
rte_memcpy.h:649:51: warning: conversion from 'size_t'
{aka 'long unsigned int'} to 'int' may change value [-Wconversion]
case 0x0B: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x0B); break;
^
rte_memcpy.h:616:15: note: in definition of macro 'MOVEUNALIGNED_LEFT47_IMM'
tmp = len;
^~~
rte_memcpy.h:793:2: note: in expansion of macro 'MOVEUNALIGNED_LEFT47'
MOVEUNALIGNED_LEFT47(dst, src, n, srcofs);
^~~~~~~~~~~~~~~~~~~~
rte_memcpy.h:618:13: warning: conversion to 'size_t'
{aka 'long unsigned int'} from 'int'
may change the sign of the result [-Wsign-conversion]
tmp -= len;
^~
rte_memcpy.h:649:16: note: in expansion of macro 'MOVEUNALIGNED_LEFT47_IMM'
case 0x0B: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x0B); break;
^~~~~~~~~~~~~~~~~~~~~~~~
rte_memcpy.h:793:2: note: in expansion of macro 'MOVEUNALIGNED_LEFT47'
MOVEUNALIGNED_LEFT47(dst, src, n, srcofs);
^~~~~~~~~~~~~~~~~~~~
rte_memcpy.h:618:13: warning: conversion to 'size_t'
{aka 'long unsigned int'} from 'int'
may change the sign of the result [-Wsign-conversion]
tmp -= len;
^~
We can eliminate the problems by setting the type of tmp to
size_t in the first place.
Fixes: d35cc1fe6a ("eal/x86: revert select optimized memcpy at run-time")
Cc: stable@dpdk.org
Suggested-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Andy Green <andy@warmcat.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
