Maxime Leroy [Tue, 13 Aug 2019 13:29:47 +0000 (15:29 +0200)]
net/vmxnet3: remove IP checksum from capabilities
The vmxnet3_prep_pkts function set rte_errno to ENOTSUP for any packets
having PKT_TX_IP_CHECKSUM set in ol_flags. But the vmxnet3 has
DEV_TX_OFFLOAD_IPV4_CKSUM set in this tx offload capa.
This issue has been introduced with the new Rx offload
API. DEV_TX_OFFLOAD_IPV4_CKSUM and DEV_RX_OFFLOAD_IPV4_CKSUM has been
added to the tx/rx offloads capa, but the vmxnet3 driver doesn't support
it.
To fix the issue, DEV_TX/RX_OFFLOAD_IPV4_CKSUM needs to be removed from
tx/rx offload capa.
Fixes: 95e4a96ccbf1 ("net/vmxnet3: convert to new Rx offload API") Cc: stable@dpdk.org Signed-off-by: Maxime Leroy <maxime.leroy@6wind.com> Acked-by: Yong Wang <yongwang@vmware.com>
Yong Wang [Fri, 30 Aug 2019 08:45:29 +0000 (16:45 +0800)]
net/ixgbe: remove redundant assignment
Since "link.link_duplex" has been assigned to ETH_LINK_FULL_DUPLEX just
before switch statement, the assignment in switch-case statement is
redundant. Remove it.
Fixes: 82113036e4e5 ("ethdev: redesign link speed config") Cc: stable@dpdk.org Signed-off-by: Yong Wang <wang.yong19@zte.com.cn> Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
Ruifeng Wang [Wed, 28 Aug 2019 08:24:54 +0000 (16:24 +0800)]
net/ixgbe: use intrinsics to count packet in NEON Rx
vPMD for aarch64 calculates the number of received packets using a loop.
Change to use NEON intrinsics for calculation. This saves CPU cycles
and has slightly better performance.
Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Ruifeng Wang [Wed, 28 Aug 2019 08:24:53 +0000 (16:24 +0800)]
net/ixgbe: remove memory barrier from NEON Rx
The memory barrier was intended for descriptor data integrity (see
comments in [1]). As later NEON loads were implemented and a whole
entry is loaded in one-run and atomic, that makes the ordering of
partial loading unnecessary. Remove it accordingly.
Corrected couple of code comments.
In terms of performance, observed slightly higher average throughput
in tests with 82599ES NIC.
[1] http://patches.dpdk.org/patch/18153/
Fixes: 989a84050542 ("net/ixgbe: fix received packets number for ARM NEON") Cc: stable@dpdk.org Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Gavin Hu [Tue, 13 Aug 2019 10:43:30 +0000 (18:43 +0800)]
net/i40e: remove memory barrier from NEON Rx
For x86, the descriptors needs to be loaded in order, so in between two
descriptors loading, there is a compiler barrier in place.[1]
For aarch64, a patch [2] is in place to survive with discontinuous DD
bits, the barriers can be removed to take full advantage of out-of-order
execution.
50% performance gain in the RFC2544 NDR test was measured on ThunderX2.
12.50% performance gain in the RFC2544 NDR test was measured on Ampere
eMAG80 platform.
error: incompatible types when initializing type
'__vector _bool int' {aka '_vector(4) __bool int'} using type 'int'
If __APPLE_ALTIVEC__ is defined, then bool type is redefined
and conflicts with stdbool.h.
There is no good solution to fix it for the whole project without
breaking something else, so a workaround is inserted in mlx5 PMD.
This workaround is not compatible with C++ but there is no C++ in DPDK.
Related to: 725f5dd0bfb5 ("net/mlx5: fix build on PPC64") Cc: stable@dpdk.org Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com> Tested-by: David Christensen <drc@linux.vnet.ibm.com> Acked-by: Matan Azrad <matan@mellanox.com>
Add multiple processes support for ice, secondary processes will share
memory and configuration with primary process, do not need further
initialization for secondary processes.
Rastislav Cernay [Tue, 27 Aug 2019 21:04:19 +0000 (23:04 +0200)]
net/szedata2: remove resources when port is closed
The rte_eth_dev_close() function now handles freeing resources for
devices (e.g., mac_addrs). To conform with the new close() behaviour we
are asserting the RTE_ETH_DEV_CLOSE_REMOVE flag so that
rte_eth_dev_close() releases all device level dynamic memory.
Signed-off-by: Rastislav Cernay <cernay@netcope.com> Acked-by: Jan Remes <remes@netcope.com>
Rastislav Cernay [Tue, 27 Aug 2019 20:37:16 +0000 (22:37 +0200)]
net/nfb: remove resources when port is closed
The rte_eth_dev_close() function now handles freeing resources for
devices (e.g., mac_addrs). To conform with the new close() behaviour we
are asserting the RTE_ETH_DEV_CLOSE_REMOVE flag so that
rte_eth_dev_close() releases all device level dynamic memory.
Signed-off-by: Rastislav Cernay <cernay@netcope.com> Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com> Acked-by: Jan Remes <remes@netcope.com>
The testpmd application provides two sets of commands for RX offload
flags configuration. The purpose of this patch is to eliminate this
duplication by removing the old set of commands:
“port config all crc-strip|scatter|rx-cksum|rx-timestamp|hw-vlan|
hw-vlan-filter|hw-vlan-strip|hw-vlan-extend on|off”
The other commands set that can be used instead in order to enable
or disable the same RX offloading flags on all RX queues of a port is:
"port config <port_id> rx_offload crc_strip|scatter|ipv4_cksum|
udp_cksum|tcp_cksum|timestamp|vlan_strip|vlan_filter|vlan_extend on|off"
This patch also fixes the "drop-en" command, which enables packets
dropping on all RX queues of all ports when no receive buffers available
“port config all drop-en on|off”
Fixes: 384161e00627 ("app/testpmd: adjust on the fly VLAN configuration") Cc: stable@dpdk.org Signed-off-by: Flavia Musatescu <flavia.musatescu@intel.com> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
David Marchand [Tue, 20 Aug 2019 13:45:03 +0000 (15:45 +0200)]
ethdev: fix endian annotation for SPI item
Security Parameters Index (SPI) should be set with network endian
values.
While 0xffffffff == htonl(0xffffffff), this missing annotation is
caught by sparse when compiling ovs (dpdk-latest branch).
Fixes: d4b684f7197a ("net: add ESP header to generic flow steering") Cc: stable@dpdk.org Signed-off-by: David Marchand <david.marchand@redhat.com> Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Sylvain Rodon [Fri, 2 Aug 2019 08:57:24 +0000 (10:57 +0200)]
net/pcap: propagate timestamp from header to mbuf
Timestamp is always set in PCAP header, whether it reads a file or
listen on an interface. This information can be important for some
applications and it cannot be obtained otherwise (especially when
reading a PCAP file, where the timestamp is not the current time).
Timestamp here is the number of microseconds since epoch.
Xiaoyu Min [Tue, 20 Aug 2019 09:47:14 +0000 (17:47 +0800)]
net/mlx5: fix crash for empty raw encap data
For the rte_flow_action_raw_encap, the header definition for
encapsulation must be available, otherwise it will lead to crash on some
OFED versions and logically it should be rejected.
Fixes: 8ba9eee4ce32 ("net/mlx5: add raw data encap/decap to Direct Verbs") Cc: stable@dpdk.org Signed-off-by: Xiaoyu Min <jackmin@mellanox.com> Acked-by: Dekel Peled <dekelp@mellanox.com>
Lu Qiuwen [Wed, 7 Aug 2019 08:24:22 +0000 (16:24 +0800)]
net/fm10k: fix stats crash in multi-process
The ops pointers in fm10k_stats_get() are set up from primary
process, when secondary process calls these ops pointers,
a segment fault will happen.
Fixes: 7223d200c227 ("fm10k: add base driver") Cc: stable@dpdk.org Signed-off-by: Lu Qiuwen <luqiuwen@iie.ac.cn> Acked-by: Xiao Wang <xiao.w.wang@intel.com> Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
Xiao Zhang [Tue, 13 Aug 2019 22:17:45 +0000 (06:17 +0800)]
net/i40e: fix VF runtime queues RSS config
I40evf queue can not work properly with kernel pf driver for X722 vf.
Eg. when configure 8 queues pair, only 4 queues can receive packets,
and half packets will be lost if using 2 queues pair.
This issue is caused by misconfiguration of look up table, the original
code of LUT configuration did not work for X722 vf, use aq command to
setup the LUT to make it work properly.
David Marchand [Wed, 10 Apr 2019 12:53:49 +0000 (14:53 +0200)]
net/bonding: prefer allmulti to promiscuous for LACP
Rather than the global promiscuous mode on all slaves, let's try to use
allmulti.
To do this, we apply the same mechanism than for promiscuous and drop in
rx_burst.
When adding a slave, we first try to use allmulti, else we try
promiscuous.
Because of this, we must be able to handle allmulti on the bonding
interface, so the associated dev_ops is added with checks on which mode
has been applied on each slave.
Rather than add a new flag in the bond private structure, we can remove
promiscuous_en since ethdev already maintains this information.
When starting the bond device, there is no promisc/allmulti re-apply
as, again, ethdev does this itself.
On reception, we must inspect if the packets are multicast or unicast
and take a decision to drop based on which mode has been enabled on the
bonding interface.
Note: in the future, if we define an API so that we can add/remove mc
addresses to a port (rather than the current global set_mc_addr_list
API), we can refine this to only register the LACP multicast mac
address.
Signed-off-by: David Marchand <david.marchand@redhat.com> Acked-by: Chas Williams <chas3@att.com>
David Marchand [Wed, 10 Apr 2019 12:53:48 +0000 (14:53 +0200)]
net/bonding: fix unicast packets filtering
By default, the 802.3ad code enables promisc mode on all slaves.
To avoid all packets going to the application (unless the application
asked for promiscuous mode), all frames are supposed to be filtered in
the rx burst handler.
However the incriminated commit broke this because non pure ethernet
frames (basically any unicast Ether()/IP() packet) are not filtered
anymore.
Fixes: 71b7b37ec959 ("net/bonding: use ptype flags for LACP Rx filtering") Cc: stable@dpdk.org Signed-off-by: David Marchand <david.marchand@redhat.com> Acked-by: Chas Williams <chas3@att.com>
David Marchand [Wed, 10 Apr 2019 12:53:46 +0000 (14:53 +0200)]
net/bonding: fix out of bound access in LACP mode
We'd better consolidate the fast queue and the normal tx burst functions
under a common inline wrapper for maintenance.
But looking closer at the bufs_slave_port_idxs[] mapping array in those
tx burst functions, its size is invalid since up to nb_bufs are handled
here.
A previous patch [1] fixed this issue for balance tx burst function
without mentioning it.
802.3ad and balance modes are functionally equivalent on transmit.
The only difference is on the slave id distribution.
Add an additional inline wrapper to consolidate even more and fix this
issue.
This patch for DPDK Contributor's Guidelines indicates the repos
against which a new PMD should be prepared; for example, for new
network ethernet PMDs it should be dpdk-next-net, and for new crypto
PMDs it should be dpdk-next-crypto. For other new PMDs, the
contributor should refer to the MAINTAINERS file. Though this may seem
obvious, it is not mentioned in DPDK documentation.
Signed-off-by: Rami Rosen <ramirose@gmail.com> Acked-by: John McNamara <john.mcnamara@intel.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
Last contribution, last commit closing the release.
Thank you Rami, you will be missed.
This reverts commit aa2c00702bad7b2c742e11a86cb9dbbb8364fd88, which
introduced the possibility of an invalid address exception when running
an application with a stopped receive queue. The issues with rxq stop/start
will be revisited in the 19.11 release timeframe.
Matan Azrad [Tue, 6 Aug 2019 14:56:22 +0000 (14:56 +0000)]
doc: announce ethdev ABI change for LRO fields
It may be needed by the user to limit the LRO session packet size.
In order to allow the above limitation, a new Rx configuration may be
added for the maximum LRO session size.
A new capability may be added too to expose the maximum LRO session size
supported by the port.
Signed-off-by: Matan Azrad <matan@mellanox.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
Pavan Nikhilesh [Fri, 9 Aug 2019 09:55:59 +0000 (15:25 +0530)]
doc: announce ethdev API changes in offload flags
Add new offload flags ``DEV_RX_OFFLOAD_RSS_HASH`` and
``DEV_RX_OFFLOAD_FLOW_MARK``.
Add new function ``rte_eth_dev_set_supported_ptypes`` to allow application
to set specific ptypes to be updated in ``rte_mbuf::packet_type``
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Thomas Monjalon [Tue, 6 Aug 2019 13:31:35 +0000 (15:31 +0200)]
doc: announce removal of old port count function
The function rte_eth_dev_count() was marked as deprecated in DPDK 18.05
in commit d9a42a69febf ("ethdev: deprecate port count function").
It is planned to be removed after the next LTS release.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: David Marchand <david.marchand@redhat.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Jerin Jacob <jerinj@marvell.com>
Andrew Rybchenko [Tue, 23 Jul 2019 14:07:51 +0000 (15:07 +0100)]
doc: announce ethdev functions will not return void
void return value is bad for get API (like rte_eth_dev_info-get())
since caller does not know if the function does its job or not and
output value is filled in.
void return value is bad for state changing API (like
rte_eth_promiscuous_enable()) since caller should use get API
to understand if state is really changed.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Ferruh Yigit <ferruh.yigit@intel.com> Acked-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Jerin Jacob <jerinj@marvell.com>
Tell users about upcoming changes to rte_ether_addr and
rte_ether_header.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Ferruh Yigit <ferruh.yigit@intel.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
David Marchand [Thu, 8 Aug 2019 09:31:29 +0000 (11:31 +0200)]
doc: announce lcore config struct removal from ABI
New accessors have been introduced to provide the hidden information.
This symbol can now be kept internal.
Signed-off-by: David Marchand <david.marchand@redhat.com> Acked-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
David Marchand [Thu, 8 Aug 2019 09:32:06 +0000 (11:32 +0200)]
doc: announce malloc virt2phys function removal
This symbol has been deprecated for quite some time.
Let's drop it as soon as possible.
Signed-off-by: David Marchand <david.marchand@redhat.com> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
David Marchand [Thu, 8 Aug 2019 09:30:46 +0000 (11:30 +0200)]
doc: announce CPU flags check function removal
This symbol has been deprecated for quite some time.
Let's drop it as soon as possible.
Signed-off-by: David Marchand <david.marchand@redhat.com> Acked-by: Aaron Conole <aconole@redhat.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
Gagandeep Singh [Thu, 8 Aug 2019 12:13:20 +0000 (17:43 +0530)]
eal: increase maximum different hugepage sizes on Arm
ARM is supporting maximum 4 hugepage sizes (64K, 2M, 32M
and 1G) when granule is 4KB since very long and DPDK
support maximum 3 hugepage sizes.
With all 4 hugepage sizes enabled, applications and some
stacks like VPP which are working over DPDK and using
"in-memory" eal option, or using separate mount points
on ARM based platform, fails at huge page initialization,
reporting error messages from eal:
EAL: FATAL: Cannot get hugepage information.
EAL: Cannot get hugepage information.
EAL: Error - exiting with code: 1
This issue is originated from Linux 5.0
(a21b0b78eaf7 "arm64: hugetlb: Register hugepages during arch init")
where kernel is by default creating directories for each supported
hugepage size in /sys/kernel/mm/hugepages/
On earlier Stable Kernel LTR's, the directories visible in
/sys/kernel/mm/hugepages/ were dependent upon what hugepage
sizes are configured at boot time.
This change increases the maximum supported hugepage sizes
to 4 for ARM based platforms.
Congwen Zhang [Thu, 8 Aug 2019 00:37:20 +0000 (08:37 +0800)]
net/ixgbe/base: fix product version check
The "and" condition offset == 0 && offset == NVM_INVALID_PTR
can never be true.
Fixes: cf3af5aa56c9 ("net/ixgbe/base: add functions to get version info") Cc: stable@dpdk.org Signed-off-by: Congwen Zhang <zhang.congwen@zte.com.cn> Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
net/mlx5: fix completion request for multi-segment
The copying of sent mbufs pointers might be deferred to the end of
tx_burst() routine to be copied in one call of rte_memcpy.
For the multi segment packets this optimization is not applicable,
because number of packets does not match with number of mbufs and
we do not have linear array of pointers in pkts parameter.
The completion request generating routine wrongly took into account
the inconsistent (for multi-segment packets) deferred pointer copying.
Aaron Conole [Wed, 7 Aug 2019 14:58:44 +0000 (10:58 -0400)]
test/mcslock: wait for lcore completion
It's possible that the mcsunlock occurs before the test_mcslock_try has
a chance to execute, which will result in the trylock being successful,
making the test case fail. Fix this by waiting until all lcores have
completed their test before unlocking the master lock.
Fixes: 32dcb9fd2a22 ("test/mcslock: add MCS queued lock unit test") Signed-off-by: Aaron Conole <aconole@redhat.com> Acked-by: Phil Yang <phil.yang@arm.com> Reviewed-by: David Marchand <david.marchand@redhat.com>
Amit Gupta [Wed, 7 Aug 2019 13:12:09 +0000 (18:42 +0530)]
app/testpmd: fix latency stats deinit on signal
On receiving signal, testpmd showing warning as
"LATENCY_STATS: failed to remove Rx/Tx callback"
because rte_latencystats_uninit is called without
checking if latencystats is enabled or not.
After this fix, rte_latencystats_uninit will be
called only if latencystats is enabled.
Fixes: 62d3216d6194 ("app/testpmd: add latency statistics calculation") Cc: stable@dpdk.org Signed-off-by: Amit Gupta <agupta3@marvell.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Tested-by: Jerin Jacob <jerinj@marvell.com>
Jerin Jacob [Wed, 7 Aug 2019 03:11:34 +0000 (08:41 +0530)]
net/memif: fix build with gcc 9.1
gcc-9 is stricter on NULL arguments for printf.
Fix the following build error by avoiding NULL argument to printf.
In file included from drivers/net/memif/memif_socket.c:26:
In function 'memif_socket_create',
inlined from 'memif_socket_init' at net/memif/memif_socket.c:965:12:
net/memif/rte_eth_memif.h:35:2: error:
'%s' directive argument is null [-Werror=format-overflow=]
35 | rte_log(RTE_LOG_ ## level, memif_logtype, \
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
36 | "%s(): " fmt "\n", __func__, ##args)
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Thomas Monjalon [Tue, 30 Jul 2019 15:57:26 +0000 (17:57 +0200)]
doc: remove deprecated ethdev features
As legacy filter API "filter_ctrl" is superseded since 2017
by the rte_flow API, and got the deprecated attribute in DPDK 19.05,
it is time to remove the associated features from the matrix.
Not documenting deprecated features as supported will avoid confusion.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Acked-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
The shared Infiniband device context should be included
into memory event callback list only once on context creation,
and removed from the list only once on context destroying.
Multiple insertions of the same object caused the infinite
loop on the list processing.
Xiaoyu Min [Wed, 24 Jul 2019 11:31:56 +0000 (19:31 +0800)]
net/mlx5: fix RSS expand for IP-in-IP
The RSS expand function for IP-in-IP tunnel type is missed,
which leads to create following flow failed:
flow create 0 ingress pattern eth / ipv4 proto is 4 /
ipv4 / udp / end actions rss queues 0 1 end level 2
types ip ipv4-other udp ipv4 ipv4-frag end /
mark id 221 / count / end
In order to make RSS expand function working correctly,
now the way to check whether a IP tunnel existing is to
check whether there is the second IPv4/IPv6 item and whether the
first IPv4/IPv6 item's next protocl is IPPROTO_IPIP/IPPROTO_IPV6.
For example:
... pattern eth / ipv4 proto is 4 / ipv4 / ....
Fixes: 5e33bebdd8d3 ("net/mlx5: support IP-in-IP tunnel") Signed-off-by: Xiaoyu Min <jackmin@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Thomas Monjalon [Mon, 5 Aug 2019 15:32:20 +0000 (17:32 +0200)]
doc: remove useless console syntax in mlx guides
It is not needed to use "console" syntax highlighting
for literal blocks.
The file is easier to read by removing the code-block lines
and simply having double colons in previous line.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Raslan Darawsheh <rasland@mellanox.com>
Thomas Monjalon [Mon, 5 Aug 2019 15:32:18 +0000 (17:32 +0200)]
doc: improve firmware configuration in mlx5 guide
The command mlxconfig was not enough explained and too much verbose
at the same time.
The syntax is now explained in introduction before listing the options,
without repeating the commands.
Some options, which are explained elsewhere in the doc,
are added to this list.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Raslan Darawsheh <rasland@mellanox.com>
Dekel Peled [Mon, 5 Aug 2019 11:51:19 +0000 (14:51 +0300)]
net/mlx5: fix UDP checksum zeroing
Function flow_dv_zero_encap_udp_csum() uses a while loop to iterate
over vlan items in flow rule.
Pointer next_hdr is incremented to the next item before it is used,
so the first item is skipped.
This patch moves the incrementing of next_hdr to the correct place.
Fixes: bf1d7d9a033a ("net/mlx5: zero out UDP checksum in encapsulation") Signed-off-by: Dekel Peled <dekelp@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Xiaoyu Min [Mon, 5 Aug 2019 08:02:03 +0000 (16:02 +0800)]
net/mlx5: fix link speed info when link is down
When the link is down, the link speed returned by ethtool is
UINT32_MAX and the link status is 0.
In this case, the DPDK ethdev link speed should be set to
ETH_SPEED_NUM_NONE.
Otherwise since link speed is non-zero but link status is zero, this
is an inconsistent situation and -EAGAIN is returned, which is not right.
Fixes: 188408719888 ("net/mlx5: fix support for newer link speeds") Cc: stable@dpdk.org Signed-off-by: Xiaoyu Min <jackmin@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
net/mlx5: fix completion queue overflow for large burst
There is the limit on completion descriptor fetch to improve
latency. If burst size is large there might be not enough
resources freed in completion processing. This fix reiterates
sending loop and allows multiple completion descriptor fetch
and processing.
This patch fixes the default settings for packet size to inline
with Enhanced Multi-Packet Write feature, allowing 256B packets
to be inlined with Out-Of-the-Box settings.
If the minimal inline data are required the data inline feature
must be engaged. There were the incorrect settings enabling the
entire small packet inline (in size up to 82B) which may result
in sending rate declining if there is no enough cores. The same
problem was raised if inline was enabled to support VLAN tag
insertion by software.
The completion loop speed optimizations for error-free
operations are done - no CQE field fetch on each loop
iteration. Also, code size is oprimized - the flush
buffers routine is invoked once.
The patch [Fixes] sets the default value of required minimal
inline data to 0 bytes. On some configurations (depends
on switchdev/legacy settings and FW version/settings)
the ConnectX-4LX NIC requires minimal 18 bytes of
Tx descriptor inline data to operate correctly.
Wrongly set to 0 default value may prevent NIC from operating
with out-of-the-box settings, this patch reverts default
value for ConnectX-4LX back to 18 bytes (inline L2).
Dekel Peled [Mon, 5 Aug 2019 07:07:05 +0000 (10:07 +0300)]
net/mlx5: fix flow rule configuration
Some flow rules were not configured.
Part of the code in function flow_dv_matcher_enable() is enclosed in
'#ifdef HAVE_MLX5DV_DR' preprocessor directive.
Using this directive is not needed here, and prevents compilation of
relevant code.
This patch removes the unnecessary preprocessor directive.
Function mlx5_flow_validate_item_vlan() validates the user setting
is supported by NIC, using a mask with TCI mask 0x0fff.
This check will reject a flow rule specifying a vlan pcp item.
This patch updates mlx5_flow_validate_item_vlan() to use mask 0xffff,
so flow rules with vlan pcp item are accepted.
net/mlx4: fix crash on info query in secondary process
mlx4_dev_info_get calls mlx4_get_ifname, but mlx4_get_ifname
uses priv->ctx which is not a valid pointer in a secondary
process. The fix is to cache the value in primary.
In the primary process, get and store the interface index of
the device so that secondary process can see it.
Bugzilla ID: 320 Fixes: 61cbdd419478 ("net/mlx4: separate device control functions") Cc: stable@dpdk.org Reported-by: Suyang Ju <sju@paloaltonetworks.com> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Matan Azrad <matan@mellanox.com>
The function mlx5_set_min_inline() includes a switch() that checks
various PCI device IDs in order to set the txq_inline_min value. No
value is set when the PCI device ID matches the ConnectX-5 adapters,
resulting in an assert() failure later in the function
mlx5_set_txlimit_params().
This error was encountered on an IBM Power 9 system running RHEL 7.6
w/o Mellanox OFED installed.
Fixes: 38b4b397a57d ("net/mlx5: add Tx configuration and setup") Signed-off-by: David Christensen <drc@linux.vnet.ibm.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
net/mlx5: add workaround for VLAN in virtual machine
On some virtual setups (particularly on ESXi) when we have SR-IOV and
E-Switch enabled there is the problem to receive VLAN traffic on VF
interfaces. The NIC driver in ESXi hypervisor does not setup E-Switch
vport setting correctly and VLAN traffic targeted to VF is dropped.
The patch provides the temporary workaround - if the rule
containing the VLAN pattern is being installed for VF the VLAN
network interface over VF is created, like the command does:
ip link add link vf.if name mlx5.wa.1.100 type vlan id 100
The PMD in DPDK maintains the database of created VLAN interfaces
for each existing VF and requested VLAN tags. When all of the RTE
Flows using the given VLAN tag are removed the created VLAN interface
with this VLAN tag is deleted.
The name of created VLAN interface follows the format:
evmlx.d1.d2, where d1 is VF interface ifindex, d2 - VLAN ifindex
Implementation limitations:
- mask in rules is ignored, rule must specify VLAN tags exactly,
no wildcards (which are implemented by the masks) are allowed
- virtual environment is detected via rte_hypervisor() call,
and the type of hypervisor is checked. Currently we engage
the workaround for ESXi and unrecognized hypervisors (which
always happen on platforms other than x86 - it means workaround
applied for the Flow over PCI VF). There are no confirmed data
the other hypervisors (HyperV, Qemu) need this workaround,
we are trying to reduce the list of configurations on those
workaround should be applied.
Xiao Zhang [Mon, 5 Aug 2019 23:20:54 +0000 (07:20 +0800)]
net/fm10k: fix address of first segment
This patch fixes (dereference after null check) coverity issue.
The address of first segmented packets was not set correctly during
reassembling packets which led to this issue.
Xiao Zhang [Mon, 5 Aug 2019 23:20:53 +0000 (07:20 +0800)]
net/iavf: fix address of first segment
This patch fixes (dereference after null check) coverity issue.
The address of first segmented packets was not set correctly during
reassembling packets which led to this issue.
Xiao Zhang [Mon, 5 Aug 2019 23:20:52 +0000 (07:20 +0800)]
net/i40e: fix address of first segment
This patch fixes (dereference after null check) coverity issue.
The address of first segmented packets was not set correctly during
reassembling packets which led to this issue.
Coverity issue: 343422, 343403 Fixes: ca74903b75cf ("net/i40e: extract non-x86 specific code from vector driver") Cc: stable@dpdk.org Signed-off-by: Xiao Zhang <xiao.zhang@intel.com> Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
Xiao Zhang [Mon, 5 Aug 2019 23:20:51 +0000 (07:20 +0800)]
net/ice: fix address of first segment
This patch fixes (dereference after null check) coverity issue.
The address of first segmented packets was not set correctly during
reassembling packets which led to this issue.
Coverity issue: 343452, 343407 Fixes: c68a52b8b38c ("net/ice: support vector SSE in Rx") Cc: stable@dpdk.org Signed-off-by: Xiao Zhang <xiao.zhang@intel.com> Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
Xiao Zhang [Mon, 5 Aug 2019 23:20:50 +0000 (07:20 +0800)]
net/ixgbe: fix address of first segment
This patch fixes (dereference after null check) coverity issue.
The address of first segmented packets was not set correctly during
reassembling packets which led to this issue.
Coverity issue: 13245 Fixes: 8a44c15aa57d ("net/ixgbe: extract non-x86 specific code from vector driver") Cc: stable@dpdk.org Signed-off-by: Xiao Zhang <xiao.zhang@intel.com> Acked-by: Xiaolong Ye <xiaolong.ye@intel.com>
Example BPF programs t2.c, t3.c in folder examples/bpf are
failing to compile with latest dpdk.org master.
The reason is changes in some core DPDK header files, that causes
now inclusion of x86 specific headers.
To overcome the issue, minimize inclusion of DPDK header files
into BPF source code.
Bugzilla ID: 321
Fixes: 9dfc06c26a8b ("test/bpf: add samples") Cc: stable@dpdk.org Reported-by: Michel Machado <michel@digirati.com.br> Suggested-by: Michel Machado <michel@digirati.com.br> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Xiaoyun Li [Mon, 5 Aug 2019 05:57:28 +0000 (13:57 +0800)]
examples/ntb: fix error handling
This patch adds return value checking for fseek function to fix
error handling issue found by coverity scan.
Coverity issue: 344996 Fixes: c5eebf85badc ("examples/ntb: add example for NTB") Cc: stable@dpdk.org Signed-off-by: Xiaoyun Li <xiaoyun.li@intel.com> Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
Vamsi Attunuru [Fri, 2 Aug 2019 06:57:16 +0000 (12:27 +0530)]
common/octeontx2: fix mbox memory access
Octeontx2 PMD's mailbox client uses device memory to send messages
to mailbox server in the admin function Linux kernel driver.
The device memory used for the mailbox communication needs to
be qualified as volatile memory type to avoid unaligned device
memory accesses because of compiler's memory access coalescing.
This patch modifies the mailbox request and responses as volatile
type which were non-volatile earlier and accessed from unaligned
memory addresses which resulted in bus errors on Fedora 30 with
gcc 9.1.1.
git.droids-corp.org / dpdk.git / log