Zhiyong Yang [Tue, 26 Dec 2017 09:25:00 +0000 (17:25 +0800)]
net/virtio: remove unnecessary macro definitions
DPDK has already the definition of Ethernet numeric link speeds in Mbps
in the file Rte_ethdev.h, it is unnecessary to rededine virtio specific
link speeds macros again.
Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com> Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
Junjie Chen [Wed, 13 Dec 2017 16:50:56 +0000 (11:50 -0500)]
vhost: fix dequeue zero copy with virtio1
This fix dequeue zero copy can not work with Qemu
version >= 2.7. Since from Qemu 2.7 virtio device
use virtio-1 protocol, the zero copy code path
forget to add offset to buffer address.
Fixes: b0a985d1f340 ("vhost: add dequeue zero copy") Cc: stable@dpdk.org Signed-off-by: Junjie Chen <junjie.j.chen@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com> Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
Jianfeng Tan [Wed, 15 Nov 2017 11:41:08 +0000 (11:41 +0000)]
vhost: fix crash
In a running VM, operations (like device attach/detach) will
trigger the QEMU to resend set_mem_table to vhost-user backend.
DPDK vhost-user handles this message rudely by unmap all existing
regions and map new ones. This might lead to segfault if there
is pmd thread just trying to touch those unmapped memory regions.
But for most cases, except VM memory hotplug, QEMU still sends the
set_mem_table message even the memory regions are not changed as
QEMU vhost-user filters out those not backed by file (fd > 0).
To fix this case, we add a check in the handler to see if the
memory regions are really changed; if not, we just keep old memory
regions.
Fixes: 8f972312b8f4 ("vhost: support vhost-user") CC: stable@dpdk.org Reported-by: Yang Zhang <zy107165@alibaba-inc.com> Reported-by: Xin Long <longxin.xl@alibaba-inc.com> Signed-off-by: Yi Yang <yi.y.yang@intel.com> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com> Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
Radu Nicolau [Thu, 21 Dec 2017 10:55:50 +0000 (10:55 +0000)]
net/ixgbe: remove ipsec keys from private data
All ipsec related settings are being held in the driver
private data to allow easy add and remove of SAS. There
is no need to keep a record of the keys, and also
storing the keys can be a security issue.
Signed-off-by: Radu Nicolau <radu.nicolau@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Beilei Xing [Thu, 4 Jan 2018 10:46:21 +0000 (18:46 +0800)]
net/i40e: improve packet type parser
Add support for PPPoE and L2TP packet types parser.
Change parser as new metadata 'OIPV4' and 'OIPV6'
added in profile to distinguish outer IP and inner
IP.
Wei Zhao [Fri, 5 Jan 2018 06:13:50 +0000 (14:13 +0800)]
net/ixgbe: add ntuple support to flow parser
ixgbe ntuple filter in rte_flow needs to support diversion data
with less than 5 tuple parameters. So it adds this new support
in parser code. This patch also adds parsering vlan pattern code
in ntuple filter in order to handle some cases like vlan in flow
API.
Signed-off-by: Wei Zhao <wei.zhao1@intel.com> Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Beilei Xing [Mon, 18 Dec 2017 05:20:57 +0000 (13:20 +0800)]
net/i40e: fix FDIR input set conflict
The first FDIR rule for some PCTYPE will configure input set and
create flow, the following flows must use the same input set,
otherwise it will cause input set conflict and fail to create flow.
If it creates the first rule after flow flush, input set should be
re-configured.
Fixes: 42044b69c67d ("net/i40e: support input set selection for FDIR") Cc: stable@dpdk.org Signed-off-by: Beilei Xing <beilei.xing@intel.com> Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Beilei Xing [Fri, 8 Dec 2017 07:40:33 +0000 (15:40 +0800)]
net/i40e: fix VLAN offload setting issue
When using kernel PF + DPDK VF, if setting VLAN strip on
or off in VF side after setting PVID for VF with ethtool in
PF side, it will cause VF Tx error. The root cause is that
Rx VLAN offload is not permitted after setting PVID for VF
in kernel PF side.
This patch fixes the issue by check VLAN offload capability
when setting VLAN offload.
Fixes: 5f0b95d59a98 ("net/i40e: support VLAN stripping for VF") Cc: stable@dpdk.org Signed-off-by: Beilei Xing <beilei.xing@intel.com> Acked-by: Jingjing Wu <jingjing.wu@intel.com>
Yangchao Zhou [Mon, 4 Dec 2017 06:38:09 +0000 (14:38 +0800)]
net/fm10k: remove RSS restriction with num of queues
FM10K HW does not have such restrictions.
Enabling RSS with single queue is not used to distribute flow, but
it can be used to compute a RSS hash value. It can reduce CPU
cycles of computing a hash value with five tuples. In addition,
there is an explicit method to disable RSS instead of an obscure
way.
Wenzhuo Lu [Mon, 20 Nov 2017 03:37:45 +0000 (11:37 +0800)]
net/ixgbe: fix wrong PBA setting
The datasheet says, if using MSI-X mode, the PBA support
bit of the GPIE register must be set to 1.
DPDK uses polling mode, we cannot hit this issue in the
scenario DPDK PF + DPDK VF. If we use DPDK PF + kernel VF,
as the kernel driver uses interrpt mode, VF may hit RX hang
after running hours.
Fixes: 00e30184daa0 ("ixgbe: add PF support") Cc: stable@dpdk.org Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com> Acked-by: Wei Dai <wei.dai@intel.com>
Xiao Wang [Sat, 9 Dec 2017 01:55:41 +0000 (17:55 -0800)]
net/fm10k: fix logical port delete
When closing port, we need to send mailbox messages to switch manager
to reset multicast mode and delete logical port. In the latest IES_SDK,
e.g. v4.3.3, switch takes longer time to handle these mailbox messages.
So this patch adds longer delay to accommodate this change. Otherwise,
the mailbox will be closed before all the messages are handled, as a
result, the logical port remains in switch manager after port is closed.
Fixes: b961fe9344dd ("net/fm10k: fix MAC address removal from switch") Cc: stable@dpdk.org Signed-off-by: Xiao Wang <xiao.w.wang@intel.com> Acked-by: Helin Zhang <helin.zhang@intel.com>
Qi Zhang [Wed, 29 Nov 2017 10:27:30 +0000 (05:27 -0500)]
net/i40e: fix VLAN offload setting
Vlan offload flag "rx_mode.hw_vlan_extend" should be considerred at
rte_eth_dev_configure stage as well as "rx_mode.hw_vlan_strip" and
"rx_mode.hw_vlan_filter".
Shahaf Shuler [Tue, 26 Dec 2017 07:40:41 +0000 (09:40 +0200)]
net/mlx5: fix RSS key configuration
The trigger for PMD to use the application RSS configuration should be
based on the validity of rss_key and not the rss_key_len. Otherwise
segmentation fault can occur if application provide valid RSS key length
but without any RSS key.
Fixes: 29c1d8bb3e79 ("net/mlx5: handle a single RSS hash key for all protocols") Cc: stable@dpdk.org Signed-off-by: Shahaf Shuler <shahafs@mellanox.com> Acked-by: Yongseok Koh <yskoh@mellanox.com>
Yongseok Koh [Wed, 27 Dec 2017 03:55:46 +0000 (19:55 -0800)]
net/mlx5: add fallback in Tx for multi-segment packet
mlx5_tx_burst_empw() falls back to legacy Tx descriptor for multi-segmented
packets without taking advantage of inlining. In many cases, the 1st
segment can be inlined and this could make device fetch only one segment
instead of two. This helps saving PCIe bandwidth when transmitting out
multi-segmented packets with still using the Enhanced Multi-Packet Send for
other packets.
Yongseok Koh [Wed, 27 Dec 2017 03:55:43 +0000 (19:55 -0800)]
net/mlx5: enable inlining data from multiple segments
mlx5_tx_burst() doesn't inline data from the 2nd segment. If there's still
enough room in the descriptor after inlining the 1st segment, further
inlining from the 2nd segment would be beneficial to save PCIe bandwidth.
Yongseok Koh [Fri, 15 Dec 2017 01:59:18 +0000 (17:59 -0800)]
net/mlx5: fix overflow of Memory Region cache
If there're more MR(Memroy Region)'s than the size of per-queue cache, the
cache can be overflowed and corrupt the following data structure in
mlx5_txq_data.
Yongseok Koh [Fri, 15 Dec 2017 01:59:17 +0000 (17:59 -0800)]
net/mlx5: fix Memory Region registration
Although granularity of chunks in a mempool is a cacheline, addresses are
extended to align to page boundary for performance reason in device when
registering a MR (Memory Region). This could make some regions overlap,
then can cause Tx completion error due to incorrect LKEY search. If the
error occurs, the Tx queue will get stuck. To avoid it, end address of a
packet segment is used in LKEY search.
Ivan Malov [Wed, 20 Dec 2017 09:52:14 +0000 (09:52 +0000)]
net/sfc: fix main MAC address handling
There is a school of thought that rte_eth_dev_default_mac_addr_set()
must call the PMD callback first and then save the new MAC address
in dev->data->mac_addrs[0]. If this concept gets approved, it will
break the current approach used in sfc driver as the latter relies
on the assumption that the new MAC address is already contained in
dev->data->mac_addrs[0], and, if adapter restart is needed to make
the HW apply the new address, the outdated value will be retrieved
from dev->data. In order to preclude any possible bugs, this patch
adds device private storage for the up-to-date copy of the address.
Fixes: c100fd464bb7 ("net/sfc: support main MAC address change") Cc: stable@dpdk.org Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Pablo de Lara [Tue, 19 Dec 2017 11:31:50 +0000 (11:31 +0000)]
ethdev: fix typo in functions comment
Fixes: fbde27f19ab8 ("ethdev: get default Rx/Tx configuration from dev info") Cc: stable@dpdk.org Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
Olivier Matz [Thu, 14 Dec 2017 14:23:00 +0000 (15:23 +0100)]
ethdev: fix missing imissed counter in xstats
imissed counter has been set as deprecated in commit 49f386542af4
("ethdev: remove driver specific stats") and removed from the
rte_eth_xstats_name_off structure.
The imissed counter has been restored few commits later but has not been
restored in the rte_eth_stats structure. Add it back.
Andrew Rybchenko [Wed, 13 Dec 2017 07:12:18 +0000 (07:12 +0000)]
net/sfc: fix DMA memory leak after kvarg processing failure
Fixes: e56fa9c23e7a ("net/sfc: add kvarg control for MAC statistics update period") Cc: stable@dpdk.org Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Andrew Rybchenko [Wed, 13 Dec 2017 07:12:17 +0000 (07:12 +0000)]
net/sfc: fix multicast address list copy memory leak
Fixes: 295f647a38a2 ("net/sfc: set multicast address list in started state only") Cc: stable@dpdk.org Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Andrew Rybchenko [Wed, 13 Dec 2017 07:12:16 +0000 (07:12 +0000)]
net/sfc: stop periodic DMA if MAC stats upload fails
Fixes: 3b257f7e6c0f ("net/sfc: request MAC stats upload immediately on port start") Cc: stable@dpdk.org Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Shahed Shaikh [Thu, 14 Dec 2017 06:36:03 +0000 (22:36 -0800)]
net/qede: add support for GENEVE tunneling offload
This patch refactors existing VXLAN tunneling offload code and enables
following features for GENEVE:
- destination UDP port configuration
- checksum offloads
- filter configuration
The previuse code took a send queue entry size for stamping from the
send queue entry pointed by completion queue entry; This 2 reads were
done per packet in completion stage.
The completion burst packets number is managed by fixed size stored in
Tx queue, so we can infer that each valid completion entry actually frees
the next fixed number packets.
The descriptors ring holds the send queue entry, so we just can infer
all the completion burst packet entries size by simple calculation and
prevent calculations per packet.
Adjust completion functions to free full completion bursts packets
by one time and prevent per packet work queue entry reads and
calculations.
Save only start of completion burst or Tx burst send queue entry
pointers in the appropriate descriptor element.
Matan Azrad [Wed, 6 Dec 2017 17:57:53 +0000 (17:57 +0000)]
net/mlx4: merge Tx queue rings management
The Tx queue send ring was managed by Tx block head,tail,count and mask
management variables which were used for managing the send queue remain
space and next places of empty or completed work queue entries.
This method suffered from an actual addresses recalculation per packet,
an unnecessary Tx block based calculations and an expensive dual
management of Tx rings.
Move send queue ring calculation to be based on actual addresses while
managing it by descriptors ring indexes.
Add new work queue entry pointer to the descriptor element to hold the
appropriate entry in the send queue.
Matan Azrad [Wed, 6 Dec 2017 17:57:52 +0000 (17:57 +0000)]
net/mlx4: optimize Tx multi-segment case
mlx4 Tx block can handle up to 4 data segments or control segment + up
to 3 data segments. The first data segment in each not first Tx block
must validate Tx queue wraparound and must use IO memory barrier before
writing the byte count.
The previous multi-segment code used "for" loop to iterate over all
packet segments and separated first Tx block data case by "if"
statements.
Use switch case and unconditional branches instead of "for" loop can
optimize the case and prevents the unnecessary checks for each data
segment; This hints to compiler to create optimized jump table.
Optimize this case by switch case and unconditional branches usage.
Matan Azrad [Wed, 6 Dec 2017 17:57:49 +0000 (17:57 +0000)]
net/mlx4: fix Tx packet drop application report
When invalid lkey is sent to HW, HW sends an error notification in
completion function.
The previous code wouldn't crash but doesn't add any application report
in case of completion error, so application cannot know that packet
actually was dropped in case of invalid lkey.
Return back the lkey validation to Tx path.
Fixes: 2eee458746bc ("net/mlx4: remove error flows from Tx fast path") Cc: stable@dpdk.org Signed-off-by: Matan Azrad <matan@mellanox.com> Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Nipun Gupta [Fri, 8 Dec 2017 05:21:27 +0000 (10:51 +0530)]
net/dpaa2: prefetch the parse results from next fd
As there were less number of CPU cycles in between when we access
the parse results and when we prefetch them, this patch modifies
the prefetch of the next parse result.
Nipun Gupta [Fri, 8 Dec 2017 05:21:25 +0000 (10:51 +0530)]
net/dpaa2: optimize Rx path packet parsing
Parsing the annotation has multiple if checks in the data path.
These are reduced for common cases like IPv4/IPv6 and UDP/TCP
packets to enhance performance of these generic cases.
Ashish Jain [Fri, 8 Dec 2017 05:21:22 +0000 (10:51 +0530)]
net/dpaa2: align the frame size in MTU set
Setting correct frame size in dpaa2_dev_mtu_set
api call. Also correcting the correct max frame size
setting in hardware while dev_configure for jumbo frames
On behalf of the DPDK Technical board, Hemant Agrawal observed that the
DPDK project's Intellectual Property Policy (http://dpdk.org/about/charter)
requires 3-Clause BSD license or an exception approval. However, two
vmxnet3 source files have 2-Clause BSD license.
This patch modifies those licenses to 3-Clause BSD license.
Declan Doherty [Thu, 30 Nov 2017 23:25:35 +0000 (23:25 +0000)]
net/bonding: do not drop LACPDUs on slaves Tx failure
In message (http://dpdk.org/ml/archives/dev/2017-November/081557.html)
it was noted that under congestion that the LACPDUs are dropped under
load.
This patch changes the drop logic to re-enqueue the LACPDU to the slaves
control message queue. This will allow resend attempts to be made in
subsequent tx_burst() calls on the bonded device.
Adrien Mazarguil [Thu, 23 Nov 2017 17:38:04 +0000 (18:38 +0100)]
net/mlx4: restore inner VXLAN RSS support
Inner VXLAN RSS was supported and performed by default prior to the entire
mlx4 refactoring that occurred in DPDK 17.11, however so far the new Verbs
RSS API did not provide means to enable it. This will be addressed in
Linux 4.15 and in RDMA core.
Thanks to RSS capabilities, the PMD can now probe for its support and
enable it again by default.
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
Adrien Mazarguil [Thu, 23 Nov 2017 17:38:02 +0000 (18:38 +0100)]
net/mlx4: restore UDP RSS by probing capabilities
Until now, UDP RSS support could not be relied on due to a problem in the
Linux kernel implementation and mlx4 RSS capabilities were not reported at
all, hence the PMD had to make assumptions.
Since both issues will be addressed simultaneously in Linux 4.15 (related
patches already upstream) and likely backported afterward, UDP RSS support
can be enabled by probing RSS capabilities.
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
Adrien Mazarguil [Thu, 23 Nov 2017 17:38:00 +0000 (18:38 +0100)]
net/mlx4: use function to get default RSS fields
Supported RSS hash fields are listed in function mlx4_conv_rss_hf() and
duplicated in mlx4_flow_prepare(); the latter are used when RSS is
requested without specifying any parameters.
This commit standardizes on mlx4_conv_rss_hf().
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
Nélio Laranjeiro [Mon, 20 Nov 2017 15:35:47 +0000 (16:35 +0100)]
net/mlx5: fix Tx checksum offloads
Tx checksum offloads are correctly handled in a single Tx burst function
whereas the capability is always set.
This causes VXLAN packet with checksum offloads request to be ignored when
the (E)MPS Tx functions are selected.
Alejandro Lucero [Fri, 24 Nov 2017 15:31:49 +0000 (15:31 +0000)]
net/nfp: configure default RSS reta table
Some apps can enable RSS but not update the reta table nor the hash.
This patch adds a default reta table setup based on total number of
configured rx queues. The hash key is dependent on how the app
configures the rx_conf struct.
Alejandro Lucero [Fri, 24 Nov 2017 14:26:02 +0000 (14:26 +0000)]
net/nfp: fix CRC strip check behaviour
NFP does CRC strip by default and it is not configurable. But, even
if an app requests not to do it, that should not be a reason for PMD
configuration failure.
Alejandro Lucero [Fri, 24 Nov 2017 14:24:37 +0000 (14:24 +0000)]
net/nfp: fix jumbo settings
When jumbo frames is configured, the hardware mtu needs to be updated to
the specified max_rx_pkt_len. Also, changing mtu should be avoided once
the PMD port started.
Nélio Laranjeiro [Thu, 23 Nov 2017 09:22:34 +0000 (10:22 +0100)]
net/mlx5: remove 32-bit support
naddr variable was introduced in
commit 9a7fa9f76d9e ("net/mlx5: use vector types to speed up processing")
to avoid compilation errors on 32bits compilation, as x86_32 is no more
supported by rdma-core nor by MLNX_OFED, this variable becomes useless and
can be safely removed.
git.droids-corp.org / dpdk.git / log