Thomas Monjalon [Fri, 29 Jul 2016 13:34:29 +0000 (15:34 +0200)]
doc: postpone mempool ABI breakage
It was planned to remove some mempool functions which are deprecated
since 16.07.
As no other mempool ABI change is planned in 16.11, it is better
to postpone and group every mempool ABI changes in 17.02.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
John McNamara [Fri, 29 Jul 2016 11:59:14 +0000 (12:59 +0100)]
doc: generate NIC overview table from ini files
Convert the NIC feature table in the overview doc into a set of ini
files and add functions into the Sphinx conf.py file to auto-generate
them back into an RST table.
The reason for doing this is to make it easier for PMD maintainers to
update the feature matrix that makes up the table and to avoid
frequent and hard to resolve conflicts in doc/guides/nics/overview.rst.
A NIC/PMD feature matrix is now an ini file like the following:
$ head doc/guides/nics/nic_features/i40e.ini
;
; Features of the i40e network driver.
;
[Features]
Link status = Y
Link status event = Y
Rx interrupt = Y
Queue start/stop = Y
...
The output RST table matches the existing table with the column
headers sorted.
Signed-off-by: John McNamara <john.mcnamara@intel.com> Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>
Yuanhan Liu [Fri, 15 Jul 2016 12:28:33 +0000 (20:28 +0800)]
doc: announce vhost-cuse removal
Vhost-cuse was invented before vhost-user exist. The both are actually
doing the same thing: a vhost-net implementation in user space. But they
are not exactly the same thing.
Firstly, vhost-cuse is harder for use; no one seems to care it, either.
Furthermore, since v2.1, a large majority of development effort has gone
to vhost-user. For example, we extended the vhost-user spec to add the
multiple queue support. We also added the vhost-user live migration at
v16.04 and the latest one, vhost-user reconnect that allows vhost app
restart without restarting the guest. Both of them are very important
features for product usage and none of them works for vhost-cuse.
You now see that the difference between vhost-user and vhost-cuse is
big (and will be bigger and bigger as time moves forward), that you
should never use vhost-cuse, that we should drop it completely.
The remove would also result to a much cleaner code base, allowing us
to do all kinds of extending easier.
So here to mark vhost-cuse as deprecated in this release and will be
removed in the next release (v16.11).
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com> Acked-by: Ciara Loftus <ciara.loftus@intel.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: Rich Lane <rich.lane@bigswitch.com> Acked-by: Jan Viktorin <viktorin@rehivetech.com> Acked-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Thomas Monjalon [Wed, 20 Jul 2016 16:35:46 +0000 (18:35 +0200)]
doc: announce ivshmem support removal
There was a prior call with an explanation of what needs to be done:
http://dpdk.org/ml/archives/dev/2016-June/040844.html
- Qemu patch upstreamed
- IVSHMEM PCI device managed by a PCI driver
- No DPDK objects (ring/mempool) allocated by EAL
As nobody seems interested, it is time to remove this code which
makes EAL improvements harder.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: David Marchand <david.marchand@6wind.com> Acked-by: Maxime Coquelin <maxime.coquelin@redhat.com> Acked-by: Jan Viktorin <viktorin@rehivetech.com> Acked-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
For 16.11, the mbuf structure will be modified implying ABI breakage.
Some discussions already took place here:
http://www.dpdk.org/dev/patchwork/patch/12878/
Signed-off-by: Olivier Matz <olivier.matz@6wind.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: John Daley <johndale@cisco.com> Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Different NIC models depending on HW offload requested might impose
different requirements on packets to be TX-ed in terms of:
- Max number of fragments per packet allowed
- Max number of fragments per TSO segments
- The way pseudo-header checksum should be pre-calculated
- L3/L4 header fields filling
- etc.
MOTIVATION:
-----------
1) Some work cannot (and didn't should) be done in rte_eth_tx_burst.
However, this work is sometimes required, and now, it's an
application issue.
2) Different hardware may have different requirements for TX offloads,
other subset can be supported and so on.
3) Some parameters (eg. number of segments in ixgbe driver) may hung
device. These parameters may be vary for different devices.
For example i40e HW allows 8 fragments per packet, but that is after
TSO segmentation. While ixgbe has a 38-fragment pre-TSO limit.
4) Fields in packet may require different initialization (like eg. will
require pseudo-header checksum precalculation, sometimes in a
different way depending on packet type, and so on). Now application
needs to care about it.
5) Using additional API (rte_eth_tx_prep) before rte_eth_tx_burst let to
prepare packet burst in acceptable form for specific device.
6) Some additional checks may be done in debug mode keeping tx_burst
implementation clean.
PROPOSAL:
---------
To help user to deal with all these varieties we propose to:
1. Introduce rte_eth_tx_prep() function to do necessary preparations of
packet burst to be safely transmitted on device for desired HW
offloads (set/reset checksum field according to the hardware
requirements) and check HW constraints (number of segments per
packet, etc).
While the limitations and requirements may differ for devices, it
requires to extend rte_eth_dev structure with new function pointer
"tx_pkt_prep" which can be implemented in the driver to prepare and
verify packets, in devices specific way, before burst, what should to
prevent application to send malformed packets.
2. Also new fields will be introduced in rte_eth_desc_lim:
nb_seg_max and nb_mtu_seg_max, providing an information about max
segments in TSO and non-TSO packets acceptable by device.
This information is useful for application to not create/limit
malicious packet.
APPLICATION (CASE OF USE):
--------------------------
1) Application should to initialize burst of packets to send, set
required tx offload flags and required fields, like l2_len, l3_len,
l4_len, and tso_segsz
2) Application passes burst to the rte_eth_tx_prep to check conditions
required to send packets through the NIC.
3) The result of rte_eth_tx_prep can be used to send valid packets
and/or restore invalid if function fails.
Thomas Monjalon [Tue, 26 Jul 2016 16:22:21 +0000 (18:22 +0200)]
doc: announce renaming of ethdev library
The right name of ethdev should be dpdk_netdev. However:
1/ We are using rte_ prefix in the code and library names.
2/ The API uses rte_ethdev
That's why 16.11 will just have the rte_ prefix prepended to
the library filename as every other libraries.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: Jan Viktorin <viktorin@rehivetech.com> Acked-by: Christian Ehrhardt <christian.ehrhardt@canonical.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Pablo de Lara [Sat, 9 Jul 2016 16:56:34 +0000 (17:56 +0100)]
doc: announce driver name changes
Driver names for all the supported devices in DPDK do not have
a naming convention. Some are using a prefix, some are not
and some have long names. Driver names are used when creating
virtual devices, so it is useful to have consistency in the names.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: Maxime Coquelin <maxime.coquelin@redhat.com> Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Jeff Guo [Wed, 27 Jul 2016 02:56:34 +0000 (22:56 -0400)]
doc: add known issue for promiscuous mode in i40e VF
When use i40e linux kernel driver as host driver and DPDK handler the i40e
VF, the promiscuous mode doesn't work in i40e VF. It is not supported by
DPDK i40e VF driver right now.
Signed-off-by: Jeff Guo <jia.guo@intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
Dumitru Ceara [Tue, 26 Jul 2016 10:46:09 +0000 (12:46 +0200)]
net/i40e: fix metadata in first mbuf of scattered Rx
The driver is incorrectly setting the RSS field in the last mbuf in
the packet chain instead of the first. Moreover, the last mbuf might
have already been freed if it only contained the Ethernet CRC.
Also, fix the call to i40e_rxd_build_fdir to store the fdir flags in
the first mbuf of the chain instead of the last.
Fixes: 4861cde46116 ("i40e: new poll mode driver") Fixes: 5a21d9715f81 ("i40e: report flow director matching") Signed-off-by: Dumitru Ceara <dumitru.ceara@gmail.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
This configuration is example configuration for flow classification.
This fix changes the offset and mask value to compute the hash correctly.
This fix does not involve code change and do not impact compilation,
build and performance.
Wei Dai [Wed, 27 Jul 2016 11:25:56 +0000 (19:25 +0800)]
eal: fix tail blank check in --lcores argument
the tail blank after a group of lcore or cpu set
will make check of its end character fail.
for example: --lcores '(0-3)@(0-3) ,(4-5)@(4-5)',
the next character after cpu set (0-3) is not ','
or '\0', which fail the check in eal_parse_lcores( ).
Fixes: 53e54bf81700 ("eal: new option --lcores for cpu assignment") Signed-off-by: Wei Dai <wei.dai@intel.com> Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
Wei Dai [Wed, 27 Jul 2016 11:23:41 +0000 (19:23 +0800)]
eal: fix parsing of option --lcores
The '-' in lcore set overrides cpu set of following
lcore set in the argument of EAL option --lcores.
for example --locres '0-2,(3-5)@(3,4),6@(5,6),7@(5-7)',
0-2 make lflags=1 which indeed suppress following
cpu set (3,4), (5,6) and (5-7) after @ .
Fixes: 53e54bf81700 ("eal: new option --lcores for cpu assignment") Signed-off-by: Wei Dai <wei.dai@intel.com> Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
Thomas Monjalon [Mon, 25 Jul 2016 19:32:03 +0000 (21:32 +0200)]
mempool: fix unsafe removal from list by callback
If a mempool is removed from the list by a callback function
during rte_mempool_walk(), the TAILQ_FOREACH loop will fail unexpectedly.
It is fixed by using the safe version of the loop macro.
Thomas Monjalon [Mon, 25 Jul 2016 12:56:37 +0000 (14:56 +0200)]
maintainers: add an entry for the stable branches
This git tree will be used to backport some fixes from the
master branch to maintain some "stable releases".
The minor version number z will be incremented for these releases:
YY.MM.z
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: John McNamara <john.mcnamara@intel.com> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
VSI structure needs to be removed from TAILQ list when releasing.
But for the child VSI it will be removed again after the structure
is freed. It will cause core dump when the DPDK i40e using as PF
host driver.
This patch fixes it to only remove child VSI from TAILQ before
send adminq command to remove it from hardware.
Maxime Coquelin [Mon, 25 Jul 2016 14:09:58 +0000 (16:09 +0200)]
vhost: fix off-by-one error on descriptor number check
nr_desc is not an index but the number of descriptors,
so can be equal to the virtqueue size.
Fixes: a436f53ebfeb ("vhost: avoid dead loop chain") Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
When timer_cb resets another running timer on the same lcore,
the list of expired timers is chained to the pending-list.
This commit prevents a running timer from being reset
by not its own timer_cb.
When timer_set_running_state() fails in rte_timer_manage(),
the failed timer is put back on pending-list.
In this case, another core tries to reset or stop the timer.
It does not need to be on pending-list.
This commit fixes incorrect pending-list manipulation
when getting list of expired timers in rte_timer_manage().
When timer_get_prev_entries() sets pending_head on prev,
the pending-list is broken.
The next of pending_head always becomes NULL.
In this depth level, it is not need to manipulate the list.
Fixes: 9b15ba895b9f ("timer: use a skip list") Signed-off-by: Hiroyuki Mikita <h.mikita89@gmail.com> Acked-by: Robert Sanford <rsanford@akamai.com>
Jerin Jacob [Sun, 24 Jul 2016 17:07:40 +0000 (22:37 +0530)]
ring: fix single consumer dequeue performance
Use of rte_smb_wmb() instead of rte_smb_rmb() in sc dequeue function
creates the additional overhead of waiting for all the STOREs
to be completed to local buffer from ring buffer memory.
The sc dequeue function demands only LOAD-STORE barrier where LOADs
from ring buffer memory needs to be completed before tail pointer update.
Changing to rte_smb_rmb() to enable the required LOAD-STORE barrier.
Fixes: ecc7d10e448e ("ring: guarantee dequeue ordering before tail update") Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Thomas Monjalon [Mon, 25 Jul 2016 10:10:55 +0000 (12:10 +0200)]
mk: fix link with glibc < 2.17
There is a dependency on librt with old glibc.
The -lrt option was added everywhere it is needed but was also
added in some applications makefiles as the first link option.
The problem is this option is really useful only if added after
the objects or libraries using it (except if using --whole-archive).
And the -lrt options put after were removed to avoid duplicates.
It was resulting in errors linking test application:
eal_timer.c:(.text+0x128): undefined reference to `clock_gettime'
eal_timer.c:(.text+0x166): undefined reference to `clock_gettime'
eal_alarm.c:(.text+0xda): undefined reference to `clock_gettime'
eal_alarm.c:(.text+0x211): undefined reference to `clock_gettime'
It is fixed by removing superfluous -lrt in app makefiles.
Fixes: 281948b4753e ("mk: fix missing librt dependencies") Fixes: 2f6414f4baf1 ("mk: fix static link with glibc < 2.17") Reported-by: Piotr Azarewicz <piotrx.t.azarewicz@intel.com> Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Thomas Monjalon [Wed, 20 Jul 2016 13:38:54 +0000 (15:38 +0200)]
unify tools naming
The following tools may be installed system-wide.
It may be cleaner and more convenient to find them with the same
dpdk- prefix (especially for autocompletion).
Moreover, the script dpdk_nic_bind.py deserves a new name because it is
not restricted to NICs and can be used for e.g. crypto.
The tools pmdinfogen, pmdinfo.py and dpdk_pdump are new in 16.07.
The scripts dpdk_nic_bind.py and setup.sh may have been used with
previous releases by end users. That's why a symbolic link still
provide the old name in the installed tools directory.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
John McNamara [Sun, 17 Jul 2016 13:19:08 +0000 (14:19 +0100)]
doc: update sphinx installation instructions
Update the Sphinx installation instructions in the documentation
contributors guide to reflect the fact that in the 1.4+ versions
of Sphinx the ReadTheDocs theme must also be installed. Previously,
in version 1.3.x, it was installed by default.
Also change 'yum' to 'dnf' for package installations.
Signed-off-by: John McNamara <john.mcnamara@intel.com>
Flow Bifurcation is a mechanism which uses features of advanced
Ethernet devices to split traffic between queues. It provides
the capability to let the kernel driver and DPDK driver co-exist
and take advantage of both.
It is achieved by using SR-IOV and the NIC's advanced filtering. This
patch describes Flow Bifurcation and adds the user guide for ixgbe
and i40e NICs.
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
This patch adds an image of the Live Migration of a VM using vhost_user
on the host, test configuration.
Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com> Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
This patch describes the procedure to be be followed to perform
Live Migration of a VM with Virtio PMD running on a host which
is running the vhost_user sample application (vhost-switch).
It includes sample host and VM scripts used in the procedure.
Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com> Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
Yuanhan Liu [Tue, 19 Jul 2016 04:17:48 +0000 (12:17 +0800)]
doc: note a pitfall on vhost reconnect feature
The vhost feature negotiation only happens at virtio reset stage, say
when a virtio-net device is firstly initiated, or when DPDK virtio PMD
initiates. That means, if vhost APP restarts after the negotiation and
reconnects, the feature negotiation process will not be triggered again,
meaning the info is lost. To make reconnect work, QEMU simply saves
the negotiated features before the restart and restores it afterwards.
Therefore, the vhost supported features must be exactly the same before
and after the restart. For example, if TSO is disabled and then enabled,
nothing will work and undefined issues might happen.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
examples/l2fwd-ivshmem/host/host.c(157):
error #3656: variable "total_vm_packets_dropped"
may be used before its value is set
total_vm_packets_dropped += ctrl->vm_ports[portid].stats.dropped;
^
Function create_mp_ring_vdev() for failure cases exits without
freeing the created rte rings, because of this, pdump tool cannot be
rerun successfully. Added rte ring cleanup logic upon failures.
Pablo de Lara [Fri, 22 Jul 2016 14:02:02 +0000 (15:02 +0100)]
net/i40e: fix unsafe tailq element removal
i40e driver was removing elements when iterating tailq lists
with TAILQ_FOREACH macro, which is not safe.
It is especially visible since the memory is zeroed on free
(commit ea0bddbd14e6).
Instead, TAILQ_FOREACH_SAFE macro is used when removing/freeing
these elements.
Fixes: 4861cde46116 ("i40e: new poll mode driver") Fixes: 440499cf5376 ("net/i40e: support floating VEB") Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Pablo de Lara [Fri, 22 Jul 2016 14:02:01 +0000 (15:02 +0100)]
eal: add tailq safe iterator macro
Removing/freeing elements elements within a TAILQ_FOREACH loop is not safe.
FreeBSD defines TAILQ_FOREACH_SAFE macro, which permits
these operations safely.
This patch defines this macro for Linux systems, where it is not defined.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
In rte_mem_virt2phy: Value returned from a function and indicating the
number of bytes was ignored. This could cause a wrong pfn (page frame
number) mask read from pagemap file.
When read returns less than the number of sizeof(uint64_t) bytes,
function rte_mem_virt2phy returns error.
Coverity issue: 13212 Fixes: 40b966a211ab ("ivshmem: library changes for mmaping using ivshmem") Signed-off-by: Michal Jastrzebski <michalx.k.jastrzebski@intel.com>
Neil Horman [Wed, 20 Jul 2016 19:02:17 +0000 (15:02 -0400)]
scripts: validate ABI faster with parallel make jobs
John Mcnamara and I were discussing enhancing the validate_abi script to
build the dpdk tree faster with multiple jobs.
Theres no reason not to do it, so this implements that requirement.
It uses a DPDK_MAKE_JOBS variable that can be set by the user to limit
the job count. By default the job count is set to the number of online
cpus.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
examples/ip_pipeline: fix performance with default config
In TM, the read size should be lesser than the write size to improve
performance.
This enables the TM ports to push maximum packets to the output port.
This fix changes the burst_read value from 64 to 24 in default_tm_params.
IP Pipeline application with the configuration for Flow Classification
IPV6 did not instantiate.
Parse error in section "PIPELINE1": entry "dma_src_mask" too long
The dma_src_mask check in pipeline_passthrough_parse_args() is wrong.
This fix increases the length of dma_src_mask by 1 for NULL termination
and corrected the validation of dma_src_mask length.
This fix is also propagated to pipeline_fc_parse_args() for key_mask_str
validation.
Pablo de Lara [Mon, 18 Jul 2016 11:20:25 +0000 (12:20 +0100)]
examples/performance-thread: add missing braces
pthread_detach() function was returning 0 even when not calling
lthread_detach(), due to missing braces in conditional
(extra indentation was applied, giving a hint this is the correct fix).
Fixes: 433ba6228f9a ("examples/performance-thread: add pthread_shim app") Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Tested-by: John McNamara <john.mcnamara@intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
Jianfeng Tan [Thu, 21 Jul 2016 00:42:45 +0000 (00:42 +0000)]
examples/vhost: fix performance
We find significant perfermance drop introduced by below commit,
when vhost example is started with --mergeable 0 and inside vm,
kernel virtio-net driver is used to do ip based forwarding.
The commit, 859b480d5afd ("vhost: add guest offload setting"), adds
support for VIRTIO_NET_F_GUEST_TSO4 and VIRTIO_NET_F_GUEST_TSO6,
in vhost lib. But inside vhost example, the way to disable tso only
excludes the direction from virtio to vhost, but not the opposite
direction. When mergeable is disabled, it triggers big_packets path
of virtio-net driver to prepare to receive possible big packets with
size of 64K. Because mergeable is off, for each entry of avail ring,
virtio driver uses 19 desc chained together, with one desc pointing
to header, other 18 desc pointing to 4K-sized pages. But QEMU only
creates 256 desc entries for each vq, which results in that only 13
packets can be received. VM kernel can quickly handle those packets
and go to sleep (HLT).
As QEMU has no option to set the desc entries of a vq, so here,
we disable VIRTIO_NET_F_GUEST_TSO4 and VIRTIO_NET_F_GUEST_TSO6
with VIRTIO_NET_F_HOST_TSO4 and VIRTIO_NET_F_HOST_TSO6 when we
disable tso of vhost example, to avoid VM kernel virtio driver
go into big_packets path.
Pablo de Lara [Mon, 18 Jul 2016 13:21:04 +0000 (14:21 +0100)]
cryptodev: fix memory leak in parameter parsing
When parsing the parameters for virtual device initialization,
rte_kvargs structure was being freed only if there was an error,
not when parsing was successful.
Coverity issue: 124568 Fixes: f3e764fa2fb7 ("cryptodev: uninline parameter parsing") Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Reshma Pattan <reshma.pattan@intel.com>
Jianfeng Tan [Fri, 22 Jul 2016 02:24:47 +0000 (02:24 +0000)]
net/virtio-user: fix inconsistent name
The commit cb6696d22023 ("drivers: update registration macro usage")
changes the name from virtio-user to virtio_user, because hyphen
cannot be used in a C symbol name. However, this commit does not
update the strings in docs and source code, which could lead to
failure to start this device as per the docs.
This patch updates related strings in the docs and source code.
Fixes: cb6696d22023 ("drivers: update registration macro usage") Reported-by: Tiwei Bie <tiwei.bie@intel.com> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Rx loop inside _recv_raw_pkts_vec() ignores nb_pkts argument and always
tries to receive RTE_I40E_VPMD_RX_BURST (32) packets. This is a violation
of rte_eth_rx_burst() API and can lead to memory corruption (out-of-bounds
writes to struct rte_mbuf **rx_pkts) if nb_pkts is less than 32.
Fix this by actually using nb_pkts inside the loop.
John Daley [Tue, 19 Jul 2016 22:43:45 +0000 (15:43 -0700)]
net/enic: heed VLAN strip flag
The configure function enicpmd_dev_configure() was not paying attention
to the rxmode VLAN strip bit. Set the VLAN strip mode according to the bit.
Fixes: fefed3d1e62c ("enic: new driver") Signed-off-by: John Daley <johndale@cisco.com> Reviewed-by: David Harton <dharton@cisco.com> Tested-by: David Harton <dharton@cisco.com>
John Daley [Tue, 19 Jul 2016 22:42:43 +0000 (15:42 -0700)]
net/enic: fix possible Rx corruption
Initialize the mbuf data offset to RTE_PKTMBUF_HEADROOM as the
enic takes ownership of them. If allocated mbufs had some offset
other than RTE_PKTMBUF_HEADROOM, the application would read mbuf
data starting at the wrong place and misinterpret the packet.
Fixes: 856d7ba7ed22 ("net/enic: support scattered Rx") Reviewed-by: Nelson Escobar <neescoba@cisco.com> Signed-off-by: John Daley <johndale@cisco.com>
drivers/net/ena/base/ena_com.c(346):
error #3656: variable "dev_node" may be used before its value is set
ENA_MEM_ALLOC_COHERENT_NODE(ena_dev->dmadev,
^
drivers/net/ena/base/ena_com.c(399):
error #3656: variable "prev_node" may be used before its value is set
ENA_MEM_ALLOC_COHERENT_NODE(ena_dev->dmadev,
^
Fixes: 3d3edc265fc8 ("net/ena: make coherent memory allocation NUMA-aware") Reported-by: Eoin Breen <eoin.breen@intel.com> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com> Acked-by: Jan Medala <jan@semihalf.com>
Disable fastpath interrupts and remove unneeded delay in
bnx2x_interrupt_action(). This patch fixes and prevents performance
degradation (upto 50% drop) for BNX2X PMD.
The support of virtio-user changed the way the mbuf dma address is
retrieved, using a physical address in case of virtio-pci and a virtual
address in case of virtio-user.
This change introduced some possible memory corruption in packets,
replacing:
m->buf_physaddr + RTE_PKTMBUF_HEADROOM
by:
m->buf_physaddr + m->data_off (through a macro)
This patch fixes this issue, restoring the original behavior.
By the way, it also rework the macros, adding a "VIRTIO_" prefix and
API comments.
Fixes: f24f8f9fee8a ("net/virtio: allow virtual address to fill vring descriptors") Signed-off-by: Olivier Matz <olivier.matz@6wind.com> Acked-by: Jianfeng Tan <jianfeng.tan@intel.com> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Currently while calling of 'rte_vhost_driver_unregister()' connection
to QEMU will not be closed. This leads to inability to register driver
again and reconnect to same virtual machine.
This scenario is reproducible with OVS. While executing of the following
command vhost port will be re-created (will be executed
'rte_vhost_driver_register()' followed by 'rte_vhost_driver_unregister()')
network will be broken and QEMU possibly will crash:
ovs-vsctl set Interface vhost1 ofport_request=15
Fix this by closing all established connections on driver unregister and
removing of pending connections from reconnection list.
If something abnormal happened to QEMU, 'connect()' can block calling
thread (e.g. main thread of OVS) forever or for a really long time.
This can break whole application or block the reconnection thread.
Example with OVS:
ovs_rcu(urcu2)|WARN|blocked 512000 ms waiting for main to quiesce
(gdb) bt
#0 connect () from /lib64/libpthread.so.0
#1 vhost_user_create_client (vsocket=0xa816e0)
#2 rte_vhost_driver_register
#3 netdev_dpdk_vhost_user_construct
#4 netdev_open (name=0xa664b0 "vhost1")
[...]
#11 main
Fix that by setting non-blocking mode for client sockets for connection.
After doing a driver callout to fill in the driver specific
parts of struct rte_eth_stats, rte_eth_stats_get() overwrites
the rx_nombuf member regardless of whether the driver itself
has assigned a value. Any driver-assigned value should take
priority.
Fixes: af75078fece3 ("first public release") Signed-off-by: Remy Horton <remy.horton@intel.com>
Thomas Monjalon [Tue, 19 Jul 2016 17:15:34 +0000 (19:15 +0200)]
app/test: increase memory allocated for greedy autotests
The autotest lists, requirements and distribution needs a big rework
to reduce the amount of cores and memory required.
The root cause is not addressed yet.
This patch just increase some memory allocation for some greedy tests
which often fail because of memory fragmentation:
LPM6 and reentrancy tests in groups 3 and 6 respectively.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com> Tested-by: Olivier Matz <olivier.matz@6wind.com>
Jerin Jacob [Mon, 18 Jul 2016 05:55:54 +0000 (11:25 +0530)]
app/test: fix ring size
rte_ring_create expects the size of the ring to
be a power of 2. REFCNT_RING_SIZE value is not
power of 2 in-case if RTE_MAX_LCORE == 96.
Fix it by aligning the size to next power of 2 value.
Fixes: af75078f ("first public release") Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
ring: guarantee dequeue ordering before tail update
Consumer queue dequeuing must be guaranteed to be done fully before
the tail is updated. This is not guaranteed with a read barrier,
changed to a write barrier just before tail update which in practice
guarantees correct order of reads and writes.
Signed-off-by: Juhamatti Kuusisaari <juhamatti.kuusisaari@coriant.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Zoltan Kiss [Wed, 20 Jul 2016 17:16:38 +0000 (18:16 +0100)]
mempool: adjust name size in related data types
A recent patch brought up an issue about the size of the 'name' fields:
85cf0079 mem: avoid memzone/mempool/ring name truncation
These relations should be observed:
1. Each ring creates a memzone with a prefixed name:
RTE_RING_NAMESIZE <= RTE_MEMZONE_NAMESIZE - strlen(RTE_RING_MZ_PREFIX)
2. There are some mempool handlers which create a ring with a prefixed
name:
RTE_MEMPOOL_NAMESIZE <= RTE_RING_NAMESIZE - strlen(RTE_MEMPOOL_MZ_PREFIX)
3. A mempool can create up to RTE_MAX_MEMZONE pre and postfixed memzones:
sprintf(postfix, "_%d", RTE_MAX_MEMZONE)
RTE_MEMPOOL_NAMESIZE <= RTE_MEMZONE_NAMESIZE -
strlen(RTE_MEMPOOL_MZ_PREFIX) - strlen(postfix)
Setting all of them to 32 hides this restriction from the application.
This patch decreases the mempool and ring string size to accommodate for
these prefixes, but it doesn't apply the 3rd constraint. Applications
relying on these constants need to be recompiled, otherwise they'll run
into ENAMETOOLONG issues.
The size of the arrays are kept 32 for ABI compatibility, it can be
decreased next time the ABI changes.
Signed-off-by: Zoltan Kiss <zoltan.kiss@schaman.hu> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Zoltan Kiss [Wed, 20 Jul 2016 17:16:39 +0000 (18:16 +0100)]
mem: allow full length name
(strlen(name) == sizeof(mz->name) - 1) is a valid case, change the
condition to reflect that.
Move it earlier to avoid lookup with invalid name.
Change errno to ENAMETOOLONG.
Fixes: 85cf0079 ("mem: avoid memzone/mempool/ring name truncation") Signed-off-by: Zoltan Kiss <zoltan.kiss@schaman.hu> Acked-by: Olivier Matz <olivier.matz@6wind.com>
On weak memory order architecture like POWER, rte_smp_wmb/rte_smp_rmb
need to use CPU instructions, not compiler barrier. This patch fixes
this. Also, to improve performance on PPC64, use light weight sync
instruction instead of sync instruction.
Thomas Monjalon [Mon, 18 Jul 2016 09:33:52 +0000 (11:33 +0200)]
mk: fix static link with glibc < 2.17
There is an error when linking static EAL library with an application:
eal_alarm.c:(.text+0xd7): undefined reference to `clock_gettime'
eal_alarm.c:(.text+0x20f): undefined reference to `clock_gettime'
eal_timer.c:(.text+0x108): undefined reference to `clock_gettime'
eal_timer.c:(.text+0x146): undefined reference to `clock_gettime'
The function clock_gettime() is in librt for old glibc.
Jeff Shaw [Wed, 13 Jul 2016 11:51:02 +0000 (04:51 -0700)]
doc: fix mailing list address typo
The correct mailing list dev@dpdk.org, not dev@dpkg.org.
Signed-off-by: Jeff Shaw <jeffrey.b.shaw@intel.com> Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
git.droids-corp.org / dpdk.git / log