net/qede/base: fix recovery from previous ungraceful exit
This patch modifies the recovery flow to allow ongoing PCIe
transactions to be completed. To achieve this, the load sequence is
changed such that the "final_cleanup" notification is sent while the
FID_enable is cleared.
This change ensures that the chip cleanup actions takes place from
previous driver instance if needed.
The corrupted code doesn't return error when probe function
fails due to error in device mac address getting.
By this way, the probe function may return success even if the
ETH dev is not allocated.
Hence, the probe caller, for example failsafe PMD, fails when it
tries to get ETH dev after the device was plugged out while mlx4
was probing it.
The fix adds error report to the probe caller when priv_get_mac fails
and in all other failure options which are missing it.
By this way, it prevents the unexpected behavior to miss ETH device
after the device was probed successfully.
Fixes: 7fae69eeff13 ("mlx4: new poll mode driver") Fixes: 001a520e419f ("net/mlx4: add port parameter") Fixes: 7b0661539229 ("mlx4: check if port is configured for ethernet") Fixes: fec3608673e6 ("mlx4: query netdevice to get initial MAC address") Cc: stable@dpdk.org Signed-off-by: Matan Azrad <matan@mellanox.com> Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
The corrupted code causes segmentation fault when user creates
flow with drop action before device starting.
For example, failsafe PMD recreates all the flows before calling
dev_start in plug-in sequence and mlx4 allocated its flow drop
queue in dev_start.
Hence, when failsafe created flow with drop action after plug-in
event, mlx4 tried to dereference flow drop queue which was
uninitialized.
The fix added check to the drop qp accessible and conditioned the
ibv_create_flow calling on device starting.
Fixes: 642fe56a1ba5 ("net/mlx4: use a single drop queue for all drop flows") Fixes: 46d5736a7049 ("net/mlx4: support basic flow items and actions") Cc: stable@dpdk.org Signed-off-by: Matan Azrad <matan@mellanox.com> Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
When advancing Tx ring index (txq->wqe_ci) in txq_scatter_v(), the title
descriptor of multi-packet send isn't taken into account if it doesn't
cross 64B boundary.
mlx5_tx_complete() polls completion queue multiple times until it
encounters an invalid entry. As Tx completions are suppressed by
MLX5_TX_COMP_THRESH, it is waste of cycles to expect multiple
completions in a poll. And freeing too many buffers in a call can
cause high jitter.
This patch improves throughput a little.
Herbert Guan [Thu, 13 Jul 2017 08:00:36 +0000 (16:00 +0800)]
test/bonding: fix parameters of a balance Tx
When test case "test_balance_l23_tx_burst_ipv4_toggle_ip_addr" is
calling balance_l23_tx_burst(), the ip_addr instead of mac_addr
should be toggled according to the test name.
Fixes: 92073ef961ee ("bond: unit tests") Cc: stable@dpdk.org Signed-off-by: Herbert Guan <herbert.guan@arm.com> Acked-by: Jianbo Liu <jianbo.liu@linaro.org> Acked-by: Declan Doherty <declan.doherty@intel.com>
Herbert Guan [Mon, 10 Jul 2017 11:13:46 +0000 (19:13 +0800)]
test/bonding: fix memory corruptions
There were double-free problems in some test cases, which will cause
a duplicated mbuf will be added into mempool. After double-free,
some new allocated mbuf will hold a same address and thus cause the
memory corruption.
Another minor issue is that in some test cases, allocated mbuf will
not be released after test case exits. Hopefully these leaked mbuf
will be released by the next test case in its setup phase when
stopping the virtual pmd ports, while this do is a memory leak of
the exited test case.
To fix above 2 issues, this patch will do:
1) Release virtual pmd ports' tx queue in the clean up function
remove_slaves_and_stop_bonded_device() of each test cases.
2) Do not release allocated mbufs for test bursts. These mbufs
will be released in remove_slaves_and_stop_bonded_device() when
test case exits.
Fixes: 92073ef961ee ("bond: unit tests") Signed-off-by: Herbert Guan <herbert.guan@arm.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Tomasz Kulasek [Wed, 26 Jul 2017 15:46:07 +0000 (17:46 +0200)]
test: fix virtual device name not set
Device name in device structure in virtual device used in
link_bonding_autotest is not set what causes segmentation fault when
rte_eth_dev_allocated is called.
Fixes: a1e7c17555e8 ("ethdev: use device name from device structure") Signed-off-by: Tomasz Kulasek <tomaszx.kulasek@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Only set the vlanonly bit if vlan_nonvlan is clear. Also, allow the
VLAN table to be cleared when vlanonly is set.
Clearing the VLAN table when vlanonly is set will stop all traffic
since it requires all frames to have a VLAN tag, and that tag to be
in the zero-length table. This is still a valid use case though,
and has been seen in the wild.
Fixes: 36735a932ca7 ("net/bnxt: support set VF QOS and MAC anti spoof") Signed-off-by: Stephen Hurd <stephen.hurd@broadcom.com> Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Tomasz Kulasek [Wed, 26 Jul 2017 15:53:30 +0000 (17:53 +0200)]
net/bonding: fix link status interrupt when down
RTE_ETH_EVENT_INTR_LSC callbacks are not called when all slaves goes down
in bond_ethdev_lsc_event_callback. It causes that link status change of
bonded device is not propagated up.
Fixes: deba8a2f8b0b ("net/bonding: fix link properties management") Signed-off-by: Tomasz Kulasek <tomaszx.kulasek@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Tomasz Kulasek [Wed, 26 Jul 2017 15:50:48 +0000 (17:50 +0200)]
net/bonding: fix crash when primary slave set
rte_eth_bond_primary_set segfaults for invalid port. This patch moves
devices check before use of internal data.
Fixes: 4c42498d916d ("net/bonding: allow slaves to also be bonded devices") Signed-off-by: Tomasz Kulasek <tomaszx.kulasek@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
net/bnxt: fix setting VLAN anti spoof with same value
If the current VLAN anti spoof setting is same as the new value,
the firmware can return an error. Call the HWRM command to update
the new setting if it is different from the current value.
Fixes: 36735a932ca7 ("net/bnxt: support set VF QOS and MAC anti spoof") Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Leonid Myravjev [Fri, 21 Jul 2017 11:25:38 +0000 (11:25 +0000)]
net/ixgbe: remove subdev ids from PCI device table
The pci_id_ixgbe_map table has only PCI_ANY_ID for
.subvend and .subdev in all entries. Using of IXGBE_SUBDEV_ID_82599_SFP
for 'dev' is not correct. It generates the wrong PMD_INFO_STRING.
The output of the program "dpdk-pmdinfo.py" is confusing.
Signed-off-by: Leonid Myravjev <myravjev@amicon.ru> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
We are wrongly using a Rx side HWRM command set_rx_mask to configure
VLAN anti-spoof. This being a Tx side feature, this patch
tries to fix it.
Since the HWRM command to do it is available only in
the newer firmware versions, the patch verifies the firmware
version before attempting to send the HWRM command to
the firmware.
Fixes: 36735a932ca7 ("net/bnxt: support set VF QOS and MAC anti spoof") Signed-off-by: Stephen Hurd <stephen.hurd@broadcom.com> Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Fix use of local variable to avoid segfault.
cnt was incorrectly tested and decremented in the loop that removes
a VLAN from the table.
Fixes: 36735a932ca7 ("net/bnxt: support set VF QOS and MAC anti spoof") Signed-off-by: Stephen Hurd <stephen.hurd@broadcom.com> Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
1) handle_async_event is a DEBUG level log message.
2) Log "Unable to get default VNIC for VF %d" at INFO level.
Fixes: 36735a932ca7 ("net/bnxt: support set VF QOS and MAC anti spoof") Signed-off-by: Stephen Hurd <stephen.hurd@broadcom.com> Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Jeff Guo [Fri, 21 Jul 2017 03:07:04 +0000 (11:07 +0800)]
net/i40e: fix link down and negotiation
Enable the functions set link down and set link up in i40e by check
phy_type, and fix the issue of auto negotiation failed in XXV710 when
bind kernel driver after unbind from DPDK driver by modify the speed
setting distinguish from set link up and down. With this fix, if unbind
DPDK to bind kernel driver, no need to set auto negotiation and ifconfig
up anymore, remove the part from doc.
Fixes: ca7e599d4506 ("net/i40e: fix link management") Fixes: 2f1e22817420 ("i40e: skip link control as firmware workaround") Fixes: 6e145fcc754b ("i40e: support autoneg or force link speed") Cc: stable@dpdk.org Signed-off-by: Jeff Guo <jia.guo@intel.com> Acked-by: Jingjing Wu <jingjing.wu@intel.com>
Jiayu Hu [Mon, 31 Jul 2017 07:56:44 +0000 (15:56 +0800)]
net/virtio: fix Rx interrupt setup
When use rte_eth_dev_configure() to enable rx queue interrupt for virtio
devices, virtio_init_device() isn't called to set up the interrupt
environment, which causes rx queue interrupt setup failed. This patch is
to fix this issue.
Fixes: 26b683b4f7d0 ("net/virtio: setup Rx queue interrupts") Cc: stable@dpdk.org Signed-off-by: Jiayu Hu <jiayu.hu@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Jianfeng Tan [Fri, 28 Jul 2017 23:01:14 +0000 (23:01 +0000)]
net/virtio: fix MAC address read
When virtio-net devices are bound to uio_pci_generic, we get
the wrong mac addr by virtio PMD. The wrong mac addr is a
addr that is 4-byte left shift of the correct addr.
It's a regression bug introduced by the cleanup patch below.
The condition of if we set use_msix should be if msix is
actually enabled. Only to check if there is a capability list
is not enough. For example, binding a transitional device
to uio_pci_device would trigger the wrong assignment of use_msix.
To correct that, we also check the flags of msix capability to
make sure it's enabled.
Pablo de Lara [Wed, 26 Jul 2017 04:20:05 +0000 (05:20 +0100)]
mbuf: fix doxygen comment of bulk alloc
When calling rte_pktmbuf_alloc_bulk, if there are
not enough objects in the mempool, it returns
a negative value, which should be reflected
in the Doxygen comments.
Fixes: 9ec201f5d6e7 ("mbuf: provide bulk allocation") Cc: stable@dpdk.org Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
The name of a device is copied in a provided buffer within
rte_eth_dev_detach(). The current sizeof is done on a pointer instead of
the intended array usually pointed to.
The name field of an rte_device is not assured however to point an
rte_devargs name field. The almost correct length to base this copy over
is thus RTE_DEV_NAME_MAX_LEN.
Almost correct, because unfortunately this function does not allow the
user to pass down a size parameter for the buffer it is meant to write.
This API should be fixed, it is broken by design.
Fixes: a1e7c17555e8 ("ethdev: use device name from device structure") Cc: stable@dpdk.org Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
If the device is a vdev, the parsing for PCI will fail with -EFAULT,
and will not try to check for a vdev.
Checking against error values returned by rte_eal_hotplug_add is
inelegant and prone to mistakes. Additionally, the failed PCI probe
prints a useless error that would throw off unsuspecting users:
ERROR: failed to parse device "pci:net_ring0"
This error is printed when attempting to probe a virtual device first
with the PCI bus (here, a net_ring0 device).
Use the relevant functions to infer the intended bus. The limitation to
PCI or vdev device is kept for strict API compatibility. Thus the PCI
probe attempt is avoided and the right function is directly called.
Fixes: 1c35f666df07 ("dev: fix attach proceeding with vdev on PCI success") Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
The corrupted code checks only RAW flow item type special case for
returning its size but doesn't deal with any other flow item type
and returns 0 for all the others.
This bug leaves the flow descriptor empty for non RAW types.
The fix takes the correct size to any regular types from appropriate
array.
The same issue, with a similar fix, is in flow action size method
which deals only with RSS special type.
This bug was already present in the original code taken from testpmd.
The corrupted code checks only RAW flow item type special case for
returning its size but doesn't deal with any other flow item type
and returns 0 for all the others.
This bug leaves the flow descriptor empty for non RAW types.
The fix takes the correct size to any regular types from appropriate
array.
The same issue, with a similar fix, is in flow action size method which
deals only with RSS special type.
Fixes: 938a184a1870 ("app/testpmd: implement basic support for flow API") Cc: stable@dpdk.org Signed-off-by: Matan Azrad <matan@mellanox.com> Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
dev: fix attach proceeding with vdev on PCI success
When rte_eal_hotplug_add() successfully probes a PCI device,
the return value is zero.
The check afterward only returns on error different from -EINVAL.
It should return also on success, as there is no need to
attempt probing the device with vdev.
Fixes: 0bba9e605048 ("eal: use new hotplug API in attach") Cc: stable@dpdk.org Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
eal: disable NUMA related warnings on non-NUMA systems
Disable multiple NUMA warnings on non-NUMA systems.
"EAL: eal_parse_sysfs_value(): cannot open sysfs value
/sys/bus/pci/devices/0000:00:00.0/numa_node
EAL: numa_node is invalid or not present. Set it 0 as default
EAL: cannot open /proc/self/numa_maps, consider that all memory is
in socket_id 0"
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
For larger packet size congestion is observed on Tx Queues.
This patch enables Tx Queue congestion state check support.
If congested, try to resend the packet few times.
Pablo de Lara [Fri, 28 Jul 2017 06:02:45 +0000 (07:02 +0100)]
doc: add cryptodev sample code
In order to illustrate in a simple way how to use
the cryptodev API to perform a basic crypto operation,
sample code has been added in the Programmer's Guide.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
Pablo de Lara [Thu, 27 Jul 2017 23:33:13 +0000 (00:33 +0100)]
app/crypto-perf: fix CSV output
When using CSV, the output of the throughput performance numbers
was not correct, representing decimal numbers incorrectly.
Fixes: f8be1786b1b8 ("app/crypto-perf: introduce performance test application") Cc: stable@dpdk.org Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Pablo de Lara [Fri, 28 Jul 2017 02:15:21 +0000 (03:15 +0100)]
app/crypto-perf: add pool size error message
If the crypto operation pool size is too small,
the pool might run out of operations, if all the crypto operations
are still being used and have not been freed.
To inform the user about this, the application should display
an error message, asking the user to increase the pool size
through the app parameters.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Pablo de Lara [Wed, 26 Jul 2017 00:59:59 +0000 (01:59 +0100)]
doc: complete crypto sample command line
Documentation of some virtual crypto PMDs have a sample command line
to show how to initialize the device on a specific application,
L2fwd-crypto.
This was meant to be used as a reference, but these lines themselves
do not work, as the sample application used required more parameters,
which are added in this commit to have a fully functional example.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
Jan Blunck [Wed, 12 Jul 2017 20:15:37 +0000 (16:15 -0400)]
cryptodev: deprecate vdev create function
The function rte_cryptodev_create_vdev is an alias
for rte_vdev_init() which is scheduled to move out of the
rte_eal library. Lets deprecate this function to be able to
remove it from the cryptodev library in 17.11.
Signed-off-by: Jan Blunck <jblunck@infradead.org> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com> Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
Pablo de Lara [Wed, 26 Jul 2017 04:27:12 +0000 (05:27 +0100)]
cryptodev: fix doxygen comment of bulk alloc
When calling rte_crypto_op_bulk_alloc, the function may
return either a 0, if not enough objects are available
in the mempool or the number of operations requested,
it there are enough available. However, the Doxygen comments
were not matching these two cases.
Fixes: c0f87eb5252b ("cryptodev: change burst API to be crypto op oriented") Cc: stable@dpdk.org Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Fiona Trahe <fiona.trahe@intel.com>
Pablo de Lara [Tue, 25 Jul 2017 06:16:50 +0000 (07:16 +0100)]
cryptodev: fix session init return value
When calling rte_cryptodev_sym_session_init(),
if there was an error, it returned -1, instead
of returning the specific error code, which can
be valuable for the application for error handling.
Fixes: b3bbd9e5f265 ("cryptodev: support device independent sessions") 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>
Commit 8a04cb612589 ("pci: set default numa node for broken systems")
added logic to default to NUMA node 0 when sysfs numa_node information
was wrong or not available.
Unfortunately there are many devices with wrong NUMA node information
that DPDK does not care about but still show warnings for them.
Instead, only check for invalid NUMA node information for devices
managed by the DPDK.
The _rte_eth_dev_callback_process function has been modified.
The return value has been changed form void to int and an
extra parameter "void *ret_param" has been added.
Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com>
APP_MAX_LCORES is hardcoded as 64.
This will cause build err when RTE_MAX_LCORE is less then 64.
"args.c:127:22: error: iteration 8 invokes undefined behavior
[-Werror=aggressive-loop-optimizations]
if (cfg->lcore_role[i] == ROLE_RTE)
~~~~~~~~~~~~~~~^~~
args.c:126:2: note: within this loop
for (i = 0; i < APP_MAX_LCORE; i++) {"
load_balancer app can also work for lower number of cores.
Limit the cores Worker and IO cores to 16 as defined in original
App. Otherwise use the actual number of lcores as MAX.
when creating a bond device in testpmd, a name for a device must meet
the correct convention described in the documentation:
The device name must start with the net_bonding prefix
followed by numbers or letters.
Change for ALB mempool allocation - mem_name was too long.
Fixes: 9bf4901d1a11 ("bus/vdev: remove probe with driver name option") Cc: stable@dpdk.org Signed-off-by: Michal Jastrzebski <michalx.k.jastrzebski@intel.com> Reviewed-by: Jingjing Wu <jingjing.wu@intel.com>
Hemant Agrawal [Thu, 22 Jun 2017 12:48:07 +0000 (18:18 +0530)]
mempool/dpaa2: fix error code for allocation failure
In case the alloc api is not able to allocate the required
number of buffer, it can return '0', which will not indicate
the failure to the calling function.
This patch fix the return value to indicate the failure.
igb_uio: issue FLR during open and release of device file
Set UIO info device file operations open and release. Call pci reset
function inside open and release to clear device state at start and end.
Copied this behaviour from vfio_pci kernel module code. With this patch,
it is not mandatory to issue FLR by PMD's during init and close.
Bus master enable and disable are added in open and release respectively
to take care of device DMA.
While matching user input against a token name or any other fixed string,
comparison stops at the end of user input if shorter (e.g. "foo" matches
token name "foobar").
Although the unintended consequence of this behavior allows users to
abbreviate command names and various parameters yet generate valid
commands, the parser was not designed to support this and does not prevent
ambiguous tokens.
For instance, entering "i" for a pattern item matches "ipv4", "ipv6" and
"icmp" then defaults to one of them in an unspecified manner.
Prevent this behavior by taking the length of fixed strings into account.
Support for dynamic tokens was added in order to implement the flow command
in testpmd, for which static tokens were not versatile enough due to the
large number of possible parameter combinations.
However, due to its reliance on a temporary array to store dynamic tokens,
this interface suffers from various limitations that need to be addressed
in order to implement more commands in the future:
- The maximum number of dynamic tokens is determined at compilation time
(CMDLINE_PARSE_DYNAMIC_TOKENS). The larger this value, the more stack
space is wasted (one pointer per potential token, i.e. 1kB of stack space
on 64-bit architectures with the default value).
- This temporary array is actually a cache in which entries already present
are not regenerated. This behavior is not documented, which makes dynamic
tokens practically unusable by applications as they do not know which
token is current.
- The cache does not really reduce the number of function calls needed to
retrieve tokens, it was mainly deemed useful to provide context about
other tokens to the generator callback.
- Like testpmd, most users will likely use repeated pointers to a fixed
token header structure (cmdline_token_hdr_t), with internal context-aware
callbacks that do not need to look at other entries; knowing the index of
the current token is enough.
Getting rid of the temporary array and properly documenting usage of the
token generator callback greatly simplifies this interface.
To avoid redundant calls to the token-generating function and provide it
with helpful context, a temporary fixed-size array allocated on the stack
of cmdline_parse() and cmdline_complete() keeps the address of preceding
tokens for the command instance being processed (cmdline_parse_inst_t).
Like the static tokens array in cmdline_parse_inst_t, it must be
NULL-terminated, however this is not properly enforced as it is initialized
once at the beginning of each function and a NULL terminator is never
appended once replaced with generated tokens.
When several commands rely on dynamic tokens, subsequent ones inherit an
already initialized array whose tokens are not regenerated, which causes
various issues such as mixed and repeated tokens from the first command.
Enforcing NULL termination of the dynamic tokens array and reinitializing
its first entry at each iteration solves these issues. Doing so is also
less expensive than a full memset() at each iteration.
Fixes: 4fffc05a2b2c ("cmdline: support dynamic tokens") Cc: stable@dpdk.org Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com> Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
git.droids-corp.org / dpdk.git / log