Moti Haimovsky [Thu, 2 Aug 2018 08:41:07 +0000 (11:41 +0300)]
net/mlx5: fix RSS flow configuration crash
This commit fixes a segmentation fault observed when configuring
mlx5 with RSS flow rule containing invalid queues indices such as
negative numbers, queue numbers bigger than the number Rx queues the
PMD or has no queues at all.
When running testpmd with an mlx5 device and then executing at testpmd
prompt in a raw: "port start all" followed by "port stop all"
a new file named /var/tmp/net_mlx5_<socket num> is created as a result
of creating a new unix domain socket (used for communication between
the primary and secondary processes).
When the new unix socket file is created the old unix socket file should
have been removed. This commit fixes it by closing the old unix socket
just before creating the new one in function mlx5_socket_init()
Fixes: f8b9a3bad467 ("net/mlx5: install a socket to exchange a file descriptor") Cc: stable@dpdk.org Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
The below commit has added a graph based expansion logic for RSS rule to
satisfy Verbs requirements. With this logic, for example, the rule:
flow create 0 ingress pattern eth / end actions rss queues 0 1 end types
ipv4-tcp ipv6-tcp end / end
will be expanded into the rules:
flow create 0 ingress pattern eth / ipv4 / tcp / end actions rss queues 0 1
end types ipv4-tcp ipv6-tcp end / end
flow create 0 ingress pattern eth / ipv6 / tcp / end actions rss queues 0 1
end types ipv4-tcp ipv6-tcp end / end
flow create 0 ingress pattern eth / end actions queue index 0 / end
The below commit defined two graphs:
1. graph for the tunnel case which starts from the ETH item
2. graph for the non-tunnel case which starts from the ETH item
The graphs are ignoring the VLAN case. Hence rules with VLAN item will
fail to traverse the graph and it will result in flow rule creation error.
Adding the VLAN item to the existing graphs will not work as the flow
engine will reject any VLAN item without a specific vid.
To solve this case two new graphs were added (for the tunnel and
non-tunnel case) which contain the VLAN item and are being used only
when the VLAN item exists in the flow pattern.
Two cases left un-covered for the inner RSS:
1. The case were VLAN exists in the pattern as part of the inner headers
2. The case were VLAN exists in the pattern both in the outer and the
inner headers
Solving those cases will require to add two more graphs.
Holding a VLAN for the overlay network is not common, the subnets are
usually defined by the tunnel protocol, for example the VXLAN vni.
Hence adding those two graphs seems like an overkill at this point.
Based on needs one can add those to provide the full support.
Matan Azrad [Wed, 1 Aug 2018 08:37:38 +0000 (08:37 +0000)]
net/mlx5: fix packet type offload for tunnels
There are dedicated QP attributes, tunnel offload flag and mask, which
must be configured in order to allow part of the HW tunnel offloads.
So, if a QP is pointed by a tunnel flow, the above QP attributes
should be configured.
The mask configuration is wrongly only performed if an internal RSS was
configured by the user, while there is no reason to condition the
tunnel offloads in RSS configurations.
Consequently, some of the tunnel offloads was not performed by the HW
when a tunnel flow was configured, for example, the packet tunnel
types was not reported to the user.
Replace the internal RSS condition with the tunnel flow condition.
Stripping binaries at build time is usually a bad thing since it makes
impossible to generate (split) debug symbols and this can lead to a more
difficult debugging.
Stripping binaries at build time is usually a bad thing since it makes
impossible to generate (split) debug symbols and this can lead to a more
difficult debugging.
net/mlx5: fix flow count action for shared counter
According to commit fb8fd96d4251 ("ethdev: add shared counter to flow
API") the counter id should be taken into account only when the shared
flag is set.
This patch verifies that the number of Rx queues configured for RSS
is supported by the device hardware.
RSS support in mlx4 requires contiguous chunk of QPs to be reserved,
there is a hardware limitation on the amount of contiguous QPs which
is reported by the hardware. Ignoring this value will cause Rx queues
creation to fail.
Gavin Hu [Tue, 31 Jul 2018 07:51:37 +0000 (15:51 +0800)]
net/dpaa2: remove loop for unused pool entries
Currently only one buffer pool is configured and in use,
looping for up to maxmum 8 times is unnecessary and might
be buggy as assigned uninititalized values.
The fix is to loop for the configured times with initialize
with valid values.
Fixes: 16bbc98a3e ("bus/fslmc: update MC to 10.3.x") Cc: stable@dpdk.org Signed-off-by: Gavin Hu <gavin.hu@arm.com> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
We were using LICENSE.bnx2x_pmd to reference inclusion of SPDX licensing
tag from all the source file. Remove the LICENSE.bnx2x_pmd file and
directly include SPDX tags in source files.
Originally the PMD had "QLogic Corporation" copyright. When we submitted
commit e3de5dad2a5d ("net/bnx2x: change copyright info to Cavium"),
the "Qlogic Corporation" copyright was accidentally replaced
with "Cavium Inc". So now we see multiple Cavium copyright messages.
We're changing it to "Broadcom Corporation" copyright.
Fixes: e3de5dad2a5d ("net/bnx2x: change copyright info to Cavium") Cc: stable@dpdk.org Signed-off-by: Rasesh Mody <rasesh.mody@cavium.com>
Rasesh Mody [Fri, 3 Aug 2018 04:42:45 +0000 (21:42 -0700)]
net/bnx2x: fix poll link status
The PMD has been modified to invoke the polling function in the link
management code which detects the peer speed/mode, configure the link
and update the status accordingly. This patch is the fix for the link
down issue seen when we do dev_stop() and dev_start() from an
application.
Rasesh Mody [Wed, 1 Aug 2018 18:19:20 +0000 (11:19 -0700)]
net/bnx2x: fix FW command timeout during stop
This patch fixes firmware command timeout error seen during device stop
while stopping queues. It patially reverts an earlier preventive change
commit 91b7e432bcef ("net/bnx2x: disable fast path interrupts") to now
enable fast path interrupts.
The original issue of performance degradation is not observed anymore,
with or without the fix.
Treat any attention which is not for timeout event as invalid
and return status accordingly. The HW error handler logs and clears
the HW attention. Without this fix we can see flood of GRC attentions.
Commit 023d7a0449f11 ("net/qede: fix legacy interrupt mode")
added a handler for legacy interrupt mode but forgot to
unregister same handler in rte_intr_callback_unregister()
Firmware can restrict access to certain resources in the hardware
depending on various factors. In such cases when the PMD tries to
configure these resources the firmware will return
HWRM_ERR_CODE_RESOURCE_ACCESS_DENIED error. Parse this and return
the standard EACCES error to the applications.
In bnxt_hwrm_flash_nvram, before attempting to allocate a buffer
we are grabbing the rte_spinlock. And if the allocation fails we
are returning before releasing the spinlock. We avoid the situation
by calling HWRM_PREP which grabs the lock after the buffer is
allocated successfully.
Fixes: 19e6af01bb36 ("net/bnxt: support get/set EEPROM") Cc: stable@dpdk.org Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com> Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
In some cases we may not be freeing up memory allocated for certain
NVM commands because the code might have bailed out before reaching
rte_free(). This patch moves some code around to ensure the allocated
memory is freed before exiting the function.
Fixes: 19e6af01bb36 ("net/bnxt: support get/set EEPROM") Cc: stable@dpdk.org Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com> Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com> Reviewed-by: Ray Jui <ray.jui@broadcom.com> Reviewed-by: Michael Wildt <michael.wildt@broadcom.com> Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
bnxt_clear_hwrm_vnic_filters() was removing the created filter from HW,
but not clearing the internal data structures by removing it from the
struct bnxt_vnic_info->filter list.
Control queues can not be allocated for VFs. So, add check to
only allocate control queues for PFs. This fixes adapter crash
when an attempt is made to allocate control queues for VFs.
Fixes: 3a3aaabc ("net/cxgbe: add control queue to communicate filter requests") Signed-off-by: Shagun Agrawal <shaguna@chelsio.com> Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
The NIC persists the vxlan port number across vNIC init/de-init
(e.g. restart testpmd). So, explicitly reset the setting to the
default value (4789) as part of the initialization.
Fixes: 8a4efd17410c ("net/enic: add handlers to add/delete vxlan port number") Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com> Reviewed-by: John Daley <johndale@cisco.com>
This reverts the patch that enabled mbuf fast free.
There are two main reasons.
First, enic_fast_free_wq_bufs is broken. When
DEV_TX_OFFLOAD_MBUF_FAST_FREE is enabled, the driver calls this
function to free transmitted mbufs. This function currently does not
reset next and nb_segs. This is simply wrong as the fast-free flag
does not imply anything about next and nb_segs.
We could fix enic_fast_free_wq_bufs by making it to call
rte_pktmbuf_prefree_seg to reset the required fields. But, it negates
most of cycle saving.
Second, there are customer applications that blindly enable all Tx
offloads supported by the device. Some of these applications do not
satisfy the requirements of mbuf fast free (i.e. a single pool per
queue and refcnt = 1), and end up crashing or behaving badly.
Fixes: bcaa54c1a148 ("net/enic: support mbuf fast free offload") Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com> Reviewed-by: John Daley <johndale@cisco.com>
On dev_stop the Rx/Tx_burst functions are being set to NULL, this causes
a SEGFAULT in cases where control path calls stop and a paket is still
being processed by a worker.
Instead clear the fastpath functions in dev_close.
Fixes: da6c687471a3 ("net/octeontx: add start and stop support") Cc: stable@dpdk.org Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com> Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Currently, nic_uio driver does not support interrupts, so any
attempts to install an interrupt handler will fail with a
not supported error, which will cause an error message that is
confusing to the user.
Silence this error by moving it to debug log level, and reword
the message to avoid containing the word "Error", to avoid
triggering DTS test failures [1].
The patch is incomplete because kernel 4.16+, while being capable
of mapping MSI-X BARs, will also report if such a capability is
available. Without checking this capability, gratuitous errors
are displayed on kernels <4.16 while VFIO is attempting to mmap
MSI-X BAR and fails, which can be confusing to the user.
Fixes: d4774a568ba0 ("vfio: fix workaround of BAR mapping") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
The forward declaraion of rte_pci_device in rte_ethdev.h
is not needed anymore.
Fixes: cd8c7c7ce241 ("ethdev: replace bus specific struct with generic dev") Cc: stable@dpdk.org Signed-off-by: Rami Rosen <rami.rosen@intel.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Ben Shelton [Tue, 10 Jul 2018 15:33:01 +0000 (10:33 -0500)]
ethdev: clarify usage of TM node parent update
The node parent update API function may be used to update the
priority/weight of an existing node. Update the documentation to
indicate that this use case is supported.
Signed-off-by: Ben Shelton <benjamin.h.shelton@intel.com> Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Gavin Hu [Wed, 1 Aug 2018 05:22:57 +0000 (13:22 +0800)]
devtools: fix checkpatch with dash
When running checkpatch.sh, it generates the following error
on some linux distributions(like Debian) with Dash as the
default shell interpreter.
trap: SIGINT: bad trap
The fix is to replace SIGINT with INT signal, it works for
both bash and dash.
Fixes: 4bec48184e ("devtools: add checks for ABI symbol addition") Cc: stable@dpdk.org Signed-off-by: Gavin Hu <gavin.hu@arm.com> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Acked-by: John McNamara <john.mcnamara@intel.com> Tested-by: Vipin Varghese <vipin.varghese@intel.com>
After adding RSS hash offload check, default rss_hf will fail on devices
that do not support all bits. This will lead to dcb config failure. The
patch fixes this issue by reading current valid rss_conf from the device.
Fixes: 8863a1fbfc66 ("ethdev: add supported hash function check") Fixes: 1a572499beb6 ("app/testpmd: setup DCB forwarding based on traffic class") Cc: stable@dpdk.org Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Bernard Iremonger <bernard.iremonger@intel.com>
If the "port config all crc-strip on" command is issued,
The DEV_RX_OFFLOAD_KEEP_CRC offload flag should be cleared.
If the "port config all crc-strip off command is issued,
The DEV_RX_OFFLOAD_KEEP_CRC offload flag should be set.
Fixes: 70815c9ecadd ("ethdev: add new offload flag to keep CRC") Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reshma Pattan [Wed, 1 Aug 2018 12:19:18 +0000 (13:19 +0100)]
test: remove CRC offload for bonding slave
Link bonding unit tests uses net_null vdev as slaves,
and trying to configure DEV_RX_OFFLOAD_CRC_STRIP as offload
flags to vdev is not correct as vdev doesn't support offloads.
The unit test is reporting this as failure, so fix this by
removing the offloads.
Fixes: c6698a3e8f ("ethdev: convert remaining apps to new offload API") Signed-off-by: Reshma Pattan <reshma.pattan@intel.com>
Makefiles are updated with new test case lists.
Test cases are classified as -
P1 - Main test cases,
P2 - Cryptodev/driver test cases,
P3 - Perf test cases which takes longer than 10s,
P4 - Logging/Dump test cases.
Makefile is updated with different targets
for the above classified groups.
Test cases for different targets are listed accordingly.
Now that everything else is in place, we can run unit tests in a
different fashion to what they were running as before. Previously,
we had all autotests as part of groups (largely obtained through
trial and error) to ensure parallel execution while still limiting
amounts of memory used by those tests.
This is no longer necessary, and as of previous commit, all tests
are now in the same group (still broken into two categories). They
still run one-by-one though. Fix this by initializing child
processes in multiprocessing Pool initialization, and putting all
tests on the queue, so that tests are executed by the first idle
worker. Tests are also affinitized to different NUMA nodes using
taskset in a round-robin fashion, to prevent over-exhausting
memory on any given NUMA node.
Non-parallel tests are executed in similar fashion, but on a
separate queue which will have only one pool worker, ensuring
non-parallel execution.
Support for FreeBSD is also added to ensure that on FreeBSD, all
tests are run sequentially even for the parallel section.
Previously, all autotests were grouped into (seemingly arbitrary)
groups. The goal was to run all tests in parallel (so that autotest
finishes faster), but we couldn't just do it willy-nilly because
DPDK couldn't allocate and free hugepages on-demand, so we had to
find autotest groupings that could work memory-wise and still be
fast enough to not hold up shorter tests. The inflexibility of
memory subsystem has now been fixed for 18.05, so grouping
autotests is no longer necessary.
Thus, this commit moves all autotests into two groups -
parallel(izable) autotests, and non-arallel(izable) autotests
(typically performance tests). Note that this particular commit
makes running autotests dog slow because while the tests are now
in a single group, the test function itself hasn't changed much,
so all autotests are now run one-by-one, starting and stopping
the DPDK test application.
Improve code for filtering test groups. Also, move reading binary
symbols into filtering stage, so that tests that are meant to be
skipped are never attempted to be executed in the first place.
Before running tests, print out any tests that were skipped because
they weren't compiled.
Autotest runner was still using python 2-style print syntax. Fix
it by importing print function from the future, and fix the calls
to be python-3 style.
There are no reports defined for any test, so this codepath was
never triggered, but it's still wrong because it's referencing
variables that aren't there. Fix it by passing target into the
test function, and reference correct log variable.
Previously, printing was done using tuple syntax, which caused
output to appear as a tuple as opposed to being one string. Fix
this by using addition operator instead.
This patch adds a new function that is called
per every checked patch,
and alerts for new instances of rte_panic/rte_exit.
The check excludes comments, and alerts in the case
of a positive balance between additions and removals.
Signed-off-by: Arnon Warshavsky <arnon@qwilt.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org> Tested-by: Kevin Traynor <ktraynor@redhat.com>
Pablo de Lara [Tue, 17 Jul 2018 13:31:51 +0000 (14:31 +0100)]
test/hash: fix potential memory leak
In the multiwriter test, if "found" array allocation failed,
the memory of "keys" array, which was successfully allocated
could not be freed, since by this time, tbl_multiwriter_test_params.keys
was not set to this array, which is the pointer freed when finishing
the test or when a failure happens.
To solve this, tbl_multiwriter_test_params.keys is set to the "keys"
address, just after allocating and filling the array.
Fixes: be856325cba3 ("hash: add scalable multi-writer insertion with Intel TSX") Cc: stable@dpdk.org Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Pablo de Lara [Tue, 17 Jul 2018 13:31:50 +0000 (14:31 +0100)]
test/hash: fix multiwriter with non consecutive cores
When non consecutive cores are passed into the test application,
the distribution of the keys that each thread needs to insert
is not correct, since it assumes that there are no cores skipped
between the master core and the worker core.
Fixes: be856325cba3 ("hash: add scalable multi-writer insertion with Intel TSX") Cc: stable@dpdk.org Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Config file has moved, but the tests weren't updated to point to
its new location. Update the code to find current prefix.
Also, this function is duplicated across multiple tests, so move it
into process.h and force compile failures for any attempts to use
it on platforms other than Linux.
Fixes: adf1d867361c ("eal: move runtime config file to new location") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Reshma Pattan <reshma.pattan@intel.com>
FreeBSD does not support running multiple primary processes
concurrently, because all DPDK instances will allocate memory
from the same place (memory provided by contigmem driver).
While it is technically possible to launch a DPDK process
using no-shconf switch, it will actually corrupt main process'
for the above reason.
Fix EAL flags autotest to not run primary processes unless
both no-shconf and no-huge are specified.
Fixes: 23eaa9059ec2 ("bus/pci: use given name as generic name") Cc: stable@dpdk.org Signed-off-by: Rami Rosen <rami.rosen@intel.com> Acked-by: John McNamara <john.mcnamara@intel.com>
If rte_metrics_init() had not been called and hence the internal
metric storage is not allocated, rte_metrics_get_values() and
rte_metrics_get_name() would silently fail by returning zero
(i.e. no metrics registered). This patch changes the result of
this scenario to an explicit fail by returning -EIO.
This patchset adds a check to rte_metrics_update_values()
that prevents the updating of metrics when presented with
an invalid metric key. Previously, doing the latter could
result in a crash.
If rte_stats_bitrate_reg() or rte_stats_bitrate_calc() are
passed NULL as the parameter for the stats structure, the
result is a crash. Fixed by adding a sanity check that makes
sure the passed-in pointer is not NULL.
Bruce Richardson [Mon, 16 Jul 2018 16:03:47 +0000 (17:03 +0100)]
examples/exception_path: fix out-of-bounds read
When printing out stats from the exception_path app, all possible
lcore_ids are iterated. However, the app only supports up to 64 cores.
To prevent possible errors, and to remove coverity warnings,
explicitly check for out-of-range lcore ids before printing.
Coverity issue: 268335 Fixes: af75078fece3 ("first public release") Cc: stable@dpdk.org Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Krzysztof Kanas [Wed, 25 Jul 2018 12:41:49 +0000 (14:41 +0200)]
app/testpmd: fix help for TM commit command
Fixes: 996cb153af06 ("app/testpmd: add commands for TM nodes and hierarchy commit") Signed-off-by: Krzysztof Kanas <krzysztof.kanas@caviumnetworks.com>
Krzysztof Kanas [Wed, 25 Jul 2018 12:41:48 +0000 (14:41 +0200)]
app/testpmd: fix crash on TM command error
print_err_msg uses message field that may be not initialized causing
segmentation fault.
Fixes: 12f76f5247e2 ("app/testpmd: add command to resume a TM node") Signed-off-by: Krzysztof Kanas <krzysztof.kanas@caviumnetworks.com> Acked-by: Bernard Iremonger <bernard.iremonger@intel.com>
Harry van Haaren [Thu, 26 Jul 2018 16:31:59 +0000 (17:31 +0100)]
service: improve checks of coremasks
This commit improves the error checking performed on the
core masks (or lists) of the service cores, in particular
with respect to the data-plane (RTE) cores of DPDK.
With this commit, invalid configurations are detected at
runtime, and warning messages are printed to inform the user.
For example specifying the coremask as 0xf, and the service
coremask as 0xff00 is invalid as not all service-cores are
contained within the coremask. A warning is now printed to
inform the user.
Reported-by: Vipin Varghese <vipin.varghese@intel.com> Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com> Acked-by: Vipin Varghese <vipin.varghese@intel.com>
Andy Green [Mon, 28 May 2018 09:03:43 +0000 (17:03 +0800)]
ring: fix sign conversion warning
On gcc 5.4.0 / native aarch64 from Ubuntu 16.04:
In function '__rte_ring_do_dequeue':
rte_ring.h: 385:35: warning:
conversion to 'int' from 'unsigned int' may change
the sign of the result [-Wsign-conversion]
n = __rte_ring_move_cons_head(r, is_sc, n, behavior,
^
Fixes: e8ed5056c8 ("ring: remove signed type flip-flopping") Cc: stable@dpdk.org Signed-off-by: Andy Green <andy@warmcat.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Andy Green [Mon, 28 May 2018 09:03:38 +0000 (17:03 +0800)]
ring: fix declaration after statement
On gcc 5.4.0 / native aarch64 from Ubuntu 16.04:
In function '__rte_ring_move_prod_head':
rte_ring_c11_mem.h:69:3: warning:
ISO C90 forbids mixed declarations and code
[-Wdeclaration-after-statement]
const uint32_t cons_tail = r->cons.tail;
^
In function '__rte_ring_move_cons_head':
rte_ring_c11_mem.h:136:3: warning:
ISO C90 forbids mixed declarations and code
[-Wdeclaration-after-statement]
const uint32_t prod_tail = r->prod.tail;
^
Fixes: 39368ebfc6 ("ring: introduce C11 memory model barrier option") Cc: stable@dpdk.org Signed-off-by: Andy Green <andy@warmcat.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
git.droids-corp.org / dpdk.git / log