dpdk.git
4 years agopipeline: add SWX pipeline action
Cristian Dumitrescu [Thu, 1 Oct 2020 10:19:34 +0000 (11:19 +0100)]
pipeline: add SWX pipeline action

Add SWX actions that are dynamically-defined through instructions as
opposed to pre-defined. The actions are subroutines of the pipeline
program that triggered by table lookup. The input arguments are the
action data from the table entry (format defined by struct), the
headers and meta-data are in/out.

Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
4 years agopipeline: add SWX extern objects and funcs
Cristian Dumitrescu [Thu, 1 Oct 2020 10:19:33 +0000 (11:19 +0100)]
pipeline: add SWX extern objects and funcs

Add extern objects and functions to plug into the SWX pipeline any
functionality that cannot be efficiently implemented with existing
instructions, e.g. special checksum/ECC, crypto, meters, stats arrays,
heuristics, etc. In/out arguments are passed through mailbox with
format defined by struct.

Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
4 years agopipeline: add SWX headers and meta-data
Cristian Dumitrescu [Thu, 1 Oct 2020 10:19:32 +0000 (11:19 +0100)]
pipeline: add SWX headers and meta-data

Add support for dynamically-defined packet headers and meta-data to
the SWX pipeline. The header and meta-data format are defined by the
struct type they instantiate.

Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
4 years agopipeline: add SWX pipeline output port
Cristian Dumitrescu [Thu, 1 Oct 2020 10:19:31 +0000 (11:19 +0100)]
pipeline: add SWX pipeline output port

Add output ports to the newly introduced SWX pipeline type. Each port
instantiates a port type that defines the port operations, e.g. ethdev
port, PCAP port, etc. The TX interface is single packet, with packet
batching internally for performance.

Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
4 years agopipeline: add SWX pipeline input port
Cristian Dumitrescu [Thu, 1 Oct 2020 10:19:30 +0000 (11:19 +0100)]
pipeline: add SWX pipeline input port

Add input ports to the newly introduced SWX pipeline type. Each port
instantiates a port type that defines the port operations, e.g. ethdev
port, PCAP port, etc. The RX interface is single packet, with packet
batching internally for performance.

Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
4 years agopipeline: add new SWX pipeline type
Cristian Dumitrescu [Thu, 1 Oct 2020 10:19:29 +0000 (11:19 +0100)]
pipeline: add new SWX pipeline type

Add new improved Software Switch (SWX) pipeline type that supports
dynamically-defined packet headers, meta-data, actions and pipelines.
Actions and pipelines are defined through instructions.

Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
4 years agodoc: remove references to make from prog guide
Ciara Power [Mon, 21 Sep 2020 13:59:17 +0000 (14:59 +0100)]
doc: remove references to make from prog guide

Make is no longer supported for compiling DPDK, references are now
removed in the documentation.

Signed-off-by: Ciara Power <ciara.power@intel.com>
Reviewed-by: Kevin Laatz <kevin.laatz@intel.com>
4 years agodoc: remove references to make from howto guides
Ciara Power [Mon, 21 Sep 2020 13:59:16 +0000 (14:59 +0100)]
doc: remove references to make from howto guides

Make is no longer supported for compiling DPDK, references are now
removed in the documentation.

Signed-off-by: Ciara Power <ciara.power@intel.com>
Reviewed-by: Kevin Laatz <kevin.laatz@intel.com>
4 years agodoc: remove references to make from FreeBSD guide
Ciara Power [Mon, 21 Sep 2020 13:59:15 +0000 (14:59 +0100)]
doc: remove references to make from FreeBSD guide

Make is no longer supported for compiling DPDK, references are now
removed in the documentation.

Signed-off-by: Ciara Power <ciara.power@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agodoc: remove references to make from Linux guide
Ciara Power [Mon, 21 Sep 2020 13:59:14 +0000 (14:59 +0100)]
doc: remove references to make from Linux guide

Make is no longer supported for compiling DPDK, references are now
removed in the documentation.

Signed-off-by: Ciara Power <ciara.power@intel.com>
Reviewed-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agoapp: remove references to make-based config
Ciara Power [Mon, 21 Sep 2020 13:59:13 +0000 (14:59 +0100)]
app: remove references to make-based config

Make is no longer supported, RTE_SDK, RTE_TARGET and CONFIG options
are no longer in use.

Signed-off-by: Ciara Power <ciara.power@intel.com>
Reviewed-by: Kevin Laatz <kevin.laatz@intel.com>
Acked-by: Nicolas Chautru <nicolas.chautru@intel.com>
4 years agodevtools: remove legacy flags from includes check
Ciara Power [Mon, 21 Sep 2020 13:59:12 +0000 (14:59 +0100)]
devtools: remove legacy flags from includes check

Make is no longer supported, the test script for make builds is no
longer required. Uses of make in other tool scripts are replaced.

Signed-off-by: Ciara Power <ciara.power@intel.com>
4 years agodoc: fix references to removed guide
Thomas Monjalon [Wed, 30 Sep 2020 17:20:18 +0000 (19:20 +0200)]
doc: fix references to removed guide

The page "Development Kit Build System" was about make,
so it has been removed. A better help is in the Linux guide
(note: mlx4/mlx5 are supported on Linux only for now).

Fixes: 3cc6ecfdfe85 ("build: remove makefiles")

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: David Marchand <david.marchand@redhat.com>
Acked-by: Ciara Power <ciara.power@intel.com>
4 years agoci: add tests jobs in aarch64 vm
Juraj Linkeš [Fri, 28 Aug 2020 11:45:37 +0000 (13:45 +0200)]
ci: add tests jobs in aarch64 vm

Tests requiring hugepages do not work outside of VM environment because
of security limitations. Add aarch64 builds which run tests to run in
a VM to avoid these limitations. Leave non-hugepage environments since
the tests may produce different results in hugepage and non-hugepage
environments.

Signed-off-by: Juraj Linkeš <juraj.linkes@pantheon.tech>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Aaron Conole <aconole@redhat.com>
4 years agostack: fix uninitialized variable
Yunjian Wang [Fri, 25 Sep 2020 05:00:50 +0000 (13:00 +0800)]
stack: fix uninitialized variable

This patch fixes an issue that uninitialized 'success'
is used to be compared with '0'.

Coverity issue: 337676
Fixes: 3340202f5954 ("stack: add lock-free implementation")
Cc: stable@dpdk.org
Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
4 years agostack: relax pop CAS ordering
Steven Lariau [Fri, 25 Sep 2020 17:43:39 +0000 (18:43 +0100)]
stack: relax pop CAS ordering

Replace the store-release by relaxed for the CAS success at the end of
pop. Release isn't needed, because there is not write to data that need
to be synchronized.
The only preceding write is when the length is decreased, but the length
CAS loop already ensures the right synchronization.
The situation to avoid is when a thread sees the old length but the new
list, that doesn't have enough items for pop to success.
But the CAS success on length before the pop loop ensures any core reads
and updates the latest length, preventing this situation.

The store-release is also used to make sure that the items are read
before the head is updated, in order to prevent a core in pop to read an
incorrect value because another core rewrites it with push.
But this isn't needed, because items are read only when removed from the
used list. Right after this, they are pushed to the free list, and the
store-release in push makes sure the items are read before they are
visible in the free list.

Signed-off-by: Steven Lariau <steven.lariau@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Gage Eads <gage.eads@intel.com>
4 years agostack: reload head when pop fails
Steven Lariau [Fri, 25 Sep 2020 17:43:38 +0000 (18:43 +0100)]
stack: reload head when pop fails

List head must be loaded right before continue (when failed to
find the new head).
Without this, one thread might keep trying and failing to pop items
without ever loading the new correct head.

Fixes: 7e6e609939a8 ("stack: add C11 atomic implementation")
Cc: gage.eads@intel.com
Cc: stable@dpdk.org
Signed-off-by: Steven Lariau <steven.lariau@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Gage Eads <gage.eads@intel.com>
4 years agostack: remove redundant orderings on pop
Steven Lariau [Fri, 25 Sep 2020 17:43:37 +0000 (18:43 +0100)]
stack: remove redundant orderings on pop

The load-acquire of list->len on pop function is redundant.
Only the CAS success needs to be load-acquire.
It synchronizes with the store release in push, to ensure that the
updated head is visible when the new length is visible.
Without this, one thread in pop could see the increased length but the
old list, which doesn't have enough items yet for pop to succeed.

Signed-off-by: Steven Lariau <steven.lariau@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Gage Eads <gage.eads@intel.com>
4 years agostack: remove acquire fence on push
Steven Lariau [Fri, 25 Sep 2020 17:43:36 +0000 (18:43 +0100)]
stack: remove acquire fence on push

An acquire fence is used to make sure loads after the fence can observe
all store operations before a specific store-release.
But push doesn't read any data, except for the head which is part of a
CAS operation (the items on the list are not read).
So there is no need for the acquire barrier.

Signed-off-by: Steven Lariau <steven.lariau@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Gage Eads <gage.eads@intel.com>
4 years agostack: fix inconsistent weak/strong CAS
Steven Lariau [Fri, 25 Sep 2020 17:43:35 +0000 (18:43 +0100)]
stack: fix inconsistent weak/strong CAS

Fix cmpexchange usage of weak / strong.
The generated code is the same on x86 and ARM (there is no weak
cmpexchange), but the old usage was inconsistent.
For push and pop update size, weak is used because cmpexchange is inside
a loop.
For pop update root, strong is used even though cmpexchange is inside a
loop, because there may be a lot of operations to do in a loop iteration
(locate the new head).

Signed-off-by: Steven Lariau <steven.lariau@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Gage Eads <gage.eads@intel.com>
4 years agotest/stack: remove thread synchronisation
Steven Lariau [Wed, 12 Aug 2020 19:18:47 +0000 (20:18 +0100)]
test/stack: remove thread synchronisation

Remove the part that checks if there is enough room in the stack, it's
always true as long as size of stack >= MAX_BULK*rte_lcore_count().
This check used an atomic cmpset, and read / write to a shared size
variable. These operations result in some form of synchronization
that might get in the way of the actual stack testing.

Signed-off-by: Steven Lariau <steven.lariau@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Gage Eads <gage.eads@intel.com>
4 years agotest/stack: check errors for multi-threads
Steven Lariau [Wed, 12 Aug 2020 19:18:46 +0000 (20:18 +0100)]
test/stack: check errors for multi-threads

Use rte_eal_wait_lcore to wait and get the return value for all cores.
This is used to propagate any error to the main core.

Signed-off-by: Steven Lariau <steven.lariau@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Gage Eads <gage.eads@intel.com>
4 years agotest/stack: remove unneeded memory allocations
Steven Lariau [Wed, 12 Aug 2020 19:18:44 +0000 (20:18 +0100)]
test/stack: remove unneeded memory allocations

Replace the arguments array by one argument.
All objects in the args array have the same values, so there is no need
to use an array, only one struct is enough.
The args object is a lot smaller, and the allocation can be replaced
with a global variable.
As a consequence of using a single argument, there is no need to use a
loop to launch the test on every core one by one. Replace it with
rte_eal_mp_remote_launch.

The allocation of obj_table isn't needed either, because MAX_BULK is
small. The allocation can instead be replaced with a static array.

Signed-off-by: Steven Lariau <steven.lariau@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Gage Eads <gage.eads@intel.com>
4 years agoethdev: fix link speed helper documentation
David Marchand [Tue, 29 Sep 2020 12:12:22 +0000 (14:12 +0200)]
ethdev: fix link speed helper documentation

When generating the documentation, a new warning can be seen:

.../dpdk/lib/librte_ethdev/rte_ethdev.h:2441:
  warning: argument 'link_speed' of command @param is not found in the
  argument list of rte_eth_link_speed_to_str(uint32_t speed_link)
.../dpdk/lib/librte_ethdev/rte_ethdev.h:2455: warning: The following
  parameters of rte_eth_link_speed_to_str(uint32_t speed_link) are not
  documented: parameter 'speed_link'

Align the function prototype to its doxygen description.

Fixes: fbf931c9c392 ("ethdev: format link status text")

Signed-off-by: David Marchand <david.marchand@redhat.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agodoc: make doxygen comply with meson werror option
Bruce Richardson [Tue, 29 Sep 2020 16:55:02 +0000 (17:55 +0100)]
doc: make doxygen comply with meson werror option

When the --werror meson build option is set, we can set the WARN_AS_ERRORS
doxygen option in the doxygen config flag to get the same behaviour for API
doc building as for building the rest of DPDK. This can help catch
documentation errors sooner in the development process.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agodoc: hide sphinx standard output
Bruce Richardson [Tue, 29 Sep 2020 16:55:00 +0000 (17:55 +0100)]
doc: hide sphinx standard output

To see only errors and warnings from the doc builds, we can send the
standard output text to a logfile and have only the stderr messages
printed. This is similar to what is done for the API documentation.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agodoc: put doxygen log file in build directory
Bruce Richardson [Tue, 29 Sep 2020 16:54:59 +0000 (17:54 +0100)]
doc: put doxygen log file in build directory

The meson documentation states that projects should not rely upon the
custom_target build commands are run from any given directory.  Therefore,
rather than writing the standout output from doxygen to the current
directory - which could be anywhere in future, put it into the api
directory, so that it is in a known location.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agodoc: align doxygen output folder with sphinx guides
Bruce Richardson [Tue, 29 Sep 2020 16:54:58 +0000 (17:54 +0100)]
doc: align doxygen output folder with sphinx guides

The API docs were output to "<build>/doc/api/api" folder, which was
ugly-looking with the repeated "api", and inconsistent with the sphinx
guides which were written to "<build>/doc/guides/html". Changing the
doxygen output folder to "html" fixes both these issues.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agodoc: hide verbose doxygen standard output
Bruce Richardson [Tue, 29 Sep 2020 16:54:57 +0000 (17:54 +0100)]
doc: hide verbose doxygen standard output

The standard output of doxygen is very verbose, and since ninja mixes
stdout and stderr together it makes it difficult to see any warnings from
the doxygen run. Therefore, we can just log the standard output to file,
and only output the stderr to make warnings clear.

Suggested-by: Thomas Monjalon <thomas@monjalon.net>
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
4 years agokni: fix build with Linux 5.9
Ferruh Yigit [Mon, 17 Aug 2020 10:32:47 +0000 (11:32 +0100)]
kni: fix build with Linux 5.9

Starting from Linux 5.9 'get_user_pages_remote()' API doesn't get
'struct task_struct' parameter:
commit 64019a2e467a ("mm/gup: remove task_struct pointer for all gup code")

The change reflected to the KNI with version check.

Cc: stable@dpdk.org
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agovhost/crypto: fix possible TOCTOU attack
Fan Zhang [Mon, 28 Sep 2020 10:59:18 +0000 (11:59 +0100)]
vhost/crypto: fix possible TOCTOU attack

This patch fixes the possible time-of-check to time-of-use (TOCTOU)
attack problem by copying request data and descriptor index to local
variable prior to process.

Also the original sequential read of descriptors may lead to TOCTOU
attack. This patch fixes the problem by loading all descriptors of a
request to local buffer before processing.

CVE-2020-14375
Fixes: 3bb595ecd682 ("vhost/crypto: add request handler")
Cc: stable@dpdk.org
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Chenbo Xia <chenbo.xia@intel.com>
4 years agovhost/crypto: fix data length check
Fan Zhang [Mon, 28 Sep 2020 10:59:17 +0000 (11:59 +0100)]
vhost/crypto: fix data length check

This patch fixes the incorrect data length check to vhost crypto.
Instead of blindly accepting the descriptor length as data length, the
change compare the request provided data length and descriptor length
first. The security issue CVE-2020-14374 is not fixed alone by this
patch, part of the fix is done through:
"vhost/crypto: fix missed request check for copy mode".

CVE-2020-14374
Fixes: 3c79609fda7c ("vhost/crypto: handle virtually non-contiguous buffers")
Cc: stable@dpdk.org
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Chenbo Xia <chenbo.xia@intel.com>
4 years agovhost/crypto: fix write back source
Fan Zhang [Mon, 28 Sep 2020 10:59:16 +0000 (11:59 +0100)]
vhost/crypto: fix write back source

This patch fixes vhost crypto library for the incorrect source and
destination buffer calculation in the copy mode.

Fixes: cd1e8f03abf0 ("vhost/crypto: fix packet copy in chaining mode")
Cc: stable@dpdk.org
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Chenbo Xia <chenbo.xia@intel.com>
4 years agovhost/crypto: fix missed request check for copy mode
Fan Zhang [Mon, 28 Sep 2020 10:59:15 +0000 (11:59 +0100)]
vhost/crypto: fix missed request check for copy mode

This patch fixes the missed request check to vhost crypto
copy mode.

CVE-2020-14376
CVE-2020-14377
Fixes: 3bb595ecd682 ("vhost/crypto: add request handler")
Cc: stable@dpdk.org
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Chenbo Xia <chenbo.xia@intel.com>
4 years agovhost/crypto: fix descriptor deduction
Fan Zhang [Mon, 28 Sep 2020 10:59:14 +0000 (11:59 +0100)]
vhost/crypto: fix descriptor deduction

This patch fixes the incorrect descriptor deduction for vhost crypto.

CVE-2020-14378
Fixes: 16d2e718b8ce ("vhost/crypto: fix possible out of bound access")
Cc: stable@dpdk.org
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Chenbo Xia <chenbo.xia@intel.com>
4 years agovhost/crypto: fix pool allocation
Fan Zhang [Mon, 28 Sep 2020 10:59:13 +0000 (11:59 +0100)]
vhost/crypto: fix pool allocation

This patch fixes the missing iv space allocation in crypto
operation mempool.

Fixes: 709521f4c2cd ("examples/vhost_crypto: support multi-core")
Cc: stable@dpdk.org
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Chenbo Xia <chenbo.xia@intel.com>
4 years agovhost: fix external backends readiness
Maxime Coquelin [Wed, 23 Sep 2020 09:49:02 +0000 (11:49 +0200)]
vhost: fix external backends readiness

Commit d0fcc38f5fa4 ("vhost: improve device readiness notifications")
makes the assumption that every Virtio devices are considered
ready for preocessing as soon as first queue pair is configured
and enabled.

While this is true for Virtio-net, it isn't for Virtio-scsi
and Virtio-blk.

This patch fixes this by only making this assumption for
the builtin Virtio-net backend, and restores back to previous
behaviour for other backends.

Fixes: d0fcc38f5fa4 ("vhost: improve device readiness notifications")

Reported-by: Changpeng Liu <changpeng.liu@intel.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
4 years agobus/pci: fix mapping BAR containing MSI-X table
Hyong Youb Kim [Fri, 25 Sep 2020 02:14:35 +0000 (19:14 -0700)]
bus/pci: fix mapping BAR containing MSI-X table

When the BAR contains MSI-X table, pci_vfio_mmap_bar() tries to skip
the table and map the rest. "map around it" is the phrase used in the
source. The function splits the BAR into two regions: the region
before the table (first part or memreg[0]) and the region after the
table (second part or memreg[1]).

For hardware that has MSI-X vector table offset 0, the first part does
not exist (memreg[0].size == 0).

  Capabilities: [60] MSI-X: Enable- Count=48 Masked-
         Vector table: BAR=2 offset=00000000
         PBA: BAR=2 offset=00001000

The mapping part of the function maps the first part, if it
exists. Then, it maps the second part, if it exists and "if mapping the
first part succeeded".

The recent change that replaces MAP_FAILED with NULL breaks the "if
mapping the first part succeeded" condition (1) in the snippet below.

    void *map_addr = NULL;
    if (memreg[0].size) {
    /* actual map of first part */
    map_addr = pci_map_resource(...);
    }

    /* if there's a second part, try to map it */
    if (map_addr != NULL                              // -- (1)
    && memreg[1].offset && memreg[1].size) {
[...]
    }

    if (map_addr == NULL) {
            RTE_LOG(ERR, EAL, "Failed to map pci BAR%d\n",
                    bar_index);
            return -1;
    }

When the first part does not exist, (1) sees map_addr is still NULL,
and the function fails. This behavior is a regression and fails
probing hardware with vector table offset 0.

Previously, (1) was "map_addr != MAP_FAILED", which meant
pci_map_resource() was actually attempted and failed. So, expand (1)
to check if the first part exists as well, to match the semantics of
MAP_FAILED.

Bugzilla ID: 539
Fixes: e200535c1ca3 ("mem: drop mapping API workaround")

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
4 years agoethdev: use C11 atomics for link status
Phil Yang [Thu, 24 Sep 2020 05:39:28 +0000 (13:39 +0800)]
ethdev: use C11 atomics for link status

Since rte_atomicXX APIs are not allowed to be used, use C11 atomic
builtins for link status update.

Signed-off-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agopower: use C11 atomics for power state
Phil Yang [Thu, 24 Sep 2020 05:39:27 +0000 (13:39 +0800)]
power: use C11 atomics for power state

Since rte_atomicXX APIs are not allowed to be used, use C11 atomic
builtins for power in use state update.

Signed-off-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: David Hunt <david.hunt@intel.com>
4 years agobbdev: use C11 atomics for device processing counter
Phil Yang [Thu, 24 Sep 2020 05:39:26 +0000 (13:39 +0800)]
bbdev: use C11 atomics for device processing counter

Since rte_atomicXX APIs are not allowed to be used, use C11 atomic builtins
for device processing counter.

Signed-off-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Nicolas Chautru <nicolas.chautru@intel.com>
4 years agoeal: use C11 atomics for initialization check
Phil Yang [Thu, 24 Sep 2020 05:39:25 +0000 (13:39 +0800)]
eal: use C11 atomics for initialization check

Since rte_atomicXX APIs are not allowed to be used, use C11 builtins to
check if EAL is already initialized.

Signed-off-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
4 years agobuild: remove deprecated cpuflag macros
Radu Nicolau [Thu, 24 Sep 2020 08:18:29 +0000 (08:18 +0000)]
build: remove deprecated cpuflag macros

Replace use of RTE_MACHINE_CPUFLAG macros with regular compiler
macros, which are more complete than those provided by DPDK, and as such
it allows new instruction sets to be leveraged without having to do
extra work to set them up in DPDK.

Signed-off-by: Sean Morrissey <sean.morrissey@intel.com>
Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
Acked-by: David Marchand <david.marchand@redhat.com>
4 years agomaintainers: update NXP email
Sachin Saxena [Mon, 14 Sep 2020 14:06:03 +0000 (19:36 +0530)]
maintainers: update NXP email

Updated email of maintainer.

Signed-off-by: Sachin Saxena <sachin.saxena@oss.nxp.com>
4 years agomaintainers: update Mellanox emails
Ori Kam [Wed, 12 Aug 2020 16:08:35 +0000 (16:08 +0000)]
maintainers: update Mellanox emails

This patch updates Mellanox maintainers mails from
the Mellanox domain to Nvidia domain.

Cc: stable@dpdk.org
Signed-off-by: Ori Kam <orika@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
4 years agomaintainers: remove documentation maintainers
John McNamara [Wed, 9 Sep 2020 17:14:49 +0000 (18:14 +0100)]
maintainers: remove documentation maintainers

Removed the documentation maintainers.
The documentation is now, currently, unmaintained.

Signed-off-by: John McNamara <john.mcnamara@intel.com>
Signed-off-by: Marko Kovacevic <marko.kovacevic@intel.com>
4 years agoeal: remove deprecated coherent IO memory barriers
Phil Yang [Wed, 23 Sep 2020 09:16:37 +0000 (17:16 +0800)]
eal: remove deprecated coherent IO memory barriers

Since the 20.08 release deprecated rte_cio_*mb APIs because these APIs
provide the same functionality as rte_io_*mb APIs on all platforms, so
remove them and use rte_io_*mb instead.

Signed-off-by: Phil Yang <phil.yang@arm.com>
Signed-off-by: Joyce Kong <joyce.kong@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: David Marchand <david.marchand@redhat.com>
4 years agotest/ring: enhance debug info in failure cases
Feifei Wang [Sun, 20 Sep 2020 11:48:56 +0000 (06:48 -0500)]
test/ring: enhance debug info in failure cases

Add more parameters into the macro TEST_RING_VERIFY and expand the scope
of application for it. Then replace all ring APIs check with
TEST_RING_VERIFY to facilitate debugging.

Furthermore, correct a spelling mistakes of the macro
TEST_RING_FULL_EMTPY_ITER.

Suggested-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
4 years agotest/ring: factorize object checks
Feifei Wang [Sun, 20 Sep 2020 11:48:55 +0000 (06:48 -0500)]
test/ring: factorize object checks

Do code clean up by moving repeated code inside 'test_ring_mem_cmp'
function to validate data and print information of enqueue/dequeue
elements if validation fails.

Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
4 years agotest/ring: validate single element enqueue/dequeue
Feifei Wang [Sun, 20 Sep 2020 11:48:54 +0000 (06:48 -0500)]
test/ring: validate single element enqueue/dequeue

Validate the return value of single element enqueue/dequeue operation in
the test.

Suggested-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
4 years agotest/ring: check dequeued object for single element
Feifei Wang [Sun, 20 Sep 2020 11:48:53 +0000 (06:48 -0500)]
test/ring: check dequeued object for single element

Add check in test_ring_basic_ex and test_ring_with_exact_size for single
element enqueue and dequeue operations to validate the dequeued objects.

Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
4 years agotest/ring: fix dequeued object checks
Feifei Wang [Sun, 20 Sep 2020 11:48:52 +0000 (06:48 -0500)]
test/ring: fix dequeued object checks

When using memcmp function to check data, the third param should be the
size of all elements, rather than the number of the elements.

Fixes: a9fe152363e2 ("test/ring: add custom element size functional tests")
Cc: stable@dpdk.org
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
4 years agotest/ring: fix number of single element enqueue/dequeue
Feifei Wang [Sun, 20 Sep 2020 11:48:51 +0000 (06:48 -0500)]
test/ring: fix number of single element enqueue/dequeue

The ring capacity is (RING_SIZE - 1), thus only (RING_SIZE - 1) number of
elements can be enqueued into the ring.

Fixes: af75078fece3 ("first public release")
Cc: stable@dpdk.org
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
4 years agotest/ring: fix object reference for single element enqueue
Feifei Wang [Sun, 20 Sep 2020 11:48:50 +0000 (06:48 -0500)]
test/ring: fix object reference for single element enqueue

When enqueue one element to ring in the performance test, a pointer
should be passed to rte_ring_[sp|mp]enqueue APIs, not the pointer
to a table of void *pointers.

Fixes: a9fe152363e2 ("test/ring: add custom element size functional tests")
Cc: stable@dpdk.org
Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
4 years agonet/enic: support VXLAN decap action combined with VLAN pop
Hyong Youb Kim [Wed, 9 Sep 2020 14:00:06 +0000 (07:00 -0700)]
net/enic: support VXLAN decap action combined with VLAN pop

Flow Manager (flowman) provides DECAP_STRIP operation which
decapsulates VXLAN header and then removes VLAN header from the inner
packet. Use this operation to support vxlan_decap followed by
of_pop_vlan.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: generate VXLAN src port if it is zero in template
Hyong Youb Kim [Wed, 9 Sep 2020 14:00:05 +0000 (07:00 -0700)]
net/enic: generate VXLAN src port if it is zero in template

When VXLAN source port in the template is zero, the adapter is
expected to generate a value based on the inner packet flow, when it
performs encapsulation. Flow Manager in the VIC adapter currently
lacks such ability. So, generate a random port when creating a flow if
the port is zero, to avoid transmitting packets with source port 0.

Fixes: ea7768b5bba8 ("net/enic: add flow implementation based on Flow Manager API")
Cc: stable@dpdk.org
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: ignore VLAN inner type when it is zero
Hyong Youb Kim [Wed, 9 Sep 2020 14:00:04 +0000 (07:00 -0700)]
net/enic: ignore VLAN inner type when it is zero

When a VLAN pattern is present, the flow handler always copies its
inner_type to the match buffer regardless of its value (i.e. HW
matches inner_type against packet's inner ethertype). When inner_type
spec and mask are both 0, adding it to the match buffer is usually
harmless but breaks the following pattern used in some applications
like OVS-DPDK.

flow create 0 ingress ... pattern eth ... type is 0x0800 /
vlan tci spec 0x2 tci mask 0xefff / ipv4 / end actions count /
of_pop_vlan / ...

The VLAN pattern's inner_type is 0. And the outer eth pattern's type
actually specifies the inner ethertype. The outer ethertype (0x0800)
is first copied to the match buffer. Then, the driver copies
inner_type (0) to the match buffer, which overwrites the existing
0x0800 with 0 and breaks the app usage above.

Simply ignore inner_type when it is 0, which is the correct
behavior. As a byproduct, the driver can support the usage like the
above.

Fixes: ea7768b5bba8 ("net/enic: add flow implementation based on Flow Manager API")
Cc: stable@dpdk.org
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: support priorities for TCAM flows
Hyong Youb Kim [Wed, 9 Sep 2020 14:00:03 +0000 (07:00 -0700)]
net/enic: support priorities for TCAM flows

Group 0 corresponds to TCAM which supports priorities. Accept non-zero
priorities for group 0 flows.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: support egress port id action
Hyong Youb Kim [Wed, 9 Sep 2020 14:00:02 +0000 (07:00 -0700)]
net/enic: support egress port id action

Use Flow Manager (flowman) to support egress PORT_ID action. It can
steer egress packets from PFs and VFs to any uplink port as long as
they are all on the same VIC adapter. It can also steer packets
between ports on the same VIC adapter (i.e. loopback).

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: remove obsolete code
Hyong Youb Kim [Wed, 9 Sep 2020 14:00:01 +0000 (07:00 -0700)]
net/enic: remove obsolete code

The 'next' field in struct enic is unused. The comment in enic_cq_rq()
is out-of-date. Remove them.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: enable flow API for VF representor
Hyong Youb Kim [Wed, 9 Sep 2020 13:56:56 +0000 (06:56 -0700)]
net/enic: enable flow API for VF representor

Use Flow Manager (flowman) to support flow API for
representors. Representor's flow handlers simply invoke PF handlers
and pass the representor's flowman structure. The PF flowman handlers
are aware of representors and perform appropriate devcmds to create
flows on the NIC.

Also use flowman to create internal flows for implicit VF-representor
path. With that, representor Tx/Rx is now functional.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: extend flow handler to support VF representors
Hyong Youb Kim [Wed, 9 Sep 2020 13:56:55 +0000 (06:56 -0700)]
net/enic: extend flow handler to support VF representors

VF representor ports can create flows on VFs through the PF flowman
(Flow Manager) instance in the firmware. These flows match packets
egressing from VFs and apply flowman actions.

1. Make flow handler aware of VF representors
When a representor port invokes flow APIs, use the PF port's flowman
instance to perform flowman devcmd. If the port ID refers to a
representor, use VF handle instead of PF handle.

2. Serialize flow API calls
Multiple application thread may invoke flow APIs through PF and VF
representor ports simultaneously. This leads to races, as ports all
share the same PF flowman instance. Use a lock to serialize API
calls. Lock is used only when representors exist.

3. Add functions to create flows for implicit representor paths
There is an implicit path between VF and its representor. The
functions below create flow rules to implement that path.
- enic_fm_add_rep2vf_flow()
- enic_fm_add_vf2rep_flow()

The flows created for representor paths are marked as internal. These
are not visible to application, and the flush API does not destroy
them. They are automatically deleted when the representor port stops
(enic_fm_destroy).

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: add single queue Tx and Rx to VF representor
Hyong Youb Kim [Wed, 9 Sep 2020 13:56:54 +0000 (06:56 -0700)]
net/enic: add single queue Tx and Rx to VF representor

A VF representor allocates queues from PF's pool of queues and use
them for its Tx and Rx. It supports 1 Tx queue and 1 Rx queue.

Implicit packet forwarding between representor queues and VF does not
yet exist. It will be enabled in subsequent commits using flowman API.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: add minimal VF representor
Hyong Youb Kim [Wed, 9 Sep 2020 13:56:53 +0000 (06:56 -0700)]
net/enic: add minimal VF representor

Enable the minimal VF representor without Tx/Rx and flow API support.

1. Enable the standard devarg 'representor'
When the devarg is specified, create VF representor ports.

2. Initialize flowman early during PF probe
Representors require the flowman API from the firmware. Initialize it
before creating VF representors, so probe can detect the flowman
support and fail if not available.

3. Add enic_fm_allocate_switch_domain() to allocate switch domain ID
PFs and VFs on the same VIC adapter can forward packets to each other,
so the switch domain is the physical adapter.

4. Create a vnic_dev lock to serialize concurrent devcmd calls
PF and VF representor ports may invoke devcmd (e.g. dump stats)
simultaneously. As they all share a single PF devcmd instance in the
firmware, use a lock to serialize devcmd calls.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/enic: extend VNIC dev API for VF representors
Hyong Youb Kim [Wed, 9 Sep 2020 13:56:52 +0000 (06:56 -0700)]
net/enic: extend VNIC dev API for VF representors

VF representors need to proxy devcmd through the PF vnic_dev
instance. Extend vnic_dev to accommodate them as follows.

1. Add vnic_vf_rep_register()
A VF representor creates its own vnic_dev instance via this function
and saves VF ID. When performing devcmd, vnic_dev uses the saved VF ID
to proxy devcmd through the PF vnic_dev instance.

2. Add vnic_register_lock()
As PF and VF representors appear as independent ports to the
application, its threads may invoke APIs on them simultaneously,
leading to race conditions on the PF vnic_dev. For example, thread A
can query stats on PF port, while thread B queries stats on a VF
representor.

The PF port invokes this function to provide a lock to vnic_dev. This
lock is used to serialize devcmd calls from PF and VF representors.

3. Add utility functions to assist VF representor settings
vnic_dev_mtu() and vnic_dev_uif() retrieve vnic MTU and UIF number
(uplink index), respectively.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
4 years agonet/hns3: add Rx buffer size to Rx queue info
Chengchang Tang [Mon, 21 Sep 2020 13:22:38 +0000 (21:22 +0800)]
net/hns3: add Rx buffer size to Rx queue info

Report hns3 PMD configured Rx buffer size in Rx queue information query.

Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agoethdev: support getting Rx buffer size in Rx queue info
Chengchang Tang [Mon, 21 Sep 2020 13:22:37 +0000 (21:22 +0800)]
ethdev: support getting Rx buffer size in Rx queue info

Add a field named rx_buf_size in rte_eth_rxq_info to indicate the buffer
size used in receiving packets for HW.

In this way, upper-layer users can get this information by calling
rte_eth_rx_queue_info_get.

Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/netvsc: fix rndis packet addresses
Long Li [Fri, 18 Sep 2020 18:53:47 +0000 (11:53 -0700)]
net/netvsc: fix rndis packet addresses

The address should be calculated before type cast, not after.

Fixes: cc0251813277 ("net/netvsc: split send buffers from Tx descriptors")
Cc: stable@dpdk.org
Reported-by: Souvik Dey <sodey@rbbn.com>
Signed-off-by: Long Li <longli@microsoft.com>
4 years agonet/iavf: fix iterator for RSS LUT
Qi Zhang [Mon, 21 Sep 2020 08:30:58 +0000 (16:30 +0800)]
net/iavf: fix iterator for RSS LUT

Change RSS LUT iterator from uint8_t to uint16_t since the
RSS LUT size could exceed 255.

Fixes: 69dd4c3d0898 ("net/avf: enable queue and device")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Ting Xu <ting.xu@intel.com>
4 years agonet/memif: relax barrier for zero copy path
Phil Yang [Fri, 11 Sep 2020 05:38:19 +0000 (13:38 +0800)]
net/memif: relax barrier for zero copy path

Using 'rte_mb' to synchronize the shared ring head/tail between producer
and consumer will stall the pipeline and damage performance on the weak
memory model platforms, such like aarch64.

Relax the expensive barrier with c11 atomic with explicit memory
ordering can improve 3.6% performance on throughput.

Signed-off-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Jakub Grajciar <jgrajcia@cisco.com>
4 years agonet/hns3: fix crash when Tx multiple buffer packets
Chengchang Tang [Wed, 9 Sep 2020 09:23:39 +0000 (17:23 +0800)]
net/hns3: fix crash when Tx multiple buffer packets

Currently, there is a possibility that segment faults occur when sending
packets whose payloads are stored in multiple buffers based on hns3
network engine. The related core dump information as follows:

Program terminated with signal 11, Segmentation fault.
0  hns3_reassemble_tx_pkts
2512                            temp = temp->next;
Missing separate debuginfos, use:
(gdb) bt
0  hns3_reassemble_tx_pkts
1  0x0000000000969c60 in hns3_check_non_tso_pkt
2  0x000000000096adbc in hns3_xmit_pkts
3  0x000000000050d4d0 in rte_eth_tx_burst
4  0x000000000050fca4 in pkt_burst_transmit
5  0x00000000004ca6b8 in run_pkt_fwd_on_lcore
6  0x00000000004ca7fc in start_pkt_forward_on_core
7  0x00000000006975a4 in eal_thread_loop
8  0x0000ffffa6f7fc48 in start_thread
9  0x0000ffffa6ed1600 in thread_start

The root cause is that hns3 PMD driver invokes the rte_pktmbuf_free_seg
API function to release the same rte_mbuf multiple times. The rte_mbuf
pointer is not set to NULL in the internal function
hns3_rx_queue_release_mbufs which is invoked during queue setup, stop
and close. As a result the rte_mbuf in Rx queues will be repeatedly
released when the user application setup queues or stop/start the dev
for multiple times.
Probably for performance reasons, DPDK mempool lib does not check for
the repeated rte_mbuf releases. The Address of released rte_mbuf are
directly stored into the per lcore cache of the mempool. This makes the
rte_mbufs obtained from mempool by calling rte_mempool_get_bulk API
function repetitively.  ultimately, it causes to access to a NULL
pointer in PMD driver.

This patch fixes this problem by setting released mbuf pointer to NULL
in the internal function named hns3_rx_queue_release_mbuf. And the other
internal function named hns3_reassemble_tx_pkts is optimized to avoid a
similar problem.

Fixes: bba636698316 ("net/hns3: support Rx/Tx and related operations")
Cc: stable@dpdk.org
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
4 years agonet/hns3: add restriction on setting VF MTU
Wei Hu (Xavier) [Wed, 9 Sep 2020 09:23:38 +0000 (17:23 +0800)]
net/hns3: add restriction on setting VF MTU

When Rx of scattered packets is off, we have some possibility of using
vector Rx process function or simple Rx functions in hns3 PMD driver.
If the input MTU is increased and the maximum length of received packets
is greater than the length of a buffer for Rx packets, the hardware
network engine needs to use multiple BDs and buffers to store these
packets. This will cause problems when still using vector Rx process
function or simple Rx function to receiving packets. So, when Rx of
scattered packets is off and device is started, it is not permitted to
increase MTU so that the maximum length of Rx packets is greater than Rx
buffer length.

Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agonet/hns3: support NEON Rx
Wei Hu (Xavier) [Wed, 9 Sep 2020 09:23:37 +0000 (17:23 +0800)]
net/hns3: support NEON Rx

This patch adds NEON vector instructions to optimize Rx burst process.

Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Huisong Li <lihuisong@huawei.com>
4 years agonet/hns3: support NEON Tx
Wei Hu (Xavier) [Wed, 9 Sep 2020 09:23:36 +0000 (17:23 +0800)]
net/hns3: support NEON Tx

This patch adds NEON vector instructions to optimize Tx burst process.

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
4 years agonet/hns3: add simple Tx path
Wei Hu (Xavier) [Wed, 9 Sep 2020 09:23:35 +0000 (17:23 +0800)]
net/hns3: add simple Tx path

This patch adds simple Tx process function. When multiple segment
packets are not needed, Which means that DEV_TX_OFFLOAD_MBUF_FAST_FREE
offload is not set, we can simple Tx process.

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
4 years agonet/hns3: add simple Rx path
Wei Hu (Xavier) [Wed, 9 Sep 2020 09:23:34 +0000 (17:23 +0800)]
net/hns3: add simple Rx path

This patch adds simple Rx process function and support chose Rx function
by real Rx offloads capability.

Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Huisong Li <lihuisong@huawei.com>
4 years agonet/hns3: reduce address calculation in Rx
Wei Hu (Xavier) [Wed, 9 Sep 2020 09:23:33 +0000 (17:23 +0800)]
net/hns3: reduce address calculation in Rx

This patch adds the internal function named hns3_write_reg_opt to avoid
performance loss from address calculation during register access in the
'.rx_pkt_burst' ops implementation function named hns3_recv_pkts.

In addition, because hardware always access register in little-endian
mode based on hns3 network engine, so driver should also call
rte_cpu_to_le_32 to convert data in little-endian mode before writing
register and call rte_le_to_cpu_32 to convert data after reading from
register. Here the driver encapsulates the data conversion operation in
the register read/write operation function as below:
  hns3_write_reg
  hns3_write_reg_opt
  hns3_read_reg
Therefore, when calling these functions, conversion is not required
again.

Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agonet/hns3: report Rx free threshold
Wei Hu (Xavier) [Wed, 9 Sep 2020 09:23:32 +0000 (17:23 +0800)]
net/hns3: report Rx free threshold

This patch reports .rx_free_thresh value in the .dev_infos_get ops
implementation function named hns3_dev_infos_get and
hns3vf_dev_infos_get.
In addition, the name of the member variable of struct hns3_rx_queue is
modified and comments are added to improve code readability.

Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
4 years agoexamples: use new link status print format
Ivan Dyukov [Tue, 15 Sep 2020 19:07:02 +0000 (22:07 +0300)]
examples: use new link status print format

Add usage of rte_eth_link_to_str function to example
applications.

Signed-off-by: Ivan Dyukov <i.dyukov@samsung.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agoapp: use new link status print format
Ivan Dyukov [Tue, 15 Sep 2020 19:06:58 +0000 (22:06 +0300)]
app: use new link status print format

Add usage of rte_eth_link_to_str function to applications and docs.

Signed-off-by: Ivan Dyukov <i.dyukov@samsung.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agoethdev: format link status text
Ivan Dyukov [Tue, 15 Sep 2020 19:06:57 +0000 (22:06 +0300)]
ethdev: format link status text

There is new link_speed value introduced. It's INT_MAX value which
means that speed is unknown. To simplify processing of the value
in application, new function is added which convert link_speed to
string. Also dpdk examples have many duplicated code which format
entire link status structure to text.

This commit adds two functions:
  * rte_eth_link_speed_to_str - format link_speed to string
  * rte_eth_link_to_str - convert link status structure to string

Signed-off-by: Ivan Dyukov <i.dyukov@samsung.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
4 years agonet/af_xdp: fix umem size
Ciara Loftus [Thu, 10 Sep 2020 09:06:47 +0000 (09:06 +0000)]
net/af_xdp: fix umem size

The kernel expects the start address of the UMEM to be page size
aligned.
Since the mempool is not guaranteed to have such alignment, we have been
aligning the address to the start of the page the mempool is on. However
when passing the 'size' of the UMEM during it's creation we did not take
this into account.

This commit adds the amount by which the address was aligned to the size
of the UMEM.

Bugzilla ID: 532
Fixes: d8a210774e1d ("net/af_xdp: support unaligned umem chunks")
Cc: stable@dpdk.org
Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
4 years agonet/virtio: fix variable assignment in helper macro
Vipul Ashri [Fri, 18 Sep 2020 09:55:04 +0000 (15:25 +0530)]
net/virtio: fix variable assignment in helper macro

Inside Macro ASSIGN_UNLESS_EQUAL(var, val), assignment to var is always
failing as assignment done using var_ having local scope only.
This leads to TX packets not going out and found broken due to cleanup
malfunctioning. This patch fixes the wrong variable assignment.

Fixes: 57f90f894588 ("net/virtio: reuse packed ring functions")
Cc: stable@dpdk.org
Signed-off-by: Vipul Ashri <vipul.ashri@oracle.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agovdpa/mlx5: fix completion queue polling
Matan Azrad [Thu, 10 Sep 2020 07:20:34 +0000 (07:20 +0000)]
vdpa/mlx5: fix completion queue polling

The CQ polling is done in order to notify the guest about new traffic
bursts and to release FW resources for the next bursts management.

When HW is faster than SW, it may be that all the FW resources are busy
in SW due to late polling.
In this case, due to wrong WQE counter masking, the fullness
calculation of the completions number is 0 while the queue is full.

Change the WQE counter masking to 16-bit wideness instead of the CQ
size mask as defined by the CQE format.

Fixes: c5f714e50b0e ("vdpa/mlx5: optimize completion queue poll")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@nvidia.com>
Acked-by: Xueming Li <xuemingl@nvidia.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agovdpa/mlx5: fix completion queue assertion
Matan Azrad [Wed, 2 Sep 2020 08:34:59 +0000 (08:34 +0000)]
vdpa/mlx5: fix completion queue assertion

The CQ configuration enables the collapse feature in HW what cause HW to
write all the completions in the first CQE.
When this feature is enabled the HW doesn't switch the owner bit when it
starts a new cycle of the CQ, not like working without the collapse
feature.

The current SW CQ polling wrongly added an assertion to validate the
owner bit switch what causes a panic in debug mode.

Remove the aforementioned assertion.

Fixes: c5f714e50b0e ("vdpa/mlx5: optimize completion queue poll")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@nvidia.com>
Acked-by: Xueming Li <xuemingl@nvidia.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agovhost: fix IOTLB mempool single-consumer flag
Eugenio Pérez [Mon, 31 Aug 2020 07:59:22 +0000 (09:59 +0200)]
vhost: fix IOTLB mempool single-consumer flag

Control thread (which handles iotlb msg) and forwarding thread
both use iotlb to translate address. The former may modify the
same entry of mempool and may cause a loop in iotlb_pending_entries
list.

Bugzilla ID: 523
Fixes: d012d1f293f4 ("vhost: add IOTLB helper functions")
Cc: stable@dpdk.org
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agovdpa/mlx5: fix event channel setup
Xueming Li [Tue, 25 Aug 2020 09:17:28 +0000 (09:17 +0000)]
vdpa/mlx5: fix event channel setup

During vDPA device setup, if some error happens, event channel
release stucks at polling event channel.

Event channel fd is set to non-blocking in cqe setup, so if any
error happens before this function and after event channel created,
the pooling before releasing resources will stuck.

This patch moves event channel to non-blocking mode right after
creation.

Fixes: 8395927cdfaf ("vdpa/mlx5: prepare HW queues")
Cc: stable@dpdk.org
Signed-off-by: Xueming Li <xuemingl@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agovhost: add device reset status
Chenbo Xia [Mon, 10 Aug 2020 13:18:02 +0000 (13:18 +0000)]
vhost: add device reset status

vhost lib now does not have definition of reset status. This patch
adds the reset status definition and changes related log.

Signed-off-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
4 years agonet/virtio-user: enable feature checking
Adrian Moreno [Wed, 5 Aug 2020 14:45:17 +0000 (16:45 +0200)]
net/virtio-user: enable feature checking

virtio 1.0 introduced a mechanism for the driver to verify that the
feature bits it sets are accepted by the device. This mechanism consists
in setting the VIRTIO_STATUS_FEATURE_OK status bit and re-reading it,
which gives a chance for the device to clear it if the features
were not accepted.

This is currently being done only in modern virtio-pci devices but since
the appropriate vhost-user messages have been added, it can also be done
in virtio-user (vhost-user only).

This patch activates this mechanism on virtio-user.

Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
4 years agonet/virtio-user: support vhost status getting
Adrian Moreno [Wed, 5 Aug 2020 14:45:16 +0000 (16:45 +0200)]
net/virtio-user: support vhost status getting

This patch adds support for VHOST_USER_GET_STATUS request.

Only vhost-user backed is supported for now

Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
4 years agonet/virtio-user: support vhost status setting
Maxime Coquelin [Wed, 5 Aug 2020 14:45:15 +0000 (16:45 +0200)]
net/virtio-user: support vhost status setting

This patch adds support for VHOST_USER_SET_STATUS
request. It is used to make the backend aware of
Virtio devices status update.

It is useful for the backend to know when the Virtio
driver is done with the Virtio device configuration.

Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
4 years agonet/virtio: add device reset status bit
Adrian Moreno [Wed, 5 Aug 2020 14:45:14 +0000 (16:45 +0200)]
net/virtio: add device reset status bit

For the sake of completeness, add the definition of the missing status
bit in accordance with the virtio spec

Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
4 years agonet/cxgbe: support RSS redirection table update
Rahul Lakkireddy [Fri, 11 Sep 2020 23:52:10 +0000 (05:22 +0530)]
net/cxgbe: support RSS redirection table update

Implement eth_dev_ops to manipulate RSS redirection table.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: improve Rx congestion control
Rahul Lakkireddy [Fri, 11 Sep 2020 23:52:09 +0000 (05:22 +0530)]
net/cxgbe: improve Rx congestion control

Chelsio T6 NIC can support up to 8 priority channels to manage
congestion. So, increase to 8 congestion channels for T6. Also,
add Rxq state to avoid unnecessarily ringing doorbell and polling
the hardware for more traffic when the Rxq is stopped.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: rework queue allocation between ports
Rahul Lakkireddy [Fri, 11 Sep 2020 23:52:08 +0000 (05:22 +0530)]
net/cxgbe: rework queue allocation between ports

Firmware returns the max queues that can be allocated on the entire
PF. The driver evenly distributes them across all the ports belonging
to the PF. However, some ports may need more queues than others and
this equal distribution scheme prevents accessing these other ports
unused queues. So, remove the equal distribution scheme and allow the
ports to allocate as many queues as they need.

Also remove the hardcoded 64 max limit on queue allocation. Instead,
use the max limit given by firmware.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: release port resources during port close
Rahul Lakkireddy [Tue, 1 Sep 2020 17:16:26 +0000 (22:46 +0530)]
net/cxgbe: release port resources during port close

Enable RTE_ETH_DEV_CLOSE_REMOVE during PCI probe for all ports
enumerated under the PF. Free up the underlying port Virtual
Identifier (VI) and associated resources during port close.
Once all the ports under the PF are closed, free up the PF-wide
shared resources. Invoke port close function of all ports under
the PF, in PCI remove too.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/cxgbe: fix queue DMA ring leaks during port close
Rahul Lakkireddy [Tue, 1 Sep 2020 17:16:25 +0000 (22:46 +0530)]
net/cxgbe: fix queue DMA ring leaks during port close

Free up the DMA memzones properly for all the port's queues during
port close. So, rework DMA ring allocation/free logic to use
rte_eth_dma_zone_reserve()/rte_eth_dma_zone_free() helper functions
for allocating/freeing the memzones.

The firmware event queue doesn't have an associated freelist queue.
So, remove check that tries to give memzone name for a non-existent
freelist queue.

Also, add a missing free for the control queue mempools.

Fixes: 0462d115441d ("cxgbe: add device related operations")
Cc: stable@dpdk.org
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
4 years agonet/sfc/base: fix tunnel configuration
Igor Romanov [Tue, 8 Sep 2020 09:20:22 +0000 (10:20 +0100)]
net/sfc/base: fix tunnel configuration

Tunnel configuration may fail because of insufficient access rights
on a virtual function. Ignore the failure if a tunnel configuration
with empty UDP ports is requested.

Fixes: 17551f6dffcc ("net/sfc/base: add API to control UDP tunnel ports")
Cc: stable@dpdk.org
Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
4 years agonet/octeontx2: support RSS hash level
Kiran Kumar K [Thu, 17 Sep 2020 02:07:35 +0000 (07:37 +0530)]
net/octeontx2: support RSS hash level

Add support to choose rss hash level from ethdev rss config.

Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
4 years agoapp/testpmd: support RSS level configuration
Kiran Kumar K [Thu, 17 Sep 2020 02:07:34 +0000 (07:37 +0530)]
app/testpmd: support RSS level configuration

Adding support to set RSS level from ethdev config.
level-default will requests the default behavior.
level-outer will requests RSS to be performed on the outermost packet
encapsulation level.
level-inner will request RSS to be performed on the specified inner
packet encapsulation level, from outermost to innermost.

Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>