There is currently a scenario test, which will continuously obtain port
statistics, causing the CPU usage to soar, which does not meet the
demand. After positioning analysis, it is found that the VF and PF
command interaction is completed through the iavf_execute_vf_cmd
function.
After the message is sent, it needs to wait for the interrupt thread to
obtain the response from the PF. For the data, the rte_delay_ms
interface is used here to wait, but the CPU will not be released during
the waiting period of this interface, which will cause the statistics to
keep occupying the CPU. This is also the root cause of the soaring CPU.
The command interaction should belong to the control plane, and there
will not be too high requirements for performance. It is recommended to
wait for the interface iavf_msec_delay to complete without taking up the
CPU time.
Ivan Ilchenko [Wed, 15 Sep 2021 12:23:27 +0000 (15:23 +0300)]
net/virtio: report Tx descriptor limits in dev info
Report max/min/align Tx descriptors limits in device info get callback.
Before calling the callback, rte_eth_dev_info_get() provides
default values of nb_min as zero and nb_max as UINT16_MAX that are
not correct for the driver, so one can't rely on them.
Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Ivan Ilchenko [Wed, 15 Sep 2021 09:19:42 +0000 (12:19 +0300)]
net/virtio: fix Tx completed mbuf leak on device stop
Free Tx completed mbufs on device stop. Not completed Tx mbufs cannot be
freed since they are still in use.
Fixes: c1f86306a026 ("virtio: add new driver") Cc: stable@dpdk.org Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Ivan Ilchenko [Wed, 15 Sep 2021 09:19:41 +0000 (12:19 +0300)]
net/virtio: fix Tx cleanup functions to have same signature
There is a family of cleanup from completed transmits functions.
Fix packed virtqueues cleanup functions to have the same signature
as split virtqueues have. This lets all functions of the family to
match the same callback prototype.
Fixes: 892dc798fa9c ("net/virtio: implement Tx path for packed queues") Cc: stable@dpdk.org Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Eugenio Pérez [Fri, 27 Aug 2021 16:12:31 +0000 (18:12 +0200)]
vhost: clean IOTLB cache on vring stop
Old IOVA cache entries are left when there is a change on virtio driver
in VM. In case that all these old entries have iova addresses lesser
than new iova entries, vhost code will need to iterate all the cache to
find the new ones. In case of just a new iova entry needed for the new
translations, this condition will last forever.
This has been observed in virtio-net to testpmd's vfio-pci driver
transition, reducing the performance from more than 10Mpps to less than
0.07Mpps if the hugepage address was higher than the networking
buffers. Since all new buffers are contained in this new gigantic page,
vhost needs to scan IOTLB_CACHE_SIZE - 1 for each translation at worst.
Tomasz Duszynski [Sun, 15 Aug 2021 23:12:02 +0000 (01:12 +0200)]
raw/cnxk_bphy: support reading NPA/SSO PF function
Add support for reading NPA/SSO pf_func which will be used
by a PSM to access NPA/SSO. PSM is a hardware block capable
of dispatching jobs to different blocks within a baseband
module.
Signed-off-by: Tomasz Duszynski <tduszynski@marvell.com> Reviewed-by: Jerin Jacob <jerinj@marvell.com>
Tomasz Duszynski [Sun, 15 Aug 2021 23:12:01 +0000 (01:12 +0200)]
common/cnxk: support reading NPA/SSO PF function
Add support for reading NPA/SSO pf_func which will be used
by a PSM to access NPA/SSO. PSM is a hardware block capable
of dispatching jobs to different blocks within a baseband
module.
Signed-off-by: Tomasz Duszynski <tduszynski@marvell.com> Reviewed-by: Jerin Jacob <jerinj@marvell.com>
NPC MCAM entry distribution is based on maximum number of PFs and LFs
available. Fixing the max no of PFs and LFs available on cn98xx to
fix the MCAM alloc entry range.
Signed-off-by: Harman Kalra <hkalra@marvell.com> Acked-by: Jerin Jacob <jerinj@marvell.com>
In case of cn98xx, 2 NIX blocks and 4 LBK blocks are present. Moreover
AF VFs are alternatively attached to NIX0 and NIX1 to ensure load
balancing. To support loopback functionality between pairs NIX0/NIX1
are attached to LBK1/LBK2 for transmission/reception respectively.
But in this default configuration NIX blocks cannot receive the
packets they sent from the same LBK, which is an important requirement
as some ODP applications only uses one AF VF for loopback functionality.
To support this scenario, NIX0 can use LBK0 (NIX1 - LBK3) by setting a
loop flag while making LF alloc mailbox request.
Signed-off-by: Harman Kalra <hkalra@marvell.com> Acked-by: Jerin Jacob <jerinj@marvell.com>
common/cnxk: add handlers to get TM hierarchy internals
Platform specific TM tree hierarchy details are part of common cnxk
driver. This patch introduces missing HAL APIs to return state of
TM hierarchy required to support ethdev TM operations inside cnxk PMD.
Different TM handlers returns various platform specific errors,
this patch introduces new API to convert these internal error
types to RTE_TM* error types.
Also updated error message API with missed TM error types.
Each NIX interface had one or more SMQs connected to SQs to send
packets. When flush enabled on SMQ, hardware will push all packets
from SMQ to physical link. This API will enable flush on all SMQs
of an interface.
Sean Morrissey [Mon, 4 Oct 2021 09:22:18 +0000 (09:22 +0000)]
ring: promote new sync modes and peek to stable
These methods were introduced in 20.05.
There has been no changes in their public API since then.
They seem mature enough to remove the experimental tag.
Signed-off-by: Sean Morrissey <sean.morrissey@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Bruce Richardson [Fri, 17 Sep 2021 15:09:17 +0000 (16:09 +0100)]
test/mem: fix memory autotests on FreeBSD
The memory autotests were failing on FreeBSD, due to an incorrect errno
variable being checked for ENOTSUP. The test checked "errno" while the
DPDK API sets "rte_errno". Changing to check the right variable makes
the test behave properly.
Fixes: c3e35a0966b8 ("test/mem: check segment fd API") Reported-by: Brandon Lo <blo@iol.unh.edu> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Reviewed-by: David Marchand <david.marchand@redhat.com>
Bruce Richardson [Mon, 13 Sep 2021 14:08:48 +0000 (15:08 +0100)]
eal/freebsd: lock memory device to prevent conflicts
Only a single DPDK process on the system can be using the /dev/contigmem
mappings at a time, but this was never explicitly enforced, e.g. when
using --in-memory flag on two processes. To prevent possible conflict
issues, we lock the dev node when it's in use, preventing other DPDK
processes from starting up and causing problems for us.
Xiao Wang [Wed, 8 Sep 2021 10:59:15 +0000 (18:59 +0800)]
net: promote make rarp packet function to stable
rte_net_make_rarp_packet was introduced in version v18.02, there was no
change in this public API since then, and it's still being used by vhost
lib and virtio driver, so promote it as stable ABI.
Signed-off-by: Xiao Wang <xiao.w.wang@intel.com> Acked-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Chenbo Xia <chenbo.xia@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Ivan Malov [Tue, 31 Aug 2021 18:14:35 +0000 (21:14 +0300)]
log: promote some function to stable
This one might be quite mature to be attested as stable.
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru> Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Acked-by: Ray Kinsella <mdr@ashroe.eu>
Bruce Richardson [Mon, 13 Sep 2021 10:51:37 +0000 (11:51 +0100)]
usertools: silence prompts for telemetry input pipe
When the input to the script is coming from a device which is not a TTY
then we become less verbose and skip the prompts and helpful messages
about what is happening.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Ciara Power <ciara.power@intel.com>
Bruce Richardson [Mon, 13 Sep 2021 10:51:36 +0000 (11:51 +0100)]
usertools: fix handling EOF for telemetry input pipe
To allow the script to take queries from input pipes e.g. "echo
/ethdev/stats,0 | dpdk-telemetry.py", we need to handle the case of EOF
correctly without crashing with an exception. Do this by using a
try-except block around the input handling.
Fixes: 6a2967c112a3 ("usertools: add new telemetry script") Cc: stable@dpdk.org Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Ciara Power <ciara.power@intel.com>
Pablo de Lara [Tue, 28 Sep 2021 13:58:39 +0000 (13:58 +0000)]
efd: allow more CPU sockets in table creation
rte_efd_create() function was using uint8_t for a socket bitmask,
for one of its parameters.
This limits the maximum of NUMA sockets to be 8.
Changing to uint64_t increases it to 64, which should be
more future-proof.
Coverity issue: 366390 Fixes: 56b6ef874f8 ("efd: new Elastic Flow Distributor library") Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: John McNamara <john.mcnamara@intel.com> Acked-by: Yipeng Wang <yipeng1.wang@intel.com> Tested-by: David Christensen <drc@linux.vnet.ibm.com>
Kevin Traynor [Fri, 9 Jul 2021 15:19:36 +0000 (16:19 +0100)]
bitrate: fix registration to match API description
rte_stats_bitrate_reg() API states it returns 'Zero on success'.
However, the implementation directly returns the return of
rte_metrics_reg_names() which may be zero or positive on success,
with a positive value also indicating the index.
The user of rte_stats_bitrate_reg() should not care about the
index as it is stored in the opaque rte_stats_bitrates struct.
Change the implementation of rte_stats_bitrate_reg() to match
the API description by always returning zero on success.
There are a number telemetry threads which are created and
there is nothing that does pthread_join() to wait for them.
Mark these threads as detached, so that the pthread library
can cleanup state when the thread exits.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Ciara Power <ciara.power@intel.com>
Cian Ferriter [Mon, 23 Aug 2021 17:28:44 +0000 (18:28 +0100)]
ring: fix Doxygen comment of internal function
Change "enqueue" to "dequeue" because the __rte_ring_move_cons_head()
function is updating the consumer head for dequeue.
Fixes: 0dfc98c507b1 ("ring: separate out head index manipulation") Cc: stable@dpdk.org Signed-off-by: Cian Ferriter <cian.ferriter@intel.com> Acked-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
William Tu [Tue, 24 Aug 2021 16:21:03 +0000 (16:21 +0000)]
eal: remove sys/queue.h from public headers
Currently there are some public headers that include 'sys/queue.h', which
is not POSIX, but usually provided by the Linux/BSD system library.
(Not in POSIX.1, POSIX.1-2001, or POSIX.1-2008. Present on the BSDs.)
The file is missing on Windows. During the Windows build, DPDK uses a
bundled copy, so building a DPDK library works fine. But when OVS or other
applications use DPDK as a library, because some DPDK public headers
include 'sys/queue.h', on Windows, it triggers an error due to no such
file.
One solution is to install the 'lib/eal/windows/include/sys/queue.h' into
Windows environment, such as [1]. However, this means DPDK exports the
functionalities of 'sys/queue.h' into the environment, which might cause
symbols, macros, headers clashing with other applications.
The patch fixes it by removing the "#include <sys/queue.h>" from
DPDK public headers, so programs including DPDK headers don't depend
on the system to provide 'sys/queue.h'. When these public headers use
macros such as TAILQ_xxx, we replace it by the ones with RTE_ prefix.
For Windows, we copy the definitions from <sys/queue.h> to rte_os.h
in Windows EAL. Note that these RTE_ macros are compatible with
<sys/queue.h>, both at the level of API (to use with <sys/queue.h>
macros in C files) and ABI (to avoid breaking it).
Additionally, the TAILQ_FOREACH_SAFE is not part of <sys/queue.h>,
the patch replaces it with RTE_TAILQ_FOREACH_SAFE.
Suggested-by: Nick Connolly <nick.connolly@mayadata.io> Suggested-by: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com> Signed-off-by: William Tu <u9012063@gmail.com> Acked-by: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com> Acked-by: Narcisa Vasile <navasile@linux.microsoft.com>
Dmitry Kozlyuk [Sun, 29 Aug 2021 02:16:02 +0000 (05:16 +0300)]
eal/windows: fix export list
* Version and randomness API were not added to .def file by mistake,
which is why they were later excluded from the export list.
* Device API stubs were added to EAL but not exported.
Fixes: edd66d57d55c ("eal/windows: add random function") Fixes: 3d2fcb0e0aec ("eal/windows: add device event stubs") Fixes: 5b637a848195 ("eal: fix querying DPDK version at runtime") Signed-off-by: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com> Acked-by: William Tu <u9012063@gmail.com> Acked-by: Ranjit Menon <ranjit.menon@intel.com>
Dmitry Kozlyuk [Sun, 29 Aug 2021 02:16:01 +0000 (05:16 +0300)]
eal: remove Windows-specific list of common files
The majority of common EAL sources that are built for all platforms were
listed separately for Windows and for other OS. It seems that developers
adding modules to EAL perceived this as if Windows supported
only a limited subset of modules and only added new ones into another.
Factor the truly common modules into a shared list,
then extend it with modules supported by different platforms.
When the two lists were created, UUID API implementation was removed
from Windows build (apparently by mistake), then excluded from the
export list for no reason other than not being built. Restore it.
Fixes: df3ff6be2b33 ("eal: simplify meson build of common directory") Signed-off-by: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com> Acked-by: William Tu <u9012063@gmail.com> Acked-by: Ranjit Menon <ranjit.menon@intel.com>
William Tu [Thu, 5 Aug 2021 17:48:19 +0000 (17:48 +0000)]
eal/windows: export version function
When OVS inits, it calls rte_version to get the DPDK's version.
The patch fixes the error below by exposing rte_version symbol.
libopenvswitch.a(dpdk.c.obj) : error LNK2019: unresolved external symbol
rte_version referenced in function dpdk_init
Fixes: 5b637a848195 ("eal: fix querying DPDK version at runtime") Cc: stable@dpdk.org Signed-off-by: William Tu <u9012063@gmail.com> Acked-by: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com>
- Enable IAVF PMD build on Windows
- Replace x86intrin.h with rte_vect.h to avoid __m_prefetchw conflicting
types
- Fix for pointer and integer sign warnings using Clang compiler on
Windows
- Add extra cflags '-fno-asynchronous-unwind-tables'
to avoid MinGW build error:
Error: invalid register for .seh_savexmm
Based on the rte_eth_dev_socket_id() documentation,
set the default numa_node to -1. When the API is unsuccessful,
set numa_node to 0.
This change more correctly resembles the Linux code.
Fixes: bf7cf1f947bd ("bus/pci: fix unknown NUMA node value on Windows") Cc: stable@dpdk.org Reported-by: Vipin Varghese <vipin.varghese@intel.com> Signed-off-by: Pallavi Kadam <pallavi.kadam@intel.com> Acked-by: Ranjit Menon <ranjit.menon@intel.com>
Fixes: e1a00536c8ed ("kvargs: add a new library to parse key/value arguments") Fixes: 3ab385063cb9 ("kvargs: add get by key") Signed-off-by: Olivier Matz <olivier.matz@6wind.com> Reviewed-by: Xueming Li <xuemingl@nvidia.com> Reviewed-by: David Marchand <david.marchand@redhat.com> Acked-by: Ray Kinsella <mdr@ashroe.eu>
A quite common scenario with kvargs is to lookup for a <key>=<value> in
a kvlist. For instance, check if name=foo is present in
name=toto,name=foo,name=bar. This is currently done in drivers/bus with
rte_kvargs_process() + the rte_kvargs_strcmp() handler.
This approach is not straightforward, and can be replaced by this new
function.
rte_kvargs_strcmp() is then removed.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com> Reviewed-by: Xueming Li <xuemingl@nvidia.com> Reviewed-by: David Marchand <david.marchand@redhat.com> Acked-by: Ray Kinsella <mdr@ashroe.eu>
The function rte_kvargs_get() is used by eal and pci bus driver since
its introduction in commit 3ab385063cb9 ("kvargs: add get by key") and
commit d2a66ad79480 ("bus: add device arguments name parsing"), in
dpdk 21.05.
Let's promote it as stable.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com> Reviewed-by: Xueming Li <xuemingl@nvidia.com> Reviewed-by: David Marchand <david.marchand@redhat.com> Acked-by: Ray Kinsella <mdr@ashroe.eu>
net/af_packet: remove timestamp from packet status
We should eliminate the timestamp status from the packet
status. This should only matter if timestamping is enabled
on the socket, but we might hit a kernel bug, which is fixed
in newer releases.
For interfaces of type 'veth', the sent skb is forwarded
to the peer and back into the network stack which timestamps
it on the RX path if timestamping is enabled globally
(which happens if any socket enables timestamping).
When the skb is destructed, tpacket_destruct_skb() is called
and it calls __packet_set_timestamp() which doesn't check
the flags on the socket and returns the timestamp if it is
set in the skb (and for veth it is, as mentioned above).
See the following kernel commit for reference [1]:
net: packetmmap: fix only tx timestamp on request
The packetmmap tx ring should only return timestamps if requested
via setsockopt PACKET_TIMESTAMP, as documented. This allows
compatibility with non-timestamp aware user-space code which checks
tp_status == TP_STATUS_AVAILABLE; not expecting additional timestamp
flags to be set in tp_status.
Junxiao Shi [Thu, 9 Sep 2021 14:42:06 +0000 (14:42 +0000)]
net/memif: fix chained mbuf determination
Previously, TX functions call rte_pktmbuf_is_contiguous to determine
whether an mbuf is chained. However, rte_pktmbuf_is_contiguous is
designed to work on the first mbuf of a packet only. In case a packet
contains three or more segment mbufs in a chain, it may cause truncated
packets or rte_mbuf_sanity_check panics.
This patch updates TX functions to determine chained mbufs using
mbuf_head->nb_segs field, which works in all cases. Moreover, it
maintains that the second cacheline is only accessed when chained mbuf
is actually present.
Fixes: 09c7e63a71f9 ("net/memif: introduce memory interface PMD") Fixes: 43b815d88188 ("net/memif: support zero-copy slave") Cc: stable@dpdk.org Signed-off-by: Junxiao Shi <git@mail1.yoursunny.com> Reviewed-by: Jakub Grajciar <jgrajcia@cisco.com>
Thomas Monjalon [Mon, 30 Aug 2021 10:42:32 +0000 (12:42 +0200)]
ethdev: group constant definitions in Doxygen
A lot of flags are parts of a group but are documented alone.
The Doxygen syntax @{ and @} for grouping is used
to make flags appear together and have a common description.
Some Rx/Tx offload flags and RSS definitions are not grouped
because they need to be all properly documented first.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Ferruh Yigit <ferruh.yigit@intel.com> Acked-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Acked-by: Kevin Traynor <ktraynor@redhat.com>
Shared RSS resources were released before checking that the shared RSS
has no more references. If it had, the destruction was aborted, leaving
the shared RSS in an invalid state where it could no longer be used.
Move reference counter check before resource release.
Fixes: d2046c09aa64 ("net/mlx5: support shared action for RSS") Cc: stable@dpdk.org Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
When an indirect action is used in a flow rule with a pattern that
causes RSS expansion, each device flow generated by the expansion
incremented the reference counter of the action. When such a flow was
destroyed, its action reference counter had been decremented only once.
The action remained marked as being used and could not be destroyed.
COUNT, AGE, and CONNTRACK indirect actions have been affected
(for AGE the error was not immediately observable).
Increment action counter only once for the original flow rule.
Fixes: 81073e1f8ce1 ("net/mlx5: support shared age action") Fixes: 2d084f69aa26 ("net/mlx5: add translation of connection tracking action") Fixes: f3191849f2c2 ("net/mlx5: support flow count action handle") Cc: stable@dpdk.org Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
net/mlx5: report error on indirect CT action destroy
When an indirect CT action of mlx5 PMD could not be destroyed,
rte_action_handle_destroy() was returning (-1), but the error
structure was not filled. This lead to a segfault in testpmd
on an attempt to print it. Fill the details for each possible
cause of this error.
Michael Baum [Sun, 12 Sep 2021 10:36:28 +0000 (13:36 +0300)]
common/mlx5: fix resource cleaning in device removal
The common remove function call in a loop to remove function for each
driver which have been registered.
If all removes are succeeded, it return 0 without to free the device
which allocated in probe function. Otherwise, it free the device.
In fact we expect exactly the opposite behavior. If all removes are
failed, it returns error without freeing the device which allocated in
probe function. Otherwise, it free the device and return 0.
Replace it with the correct behavior.
Fixes: 8a41f4deccc3 ("common/mlx5: introduce layer for multiple class drivers") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Matan Azrad <matan@nvidia.com>
Michael Baum [Sun, 12 Sep 2021 10:36:27 +0000 (13:36 +0300)]
common/mlx5: fix device list operations concurrency
The mlx5 common driver has a global list of mlx5 devices which are
probed.
In probe function it creates one and insert it to the list. Similarly it
removes the device in remove function.
These operations are not safe as there can be such operations in
parallel, by different threads.
Add global lock for the list and use it to insert or remove.
Fixes: 8a41f4deccc3 ("common/mlx5: introduce layer for multiple class drivers") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Matan Azrad <matan@nvidia.com>
Michael Baum [Sun, 12 Sep 2021 10:36:26 +0000 (13:36 +0300)]
common/mlx5: fix class combination validation
The common probe function gets as a user argument the classes it should
create, and checks whether the combination is valid.
In case the device already exists, it checks the integration of the
above with the classes that the device has.
However, the function does not check the combination when the device
does not exist and it has to create it.
Michael Baum [Sun, 12 Sep 2021 10:36:25 +0000 (13:36 +0300)]
net/mlx5: fix duplicate pattern option default
In order to allow/disallow configuring rules with identical patterns,
the new device argument 'allow_duplicate_pattern' was introduced.
The default is to allow, and it is initialized to 1 in PCI probe
function.
However, on auxiliary bus probing (for Sub-Function) it is not
initialized at all, so it's actually initialized to 0.
Move the initialization to default config function which is called from
both.
Fixes: 919488fbfa71 ("net/mlx5: support Sub-Function") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Matan Azrad <matan@nvidia.com>
Michael Baum [Sun, 12 Sep 2021 10:36:24 +0000 (13:36 +0300)]
net/mlx5: fix PF leak on PCI probing failure
During PCI probe, the internal probe function is called per PF.
If one of them fails, it was missing a proper destroy for the previously
probed PFs.
This fixes the behavior by destroying all previously probed PFs.
Fixes: 08c2772fc747 ("net/mlx5: support list of representor PF") Cc: stable@dpdk.org Signed-off-by: Michael Baum <michaelba@nvidia.com> Acked-by: Matan Azrad <matan@nvidia.com>
1. Added support to specify l4 port masks in the template. Also enabled
source mac in the wild card key for ingress flows.
2. Added support to enable offload for ipv6 traffic within the vxlan
tunnel connection.
3. The flow counters is reduced from 7168 to 6912 for Whitney.
The stats operation is updated to reflect counts for packets
at egress from CFA instead of ingress to CFA
4. The miss path for the l2 context table is updated with correct
parif and default action handler to handle the miss path for
egress flows.
5. This support enables allocation of encapsulation, modification and
action records dynamically based on a given flow actions.
6. Reduce the l2context resource requests during open_session. Move the
SMAC from the L2Context to the EM/WM
7. Remap the parif in the bd action in order to eliminate incorrect
replication of broadcast packets. The layer 4 source port mask
was incorrectly updated in the outer layer 4 source port mask
instead of inner layer 4. Add the l3 proto to egress rules, switch
to using computed fields for l4 ports, add internal smac to f1/f2
flows, add l3 proto to ingress ipv6 flows
Signed-off-by: Kishore Padmanabha <kishore.padmanabha@broadcom.com> Signed-off-by: Mike Baucom <michael.baucom@broadcom.com> Signed-off-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com> Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com> Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Farah Smith [Mon, 20 Sep 2021 07:42:13 +0000 (13:12 +0530)]
net/bnxt: add SRAM manager shared session
Fix shared session support issues due to SRAM manager
additions. Shared session does not support slices within
RM blocks. Calculate resources required without slices
and determine base addresses using old methods for the
shared session.
Randy Schacher [Mon, 20 Sep 2021 07:42:12 +0000 (13:12 +0530)]
net/bnxt: allocate space dynamically for EM defrag
The dynamic pool allocation defrag function currently uses stack
allocation. To improve use of stack space, dynamically allocate
and deallocate memory for use to defragment the dynamic pool of
EM resources.
Signed-off-by: Randy Schacher <stuart.schacher@broadcom.com> Reviewed-by: Peter Spreadborough <peter.spreadborough@broadcom.com> Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
1. Add support for egress flows with port and count action for
Thor platform.
2. Added templates to support VXLAN encapsulation feature for Thor.
3. Added support for VXLAN decap and VLAN pop actions along with
the ingress flow.
4. Added templates to enable VXLAN decap support for f1 and f2 flows.
5. Added templates Thor VF Rep support
6. Added Thor ingress mod table actions for NAT, NAPT, and TTL.
7. Added mirror/sample table support
8. Added supported for IPv6 flows for Thor.
The encapsulation record processing is enhanced to handle data
dynamically. Different combinations of VXLAN encapsulation using
no VLAN or single or double VLAN can be supported and also supports
both IPv4 and IPv6 versions.
Add support for tunnel offload APIs. Specifically the following
are supported.
tunnel_decap_set, tunnel_match, tunnel_action_decap_release,
tunnel_item_release.
This provides support for VXLAN decap action where two flows
can indicate tunnel offload rule. The first flow indicates the
tunnel properties and second flow indicates the inner packet
structure. The templates are updated to support this
feature.
Template adds non-VFR based support for testpmd with:
matches to include
- DMAC, SIP, DIP, Proto, Sport, Dport
- SIP, DIP, Proto, Sport, Dport
actions:
- count, drop
Farah Smith [Mon, 20 Sep 2021 07:42:05 +0000 (13:12 +0530)]
net/bnxt: add SRAM manager model
The SRAM manager supports allocation and free of variable sized
records within SRAM memory. These record sizes are 8, 16, 32, or
64B. The SRAM manager algorithm will not fragment memory during
run time. Previous implementation only included fixed size 64B
records regardless of the size required.
Signed-off-by: Farah Smith <farah.smith@broadcom.com> Reviewed-by: Shahaji Bhosle <sbhosle@broadcom.com> Reviewed-by: Peter Spreadborough <peter.spreadborough@broadcom.com> Acked-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
git.droids-corp.org / dpdk.git / log