The pkt_type field in the profile TCAM table needs to be ignored and
should not be set to normal packet type. The pkt_type for the packets
that are segmented due to transmit segment offload feature in the driver
are not marked as normal pkt_type and this shall result in profile TCAM
table miss and flow not being offloaded hence resulting in the reduction
of the throughput.
Fixes: fe82f3e02701 ("net/bnxt: support exact match templates") Cc: stable@dpdk.org Signed-off-by: Kishore Padmanabha <kishore.padmanabha@broadcom.com> Reviewed-by: Mike Baucom <michael.baucom@broadcom.com> Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
net/bnxt: support representors on remote host domain
In the Stingray use case, representors are conventionally run
inside the SoC domain representing functions that are on the
X86 domain. In order to support this mechanism of building
representors for endpoints that are not in the same host domain,
additional dev args have been in the PMD like so:
rep-based-pf=<physical index> rep-is-pf=<VF=0 or PF=1>
where `rep-based-pf` specifies the physical index of the base PF
that the representor is derived off of.
Since representor(s) can be created for endpoint PFs as well,
rename struct bnxt_vf_representor to bnxt_representor and other such
dev_ops and function names.
devargs have also been extended to specify the exact CoS queue along
with flow control enablement to be used for the conduit between the
representor and the endpoint function.
This is how a sample devargs would look with all the extended devargs
In page_roundup() left shifting by more than 31 bits could have
undefined behavior as the return value is int and in page_getenum()
it is possible to return a value as high as 63.
Fix that to cap the return value to less than 32.
rx_queue_setup_op for representor was using a common function to
initialize the software data structures for the Rx ring. But that
routine has code to init other rings not needed for representors like
cp/agg ring etc.
Define and invoke a new function to setup structures just for the
representor Rx ring
Fixes: 6dc83230b43b ("net/bnxt: support port representor data path") Cc: stable@dpdk.org Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com> Reviewed-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com> Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Kernel v5.10 will introduce the ability to efficiently share a UMEM
between AF_XDP sockets bound to different queue ids on the same or
different devices. This patch integrates that functionality into the AF_XDP
PMD.
A PMD will attempt to share a UMEM with others if the shared_umem=1 vdev
arg is set. UMEMs can only be shared across PMDs with the same mempool, up
to a limited number of PMDs goverened by the size of the given mempool.
Sharing UMEMs is not supported for non-zero-copy (aligned) mode.
The benefit of sharing UMEM across PMDs is a saving in memory due to not
having to register the UMEM multiple times. Throughput was measured to
remain within 2% of the default mode (not sharing UMEM).
A version of libbpf >= v0.2.0 is required and the appropriate pkg-config
file for libbpf must be installed such that meson can determine the
version.
net/mlx5: manage header reformat actions with hashed list
To manage encap decap header format actions mlx5 PMD used the single
linked list and lookup and insertion operations took too long times if
there were millions of objects and this impacted the flow
insertion/deletion rate.
In order to optimize the performance the hashed list is engaged. The
list implementation is updated to support non-unique keys with few
collisions.
Xueming Li [Tue, 15 Sep 2020 03:05:53 +0000 (03:05 +0000)]
net/mlx5: use bond index for netdev operations
In case of bonding, device ifindex was detected as the PF ifindex, so
any operation using ifindex applied to PF instead of the bond device.
These operations includes MTU get/set, up/down and mac address
manipulation, etc.
This patch detects bond interface ifindex and name for PF that join a
bond interface, uses it by default for netdev operations.
Cc: stable@dpdk.org Signed-off-by: Xueming Li <xuemingl@nvidia.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
The Rx queue start/stop feature is not supported if vectorized
rx_burst routine is engaged. There was a routine address typo
and rx_burst type check was wrong.
Kalesh AP [Tue, 22 Sep 2020 05:34:16 +0000 (11:04 +0530)]
net/bnxt: fix link status during device recovery
Driver should not send the phy_cfg request to bring link down
during reset recovery. If the driver sends the phy_cfg request
in recovery process, then FW needs to re-establish the link which
in turn increases the recovery time based on PHY type and link partners.
Jeff Guo [Fri, 18 Sep 2020 05:46:56 +0000 (13:46 +0800)]
net/iavf: support GTPU outer and inner co-exist
Although currently only the gtpu inner hash be enabled while not the
gtpu outer hash, but the outer protocol still needed to co-exist with
inner protocol when configure the gtpu inner hash rule, that would
allow the gtpu inner hash support for the different outer protocols.
Signed-off-by: Jeff Guo <jia.guo@intel.com> Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Haiyue Wang [Tue, 22 Sep 2020 06:40:24 +0000 (14:40 +0800)]
net/ice: refactor Rx FlexiMD handling
The hardware supports many kinds of FlexiMDs set into Rx descriptor, and
the FlexiMDs can have different offsets in the descriptor according the
DDP package setting.
The FlexiMDs type and offset are identified by the RXDID, which will be
used to setup the queue.
For expanding to support different RXDIDs in the future, refactor the Rx
FlexiMD handling by the functions mapped to related RXDIDs.
Signed-off-by: Haiyue Wang <haiyue.wang@intel.com> Acked-by: Jeff Guo <jia.guo@intel.com>
RongQing Li [Fri, 18 Sep 2020 11:32:31 +0000 (19:32 +0800)]
net/af_xdp: avoid deadlock due to empty fill queue
While receiving packets, it is possible to fail to reserve
fill queue, since buffer ring is shared between tx and rx,
and maybe not available temporary. As a result both fill
queue and Rx queue will be empty.
Then kernel side will not be able to receive packets due to
empty fill queue, and dpdk will not be able to reserve fill
queue because dpdk doesn't have packets to receive, finally
deadlock will happen.
So move reserve fill queue before xsk_ring_cons__peek to fix it.
Cc: stable@dpdk.org Signed-off-by: RongQing Li <lirongqing@baidu.com> Signed-off-by: Dongsheng Rong <rongdongsheng@baidu.com> Acked-by: Ciara Loftus <ciara.loftus@intel.com>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:35 +0000 (07:30 +0200)]
net/ena: expose ENI stats as additional xstats
New HAL allows driver to read extra ENI stats. Exact meaning of each of
them can be found in base/ena_defs/ena_admin_defs.h file and structure
ena_admin_eni_stats.
The ena_eni_stats structure is exactly the same as ena_admin_eni_stats,
but it was required to be added for compatibility with xstats macros.
Reading ENI stats requires communication with the admin queue.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:34 +0000 (07:30 +0200)]
net/ena: lock dynamic usages of admin queue
There are some cases, where the admin queue commands after the
configuration phase finished - for example, the application could ask
for the driver statistics from multiple cores at once.
As by the design, the admin queue is not multithread safe, the spinlock
was added to protect all usages of the admin queue after the
configuration is done.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:33 +0000 (07:30 +0200)]
net/ena/base: update generation date and commit
The current ena_com version was generated on 26.04.2020.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Checking for the cdesc not being NULL doesn't have any sense if the idx
argument is not 0, so it can be skipped, as the error won't be detected
anyway.
To simplify that, only the 'i' value is being verified and the code is
breaking from the infinite loop in case when all descriptors were copied
into the buffer.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:31 +0000 (07:30 +0200)]
net/ena/base: convert values to 32-bit before shifting
When filling out meta descriptor, all values should be converted to the
desired type (u32 in case of the meta descriptor) to prevent losing the
data.
For example, io_sq->phase is of type u8. If
ENA_ETH_IO_TX_META_DESC_PHASE_SHIFT would be greater then 8, all data
would be lost.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:30 +0000 (07:30 +0200)]
net/ena/base: check null meta desc
Static code analysis showed up, that it's possible for meta_desc being
NULL. To avoid dereference of the NULL pointer, extra check was added if
the pointer is in fact valid.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:29 +0000 (07:30 +0200)]
net/ena/base: store admin stats as 64-bit
To minimize chance of integer overflow, the type of admin statistics was
changed from u32 to u64.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:28 +0000 (07:30 +0200)]
net/ena/base: add missing unlikely
To align the error checking code with other parts of the ena_com,
the conditional check is being tested for the error was wrapped inside
unlikely().
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:27 +0000 (07:30 +0200)]
net/ena/base: cleanup coding style
* Function argument style improvement (space after *)
* Align indentation of the define
* Typo fix in the documentation
* Remove extra empty line after license (aligned with other files)
* Extra alignment of one line was fixed
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:26 +0000 (07:30 +0200)]
net/ena/base: check for RSS key configuration support
Setting RSS hash function could not be supported by the device. In that
situation there is no need to fill in default hash key or even allocate
hash key.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:25 +0000 (07:30 +0200)]
net/ena/base: do not use hardcoded RSS key buffer size
It's well defined how the RSS key buffer looks from the device
perspective, so the constant value should be used instead of magic
number. Also it doesn't has to be calculated dynamically.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:24 +0000 (07:30 +0200)]
net/ena/base: split RSS function and hash getters
There is no need to keep single function for both hash function and
the key. If the caller want's to get only single value, then it had to
pass NULL as one of the values, making the API harder to use.
Except reading functions from the device, one can also use function
ena_com_get_current_hash_function() to get the integer value, which
is representing current hash function stored in the ena_com layer.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:23 +0000 (07:30 +0200)]
net/ena/base: add ENI stats
The Elastic Netfwork Interface (ENI) stats can be acquired from the HW.
They can provide advanced values which can be further used by the
application for better flow management.
It isn't available to the DPDK application, yet. The PMD must expose
them directly.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:22 +0000 (07:30 +0200)]
net/ena/base: rework setup of accelerated LLQ mode
The purpose of this change is general code simplification and
type safety improvement for the logical values.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:21 +0000 (07:30 +0200)]
net/ena/base: remove mmiowb not defined macro
As there is no replacement for mmiowb() and there is no need to use both
versions in the DPDK, this ifdef was simply removed.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:20 +0000 (07:30 +0200)]
net/ena/base: fix release of wait event
The wait event is being accessed without making sure it the completion
context exists. The check for that is just below, so it could be used
for releasing wait even safely.
Fixes: 3adcba9a8987 ("net/ena: update HAL to the newer version") Cc: stable@dpdk.org Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:19 +0000 (07:30 +0200)]
net/ena/base: make delay exponential in polling functions
Instead of the fixes, 5 ms delay in the polling functions, use
values into given range (by default from 100 us 5000 us) and increase
them exponentially each time, the operation isn't finished.
This change can improve responsiveness of the driver for the fast
operations.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:18 +0000 (07:30 +0200)]
net/ena/base: support admin status for resource busy
The admin command could return ENA_ADMIN_RESOURCE_BUSY status, which
is meaning that currently the given resource cannot be used.
However, the request can be repeated, so it's being converted to the
ENA_COM_TRY_AGAIN error code.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:17 +0000 (07:30 +0200)]
net/ena/base: specify delay operations
ENA_MSLEEP() and ENA_UDELAY() were expecting different behavior - the
first one is expecting driver to sleep, while the other, to busy wait.
For both cases, the rte_delay_(u|m)s() function was used, which could
be either sleep or block, depending on the configuration.
To make the macros valid, the operations should be specified directly.
Because of that, the rte_delay_us_sleep() and rte_delay_us_block() are
now being used.
Fixes: 9ba7981ec992 ("ena: add communication layer for DPDK") Cc: stable@dpdk.org Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Michal Krawczyk [Thu, 17 Sep 2020 05:30:16 +0000 (07:30 +0200)]
net/ena/base: use min/max macros with type conversion
Usage of RTE_MIN(MAX) in ENA_MIN32, ENA_MIN16, ENA_MIN8 (and same for
the MAX), was not enough, as the HAL code is assuming that those macros
will convert both arguments to the specified uintX_t type.
As RTE_MIN(MAX) is using 'typeof' operator, the behavior won't be the
same, especially if arguments has different types (and it could cause
compilation warnings).
To satisfy that, the ENA_MIN_T and ENA_MAX_T macros were added, which
are converting both arguments to the type which is being passed as an
argument.
Fixes: 9ba7981ec992 ("ena: add communication layer for DPDK") Cc: stable@dpdk.org Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
The macro defined for milliseconds sleep was not putting the thread
to sleep and was simply calling a delay routine. This fix redefines
the macro to call the correct rte sleep API.
Xiaoyun Wang [Mon, 14 Sep 2020 14:31:43 +0000 (22:31 +0800)]
net/hinic: fix TCAM filter set
hinic supports two methods: linear table and tcam table,
if tcam filter enables failed but linear table is ok,
which also needs to enable filter, so for this scene,
driver should not close fdir switch.
Fixes: f4ca3fd54c4d ("net/hinic: create and destroy flow director filter") Cc: stable@dpdk.org Signed-off-by: Xiaoyun Wang <cloud.wangxiaoyun@huawei.com>
The decorator may be used in the future to instruct linker
to put it into dedicated sections or hide.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Andy Moreton <amoreton@xilinx.com> Reviewed-by: Mark Spender <mspender@xilinx.com> Reviewed-by: Richard Houldsworth <rhouldsw@xilinx.com>
Andrew Rybchenko [Thu, 17 Sep 2020 06:34:28 +0000 (07:34 +0100)]
net/sfc/base: decorate libefx API functions
The decorators will be used in the future to mark libefx API
functions as __rte_internal.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Andy Moreton <amoreton@solarflare.com> Reviewed-by: Richard Houldsworth <rhouldsw@xilinx.com>
libefx coding standard requires it and the specifier is
used for almost all functions in the header file.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Andy Moreton <amoreton@solarflare.com> Reviewed-by: Richard Houldsworth <rhouldsw@xilinx.com>
examples/pipeline: add VXLAN encapsulation example
Add VXLAN encapsulation example to the SWX pipeline application. The
VXLAN tunnels can be generated with the vxlan_table.py script. Example
command line: ./build/pipeline -l0-1 -- -s ./examples/vxlan.cli
examples/pipeline: add l2fwd with MAC swap example
Add L2 Forwarding example with MAC destination and source address swap
to the SWX pipeline application. Example command line:
./build/pipeline -l0-1 -- -s ./examples/l2fwd_macswp.cli
Add network-based connectivity mechanism for the application to allow
for the exchange of configuration messages through the network as
opposed to local CLI only.
Add the PCAP file-based source (input) and sink (output) port types
for the SWX pipeline. The sink port is typically used to implement the
packet drop pipeline action. Used under the hood by the pipeline rx
and tx instructions.
Add support for building the SWX pipeline based on specification file
with syntax aligned to the P4 language. The specification file may be
generated by the P4C compiler in the future.
High-level transaction-oriented API for SWX pipeline table updates. It
supports multi-table atomic updates, i.e. multiple tables can be
updated in a single step with only the before and after table set
visible to the packets. Uses the lower-level table update mechanisms.
Instruction optimizer. Detects frequent patterns and replaces them
with some more powerful vector-like pipeline instructions without any
user effort. Executes at instruction translation, not at run-time.
pipeline: introduce SWX jump and return instructions
The jump instructions are either unconditional (jmp) or conditional on
positive/negative tests such as header validity (jmpv/jmpnv), table
lookup hit/miss (jmph/jmpnh), executed action (jmpa/jmpna), equality
(jmpeq/jmpneq), comparison result (jmplt/jmpgt). The return
instruction resumes the pipeline execution after action subroutine.
The extern instruction calls one of the member functions of a given
extern object or it calls the given extern function. The function
arguments must be written in advance to the mailbox. The results
are available in the same place after execution.
The table instruction looks up the input key into the table and then
it triggers the execution of the action found in the table entry. On
lookup miss, the default table action is executed.
The shr (i.e. shift right) instruction source can be header field (H),
meta-data field (M), extern object (E) or function (F) mailbox field,
table entry action data field (T) or immediate value (I). The
destination is HMEF.
The shl (i.e. shift left) instruction source can be header field (H),
meta-data field (M), extern object (E) or function (F) mailbox field,
table entry action data field (T) or immediate value (I). The
destination is HMEF.
The xor (i.e. bitwise exclusive or) instruction source can be header
field (H), meta-data field (M), extern object (E) or function (F)
mailbox field, table entry action data field (T) or immediate value
(I). The destination is HMEF.
The or (i.e. bitwise or) instruction source can be header field (H),
meta-data field (M), extern object (E) or function (F) mailbox field,
table entry action data field (T) or immediate value (I). The
destination is HMEF.
The and (i.e. bitwise and) instruction source can be header field (H),
meta-data field (M), extern object (E) or function (F) mailbox field,
table entry action data field (T) or immediate value (I). The
destination is HMEF.
The ckadd (i.e. checksum add) instruction is used to either compute,
verify or update the 1's complement sum commonly used by protocols
such as IPv4, TCP or UDP.
The sub (i.e. subtract) instruction source can be header field (H),
meta-data field (M), extern object (E) or function (F) mailbox field,
table entry action data field (T) or immediate value (I). The
destination is HMEF.
The add instruction source can be header field (H), meta-data field
(M), extern object (E) or function (F) mailbox field, table entry
action data field (T) or immediate value (I). The destination is HMEF.
The DMA instruction handles the bulk read transfer of one header from
the table entry action data. Typically used to generate headers, i.e.
headers that are not extracted from the input packet.
The mov (i.e. move) instruction source can be header field (H),
meta-data field (M), extern object (E) or function (F) mailbox field,
table entry action data field (T) or immediate value (I). The
destination is HMEF.
pipeline: add header validate and invalidate SWX instructions
Add instructions to flag a header as valid or invalid. This flag can
be tested by the jmpv (jump if header valid) and jmpnv (jump if header
not valid) instructions.
Add header emit and packet transmission instructions. Emit adds to the
output packet a header that is either generated (e.g. read from table
entry by action) or extracted from the input packet. Tx ends the
pipeline processing; discard is implemented by tx to special port.
Add packet reception and header extraction instructions. The Rx must
be the first pipeline instruction. Each extracted header is logically
removed from the packet, then it can be read/written by instructions,
emitted into the outgoing packet or discarded.
The SWX pipeline instructions represent the main program that defines
the life of the packet. As packets go through tables that trigger
action subroutines, the headers and meta-data get transformed along
the way.
Add tables to the SWX pipeline. The match fields are flexibly selected
from the headers and meta-data. The set of table actions is flexibly
selected for each table from the set of pipeline actions.
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.
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.
git.droids-corp.org / dpdk.git / log