Support to calculate PDCP checksums. Includes changes to
pass in dictionaries. Adds support for sync flush
Signed-off-by: Sameer Vaze
---
v3:
* Moved variable defintions to start of function
* Added macros for nibble in double word
* Removed extra lines
drivers/compress/zlib/zlib_pmd.c
Adds support to read and apply dictionarie
Signed-off-by: Sameer Vaze
---
v3:
* Removed extra line
* Moved variable defitions to start of function
* Free dictionary if erroring out
app/test-compress-perf/comp_perf_options.h| 4 +
.../comp_perf_options_parse.c | 15 +++
.../
Adds definitions for PDCP checksums and apis to pass in
dictionaries
Signed-off-by: Sameer Vaze
---
v3:
* Renamed checksum validation enum
* Added missed comments for window size enum
lib/compressdev/rte_comp.h | 27 +++
1 file changed, 27 insertions(+)
diff --git a/lib
> Adds support to read and apply dictionaries
>
> Signed-off-by: Sameer Vaze
> ---
> v2:
> * Handles relocated dictionary params
>
> app/test-compress-perf/comp_perf_options.h| 4 +
> .../comp_perf_options_parse.c | 16
> .../comp_perf_test_verify.c |
> Adds definitions for PDCP checksums and apis to pass in
> dictionaries
>
> Signed-off-by: Sameer Vaze
> ---
> v2:
> * Moved dictionary params to deflate params
> * Added infalte params to union
>
> lib/compressdev/rte_comp.h | 26 ++
> 1 file changed, 26 insertions(+)
Hey Akhil,
I pushed a reworked v2 patch based on the comments below:
https://patches.dpdk.org/project/dpdk/list/?series=36288
Even though I included --in-reply-to to the older series it seems to have
created a new series. Is that expected or is the command I used incorrect? The
command I used
Adds definitions for PDCP checksums and apis to pass in
dictionaries
Signed-off-by: Sameer Vaze
---
v2:
* Moved dictionary params to deflate params
* Added infalte params to union
lib/compressdev/rte_comp.h | 26 ++
1 file changed, 26 insertions(+)
diff --git a/lib/comp
The queue stats part of the rte_eth_stats structure has been deprecated
for many years now, since 2020 [1]. Therefore we look to remove these
fields from the stats structure.
Unfortunately, the flag introduced by the deprecation,
"RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS", means that for drivers using it
>
> >
> > On Mon, Sep 29, 2025 at 02:30:38PM +, Ciara Loftus wrote:
> > > Introduce a function that allows a VF to request the PF to reset itself.
> > > This is useful for example when the application detects that one of the
> > > queues have hung or any event where a reset is required and the
Expose PFC statistics in xstats.
Signed-off-by: Vladimir Medvedkin
---
drivers/net/intel/ice/ice_ethdev.c | 63 ++
1 file changed, 63 insertions(+)
diff --git a/drivers/net/intel/ice/ice_ethdev.c
b/drivers/net/intel/ice/ice_ethdev.c
index 6cfcc3a46e..0b193b1c31 1006
This series introduces an initial implementation of DCB and PFC to the ICE PMD.
This includes adding support for TCs in ice_aq_set_mac_cfg() and asymmetric
DCB/PFC configuration in base code.
Current implementation relies on current API and has a number of limitation and
assumptions such as:
-
This patch add support for Priority Flow Control (PFC)
Signed-off-by: Vladimir Medvedkin
---
drivers/net/intel/ice/ice_ethdev.c | 125 -
1 file changed, 124 insertions(+), 1 deletion(-)
diff --git a/drivers/net/intel/ice/ice_ethdev.c
b/drivers/net/intel/ice/ice_ethd
This patch adds support for Data Center Bridging (DCB)
Signed-off-by: Vladimir Medvedkin
---
drivers/net/intel/ice/ice_ethdev.c | 184 +
drivers/net/intel/ice/ice_rxtx.c | 21
2 files changed, 205 insertions(+)
diff --git a/drivers/net/intel/ice/ice_ethdev.c
In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
In the single queue model, the same descriptor queue is used by SW
to post descriptors to the device and used by device to report completed
descriptors to SW. While as the split queue model separates t
Use the new cpuset to string function to replace other thread-specific
functions which did similar work. Also use it for additional debug logs.
Signed-off-by: Bruce Richardson
---
lib/eal/common/eal_common_lcore.c | 13 -
lib/eal/common/eal_common_options.c | 22 +++
lib/ea
On 10/2/25 7:28 PM, Bruce Richardson wrote:
On Thu, Oct 02, 2025 at 06:42:15PM +0300, Ali Alnubani wrote:
On 10/2/25 3:45 PM, David Marchand wrote:
On Thu, 2 Oct 2025 at 13:54, David Marchand wrote:
On Wed, 24 Sept 2025 at 13:13, Bruce Richardson
wrote:
The meson find_library() API call
http://bugs.dpdk.org/show_bug.cgi?id=1798
Bug ID: 1798
Summary: Meson can't find MLX5 build dependencies with clang
Product: DPDK
Version: unspecified
Hardware: All
OS: All
Status: UNCONFIRMED
Severity: no
Now that we have removed the queue stats from the ethdev stats, we need
to update the various PMDs to take account of the changes.
Update each stats_get function to take the extra parameters for qstats,
just marking it as unused if the driver does not fill in queue stats.
For those that do complet
The ipn3ke driver does not actually report per-queue stats, it always
sets the values to zero, so drop the flag causing auto-generation of
these meaningless xstats.
Signed-off-by: Bruce Richardson
Reviewed-by: Rosen Xu
---
drivers/net/intel/ipn3ke/ipn3ke_representor.c | 11 +--
1 file c
Move text from deprecation notice to release note, and update.
Signed-off-by: Bruce Richardson
---
doc/guides/rel_notes/deprecation.rst | 7 ---
doc/guides/rel_notes/release_25_11.rst | 6 ++
2 files changed, 6 insertions(+), 7 deletions(-)
diff --git a/doc/guides/rel_notes/deprecatio
The ice driver does not fill in the queue statistics in the
rte_eth_stats structure, meaning that the per-queue xstats reported are
meaningless, and always zero. Don't set the device flag
"RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS" so these zero-stats are not
automatically created by the ethdev layer.
Sig
The description of the rx_queue_id parameter was copy-pasted from the
queue mapping function without being updated. Fix the text.
Fixes: 81f9db8ecc2c ("ethdev: add vlan offload support")
Cc: sta...@dpdk.org
Signed-off-by: Bruce Richardson
---
lib/ethdev/rte_ethdev.h | 2 +-
1 file changed, 1 in
On Mon, Sep 29, 2025 at 09:41:16AM -0700, Stephen Hemminger wrote:
> On Mon, 29 Sep 2025 16:00:03 +0100
> Bruce Richardson wrote:
>
> > NOTE: as called out below, on apply patches 3-6 should be squashed. They
> > are separated out here for easier review!
> >
> > Since DPDK 20.11 release, t
Deprecate the queue stats mapping functions in the ethdev API, since we
no longer have queue stats in the main device statistics. As part of the
deprecation, rather than allowing test-pmd to use the deprecated
functions just remove the mapping functionality from testpmd itself.
Without the queue st
Adjust applications to take account of the fact that queue stats are no
longer part of ethdev stats.
- For proc-info, we have to drop the output of the stats, since they are
no longer present.
- For the virtual_pmd in the test app, we adjust our stats_get function.
- For testpmd, we can no longe
NOTE: as called out below, on apply patches 3-6 should be squashed. They
are separated out here for easier review!
Since DPDK 20.11 release, the use of queue stats inside the rte_eth_stats
structure has been deprecated with the intention to remove them. Sadly,
despite 5 years passing that ha
This patch adds an IAVF testpmd command "port restore " which
will send a request to the PF to reset the given VF, followed by the
VF then reinitialising and restarting itself.
Signed-off-by: Ciara Loftus
---
drivers/net/intel/iavf/iavf_testpmd.c | 52 +++
1 file changed,
In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
In the single queue model, the same descriptor queue is used by SW
to post descriptors to the device and used by device to report completed
descriptors to SW. While as the split queue model separates t
In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
In the single queue model, the same descriptor queue is used by SW
to post descriptors to the device and used by device to report completed
descriptors to SW. While as the split queue model separates t
When run without an coremask or core list parameters, EAL detects the
working set of cores allowed for the current process and uses that as a
core list for use. Currently, this does not work for cores which are
numbered >= RTE_MAX_LCORE.
We can fix this by allowing the remap option to also be used
TB Members Attending: 7/11
--
- Bruce
- Thomas
- Kevin
- Konstantin
- Morten
- Jerin
- Maxime
NOTE: The Technical Board meetings take place every second Wednesday at 3 pm UTC
on
https://zoom-lfx.platform.linuxfoundation.org/meeting/96459488340?password=d808f1f6-0a28-4165-9
When the arguments passed to argparse include -h/--help then usage
information is automatically printed. Provide the capability for the
user to supply their own help function for this.
Signed-off-by: Bruce Richardson
---
doc/guides/prog_guide/argparse_lib.rst | 16
doc/guides/re
Use a utility function and macro to simplify the code for checking for
conflicting cmdline options. The checking can also be done at the
initial argument collating stage, shortening the argument
processing function which is very much on the long side.
Signed-off-by: Bruce Richardson
---
lib/eal/
The existing "available_cores" function in eal_common_options.c has
general code for converting a set of cores to a printable string.
Generalize this code into an "eal_cpuset_to_str" function which takes a
cpuset as parameter, and create a new "available_cores" function using
this new utility funct
Rather than working off arrays of lcore values, update the coremask
handling to use rte_cpuset_t. The use of the array provides no
additional value for coremasks, since the order of the bits set will
always be from lowest core number to highest, making the tracking of
what core was 0, 1, 2 etc. poi
Allow the "-R" flag to be used with core lists so that we can specify
lists with core ids >= RTE_MAX_LCORE in them, e.g. -l 140-148. In order
to do so, we limit the use of the "@" symbol for per-core remapping when
specifying a core list for automatic/global remapping, rather than
dealing with the
Add the new "--remap-lcore-ids" or "-R" option to signify that lcore ids
are to start at a low range. This takes an optional parameter to specify
the value to start from. Initially, use this to adjust the core mappings
when using a coremask.
As well as remapping within the range of RTE_MAX_LCORE,
The tailq list options from commandline require initialization before
being used. This was being done by explicitly initing those values in
the collate function. A better option is to do so using macros at time
of structure definition, that way, adding a new list option does not
require any other c
When rte_eal_init fails part way through, any saved EAL arguments need
to be freed, and the run_once flag needs to be set back to zero again.
The former task was never done on failure, and the latter was only done
on some occasions. Rework the error handling to always go to an err_out
label where c
Remove the separate function to check combinations of cmdline
parameters. Instead, just do those checks when parsing the parameters
since we have all info about what parameters are provided at that point.
Signed-off-by: Bruce Richardson
---
lib/eal/common/eal_common_memory.c | 3 +-
lib/eal/c
DPDK traditionally has iterated through all args and processed them as
they appear in the commandline. The arg processing logic can be
simplified if instead we initially gather all arguments into a structure
which is then processed with the arguments dealt with in a fixed/known
order.
Signed-off-b
Create eal_option_list.h, containing all the possible EAL parameters,
and basic info about them, such as type, whether they take a parameter
or not. Each entry is defined using a macro, which will be then
interpreted when the file is included.
First time this header in included in the eal_common_o
To simplify future rework of the EAL arg handling, add a long-option
equivalent for each short option that doesn't already have one.
When updating the docs with the new long options, standardize the format
of options which have both short and long variants, and drop the
deprecated service-coremask
The ultimate of this patchset is to make it easier to run on systems
with large numbers of cores, by simplifying the process of using core
numbers >RTE_MAX_LCORE. The new EAL arg ``-remap-lcore-ids``, also
shortened to ``-R``, is added to DPDK to support this.
However, in order to add this new fla
Make the function to print out the help text for an argparse object a
public function, which takes as a new parameter the file stream on which
to print. This can be used in future to allow application to extend
their own help information.
Signed-off-by: Bruce Richardson
---
lib/argparse/rte_argp
Core lists are widely used in DPDK, so add support for parsing them.
Signed-off-by: Bruce Richardson
---
app/test/test_argparse.c | 188 +
doc/guides/prog_guide/argparse_lib.rst | 32 +
lib/argparse/rte_argparse.c| 63 +
lib/argpars
The different value types supported by the argparse library are not
discussed in the documentation, so add some reference material about
them.
Signed-off-by: Bruce Richardson
---
doc/guides/prog_guide/argparse_lib.rst | 73 +-
1 file changed, 71 insertions(+), 2 deletions
Introduce a string define for the currently running OS, or execution
environment.
Originally, with old make build system, CONFIG_RTE_EXEC_ENV used to hold
this name string, but the variable seems to have been missed in the
meson build system, until commit cadb255e25d6 ("eal: add OS defines for
C c
Running rte_exit() in a forked process means that shared memory will be
released by the child process before the parent process does the same.
This issue has been seen recently when some GHA virtual machine (with
some mlx5 devices) runs the debug_autotest unit test.
Instead, run rte_panic() and rt
On Fri, Oct 03, 2025 at 08:51:01AM +0200, David Marchand wrote:
> Running rte_exit() in a forked process means that shared memory will be
> released by the child process before the parent process does the same.
> This issue has been seen recently when some GHA virtual machine (with
> some mlx5 devi
> Sorry for the various series I pushed. I attempted in reply to but it seemed
> to
> create a new series. If I am making changes to one commit in a series, do I
> need
> to resend all commits with an updated version? In such a case do I reply to
> the
> series or the specific change I am updati
51 matches
Mail list logo
