| 9eb60580 | 24-Jan-2025 |
Bruce Richardson <bruce.richardson@intel.com> |
net/intel: extract common Rx vector criteria
While some drivers have specific criteria for when a vector driver can
be enabled on the Rx path, there are a number of basic criteria which
apply across
net/intel: extract common Rx vector criteria
While some drivers have specific criteria for when a vector driver can
be enabled on the Rx path, there are a number of basic criteria which
apply across all drivers. Centralize those in the intel/common folder,
and then update drivers to use the common conditional checks. This adds
some additional restrictions to some drivers like ixgbe, where those
conditions were necessary but never checked.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Praveen Shetty <praveen.shetty@intel.com>
show more ...
|
| 61dcf278 | 24-Jan-2025 |
Bruce Richardson <bruce.richardson@intel.com> |
net/intel: create common mbuf initializer
Across a number of drivers, the same code is used for initializing the
"mbuf_initializer" value inside the rx queue structure for use with the
vector driver
net/intel: create common mbuf initializer
Across a number of drivers, the same code is used for initializing the
"mbuf_initializer" value inside the rx queue structure for use with the
vector drivers. Since the rx queue structures are (currently) different
across the drivers, we cannot just move a single copy of the function to
a common location. Instead, we create a dedicated function which just
creates the mbuf initializer for a particular port.
In creating this function, we can shorten it vs the original versions by
initializing the mbuf fields as they are defined, rather than
afterwards. We can also remove the use of the barrier and temporary
uintptr_t variable, because the mbuf has been reworked so that
rearm_data is a proper single-element array in a union.
Across ixgbe, i40e, iavf and i40e, we can call this function to
initialize the rxq data, replacing the "*_rxq_vec_setup_default"
functions. Only the i40e was slightly different, having an extra
assignment in it, to set the "sse" flag (even in case of neon and
altivec paths). This assignment was just duplicated to the calling sites
for simplicity and to keep existing behaviour.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Praveen Shetty <praveen.shetty@intel.com>
show more ...
|
| d3bb1c9e | 24-Jan-2025 |
Bruce Richardson <bruce.richardson@intel.com> |
net/intel: remove unneeded vector flags and cleanup code
With all drivers using the common Tx structure updated so that their
vector paths all use the simplified Tx mbuf ring format, it's no longer
net/intel: remove unneeded vector flags and cleanup code
With all drivers using the common Tx structure updated so that their
vector paths all use the simplified Tx mbuf ring format, it's no longer
necessary to have a separate flag for the ring format and for use of a
vector driver.
Remove the former flag and base all decisions off the vector flag. With
that done, we go from having only two paths to consider for releasing
all mbufs in the ring, not three. That allows further simplification of
the "ci_txq_release_all_mbufs" function.
The separate function to free buffers from the vector driver not using
the simplified ring format can similarly be removed as no longer
necessary.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Praveen Shetty <praveen.shetty@intel.com>
show more ...
|
| 7e230d56 | 24-Jan-2025 |
Bruce Richardson <bruce.richardson@intel.com> |
net/ice: use vector SW ring for all vector paths
The AVX-512 code path used a smaller SW ring structure only containing
the mbuf pointer, but no other fields. The other fields are only used in
the s
net/ice: use vector SW ring for all vector paths
The AVX-512 code path used a smaller SW ring structure only containing
the mbuf pointer, but no other fields. The other fields are only used in
the scalar code path, so update all vector driver code paths to use the
smaller, faster structure.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Praveen Shetty <praveen.shetty@intel.com>
show more ...
|
| 552979df | 24-Jan-2025 |
Bruce Richardson <bruce.richardson@intel.com> |
net/intel: create common Tx queue mbuf cleanup
The function to loop over the Tx queue and clean up all the mbufs on
it, e.g. for queue shutdown, is not device specific and so can move into
the commo
net/intel: create common Tx queue mbuf cleanup
The function to loop over the Tx queue and clean up all the mbufs on
it, e.g. for queue shutdown, is not device specific and so can move into
the common/intel headers. Only complication is ensuring that the
correct ring format, either minimal vector or full structure, is used.
Ice driver currently uses two functions and a function pointer to help
with this - though actually one of those functions uses a further check
inside it - so we can simplify this down to just one common function,
with a flag set in the appropriate place. This avoids checking for
AVX-512-specific functions, which were the only function using the
smaller struct in this driver.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Praveen Shetty <praveen.shetty@intel.com>
show more ...
|
| 0f62bbef | 24-Jan-2025 |
Bruce Richardson <bruce.richardson@intel.com> |
net/iavf: use common Tx free for AVX-512
Switch the iavf driver to use the common Tx free function. This requires
one additional parameter to that function, since iavf sometimes uses
context descrip
net/iavf: use common Tx free for AVX-512
Switch the iavf driver to use the common Tx free function. This requires
one additional parameter to that function, since iavf sometimes uses
context descriptors which means that we have double the descriptors per
SW ring slot.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Praveen Shetty <praveen.shetty@intel.com>
show more ...
|
| f6f34a5c | 24-Jan-2025 |
Bruce Richardson <bruce.richardson@intel.com> |
net/intel: create common Tx buffer free for AVX-512
AVX-512 code paths for ice and i40e drivers are common, and differ from
the regular post-Tx free function in that the SW ring from which the
buffe
net/intel: create common Tx buffer free for AVX-512
AVX-512 code paths for ice and i40e drivers are common, and differ from
the regular post-Tx free function in that the SW ring from which the
buffers are freed does not contain anything other than the mbuf pointer.
Merge these into a common function in intel/common to reduce
duplication.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Praveen Shetty <praveen.shetty@intel.com>
show more ...
|
| b87fc211 | 24-Jan-2025 |
Bruce Richardson <bruce.richardson@intel.com> |
net/intel: create common post-Tx buffer free
The actions taken for post-Tx buffer free for the SSE and AVX drivers
for i40e, iavf and ice drivers are all common, so centralize those in
net/intel/com
net/intel: create common post-Tx buffer free
The actions taken for post-Tx buffer free for the SSE and AVX drivers
for i40e, iavf and ice drivers are all common, so centralize those in
net/intel/common driver.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Praveen Shetty <praveen.shetty@intel.com>
show more ...
|