Age | Commit message (Collapse) | Author |
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
The previous patch made this unused.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
This patch reuses odp_packet_io.c and pktio/loop.c from linux-generic's
codebase, and plugs the DPDK specific pieces from odp_packet_dpdk.c into
that framework. In order to do that the following steps were taken:
- sync up odp_packet_io_internal.h: this file is nearly the same, except
we use only loop and our DPDK specific parts
- move DPDK specific functionality to odp_packet_dpdk.c, and use those
functions in dpdk_pktio_ops. The old pktio code will be deleted in a
separate patch
- enable classification in Makefiles and init: linux-generic code refers
to that, that's why we can't do this in separate patches
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
But don't parse L2 header by default, only on demand.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
It's an obsolete garbage.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
rte_eal_init() returns the consumed arguments, but it doesn't count the
first one, which is used only as a logging prefix. Count this when
checking return value.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
DPDK now takes care of this global variable, no need to save and reset
here as well.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
rte_eal_init() expects the first argument is the "program name", it is
used for logging purposes.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
This is now taken care by DPDK itself.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Quite a while ago, but the README referred to old data.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
This platform doesn't support that, therefore we have to create a local
copy of this Makefile.am
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
The DPDK memcpy version is also exposed through function pointer.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
There is a type compatibility issue in recv_pkt_dpdk(), although it
doesn't have an effect as both of them are pointers.
The rte_mbuf accessor functions are bringing up alignment change warnings,
which are false-positives in this case. These fields are defined with
their appropriate type in the original structure, so their alignment is
guaranteed, but the reason for these functions is to hide that structure.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
That way the helper tests can pick it up too.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
The only reason for a separate one is because odp_cpumask_default_worker()
goes through the lcore list to get this data for us. But with the master
thread's affinity fixed there is no need for that, the linux-generic way of
using the thread affinity is good for linux-dpdk as well.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
The others would be used for DPDK lcore threads, which are not accessible
for an ODP application. It requires ODP thread init to set up lcore_role,
though it's not widely used in other DPDK libs.
Also reset the original affinity of the calling thread, as DPDK limits it
down to one core.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
helper/linux.c is used instead of this.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
interface is stopped"
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
before receiving or sending"
And "45259bf4 linux-generic: pktio: factor state management into packet_io",
plus add some extra logging.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
stopped
DPDK checks that too but it doesn't return an error, so we should maintain the
status too.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Plus set the errno as well, and codify further that
odp_buffer_t == (struct rte_mbuf *) .
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
It doesn't make any sense to allow this function allocate from non-buffer
pools.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
common to all platforms"
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
DPDK already implements rte_strsplit() which can be used instead of the
parse_dpdk_args() function. This saves some dynamic memory allocation
and simplifies the code.
Signed-off-by: Miroslav Kiradzhiyski <miroslav@virtualopensystems.com>
Signed-off-by: Nikolay Nikolaev <n.nikolaev@virtualopensystems.com>
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
The calloc() has several issues in calculating the string size:
- doesn't account for the "0x" prefix
- nor the space after the core mask
- sizeof(core_mask) should be doubled, as each byte can be represented as two
hex characters
But also the width of coremask cause a limitation. This patch uses the
odp_cpumask_*() functions to create a proper mask.
Reported-by: Stuart Haslam <stuart.haslam@linaro.org>
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
The platform_params parameter will convey that through.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
- ixgbe vPMD now has lower requirement for recv burst
- mlx5 supported
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
Reported-by: P D, Arun <arun.p_d@nokia.com>
|
|
It was forgotten at the time of upgrade to 2.1.0.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
It is DPDK's errno variable. odp_pktio_send() also makes sure that it is set
even if the PMD doesn't do that.
The macro in odp_buffer_inlines.h is a workaround to make sure linux-generic
timer implementation doesn't have a linking issue. It would be slightly
better to have it in a local copy of odp_internal.h, but having it in sync
with linux-generic is more important.
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|
|
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
|