[dpdk-dev] [Bug 333] PCI devices not found when DPDK is linked as dynamic libraries
Stephen Hemminger
stephen at networkplumber.org
Sat Jul 20 01:17:24 CEST 2019
On Fri, 19 Jul 2019 20:43:04 +0000
bugzilla at dpdk.org wrote:
> https://bugs.dpdk.org/show_bug.cgi?id=333
>
> Bug ID: 333
> Summary: PCI devices not found when DPDK is linked as dynamic
> libraries
> Product: DPDK
> Version: 19.08
> Hardware: All
> OS: Linux
> Status: UNCONFIRMED
> Severity: critical
> Priority: Normal
> Component: core
> Assignee: dev at dpdk.org
> Reporter: stephen at networkplumber.org
> Target Milestone: ---
>
> If DPDK is configured with RTE_SHARED_LIB=y then applications are not loading
> the required PCI library and therefore devices are not found.
>
> Example is that l3fwd works if linked statically but fails if built with shared
> libraries.
>
> # ./examples/l3fwd/build/l3fwd -n4 -l0-3 -w 02:00.0
> EAL: Detected 8 lcore(s)
> EAL: Detected 1 NUMA nodes
> EAL: failed to parse device "02:00.0"
> EAL: Unable to parse device '02:00.0'
> EAL: Error - exiting with code: 1
> Cause: Invalid EAL parameters
>
> The string 02:00.0 should have been parsed by PCI library and that will find
> the device. but since library is not pulled in, the device is not found.
>
There are two levels of problems here.
#1 All the bus drivers work by registering themselves in the bus subsystem.
But with the --as-needed flag the RTE_INIT() that registers is not run on
PCI (or other bus) when built as shared library. This is resolved by flag
changes in mk.rteapp.mk.
#2 All the PMD drivers have the same problem. Even though they get built,
their RTE_INIT does not get called either since they are not on the application
link line.
#3 Similar problems with mempools.
Not sure if the real answer is to always link application with the
libdpdk.so (combined group) instead of the laundry list of libraries.
Lastly, why is shared library build not tested as part of the CI
infrastructure?
More information about the dev
mailing list