>-----Original Message----- >From: dev [mailto:dev-boun...@openvswitch.org] On Behalf Of Kavanagh, Mark B >Sent: Wednesday, April 8, 2015 10:26 PM >To: Daniele Di Proietto >Cc: dev@openvswitch.org; Butler, Siobhan A >Subject: Re: [ovs-dev] [PATCH] DPDK: add support for v2.0.0 > >>Thanks for the patch, I’m looking forward to use new >>DPDK 2.0 features! >> >Hi Daniele - sure, no problem :) > >>I have some doubts about the required -mssse3 option. >>The main point is: does DPDK always requires a CPU >>with SSSE3? >> >>I believe the answer is yes (Please correct me if I’m wrong). >>If CONFIG_RTE_MACHINE is set to “default", MACHINE_CFLAGS are >>set to ‘-march=core2’ in mk/machine/default/rte.vars.mk. >>If I remove -march=core2, rte_memcpy.h fails to compile. >>(I’m using GCC 4.9.1 from Debian testing) >> > >Your hypothesis seems sound - I haven't found any concrete documentation >regarding the >minimum-required CPU microarchitecture for DPDK, but I've spoken briefly to >some >colleagues in the DPDK team who have anecdotally confirmed this; I'll dig >around tomorrow >and come back with something concrete. > >>If this is the case (DPDK requires at least a core2 CPU, which >>should support SSSE3 extension according to >>https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gcc/i386-and-x86-64-Options.html), >>I would prefer adding the ‘-mssse3’ option to OVS_CFLAGS in >>OVS_CHECK_DPDK (acinclude.m4) >>
DPDK team have confirmed this - I've updated acinclude.m4 accordingly. > >This is the first solution that I'd considered, but tbh I wasn't sure if it >was the right >approach at the time. In the end I implemented an additional config option >instead - >there's some discussion on that here: >http://openvswitch.org/pipermail/dev/2015- >March/053091.html (FYI, this thread spans both the March and April archives). >For what >it's worth, I agree with your approach - we probably need Ben and/or Pravin to >weigh in on >the preferred solution. > >> >>Last thing: on my system, clang (3.5.0) still needs >>-Wno-error=cast-align to compile rte_memcpy.h. >> >I generally use gcc, but I'll look into this tomorrow also. Thanks for your >input! > I've adjusted INSTALL.DPDK.md and the travis build script to reflect same. Thanks to everyone for the review comments. >>What do you think? >> >>Daniele >> >>> On 8 Apr 2015, at 16:34, Mark Kavanagh <mark.b.kavan...@intel.com> wrote: >>> >>> Update relevant artifacts to add support for DPDK v2.0.0 >>> - INSTALL.DPDK.md >>> - travis build script >>> - netdev-dpdk: fix build with unified offload types in DPDK v2.0.0 >>> >>> Note that this breaks compatibility with DPDK v1.8.0 >>> >>> Signed-off-by: Mark Kavanagh <mark.b.kavan...@intel.com> >>> Signed-off-by: Panu Matilainen <pmati...@redhat.com> >>> --- >>> .travis/build.sh | 6 +++--- >>> INSTALL.DPDK.md | 10 +++++++--- >>> lib/netdev-dpdk.c | 3 +-- >>> 3 files changed, 11 insertions(+), 8 deletions(-) >>> >>> diff --git a/.travis/build.sh b/.travis/build.sh >>> index b6b701c..b6fe0eb 100755 >>> --- a/.travis/build.sh >>> +++ b/.travis/build.sh >>> @@ -68,11 +68,11 @@ fi >>> >>> if [ "$DPDK" ]; then >>> if [ -z "$DPDK_VER" ]; then >>> - DPDK_VER="1.8.0" >>> + DPDK_VER="2.0.0" >>> fi >>> install_dpdk $DPDK_VER >>> - # Disregard bad function casts until DPDK is fixed >>> - CFLAGS="$CFLAGS -Wno-error=bad-function-cast -Wno-error=cast-align" >>> + # Support older versions of GCC >>> + CFLAGS="-mssse3" >>> EXTRA_OPTS+="--with-dpdk=./dpdk-$DPDK_VER/build" >>> elif [ $CC != "clang" ]; then >>> # DPDK headers currently trigger sparse errors >>> diff --git a/INSTALL.DPDK.md b/INSTALL.DPDK.md >>> index 60889d0..8e6f6af 100644 >>> --- a/INSTALL.DPDK.md >>> +++ b/INSTALL.DPDK.md >>> @@ -16,13 +16,13 @@ OVS needs a system with 1GB hugepages support. >>> Building and Installing: >>> ------------------------ >>> >>> -Required DPDK 1.8.0, `fuse`, `fuse-devel` (`libfuse-dev` on Debian/Ubuntu) >>> +Required DPDK 2.0, `fuse`, `fuse-devel` (`libfuse-dev` on Debian/Ubuntu) >>> >>> 1. Configure build & install DPDK: >>> 1. Set `$DPDK_DIR` >>> >>> ``` >>> - export DPDK_DIR=/usr/src/dpdk-1.8.0 >>> + export DPDK_DIR=/usr/src/dpdk-2.0 >>> cd $DPDK_DIR >>> ``` >>> >>> @@ -66,8 +66,12 @@ Required DPDK 1.8.0, `fuse`, `fuse-devel` (`libfuse-dev` >>> on >>Debian/Ubuntu) >>> cd $(OVS_DIR)/openvswitch >>> ./boot.sh >>> ./configure --with-dpdk=$DPDK_BUILD >>> - make >>> + make [CFLAGS="-g -O2 -mssse3"] >>> ``` >>> +Note that users of older versions of GCC must pass additional CFLAGS >>> +("-g -O2 -mssse3") to 'make'. This issue has been observed with GCC >>> versions >>> +<= 4.8.3, and is not present in GCC v4.9.2 - applicability to interim >>> versions is >>> +untested. >>> >>> To have better performance one can enable aggressive compiler optimizations >>> and >>> use the special instructions(popcnt, crc32) that may not be available on all >>> diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c >>> index f69154b..5ab119a 100644 >>> --- a/lib/netdev-dpdk.c >>> +++ b/lib/netdev-dpdk.c >>> @@ -117,8 +117,7 @@ static const struct rte_eth_conf port_conf = { >>> .rx_adv_conf = { >>> .rss_conf = { >>> .rss_key = NULL, >>> - .rss_hf = ETH_RSS_IPV4_TCP | ETH_RSS_IPV4 | ETH_RSS_IPV6 >>> - | ETH_RSS_IPV4_UDP | ETH_RSS_IPV6_TCP | >>> ETH_RSS_IPV6_UDP, >>> + .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, >>> }, >>> }, >>> .txmode = { >>> -- >>> 1.7.4.1 >>> >>> _______________________________________________ >>> dev mailing list >>> dev@openvswitch.org >>> https://urldefense.proofpoint.com/v2/url?u=http- >>3A__openvswitch.org_mailman_listinfo_dev&d=AwIGaQ&c=Sqcl0Ez6M0X8aeM67LKIiDJAXVeAw- >>YihVMNtXt- >>uEs&r=SmB5nZacmXNq0gKCC1s_Cw5yUNjxgD4v5kJqZ2uWLlE&m=b6TCcALkXJRwrk15b5jrDqbBpGEWbDBxnXBSK >d >>AYnjo&s=cHDYfYnLsmOAsNIfukJ5AjjEtU6JOjTPX-tyPT7-4kY&e= > >_______________________________________________ >dev mailing list >dev@openvswitch.org >http://openvswitch.org/mailman/listinfo/dev _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
