[dpdk-dev] [PATCH 1/1] eal: Don't fail secondary if primary is missing tailqs

[Bruce Richardson]

On Tue, Nov 13, 2018 at 11:42:51PM +0000, Burdick, Cliff wrote:
> 
> ________________________________________
> From: Thomas Monjalon [thomas at monjalon.net]
> Sent: Tuesday, November 13, 2018 2:18 PM
> To: Burdick, Cliff
> Cc: Burakov, Anatoly; dev at dpdk.org; bruce.richardson at intel.com
> Subject: Re: [dpdk-dev] [PATCH 1/1] eal: Don't fail secondary if primary is missing tailqs
> 
> 13/11/2018 23:08, Burdick, Cliff:
> > From: Thomas Monjalon [mailto:thomas at monjalon.net]
> > > 13/11/2018 17:38, Burdick, Cliff:
> > > > From: Thomas Monjalon [mailto:thomas at monjalon.net]
> > > > 13/11/2018 16:45, Burdick, Cliff:
> > > > > From: Burakov, Anatoly [mailto:anatoly.burakov at intel.com]
> > > > > > On 13-Nov-18 9:21 AM, Thomas Monjalon wrote:
> > > > > > > 13/11/2018 00:33, Burdick, Cliff:
> > > > > > >> This patch was submitted by Jean Tourrilhes over two years ago,
> > > > > > >> but didn't receive any responses. I hit the same issue recently
> > > > > > >> when trying to use cgo (Golang) as a primary process linked to
> > > > > > >> libdpdk.a against a C++ application linked against the same
> > > > > > >> library.> > >
> > > > > > >
> > > > > > > The question is to know why you don't have the same constructors
> > > > > > > in primary and seconday?
> > > > > >
> > > > > > I've hit similar things in the past. I believe it was caused by our
> > > > > > build system stripping out unused libraries (such as rte_hash) from
> > > > > > the binary and thus not calling the constructor in the primary, but
> > > > > > doing so in the secondary (or something to that effect). In any
> > > > > > case, this is caused by linking different number of libraries to
> > > > > > primary and secondary, and should probably be fixed in the build
> > > > > > system, not in the tailqs code (unless we specifically support
> > > > > > having different linked libraries to primary and secondary?).> > > >
> > > > > Right, I think the original author of the patch stated the reasons in
> > > > > the link I provided. The build system seems like the most appropriate
> > > > > place to fix it, but the patch got me going quickly. I think the
> > > > > question is whether you want DPDK to support these other ways of
> > > > > linking. I'm certainly not the first to use cgo, since there's a
> > > > > virtual switch project doing the same:
> > > > >
> > > > > https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_lago
> > > > > pu
> > > > > s_vsw&d=DwICAg&c=jcv3orpCsv7C4ly8-ubDoUfrxF5xIGWmptxGWP5vi5w&r=m1RLQ
> > > > > OG
> > > > > Okz9MauvVLZmiGtyWc5mA7DejbPFNE1IDj-4&m=hQqVCNwW7eoEzB_hLFK97i8idS8FI
> > > > > qX oPeclwsIZq7Y&s=BMoBlwkqljwWIBY3SE3pPMCfVnOUlDuZLrno4-SojKM&e=
> > > > >
> > > > > They don't use primary/secondary processes, though, so the issue is
> > > > > never hit. I'm in a situation where using cgo seemed like the easiest
> > > > > path to accomplish what I'm doing since I needed specialized
> > > > > libraries for it that were not available in C/C++. At some point I
> > > > > bet someone would use Cython to start linking against DPDK as well,
> > > > > and we'd likely run into the same issue.> > > >
> > > > >For sure, we want to support using DPDK with cgo or cython.
> > > > >But it is not clear what is the relation with not having the same
> > > > >compilation for primary and secondary. Please could you elaborate?> > >
> > > > Hi Thomas, I think Jean explained it well here:
> > > > https://urldefense.proofpoint.com/v2/url?u=https-3A__dev.dpdk.narkive.
> > > > com_ZM3a7QD1_dpdk-2Ddev-2Dbug-2Dstatic-2Dconstructors-2Dconsidered-2De
> > > > vil&d=DwICAg&c=jcv3orpCsv7C4ly8-ubDoUfrxF5xIGWmptxGWP5vi5w&r=m1RLQOGOk
> > > > z9MauvVLZmiGtyWc5mA7DejbPFNE1IDj-4&m=C69wDgrjDX8_oXp1M_3bnmWOOZdGqwBBG
> > > > 9vzkyGDWGQ&s=YYn2N7WrzJpH1ptNrLZad0nPAQdrUyqBckk2uFuWYPQ&e=
> > > >
> > > > "The build system of the application does not have all the subtelties
> > > > of the DPDK build system, and ends up including *all* the
> > > > constructors, wether they are used or not in the code. Moreover, they
> > > > are included in a different order. Actually, by default the builds
> > > > include no constructors at all (which is a big fail), so the library
> > > > needs to be included with --whole-archive (see Snort DPDK
> > > > instructions)."
> > > >
> > > > I will get to the bottom of my exact case to understand what's
> > > > happening, but my primary application is a cgo application that I'm
> > > > linking to by using almost exactly the same flags that are used in the
> > > > DPDK build system to build examples. The DPDK libraries I'm linking
> > > > against is a single location for both primary and secondary; in other
> > > > words, I don't build DPDK twice.
> > > >
> > > > OK I understand, thanks.
> > > >
> > > > You had alluded to a pkg-config for DPDK in the 2015 thread, which cgo
> > > > can use, but I don't know if that ever was implemented. Cgo can use
> > > > pkg-config if it's available, otherwise the only tools are specifying
> > > > LDFLAGS and CFLAGS into their build system.
> > >Yes, the right answer is still pkg-config :) The good news is that it is now implemented thanks to the meson build system:
> > >     https://urldefense.proofpoint.com/v2/url?u=http-3A__git.dpdk.org_dpdk_tree_doc_build-2Dsdk-2Dmeson.txt-23n182&d=DwICAg&c=jcv3orpCsv7C4ly8-ubDoUfrxF5xIGWmptxGWP5vi5w&r=m1RLQOGOkz9MauvVLZmiGtyWc5mA7DejbPFNE1IDj->4&m=C69wDgrjDX8_oXp1M_3bnmWOOZdGqwBBG9vzkyGDWGQ&s=oC86KD_RJ1T6rfzi3x5zFT1Ri13ELpKmsyFqpgDbgFg&e=
> >
> > Hi Thomas, are there instructions on how to use pkg-config to build the mlx4/5 PMD? I noticed a patch was submitted in September to add support for it, but that link you provided on using meson doesn't say how to build specific drivers. It appears to be disabled by default.
> 
> > If the dependency is found, meson will enable the PMD when building DPDK.
> 
> Do you know where exactly that is? I've been using mlx5 for a while on this system, and I can see that 18.11-rc2 meson build+ninja built the pmd, but it's not on the --libs listing for pkg-config. Does it tell me what I was missing?
> 
For dynamic linking of applications, the drivers are not normally linked
in. Instead, they should be loaded from the drivers directory as .so files
- normally by default in EAL as the driver .so's should be copied to the
EAL_PMD_PATH where EAL finds them automatically. [This applies to both
meson and make builds, though only meson generates the .pc file for
pkg-config]

If you are doing a static build, then you need to explicitly link in the
drivers. You can get a list from pkg-config using the "--static" flag in
this case. A good example of how to use pkg-config in this way can be found
in the Makefiles for most examples, e.g. examples/helloworld/Makefile,
reproduced below.

Regards,
/Bruce

all: shared
.PHONY: shared static
shared: build/$(APP)-shared
        ln -sf $(APP)-shared build/$(APP)
static: build/$(APP)-static
        ln -sf $(APP)-static build/$(APP)

PC_FILE := $(shell pkg-config --path libdpdk)
CFLAGS += -O3 $(shell pkg-config --cflags libdpdk)
LDFLAGS_SHARED = $(shell pkg-config --libs libdpdk)
LDFLAGS_STATIC = -Wl,-Bstatic $(shell pkg-config --static --libs libdpdk)

build/$(APP)-shared: $(SRCS-y) Makefile $(PC_FILE) | build
        $(CC) $(CFLAGS) $(SRCS-y) -o $@ $(LDFLAGS) $(LDFLAGS_SHARED)

build/$(APP)-static: $(SRCS-y) Makefile $(PC_FILE) | build
        $(CC) $(CFLAGS) $(SRCS-y) -o $@ $(LDFLAGS) $(LDFLAGS_STATIC)

build:
        @mkdir -p $@