From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtp.tuxdriver.com (charlotte.tuxdriver.com [70.61.120.58]) by dpdk.org (Postfix) with ESMTP id 992362E81 for ; Fri, 30 Jan 2015 19:13:00 +0100 (CET) Received: from hmsreliant.think-freely.org ([2001:470:8:a08:7aac:c0ff:fec2:933b] helo=localhost) by smtp.tuxdriver.com with esmtpsa (TLSv1:AES128-SHA:128) (Exim 4.63) (envelope-from ) id 1YHG3n-0008HW-Iv; Fri, 30 Jan 2015 13:12:55 -0500 Date: Fri, 30 Jan 2015 13:12:49 -0500 From: Neil Horman To: "Gonzalez Monroy, Sergio" Message-ID: <20150130181249.GC2664@hmsreliant.think-freely.org> References: <1422544811-26385-1-git-send-email-sergio.gonzalez.monroy@intel.com> <20150129163859.GE1999@hmsreliant.think-freely.org> <91383E96CE459D47BCE92EFBF5CE73B004F43D9B@IRSMSX108.ger.corp.intel.com> <20150129194539.GG1999@hmsreliant.think-freely.org> <91383E96CE459D47BCE92EFBF5CE73B004F453D7@IRSMSX108.ger.corp.intel.com> <20150130140507.GA2664@hmsreliant.think-freely.org> <91383E96CE459D47BCE92EFBF5CE73B004F45534@IRSMSX108.ger.corp.intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <91383E96CE459D47BCE92EFBF5CE73B004F45534@IRSMSX108.ger.corp.intel.com> User-Agent: Mutt/1.5.23 (2014-03-12) X-Spam-Score: -2.9 (--) X-Spam-Status: No Cc: "dev@dpdk.org" Subject: Re: [dpdk-dev] [PATCH 0/8] Improve build process X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 30 Jan 2015 18:13:01 -0000 On Fri, Jan 30, 2015 at 05:38:49PM +0000, Gonzalez Monroy, Sergio wrote: > > From: Neil Horman [mailto:nhorman@tuxdriver.com] > > Sent: Friday, January 30, 2015 2:05 PM > > To: Gonzalez Monroy, Sergio > > Cc: Thomas Monjalon; dev@dpdk.org > > Subject: Re: [dpdk-dev] [PATCH 0/8] Improve build process > > > > On Fri, Jan 30, 2015 at 01:39:28PM +0000, Gonzalez Monroy, Sergio wrote: > > > > From: Neil Horman [mailto:nhorman@tuxdriver.com] > > > > Sent: Thursday, January 29, 2015 7:46 PM > > > > To: Gonzalez Monroy, Sergio > > > > Cc: dev@dpdk.org > > > > Subject: Re: [dpdk-dev] [PATCH 0/8] Improve build process > > > > > > > > On Thu, Jan 29, 2015 at 05:04:20PM +0000, Gonzalez Monroy, Sergio > > wrote: > > > > > > From: Neil Horman [mailto:nhorman@tuxdriver.com] > > > > > > Sent: Thursday, January 29, 2015 4:39 PM > > > > > > To: Gonzalez Monroy, Sergio > > > > > > Cc: dev@dpdk.org > > > > > > Subject: Re: [dpdk-dev] [PATCH 0/8] Improve build process > > > > > > > > > > > > On Thu, Jan 29, 2015 at 03:20:03PM +0000, Sergio Gonzalez Monroy > > > > wrote: > > > > > > > This patch series improves the DPDK build system mostly for > > > > > > > shared libraries (and a few nits for static libraries) with the following > > goals: > > > > > > > - Create a library containing core DPDK libraries (librte_eal, > > > > > > > librte_malloc, librte_mempool, librte_mbuf and librte_ring). > > > > > > > The idea of core libraries is to group those libraries that are > > > > > > > always required (and have interdependencies) for any DPDK > > > > application. > > > > > > > - Remove config option to build a combined library. > > > > > > > - For shared libraries, explicitly link against dependant > > > > > > > libraries (adding entries to DT_NEEDED). > > > > > > > - Update app linking flags for static/shared DPDK libs. > > > > > > > > > > > > > > Sergio Gonzalez Monroy (8): > > > > > > > mk: remove combined library and related options > > > > > > > core: create new librte_core > > > > > > > mk: new corelib makefile > > > > > > > lib: update DEPDIRS variable > > > > > > > lib: set LDLIBS for each library > > > > > > > mk: use LDLIBS when linking shared libraries > > > > > > > mk: update LDLIBS for app building > > > > > > > mk: add -lpthread to linuxapp EXECENV_LDLIBS > > > > > > > > > > > > > > config/common_bsdapp | 6 -- > > > > > > > config/common_linuxapp | 6 -- > > > > > > > config/defconfig_ppc_64-power8-linuxapp-gcc | 2 - > > > > > > > lib/Makefile | 1 - > > > > > > > lib/librte_acl/Makefile | 5 +- > > > > > > > lib/librte_cfgfile/Makefile | 4 +- > > > > > > > lib/librte_cmdline/Makefile | 6 +- > > > > > > > lib/librte_core/Makefile | 45 +++++++++++++ > > > > > > > lib/librte_distributor/Makefile | 5 +- > > > > > > > lib/librte_eal/bsdapp/eal/Makefile | 3 +- > > > > > > > lib/librte_eal/linuxapp/eal/Makefile | 3 +- > > > > > > > lib/librte_ether/Makefile | 4 +- > > > > > > > lib/librte_hash/Makefile | 4 +- > > > > > > > lib/librte_ip_frag/Makefile | 6 +- > > > > > > > lib/librte_ivshmem/Makefile | 4 +- > > > > > > > lib/librte_kni/Makefile | 6 +- > > > > > > > lib/librte_kvargs/Makefile | 6 +- > > > > > > > lib/librte_lpm/Makefile | 6 +- > > > > > > > lib/librte_malloc/Makefile | 2 +- > > > > > > > lib/librte_mbuf/Makefile | 2 +- > > > > > > > lib/librte_mempool/Makefile | 2 +- > > > > > > > lib/librte_meter/Makefile | 4 +- > > > > > > > lib/librte_pipeline/Makefile | 3 + > > > > > > > lib/librte_pmd_af_packet/Makefile | 5 +- > > > > > > > lib/librte_pmd_bond/Makefile | 7 +- > > > > > > > lib/librte_pmd_e1000/Makefile | 8 ++- > > > > > > > lib/librte_pmd_enic/Makefile | 8 ++- > > > > > > > lib/librte_pmd_i40e/Makefile | 8 ++- > > > > > > > lib/librte_pmd_ixgbe/Makefile | 8 ++- > > > > > > > lib/librte_pmd_pcap/Makefile | 5 +- > > > > > > > lib/librte_pmd_ring/Makefile | 6 +- > > > > > > > lib/librte_pmd_virtio/Makefile | 7 +- > > > > > > > lib/librte_pmd_vmxnet3/Makefile | 8 ++- > > > > > > > lib/librte_pmd_xenvirt/Makefile | 8 ++- > > > > > > > lib/librte_port/Makefile | 8 +-- > > > > > > > lib/librte_power/Makefile | 4 +- > > > > > > > lib/librte_ring/Makefile | 2 +- > > > > > > > lib/librte_sched/Makefile | 7 +- > > > > > > > lib/librte_table/Makefile | 8 +-- > > > > > > > lib/librte_timer/Makefile | 6 +- > > > > > > > lib/librte_vhost/Makefile | 9 +-- > > > > > > > mk/exec-env/linuxapp/rte.vars.mk | 2 + > > > > > > > mk/rte.app.mk | 53 ++++----------- > > > > > > > mk/rte.corelib.mk | 84 +++++++++++++++++++++++ > > > > > > > mk/rte.lib.mk | 49 +++----------- > > > > > > > mk/rte.sdkbuild.mk | 3 - > > > > > > > mk/rte.sharelib.mk | 101 ---------------------------- > > > > > > > mk/rte.vars.mk | 9 --- > > > > > > > 48 files changed, 276 insertions(+), 282 deletions(-) create > > > > > > > mode > > > > > > > 100644 lib/librte_core/Makefile create mode 100644 > > > > > > > mk/rte.corelib.mk delete mode 100644 mk/rte.sharelib.mk > > > > > > > > > > > > > > -- > > > > > > > 1.9.3 > > > > > > > > > > > > > > > > > > > > Something occured to me thinking about this patch set. I > > > > > > noticed recently that different rules are used to build the > > > > > > shared combined lib from the individual shared objects. The > > > > > > implication here is that linker options specified in individual > > > > > > make files (like the LIBABIVER and EXPORT_MAP options in my ABI > > > > > > versioning script) get ignored, which is bad. Any other file > > > > > > specific linker options (like _LDFLAGS specified in > > > > > > individual library makefiles are getting > > > > dropped for the combined lib. > > > > > > > > > > > > It seems like it would be better if the combined libs were > > > > > > manufactured as linker scripts themselves (textfiles that used > > > > > > linker directives to include individual libraries under the > > > > > > covers (see > > > > /lib64/libc.so for an example). > > > > > > > > > > > > The disadvantage of such an approach are fairly minimal. With > > > > > > such a combined library, you still need to install individual > > > > > > libraries, but for applications that wish to link and run > > > > > > against a single dpdk library will still work just as they > > > > > > currently do, you can link to just a single > > > > library. > > > > > > > > > > > > The advantage is clear however. By following a linker script > > > > > > aproach, objects build as separate libraries are built exactly > > > > > > the same way, using the same rules with the same options. It > > > > > > reduces the dpdk build environment size and complexity, and > > > > > > reduces the opportunity for bugs to creep in from forgetting to > > > > > > add build options to multiple locations. It also provides a > > > > > > more granular approach for grouping files. Creating a dpdk core > > > > > > library becomes a matter of creating a one line linker script > > > > > > named libdpdk_core.so, rather > > > > than re- arraning sections of the build system. > > > > > > > > > > > > Thoughts? > > > > > > Neil > > > > > > > > > > > Hi Neil, > > > > > > > > > > I think that is a very interesting approach. > > > > > I have tried to do something similar in this patch by removing > > > > > rte.sharelib.mk and just having rte.lib.mk to do the linking, > > > > > leaving as you suggest a single file to modify anything related to building > > libs. > > > > > > > > > > I do think however that your proposal is an improvement over the > > > > > current > > > > patch. > > > > > > > > > > So basically we want: > > > > > - get rid of rte.corelib.mk > > > > > - generate librte_core.so linker script grouping core libs > > > > > - we do not modify DEPDIR variables > > > > > - when setting LDLIBS to each lib, we do specify -lrte_core, right? > > > > > > > > > Exactly, and librte_core.so is really just a text file containing > > > > the following line > > > > : > > > > INPUT(-lrte_malloc -lrte_mbuf -lrte_eal ....) > > > > > > > > Adding in whatever libraries you want librte_core to consist of. > > > > Truthfully, you could almost get rid of the COMBINE_LIBS option > > > > entirely, and just create this file statically if you wanted to (not > > > > sure thats the best approach, but its definately do-able). > > > > > > > Hi Neil, > > > > > > Actually, the first patch series does get rid of COMBINE_LIBS entirely. > > > > > Sorry, I didn't mean to imply your patch wasn't, just re-iterating that the > > option is not needed using the alternate method we're discussing, but I really > > wasn't very clear on that. > > > > > So as I was looking into this, by using this approach we do not resolve the > > interdependencies issue of the core libraries. > > > We would effectively leave all core libraries (or at least EAL) without proper > > DT_NEEDED entries. > > > > > > Thoughts? > > > > > You're correct, or at least what you assert is possible, depending on the > > implementation. Adding DT_NEEDED entries is something of an orthogonal > > problem (though your current implementation I think handles it well). You > > could specify linker directives when building each library so that each DSO > > contains the proper DT_NEEDED entries (using -l and --no-as-needed). > > using a linker script approach doesn't preclude you from doing that, though > > its not strictly speaking necessecary. When you write the linker script, you > > implicitly specify the link dependencies by the order in whcih you list the > > inferior libraries in the scripts INPUT line. It doesn't give you the DT_NEEDED > > entries, but from an application build/run standpoint, it won't matter, > > because the libraries will be linked/loaded in the order specified. You can still > > do the --no-as-needed method though if you like for safety on the part of > > those using libraries independently. > > So would it be reasonable to add DT_NEEDED entries to all DPDK libraries but EAL? > If I understood what you were saying right, we could enforce the 'dependency' in the > linker script with something like this: > $ cat librte_eal.so > INPUT( librte_eal.so.1 -lrte_mempool -lrte_malloc) > We could have such linker script for librte_eal.so instead of the soft link once > versioning is in place. > Correct. > Things that would be missing versus the proposed patch: > - As I have mention in previous post, ldd info for EAL library would not reflect > its dependency to other DPDK libs. librte_eal.so would no show those dependencies, as far as I know (though I haven't explicitly checked). The subordunate libraries included in the input line, may or may not show dependencies among themselves, depending on your build setup (and the use of --no-as-needed and -l when linking the individual .so libraries. > - I was enforcing resolving all references when building the libraries (-z defs), so > we either remove it altogether or skip eal. I think thats correct, yes. > - All apps would show DT_NEEDED entries for a set of DPDK libraries that > in most cases are required (eal, mempool, malloc, mbuf, ring VS dpdk_core) > I think apps linked to libdpdk_core would have DT_NEEDED entries for libdpdk_core, not the subordonate libraries (though check me on that to be sure). > I think that the linker script approach is reasonable if we prefer to go that way > instead of creating a core library. > I think it would make sense from a build environment point of view, in that it allows library specific flags to be incorporated properly. I think the only downside is that the individual libraries still need to be carried around (though they can be ignored from an application build/run standpoint). You're question should probably be asked of people using COMBINED_LIBS currently to make sure that meets their needs, though I think it will. Neil > Regards, > Sergio > > > Neil > > > > > Regards, > > > Sergio > > > > > > > Regards > > > > Neil > > > > > > > >