[dpdk-dev] Issue with rte_compat versioning macros

Gajdzica, MaciejX T maciejx.t.gajdzica at intel.com
Tue Jun 23 13:28:29 CEST 2015


> -----Original Message-----
> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Gajdzica, MaciejX T
> Sent: Tuesday, June 23, 2015 9:46 AM
> To: Neil Horman
> Cc: dev at dpdk.org
> Subject: Re: [dpdk-dev] Issue with rte_compat versioning macros
> 
> 
> 
> > -----Original Message-----
> > From: Neil Horman [mailto:nhorman at tuxdriver.com]
> > Sent: Monday, June 22, 2015 6:52 PM
> > To: Gajdzica, MaciejX T
> > Cc: Thomas Monjalon; dev at dpdk.org
> > Subject: Re: [dpdk-dev] Issue with rte_compat versioning macros
> >
> > On Mon, Jun 22, 2015 at 03:32:01PM +0000, Gajdzica, MaciejX T wrote:
> > > Hi Neil
> > >
> > > > -----Original Message-----
> > > > From: Neil Horman [mailto:nhorman at tuxdriver.com]
> > > > Sent: Friday, June 19, 2015 5:32 PM
> > > > To: Gajdzica, MaciejX T
> > > > Cc: Thomas Monjalon; dev at dpdk.org
> > > > Subject: Re: [dpdk-dev] Issue with rte_compat versioning macros
> > > >
> > > > On Fri, Jun 19, 2015 at 03:04:17PM +0000, Gajdzica, MaciejX T wrote:
> > > > > > -----Original Message-----
> > > > > > From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> > > > > > Sent: Friday, June 19, 2015 4:48 PM
> > > > > > To: Gajdzica, MaciejX T
> > > > > > Cc: dev at dpdk.org; nhorman at tuxdriver.com
> > > > > > Subject: Re: [dpdk-dev] Issue with rte_compat versioning
> > > > > > macros
> > > > > >
> > > > > > 2015-06-19 14:38, Gajdzica, MaciejX T:
> > > > > > > There is an issue with macros in rte_compat.h. For shared
> > > > > > > library case, macro BIND_DEFAULT_SYMBOL takes three
> > > > > > > arguments and for other case it takes only two arguments.
> > > > > > > Also letters for macro variable names are
> > > > > > not consistent in these two cases.
> > > > > >
> > > > > > Yes, and your patch fix it:
> > > > > > 	http://dpdk.org/dev/patchwork/patch/5475/
> > > > > > But it is part of a series which is not accepted yet.
> > > > > >
> > > > > > It would be faster merged if you send it as a standalone patch.
> > > > > > Thanks
> > > > >
> > > > > But simple solution with adding third argument to static library
> > > > > case doesn't
> > > > work. Comment in rte_compat.h file describes steps needed to add
> > > > new version of the function and it says:
> > > > >
> > > > > * 2) rename the existing function int foo(char *string) to
> > > > > * 	int __vsym foo_v20(char *string)
> > > > > *
> > > > > * 3) Add this macro immediately below the function
> > > > > * 	VERSION_SYMBOL(foo, _v20, 2.0);
> > > > > *
> > > > > * 4) Implement a new version of foo.
> > > > > * 	char foo(int value, int otherval) { ...}
> > > > > *
> > > > > * 5) Mark the newest version as the default version
> > > > > * 	BIND_DEFAULT_SYMBOL(foo, 2.1);
> > > > >
> > > > > So probably BIND_DEFAULT_SYMBOL macro for shared library case
> > > > > needs to
> > > > be modified to have two arguments.
> > > > > I'm not familiar with that symver syntax so I need some help. It
> > > > > would be
> > > > better when original author say how it should look like.
> > > > >
> > > > > Best Regards
> > > > > Maciek
> > > > >
> > > > >
> > > >
> > > > No adding the third parameter will work just fine.  The
> > > > documentation needs to be updated as well to reflect the 3rd argument:
> > > > * 5) Mark the newest version as the default version
> > > > *      BIND_DEFAULT_SYMBOL(foo, foo, 2.1);
> > > >
> > > >
> > > > Neil
> > > >
> > >
> > > For me versioning still don't work, even though I try to do it as
> > > you say. I had
> > function:
> > >
> > > int
> > > rte_cfgfile_section_entries(struct rte_cfgfile *cfg,
> > > 	const char *sectionname,
> > > 	struct rte_cfgfile_entry *entries,
> > > 	int max_entries)
> > > {
> > > [...]
> > > }
> > >
> > > And I wanted to add new version of it. So I marked it as _v20, added
> > > version macro, added new implementation and default symbol macro:
> > >
> > > int __vsym
> > > rte_cfgfile_section_entries_v20(struct rte_cfgfile *cfg,
> > > 	const char *sectionname,
> > > 	struct rte_cfgfile_entry *entries,
> > > 	int max_entries)
> > > {
> > > [...]
> > > }
> > > VERSION_SYMBOL(rte_cfgfile_section_entries, _v20, 2.0);
> > >
> > > int __vsym
> > > rte_cfgfile_section_entries(struct rte_cfgfile *cfg,
> > > 		const char *sectionname, struct rte_cfgfile_entry2 **entries,
> > > 		int max_entries)
> > > {
> > > [...]
> > > }
> > > BIND_DEFAULT_SYMBOL(rte_cfgfile_section_entries,
> > > rte_cfgfile_section_entries, 2.1);
> > >
> > > I edited map file so it looks like this:
> > >
> > > DPDK_2.0 {
> > > 	global:
> > >
> > > [...]
> > > 	rte_cfgfile_section_entries;
> > > [...]
> > >
> > > 	local: *;
> > > };
> > >
> > > DPDK_2.1 {
> > > 	global:
> > >
> > > 	rte_cfgfile_section_entries;
> > >
> > > 	local: *;
> > > };
> > >
> > > Then I try to build dpdk and qos_sched example which uses this
> > > function. When I build dpdk as static library, everything works fine.
> > > When I build dpdk as shared library, compilation of example returns
> > > error, that
> > it doesn't see rte_cfgfile_section_entries function.
> > >
> > Soooooo.....What are your errors?  Complaining that it doesn't work
> > doesn't help anyone if you don't tell us whats wrong.
> >
> >
> > > Maybe I do something wrong, but it's obvious that better
> > > documentation is needed. In current dpdk master code, versioning is
> > > not used. There are only
> > 2 patchsets that tries to use it. This one and:
> > > http://dpdk.org/ml/archives/dev/2015-May/018169.html
> > > That second one probably uses it wrong, because second argument of
> > BIND_DEFAULT_SYMBOL is _v21.
> > > And you said, it should be the same as function name. I don't need
> > > versioning for now as we decided That rte_cfgfile modifications
> > > should be done in next release. But knowing that versioning macros
> > > are well
> > documented and work properly could encourage more people to use it.
> > >
> > Clearly you did something wrong, because its not building.  I agree
> > that more documentation is needed, but unless you tell me whats
> > broken, enhancing the documentation isn't going to be much more than a
> > rock fetch problem Neil
> 
> Here is my qos_sched example build log:
> 
> gcc -m64 -pthread  -march=native -DRTE_MACHINE_CPUFLAG_SSE -
> DRTE_MACHINE_CPUFLAG_SSE2 -DRTE_MACHINE_CPUFLAG_SSE3 -
> DRTE_MACHINE_CPUFLAG_SSSE3 -DRTE_MACHINE_CPUFLAG_SSE4_1 -
> DRTE_MACHINE_CPUFLAG_SSE4_2 -DRTE_MACHINE_CPUFLAG_AES -
> DRTE_MACHINE_CPUFLAG_PCLMULQDQ -DRTE_MACHINE_CPUFLAG_AVX -
> DRTE_MACHINE_CPUFLAG_RDRAND -DRTE_MACHINE_CPUFLAG_FSGSBASE -
> DRTE_MACHINE_CPUFLAG_F16C -
> DRTE_COMPILE_TIME_CPUFLAGS=RTE_CPUFLAG_SSE,RTE_CPUFLAG_SSE2,RTE
> _CPUFLAG_SSE3,RTE_CPUFLAG_SSSE3,RTE_CPUFLAG_SSE4_1,RTE_CPUFLAG_SS
> E4_2,RTE_CPUFLAG_AES,RTE_CPUFLAG_PCLMULQDQ,RTE_CPUFLAG_AVX,RTE_
> CPUFLAG_RDRAND,RTE_CPUFLAG_FSGSBASE,RTE_CPUFLAG_F16C  -
> I/mnt/shared/mtgajdzx/workspace/cmd_patch/examples/qos_sched/build/inclu
> de -I/mnt/shared/mtgajdzx/workspace/cmd_patch/x86_64-native-linuxapp-
> gcc/include -include /mnt/shared/mtgajdzx/workspace/cmd_patch/x86_64-
> native-linuxapp-gcc/include/rte_config.h -O3 -W -Wall -Werror -Wstrict-
> prototypes -Wmissing-prototypes -Wmissing-declarations -Wold-style-definition
> -Wpointer-arith -Wcast-align -Wnested-externs -Wcast-qual -Wformat-
> nonliteral -Wformat-security -Wundef -Wwrite-strings -Wno-missing-field-
> initializers -Wno-uninitialized  -Wl,-Map=qos_sched.map,--cref -o qos_sched
> main.o args.o init.o app_thread.o cfg_file.o cmdline.o stats.o -Wl,--no-as-
> needed -Wl,-export-dynamic -
> L/mnt/shared/mtgajdzx/workspace/cmd_patch/examples/qos_sched/build/lib -
> L/mnt/shared/mtgajdzx/workspace/cmd_patch/x86_64-native-linuxapp-gcc/lib
> -L/mnt/shared/mtgajdzx/workspace/cmd_patch/x86_64-native-linuxapp-gcc/lib
> -Wl,--whole-archive -Wl,-lrte_distributor -Wl,-lrte_reorder -Wl,-lrte_kni -Wl,-
> lrte_pipeline -Wl,-lrte_table -Wl,-lrte_port -Wl,-lrte_timer -Wl,-lrte_hash -Wl,-
> lrte_jobstats -Wl,-lrte_lpm -Wl,-lrte_power -Wl,-lrte_acl -Wl,-lrte_meter -Wl,-
> lrte_sched -Wl,-lm -Wl,-lrt -Wl,--start-group -Wl,-lrte_kvargs -Wl,-lrte_mbuf -
> Wl,-lrte_ip_frag -Wl,-lethdev -Wl,-lrte_malloc -Wl,-lrte_mempool -Wl,-lrte_ring
> -Wl,-lrte_eal -Wl,-lrte_cmdline -Wl,-lrte_cfgfile -Wl,-lrte_pmd_bond -Wl,-lrt -
> Wl,-lm -Wl,-ldl -Wl,--end-group -Wl,--no-whole-archive
> cfg_file.o: In function `cfg_load_subport':
> /mnt/shared/mtgajdzx/workspace/cmd_patch/examples/qos_sched/cfg_file.c:4
> 85: undefined reference to `rte_cfgfile_section_entries'
> collect2: ld returned 1 exit status
> make[1]: *** [qos_sched] Error 1
> make: *** [all] Error 2
> 
> Linker couldn't find versioned function when dpdk is build as shared lib.

After further investigation I came up with solution working for both shared and dynamic library cases.
Old function should have added suffix _v20. New function should have suffix _v21 and bind default macro:

BIND_DEFAULT_SYMBOL(rte_cfgfile_section_entries, _v21, 2.1);

In rte_compat.h versioning macros should have the same form for both shared and static libraries. I removed
empty macros for static case and ifdef. I will send my changes as a patch.

Best Regards
Maciek


More information about the dev mailing list