Re: [dpdk-dev] spinlock: Move constructor function out of header file

[Thomas Monjalon <thomas.monjalon@6wind.com>]

I will apply it with trivial changes suggested by Jan and
the small needed changes that I describe below:

2016-07-14 20:03, Jan Viktorin:
> On Thu, 14 Jul 2016 15:27:29 +0200
> damarion@cisco.com wrote:
> > --- /dev/null
> > +++ b/lib/librte_eal/common/arch/x86/rte_spinlock.c
[...]
> > +uint8_t rte_rtm_supported; /* cache the flag to avoid the overhead
> > +			      of the rte_cpu_get_flag_enabled function */

This variable must be exported in the .map file.

> > --- a/lib/librte_eal/linuxapp/eal/Makefile
> > +++ b/lib/librte_eal/linuxapp/eal/Makefile
> > @@ -106,6 +106,7 @@ SRCS-$(CONFIG_RTE_EXEC_ENV_LINUXAPP) += rte_keepalive.c
> >  
> >  # from arch dir
> >  SRCS-$(CONFIG_RTE_EXEC_ENV_LINUXAPP) += rte_cpuflags.c
> > +SRCS-$(CONFIG_RTE_EXEC_ENV_LINUXAPP) += rte_spinlock.c
> 
> This is not good, you provide rte_spinlock.c only for x86. Building
> for any other arch would fail to find this file.

This change do the trick:

-SRCS-$(CONFIG_RTE_EXEC_ENV_LINUXAPP) += rte_spinlock.c
+SRCS-$(CONFIG_RTE_ARCH_X86) += rte_spinlock.c

(Note that CONFIG_RTE_EXEC_ENV_LINUXAPP check is not needed inside linuxapp EAL)

> Moreover, the bsdapp/eal/Makefile should reflect this situation as
> well.

Yes