[ofa-general] Re: [PATCH 20/21] RDS: Kconfig and Makefile

[Andy Grover]

Roland Dreier wrote:
>  > +obj-$(CONFIG_INFINIBAND_ISER)		+= ulp/rds/
> 
> Typo for ..._RDS

Whups. :)

>  > +config INFINIBAND_RDS_DEBUG
>  > +        bool "Debugging messages"
>  > +	depends on INFINIBAND_RDS
>  > +        default n
> 
> No way to enable this?  Disabled by default?
> 
> You really want debugging messages to be built by default and controlled
> at runtime ... otherwise debugging end-user installations is a pain
> (they just install what the distro gives them, and it's very hard for
> them to rebuild just to enable debugging).

So the solution is just to base debug message output on a variable,
instead of a config option? RDS actually does do this a little already,
so converting totally isn't hard. I hadn't seen mention this was
preferable -- indeed, tons of drivers and subsystems have options for
compile-time debug statements, should these be converted?

>  > +ib_rds-y :=	af_rds.o bind.o cong.o connection.o info.o message.o   \
>  > +			recv.o send.o stats.o sysctl.o threads.o transport.o \
>  > +			loop.o page.o rdma.o
>  > +
>  > +ib_rds-y += ib.o ib_cm.o ib_recv.o ib_ring.o ib_send.o ib_stats.o \
>  > +			ib_sysctl.o ib_rdma.o
> 
> a very strange way to write an assignment statement...

RDS is implemented as a core sockets layer and then a transport layer.
IB is currently the only transport so I thought it made sense to just
compile them together, but once there are >1 then RDS's IB support could
be broken out into its own module.

Regards -- Andy