Re: [RFC PATCH 3/4] gcov: compile specific gcov implementation based on gcc version

[Frantisek Hrbata <fhrbata@redhat.com>]

On Fri, Aug 23, 2013 at 05:21:12PM +0200, Peter Oberparleiter wrote:
> On 23.08.2013 17:15, Peter Oberparleiter wrote:
> > On 23.08.2013 10:39, Frantisek Hrbata wrote:
> >> Compile the correct gcov implementation file for a specific gcc version. In
> >> the future, if another file is added, the conditions will need to be somehow
> >> adjusted to if-elif-else case, but at this point the simple cc-ifversion should
> >> be enough.
> 
> As promised, I'm also adding the patch that makes the format-specific part
> of gcov-kernel a loadable kernel module:
> 
> ---
> kernel: gcov: make format-specific code loadable
> 
> Turn the format-specific part of gcov-kernel into a loadable kernel
> module. This enables the use of gcov-kernel with kernel modules
> that were compiled with a version of GCC that produces a different
> gcov format when compared to the version of GCC that was used to
> compile the kernel.

If I understand it correctly, this would mean that you will be able to use only
one implementation of gcov format at the time. Meaning you will be able to get
coverage data for module, but not for kernel if it was compiled with different
gcc(gcda format). This is probably ok if you work only on your module, but I'm
not sure this is generally the right approach. In this case I would probably
rather see some support for more gcov formats at the same time(e.g. set of
callback operations per gcov version). Again I'm probably missing something, but
I still cannot see reason why to add such feature. If you want gcov support just
compile your kernel and modules with the same gcc version(gcda format). But if
this is really needed maybe it would be better to consider some parallel support
for more gcov formats based on the gcov_info version.

Would it be possible to add support for the modified gcc 4.7 gcov format and
deal with this later? I can incorporate your changes: iter to use buffer,
.init_array for modules and possibility to explicitly select the gcda format.
In this case we will have at least the basic support in kernel. This is just me
thinking out loud.

Many thanks Peter!

> 
> Signed-off-by: Peter Oberparleiter <oberpar@linux.vnet.ibm.com>
> ---
>  kernel/gcov/Kconfig  |   19 +++++++++++++++++--
>  kernel/gcov/Makefile |    7 ++++---
>  2 files changed, 21 insertions(+), 5 deletions(-)
> 
> --- a/kernel/gcov/Kconfig
> +++ b/kernel/gcov/Kconfig
> @@ -29,8 +29,23 @@ config GCOV_KERNEL
>  	and:
>  	        GCOV_PROFILE := n
> 
> -	Note that the debugfs filesystem has to be mounted to access
> -	profiling data.
> +	Note that GCOV_KERNEL_FS has to specified as well and the debugfs
> +	filesystem has to be mounted to access profiling data.
> +
> +config GCOV_KERNEL_FS
> +	tristate "Provide gcov data files in debugfs"
> +	depends on GCOV_KERNEL
> +	default y
> +	---help---
> +	Make profiling data available in debugfs at /sys/kernel/debug/gcov.
> +
> +	Say M if you want to enable collecting coverage data for kernel modules
> +	which are compiled using a gcc version different from the one that
> +	was used to compile the kernel. In that case, re-compile the gcov
> +	kernel module with corresponding format support and load that module
> +	instead.
> +
> +	If unsure, say Y.
> 
>  config GCOV_PROFILE_ALL
>  	bool "Profile entire Kernel"
> --- a/kernel/gcov/Makefile
> +++ b/kernel/gcov/Makefile
> @@ -13,10 +13,11 @@ else
>    cc-ver := $(call cc-version)
>  endif
> 
> -obj-$(CONFIG_GCOV_KERNEL) := base.o fs.o
> +obj-$(CONFIG_GCOV_KERNEL) += base.o
> +obj-$(CONFIG_GCOV_KERNEL_FS) += gcov.o
> 
>  ifeq ($(call if-lt, $(cc-ver), 0407),1)
> -  obj-$(CONFIG_GCOV_KERNEL) += gcc_3_4.o
> +  gcov-objs += fs.o gcc_3_4.o
>  else
> -  obj-$(CONFIG_GCOV_KERNEL) += gcc_4_7.o
> +  gcov-objs += fs.o gcc_4_7.o
>  endif
> 
> -- 
> Peter Oberparleiter
> Linux on System z Development - IBM Germany
> 

-- 
Frantisek Hrbata