Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <0820010f-e9dc-779d-7924-49c7df446bce@linux.ibm.com>
Date:   Wed, 21 Jun 2023 14:28:18 +0200
MIME-Version: 1.0
Subject: Re: [PATCH] gcov: add prototypes for internal functions
Content-Language: en-US
To:     Arnd Bergmann <arnd@kernel.org>
Cc:     Arnd Bergmann <arnd@arndb.de>, linux-kernel@vger.kernel.org
References: <20230517125011.930274-1-arnd@kernel.org>
From:   Peter Oberparleiter <oberpar@linux.ibm.com>
In-Reply-To: <20230517125011.930274-1-arnd@kernel.org>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 17.05.2023 14:50, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> gcov uses global functions that are called from generated code,
> but these have no prototype in a header, which causes a W=1
> build warning:

Sorry for the late reply. I can confirm that the problem exists, both
with GCC and Clang compiles. I see some issues with your fix though:

1. It contains extraneous prototypes that are duplicates of existing
prototypes in gcov.h (everything that doesn't start with __gcov or llvm)

2. Adding the list of prototypes, though not actually needed by kernel
source code, will require updating gcov.h if new calls from generated
code are introduced - this is extra work I'd like to avoid if possible

Because this situation is special, I would rather prefer to silence
these warnings by removing the associated compiler flags
(-Wmissing-prototypes and I think also -Wmissing-declarations) for
gcc_base.c and clang.c using CFLAGS_REMOVE.

If you agree, do you want to provide a v2 along these lines?

[...]

> --- a/kernel/gcov/gcov.h
> +++ b/kernel/gcov/gcov.h
> @@ -82,4 +82,34 @@ extern const struct gcov_link gcov_link[];
>  extern int gcov_events_enabled;
>  extern struct mutex gcov_lock;
>  

FYI - these are defined in gcc_base.c and called only by GCC generated code:

> +void __gcov_init(struct gcov_info *info);
> +void __gcov_flush(void);
> +void __gcov_merge_add(gcov_type *counters, unsigned int n_counters);
> +void __gcov_merge_single(gcov_type *counters, unsigned int n_counters);
> +void __gcov_merge_delta(gcov_type *counters, unsigned int n_counters);
> +void __gcov_merge_ior(gcov_type *counters, unsigned int n_counters);
> +void __gcov_merge_time_profile(gcov_type *counters, unsigned int n_counters);
> +void __gcov_merge_icall_topn(gcov_type *counters, unsigned int n_counters);
> +void __gcov_exit(void);

These are defined in both gcc_4_7.c and clang.c and used by other code
in kernel/gcov/*:

> +size_t convert_to_gcda(char *buffer, struct gcov_info *info);
> +void gcov_info_add(struct gcov_info *dst, struct gcov_info *src);
> +struct gcov_info *gcov_info_dup(struct gcov_info *info);
> +const char *gcov_info_filename(struct gcov_info *info);
> +void gcov_info_free(struct gcov_info *info);
> +int gcov_info_is_compatible(struct gcov_info *info1, struct gcov_info *info2);
> +void gcov_info_link(struct gcov_info *info);
> +struct gcov_info *gcov_info_next(struct gcov_info *info);
> +void gcov_info_reset(struct gcov_info *info);
> +void gcov_info_unlink(struct gcov_info *prev, struct gcov_info *info);
> +unsigned int gcov_info_version(struct gcov_info *info);
> +bool gcov_info_within_module(struct gcov_info *info, struct module *mod);

And these are defined in clang.c and called only by Clang generated code:

> +void llvm_gcda_emit_arcs(u32 num_counters, u64 *counters);
> +void llvm_gcda_emit_function(u32 ident, u32 func_checksum, u32 cfg_checksum);
> +void llvm_gcda_end_file(void);
> +void llvm_gcda_start_file(const char *orig_filename, u32 version, u32 checksum);
> +void llvm_gcda_summary_info(void);
> +typedef void (*llvm_gcov_callback)(void);
> +void llvm_gcov_init(llvm_gcov_callback writeout, llvm_gcov_callback flush);

-- 
Peter Oberparleiter
Linux on IBM Z Development - IBM Germany R&D