Re: [PATCH] arm/mach-hisi: Use BUG_ON instead of if condition followed by BUG

[Robin Murphy <robin.murphy@arm.com>]

On 2021-04-23 09:14, zhouchuangao wrote:
> BUG_ON uses unlikely in if(). Through disassembly, we can see that
> brk #0x800 is compiled to the end of the function.
> As you can see below:
>      ......
>      ffffff8008660bec:   d65f03c0    ret
>      ffffff8008660bf0:   d4210000    brk #0x800
> 
> Usually, the condition in if () is not satisfied. For the
> multi-stage pipeline, we do not need to perform fetch decode
> and excute operation on brk instruction.

32-bit Arm does not have "ret" and "brk" instructions, and either way 
the relevant BUG() instruction(s) aren't executed unless the condition 
is met, so this really makes very little sense.

> In my opinion, this can improve the efficiency of the
> multi-stage pipeline.

It has very little to do with the pipeline - modern cores are 
considerably more sophisticated than the 3-stage Acorn RISC Machine of 
1985, and are not usually limited by frontend throughput. The point of 
unlikely() is to avoid having a normally-taken forward branch to skip 
over in-line code, and instead make sure the only thing in the normal 
execution path is a normally-not-taken branch to handle the condition 
out-of-line. Yes, the impact of branches - and thus why it can be 
desirable to avoid them - is indeed *related* to pipelining, but that's 
rather tangential.

Even then, it's only worth considering things at this level in 
frequently-executed and/or performance-critical code. Saving a couple of 
CPU cycles in something that is effectively a one-time operation is 
utterly immaterial.

The realistic justification for these patches is that that BUG_ON() 
exists for implementing conditional BUG()s, so we may as well use it if 
it makes the source code more readable.

> Signed-off-by: zhouchuangao <zhouchuangao@vivo.com>
> ---
>   arch/arm/mach-hisi/hotplug.c  | 3 +--
>   arch/arm/mach-hisi/platmcpm.c | 4 ++--
>   2 files changed, 3 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/arm/mach-hisi/hotplug.c b/arch/arm/mach-hisi/hotplug.c
> index c517941..b9ced60 100644
> --- a/arch/arm/mach-hisi/hotplug.c
> +++ b/arch/arm/mach-hisi/hotplug.c
> @@ -193,8 +193,7 @@ void hix5hd2_set_cpu(int cpu, bool enable)
>   	u32 val = 0;
>   
>   	if (!ctrl_base)
> -		if (!hix5hd2_hotplug_init())
> -			BUG();
> +		BUG_ON(!hix5hd2_hotplug_init());

Whatever tool you're using to detect these patterns, consider improving 
it, or at least giving a bit more thought to the results beyond blindly 
applying one single rule - "if(x) BUG_ON(y);" arguably makes even less 
sense since it's now neither one thing nor the other.

Robin.

>   	if (enable) {
>   		/* power on cpu1 */
> diff --git a/arch/arm/mach-hisi/platmcpm.c b/arch/arm/mach-hisi/platmcpm.c
> index 96a4840..6c90039 100644
> --- a/arch/arm/mach-hisi/platmcpm.c
> +++ b/arch/arm/mach-hisi/platmcpm.c
> @@ -82,8 +82,8 @@ static void hip04_set_snoop_filter(unsigned int cluster, unsigned int on)
>   {
>   	unsigned long data;
>   
> -	if (!fabric)
> -		BUG();
> +	BUG_ON(!fabric);
> +
>   	data = readl_relaxed(fabric + FAB_SF_MODE);
>   	if (on)
>   		data |= 1 << cluster;
>