Re: [PATCH] x86/boot/arch/variable: I don't think (void *) Pointers need to be cast

[Ingo Molnar]

* Li kunyu <kunyu@xxxxxxxxxxxx> wrote:

> I first observe (void *) type coercion and non coercion through assembly
> language. It seems that there is no difference.
> Then I output the assigned information through the print function and
> found that the pointer that is not coerced is directly assigned when
> executing the print function (opcode a1), while the coerced pointer
> needs to execute the assembly instruction xlat (opcode d7), which seems
> to be more efficient without coerced conversion.
> At present, I just started to try to analyze this part of knowledge
> (machine code), please forgive me if the analysis is wrong.
> 
> Signed-off-by: Li kunyu <kunyu@xxxxxxxxxxxx>
> ---
>  arch/x86/boot/bitops.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/x86/boot/bitops.h b/arch/x86/boot/bitops.h
> index 02e1dea11d94..8518ae214c9b 100644
> --- a/arch/x86/boot/bitops.h
> +++ b/arch/x86/boot/bitops.h
> @@ -19,13 +19,13 @@
>  
>  static inline bool constant_test_bit(int nr, const void *addr)
>  {
> -	const u32 *p = (const u32 *)addr;
> +	const u32 *p = addr;
>  	return ((1UL << (nr & 31)) & (p[nr >> 5])) != 0;
>  }
>  static inline bool variable_test_bit(int nr, const void *addr)
>  {
>  	bool v;
> -	const u32 *p = (const u32 *)addr;
> +	const u32 *p = addr;
>  
>  	asm("btl %2,%1" CC_SET(c) : CC_OUT(c) (v) : "m" (*p), "Ir" (nr));
>  	return v;

It's true that the forced-type casting of 'addr' is unnecessary in the 
cases above, I'm not sure how the kernel would end up with an XLAT 
instruction being generated in that sequence.

But your patch is a good cleanup in its own right - I've applied the patch 
below to tip:x86/cleanups, with a different changelog.

Thanks,

	Ingo

===================>
From: Li kunyu <kunyu@xxxxxxxxxxxx>
Date: Mon, 25 Jul 2022 12:23:58 +0800
Subject: [PATCH] x86/boot: Remove superfluous type casting from arch/x86/boot/bitops.h

'const void *' will auto-type-convert to just about any other const pointer type,
no need to force it.

[ mingo: Rewrote the changelog. ]

Signed-off-by: Li kunyu <kunyu@xxxxxxxxxxxx>
Signed-off-by: Ingo Molnar <mingo@xxxxxxxxxx>
Link: https://lore.kernel.org/r/20220725042358.3377-1-kunyu@xxxxxxxxxxxx
---
 arch/x86/boot/bitops.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/boot/bitops.h b/arch/x86/boot/bitops.h
index 02e1dea11d94..8518ae214c9b 100644
--- a/arch/x86/boot/bitops.h
+++ b/arch/x86/boot/bitops.h
@@ -19,13 +19,13 @@
 
 static inline bool constant_test_bit(int nr, const void *addr)
 {
-	const u32 *p = (const u32 *)addr;
+	const u32 *p = addr;
 	return ((1UL << (nr & 31)) & (p[nr >> 5])) != 0;
 }
 static inline bool variable_test_bit(int nr, const void *addr)
 {
 	bool v;
-	const u32 *p = (const u32 *)addr;
+	const u32 *p = addr;
 
 	asm("btl %2,%1" CC_SET(c) : CC_OUT(c) (v) : "m" (*p), "Ir" (nr));
 	return v;