Re: [RFC][PATCH 2/3] x86/uaccess: Implement unsafe_try_cmpxchg_user()

[Sean Christopherson <seanjc@google.com>]

On Tue, Dec 14, 2021, Peter Zijlstra wrote:
> Do try_cmpxchg() loops on userspace addresses.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
> @@ -501,6 +543,21 @@ do {										\
>  } while (0)
>  #endif // CONFIG_CC_HAS_ASM_GOTO_OUTPUT
>  
> +extern void __try_cmpxchg_user_wrong_size(void);
> +
> +#define unsafe_try_cmpxchg_user(_ptr, _oldp, _nval, _label) ({		\
> +	__typeof__(*(_ptr)) __ret;					\
> +	switch (sizeof(__ret)) {					\
> +	case 4:	__ret = __try_cmpxchg_user_asm("l", (_ptr), (_oldp),	\
> +					       (_nval), _label);	\
> +		break;							\
> +	case 8:	__ret = __try_cmpxchg_user_asm("q", (_ptr), (_oldp),	\
> +					       (_nval), _label);	\
> +		break;							\

Can we add support for 1-byte and 2-byte cmpxchg, and for using cmpxchg8b to handle
8-byte operations in 32-bit mode?  Support for all the flavors (except 16-byte)
would allow KVM to use this in an emulator path that currently kmaps the target.
I'd be more than happy to help test the result.

Thanks!

> +	default: __try_cmpxchg_user_wrong_size();			\
> +	}								\
> +	__ret;						})
> +
>  /*
>   * We want the unsafe accessors to always be inlined and use
>   * the error labels - thus the macro games.
> 
>