The barrier_nospec() in 32-bit copy_from_user() is slow. Instead use pointer masking to force the user pointer to all 1's if a previous access_ok() mispredicted true for an invalid address.
Signed-off-by: Josh Poimboeuf <jpoim...@kernel.org> --- arch/x86/include/asm/uaccess.h | 34 +++++++++++++++++++++++++++++++ arch/x86/include/asm/uaccess_32.h | 2 +- arch/x86/include/asm/uaccess_64.h | 29 +------------------------- 3 files changed, 36 insertions(+), 29 deletions(-) diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h index 3a7755c1a441..e7ac97d42bc2 100644 --- a/arch/x86/include/asm/uaccess.h +++ b/arch/x86/include/asm/uaccess.h @@ -15,6 +15,40 @@ #include <asm/smap.h> #include <asm/extable.h> #include <asm/tlbflush.h> +#include <asm/runtime-const.h> + +#ifdef CONFIG_X86_64 +/* + * Virtual variable: there's no actual backing store for this, + * it can purely be used as 'runtime_const_ptr(USER_PTR_MAX)' + */ +extern unsigned long USER_PTR_MAX; +# define USER_PTR_MAX_CONST runtime_const_ptr(USER_PTR_MAX) +#else +# define USER_PTR_MAX_CONST TASK_SIZE_MAX-1 +#endif + +/* + * Masking the user address is an alternative to a conditional + * user_access_begin that can avoid the fencing. This only works + * for dense accesses starting at the address. + */ +static inline void __user *mask_user_address(const void __user *ptr) +{ + unsigned long mask; + asm("cmp %1,%0\n\t" + "sbb %0,%0" + :"=r" (mask) + :"r" (ptr), + "0" (USER_PTR_MAX_CONST)); + return (__force void __user *)(mask | (__force unsigned long)ptr); +} + +#define masked_user_access_begin(x) ({ \ + __auto_type __masked_ptr = (x); \ + __masked_ptr = mask_user_address(__masked_ptr); \ + __uaccess_begin(); __masked_ptr; }) + #ifdef CONFIG_X86_32 # include <asm/uaccess_32.h> diff --git a/arch/x86/include/asm/uaccess_32.h b/arch/x86/include/asm/uaccess_32.h index 8393ba104b2c..6ec2d73f8bba 100644 --- a/arch/x86/include/asm/uaccess_32.h +++ b/arch/x86/include/asm/uaccess_32.h @@ -23,7 +23,7 @@ raw_copy_to_user(void __user *to, const void *from, unsigned long n) static __always_inline unsigned long raw_copy_from_user(void *to, const void __user *from, unsigned long n) { - barrier_nospec(); + from = mask_user_address(from); return __copy_user_ll(to, (__force const void *)from, n); } diff --git a/arch/x86/include/asm/uaccess_64.h b/arch/x86/include/asm/uaccess_64.h index 7ce84090f0ec..dfb78154ac26 100644 --- a/arch/x86/include/asm/uaccess_64.h +++ b/arch/x86/include/asm/uaccess_64.h @@ -12,13 +12,6 @@ #include <asm/cpufeatures.h> #include <asm/page.h> #include <asm/percpu.h> -#include <asm/runtime-const.h> - -/* - * Virtual variable: there's no actual backing store for this, - * it can purely be used as 'runtime_const_ptr(USER_PTR_MAX)' - */ -extern unsigned long USER_PTR_MAX; #ifdef CONFIG_ADDRESS_MASKING /* @@ -54,27 +47,7 @@ static inline unsigned long __untagged_addr_remote(struct mm_struct *mm, #endif #define valid_user_address(x) \ - ((__force unsigned long)(x) <= runtime_const_ptr(USER_PTR_MAX)) - -/* - * Masking the user address is an alternative to a conditional - * user_access_begin that can avoid the fencing. This only works - * for dense accesses starting at the address. - */ -static inline void __user *mask_user_address(const void __user *ptr) -{ - unsigned long mask; - asm("cmp %1,%0\n\t" - "sbb %0,%0" - :"=r" (mask) - :"r" (ptr), - "0" (runtime_const_ptr(USER_PTR_MAX))); - return (__force void __user *)(mask | (__force unsigned long)ptr); -} -#define masked_user_access_begin(x) ({ \ - __auto_type __masked_ptr = (x); \ - __masked_ptr = mask_user_address(__masked_ptr); \ - __uaccess_begin(); __masked_ptr; }) + ((__force unsigned long)(x) <= USER_PTR_MAX_CONST) /* * User pointers can have tag bits on x86-64. This scheme tolerates -- 2.47.0