Re: [PATCH] arm64/traps: Avoid unnecessary kernel/user pointer conversion

From: Mark Rutland
Date: Tue Sep 14 2021 - 12:01:08 EST


On Tue, Sep 14, 2021 at 08:57:42PM +0530, Amit Daniel Kachhap wrote:
> Annotating a pointer from kernel to __user and then back again might
> confuse sparse. In call_undef_hook() it can be avoided by not using the
> intermediate user pointer variable.

When you say "might confuse sparse", does it complain today? If so, can
you include an example of what goes wrong?

> Note: This patch adds no functional changes to code.
>
> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> Cc: Will Deacon <will@xxxxxxxxxx>
> Signed-off-by: Amit Daniel Kachhap <amit.kachhap@xxxxxxx>
> ---
> arch/arm64/kernel/traps.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arm64/kernel/traps.c b/arch/arm64/kernel/traps.c
> index b03e383d944a..357d10a8bbf5 100644
> --- a/arch/arm64/kernel/traps.c
> +++ b/arch/arm64/kernel/traps.c
> @@ -404,7 +404,8 @@ static int call_undef_hook(struct pt_regs *regs)
>
> if (!user_mode(regs)) {
> __le32 instr_le;
> - if (get_kernel_nofault(instr_le, (__force __le32 *)pc))
> + if (get_kernel_nofault(instr_le,
> + (__le32 *)instruction_pointer(regs)))

Can we make `pc` an unsigned long, instead?

It'd be nice to handle all three cases consistently, even if that means
adding __force to the two user cases.

Thanks,
Mark.

> goto exit;
> instr = le32_to_cpu(instr_le);
> } else if (compat_thumb_mode(regs)) {
> --
> 2.17.1
>