Re: [PATCHv4] uprobes: simplify rip-relative handling

[Oleg Nesterov <oleg@redhat.com>]

Thanks, I hope that Jim's ack still applies to this version.

On 05/01, Denys Vlasenko wrote:
>
> v4: Changed arch_uprobe_xol_was_trapped() comment to reflect new logic.

Hmm. I guess you meant arch_uprobe_post_xol()... please see below.

>  static int default_post_xol_op(struct arch_uprobe *auprobe, struct pt_regs *regs)
>  {
>  	struct uprobe_task *utask = current->utask;
> -	long correction = (long)(utask->vaddr - utask->xol_vaddr);
>  
> -	riprel_post_xol(auprobe, regs, &correction);
> +	riprel_post_xol(auprobe, regs);
>  	if (auprobe->def.fixups & UPROBE_FIX_IP) {
> +		long correction = (long)(utask->vaddr - utask->xol_vaddr);

Can't resist, I'll remove this pointless cast ;)

>   * If the original instruction was a rip-relative instruction such as
>   * "movl %edx,0xnnnn(%rip)", we have instead executed an equivalent
> - * instruction using a scratch register -- e.g., "movl %edx,(%rax)".
> - * We need to restore the contents of the scratch register and adjust
> - * the ip, keeping in mind that the instruction we executed is 4 bytes
> - * shorter than the original instruction (since we squeezed out the offset
> - * field).  (FIX_RIP_AX or FIX_RIP_CX)
> + * instruction using a scratch register -- e.g., "movl %edx,0xnnnn(%rax)".
> + * We need to restore the contents of the scratch register
> + * (FIX_RIP_AX or FIX_RIP_CX).
>   */
>  int arch_uprobe_post_xol(struct arch_uprobe *auprobe, struct pt_regs *regs)

Perhaps it makes sense to move this part of the comment above
default_post_xol_op() which actually does this...

I won't insist, I do not really care because I almost never read the comments
anyway ;)

Oleg.