Re: [PATCH] RFC x86_64 more accurate KSTK_ESP implementation

[Andi Kleen]

> +void update_usersp(struct pt_regs *regs)
> +{
> +	unsigned long stk = (unsigned long)task_stack_page(current);
> +	unsigned long stkp = (regs)->sp; 
> +
> +	if (((stkp < stk) || (stkp >= stk + THREAD_SIZE))
> +	    && regs->ip < PAGE_OFFSET)
> +		percpu_write(old_rsp, stkp);

This does not handle interrupt and exception stacks correctly.

Also regs->ip is never a safe check for running in user space,
because a program can set the IP to a arbitrary value for a one
instruction window.

The larger problem is also if the kernel moves to no-tick-for-non-idle
(which I guess will happen sooner or later) your method won't
work anyways, or again be arbitarily inaccurate. Even today 10ms
worst time inaccuracy for HZ=100 is rather bad, there can be a lot of stack
allocations in that time. And adding new dependencies on a regular
timer when everything else is moving away from that doesn't seem right.

Also I suspect this method won't work on preempt-rt without 
additional tweaks.

-Andi
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/