Re: [PATCH v4] introduce macro spin_event_timeout()

[Timur Tabi <timur@freescale.com>]

On Tue, Mar 10, 2009 at 7:37 PM, Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:

> Given the proposed helper isn't a sane default for x86 I think it needs a
> good deal more work.

Alan, I'm happy to put whatever effort is requirement to make this
code satisfactory.  I believe that this is a useful macro to have in
the kernel.  I do appreciate the attention you are giving it.

> It also hides details like that timing which is bad
> sometimes.

Are you talking about the udelay() inside the loop?  If so, I agree
that this is bad and have removed it in the PowerPC-specific version:

#define spin_event_timeout(condition, timeout)			\
({								\
	unsigned long __start = get_tbl();			\
	unsigned long __loops = tb_ticks_per_usec * timeout;	\
	int __ret = 1;						\
	while (!(condition)) {					\
		if (tb_ticks_since(__start) > __loops) {	\
			__ret = 0;				\
			break;					\
		}						\
		cpu_relax();					\
	}							\
	__ret;							\
})

tb_ticks_since() is a front-end to get_tbl(), which is similar to the
rdtsc instruction.  In this case, we're not adding arbitrary delays
into the loop, and we're not using jiffies, but we are
architecture-dependent.

-- 
Timur Tabi
Linux kernel developer at Freescale