|
Hello,
On Tue, 15 Jan 2002, Isaac Claymore wrote:
> Hi, all
> This might not be called a bug, however, I'd like to point it out.
> I met a comment in internal_add_sltimer() of file ip_vs_timer.c:
> /*
> * must be cli-ed when calling this
> */
> But later I found when internal_add_sltimer() is invoked in add_sltimer()
> and
> other places, HardIRQ isn't blocked, and plain vanilla spin_[un]lock() is
> used.
> And according to that comment above, spin_lock_irqsave() &
> spin_unlock_irqrestore()
> should be used instead. This confused me :)
> Later after I checked the timer.c in kernel source, I found that you LVS
> developers
> just copied the comment from timer.c. And since LVS slow timers isn't used in
> HardIRQ
> context like those kernel timers, we need not cli().
> You guys had successfully converted the irq-block version of spinlock
> routines of
Thanks
> timer.c in ip_vs_timer.c, but missed that comment.
Yep, the comment should be fixed. The sltimer functions are
not called from hardirq (LVS does not run there).
> OK, this is never a serious problem. But it might cause confusion among
> us newbies.
> If I'm right about this, would you maintainers correct it? Or, otherwise,
> beat me :)
You are right, for both 2.2 and 2.4
Regards
--
Julian Anastasov <ja@xxxxxx>
|
