[lkml]   [2009]   [Jan]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH -v2] use per cpu data for single cpu ipi calls
On Thu, 2009-01-29 at 09:21 -0800, Linus Torvalds wrote:
> On Thu, 29 Jan 2009, Steven Rostedt wrote:
> >
> > The caller must wait till the LOCK bit is cleared before setting
> > it. When it is cleared, there is no IPI function using it.
> > A spinlock is used to synchronize the setting of the bit between
> > callers. Since only one callee can be called at a time, and it
> > is the only thing to clear it, the IPI does not need to use
> > any locking.
> That spinlock cannot be right. It is provably wrong for so many reasons..
> Think about it. We're talking about a per-CPU lock, which already makes no
> sense: we're only locking against our own CPU, and we've already disabled
> preemption for totally unrelated reasons.
> And the only way locking can make sense against our own CPU is if we lock
> against interrupts - but the lock isn't actually irq-safe, so if you are
> trying to lock against interrupts, you are (a) doing it wrong (you should
> disable interrupts, not use a spinlock) and (b) causing a deadlock if it
> ever happens.

> + else {
> + data = &per_cpu(csd_data, cpu);
> + spin_lock(&per_cpu(csd_data_lock, cpu));
> + while (data->flags & CSD_FLAG_LOCK)
> + cpu_relax();
> + data->flags = CSD_FLAG_LOCK;
> + spin_unlock(&per_cpu(csd_data_lock, cpu));
> + }

I think your argument would hold if he did:

data = &__get_cpu_var(csd_data);

But now he's actually grabbing the remote cpu's csd, and thus needs
atomicy around that remote csd -- which two cpus could contend for.

 \ /
  Last update: 2009-01-29 18:49    [W:0.098 / U:9.176 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site