At Sat, 26 Sep 2009 17:06:23 +0100 (BST), Mark Hills wrote:
On Fri, 25 Sep 2009, Daniel Mack wrote:
On Fri, Sep 25, 2009 at 04:22:15PM +0100, Mark Hills wrote:
The driver documentation states that local interrupts are already disabled when trigger() is called, so only a spinlock is required.
Hmm. I assume you tested them well. But I remember to have had quite some trouble at this point on SMP machines and ended up with that irqsave code. However, if you're sure that fixes a bug or is not necessary, let's take the patches and I'll test them once I'm back home around mid Oct.
I did check both non and SMP machines, and it corresponds to the docs. I also removed the patches to verify I could reproduce the original problems.
The irqsave is needed for deactivate_substream() because it's called also in your hw_free callback, which is non-atomic. In the real case, it's hard to hit this codepath, though.
Meanwhile, activate_substream() is called only from the trigger callback, and thus it's safe to use without irqsave version.
thanks,
Takashi