On Tue, Sep 16, 2014 at 12:28:18PM +0200, Daniel Vetter wrote:
On Tue, Sep 16, 2014 at 11:21:52AM +0100, Chris Wilson wrote:
On Mon, Sep 15, 2014 at 02:35:07PM +0200, Daniel Vetter wrote:
irq handlers always run with interrupts locally disabled, so plain spinlocks is all we need. I've also reviewed again that they all follow the _irq_handler postfix convention.
Hmm, we still have the full irq dance inside the reg read/write macros, which themselves should never be used from inside the irq handlers.
(Modulo the misgivings in execlists_irq_handler).
Hm, we still have the ACTHEAD hack to read somewhat coherent-ish seqnos. Dunno whether open-coding that would make sense really, and whether it's really beneficial to force register writes to never acquire forcewake from irq context. Definitely material for different patches though.
ATCHD inside the irq handler? I am pretty sure that even -nightly doesn't have the irq-barrier inside the irq handler. And yes, we do want to avoid having a massive delay whilst we apply the forcewake workarounds from inside irq handlers. -Chris