On Fri, 19 Sep 2025, Ville Syrjälä <[email protected]> wrote:
> On Fri, Sep 19, 2025 at 12:51:49PM +0300, Jani Nikula wrote:
>> +void ilk_display_irq_master_disable(struct intel_display *display, u32 
>> *de_ier, u32 *sde_ier)
>> +{
>> +    /* disable master interrupt before clearing iir  */
>> +    *de_ier = intel_de_read(display, DEIER);
>> +    intel_de_write(display, DEIER, *de_ier & ~DE_MASTER_IRQ_CONTROL);
>
> The original used completely unlocked register accessors. Assuming that
> has been working all along without hangs I don't really want to
> re-introduce the lock again. I recently looked at some profiles and
> the irq handlers showed up fairly prominently so making these a lot
> more expensive isn't really desirable. So I'd go with the _fw() variant
> for this stuff.
>
> I've already written a patch that I was planning to send at some point
> that switches all the bdw+ display irq handlers to _fw() to reduce the
> overhead a bit. Would be nice if we could do the same for all the
> earlier platforms too, or at least not make it much more expensive
> than it already is.

Ack. I was actually going to use _fw, but I don't know what happened
between brain and keyboard. :/

BR,
Jani.


-- 
Jani Nikula, Intel

Reply via email to