Heiko,

On Wed, Nov 19, 2014 at 10:46 AM, Heiko St?bner [off-list ref] wrote:

Am Mittwoch, 19. November 2014, 09:54:13 schrieb Doug Anderson:

quoted

Hi,

On Tue, Nov 18, 2014 at 3:49 PM, Doug Anderson [off-list ref]

wrote:

quoted

quoted

+static void rockchip_irq_disable(struct irq_data *d)
+{
+       struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d);
+       u32 val;
+
+       irq_gc_lock(gc);
+       val = irq_reg_readl(gc, GPIO_INTEN);
+       irq_reg_writel(gc, val & ~d->mask, GPIO_INTEN);
+       irq_gc_unlock(gc);
+}

Off list, Dmitry asked me why I didn't use irq_gc_mask_disable_reg()
and irq_gc_unmask_enable_reg() (AKA why I coded up my own function
here).  Originally I tried to use irq_gc_mask_disable_reg() and
irq_gc_unmask_enable_reg().  ..but they're really not designed to work
in tandem with the irq_gc_mask_set_bit() and irq_gc_mask_clr_bit().

Specifically if you try to use one set of functions for your
mask/unmask and the other for your disable/enable you'll find that
they stomp on each other.  Both functions upkeep the exact same
"mask_cache" variable.

Personally I'm totally baffled by how irq_gc_mask_disable_reg() and
irq_gc_unmask_enable_reg() could actually be sane, but that's maybe a
topic for another discussion.

I don't think irq_gc_mask_disable_reg and irq_gc_unmask_enable_reg are meant
as callbacks for irq_enable/irq_disable. As the name implies they are
standardized callbacks for irq_mask/irq_unmask on machines using a different
scheme for masking. So I would expected that they operate on the same
mask_cache because both types of functions handle masking on different types of
interrupt controllers.

Agreed that they aren't meant for irq_enable / irq_disable and that
it's not a bug.  It was just so tempting to use them and Dmitry
wondered the same thing, so I wrote an email detailing this.  Even
though these aren't for use as enable/disable, I will point out that
they don't seem sane (to me) for masking...

There don't seem to be any generalized callbacks for irq_enable/irq_disable
themself, probably because machines do the most uncommon things there :-)

Fair enough.  If it becomes common someone can move my functions
somewhere common.  ;)


Do you think this patch is something that should land?  Do I need to
prove that it's useful before we actually land it?  Right now I just
posted it because it seemed better, not because it fixes anything that
I know of.

-Doug