[PATCH 5/5] mailbox: mtk-cmdq: Add support runtime get and set GCE event

Jason-JH Lin (林睿祥) Jason-JH.Lin at mediatek.com
Mon Mar 4 19:09:31 PST 2024


Hi Angelo,

Thanks for the reviews.

On Mon, 2024-03-04 at 11:06 +0100, AngeloGioacchino Del Regno wrote:
> Il 01/03/24 12:11, Jason-JH.Lin ha scritto:
> > ISP drivers need to get and set GCE event in their runtime contorl
> > flow.
> > So add these functions to support get and set GCE by CPU.
> > 
> > Signed-off-by: Jason-JH.Lin <jason-jh.lin at mediatek.com>
> > Change-Id: I494c34ebc5ec26c82213f2bc03d2033d60652523
> 
> Change-Id makes no sense upstream. Please drop.

OK, I'll drop it.

> 
> > ---
> >   drivers/mailbox/mtk-cmdq-mailbox.c       | 37
> > ++++++++++++++++++++++++
> >   include/linux/mailbox/mtk-cmdq-mailbox.h |  2 ++
> >   2 files changed, 39 insertions(+)
> > 
> > diff --git a/drivers/mailbox/mtk-cmdq-mailbox.c
> > b/drivers/mailbox/mtk-cmdq-mailbox.c
> > index ead2200f39ba..d7c08249c898 100644
> > --- a/drivers/mailbox/mtk-cmdq-mailbox.c
> > +++ b/drivers/mailbox/mtk-cmdq-mailbox.c
> > @@ -25,7 +25,11 @@
> >   #define CMDQ_GCE_NUM_MAX		(2)
> >   
> >   #define CMDQ_CURR_IRQ_STATUS		0x10
> > +#define CMDQ_SYNC_TOKEN_ID		0x60
> > +#define CMDQ_SYNC_TOKEN_VALUE		0x64
> > +#define CMDQ_TOKEN_ID_MASK			GENMASK(9, 0)
> >   #define CMDQ_SYNC_TOKEN_UPDATE		0x68
> > +#define CMDQ_TOKEN_UPDATE_VALUE			BIT(16)
> >   #define CMDQ_THR_SLOT_CYCLES		0x30
> >   #define CMDQ_THR_BASE			0x100
> >   #define CMDQ_THR_SIZE			0x80
> > @@ -83,6 +87,7 @@ struct cmdq {
> >   	struct cmdq_thread	*thread;
> >   	struct clk_bulk_data	clocks[CMDQ_GCE_NUM_MAX];
> >   	bool			suspended;
> > +	spinlock_t		event_lock; /* lock for gce event */
> >   };
> >   
> >   struct gce_plat {
> > @@ -113,6 +118,38 @@ u8 cmdq_get_shift_pa(struct mbox_chan *chan)
> >   }
> >   EXPORT_SYMBOL(cmdq_get_shift_pa);
> >   
> > +void cmdq_set_event(void *chan, u16 event_id)
> > +{
> > +	struct cmdq *cmdq = container_of(((struct mbox_chan *)chan)-
> > >mbox,
> > +		typeof(*cmdq), mbox);
> 
> struct mbox_chan *mbc = chan;
> struct cmdq *cmdq = container_of(mbc->mbox, ... etc); (and this fits
> in one line)
> 
OK, I'll change it.

> > +	unsigned long flags;
> > +
> > +	spin_lock_irqsave(&cmdq->event_lock, flags);
> 
> Why do you need irqsave/irqrestore? I think I know, but please
> explain.
> 
Because ISP driver may call cmdq_get_event() first than use
cmdq_set_event() to update the event status in one
mtk_imgsys_setevent() function frequently.

And mtk_imgsys_setevent() will be called in SW multi-thread after cmdq
callback from cmdq_irq_handler, so we use the spin_lock_irqsave to
avoid the race condition.

But after discussing with ISP owners, they dropped
mtk_imgsys_setevent() recently, so I'll drop this patch in the next version.

Regards,
Jason-JH.Lin



More information about the Linux-mediatek mailing list