[PATCH v2 2/4] dt-bindings: mailbox: mediatek: gce-mailbox: Add reference to gce-props.yaml

Jason-JH Lin (林睿祥) Jason-JH.Lin at mediatek.com
Thu Jan 11 23:44:13 PST 2024


On Thu, 2024-01-11 at 17:31 +0000, Conor Dooley wrote:
> On Wed, Jan 10, 2024 at 04:36:20PM +0000, Jason-JH Lin (林睿祥) wrote:
> > Hi Conor,
> > 
> > Thanks for the reviews.
> > 
> > On Wed, 2024-01-10 at 10:36 +0000, Conor Dooley wrote:
> > > On Wed, Jan 10, 2024 at 02:35:30PM +0800, Jason-JH.Lin wrote:
> > > > 1. Add "Provider" to the title to make it clearer.
> > > > 2. Add reference to gce-props.yaml for adding mediatek,gce-
> > > > events
> > > > property.
> > > 
> > > I can see this from the diff. There's still no explanation here
> > > as to
> > > why the mailbox provider needs to have a gce-event id. NAK until
> > > you
> > > can
> > > explain that.
> > > 
> > 
> > Sorry for missing the reason in commit message, I'll add it in the
> > next
> > version.
> > 
> > There are 2 reasons why the mailbox provider needs gce-events:
> > 1. The mailbox provider here is CMDQ mailbox driver. It configures
> > GCE
> > hardware register by CPU directly. If we want to set the default
> > value
> > from 0 to 1 for specific gce-events during the initialization of
> > CMDQ
> > driver. We need to tell CMDQ driver what gce-events need to be set
> > and
> > I think such GCE hardware setting can get from its device node.
> 
> Why would someone want to set it to 1 or 0?

GCE HW have an event table in SRAM. Each event ID has a boolean event
value and the default value is 0. There are 1024 event IDs (0~1023) in
the event table. The mediatek,gce-events is the property to get the
event IDs.

E.g.
In some camera suspend/resume use cases, the may set the event ID: 100
to 1 before suspend. When CMDQ driver resumes, all the event value of
event IDs will be clear to 0. But camera driver won't know the event
IDs:100 has been cleared to 0 and still send a cmd to wait for the
event ID:100. So CMDQ driver may need to keep the value of event ID:
100 before suspend and set back the value to 1 after CMDQ driver
clearing all the event value of event IDs.

CMDQ driver will need to know which event IDs' values need to be
backupped and restored in that camera suspend/resume use case.

> At what level will that vary? Per SoC? Per board? Something else?
> 

I think the SoC supports camera feature may need this property.

> > 2. We'll have the secure CMDQ mailbox driver in the future patch
> > [1].
> > It will request or reserve a mailbox channel, which is a dedicate
> > GCE
> > thread, as a secure IRQ handler. This GCE thread executes a looping
> > instruction set that keeps waiting for the gce-event set from
> > another
> > GCE thread in the secure world. So we also need to tell the CMDQ
> > driver
> > what gce-event need to be waited.
> 
> Ditto here, what level does this vary at? Do different SoCs or
> different
> boards/platforms dictate the value?

It's a SoC level, the SoC supports secure feature will need this
property.

> Could this channel be determined from the soc-specific compatible?
> 
> In other words, please explain in your commit message why this
> requires
> a property and is not detectable from any existing mechanism. From
> reading this I don't know what is preventing the secure mailbox
> channel
> from picking a "random" unused channel.

The secure channel could be dedicated from the soc-specific compatible,
but the event ID couldn't.

The same event signal corresponding event ID may changes in different
SoC.
E.g.
The HW event signal for CMDQ_EVENT_VDO0_MUTEX_STREAM_DONE_0 is
corresponding to GCE event ID: 574 in MT8188, but it's corresponding to
eventID: 597 in MT8195.

I can take any of the "unused" mailbox channel for the secure irq
handler. But without the property mediatek,gce-events, the secure
channel might not know what event ID to wait for.

Regards,
Jason-JH.Lin
> 
> Thanks,
> Conor.
> 
> > [1] cmdq_sec_irq_notify_start() is where the GCE thread is
> > requested to
> > prepare a looping instruction set to wait for the gce-event.
> > - 
> > 
https://patchwork.kernel.org/project/linux-mediatek/patch/20231222045228.27826-9-jason-jh.lin@mediatek.com/
> > 
> > Regards,
> > Jason-JH.Lin
> > 
> > > Cheers,
> > > Conor.
> > > 
> > > > 
> > > > Signed-off-by: Jason-JH.Lin <jason-jh.lin at mediatek.com>
> > > > ---
> > > >  .../devicetree/bindings/mailbox/mediatek,gce-mailbox.yaml   |
> > > > 6
> > > > ++++--
> > > >  1 file changed, 4 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git
> > > > a/Documentation/devicetree/bindings/mailbox/mediatek,gce-
> > > > mailbox.yaml
> > > > b/Documentation/devicetree/bindings/mailbox/mediatek,gce-
> > > > mailbox.yaml
> > > > index cef9d7601398..728dc93117a6 100644
> > > > --- a/Documentation/devicetree/bindings/mailbox/mediatek,gce-
> > > > mailbox.yaml
> > > > +++ b/Documentation/devicetree/bindings/mailbox/mediatek,gce-
> > > > mailbox.yaml
> > > > @@ -4,7 +4,7 @@
> > > >  $id: 
> > > > 
http://devicetree.org/schemas/mailbox/mediatek,gce-mailbox.yaml#
> > > >  $schema: http://devicetree.org/meta-schemas/core.yaml#
> > > >  
> > > > -title: Mediatek Global Command Engine Mailbox
> > > > +title: MediaTek Global Command Engine Mailbox Provider
> > > >  
> > > >  maintainers:
> > > >    - Houlong Wei <houlong.wei at mediatek.com>
> > > > @@ -57,6 +57,8 @@ required:
> > > >    - clocks
> > > >  
> > > >  allOf:
> > > > +  - $ref: mediatek,gce-props.yaml
> > > > +
> > > >    - if:
> > > >        not:
> > > >          properties:
> > > > @@ -67,7 +69,7 @@ allOf:
> > > >        required:
> > > >          - clock-names
> > > >  
> > > > -additionalProperties: false
> > > > +unevaluatedProperties: false
> > > >  
> > > >  examples:
> > > >    - |
> > > > -- 
> > > > 2.18.0
> > > > 


More information about the Linux-mediatek mailing list