Re: [PATCH v21 04/14] media: qcom: camss: Issue stream-on REG_UPDATE from CSID

[Hangxiang Ma]

On 10/21/2026 9:00 PM, Loic Poulain wrote:
> On Wed, Oct 21, 2026 at 1:24 PM Hangxiang Ma
> <hangxiang.ma@xxxxxxxxxxxxxxxx> wrote:
> > On 10/21/2026 4:53 PM, Bryan O'Donoghue wrote:
> > > On 21/10/2026 08:37, Wenmeng Liu wrote:
> > > > Hi Bryan, Hangxiang,
> > > >
> > > > On 10/12/2026 2:45 PM, Hangxiang Ma wrote:
> > > > > On stream-on, CAMSS starts the subdevices upstream from the video node
> > > > > through their sink pads, so the VFE is started before the CSID.
> > > > >
> > > > > Starting with CSID 680, the REG_UPDATE is issued through the CSID
> > > > > rather than the VFE. However, vfe_enable_output_v2() still issues a
> > > > > REG_UPDATE after queueing each initial buffer, before the CSID has
> > > > > configured the stream. The VFE then never signals BUF_DONE. Therefore,
> > > > > the CAMSS now issues the REG_UPDATE from configure_stream() on CSID 680
> > > > > and CSID Gen3 instead. The operation happens after the RX configuration
> > > > > and before the RDI path is enabled.
> > > > >
> > > > > Skip the VFE-side stream-on REG_UPDATE only when the CSID connected to
> > > > > the VFE sink pad implements the reg_update callback. Platforms whose
> > > > > CSID does not provide this callback, as well as lines fed by ISPIF,
> > > > > keep the existing VFE-side REG_UPDATE.
> > > > >
> > > > > Signed-off-by: Hangxiang Ma <hangxiang.ma@xxxxxxxxxxxxxxxx>
> > > > > ---
> > > > >    drivers/media/platform/qcom/camss/camss-csid-680.c |  3 ++
> > > > >    .../media/platform/qcom/camss/camss-csid-gen3.c    |  3 ++
> > > > >    drivers/media/platform/qcom/camss/camss-vfe.c      | 37 +++++++++++
> > > > > + +++++++++-
> > > > >    3 files changed, 42 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/drivers/media/platform/qcom/camss/camss-csid-680.c b/
> > > > > drivers/media/platform/qcom/camss/camss-csid-680.c
> > > > > index 345a67c8fb94..556cdfef6207 100644
> > > > > --- a/drivers/media/platform/qcom/camss/camss-csid-680.c
> > > > > +++ b/drivers/media/platform/qcom/camss/camss-csid-680.c
> > > > > @@ -296,6 +296,8 @@ static void __csid_configure_rdi_stream(struct
> > > > > csid_device *csid, u8 enable, u8
> > > > >        writel(val, csid->base + CSID_RDI_CFG0(port));
> > > > >    }
> > > > > +static void csid_subdev_reg_update(struct csid_device *csid, int
> > > > > port_id, bool is_clear);
> > > > > +
> > > > >    static void csid_configure_stream(struct csid_device *csid, u8 enable)
> > > > >    {
> > > > >        int i;
> > > > > @@ -307,6 +309,7 @@ static void csid_configure_stream(struct
> > > > > csid_device *csid, u8 enable)
> > > > >            if (csid->phy.en_vc & BIT(i)) {
> > > > >                __csid_configure_rdi_stream(csid, enable, i, 0);
> > > > >                __csid_configure_rx(csid, &csid->phy, 0);
> > > > > +            csid_subdev_reg_update(csid, i, false);
> > > > >                __csid_ctrl_rdi(csid, enable, i);
> > > > >            }
> > > > >        }
> > > > > diff --git a/drivers/media/platform/qcom/camss/camss-csid-gen3.c b/
> > > > > drivers/media/platform/qcom/camss/camss-csid-gen3.c
> > > > > index 0fdbf75fb27d..830c9c818d3d 100644
> > > > > --- a/drivers/media/platform/qcom/camss/camss-csid-gen3.c
> > > > > +++ b/drivers/media/platform/qcom/camss/camss-csid-gen3.c
> > > > > @@ -219,6 +219,8 @@ static void __csid_configure_rdi_stream(struct
> > > > > csid_device *csid, u8 enable, u8
> > > > >        writel(val, csid->base + CSID_RDI_CFG0(port));
> > > > >    }
> > > > > +static void csid_subdev_reg_update(struct csid_device *csid, int
> > > > > port_id, bool is_clear);
> > > > > +
> > > > >    static void csid_configure_stream(struct csid_device *csid, u8 enable)
> > > > >    {
> > > > >        u8 i;
> > > > > @@ -230,6 +232,7 @@ static void csid_configure_stream(struct
> > > > > csid_device *csid, u8 enable)
> > > > >            if (csid->phy.en_vc & BIT(i)) {
> > > > >                __csid_configure_rdi_stream(csid, enable, i, 0);
> > > > >                __csid_configure_rx(csid, &csid->phy, 0);
> > > > > +            csid_subdev_reg_update(csid, i, false);
> > > > >                __csid_ctrl_rdi(csid, enable, i);
> > > > >            }
> > > > >    }
> > > > > diff --git a/drivers/media/platform/qcom/camss/camss-vfe.c b/drivers/
> > > > > media/platform/qcom/camss/camss-vfe.c
> > > > > index 2680a9914765..16a0b5663fe2 100644
> > > > > --- a/drivers/media/platform/qcom/camss/camss-vfe.c
> > > > > +++ b/drivers/media/platform/qcom/camss/camss-vfe.c
> > > > > @@ -502,6 +502,37 @@ void vfe_buf_done(struct vfe_device *vfe, int wm)
> > > > >        spin_unlock_irqrestore(&vfe->output_lock, flags);
> > > > >    }
> > > > > +/*
> > > > > + * vfe_upstream_csid_has_reg_update - Check whether the CSID feeding
> > > > > this line
> > > > > + * issues its own streamon RUP (i.e. its configure_stream does the
> > > > > RUP).
> > > > > + *
> > > > > + * The CSID->VFE link is created in camss.c for all CSIDs and VFEs,
> > > > > so the
> > > > > + * upstream entity of the sink pad is either a CSID (possibly with
> > > > > RUP) or an
> > > > > + * ISPIF (no RUP). Only enabled links are considered.
> > > > > + */
> > > > > +static bool vfe_upstream_csid_has_reg_update(struct vfe_line *line)
> > > > > +{
> > > > > +    struct vfe_device *vfe = to_vfe(line);
> > > > > +    struct camss *camss = vfe->camss;
> > > > > +    struct media_pad *remote;
> > > > > +    unsigned int i;
> > > > > +
> > > > > +    remote = media_pad_remote_pad_first(&line->pads[MSM_VFE_PAD_SINK]);
> > > > > +    if (!remote || !is_media_entity_v4l2_subdev(remote->entity))
> > > > > +        return false;
> > > > > +
> > > > > +    for (i = 0; i < camss->res->csid_num; i++) {
> > > > > +        struct csid_device *csid = &camss->csid[i];
> > > > > +
> > > > > +        if (&csid->subdev.entity != remote->entity)
> > > > > +            continue;
> > > > > +
> > > > > +        return !!csid->res->hw_ops->reg_update;
> > > > > +    }
> > > > > +
> > > > > +    return false;
> > > > > +}
> > > > > +
> > > > >    int vfe_enable_output_v2(struct vfe_line *line)
> > > > >    {
> > > > >        struct vfe_device *vfe = to_vfe(line);
> > > > > @@ -511,6 +542,7 @@ int vfe_enable_output_v2(struct vfe_line *line)
> > > > >        unsigned long flags;
> > > > >        unsigned int frame_skip = 0;
> > > > >        unsigned int i;
> > > > > +    bool rup_in_csid;
> > > > >        sensor_pad = camss_find_sensor_pad(&line->subdev.entity);
> > > > >        if (sensor_pad) {
> > > > > @@ -523,6 +555,8 @@ int vfe_enable_output_v2(struct vfe_line *line)
> > > > >                frame_skip = VFE_FRAME_DROP_VAL - 1;
> > > > >        }
> > > > > +    rup_in_csid = vfe_upstream_csid_has_reg_update(line);
> > > > > +
> > > > >        spin_lock_irqsave(&vfe->output_lock, flags);
> > > > >        ops->reg_update_clear(vfe, line->id);
> > > > > @@ -552,7 +586,8 @@ int vfe_enable_output_v2(struct vfe_line *line)
> > > > >            output->gen2.active_num++;
> > > > >            ops->vfe_wm_update(vfe, output->wm_idx[0],
> > > > >                       output->buf[i]->addr[0], line);
> > > > > -        ops->reg_update(vfe, line->id);
> > > > > +        if (!rup_in_csid)
> > > > > +            ops->reg_update(vfe, line->id);
> > > > >        }
> > > > >        spin_unlock_irqrestore(&vfe->output_lock, flags);
> > > >
> > > > According to the hardware documentation:
> > > > Starting with CSID 680, CAMIF has moved into the CSID. As a result,
> > > > RUP is now issued by the CSID and propagated to the IFE.
> > > >
> > > > Starting with CSID 980, software must also set the SET bit in
> > > > RUP_AUP_CMD after issuing the RUP/AUP command. Otherwise, the hardware
> > > > ignores the command and the corresponding register update does not
> > > > take effect.
> > > >
> > > >
> > > > Bryan, according to the HW documentation, the change itself is
> > > > consistent with the required programming sequence. The main question
> > > > is whether this is the right implementation from the driver
> > > > perspective, or if there is a cleaner way to handle it.
> > > >
> > > > I have tested the change on the Hamoa EVK board.
> > > > Tested-by: Wenmeng Liu <wenmeng.liu@xxxxxxxxxxxxxxxx> # Hamoa EVK
> > > >
> > > >
> > > > Thanks,
> > > > Wenmeng
> > >
> > > OK thank you for the explanation.
> > >
> > > OTOH I agree with Loic's statement - it is cleaner to subtract the
> > > callback and disjunct on its presence than to add a flag to the same
> > > effect in reverse.
> > >
> > > ---
> > > bod
> >
> > Hi Loic, Bryan, thanks for the review. I wonder if it would be better to
> > keep the check on the CSID side?
> >
> > vfe->reg_update() is also used outside stream-on, such as bufdone, queue
> > buffer. It must be implemented in VFE. We only want to change the VFE
> > behavior at stream-on stage, in function vfe_enable_output_v2(). So
> > making vfe->reg_update() optional could affect runtime updates.
>
> But the patch says:
> 'The REG_UPDATE is issued through the CSID rather than the VFE'
>
> So you mean it's not always driven by CSID?
>
> Also, today we perform a reg_update after queuing a buffer (i.e. after
> programming ADDR). However, as far as I know, this is no longer
> necessary in recent VFE/TFE/IFE generations. In those implementations,
> the ADDR register is not double-buffered and is not part of the
> reg-update set. Instead, it behaves as a FIFO entry, so issuing a
> reg_update only results in a useless RUP_DONE interrupt.
>
> > Maybe the cheapest way is still to check the CSID side reg_update
> > implementation at stream-on stage? I'm very glad to hear your
> > suggestions. Thanks a lot.
> >
> > ---
> > Best Regards,
> > Hangxiang

Yes. Before the 680 platform, VFE issued the REG_UPDATE itself. Starting 
with 680, however, the REG_UPDATE signal is controlled by CSID rather 
than VFE. The CSID stream-on RUP is issued from the CSID start path 
after VFE has been started.

Wenmeng previously described the hardware evolution and the documented 
initialization sequence in this thread. I hope this helps clarify the 
background.

https://lore.kernel.org/all/3843110c-4766-4352-ab2a-6d3017336534@xxxxxxxxxxxxxxxx/

Going back to the split AUP/RUP design, this was introduced in the 
Pakala (980) generation. The split AUP/RUP design allows us to select 
ADDR or REG updates independently. However, since the 680 generation, 
REG_UPDATE has been moved to the CSID side, which does not have the same 
split capability.

I agree that combining AUP and RUP would generate an unnecessary 
RUP_DONE interrupt. However, please take a look at the implementation of 
csid_subdev_reg_updatem() in camss-csid-1080.c. In this series, we have 
already followed the approach you suggested above: during runtime, we 
configure and trigger only AUP, without RUP. Therefore, no unnecessary 
RUP_DONE interrupt will be generated.

I'm very open to further discussion and sincerely appreciate your 
feedback. If there is anything I have misunderstood, or if you have a 
better approach in mind, please feel free to point it out. I'd be more 
than happy to clarify any questions and learn from your suggestions.

---
Best Regards,
Hangxiang