Hi Laruent,
> -----Original Message-----
> From: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
> Sent: Tuesday, October 8, 2024 5:44 AM
> To: G.N. Zhou (OSS) <guoniu.zhou@xxxxxxxxxxx>
> Cc: rmfrfs@xxxxxxxxx; martink@xxxxxxxxx; kernel@xxxxxxx;
> mchehab@xxxxxxxxxx; shawnguo@xxxxxxxxxx; s.hauer@xxxxxxxxxxxxxx;
> kernel@xxxxxxxxxxxxxx; festevam@xxxxxxxxx; imx@xxxxxxxxxxxxxxx; linux-
> media@xxxxxxxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH 0/3] media: imx8mq-mipi-csi2: Simplify power
> management handling
>
> Hi Guo,
>
> On Mon, Sep 30, 2024 at 07:51:45AM +0000, G.N. Zhou (OSS) wrote:
> > On Monday, September 30, 2024 3:22 PM, Laurent Pinchart wrote:
> > > On Mon, Sep 30, 2024 at 07:08:09AM +0000, G.N. Zhou (OSS) wrote:
> > > > On Sunday, September 29, 2024 9:44 PM, Laurent Pinchart wrote:
> > > > >
> > > > > Hello,
> > > > >
> > > > > This small patch series is a reaction to "[PATCH] media: nxp:
> > > > > imx8mq-mipi-csi2: Fix CSI clocks always enabled issue" ([1]).
> > > > > Instead of making the PM handling more complex, I think it can be
> greatly simplified.
> > > > >
> > > > > I have only compile-tested the patches. Guoniu, could you give this a
> try ?
> > > >
> > > > After applying the patches and test both on iMX8ULP.
> > > >
> > > > For iMX8ULP, it will cause kernel dump when access CSI registers
> > > > and system hang during do suspend/resume while streaming Need to
> > > > add system suspend/resume handlers and call
> > > > pm_runtime_force_suspend/resume in the handlers.
> > > >
> > > > I tried to debug this issue and found pm runtime callback won't be
> > > > called when system resume. The state of power domain won't enabled.
> > >
> > > Thank you for testing.
> > >
> > > I wonder if this could be caused by the CSI bridge being resumed
> > > from system sleep before the CSI-2 receiver. Could you check if
> > > that's the case ? If so, does the following change fix the issue ?
> >
> > I tested on iMX8ULP which don't use CSI bridge but ISI, not iMX8MQ. In
> > ISI driver, I notice that it already handler the device relationship when subdev
> bound like bellow:
> >
> > link = device_link_add(isi->dev, sd->dev, DL_FLAG_STATELESS); if
> > (!link) {
> > dev_err(isi->dev,
> > "Failed to create device link to source %s\n", sd->name);
> > return -EINVAL;
> > }
>
> Ah yes indeed with the ISI it should already be handled.
>
> I can't test this on hardware now as I'm travelling. Is the system hang happening
> at suspend or resume time ? What is the order of the suspend/resume handlers
> calls for the imx8-isi driver and the
> imx8mq-mipi-csi2 driver ?
#1) Is the system hang happening at suspend or resume time ?
After applying your patches, system hang will happened at resume time. Adding
log and enable no_console_suspend, I can know the the point of causing issue is
accessing CSI registers when top-level(ISI) device try to enable stream from CSI in
ISI resume callback like bellow:
mxc_isi_pm_resume(dev)
mxc_isi_video_resume(pipe)
mxc_isi_pipe_enable(pipe)
v4l2_subdev_enable_streams(...)
mx8mq_mipi_csi_s_stream()
imx8mq_mipi_csi_start_stream()
imx8mq_mipi_csi_set_params(state)
imx8mq_mipi_csi_write(state, CSI2RX_CFG_NUM_LANES, lanes - 1); <--- causing kernel panic and system don't have any response
#2) What is the order of the suspend/resume handlers calls for the imx8-isi driver and the
imx8mq-mipi-csi2 driver ?
Suspend: ISI -> CSI
Resume: CSI -> ISI
One more thing when I debugging this issue that the value of CSI device "power.usage_count" in two cases
Case 1: Don't do system suspend/resume and only run camera test.
Top-level device(ISI) can call CSI s_stream when start/stop capture and CSI runtime_suspend/resume will
be called in this case since the value of CSI device power.usage_count is 1
Case 2: Run camera test first and keep it run in backend, then run system suspend/resume.
Top-level device(ISI) will CSI s_stream() to start capture when run camera test and then, ISI system suspend
callback will disable stream from CSI by calling s_stream(), the pm_runtime_put(dev) doesn't call CSI runtime
suspend callback since the value of CSI device power.usage_count is 2.
I'm not sure if it's helpfull, just for your reference.
>
> > For iMX8MQ, I'm trying to enable it, but meet some problems, so can't
> > give you the results in short time.
> >
> > > diff --git a/drivers/media/platform/nxp/imx7-media-csi.c
> > > b/drivers/media/platform/nxp/imx7-media-csi.c
> > > index 9566ff738818..c66b0621e395 100644
> > > --- a/drivers/media/platform/nxp/imx7-media-csi.c
> > > +++ b/drivers/media/platform/nxp/imx7-media-csi.c
> > > @@ -2057,9 +2057,22 @@ static int imx7_csi_notify_bound(struct
> > > v4l2_async_notifier *notifier, {
> > > struct imx7_csi *csi = imx7_csi_notifier_to_dev(notifier);
> > > struct media_pad *sink = &csi->sd.entity.pads[IMX7_CSI_PAD_SINK];
> > > + struct device_link *link;
> > >
> > > csi->src_sd = sd;
> > >
> > > + /*
> > > + * Enforce suspend/resume ordering between the source (supplier) and
> > > + * the CSI (consumer). The source will be suspended before and resume
> > > + * after the CSI.
> > > + */
> > > + link = device_link_add(csi->dev, sd->dev, DL_FLAG_STATELESS);
> > > + if (!link) {
> > > + dev_err(csi->dev,
> > > + "Failed to create device link to source %s\n", sd-
> >name);
> > > + return -EINVAL;
> > > + }
> > > +
> > > return v4l2_create_fwnode_links_to_pad(sd, sink,
> MEDIA_LNK_FL_ENABLED |
> > > MEDIA_LNK_FL_IMMUTABLE); }
> > >
> > > > > [1]
> > > > > https://lore.kernel.org/r/20240929101635.1648234-1-guoniu.zhou@o
> > > > > ss.nxp.com
> > > > >
> > > > > Laurent Pinchart (3):
> > > > > media: imx8mq-mipi-csi2: Drop stream stop/restart at suspend/resume
> time
> > > > > media: imx8mq-mipi-csi2: Drop ST_SUSPENDED guard
> > > > > media: imx8mq-mipi-csi2: Drop system suspend/resume handlers
> > > > >
> > > > > drivers/media/platform/nxp/imx8mq-mipi-csi2.c | 113
> > > > > ++----------------
> > > > > 1 file changed, 10 insertions(+), 103 deletions(-)
> > > > >
> > > > >
> > > > > base-commit: 81ee62e8d09ee3c7107d11c8bbfd64073ab601ad
>
> --
> Regards,
>
> Laurent Pinchart
