Re: [PATCH v3 7/7] remoteproc: qcom: Always assert and deassert reset signals in SDM845

[Sibi S]

Hi Bjorn,
Thanks for the review

On 03/19/2018 04:16 AM, Bjorn Andersson wrote:
> On Wed 14 Mar 02:21 PDT 2018, Sibi S wrote:
> > diff --git a/drivers/remoteproc/qcom_q6v5_pil.c b/drivers/remoteproc/qcom_q6v5_pil.c
> [..]
> > +struct q6v5_reset_ops {
> > +	int (*reset_start)(struct q6v5 *qproc);
> > +	int (*reset_stop)(struct q6v5 *qproc);
> > +};
>
> Please add these two ops directly in q6v5 instead and please keep the
> naming "reset_assert" and "reset_deassert".

sure will do

> > +
> >   enum {
> >   	MSS_MSM8916,
> >   	MSS_MSM8974,
> > @@ -343,6 +354,52 @@ static int q6v5_load(struct rproc *rproc, const struct firmware *fw)
> >   	return 0;
> >   }
> >   
> > +static void alt_reset_restart(struct q6v5 *qproc, u32 restart)
> > +{
> > +	writel(restart, qproc->rmb_base + RMB_MBA_ALT_RESET);
>
> Just move this write into q6v5_sdm_reset_start()

sure will do

> > +}
> > +
> > +static int q6v5_msm_reset_stop(struct q6v5 *qproc)
> > +{
> > +	return reset_control_assert(qproc->mss_restart);
> > +}
> > +
> > +static int q6v5_msm_reset_start(struct q6v5 *qproc)
> > +{
> > +	return reset_control_deassert(qproc->mss_restart);
> > +}
> > +
> > +static int q6v5_sdm_reset_stop(struct q6v5 *qproc)
> > +{
> > +	return reset_control_reset(qproc->mss_restart);
> > +}
> > +
> > +static int q6v5_sdm_reset_start(struct q6v5 *qproc)
> > +{
> > +	int ret;
> > +
> > +	alt_reset_restart(qproc, 1);
> > +	/* Ensure alt reset is written before restart reg */
>
> That's not what udelay does ;)
>
> If you want to make sure that the register is written and then give it
> 100us delay until you reset the reset you have to readl() the same
> register back after the writel()
>
> I think the delay deserves a comment on what we're waiting for.

the delay can be removed for the reasons stated below

> > +	udelay(100);
>
> Use usleep_range() for delays longer than 10us.
>
> > +
> > +	ret = reset_control_reset(qproc->mss_restart);
> > +
> > +	udelay(100);
>
> Same as the delay above, what are we waiting for?

The point is to wait for 6 32khz sleep cycles both after assert and
after de-assert of the aoss mss reset line. So moving the udelay to
the aoss reset controller should have been right thing to do. alt_reset
shouldn't need delays in between will remove them.

> > +	alt_reset_restart(qproc, 0);
> > +
> > +	return ret;
> > +}
> > +
> [..]
> > @@ -1179,6 +1251,12 @@ static int q6v5_probe(struct platform_device *pdev)
> >   	qproc->dev = &pdev->dev;
> >   	qproc->rproc = rproc;
> >   	platform_set_drvdata(pdev, qproc);
> > +	qproc->version = desc->version;
> > +
> > +	if (qproc->version == MSS_SDM845)
> > +		qproc->ops = &q6v5_sdm_ops;
> > +	else
> > +		qproc->ops = &q6v5_msm_ops;
>
> Can we carry a boolean for "has_alt_reset" or something that picks the
> new reset ops, rather than picking by version?

Though alt_reset is specific to SDM845 SoCs it also includes another
reset controller (pdc_sync) in the modem bringup sequence, it isn't
included it in the patch series since it doesn't seem to affect the
modem start/stop functionality. Knowing that it may be added wouldn't
version be a better choice here?

> Regards,
> Bjorn

--
Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc, is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project