Re: [PATCH] remoteproc: qcom: pas: Add decrypt shutdown support for modem

[Sibi Sankar]

Hey Bjorn,
Thanks for taking time to review the series.

On 7/1/22 12:58 AM, Bjorn Andersson wrote:
> On Fri 20 May 02:28 CDT 2022, Sibi Sankar wrote:
>
> > The initial shutdown request to modem on SM8450 SoCs would start the
> > decryption process and will keep returning errors until the modem shutdown
> > is complete. Fix this by retrying shutdowns in fixed intervals.
> >
> > Err Logs on modem shutdown:
> > qcom_q6v5_pas 4080000.remoteproc: failed to shutdown: -22
> > remoteproc remoteproc3: can't stop rproc: -22
> >
> > Fixes: 5cef9b48458d ("remoteproc: qcom: pas: Add SM8450 remoteproc support")
> > Signed-off-by: Sibi Sankar <quic_sibis@xxxxxxxxxxx>
>
> Looks reasonable, just two inquiries below.
>
> > ---
> >   drivers/remoteproc/qcom_q6v5_pas.c | 67 +++++++++++++++++++++++++++++++++++++-
> >   1 file changed, 66 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/remoteproc/qcom_q6v5_pas.c b/drivers/remoteproc/qcom_q6v5_pas.c
> > index 6ae39c5653b1..d04c4b877e12 100644
> > --- a/drivers/remoteproc/qcom_q6v5_pas.c
> > +++ b/drivers/remoteproc/qcom_q6v5_pas.c
> > @@ -8,6 +8,7 @@
> >    */
> >   
> >   #include <linux/clk.h>
> > +#include <linux/delay.h>
> >   #include <linux/firmware.h>
> >   #include <linux/interrupt.h>
> >   #include <linux/kernel.h>
> > @@ -29,6 +30,8 @@
> >   #include "qcom_q6v5.h"
> >   #include "remoteproc_internal.h"
> >   
> > +#define ADSP_DECRYPT_SHUTDOWN_DELAY_MS	100
> > +
> >   struct adsp_data {
> >   	int crash_reason_smem;
> >   	const char *firmware_name;
> > @@ -36,6 +39,7 @@ struct adsp_data {
> >   	unsigned int minidump_id;
> >   	bool has_aggre2_clk;
> >   	bool auto_boot;
> > +	bool decrypt_shutdown;
> >   
> >   	char **proxy_pd_names;
> >   
> > @@ -65,6 +69,7 @@ struct qcom_adsp {
> >   	unsigned int minidump_id;
> >   	int crash_reason_smem;
> >   	bool has_aggre2_clk;
> > +	bool decrypt_shutdown;
> >   	const char *info_name;
> >   
> >   	struct completion start_done;
> > @@ -128,6 +133,20 @@ static void adsp_pds_disable(struct qcom_adsp *adsp, struct device **pds,
> >   	}
> >   }
> >   
> > +static int adsp_decrypt_shutdown(struct qcom_adsp *adsp)
> > +{
> > +	int retry_num = 50;
>
> Seems unsigned to me.

ack

> > +	int ret = -EINVAL;
> > +
> > +	while (retry_num && ret) {
> > +		msleep(ADSP_DECRYPT_SHUTDOWN_DELAY_MS);
> > +		ret = qcom_scm_pas_shutdown(adsp->pas_id);
> > +		retry_num--;
> > +	}
>
> Will qcom_scm_pas_shutdown() ever return any other errors than -EINVAL?
>
> Would it make sense to make this:
>
> 	do {
> 		...;
> 	} while (ret == -EINVAL && --retry_num);

Just checking on ret would cover the -EINVAL case as well but like you
said pas_shutdown() won't return any other error. So I'll just stick
with your suggestion.

> > +
> > +	return ret;
> > +}
> > +
> >   static int adsp_unprepare(struct rproc *rproc)
> >   {
> >   	struct qcom_adsp *adsp = (struct qcom_adsp *)rproc->priv;
> > @@ -249,6 +268,9 @@ static int adsp_stop(struct rproc *rproc)
> >   		dev_err(adsp->dev, "timed out on wait\n");
> >   
> >   	ret = qcom_scm_pas_shutdown(adsp->pas_id);
> > +	if (ret && adsp->decrypt_shutdown)
> > +		ret = adsp_decrypt_shutdown(adsp);
> > +
> >   	if (ret)
> >   		dev_err(adsp->dev, "failed to shutdown: %d\n", ret);
> >   
> > @@ -459,6 +481,7 @@ static int adsp_probe(struct platform_device *pdev)
> >   	adsp->pas_id = desc->pas_id;
> >   	adsp->has_aggre2_clk = desc->has_aggre2_clk;
> >   	adsp->info_name = desc->sysmon_name;
> > +	adsp->decrypt_shutdown = desc->decrypt_shutdown;
> >   	platform_set_drvdata(pdev, adsp);
> >   
> >   	device_wakeup_enable(adsp->dev);
> > @@ -533,6 +556,7 @@ static const struct adsp_data adsp_resource_init = {
> >   		.pas_id = 1,
> >   		.has_aggre2_clk = false,
> >   		.auto_boot = true,
> > +		.decrypt_shutdown = false,
>
> With all these booleans, I would prefer if we cleaned it up to not list
> the disabled options. That would make it quicker to spot which features
> are actually enabled for each remoteproc.

ack and ack to the adsp_shutdown_poll_decrypt() as well.

> Regards,
> Bjorn