Hi Sai,
On Tue, May 19, 2020 at 07:06:02PM +0530, Sai Prakash Ranjan wrote:
> On some QCOM SoCs, replicators in Always-On domain loses its
> context as soon as the clock is disabled. Currently as a part
> of pm_runtime workqueue, clock is disabled after the replicator
> is initialized by amba_pm_runtime_suspend assuming that context
> is not lost which is not true for replicators with such
> limitations. So add a new property "qcom,replicator-loses-context"
> to identify such replicators and reset them.
>
> Suggested-by: Mike Leach <mike.leach@xxxxxxxxxx>
> Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@xxxxxxxxxxxxxx>
> ---
>
> Added Mike's suggested by for parts other than the DT property.
> Perhaps I should add Co-developed-by Mike since the full skeletal
> was given by Mike. I can add that if required on the next version.
I will let Mike decide what he wants to do - I'm fine either way.
>
> ---
> .../coresight/coresight-replicator.c | 53 +++++++++++++------
> 1 file changed, 36 insertions(+), 17 deletions(-)
>
> diff --git a/drivers/hwtracing/coresight/coresight-replicator.c b/drivers/hwtracing/coresight/coresight-replicator.c
> index c619b456f55a..ba66160c8140 100644
> --- a/drivers/hwtracing/coresight/coresight-replicator.c
> +++ b/drivers/hwtracing/coresight/coresight-replicator.c
> @@ -38,6 +38,7 @@ struct replicator_drvdata {
> struct clk *atclk;
> struct coresight_device *csdev;
> spinlock_t spinlock;
> + bool check_idfilter_val;
Please add documentation for the new field, the same way other fields are
documented.
> };
>
> static void dynamic_replicator_reset(struct replicator_drvdata *drvdata)
> @@ -66,29 +67,43 @@ static int dynamic_replicator_enable(struct replicator_drvdata *drvdata,
> int inport, int outport)
> {
> int rc = 0;
> - u32 reg;
> -
> - switch (outport) {
> - case 0:
> - reg = REPLICATOR_IDFILTER0;
> - break;
> - case 1:
> - reg = REPLICATOR_IDFILTER1;
> - break;
> - default:
> - WARN_ON(1);
> - return -EINVAL;
> - }
> + u32 id0val, id1val;
>
> CS_UNLOCK(drvdata->base);
>
> - if ((readl_relaxed(drvdata->base + REPLICATOR_IDFILTER0) == 0xff) &&
> - (readl_relaxed(drvdata->base + REPLICATOR_IDFILTER1) == 0xff))
> + id0val = readl_relaxed(drvdata->base + REPLICATOR_IDFILTER0);
> + id1val = readl_relaxed(drvdata->base + REPLICATOR_IDFILTER1);
> +
> + /*
> + * Some replicator designs lose context when AMBA clocks are removed,
> + * so have a check for this.
> + */
> + if (drvdata->check_idfilter_val && id0val == 0x0 && id1val == 0x0)
> + id0val = id1val = 0xff;
> +
> + if (id0val == 0xff && id1val == 0xff)
> rc = coresight_claim_device_unlocked(drvdata->base);
>
> + if (!rc) {
> + switch (outport) {
> + case 0:
> + id0val = 0x0;
> + break;
> + case 1:
> + id1val = 0x0;
> + break;
> + default:
> + WARN_ON(1);
> + rc = -EINVAL;
> + }
> + }
> +
> /* Ensure that the outport is enabled. */
> - if (!rc)
> - writel_relaxed(0x00, drvdata->base + reg);
> + if (!rc) {
> + writel_relaxed(id0val, drvdata->base + REPLICATOR_IDFILTER0);
> + writel_relaxed(id1val, drvdata->base + REPLICATOR_IDFILTER1);
> + }
> +
> CS_LOCK(drvdata->base);
>
> return rc;
> @@ -239,6 +254,10 @@ static int replicator_probe(struct device *dev, struct resource *res)
> desc.groups = replicator_groups;
> }
>
> + if (fwnode_property_present(dev_fwnode(dev),
> + "qcom,replicator-loses-context"))
> + drvdata->check_idfilter_val = true;
> +
The header <linux/property.h> needs to be added for function
fwnode_property_present().
What is the clock situation with other QC components like funnels? Have they
also been designed the same way? If so the binding should probably be
"qcom,component-loses-context", otherwise what you have suggested will work just
fine. My goal here is to avoid having "qcom,replicator-loses-context" and
"qcom,funnel-loses-context".
Lastly, I have applied patch 1 and 2 of this set to my tree so no need to resend
them again with the next revision.
Thanks,
Mathieu
> dev_set_drvdata(dev, drvdata);
>
> pdata = coresight_get_platform_data(dev);
> --
> QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
> of Code Aurora Forum, hosted by The Linux Foundation
>
