Re: [PATCH v4 04/13] remoteproc: qcom: q6v5: Use qmp_send to update co-processor load state

[Bjorn Andersson <bjorn.andersson@linaro.org>]

On Mon 19 Jul 23:36 CDT 2021, Sibi Sankar wrote:

> The power domains exposed by the AOSS QMP driver control the load state
> resources linked to modem, adsp, cdsp remoteprocs. These are used to
> notify the Always on Subsystem (AOSS) that a particular co-processor is
> up/down. AOSS uses this information to wait for the co-processors to
> suspend before starting its sleep sequence.
> 
> These co-processors enter low-power modes independent to that of the
> application processor and the load state resources linked to them are
> expected to remain unaltered across system suspend/resume cycles. To
> achieve this behavior lets stop using the power-domains exposed by the
> AOSS QMP node and replace them with generic qmp_send interface instead.
> 
> Signed-off-by: Sibi Sankar <sibis@codeaurora.org>
> ---
> 
> v4:
>  * Drop unused pdev and kfree the load state string in q6v5_deinit
>    /probe path for patch 4. [Matthias]
> 
>  drivers/remoteproc/qcom_q6v5.c      | 57 ++++++++++++++++++++++++-
>  drivers/remoteproc/qcom_q6v5.h      |  7 ++-
>  drivers/remoteproc/qcom_q6v5_adsp.c |  7 ++-
>  drivers/remoteproc/qcom_q6v5_mss.c  | 44 ++++---------------
>  drivers/remoteproc/qcom_q6v5_pas.c  | 85 +++++++++----------------------------
>  drivers/remoteproc/qcom_q6v5_wcss.c |  4 +-
>  6 files changed, 96 insertions(+), 108 deletions(-)
> 
> diff --git a/drivers/remoteproc/qcom_q6v5.c b/drivers/remoteproc/qcom_q6v5.c
> index 7e9244c748da..997ff21271f7 100644
> --- a/drivers/remoteproc/qcom_q6v5.c
> +++ b/drivers/remoteproc/qcom_q6v5.c
> @@ -16,8 +16,28 @@
>  #include "qcom_common.h"
>  #include "qcom_q6v5.h"
>  
> +#define Q6V5_LOAD_STATE_MSG_LEN	64
>  #define Q6V5_PANIC_DELAY_MS	200
>  
> +static int q6v5_load_state_toggle(struct qcom_q6v5 *q6v5, bool enable)
> +{
> +	char buf[Q6V5_LOAD_STATE_MSG_LEN] = {};
> +	int ret;
> +
> +	if (IS_ERR(q6v5->qmp))

I would prefer that you keep it NULL when there's no qmp.

> +		return 0;
> +
> +	snprintf(buf, sizeof(buf),
> +		 "{class: image, res: load_state, name: %s, val: %s}",
> +		 q6v5->load_state, enable ? "on" : "off");
> +
> +	ret = qmp_send(q6v5->qmp, buf, sizeof(buf));
> +	if (ret)
> +		dev_err(q6v5->dev, "failed to toggle load state\n");
> +
> +	return ret;
> +}
> +
>  /**
>   * qcom_q6v5_prepare() - reinitialize the qcom_q6v5 context before start
>   * @q6v5:	reference to qcom_q6v5 context to be reinitialized
> @@ -26,6 +46,12 @@
>   */
>  int qcom_q6v5_prepare(struct qcom_q6v5 *q6v5)
>  {
> +	int ret;
> +
> +	ret = q6v5_load_state_toggle(q6v5, true);
> +	if (ret)
> +		return ret;
> +
>  	reinit_completion(&q6v5->start_done);
>  	reinit_completion(&q6v5->stop_done);
>  
> @@ -47,6 +73,7 @@ EXPORT_SYMBOL_GPL(qcom_q6v5_prepare);
>  int qcom_q6v5_unprepare(struct qcom_q6v5 *q6v5)
>  {
>  	disable_irq(q6v5->handover_irq);
> +	q6v5_load_state_toggle(q6v5, false);
>  
>  	return !q6v5->handover_issued;
>  }
> @@ -196,12 +223,13 @@ EXPORT_SYMBOL_GPL(qcom_q6v5_panic);
>   * @pdev:	platform_device reference for acquiring resources
>   * @rproc:	associated remoteproc instance
>   * @crash_reason: SMEM id for crash reason string, or 0 if none
> + * @load_state: load state resource string
>   * @handover:	function to be called when proxy resources should be released
>   *
>   * Return: 0 on success, negative errno on failure
>   */
>  int qcom_q6v5_init(struct qcom_q6v5 *q6v5, struct platform_device *pdev,
> -		   struct rproc *rproc, int crash_reason,
> +		   struct rproc *rproc, int crash_reason, const char *load_state,
>  		   void (*handover)(struct qcom_q6v5 *q6v5))
>  {
>  	int ret;
> @@ -286,9 +314,36 @@ int qcom_q6v5_init(struct qcom_q6v5 *q6v5, struct platform_device *pdev,
>  		return PTR_ERR(q6v5->state);
>  	}
>  
> +	q6v5->load_state = kstrdup_const(load_state, GFP_KERNEL);

Can't you use devm_kstrdup_const(&pdev->dev, ...) for this?

> +	q6v5->qmp = qmp_get(&pdev->dev);
> +	if (IS_ERR(q6v5->qmp)) {
> +		if (PTR_ERR(q6v5->qmp) != -ENODEV) {
> +			if (PTR_ERR(q6v5->qmp) != -EPROBE_DEFER)
> +				dev_err(&pdev->dev, "failed to acquire load state\n");
> +			kfree_const(q6v5->load_state);

Then you don't need to free it here.

> +			return PTR_ERR(q6v5->qmp);
> +		}
> +	} else {
> +		if (!q6v5->load_state) {
> +			dev_err(&pdev->dev, "load state resource string empty\n");
> +			return -EINVAL;

I see two cases here:

1) kstrdup_const() failed to allocate memory, the error print is
unnecessary and misleading and it would be more appropriate to return
-ENOMEM.

2) kstrdup_const() failed because you passed load_state == NULL, in
which case the error message could be more helpful by saying "unexpected
qcom,qmp property found" or something like that.

> +		}
> +	}
> +
>  	return 0;
>  }
>  EXPORT_SYMBOL_GPL(qcom_q6v5_init);
>  
> +/**
> + * qcom_q6v5_deinit() - deinitialize the q6v5 common struct
> + * @q6v5:	reference to qcom_q6v5 context to be deinitialized
> + */
> +void qcom_q6v5_deinit(struct qcom_q6v5 *q6v5)
> +{
> +	kfree_const(q6v5->load_state);
> +	qmp_put(q6v5->qmp);
> +}
> +EXPORT_SYMBOL_GPL(qcom_q6v5_deinit);
> +
>  MODULE_LICENSE("GPL v2");
>  MODULE_DESCRIPTION("Qualcomm Peripheral Image Loader for Q6V5");
> diff --git a/drivers/remoteproc/qcom_q6v5.h b/drivers/remoteproc/qcom_q6v5.h
> index 1c212f670cbc..f35e04471ed7 100644
> --- a/drivers/remoteproc/qcom_q6v5.h
> +++ b/drivers/remoteproc/qcom_q6v5.h
> @@ -5,6 +5,7 @@
>  
>  #include <linux/kernel.h>
>  #include <linux/completion.h>
> +#include <linux/soc/qcom/qcom_aoss.h>
>  
>  struct rproc;
>  struct qcom_smem_state;
> @@ -15,6 +16,8 @@ struct qcom_q6v5 {
>  	struct rproc *rproc;
>  
>  	struct qcom_smem_state *state;
> +	struct qmp *qmp;
> +
>  	unsigned stop_bit;
>  
>  	int wdog_irq;
> @@ -32,12 +35,14 @@ struct qcom_q6v5 {
>  
>  	bool running;
>  
> +	const char *load_state;
>  	void (*handover)(struct qcom_q6v5 *q6v5);
>  };
>  
>  int qcom_q6v5_init(struct qcom_q6v5 *q6v5, struct platform_device *pdev,
> -		   struct rproc *rproc, int crash_reason,
> +		   struct rproc *rproc, int crash_reason, const char *load_state,
>  		   void (*handover)(struct qcom_q6v5 *q6v5));
> +void qcom_q6v5_deinit(struct qcom_q6v5 *q6v5);
>  
>  int qcom_q6v5_prepare(struct qcom_q6v5 *q6v5);
>  int qcom_q6v5_unprepare(struct qcom_q6v5 *q6v5);
> diff --git a/drivers/remoteproc/qcom_q6v5_adsp.c b/drivers/remoteproc/qcom_q6v5_adsp.c
> index 8b0d8bbacd2e..098362e6e233 100644
> --- a/drivers/remoteproc/qcom_q6v5_adsp.c
> +++ b/drivers/remoteproc/qcom_q6v5_adsp.c
> @@ -185,7 +185,9 @@ static int adsp_start(struct rproc *rproc)
>  	int ret;
>  	unsigned int val;
>  
> -	qcom_q6v5_prepare(&adsp->q6v5);
> +	ret = qcom_q6v5_prepare(&adsp->q6v5);
> +	if (ret)
> +		return ret;

In the (hopefully unlikely) case that we have instances of
qcom_q6v5_prepare() failing today, the switch to the new qmp interface
would also introduce a regression in the same commit.

Could you please add the error handling in a separate commit? Just so
that we can pinpoint which of the two changes caused any issues if we
need to bisect this?

Regards,
Bjorn