Re: [PATCH 21/22] xlink-core: add async channel and events

[Joe Perches <joe@perches.com>]

On Tue, 2020-12-01 at 14:35 -0800, mgross@linux.intel.com wrote:
> Enable asynchronous channel and event communication.
[]
> diff --git a/drivers/misc/xlink-core/xlink-core.c b/drivers/misc/xlink-core/xlink-core.c
[]
> +
> +// sysfs attribute functions
> +
> +static ssize_t event0_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct keembay_xlink_dev *xlink_dev = dev_get_drvdata(dev);
> +	struct xlink_attr *a = &xlink_dev->eventx[0];
> +
> +	return scnprintf(buf, PAGE_SIZE, "0x%x 0x%lx\n", a->sw_dev_id, a->value);
> +}
> +
> +static ssize_t event1_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct keembay_xlink_dev *xlink_dev = dev_get_drvdata(dev);
> +	struct xlink_attr *a = &xlink_dev->eventx[1];
> +
> +	return scnprintf(buf, PAGE_SIZE, "0x%x 0x%lx\n", a->sw_dev_id, a->value);
> +}
> +
> +static ssize_t event2_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct keembay_xlink_dev *xlink_dev = dev_get_drvdata(dev);
> +	struct xlink_attr *a = &xlink_dev->eventx[2];
> +
> +	return scnprintf(buf, PAGE_SIZE, "0x%x 0x%lx\n", a->sw_dev_id, a->value);
> +}
> +
> +static ssize_t event3_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct keembay_xlink_dev *xlink_dev = dev_get_drvdata(dev);
> +	struct xlink_attr *a = &xlink_dev->eventx[3];
> +
> +	return scnprintf(buf, PAGE_SIZE, "0x%x 0x%lx\n", a->sw_dev_id, a->value);
> +}

These could use the sysfs_emit function and not scnprintf
and as well could use a single shared function with an index.

Something like:

static ssize_t eventx_show(struct device *dev, struct device_attr *attr, int index, char *buf)
{
	struct keembay_xlink_dev *xlink_dev = dev_get_drvdata(dev);
	struct xlink_attr *a = &xlink_dev->eventx[index];

	return sysfs_emit(buf, "0x%x 0x%lx\n", a->sw_dev_id, a->value);
}

static ssize_t event0_show(struct device *dev, struct device_attribute *attr, char *buf)
{
	return eventx_show(dev, attr, 0, buf);
}

etc...


> @@ -270,19 +353,23 @@ static long xlink_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
>  	struct xlinkconnect		con	= {0};
>  	struct xlinkrelease		rel	= {0};
>  	struct xlinkstartvpu		startvpu = {0};
> +	struct xlinkcallback		cb	= {0};
>  	struct xlinkgetdevicename	devn	= {0};
>  	struct xlinkgetdevicelist	devl	= {0};
>  	struct xlinkgetdevicestatus	devs	= {0};
>  	struct xlinkbootdevice		boot	= {0};
>  	struct xlinkresetdevice		res	= {0};
>  	struct xlinkdevmode		devm	= {0};
> +	struct xlinkregdevevent		regdevevent = {0};

Perhaps significantly less stack would be used if these declarations were
moved into the case blocks where used.  Less initialization would be done
per ioctl too.

> @@ -318,6 +405,30 @@ static long xlink_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
>  		if (copy_to_user((void __user *)op.return_code, (void *)&rc, sizeof(rc)))
>  			return -EFAULT;
>  		break;
> +	case XL_DATA_READY_CALLBACK:
> +		if (copy_from_user(&cb, (void __user *)arg,
> +				   sizeof(struct xlinkcallback)))
> +			return -EFAULT;
> +		if (copy_from_user(&devh, (void __user *)cb.handle,
> +				   sizeof(struct xlink_handle)))
> +			return -EFAULT;
> +		CHANNEL_SET_USER_BIT(cb.chan); // set MSbit for user space call
> +		rc = xlink_data_available_event(&devh, cb.chan, cb.callback);
> +		if (copy_to_user((void __user *)cb.return_code, (void *)&rc, sizeof(rc)))
> +			return -EFAULT;
> +		break;

	case XL_DATA_READY_CALLBACK: {
		struct xlinkcallback cb	= {};
		etc...
		break;
	}