Re: [PATCH] ieee802154: Add trace events for rdev->ops

[Alexander Aring <alex.aring@gmail.com>]

On Sun, Apr 12, 2015 at 02:07:33PM +0200, Guido Günther wrote:
...
> +DECLARE_EVENT_CLASS(802154_le16_template,
> +	TP_PROTO(struct wpan_phy *wpan_phy, struct wpan_dev *wpan_dev,
> +		 __le16 le16arg),
> +	TP_ARGS(wpan_phy, wpan_dev, le16arg),
> +	TP_STRUCT__entry(
> +		WPAN_PHY_ENTRY
> +		WPAN_DEV_ENTRY
> +		__field(__le16, le16arg)
> +	),
> +	TP_fast_assign(
> +		WPAN_PHY_ASSIGN;
> +		WPAN_DEV_ASSIGN;
> +		__entry->le16arg = __le16_to_cpu(le16arg);

When "__entry->le16arg" is type of "__le16" which is given by
"__field(__le16, le16arg)". Then translation of __le16_to_cpu is here
wrong because __le16_to_cpu return u16.

> +	),
> +	TP_printk(WPAN_PHY_PR_FMT ", " WPAN_DEV_PR_FMT ", pan id: 0x%04x",
> +		  WPAN_PHY_PR_ARG, WPAN_DEV_PR_ARG, __entry->le16arg)

When then we need to do the translation here. The format string requires
a host byte-order type. So it's __le16_to_cpu(__entry->le16arg), I don't
know why sparse (the C=2) doesn't report about that.

This is like wireless it does at [0], but with a pointer. When wireless
do that in kernelspace, then I have no idea why there is a cfg80211
plugin which doing byte order translation at [1]. I think this not for
parsing the argument, it's for correct handling of handling the display
of __le16_to_cpup (We use __le16_to_cpu here).

Then we need a cfg802154 plugin in trace_cmd for handling "__le16_to_cpu"
correct.

- Alex

[0] http://lxr.free-electrons.com/source/net/wireless/trace.h?v=3.19#L2160
[1] http://git.kernel.org/cgit/linux/kernel/git/rostedt/trace-cmd.git/tree/plugin_cfg80211.c