On Sun, Mar 09, 2025 at 06:46:48AM -0400, Faizal Rahim wrote:
> +/**
> + * igc_ethtool_get_frame_ass_error - Get the frame assembly error count.
> + * @reg_value: Register value for IGC_PRMEXCPRCNT
> + * Return: The count of frame assembly errors.
> + */
> +static u64 igc_ethtool_get_frame_ass_error(u32 reg_value)
> +{
> + /* Out of order statistics */
> + u32 ooo_frame_cnt, ooo_frag_cnt;
> + u32 miss_frame_frag_cnt;
> +
> + ooo_frame_cnt = FIELD_GET(IGC_PRMEXCPRCNT_OOO_FRAME_CNT, reg_value);
> + ooo_frag_cnt = FIELD_GET(IGC_PRMEXCPRCNT_OOO_FRAG_CNT, reg_value);
> + miss_frame_frag_cnt = FIELD_GET(IGC_PRMEXCPRCNT_MISS_FRAME_FRAG_CNT, reg_value);
> +
> + return ooo_frame_cnt + ooo_frag_cnt + miss_frame_frag_cnt;
> +}
Counters should be monotonically increasing 64-bit values, and as per your
response in v8, these are 8-bit saturating, clear-on-read values, which
I have some doubts about. But I don't think there's anything that
software can do to make that situation any better. Even reading them
periodically and integrating them into a software sum risks losing
events due to the saturating behavior, which is confusing in its own
way. So I'll have to tolerate this.
Reviewed-by: Vladimir Oltean <vladimir.oltean@xxxxxxx>
