> +static uint64_t
> +stats_get_diff(uint64_t stats, uint64_t offset)
> +{
> + if (stats >= offset)
> + return stats - offset;
> + /* unlikely wraparound case */
> + return UINT64_MAX + stats - offset;
The numbers are unsigned, so wrapping comes for free.
Remove the comparison and always return stats - offset.
Using uint8_t for easier explanation, if offset is 255 and stats is 0, then the
diff should be 1.
Returning stats - offset:
stats - offset = 0 - 255 = 0 - 0xFF = 1.
Returning UINT8_MAX + stats - offset is wrong:
UINT8_MAX + stats - offset = 255 - 0 - 255 = 0.
Besides that, it looks good to me.
While reviewing, I came across the rx_mbuf_alloc_failed counter in the
rte_eth_dev_data structure:
https://elixir.bootlin.com/dpdk/v24.03/source/lib/ethdev/rte_ethdev.c#L3145
https://elixir.bootlin.com/dpdk/v24.03/source/lib/ethdev/ethdev_driver.h#L127
Doesn't it have the same problem?
BTW, the af_packet PMD doesn't increase the rx_mbuf_alloc_failed counter on
mbuf allocation failures. But that's a separate bug.
