Re: [net-next] net: airoha: Use u64_stats_t with u64_stats_sync properly

[Jakub Kicinski <kuba@kernel.org>]

This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.

For local reproduction steps see https://netdev-ai.bots.linux.dev/ai-local.html
---
net: airoha: Use u64_stats_t with u64_stats_sync properly

This commit converts statistics fields from u64 to u64_stats_t and changes
direct arithmetic operations to u64_stats_add()/u64_stats_read() calls
to ensure atomic operations on 64-bit architectures.

> diff --git a/drivers/net/ethernet/airoha/airoha_eth.c b/drivers/net/ethernet/airoha/airoha_eth.c
> index 62bcbbbe2a95..6ed220e5a094 100644
> --- a/drivers/net/ethernet/airoha/airoha_eth.c
> +++ b/drivers/net/ethernet/airoha/airoha_eth.c
> @@ -1472,131 +1472,131 @@ static void airoha_update_hw_stats(struct airoha_gdm_port *port)
>
>  	/* TX */
>  	val = airoha_fe_rr(eth, REG_FE_GDM_TX_OK_PKT_CNT_H(port->id));
> -	port->stats.tx_ok_pkts += ((u64)val << 32);
> +	u64_stats_add(&port->stats.tx_ok_pkts, (u64)val << 32);

On 32-bit systems (ARCH_AIROHA is supported on ARM32 per
arch/arm/Kconfig.platforms), does this lose the upper 32 bits of the
counter?

The u64_stats_add() signature is:

  static inline void u64_stats_add(u64_stats_t *p, unsigned long val)

On 32-bit architectures, unsigned long is 32 bits. When (u64)val << 32
is passed, the resulting 64-bit value is truncated to unsigned long.
Since all the non-zero bits are in the upper half, the truncation yields
zero.

The original code worked because the destination was u64:

  port->stats.tx_ok_pkts += ((u64)val << 32);

This pattern appears 16 times in airoha_update_hw_stats() for all the
_CNT_H registers (tx_ok_pkts, tx_ok_bytes, tx_len[], rx_ok_pkts,
rx_ok_bytes, rx_len[]). On 32-bit builds, these counters would only
accumulate the low 32 bits from the hardware.

[ ... ]
-- 
pw-bot: cr