| Message ID | e82f7f4d-8d45-1e7c-a2ef-5a8bfc3992c6@gmail.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net] r8169: disable hw csum for short packets and chip versions with hw padding bug | expand |
From: Heiner Kallweit > Sent: 05 November 2020 13:58 > > RTL8125B has same or similar short packet hw padding bug as RTL8168evl. > The main workaround has been extended accordingly, however we have to > disable also hw checksumming for short packets on affected new chip > versions. Change the code in a way that in case of further affected > chip versions we have to add them in one place only. Why not just disable hw checksumming for short packets on all devices (that use this driver). It can't make much difference to the performance. The lack of conditionals may even make it faster. David - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
On 05.11.2020 15:20, David Laight wrote: > From: Heiner Kallweit >> Sent: 05 November 2020 13:58 >> >> RTL8125B has same or similar short packet hw padding bug as RTL8168evl. >> The main workaround has been extended accordingly, however we have to >> disable also hw checksumming for short packets on affected new chip >> versions. Change the code in a way that in case of further affected >> chip versions we have to add them in one place only. > > Why not just disable hw checksumming for short packets on > all devices (that use this driver). > Thanks for the hint. Briefly thought about that too but then decided against it as I don't have performance figures. But if everybody is fine with it, then I'd be happy to simplify short packet handling in the described way (in a v2). > It can't make much difference to the performance. > The lack of conditionals may even make it faster. > > David > > - > Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK > Registration No: 1397386 (Wales) > Heiner
diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c index c5d5c1cfc..56f84b597 100644 --- a/drivers/net/ethernet/realtek/r8169_main.c +++ b/drivers/net/ethernet/realtek/r8169_main.c @@ -4315,11 +4315,13 @@ static netdev_features_t rtl8169_features_check(struct sk_buff *skb, features &= ~NETIF_F_ALL_TSO; } else if (skb->ip_summed == CHECKSUM_PARTIAL) { if (skb->len < ETH_ZLEN) { + if (rtl_test_hw_pad_bug(tp)) + features &= ~NETIF_F_CSUM_MASK; + switch (tp->mac_version) { case RTL_GIGA_MAC_VER_11: case RTL_GIGA_MAC_VER_12: case RTL_GIGA_MAC_VER_17: - case RTL_GIGA_MAC_VER_34: features &= ~NETIF_F_CSUM_MASK; break; default:
RTL8125B has same or similar short packet hw padding bug as RTL8168evl. The main workaround has been extended accordingly, however we have to disable also hw checksumming for short packets on affected new chip versions. Change the code in a way that in case of further affected chip versions we have to add them in one place only. Fixes: 0439297be951 ("r8169: add support for RTL8125B") Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com> --- drivers/net/ethernet/realtek/r8169_main.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)