On Fri, Nov 9, 2018 at 6:02 PM Yunsheng Lin <linyunsh...@huawei.com> wrote:
>
> On 2018/11/10 9:42, Cong Wang wrote:
> > On Fri, Nov 9, 2018 at 5:39 PM Yunsheng Lin <linyunsh...@huawei.com> wrote:
> >>
> >> On 2018/11/10 3:43, Cong Wang wrote:
> >>> Currently netdev_rx_csum_fault() only shows a device name,
> >>> we need more information about the skb for debugging.
> >>>
> >>> Sample output:
> >>>
> >>> ens3: hw csum failure
> >>> dev features: 0x0000000000014b89
> >>> skb len=84 data_len=0 gso_size=0 gso_type=0 ip_summed=0 csum=0,
> >>> csum_complete_sw=0, csum_valid=0
> >>>
> >>> Signed-off-by: Cong Wang <xiyou.wangc...@gmail.com>
> >>> ---
> >>> include/linux/netdevice.h | 5 +++--
> >>> net/core/datagram.c | 6 +++---
> >>> net/core/dev.c | 10 ++++++++--
> >>> net/sunrpc/socklib.c | 2 +-
> >>> 4 files changed, 15 insertions(+), 8 deletions(-)
> >>>
> >>> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> >>> index 857f8abf7b91..fabcd9fa6cf7 100644
> >>> --- a/include/linux/netdevice.h
> >>> +++ b/include/linux/netdevice.h
> >>> @@ -4332,9 +4332,10 @@ static inline bool
> >>> can_checksum_protocol(netdev_features_t features,
> >>> }
> >>>
> >>> #ifdef CONFIG_BUG
> >>> -void netdev_rx_csum_fault(struct net_device *dev);
> >>> +void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb);
> >>> #else
> >>> -static inline void netdev_rx_csum_fault(struct net_device *dev)
> >>> +static inline void netdev_rx_csum_fault(struct net_device *dev,
> >>> + struct sk_buff *skb)
> >>> {
> >>> }
> >>> #endif
> >>> diff --git a/net/core/datagram.c b/net/core/datagram.c
> >>> index 57f3a6fcfc1e..d8f4d55cd6c5 100644
> >>> --- a/net/core/datagram.c
> >>> +++ b/net/core/datagram.c
> >>> @@ -736,7 +736,7 @@ __sum16 __skb_checksum_complete_head(struct sk_buff
> >>> *skb, int len)
> >>> if (likely(!sum)) {
> >>> if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
> >>> !skb->csum_complete_sw)
> >>> - netdev_rx_csum_fault(skb->dev);
> >>> + netdev_rx_csum_fault(skb->dev, skb);
> >>> }
> >>> if (!skb_shared(skb))
> >>> skb->csum_valid = !sum;
> >>> @@ -756,7 +756,7 @@ __sum16 __skb_checksum_complete(struct sk_buff *skb)
> >>> if (likely(!sum)) {
> >>> if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
> >>> !skb->csum_complete_sw)
> >>> - netdev_rx_csum_fault(skb->dev);
> >>> + netdev_rx_csum_fault(skb->dev, skb);
> >>> }
> >>>
> >>> if (!skb_shared(skb)) {
> >>> @@ -810,7 +810,7 @@ int skb_copy_and_csum_datagram_msg(struct sk_buff
> >>> *skb,
> >>>
> >>> if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
> >>> !skb->csum_complete_sw)
> >>> - netdev_rx_csum_fault(NULL);
> >>> + netdev_rx_csum_fault(NULL, skb);
> >>> }
> >>> return 0;
> >>> fault:
> >>> diff --git a/net/core/dev.c b/net/core/dev.c
> >>> index 0ffcbdd55fa9..2b337df26117 100644
> >>> --- a/net/core/dev.c
> >>> +++ b/net/core/dev.c
> >>> @@ -3091,10 +3091,16 @@ EXPORT_SYMBOL(__skb_gso_segment);
> >>>
> >>> /* Take action when hardware reception checksum errors are detected. */
> >>> #ifdef CONFIG_BUG
> >>> -void netdev_rx_csum_fault(struct net_device *dev)
> >>> +void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
> >>> {
> >>> if (net_ratelimit()) {
> >>> pr_err("%s: hw csum failure\n", dev ? dev->name :
> >>> "<unknown>");
> >>> + if (dev)
> >>> + pr_err("dev features: %pNF\n", &dev->features);
> >>> + pr_err("skb len=%d data_len=%d gso_size=%d gso_type=%d
> >>> ip_summed=%d csum=%x, csum_complete_sw=%d, csum_valid=%d\n",
> >>> + skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
> >>> + skb_shinfo(skb)->gso_type, skb->ip_summed, skb->csum,
> >>> + skb->csum_complete_sw, skb->csum_valid);
> >>
> >>
> >> This function also have the netdev available, use netdev_err to log the
> >> error?
> >
> > It is apparently not me who picked pr_err() from the beginning,
> > I just follow that pr_err(). If you are not happy with it, please send
> > a followup.
>
> Yes, but perhaps it is something to improve.
Sure, no one stops you from improving it in a followup patch. :)
> When using the netdev, then maybe it does not have to check if dev is null,
> because
> netdev_err has handled the netdev being NULL case.
> Maybe I missed something that netdev can not be used here?
> If not, maybe I can send a followup.
>
Maybe. Again, my patch intends to add a few debugging logs,
not to convert pr_err() to whatever else, they are totally different
goals. I choose pr_err() only because I follow the existing one,
not to say which one is better than the other.
Thanks.
