Skip to contentSkip to conversation

ipv4: Fix updating SOF_TIMESTAMPING_OPT_ID when SKBTX_HW_TSTAMP is enabled

PATCH In review Stephen Mallon 3 participants 7 messages Apr 28, 2019 – Apr 30, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ipv4: Fix updating SOF_TIMESTAMPING_OPT_ID when SKBTX_HW_TSTAMP is enabled

Ensure that the unique timestamp identifier is incremented for skb hardware timestamps, not just software timestamps.

---
 net/ipv4/ip_output.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/ipv4/ip_output.c +1 -1
index c80188875f39..96ef1d467ba7 100644
@@ -891,7 +891,7 @@ static int __ip_append_data(struct sock *sk,
 	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
 	paged = !!cork->gso_size;
 
-	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
+	if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
 	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
 		tskey = sk->sk_tskey++;
 
Signature · 1 line
2.18.1
#1 lore raw
Quoted 2 lines · Stephen Mallon

Ensure that the unique timestamp identifier is incremented for skb hardware timestamps, not just software timestamps.

Thanks for fixing this. It has been at the back of my mind for a long time, but since I don't use this feature... oh well.

This patch would be a candidate for stable kernels. Can you please add a Fixes: tag?

Thanks,
Richard

#2 lore raw
Quoted 10 lines · Richard Cochran
Quoted 2 lines · Stephen Mallon

Ensure that the unique timestamp identifier is incremented for skb hardware timestamps, not just software timestamps.

Thanks for fixing this. It has been at the back of my mind for a long time, but since I don't use this feature... oh well.

This patch would be a candidate for stable kernels. Can you please add a Fixes: tag?

It is debatable whether this is a fix or a new feature. It extends SOF_TIMESTAMPING_OPT_ID to hardware timestamps. I don't think this would be a stable candidate.

More importantly, note that __ip6_append_data has similar logic. For consistency the two should be updated at the same time.

#3 lore raw
Quoted 3 lines · Willem de Bruijn

It is debatable whether this is a fix or a new feature. It extends SOF_TIMESTAMPING_OPT_ID to hardware timestamps. I don't think this would be a stable candidate.

Was the original series advertised as SW timestamping only?

If so, I missed that at the time. After seeing it not work, I meant to fix it, but never got around to it. So to me this is a known issue.

Quoted 2 lines

More importantly, note that __ip6_append_data has similar logic. For consistency the two should be updated at the same time.

+1

Thanks,
Richard

#4 lore raw
Quoted 7 lines · Richard Cochran
Quoted 3 lines · Willem de Bruijn

It is debatable whether this is a fix or a new feature. It extends SOF_TIMESTAMPING_OPT_ID to hardware timestamps. I don't think this would be a stable candidate.

Was the original series advertised as SW timestamping only?

I did not intend to cover hardware timestamps at the time.

Quoted 3 lines

If so, I missed that at the time. After seeing it not work, I meant to fix it, but never got around to it. So to me this is a known issue.

Understood. I certainly understand that view. I never use hw timestamps, so it is a bit of a blind spot for me. If this is a safe and predictable change, I don't care strongly about net vs net-next. I don't think it meets the bar for stable, but that is not my call.

Quoted 7 lines
Quoted 2 lines

More importantly, note that __ip6_append_data has similar logic. For consistency the two should be updated at the same time.

+1

Thanks,
Richard

#5 lore raw
Quoted 12 lines · Willem de Bruijn
Quoted 7 lines · Richard Cochran
Quoted 3 lines · Willem de Bruijn

It is debatable whether this is a fix or a new feature. It extends SOF_TIMESTAMPING_OPT_ID to hardware timestamps. I don't think this would be a stable candidate.

Was the original series advertised as SW timestamping only?

I did not intend to cover hardware timestamps at the time.

Quoted 8 lines
Quoted 3 lines

If so, I missed that at the time. After seeing it not work, I meant to fix it, but never got around to it. So to me this is a known issue.

Understood. I certainly understand that view. I never use hw timestamps, so it is a bit of a blind spot for me. If this is a safe and predictable change, I don't care strongly about net vs net-next. I don't think it meets the bar for stable, but that is not my call.

I've found that SOF_TIMESTAMPING_OPT_ID already works with hardware timestamps for TCP just not datagram sockets and so I though this was a fix.

Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

More importantly, note that __ip6_append_data has similar logic. For consistency the two should be updated at the same time.

+1

Thanks,
Richard

Thanks for the feedback, I'll update with __ip6_append_data.

#6 lore raw
Quoted 3 lines · Stephen Mallon

I've found that SOF_TIMESTAMPING_OPT_ID already works with hardware timestamps for TCP just not datagram sockets and so I though this was a fix.

Right, so if HW time stamping worked for TCP but not for UDP, then I would call it a regression to be fixed in stable.

Thanks,
Richard