Hello!
> > > + tp->lost_out -= diff;
> > > + if ((int)tp->lost_out < 0)
> > > + tp->lost_out = 0;
> >
> > These checks aren't necessary.
>
> Are you sure this can't happen if the MSS changes?
Anyway, I think this ca
On Thu, Sep 01, 2005 at 09:44:09PM -0700, David S. Miller wrote:
>
> Ok, how does this look?
Thanks Dave, this one looks perfect.
--
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[EMAIL PROTECTED]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.
From: Herbert Xu <[EMAIL PROTECTED]>
Subject: Re: [PATCHES]: Two TSO refinements
Date: Fri, 2 Sep 2005 14:30:26 +1000
> On Thu, Sep 01, 2005 at 09:28:16PM -0700, David S. Miller wrote:
> >
> > > Therefore,
> > >
> > > tp->lost_out >= diff
>
On Thu, Sep 01, 2005 at 09:28:16PM -0700, David S. Miller wrote:
>
> > Therefore,
> >
> > tp->lost_out >= diff
>
> I assume the same applies to tp->left_out as well?
Yes it does.
Cheers,
--
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[EMAIL PROTECTED]>
Home Pag
From: Herbert Xu <[EMAIL PROTECTED]>
Date: Fri, 2 Sep 2005 11:08:12 +1000
> Yes, because
>
> diff = pcount(orig_skb) - (pcount(skb) + pcount(buff))
><= pcount(orig_skb)
>
> Now if orig_skb is marked as TCPCB_LOST, then by definition
>
> tp->lost_out >= pcount(orig_skb)
>
On Thu, Sep 01, 2005 at 05:53:03PM -0700, David S. Miller wrote:
>
> > > + tp->lost_out -= diff;
> > > + if ((int)tp->lost_out < 0)
> > > + tp->lost_out = 0;
> >
> > These checks aren't necessary.
>
> Are you sure th
From: Herbert Xu <[EMAIL PROTECTED]>
Date: Fri, 2 Sep 2005 08:11:37 +1000
> On Thu, Sep 01, 2005 at 03:06:47PM -0700, David S. Miller wrote:
> >
> > - if (TCP_SKB_CB(buff)->sacked&TCPCB_LOST) {
> > - tp->lost_out += tcp_skb_pcount(buff);
> > - tp->left_out += tcp_skb_pcount(b
Ok, here is the current version of the patch, and what I intend
to push to Linus:
diff-tree 4980a059ef42741e80e9efa0dabdf520f9ba0c5a (from
6b39374a27eb4be7e9d82145ae270ba02ea90dc8)
Author: David S. Miller <[EMAIL PROTECTED]>
Date: Thu Sep 1 15:06:18 2005 -0700
[TCP]: Keep TSO enabled even
On Thu, Sep 01, 2005 at 03:06:47PM -0700, David S. Miller wrote:
>
> - if (TCP_SKB_CB(buff)->sacked&TCPCB_LOST) {
> - tp->lost_out += tcp_skb_pcount(buff);
> - tp->left_out += tcp_skb_pcount(buff);
> + tp->packets_out -= diff;
> + if (diff > 0) {
From: Herbert Xu <[EMAIL PROTECTED]>
Date: Tue, 23 Aug 2005 12:50:01 +1000
> Sorry, I didn't mean to say that pkt_len is an incorrect cut-off point,
> but that it's not a valid value for the MSS. So I'd like to see
>
> if (tcp_fragment(sk, skb, pkt_len, TCP_SKB_CB(skb)->tso_size))
>
On Mon, Aug 22, 2005 at 07:44:59PM -0700, David S. Miller wrote:
>
> It makes perfect sense. In your example, we want a 2*MSS TSO
> frame for the "not-SACKed" area, and the tail 1*MSS frame for
> the SACKed area. This provides exactly the necessary granularity
> in order to record the received SA
From: Herbert Xu <[EMAIL PROTECTED]>
Date: Mon, 22 Aug 2005 22:54:50 +1000
> > + if (pcount > 1 &&
> > + (after(start_seq, TCP_SKB_CB(skb)->seq) ||
> > +before(end_seq, TCP_SKB_CB(skb)->end_seq))) {
> > + uns
From: Herbert Xu <[EMAIL PROTECTED]>
Date: Mon, 22 Aug 2005 13:04:31 +1000
> You need to update various counters likes packets_out since you're
> changing the skb_pcount of a packet that's on the retransmit queue.
Good point.
We would now have two spots which need to do the packets_out
updates,
David S. Miller <[EMAIL PROTECTED]> wrote:
> + if (pcount > 1 &&
> + (after(start_seq, TCP_SKB_CB(skb)->seq) ||
> +before(end_seq, TCP_SKB_CB(skb)->end_seq))) {
> + unsigned int pkt_len;
> +
>
On Sun, Aug 21, 2005 at 08:57:15PM -0700, David S. Miller wrote:
>
> Thinking about this some more... why would I need to use
> anyting other than "SKB"? SKB is where the front part of
> the split data is placed, and skb->next contains the
> subsequent data.
>
> So just using "skb" should be fine
From: Herbert Xu <[EMAIL PROTECTED]>
Date: Mon, 22 Aug 2005 14:21:40 +1000
> > A further simplification seems possible, in that all of the
> > tp->*_out modifications can be keyed upon the "skb_ever_sent"
> > arg to tcp_fragment().
>
> Yes this is a good idea.
>
> > + if (skb_ever_sent) {
>
From: Herbert Xu <[EMAIL PROTECTED]>
Date: Mon, 22 Aug 2005 13:04:31 +1000
> > +
> > + fack_count += pcount;
> >
> >in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&
> >!before(end_seq, TCP_SKB_CB(skb)->end_seq);
>
On Sun, Aug 21, 2005 at 11:29:11PM -0700, David S. Miller wrote:
>
> Ok, this is a full redo of the "do not disable TSO on loss" patch
> including all of the refinements and fixes we've discussed so
> far.
Thanks Dave.
> +extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int
From: "David S. Miller" <[EMAIL PROTECTED]>
Date: Sun, 21 Aug 2005 16:35:06 -0700 (PDT)
> Attached are two patches dealing with two issues I think should be
> addressed in 2.6.13 if possible.
Sorry, I attached the wrong patches, let's try this instead.
diff-tree b94c07bcb646075b82302616c369b1f30
Attached are two patches dealing with two issues I think should be
addressed in 2.6.13 if possible.
The first patch is the TSO deferral logic fix from Dmitry Yusupov. I
think it's correct since even if no new data is tacked on, we will
keep the pipe filled to the end since the percentage-of-wind
20 matches
Mail list logo
