[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANn89iKm5X8V7fMD=oLwBBdX2=JuBv3VNQ5_7-G7yFaENYJrjg@mail.gmail.com>
Date: Tue, 2 Apr 2024 18:23:32 +0200
From: Eric Dumazet <edumazet@...gle.com>
To: shironeko@...aguri.club, Jose Alonso <joalonsof@...il.com>
Cc: "David S . Miller" <davem@...emloft.net>, Jakub Kicinski <kuba@...nel.org>,
Paolo Abeni <pabeni@...hat.com>, netdev@...r.kernel.org,
Soheil Hassas Yeganeh <soheil@...gle.com>, Neal Cardwell <ncardwell@...gle.com>,
Yuchung Cheng <ycheng@...gle.com>, eric.dumazet@...il.com
Subject: Re: [PATCH net-next] tcp: get rid of sysctl_tcp_adv_win_scale
On Tue, Apr 2, 2024 at 5:50 PM Eric Dumazet <edumazet@...gle.com> wrote:
>
> On Tue, Apr 2, 2024 at 5:28 PM <shironeko@...aguri.club> wrote:
> >
> > Hi all,
> >
> > These parts seems to be causing a regression for a specific USB NIC, I
> > have this on one of home server, and it's network will randomly cut out
> > a few times a week.
> > Seems others have ran into the same issue with this particular NIC as
> > well https://bugzilla.kernel.org/show_bug.cgi?id=218536
> >
> > > +/* inverse of tcp_win_from_space() */
> > > +static inline int tcp_space_from_win(const struct sock *sk, int win)
> > > +{
> > > + u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE;
> > > +
> > > + do_div(val, tcp_sk(sk)->scaling_ratio);
> > > + return val;
> > > +}
> > > +
> > > +static inline void tcp_scaling_ratio_init(struct sock *sk)
> > > +{
> > > + /* Assume a conservative default of 1200 bytes of payload per 4K
> > > page.
> > > + * This may be adjusted later in tcp_measure_rcv_mss().
> > > + */
> > > + tcp_sk(sk)->scaling_ratio = (1200 << TCP_RMEM_TO_WIN_SCALE) /
> > > + SKB_TRUESIZE(4096);
> > > }
> > ...
> > > @@ -740,12 +750,7 @@ void tcp_rcv_space_adjust(struct sock *sk)
> > > do_div(grow, tp->rcvq_space.space);
> > > rcvwin += (grow << 1);
> > >
> > > - rcvmem = SKB_TRUESIZE(tp->advmss + MAX_TCP_HEADER);
> > > - while (tcp_win_from_space(sk, rcvmem) < tp->advmss)
> > > - rcvmem += 128;
> > > -
> > > - do_div(rcvwin, tp->advmss);
> > > - rcvbuf = min_t(u64, rcvwin * rcvmem,
> > > + rcvbuf = min_t(u64, tcp_space_from_win(sk, rcvwin),
> > > READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
> > > if (rcvbuf > sk->sk_rcvbuf) {
> > > WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
> >
> > The NIC:
> > usb 2-2: new SuperSpeed USB device number 4 using xhci_hcd
> > usb 2-2: New USB device found, idVendor=0b95, idProduct=1790, bcdDevice=
> > 1.00
> > usb 2-2: New USB device strings: Mfr=1, Product=2, SerialNumber=3
> > usb 2-2: Product: AX88179
> > usb 2-2: Manufacturer: ASIX Elec. Corp.
> > usb 2-2: SerialNumber: 0000000000009D
> > ax88179_178a 2-2:1.0 eth0: register 'ax88179_178a' at
> > usb-0000:00:14.0-2, ASIX AX88179 USB 3.0 Gigabit Ethernet,
> > 02:5e:c0:4b:a4:f7
> > ax88179_178a 2-2:1.0 eth0: ax88179 - Link status is: 1
> >
> > The dmesg error I get:
> >
>
> Thanks for the report. This driver is probably lying on skb->truesize.
>
> This commit looks suspicious
>
> commit f8ebb3ac881b17712e1d5967c97ab1806b16d3d6
> Author: Jose Alonso <joalonsof@...il.com>
> Date: Tue Jun 28 12:13:02 2022 -0300
>
> net: usb: ax88179_178a: Fix packet receiving
Could you try this patch ?
diff --git a/drivers/net/usb/ax88179_178a.c b/drivers/net/usb/ax88179_178a.c
index 88e084534853dd50505fd730e7ccd07c70f2d8ee..ca33365e49cc3993a974ddbdbf68189ce4df2e82
100644
--- a/drivers/net/usb/ax88179_178a.c
+++ b/drivers/net/usb/ax88179_178a.c
@@ -1452,21 +1452,16 @@ static int ax88179_rx_fixup(struct usbnet
*dev, struct sk_buff *skb)
/* Skip IP alignment pseudo header */
skb_pull(skb, 2);
- skb->truesize = SKB_TRUESIZE(pkt_len_plus_padd);
ax88179_rx_checksum(skb, pkt_hdr);
return 1;
}
- ax_skb = skb_clone(skb, GFP_ATOMIC);
+ ax_skb = netdev_alloc_skb_ip_align(dev->net, pkt_len);
if (!ax_skb)
return 0;
- skb_trim(ax_skb, pkt_len);
+ skb_put(ax_skb, pkt_len);
+ memcpy(ax_skb->data, skb->data + 2, pkt_len);
- /* Skip IP alignment pseudo header */
- skb_pull(ax_skb, 2);
-
- skb->truesize = pkt_len_plus_padd +
- SKB_DATA_ALIGN(sizeof(struct sk_buff));
ax88179_rx_checksum(ax_skb, pkt_hdr);
usbnet_skb_return(dev, ax_skb);
Powered by blists - more mailing lists