[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 11 Nov 2019 18:01:30 +0000
From: Edward Cree <ecree@...arflare.com>
To: Arthur Fabre <afabre@...udflare.com>
CC: Solarflare linux maintainers <linux-net-drivers@...arflare.com>,
"Charles McLachlan" <cmclachlan@...arflare.com>,
Martin Habets <mhabets@...arflare.com>,
David Miller <davem@...emloft.net>,
"Alexei Starovoitov" <ast@...nel.org>,
Daniel Borkmann <daniel@...earbox.net>,
"Jakub Kicinski" <jakub.kicinski@...ronome.com>,
Jesper Dangaard Brouer <hawk@...nel.org>,
John Fastabend <john.fastabend@...il.com>,
netdev <netdev@...r.kernel.org>, bpf <bpf@...r.kernel.org>,
kernel-team <kernel-team@...udflare.com>
Subject: Re: [PATCH net-next] sfc: trace_xdp_exception on XDP failure
On 11/11/2019 17:38, Arthur Fabre wrote:
> On Mon, Nov 11, 2019 at 5:27 PM Edward Cree <ecree@...arflare.com> wrote:
>>
>> On 11/11/2019 10:51, Arthur Fabre wrote:
>>> diff --git a/drivers/net/ethernet/sfc/rx.c b/drivers/net/ethernet/sfc/rx.c
>>> index a7d9841105d8..5bfe1f6112a1 100644
>>> --- a/drivers/net/ethernet/sfc/rx.c
>>> +++ b/drivers/net/ethernet/sfc/rx.c
>>> @@ -678,6 +678,7 @@ static bool efx_do_xdp(struct efx_nic *efx, struct efx_channel *channel,
>>> "XDP is not possible with multiple receive fragments (%d)\n",
>>> channel->rx_pkt_n_frags);
>>> channel->n_rx_xdp_bad_drops++;
>>> + trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
>>> return false;
>>> }
>> AIUI trace_xdp_exception() is improper here as we have not run
>> the XDP program (and xdp_act is thus uninitialised).
>>
>> The other three, below, appear to be correct.
>> -Ed
>>
>
> Good point. Do you know under what conditions we'd end up with
> "fragmented" packets? As far as I can tell this isn't IP
> fragmentation?
Fragments in this case means that the packet data are spread across
multiple RX buffers (~= memory pages). This should only happen if
the RX packet is too big to fit in a single buffer, and when
enabling XDP we ensure that the MTU is small enough to prevent
that. So in theory this can't happen if the NIC is functioning
correctly.
-Ed
Powered by blists - more mailing lists