[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20201119225842.GK15137@breakpoint.cc>
Date: Thu, 19 Nov 2020 23:58:42 +0100
From: Florian Westphal <fw@...len.de>
To: "Ramsay, Lincoln" <Lincoln.Ramsay@...i.com>
Cc: Florian Westphal <fw@...len.de>,
Igor Russkikh <irusskikh@...vell.com>,
"David S. Miller" <davem@...emloft.net>,
Jakub Kicinski <kuba@...nel.org>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
Dmitry Bogdanov <dbogdanov@...vell.com>
Subject: Re: [PATCH v3] aquantia: Remove the build_skb path
Ramsay, Lincoln <Lincoln.Ramsay@...i.com> wrote:
> When performing IPv6 forwarding, there is an expectation that SKBs
> will have some headroom. When forwarding a packet from the aquantia
> driver, this does not always happen, triggering a kernel warning.
>
> The build_skb path fails to allow for an SKB header, but the hardware
For build_skb path to work the buffer scheme would need to be changed
to reserve headroom, so yes, I think that the proposed patch is the
most convenient solution.
> buffer it is built around won't allow for this anyway. Just always use the
> slower codepath that copies memory into an allocated SKB.
I thought this changes the driver to always copy the entire packet, but
thats not true, see below.
> It seems that skb_headroom is only 14, when it is expected to be >= 16.
Yes, kernel expects to have some headroom in skbs.
> aq_ring.c has this code (edited slightly for brevity):
>
> if (buff->is_eop && buff->len <= AQ_CFG_RX_FRAME_MAX - AQ_SKB_ALIGN) {
> skb = build_skb(aq_buf_vaddr(&buff->rxdata), AQ_CFG_RX_FRAME_MAX);
> skb_put(skb, buff->len);
> } else {
> skb = napi_alloc_skb(napi, AQ_CFG_RX_HDR_SIZE);
>
> There is a significant difference between the SKB produced by these 2 code paths. When napi_alloc_skb creates an SKB, there is a certain amount of headroom reserved. The same pattern appears to be used in all of the other ethernet drivers I have looked at. However, this is not done in the build_skb codepath.
I think the above should be part of the commit message rather than this
meta-space (which gets removed by git-am).
> + skb = napi_alloc_skb(napi, AQ_CFG_RX_HDR_SIZE);
> + if (unlikely(!skb)) {
AQ_CFG_RX_HDR_SIZE is 256 byte, so for larger packets ...
> + memcpy(__skb_put(skb, hdr_len), aq_buf_vaddr(&buff->rxdata),
> + ALIGN(hdr_len, sizeof(long)));
This only copies the initial part and then...
> + if (buff->len - hdr_len > 0) {
> + skb_add_rx_frag(skb, 0, buff->rxdata.page,
> + buff->rxdata.pg_off + hdr_len,
> + buff->len - hdr_len,
> AQ_CFG_RX_FRAME_MAX);
The rest is added as a frag.
IOW, this patch looks good to me, but could you update the
commit message so it becomes clear that this doesn't result in a full
copy?
Perhaps something like:
'Just always use the napi_alloc_skb() code path that passes the buffer
as a page fragment', or similar.
Thanks.
Powered by blists - more mailing lists