[PATCH net-next, v2, 1/3] hv_netvsc: Add support for LRO/RSC in the vSwitch

Stephen Hemminger stephen at networkplumber.org
Fri Sep 21 20:22:06 UTC 2018


On Fri, 21 Sep 2018 18:51:54 +0000
Haiyang Zhang <haiyangz at microsoft.com> wrote:

> > -----Original Message-----
> > From: Stephen Hemminger <stephen at networkplumber.org>
> > Sent: Friday, September 21, 2018 2:37 PM
> > To: Haiyang Zhang <haiyangz at linuxonhyperv.com>
> > Cc: Haiyang Zhang <haiyangz at microsoft.com>; davem at davemloft.net;
> > netdev at vger.kernel.org; olaf at aepfle.de; linux-kernel at vger.kernel.org;
> > devel at linuxdriverproject.org; vkuznets <vkuznets at redhat.com>
> > Subject: Re: [PATCH net-next, v2, 1/3] hv_netvsc: Add support for LRO/RSC in
> > the vSwitch
> > 
> > On Fri, 21 Sep 2018 18:20:35 +0000
> > Haiyang Zhang <haiyangz at linuxonhyperv.com> wrote:
> > 
> > Overall, this looks good. Some minor suggestions.
> >   
> > > +struct nvsc_rsc {
> > > +	const struct ndis_pkt_8021q_info *vlan;
> > > +	const struct ndis_tcp_ip_checksum_info *csum_info;
> > > +	u8 is_last; /* last RNDIS msg in a vmtransfer_page */
> > > +	u32 cnt; /* #fragments in an RSC packet */
> > > +	u32 pktlen; /* Full packet length */
> > > +	void *data[NVSP_RSC_MAX];
> > > +	u32 len[NVSP_RSC_MAX];
> > > +};
> > > +  
> > 
> > This new state structure is state on a per-channel basis.
> > Do you really need this to be persistent across packets?
> > 
> > Could this be on stack or do you need it to handle split packets arriving in
> > different polls? Or is the stack space a problem?
> > 
> > Also, maybe data and length could be in one structure since they are related.  
> 
> The stack space is a problem. NVSP_RSC_MAX is 562, which is defined by host.
> It will be too large for limited stack space. 
> 
> struct nvsc_rsc includes the data, len, cnt, chksum, vlan for one RSC packet. They
> are all related to construction of one SKB and its meta data. So I put them in
> one structure.
> 
> Thanks,
> - Haiyang
> 

That makes sense. How big is sizeof(struct net_device) + netdev_priv now?
Need to make sure it doesn't become an order 2 (ie keep it less than 4K).


More information about the devel mailing list