Re: [PATCH net-next 02/12] net: dsa: add Renesas RZ/N1 switch tag driver

[Clément Léger]

Le Thu, 14 Apr 2022 18:11:46 +0300,
Vladimir Oltean <olteanv@xxxxxxxxx> a écrit :

> On Thu, Apr 14, 2022 at 04:35:46PM +0200, Clément Léger wrote:
> > > Please keep variable declarations sorted in decreasing order of line
> > > length (applies throughout the patch series, I won't repeat this comment).  
> > 
> > Acked, both PCS and DSA driver are ok with that rule. Missed that one
> > though.  
> 
> Are you sure? Because a5psw_port_stp_state_set() says otherwise.

Weeeeell, ok let's say I missed these two. Would be useful to have such
checks in checkpatch.pl.

> 
> > > sizeof(tag), to be consistent with the other use of sizeof() above?
> > > Although, hmm, I think you could get away with editing "ptag" in place.  
> > 
> > I was not sure of the alignment guarantee I would have here. If the
> > VLAN header is guaranteed to be aligned on 2 bytes, then I guess it's
> > ok to do that in-place.  
> 
> If I look at Documentation/core-api/unaligned-memory-access.rst
> 
> | Alignment vs. Networking
> | ========================
> | 
> | On architectures that require aligned loads, networking requires that the IP
> | header is aligned on a four-byte boundary to optimise the IP stack. For
> | regular ethernet hardware, the constant NET_IP_ALIGN is used. On most
> | architectures this constant has the value 2 because the normal ethernet
> | header is 14 bytes long, so in order to get proper alignment one needs to
> | DMA to an address which can be expressed as 4*n + 2. One notable exception
> | here is powerpc which defines NET_IP_ALIGN to 0 because DMA to unaligned
> | addresses can be very expensive and dwarf the cost of unaligned loads.
> 
> Your struct a5psw_tag *ptag starts at 10 bytes (8 for tag, 2 for Ethertype)
> before the IP header, so I'm thinking it is aligned at a 2 byte boundary
> as well. A VLAN header between the DSA header and the IP header should
> also not affect that alignment, since its length is 4 bytes.
> 
> If "ctrl_tag" is aligned at a 4 byte boundary - 10, it means "ctrl_data"
> is aligned at a 4 byte boundary - 8, so also a 4 byte boundary.
> 
> But "ctrl_data2" is aligned at a 4 byte boundary + 2, so you might want
> to break up the u32 access into 2 u16 accesses, just to be on the safe
> side?

Thanks for finding these, looks like a good compromise, let's go that
way then.

-- 
Clément Léger,
Embedded Linux and Kernel engineer at Bootlin
https://bootlin.com