Re: [PATCH] [V3] Add non-Virtex5 support for LL TEMAC driver

[Grant Likely <grant.likely@secretlab.ca>]

On Tue, Apr 6, 2010 at 11:11 AM, John Linn <John.Linn@xilinx.com> wrote:
>> -----Original Message-----
>> From: Eric Dumazet [mailto:eric.dumazet@gmail.com]
>> Sent: Tuesday, April 06, 2010 11:01 AM
>> To: John Linn
>> Cc: netdev@vger.kernel.org; linuxppc-dev@ozlabs.org; grant.likely@secret=
lab.ca;
>> jwboyer@linux.vnet.ibm.com; john.williams@petalogix.com; michal.simek@pe=
talogix.com; John Tyner
>> Subject: RE: [PATCH] [V3] Add non-Virtex5 support for LL TEMAC driver
>>
>> Le mardi 06 avril 2010 =E0 10:12 -0600, John Linn a =E9crit :
>> > > -----Original Message-----
>> > > From: Eric Dumazet [mailto:eric.dumazet@gmail.com]
>> > > Sent: Monday, April 05, 2010 3:30 PM
>> > > To: John Linn
>> > > Cc: netdev@vger.kernel.org; linuxppc-dev@ozlabs.org; grant.likely@se=
cretlab.ca;
>> > > jwboyer@linux.vnet.ibm.com; john.williams@petalogix.com; michal.sime=
k@petalogix.com; John Tyner
>> > > Subject: Re: [PATCH] [V3] Add non-Virtex5 support for LL TEMAC drive=
r
>> > >
>> > > Le lundi 05 avril 2010 =E0 15:11 -0600, John Linn a =E9crit :
>> > > > This patch adds support for using the LL TEMAC Ethernet driver on
>> > > > non-Virtex 5 platforms by adding support for accessing the Soft DM=
A
>> > > > registers as if they were memory mapped instead of solely through =
the
>> > > > DCR's (available on the Virtex 5).
>> > > >
>> > > > The patch also updates the driver so that it runs on the MicroBlaz=
e.
>> > > > The changes were tested on the PowerPC 440, PowerPC 405, and the
>> > > > MicroBlaze platforms.
>> > > >
>> > > > Signed-off-by: John Tyner <jtyner@cs.ucr.edu>
>> > > > Signed-off-by: John Linn <john.linn@xilinx.com>
>> > > >
>> > > > ---
>> > >
>> > > > +/* Align the IP data in the packet on word boundaries as MicroBla=
ze
>> > > > + * needs it.
>> > > > + */
>> > > > +
>> > > > =A0#define XTE_ALIGN =A0 =A0 =A0 32
>> > > > -#define BUFFER_ALIGN(adr) ((XTE_ALIGN - ((u32) adr)) % XTE_ALIGN)
>> > > > +#define BUFFER_ALIGN(adr) ((34 - ((u32) adr)) % XTE_ALIGN)
>> > > >
>> > >
>> > > Very interesting way of doing this, but why such convoluted thing ?
>> >
>> > This is trying to align for a cache line (32 bytes) before my change.
>> >
>> > My change was then also making it align the IP data on a word boundary=
.
>> >
>> > >
>> > > Because of the % 32, this is equivalent to :
>> > >
>> > > #define BUFFER_ALIGN(adr) ((2 - ((u32) adr)) % XTE_ALIGN)
>> > >
>> >
>> > Yes, but I'm not sure that's clearer IMHO.
>> >
>> > > But wait, dont we recognise the magic constant NET_IP_ALIGN ?
>> >
>> > Yes it could be used. =A0I'm struggling with how to make this all be c=
learer.
>> >
>>
>> I am not saying its clearer, I am saying we have a standard way to
>> handle this exact problem (aligning rcvs buffer so that IP header is
>> aligned)
>>
>> There is no need to invent new ones, this makes reviewing of this driver
>> more difficult.

Hold on.... BUFFER_ALIGN is being used to align the DMA buffer on a
cache line boundary.  I don't think netdev_alloc_skb() makes any
guarantees about how the start of the IP header lines up against cache
line boundaries.  The amount of padding needed is not known until an
skbuff is obtained from netdev_alloc_skb(), and
netdev_alloc_skb_ip_align() can only handle a fixed size padding,

It doesn't look like netdev_alloc_skb_ip_align() is the right thing in
this regard.

>> > How about this?
>> > #define BUFFER_ALIGN(adr) (((XTE_ALIGN + NET_IP_ALIGN) - ((u32) adr)) =
% XTE_ALIGN)
>> >
>>
>> Sorry, I still dont understand why you need XTE_ALIGN + ...
>>
>> ((A + B) - C) % A =A0 is equal to (B - C) % A
>>
>> Which one is more readable ?
>
> I'm fine with your suggestion.
>
> #define BUFFER_ALIGN(adr) ((2 - ((u32) adr)) % XTE_ALIGN)
>
>>
>> Please take a look at existing and clean code, no magic macro, and we
>> can understand the intention.
>>
>> find drivers/net | xargs grep -n netdev_alloc_skb_ip_align
>>
>>
>
> Yes I see how it's used, but it only allows you to reserve 2 bytes in the=
 skb with no options.

Eric is here.  The mod operation means that BUFFER_ALIGN using either
2 or 34 is equivalent.

g.