Re: [PATCH] net: Abstract features usage.

["Michał Mirosław" <mirqus@gmail.com>]

W dniu 24 maja 2011 22:29 użytkownik Mahesh Bandewar
<maheshb@google.com> napisał:
> On Tue, May 24, 2011 at 12:37 PM, Michał Mirosław <mirqus@gmail.com> wrote:
>>
>> 2011/5/24 Mahesh Bandewar <maheshb@google.com>:
>> > Define macros to set/clear/test bits for feature set usage. This will eliminate
>> > the direct use of these fields and enable future ease in managing these fields.
>> >
>> > Signed-off-by: Mahesh Bandewar <maheshb@google.com>
>> > ---
>> >  include/linux/netdev_features.h |  137 +++++++++++++++++++++++++++++++++++++++
>> >  include/linux/netdevice.h       |   35 ++---------
>> >  2 files changed, 142 insertions(+), 30 deletions(-)
>> >  create mode 100644 include/linux/netdev_features.h
>> >
>> > diff --git a/include/linux/netdev_features.h b/include/linux/netdev_features.h
>> > new file mode 100644
>> > index 0000000..97bf8c4
>> > --- /dev/null
>> > +++ b/include/linux/netdev_features.h
>> > @@ -0,0 +1,137 @@
>> > +#ifndef        _NETDEV_FEATURES_H
>> > +#define        _NETDEV_FEATURES_H
>> > +
>> > +/* Forward declarations */
>> > +struct net_device;
>> > +
>> > +typedef        unsigned long *nd_feature_t;
>> > +
>> > +/* Net device feature bits; if you change something,
>> > + * also update netdev_features_strings[] in ethtool.c */
>> > +enum netdev_features {
>> > +       SG_BIT,                 /* Scatter/gather IO. */
>> [...]
>>
>> Please split this change (introducing enum + converting NETIF_F_*
>> defines to use it). This part is a nice cleanup, but I think the
>> bitmap idea is still not worth the trouble until u64 runs out.
> What will we achieve by this split? May be I'm not getting your point
> about split. Do you want to see the cleanup part as a separate patch?
> These enums and NETIF_F_* values have to be in sync.

Exactly. You already did the cleanup (defining enum and NETIF_F_*
#defines), so just split it into separate patch. This is a
prerequisite to any change to the type holding features from u32 to
either u64 or bitmap. And this will separate the cleanup from your
proposed idea.

> The macros that I have defined are expecting this bit numbers rather
> than the flag value.
> Whether it is u64 or bitmap, the changes should be limited to this
> file. Going forward that can be achieved by
> simply changing few lines in these macros / inlines.

One thing I don't like in this implementation is that it leaves old
fields around - so you'll have two code templates in the kernel doing
the same thing (unless you fix all users, but then you won't need old
fields anyway).

Best Regards,
Michał Mirosław