[Click] [PATCH] patches for linux 2.6.24

[Adam Greenhalgh]

Joonwoo,

Thats great news ! Well done.

The move to 2.6.24 brings with it the ability to support multiple
receive and transmit queues on newer intel hardware. For example you
might want to modify rx_refill to :-

 int                     (*rx_refill)(struct net_device*, int
ring_index, struct sk_buff**);

but perhaps it would be better to have a new function e.g.

 int                     (*rx_refill_mq)(struct net_device*, struct sk_buff**);

Does anyone have any comments ?

Adam

On 21/04/2008, Joonwoo Park <joonwpark81 at gmail.com> wrote:
> 2008/3/20, Joonwoo Park <joonwpark81 at gmail.com>:
>
> > 2008/2/14, Joonwoo Park <joonwpark81 at gmail.com>:
>  >
>  > > On Mon, Feb 11, 2008 at 05:10:18PM -0000, Egi, Norbert wrote:
>  >  > > >I guess the assertion issue  is releated locking problem, I sent a patch for it.
>  >  > > >Since I'm off for this week, I can't test your patch at this moment.
>  >  > > >I'll dig these things at next week and I'll try NAPI as a possible
>  >  > > >option for e1000 that you patched.
>  >  > >
>  >  > > Joonwoo,
>  >  > >
>  >  > > Yes, it was related to a locking problem when I wanted to have the interfaces in promiscuous mode. It looks so that the dev_set_promiscuity function (in net/core/dev.c) requires the calling thread to hold rtnl mutex by calling rtnl_lock(), so I added this to anydevice.cc with the corresponding rtnl_unlock(). This eliminated the assertion problem I had. So far I haven't experienced any other problems with the patches yet. (By the way, I didn't get the patch you mentioned above.)
>  >  > >
>  >  > > Cheers,
>  >  > > Norbert
>  >  >
>  >  > Hi Norbert,
>  >  > I think you made a indentation mistake.
>  >  > After fixing it, the RTNL_LOCK assertion was disappeared.
>  >  > I'm attaching a mergered patch as click-1.6.0-linux-2.6.24-3.patch.gz as
>  >  > well.
>  >  >
>  >  > ---
>  >
>  >
>  > Hi forks,
>  >  I've updated draft patch for linux 2.6.24.
>  >  - fix spinlock init macro
>  >  - re-enable skb_recycle
>  >  - fix enormous memory leak in skb_recycle
>  >
>  >  Eddie,
>  >  I guess, also the version of 2.6.19's skb_recycle needs
>  >  skb_release_data for fix potential memory leakage.
>  >  Please consider it.
>  >
>
>
> Hi,
>
>  I've overhauled the patchset.
>  There was improvements about support non-x86 arch, backward
>  compatibility for older versions of linux and cleaning up.
>  IMHO, It might be appliable to tree.
>  Eddie, please consider apply this patchset.
>
>  The things changed after the last patch.
>  * Linux-2.6.24-click:
>  1. Initial Linux-2.6.24 patch for Click.
>  2. Fix build failure for !CONFIG_X86_PAE
>  3. Fix incorrect type casting for non-32bit arch
>  3. Fix x86_64 build error
>  4. Fix undefined symbol hweight64
>  5. Fix build error about netfilter
>  6. Use ':' for initializing constant struct member
>  7. Undo unnecessary '::' -> ': :'
>  8. Undo unnecessary '::' -> ': :' #2
>  9. Revert off-topic patch for ipv4/arp
>  10. Undo unnecessary '::' -> ': :' #3
>  11. Revert off-topic patch for hwmon/w83627ehf.c
>
>  * Click-1.6:
>  1. Fix packet::set_network_header
>  2. Fix ToHost set_mac_header
>  3. Fix ToHost set_mac_header
>  4. Improve support 64bit machines
>  5. Improve 64bit machines support #2
>  6. Kill build warning
>  7. Improve Backward compatibility for 2.6.19.2
>  8. Improve Backward compatibility for 2.6.19.2 #2
>  9. Fix regression for build non-32bit arch
>
>  Both of patches contain one-shot patch as well.
>
>  Thanks,
>
> Joonwoo
>
>