[ewg] [GIT PULL OFED-3.2] cxgb3/cxgb4/iw_cxgb3/iw_cxgb4backports

[Vipul Pandya]

-----Original Message-----
From: Marciniszyn, Mike [mailto:mike.marciniszyn at intel.com] 
Sent: 22 May 2012 19:03
To: Vipul Pandya
Cc: Kumar A S; Abhishek Agrawal; ewg at lists.openfabrics.org; Divy Le Ray;
Vladimir Sokolovsky
Subject: RE: [ewg] [GIT PULL OFED-3.2]
cxgb3/cxgb4/iw_cxgb3/iw_cxgb4backports

Ok.

I tried to more or less do the same thing as the upstream fix (eliminate
the __netlink* call in favor of alloc_page()) in this specific instance.

The alloc_page() call upstream, however, loses the gfp parameter.   That
is why I changed it here to alloc_pages(gfp, 0) to preserve that aspect
of the current code.

[VIPUL] I just double checked this, alloc_page() call upstream has a gfp
parameter. It doesn't lose it.

Mike

> -----Original Message-----
> From: Vipul Pandya [mailto:vipul at chelsio.com]
> Sent: Tuesday, May 22, 2012 9:01 AM
> To: Marciniszyn, Mike
> Cc: Kumar A S; Abhishek Agrawal; ewg at lists.openfabrics.org; Divy Le
Ray;
> Vladimir Sokolovsky
> Subject: RE: [ewg] [GIT PULL OFED-3.2]
> cxgb3/cxgb4/iw_cxgb3/iw_cxgb4backports
> 
> 
> 
> -----Original Message-----
> From: Marciniszyn, Mike [mailto:mike.marciniszyn at intel.com]
> Sent: 21 May 2012 22:08
> To: Vipul Pandya
> Cc: Kumar A S; Abhishek Agrawal; ewg at lists.openfabrics.org; Divy Le
Ray;
> Vladimir Sokolovsky
> Subject: RE: [ewg] [GIT PULL OFED-3.2]
> cxgb3/cxgb4/iw_cxgb3/iw_cxgb4backports
> 
> Vipul,
> 
> I noticed that this patch gets a build failure for cxgb4 on SLES11SP2
with the
> current compat, compat-rdma, and linux-3.2 trees.
> 
> There is an API difference between SLES11SP2 and 3.0.13 with
> __netdev_alloc_page() taking one more argument:
> 
> SLES11SP2:
> struct page *_netdev_alloc_page(struct net_device, gfp_t, struct
sk_buff
> *)
> 
> 3.0.13:
> struct page *_netdev_alloc_page(struct net_device, gfp_t)
> 
> The cxgb4 driver builds for everything else.
> 
> How do you want to fix this?
> 
> [VIPUL] I have seen in your pull message that you have already fixed
this
> thing. Thank you for this. However I see that it has been fixed in
upstream
> Linux by commit 1f2149c1df50c8c712950872675f46e6e44629f0. This commit
> does couple of additional changes apart from what you have done.
> I would like this to fix the same way. Once your commits are pulled I
will do
> the necessary changes.
> 
> This is the second API different I've seen with SLES11SP2.   The first
> was in the netlink_dump_start() used in core/netlink.c.
> 
> I'm suspecting that a distinct backport branch is needed for
SLES11SP2, which
> I have in a patch to add that to fix the netlink_dump_start()
difference.
> 
> Mike
>