On 06/27/2011 04:43 PM, Rafał Miłecki wrote:

W dniu 27 czerwca 2011 16:24 użytkownik Alexey Dobriyan
[off-list ref]  napisał:

quoted

2011/6/27 Rafał Miłecki[off-list ref]:

quoted

2011/6/26 Geert Uytterhoeven[off-list ref]:

quoted

m68k allmodconfig:

drivers/bcma/main.c: In function ‘bcma_release_core_dev’:
drivers/bcma/main.c:68: error: implicit declaration of function ‘kfree’

We already include slab.h in:
host_pci.c
scan.c
sprom.c

Maybe we can just include this in bcma.h as a better solution?

It isn't better solution.
It results in situation where unnecessary inclusion will be done.
Maybe it's not the case now, but it will be in future.

Scanning code is required for every BCMA board, so we already include
linux/slab.h on every configuration. No matter if this is PCI host
board, or SoC, or whatever we will support in the future.
Now we discovered this is also needed in main.c, which will be always compiled.

That's why I think it's safe to include linux/slab.h in bcma_private.h.
But if that's just my opinion, everybody think it's wrong idea, I'm OK with it.

My rule of thumb is: Header file a.h may only include header b.h when 
a.h needs some definition from b.h. Convenience is never a good reason 
for nested includes.

Gr. AvS

-- 
Almost nobody dances sober, unless they happen to be insane.
-- H.P. Lovecraft --