Re: [PATCH] new kqueue API v.08

[Stefani Seibold <stefani@seibold.net>]

Am Samstag, den 19.12.2009, 12:36 +0100 schrieb Andi Kleen:
> I like the basic idea of a type safe FIFO.
> 
> > #define	DYNAMIC
> > #ifdef DYNAMIC
> > static DECLARE_KFIFO_PTR(test[1], int);
> > #else
> > static DECLARE_KFIFO(test[1], int, FIFO_SIZE);
> 
> The [1] looks weird.  Is that really needed and what does it mean?
> The callers below don't seem to use it like an array.

I am a lazy girl. This is only for convenient, because i don't want to
write always kfifo_....(&test...). Using an array of [1] provide the
pointer automaticly. Of course you can also write

static DECLARE_KFIFO(test, int, FIFO_SIZE);

and then call the kfifo macros with the address of the variable.
 
> 
> > I know that this kind of macros are very sophisticated and not easy to
> > maintain. But i have all tested and it works as expected. I analyzed the
> > output of the compiler and for the x86 the code is as good as hand
> > written assembler code.
> 
> Linux has a long tradition of complicated macros in headers, that shouldn't be a 
> problem.
> 

I know, but this are in my opinion the most complicated macros for
linux.

> >  include/linux/kfifo.h | 1107 +++++++++++++++++++++++++++++---------------------
> >  kernel/kfifo.c        |  768 +++++++++++++++++++++++-----------
> >  2 files changed, 1174 insertions(+), 701 deletions(-)
> > 
> > diff -u -N -r -p mmotm.orig/include/linux/kfifo.h mmotm.new/include/linux/kfifo.h
> > --- mmotm.orig/include/linux/kfifo.h	2009-12-19 00:23:12.510334931 +0100
> > +++ mmotm.new/include/linux/kfifo.h	2009-12-19 00:23:04.375307229 +0100
> > @@ -1,8 +1,7 @@
> >  /*
> > - * A generic kernel FIFO implementation.
> > + * A generic kernel fifo implementation
> >   *
> >   * Copyright (C) 2009 Stefani Seibold <stefani@seibold.net>
> > - * Copyright (C) 2004 Stelian Pop <stelian@popies.net>
> 
> You should probably keep the old copyright, even if not much code remains.
> 

No, all previous code has gone. I use the counter fifo technique since
many years, so the basic idea is very old. 
 
> > +#ifdef __KERNEL__
> >  #include <linux/kernel.h>
> >  #include <linux/spinlock.h>
> > +#include <linux/stddef.h>
> > +#include <linux/scatterlist.h>
> > +#else
> > +#include "helper.h"
> > +#endif
> 
> Such ifdefs should not make it into submitted code. Better use more glue
> in the test program.

Will be fixed after review.

> 
> 
> ... didn't review the whole thing at this point ...

Would be great if you can do this.

Stefani