Re: [PATCH] [2.5] include/asm-generic/bitops.h {set,clear}_bit return void

Re: [PATCH] [2.5] include/asm-generic/bitops.h {set,clear}_bit return void

[Arnd Bergmann]

Carl-Daniel Hailfinger wrote:

> +     mask = 1 << (nr & 0x1f);
> +     cli();
> +     *addr |= mask;
> +     sti();

cli() and sti() are no more. Moreover, the file you are trying to fix is
not even used anywhere. Better submit a patch to remove it completely.

        Arnd <><

Re: [PATCH] [2.5] include/asm-generic/bitops.h {set,clear}_bit return void

[Carl-Daniel Hailfinger]

Arnd Bergmann wrote:
> Carl-Daniel Hailfinger wrote:
> 
> 
>>+     mask = 1 << (nr & 0x1f);
>>+     cli();
>>+     *addr |= mask;
>>+     sti();
> 
> 
> cli() and sti() are no more. Moreover, the file you are trying to fix is

What is the preferred way to achieve atomicity in an operation now that
cli() and sti() are gone?

> not even used anywhere. Better submit a patch to remove it completely.

The point of asm-generic is not to use the files, but to give porters a
hint about the functionality. Quoting asm-generic/bitops.h:

/* For the benefit of those who are trying to port Linux to another
 * architecture, here are some C-language equivalents.  You should
 * recode these in the native assembly language, if at all possible.
 * To guarantee atomicity, these routines call cli() and sti() to
 * disable interrupts while they operate.  (You have to provide inline
 * routines to cli() and sti().) */

Or is this comment wrong, too?

Regards,
Carl-Daniel

-- 
http://www.hailfinger.org/

Re: [PATCH] [2.5] include/asm-generic/bitops.h {set,clear}_bit return void

[Robert Love]

On Tue, 2003-04-15 at 17:27, Carl-Daniel Hailfinger wrote:

> What is the preferred way to achieve atomicity in an operation now that
> cli() and sti() are gone?

spin locks.

> The point of asm-generic is not to use the files, but to give porters a
> hint about the functionality. Quoting asm-generic/bitops.h:
> 
> /* For the benefit of those who are trying to port Linux to another
>  * architecture, here are some C-language equivalents.  You should
>  * recode these in the native assembly language, if at all possible.
>  * To guarantee atomicity, these routines call cli() and sti() to
>  * disable interrupts while they operate.  (You have to provide inline
>  * routines to cli() and sti().) */
> 
> Or is this comment wrong, too?

Well, the cli() and sti() part is definitely wrong for 2.5.

It is wrong though to assume that nothing will use these; someone may
copy them directly (and then they do not work) or someone may #include
this file.

I like Arnd's suggestion to just remove these functions and all other
instances of them -- assuming in fact they are never used.

	Robert Love

Re: [PATCH] [2.5] include/asm-generic/bitops.h {set,clear}_bit return void

[Arnd Bergmann]

On Tuesday 15 April 2003 23:34, Robert Love wrote:
> > The point of asm-generic is not to use the files, but to give porters a
> > hint about the functionality. Quoting asm-generic/bitops.h:
> >
> > /* For the benefit of those who are trying to port Linux to another
> >  * architecture, here are some C-language equivalents.  You should
> >  * recode these in the native assembly language, if at all possible.
> >  * To guarantee atomicity, these routines call cli() and sti() to
> >  * disable interrupts while they operate.  (You have to provide inline
> >  * routines to cli() and sti().) */
> >
> > Or is this comment wrong, too?
>
> Well, the cli() and sti() part is definitely wrong for 2.5.
>
> It is wrong though to assume that nothing will use these; someone may
> copy them directly (and then they do not work) or someone may #include
> this file.

AFAICS, the meaning of the asm-generic directory has completely changed
during the 2.4/2.5 timeframe. They are now meant to be included from 
other header files, with the exception of unaligned.h and bitops.h.

These two files are now worthless for the purpose they were meant for. When
someone wants to start a new architecture, the maintained architectures
are the place to look at. For example, the parisc and ia64 implementations 
of bitops.h are both generic enough to be used for a new architecture 
and they actually are SMP safe and complete, while the asm-generic 
variant has been broken since linux-2.0!

> I like Arnd's suggestion to just remove these functions and all other
> instances of them -- assuming in fact they are never used.

I actually meant removing the asm-generic/bitops.h file, not the functions.

	Arnd <><

Re: [PATCH] [2.5] include/asm-generic/bitops.h {set,clear}_bit return void

[Pavel Machek]

Hi!

> > 
> > 
> >>+     mask = 1 << (nr & 0x1f);
> >>+     cli();
> >>+     *addr |= mask;
> >>+     sti();
> > 
> > 
> > cli() and sti() are no more. Moreover, the file you are trying to fix is

No. That file is still usefull.

> What is the preferred way to achieve atomicity in an operation now that
> cli() and sti() are gone?

spin_lock_irqsave(&bitops_lock).

> > not even used anywhere. Better submit a patch to remove it completely.
> 
> The point of asm-generic is not to use the files, but to give porters a
> hint about the functionality. Quoting asm-generic/bitops.h:
> 
> /* For the benefit of those who are trying to port Linux to another
>  * architecture, here are some C-language equivalents.  You should
>  * recode these in the native assembly language, if at all possible.
>  * To guarantee atomicity, these routines call cli() and sti() to
>  * disable interrupts while they operate.  (You have to provide inline
>  * routines to cli() and sti().) */
> 
> Or is this comment wrong, too?
-- 
				Pavel
Written on sharp zaurus, because my Velo1 broke. If you have Velo you don't need...

[PATCH] [2.5] include/asm-generic/bitops.h {set,clear}_bit return void

[Carl-Daniel Hailfinger]

Linus,

{set,clear}_bit for all arches no longer return int, but void.
This patch renames the old generic implementations to
test_and_{set,clear}_bit and adds new-style {set,clear}_bit.

Regards,
Carl-Daniel

===== include/asm-generic/bitops.h 1.2 vs edited =====
--- 1.2/include/asm-generic/bitops.h	Fri May  3 02:08:35 2002
+++ edited/include/asm-generic/bitops.h	Thu Apr 10 09:12:41 2003
@@ -16,7 +16,31 @@
  * C language equivalents written by Theodore Ts'o, 9/26/92
  */

+extern __inline__ void set_bit(int nr,long * addr)
+{
+	int	mask;
+
+	addr += nr >> 5;
+	mask = 1 << (nr & 0x1f);
+	cli();
+	*addr |= mask;
+	sti();
+	return;
+}
+
+extern __inline__ void clear_bit(int nr, long * addr)
+{
+	int	mask;
+
+	addr += nr >> 5;
+	mask = 1 << (nr & 0x1f);
+	cli();
+	*addr &= ~mask;
+	sti();
+	return;
+}
+
-extern __inline__ int set_bit(int nr,long * addr)
+extern __inline__ int test_and_set_bit(int nr,long * addr)
 {
 	int	mask, retval;

@@ -29,7 +53,7 @@
 	return retval;
 }

-extern __inline__ int clear_bit(int nr, long * addr)
+extern __inline__ int test_and_clear_bit(int nr, long * addr)
 {
 	int	mask, retval;


-- 
Linux scales to much more than 64 CPUs!
See include/linux/smp.h:64
#define MSG_ALL_BUT_SELF 0x8000 /* Assume <32768 CPU's */