[PATCH 0/6] RFC: use include/asm-generic/bitops.h

[PATCH 0/6] RFC: use include/asm-generic/bitops.h

[Akinobu Mita]

Large number of boilerplate bit operations written in C-language
are scattered around include/asm-*/bitops.h.
These patch series gather them into include/asm-generic/bitops.h. And

- kill duplicated code and comment (about 4000lines)
- use better C-language equivalents
- help porting new architecture (now include/asm-generic/bitops.h is not
  referenced from anywhere)

[PATCH 1/6] {set,clear,test}_bit() related cleanup

[Akinobu Mita]

While working on these patch set, I found several possible cleanup
on x86-64 and ia64.

Signed-off-by: Akinobu Mita <mita@miraclelinux.com>
---
 arch/ia64/kernel/mca.c           |    3 ++-
 arch/x86_64/kernel/mce.c         |    3 +--
 arch/x86_64/kernel/setup.c       |    3 +--
 arch/x86_64/pci/mmconfig.c       |    4 ++--
 include/asm-x86_64/mmu_context.h |    6 +++---
 include/asm-x86_64/pgtable.h     |    6 +++---
 6 files changed, 12 insertions(+), 13 deletions(-)

Index: 2.6-git/arch/x86_64/kernel/mce.c
===================================================================
--- 2.6-git.orig/arch/x86_64/kernel/mce.c	2006-01-25 19:07:15.000000000 +0900
+++ 2.6-git/arch/x86_64/kernel/mce.c	2006-01-25 19:13:59.000000000 +0900
@@ -139,8 +139,7 @@
 
 static int mce_available(struct cpuinfo_x86 *c)
 {
-	return test_bit(X86_FEATURE_MCE, &c->x86_capability) &&
-	       test_bit(X86_FEATURE_MCA, &c->x86_capability);
+	return cpu_has(c, X86_FEATURE_MCE) && cpu_has(c, X86_FEATURE_MCA);
 }
 
 static inline void mce_get_rip(struct mce *m, struct pt_regs *regs)
Index: 2.6-git/arch/x86_64/kernel/setup.c
===================================================================
--- 2.6-git.orig/arch/x86_64/kernel/setup.c	2006-01-25 19:07:15.000000000 +0900
+++ 2.6-git/arch/x86_64/kernel/setup.c	2006-01-25 19:13:59.000000000 +0900
@@ -1334,8 +1334,7 @@
 	{ 
 		int i; 
 		for ( i = 0 ; i < 32*NCAPINTS ; i++ )
-			if ( test_bit(i, &c->x86_capability) &&
-			     x86_cap_flags[i] != NULL )
+			if (cpu_has(c, i) && x86_cap_flags[i] != NULL )
 				seq_printf(m, " %s", x86_cap_flags[i]);
 	}
 		
Index: 2.6-git/include/asm-x86_64/mmu_context.h
===================================================================
--- 2.6-git.orig/include/asm-x86_64/mmu_context.h	2006-01-25 19:07:15.000000000 +0900
+++ 2.6-git/include/asm-x86_64/mmu_context.h	2006-01-25 19:13:59.000000000 +0900
@@ -34,12 +34,12 @@
 	unsigned cpu = smp_processor_id();
 	if (likely(prev != next)) {
 		/* stop flush ipis for the previous mm */
-		clear_bit(cpu, &prev->cpu_vm_mask);
+		cpu_clear(cpu, prev->cpu_vm_mask);
 #ifdef CONFIG_SMP
 		write_pda(mmu_state, TLBSTATE_OK);
 		write_pda(active_mm, next);
 #endif
-		set_bit(cpu, &next->cpu_vm_mask);
+		cpu_set(cpu, next->cpu_vm_mask);
 		load_cr3(next->pgd);
 
 		if (unlikely(next->context.ldt != prev->context.ldt)) 
@@ -50,7 +50,7 @@
 		write_pda(mmu_state, TLBSTATE_OK);
 		if (read_pda(active_mm) != next)
 			out_of_line_bug();
-		if(!test_and_set_bit(cpu, &next->cpu_vm_mask)) {
+		if(!cpu_test_and_set(cpu, next->cpu_vm_mask)) {
 			/* We were in lazy tlb mode and leave_mm disabled 
 			 * tlb flush IPI delivery. We must reload CR3
 			 * to make sure to use no freed page tables.
Index: 2.6-git/arch/x86_64/pci/mmconfig.c
===================================================================
--- 2.6-git.orig/arch/x86_64/pci/mmconfig.c	2006-01-25 19:07:15.000000000 +0900
+++ 2.6-git/arch/x86_64/pci/mmconfig.c	2006-01-25 19:13:59.000000000 +0900
@@ -46,7 +46,7 @@
 static char __iomem *pci_dev_base(unsigned int seg, unsigned int bus, unsigned int devfn)
 {
 	char __iomem *addr;
-	if (seg == 0 && bus == 0 && test_bit(PCI_SLOT(devfn), &fallback_slots))
+	if (seg == 0 && bus == 0 && test_bit(PCI_SLOT(devfn), fallback_slots))
 		return NULL;
 	addr = get_virt(seg, bus);
 	if (!addr)
@@ -134,7 +134,7 @@
 			continue;
 		addr = pci_dev_base(0, 0, PCI_DEVFN(i, 0));
 		if (addr == NULL|| readl(addr) != val1) {
-			set_bit(i, &fallback_slots);
+			set_bit(i, fallback_slots);
 		}
 	}
 }
Index: 2.6-git/include/asm-x86_64/pgtable.h
===================================================================
--- 2.6-git.orig/include/asm-x86_64/pgtable.h	2006-01-25 19:07:15.000000000 +0900
+++ 2.6-git/include/asm-x86_64/pgtable.h	2006-01-25 19:13:59.000000000 +0900
@@ -293,19 +293,19 @@
 {
 	if (!pte_dirty(*ptep))
 		return 0;
-	return test_and_clear_bit(_PAGE_BIT_DIRTY, ptep);
+	return test_and_clear_bit(_PAGE_BIT_DIRTY, &ptep->pte);
 }
 
 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
 	if (!pte_young(*ptep))
 		return 0;
-	return test_and_clear_bit(_PAGE_BIT_ACCESSED, ptep);
+	return test_and_clear_bit(_PAGE_BIT_ACCESSED, &ptep->pte);
 }
 
 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
-	clear_bit(_PAGE_BIT_RW, ptep);
+	clear_bit(_PAGE_BIT_RW, &ptep->pte);
 }
 
 /*
Index: 2.6-git/arch/ia64/kernel/mca.c
===================================================================
--- 2.6-git.orig/arch/ia64/kernel/mca.c	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/arch/ia64/kernel/mca.c	2006-01-25 19:14:01.000000000 +0900
@@ -69,6 +69,7 @@
 #include <linux/kernel.h>
 #include <linux/smp.h>
 #include <linux/workqueue.h>
+#include <linux/cpumask.h>
 
 #include <asm/delay.h>
 #include <asm/kdebug.h>
@@ -1430,7 +1431,7 @@
 	ti->cpu = cpu;
 	p->thread_info = ti;
 	p->state = TASK_UNINTERRUPTIBLE;
-	__set_bit(cpu, &p->cpus_allowed);
+	cpu_set(cpu, p->cpus_allowed);
 	INIT_LIST_HEAD(&p->tasks);
 	p->parent = p->real_parent = p->group_leader = p;
 	INIT_LIST_HEAD(&p->children);

Re: [PATCH 1/6] {set,clear,test}_bit() related cleanup

[Pavel Machek]

Hi!

> While working on these patch set, I found several possible cleanup
> on x86-64 and ia64.

It is probably not your fault, but...

> Index: 2.6-git/include/asm-x86_64/mmu_context.h
> ===================================================================
> --- 2.6-git.orig/include/asm-x86_64/mmu_context.h	2006-01-25 19:07:15.000000000 +0900
> +++ 2.6-git/include/asm-x86_64/mmu_context.h	2006-01-25 19:13:59.000000000 +0900
> @@ -34,12 +34,12 @@
>  	unsigned cpu = smp_processor_id();
>  	if (likely(prev != next)) {
>  		/* stop flush ipis for the previous mm */
> -		clear_bit(cpu, &prev->cpu_vm_mask);
> +		cpu_clear(cpu, prev->cpu_vm_mask);
>  #ifdef CONFIG_SMP
>  		write_pda(mmu_state, TLBSTATE_OK);
>  		write_pda(active_mm, next);
>  #endif
> -		set_bit(cpu, &next->cpu_vm_mask);
> +		cpu_set(cpu, next->cpu_vm_mask);
>  		load_cr3(next->pgd);
>  
>  		if (unlikely(next->context.ldt != prev->context.ldt)) 

cpu_set sounds *very* ambiguous. We have thing called cpusets, for
example. I'd not guess that is set_bit in cpu endianity (is it?).

								Pavel
-- 
Thanks, Sharp!

Re: [PATCH 1/6] {set,clear,test}_bit() related cleanup

[Russell King]

On Thu, Jan 26, 2006 at 05:14:27PM +0100, Pavel Machek wrote:
> > Index: 2.6-git/include/asm-x86_64/mmu_context.h
> > ===================================================================
> > --- 2.6-git.orig/include/asm-x86_64/mmu_context.h	2006-01-25 19:07:15.000000000 +0900
> > +++ 2.6-git/include/asm-x86_64/mmu_context.h	2006-01-25 19:13:59.000000000 +0900
> > @@ -34,12 +34,12 @@
> >  	unsigned cpu = smp_processor_id();
> >  	if (likely(prev != next)) {
> >  		/* stop flush ipis for the previous mm */
> > -		clear_bit(cpu, &prev->cpu_vm_mask);
> > +		cpu_clear(cpu, prev->cpu_vm_mask);
> >  #ifdef CONFIG_SMP
> >  		write_pda(mmu_state, TLBSTATE_OK);
> >  		write_pda(active_mm, next);
> >  #endif
> > -		set_bit(cpu, &next->cpu_vm_mask);
> > +		cpu_set(cpu, next->cpu_vm_mask);
> >  		load_cr3(next->pgd);
> >  
> >  		if (unlikely(next->context.ldt != prev->context.ldt)) 
> 
> cpu_set sounds *very* ambiguous. We have thing called cpusets, for
> example. I'd not guess that is set_bit in cpu endianity (is it?).

That's a problem for the cpusets folk - cpu_set predates them by a
fair time - it's part of the cpumask API.  See include/linux/cpumask.h

Also, since cpu_vm_mask is a cpumask_t, the above change to me looks
like a bug fix in its own right.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: [PATCH 1/6] {set,clear,test}_bit() related cleanup

[Paul Jackson]

Pavel wrote:
> cpu_set sounds *very* ambiguous. We have thing called cpusets,

Hmmm ... you're right.  I've worked for quite some time on both
of these, and hadn't noticed this similarity before.

Oh well.  Such is the nature of naming things.  Sometimes nice
names resemble other nice names in unexpected ways.

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.925.600.0401

[PATCH 2/6] use non atomic operations for minix_*_bit() and ext2_*_bit()

[Akinobu Mita]

Bitmap functions for the minix filesystem and the ext2 filesystem do not
require the atomic guarantees except ext2_set_bit_atomic() and
ext2_clear_bit_atomic().

But they are defined by using atomic bit operations on several architectures.
(h8300, ia64, mips, s390, sh, sh64, sparc, v850, and xtensa)
This patch switches to non atomic bit operation.

Signed-off-by: Akinobu Mita <mita@miraclelinux.com>
---
 asm-h8300/bitops.h   |    6 +++---
 asm-ia64/bitops.h    |   10 +++++-----
 asm-mips/bitops.h    |    6 +++---
 asm-s390/bitops.h    |   10 +++++-----
 asm-sh/bitops.h      |   16 +++++-----------
 asm-sh64/bitops.h    |   16 +++++-----------
 asm-sparc/bitops.h   |    6 +++---
 asm-sparc64/bitops.h |    6 +++---
 asm-v850/bitops.h    |   10 +++++-----
 asm-xtensa/bitops.h  |    6 +++---
 10 files changed, 40 insertions(+), 52 deletions(-)

Index: 2.6-git/include/asm-h8300/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-h8300/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-h8300/bitops.h	2006-01-25 19:14:01.000000000 +0900
@@ -397,9 +397,9 @@
 }
 
 /* Bitmap functions for the minix filesystem.  */
-#define minix_test_and_set_bit(nr,addr) test_and_set_bit(nr,addr)
-#define minix_set_bit(nr,addr) set_bit(nr,addr)
-#define minix_test_and_clear_bit(nr,addr) test_and_clear_bit(nr,addr)
+#define minix_test_and_set_bit(nr,addr) __test_and_set_bit(nr,addr)
+#define minix_set_bit(nr,addr) __set_bit(nr,addr)
+#define minix_test_and_clear_bit(nr,addr) __test_and_clear_bit(nr,addr)
 #define minix_test_bit(nr,addr) test_bit(nr,addr)
 #define minix_find_first_zero_bit(addr,size) find_first_zero_bit(addr,size)
 
Index: 2.6-git/include/asm-ia64/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-ia64/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-ia64/bitops.h	2006-01-25 19:14:02.000000000 +0900
@@ -394,18 +394,18 @@
 
 #define __clear_bit(nr, addr)		clear_bit(nr, addr)
 
-#define ext2_set_bit			test_and_set_bit
+#define ext2_set_bit			__test_and_set_bit
 #define ext2_set_bit_atomic(l,n,a)	test_and_set_bit(n,a)
-#define ext2_clear_bit			test_and_clear_bit
+#define ext2_clear_bit			__test_and_clear_bit
 #define ext2_clear_bit_atomic(l,n,a)	test_and_clear_bit(n,a)
 #define ext2_test_bit			test_bit
 #define ext2_find_first_zero_bit	find_first_zero_bit
 #define ext2_find_next_zero_bit		find_next_zero_bit
 
 /* Bitmap functions for the minix filesystem.  */
-#define minix_test_and_set_bit(nr,addr)		test_and_set_bit(nr,addr)
-#define minix_set_bit(nr,addr)			set_bit(nr,addr)
-#define minix_test_and_clear_bit(nr,addr)	test_and_clear_bit(nr,addr)
+#define minix_test_and_set_bit(nr,addr)		__test_and_set_bit(nr,addr)
+#define minix_set_bit(nr,addr)			__set_bit(nr,addr)
+#define minix_test_and_clear_bit(nr,addr)	__test_and_clear_bit(nr,addr)
 #define minix_test_bit(nr,addr)			test_bit(nr,addr)
 #define minix_find_first_zero_bit(addr,size)	find_first_zero_bit(addr,size)
 
Index: 2.6-git/include/asm-mips/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-mips/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-mips/bitops.h	2006-01-25 19:14:05.000000000 +0900
@@ -956,9 +956,9 @@
  * FIXME: These assume that Minix uses the native byte/bitorder.
  * This limits the Minix filesystem's value for data exchange very much.
  */
-#define minix_test_and_set_bit(nr,addr) test_and_set_bit(nr,addr)
-#define minix_set_bit(nr,addr) set_bit(nr,addr)
-#define minix_test_and_clear_bit(nr,addr) test_and_clear_bit(nr,addr)
+#define minix_test_and_set_bit(nr,addr) __test_and_set_bit(nr,addr)
+#define minix_set_bit(nr,addr) __set_bit(nr,addr)
+#define minix_test_and_clear_bit(nr,addr) __test_and_clear_bit(nr,addr)
 #define minix_test_bit(nr,addr) test_bit(nr,addr)
 #define minix_find_first_zero_bit(addr,size) find_first_zero_bit(addr,size)
 
Index: 2.6-git/include/asm-s390/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-s390/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-s390/bitops.h	2006-01-25 19:14:05.000000000 +0900
@@ -871,11 +871,11 @@
  */
 
 #define ext2_set_bit(nr, addr)       \
-	test_and_set_bit((nr)^(__BITOPS_WORDSIZE - 8), (unsigned long *)addr)
+	__test_and_set_bit((nr)^(__BITOPS_WORDSIZE - 8), (unsigned long *)addr)
 #define ext2_set_bit_atomic(lock, nr, addr)       \
 	test_and_set_bit((nr)^(__BITOPS_WORDSIZE - 8), (unsigned long *)addr)
 #define ext2_clear_bit(nr, addr)     \
-	test_and_clear_bit((nr)^(__BITOPS_WORDSIZE - 8), (unsigned long *)addr)
+	__test_and_clear_bit((nr)^(__BITOPS_WORDSIZE - 8), (unsigned long *)addr)
 #define ext2_clear_bit_atomic(lock, nr, addr)     \
 	test_and_clear_bit((nr)^(__BITOPS_WORDSIZE - 8), (unsigned long *)addr)
 #define ext2_test_bit(nr, addr)      \
@@ -1014,11 +1014,11 @@
 /* Bitmap functions for the minix filesystem.  */
 /* FIXME !!! */
 #define minix_test_and_set_bit(nr,addr) \
-	test_and_set_bit(nr,(unsigned long *)addr)
+	__test_and_set_bit(nr,(unsigned long *)addr)
 #define minix_set_bit(nr,addr) \
-	set_bit(nr,(unsigned long *)addr)
+	__set_bit(nr,(unsigned long *)addr)
 #define minix_test_and_clear_bit(nr,addr) \
-	test_and_clear_bit(nr,(unsigned long *)addr)
+	__test_and_clear_bit(nr,(unsigned long *)addr)
 #define minix_test_bit(nr,addr) \
 	test_bit(nr,(unsigned long *)addr)
 #define minix_find_first_zero_bit(addr,size) \
Index: 2.6-git/include/asm-sh/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-sh/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-sh/bitops.h	2006-01-25 19:14:06.000000000 +0900
@@ -339,8 +339,8 @@
 }
 
 #ifdef __LITTLE_ENDIAN__
-#define ext2_set_bit(nr, addr) test_and_set_bit((nr), (addr))
-#define ext2_clear_bit(nr, addr) test_and_clear_bit((nr), (addr))
+#define ext2_set_bit(nr, addr) __test_and_set_bit((nr), (addr))
+#define ext2_clear_bit(nr, addr) __test_and_clear_bit((nr), (addr))
 #define ext2_test_bit(nr, addr) test_bit((nr), (addr))
 #define ext2_find_first_zero_bit(addr, size) find_first_zero_bit((addr), (size))
 #define ext2_find_next_zero_bit(addr, size, offset) \
@@ -349,30 +349,24 @@
 static __inline__ int ext2_set_bit(int nr, volatile void * addr)
 {
 	int		mask, retval;
-	unsigned long	flags;
 	volatile unsigned char	*ADDR = (unsigned char *) addr;
 
 	ADDR += nr >> 3;
 	mask = 1 << (nr & 0x07);
-	local_irq_save(flags);
 	retval = (mask & *ADDR) != 0;
 	*ADDR |= mask;
-	local_irq_restore(flags);
 	return retval;
 }
 
 static __inline__ int ext2_clear_bit(int nr, volatile void * addr)
 {
 	int		mask, retval;
-	unsigned long	flags;
 	volatile unsigned char	*ADDR = (unsigned char *) addr;
 
 	ADDR += nr >> 3;
 	mask = 1 << (nr & 0x07);
-	local_irq_save(flags);
 	retval = (mask & *ADDR) != 0;
 	*ADDR &= ~mask;
-	local_irq_restore(flags);
 	return retval;
 }
 
@@ -459,9 +453,9 @@
 	})
 
 /* Bitmap functions for the minix filesystem.  */
-#define minix_test_and_set_bit(nr,addr) test_and_set_bit(nr,addr)
-#define minix_set_bit(nr,addr) set_bit(nr,addr)
-#define minix_test_and_clear_bit(nr,addr) test_and_clear_bit(nr,addr)
+#define minix_test_and_set_bit(nr,addr) __test_and_set_bit(nr,addr)
+#define minix_set_bit(nr,addr) __set_bit(nr,addr)
+#define minix_test_and_clear_bit(nr,addr) __test_and_clear_bit(nr,addr)
 #define minix_test_bit(nr,addr) test_bit(nr,addr)
 #define minix_find_first_zero_bit(addr,size) find_first_zero_bit(addr,size)
 
Index: 2.6-git/include/asm-sh64/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-sh64/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-sh64/bitops.h	2006-01-25 19:14:07.000000000 +0900
@@ -382,8 +382,8 @@
 #define hweight8(x) generic_hweight8(x)
 
 #ifdef __LITTLE_ENDIAN__
-#define ext2_set_bit(nr, addr) test_and_set_bit((nr), (addr))
-#define ext2_clear_bit(nr, addr) test_and_clear_bit((nr), (addr))
+#define ext2_set_bit(nr, addr) __test_and_set_bit((nr), (addr))
+#define ext2_clear_bit(nr, addr) __test_and_clear_bit((nr), (addr))
 #define ext2_test_bit(nr, addr) test_bit((nr), (addr))
 #define ext2_find_first_zero_bit(addr, size) find_first_zero_bit((addr), (size))
 #define ext2_find_next_zero_bit(addr, size, offset) \
@@ -392,30 +392,24 @@
 static __inline__ int ext2_set_bit(int nr, volatile void * addr)
 {
 	int		mask, retval;
-	unsigned long	flags;
 	volatile unsigned char	*ADDR = (unsigned char *) addr;
 
 	ADDR += nr >> 3;
 	mask = 1 << (nr & 0x07);
-	local_irq_save(flags);
 	retval = (mask & *ADDR) != 0;
 	*ADDR |= mask;
-	local_irq_restore(flags);
 	return retval;
 }
 
 static __inline__ int ext2_clear_bit(int nr, volatile void * addr)
 {
 	int		mask, retval;
-	unsigned long	flags;
 	volatile unsigned char	*ADDR = (unsigned char *) addr;
 
 	ADDR += nr >> 3;
 	mask = 1 << (nr & 0x07);
-	local_irq_save(flags);
 	retval = (mask & *ADDR) != 0;
 	*ADDR &= ~mask;
-	local_irq_restore(flags);
 	return retval;
 }
 
@@ -502,9 +496,9 @@
 	})
 
 /* Bitmap functions for the minix filesystem.  */
-#define minix_test_and_set_bit(nr,addr) test_and_set_bit(nr,addr)
-#define minix_set_bit(nr,addr) set_bit(nr,addr)
-#define minix_test_and_clear_bit(nr,addr) test_and_clear_bit(nr,addr)
+#define minix_test_and_set_bit(nr,addr) __test_and_set_bit(nr,addr)
+#define minix_set_bit(nr,addr) __set_bit(nr,addr)
+#define minix_test_and_clear_bit(nr,addr) __test_and_clear_bit(nr,addr)
 #define minix_test_bit(nr,addr) test_bit(nr,addr)
 #define minix_find_first_zero_bit(addr,size) find_first_zero_bit(addr,size)
 
Index: 2.6-git/include/asm-sparc/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-sparc/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-sparc/bitops.h	2006-01-25 19:14:08.000000000 +0900
@@ -523,11 +523,11 @@
 
 /* Bitmap functions for the minix filesystem.  */
 #define minix_test_and_set_bit(nr,addr)	\
-	test_and_set_bit((nr),(unsigned long *)(addr))
+	__test_and_set_bit((nr),(unsigned long *)(addr))
 #define minix_set_bit(nr,addr)		\
-	set_bit((nr),(unsigned long *)(addr))
+	__set_bit((nr),(unsigned long *)(addr))
 #define minix_test_and_clear_bit(nr,addr) \
-	test_and_clear_bit((nr),(unsigned long *)(addr))
+	__test_and_clear_bit((nr),(unsigned long *)(addr))
 #define minix_test_bit(nr,addr)		\
 	test_bit((nr),(unsigned long *)(addr))
 #define minix_find_first_zero_bit(addr,size) \
Index: 2.6-git/include/asm-sparc64/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-sparc64/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-sparc64/bitops.h	2006-01-25 19:14:08.000000000 +0900
@@ -280,11 +280,11 @@
 
 /* Bitmap functions for the minix filesystem.  */
 #define minix_test_and_set_bit(nr,addr)	\
-	test_and_set_bit((nr),(unsigned long *)(addr))
+	__test_and_set_bit((nr),(unsigned long *)(addr))
 #define minix_set_bit(nr,addr)	\
-	set_bit((nr),(unsigned long *)(addr))
+	__set_bit((nr),(unsigned long *)(addr))
 #define minix_test_and_clear_bit(nr,addr) \
-	test_and_clear_bit((nr),(unsigned long *)(addr))
+	__test_and_clear_bit((nr),(unsigned long *)(addr))
 #define minix_test_bit(nr,addr)	\
 	test_bit((nr),(unsigned long *)(addr))
 #define minix_find_first_zero_bit(addr,size) \
Index: 2.6-git/include/asm-v850/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-v850/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-v850/bitops.h	2006-01-25 19:14:08.000000000 +0900
@@ -336,18 +336,18 @@
 #define hweight16(x) 			generic_hweight16 (x)
 #define hweight8(x) 			generic_hweight8 (x)
 
-#define ext2_set_bit			test_and_set_bit
+#define ext2_set_bit			__test_and_set_bit
 #define ext2_set_bit_atomic(l,n,a)      test_and_set_bit(n,a)
-#define ext2_clear_bit			test_and_clear_bit
+#define ext2_clear_bit			__test_and_clear_bit
 #define ext2_clear_bit_atomic(l,n,a)    test_and_clear_bit(n,a)
 #define ext2_test_bit			test_bit
 #define ext2_find_first_zero_bit	find_first_zero_bit
 #define ext2_find_next_zero_bit		find_next_zero_bit
 
 /* Bitmap functions for the minix filesystem.  */
-#define minix_test_and_set_bit		test_and_set_bit
-#define minix_set_bit			set_bit
-#define minix_test_and_clear_bit	test_and_clear_bit
+#define minix_test_and_set_bit		__test_and_set_bit
+#define minix_set_bit			__set_bit
+#define minix_test_and_clear_bit	__test_and_clear_bit
 #define minix_test_bit 			test_bit
 #define minix_find_first_zero_bit 	find_first_zero_bit
 
Index: 2.6-git/include/asm-xtensa/bitops.h
===================================================================
--- 2.6-git.orig/include/asm-xtensa/bitops.h	2006-01-25 19:07:14.000000000 +0900
+++ 2.6-git/include/asm-xtensa/bitops.h	2006-01-25 19:14:08.000000000 +0900
@@ -436,9 +436,9 @@
 
 /* Bitmap functions for the minix filesystem.  */
 
-#define minix_test_and_set_bit(nr,addr) test_and_set_bit(nr,addr)
-#define minix_set_bit(nr,addr) set_bit(nr,addr)
-#define minix_test_and_clear_bit(nr,addr) test_and_clear_bit(nr,addr)
+#define minix_test_and_set_bit(nr,addr) __test_and_set_bit(nr,addr)
+#define minix_set_bit(nr,addr) __set_bit(nr,addr)
+#define minix_test_and_clear_bit(nr,addr) __test_and_clear_bit(nr,addr)
 #define minix_test_bit(nr,addr) test_bit(nr,addr)
 #define minix_find_first_zero_bit(addr,size) find_first_zero_bit(addr,size)
 

[PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Akinobu Mita]

o generic {,test_and_}{set,clear,change}_bit() (atomic bitops)

This patch introduces the C-language equivalents of the functions below:
void set_bit(int nr, volatile unsigned long *addr);
void clear_bit(int nr, volatile unsigned long *addr);
void change_bit(int nr, volatile unsigned long *addr);
int test_and_set_bit(int nr, volatile unsigned long *addr);
int test_and_clear_bit(int nr, volatile unsigned long *addr);
int test_and_change_bit(int nr, volatile unsigned long *addr);

HAVE_ARCH_ATOMIC_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/asm-powerpc/bitops.h
include/asm-parisc/bitops.h
include/asm-parisc/atomic.h

o generic __{,test_and_}{set,clear,change}_bit() and test_bit()

This patch introduces the C-language equivalents of the functions below:
void __set_bit(int nr, volatile unsigned long *addr);
void __clear_bit(int nr, volatile unsigned long *addr);
void __change_bit(int nr, volatile unsigned long *addr);
int __test_and_set_bit(int nr, volatile unsigned long *addr);
int __test_and_clear_bit(int nr, volatile unsigned long *addr);
int __test_and_change_bit(int nr, volatile unsigned long *addr);
int test_bit(int nr, const volatile unsigned long *addr);

HAVE_ARCH_NON_ATOMIC_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
asm-powerpc/bitops.h

o generic __ffs()

This patch introduces the C-language equivalent of the function:
unsigned long __ffs(unsigned long word);

HAVE_ARCH___FFS_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/asm-sparc64/bitops.h

o generic ffz()

This patch introduces the C-language equivalent of the function:
unsigned long ffz(unsigned long word);

HAVE_ARCH_FFZ_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/asm-sparc64/bitops.h

o generic fls()

This patch introduces the C-language equivalent of the function:
int fls(int x);

HAVE_ARCH_FLS_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/linux/bitops.h

o generic fls64()

This patch introduces the C-language equivalent of the function:
int fls64(__u64 x);

HAVE_ARCH_FLS64_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/linux/bitops.h

o generic find_{next,first}{,_zero}_bit()

This patch introduces the C-language equivalents of the functions below:

unsigned logn find_next_bit(const unsigned long *addr, unsigned long size,
			    unsigned long offset);
unsigned long find_next_zero_bit(const unsigned long *addr, unsigned long size,
				 unsigned long offset);
unsigned long find_first_zero_bit(const unsigned long *addr,
				  unsigned long size);
unsigned long find_first_bit(const unsigned long *addr, unsigned long size);

HAVE_ARCH_FIND_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
arch/powerpc/lib/bitops.c

==== KERNEL

o generic sched_find_first_bit()

This patch introduces the C-language equivalent of the function:
int sched_find_first_bit(const unsigned long *b);

HAVE_ARCH_SCHED_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/asm-powerpc/bitops.h

o generic ffs()

This patch introduces the C-language equivalent of the function:
int ffs(int x);

HAVE_ARCH_FFS_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/linux/bitops.h

o generic hweight{32,16,8}()

This patch introduces the C-language equivalents of the functions below:
unsigned int hweight32(unsigned int w);
unsigned int hweight16(unsigned int w);
unsigned int hweight8(unsigned int w);

HAVE_ARCH_HWEIGHT_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/linux/bitops.h

o generic hweight64()

This patch introduces the C-language equivalent of the function:
unsigned long hweight64(__u64 w);

HAVE_ARCH_HWEIGHT64_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/linux/bitops.h

o generic ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()

This patch introduces the C-language equivalents of the functions below:

int ext2_set_bit(int nr, volatile unsigned long *addr);
int ext2_clear_bit(int nr, volatile unsigned long *addr);
int ext2_test_bit(int nr, const volatile unsigned long *addr);
unsigned long ext2_find_first_zero_bit(const unsigned long *addr,
				       unsigned long size);

HAVE_ARCH_EXT2_NON_ATOMIC_BITOPS is defined when the architecture has its own
version of these functions.

unsinged long ext2_find_next_zero_bit(const unsigned long *addr,
				      unsigned long size);

This code largely copied from:
include/asm-powerpc/bitops.h
include/asm-parisc/bitops.h

o generic ext2_{set,clear}_bit_atomic()

This patch introduces the C-language equivalents of the functions below:
int ext2_set_bit_atomic(int nr, volatile unsigned long *addr);
int ext2_clear_bit_atomic(int nr, volatile unsigned long *addr);

HAVE_ARCH_EXT2_ATOMIC_BITOPS is defined when the architecture has its own
version of these functions.

This code largely copied from:
include/asm-sparc/bitops.h

o generic minix_{test,set,test_and_clear,test,find_first_zero}_bit()

This patch introduces the C-language equivalents of the functions below:

HAVE_ARCH_MINIX_BITOPS is defined when the architecture has its own
version of these functions.

int minix_test_and_set_bit(int nr, volatile unsigned long *addr);
int minix_set_bit(int nr, volatile unsigned long *addr);
int minix_test_and_clear_bit(int nr, volatile unsigned long *addr);
int minix_test_bit(int nr, const volatile unsigned long *addr);
unsigned long minix_find_first_zero_bit(const unsigned long *addr,
					unsigned long size);

This code largely copied from:
include/asm-sparc/bitops.h

Signed-off-by: Akinobu Mita <mita@miraclelinux.com>
---
 bitops.h |  677 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++----
 1 files changed, 641 insertions(+), 36 deletions(-)

Index: work/include/asm-generic/bitops.h
===================================================================
--- work.orig/include/asm-generic/bitops.h	2006-01-25 19:14:27.000000000 +0900
+++ work/include/asm-generic/bitops.h	2006-01-25 19:32:48.000000000 +0900
@@ -1,81 +1,686 @@
 #ifndef _ASM_GENERIC_BITOPS_H_
 #define _ASM_GENERIC_BITOPS_H_
 
+#include <asm/types.h>
+
+#define BITOP_MASK(nr)		(1UL << ((nr) % BITS_PER_LONG))
+#define BITOP_WORD(nr)		((nr) / BITS_PER_LONG)
+#define BITOP_LE_SWIZZLE	((BITS_PER_LONG-1) & ~0x7)
+
+#ifndef HAVE_ARCH_ATOMIC_BITOPS
+
+#ifdef CONFIG_SMP
+#include <asm/spinlock.h>
+#include <asm/cache.h>		/* we use L1_CACHE_BYTES */
+
+/* Use an array of spinlocks for our atomic_ts.
+ * Hash function to index into a different SPINLOCK.
+ * Since "a" is usually an address, use one spinlock per cacheline.
+ */
+#  define ATOMIC_HASH_SIZE 4
+#  define ATOMIC_HASH(a) (&(__atomic_hash[ (((unsigned long) a)/L1_CACHE_BYTES) & (ATOMIC_HASH_SIZE-1) ]))
+
+extern raw_spinlock_t __atomic_hash[ATOMIC_HASH_SIZE] __lock_aligned;
+
+/* Can't use raw_spin_lock_irq because of #include problems, so
+ * this is the substitute */
+#define _atomic_spin_lock_irqsave(l,f) do {	\
+	raw_spinlock_t *s = ATOMIC_HASH(l);	\
+	local_irq_save(f);			\
+	__raw_spin_lock(s);			\
+} while(0)
+
+#define _atomic_spin_unlock_irqrestore(l,f) do {	\
+	raw_spinlock_t *s = ATOMIC_HASH(l);		\
+	__raw_spin_unlock(s);				\
+	local_irq_restore(f);				\
+} while(0)
+
+
+#else
+#  define _atomic_spin_lock_irqsave(l,f) do { local_irq_save(f); } while (0)
+#  define _atomic_spin_unlock_irqrestore(l,f) do { local_irq_restore(f); } while (0)
+#endif
+
 /*
  * 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().)
  *
- * Also note, these routines assume that you have 32 bit longs.
- * You will have to change this if you are trying to port Linux to the
- * Alpha architecture or to a Cray.  :-)
- * 
  * C language equivalents written by Theodore Ts'o, 9/26/92
  */
 
-extern __inline__ int set_bit(int nr,long * addr)
+static __inline__ void set_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long flags;
+
+	_atomic_spin_lock_irqsave(p, flags);
+	*p  |= mask;
+	_atomic_spin_unlock_irqrestore(p, flags);
+}
+
+static __inline__ void clear_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long flags;
+
+	_atomic_spin_lock_irqsave(p, flags);
+	*p &= ~mask;
+	_atomic_spin_unlock_irqrestore(p, flags);
+}
+
+static __inline__ void change_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long flags;
+
+	_atomic_spin_lock_irqsave(p, flags);
+	*p ^= mask;
+	_atomic_spin_unlock_irqrestore(p, flags);
+}
+
+static __inline__ int test_and_set_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long old;
+	unsigned long flags;
+
+	_atomic_spin_lock_irqsave(p, flags);
+	old = *p;
+	*p = old | mask;
+	_atomic_spin_unlock_irqrestore(p, flags);
+
+	return (old & mask) != 0;
+}
+
+static __inline__ int test_and_clear_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long old;
+	unsigned long flags;
+
+	_atomic_spin_lock_irqsave(p, flags);
+	old = *p;
+	*p = old & ~mask;
+	_atomic_spin_unlock_irqrestore(p, flags);
+
+	return (old & mask) != 0;
+}
+
+static __inline__ int test_and_change_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long old;
+	unsigned long flags;
+
+	_atomic_spin_lock_irqsave(p, flags);
+	old = *p;
+	*p = old ^ mask;
+	_atomic_spin_unlock_irqrestore(p, flags);
+
+	return (old & mask) != 0;
+}
+
+#endif /* HAVE_ARCH_ATOMIC_BITOPS */
+
+#ifndef HAVE_ARCH_NON_ATOMIC_BITOPS
+
+static __inline__ void __set_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+
+	*p  |= mask;
+}
+
+static __inline__ void __clear_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+
+	*p &= ~mask;
+}
+
+static __inline__ void __change_bit(int nr, volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+
+	*p ^= mask;
+}
+
+static __inline__ int __test_and_set_bit(int nr, volatile unsigned long *addr)
 {
-	int	mask, retval;
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long old = *p;
 
-	addr += nr >> 5;
-	mask = 1 << (nr & 0x1f);
-	cli();
-	retval = (mask & *addr) != 0;
-	*addr |= mask;
-	sti();
-	return retval;
+	*p = old | mask;
+	return (old & mask) != 0;
 }
 
-extern __inline__ int clear_bit(int nr, long * addr)
+static __inline__ int __test_and_clear_bit(int nr, volatile unsigned long *addr)
 {
-	int	mask, retval;
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long old = *p;
 
-	addr += nr >> 5;
-	mask = 1 << (nr & 0x1f);
-	cli();
-	retval = (mask & *addr) != 0;
-	*addr &= ~mask;
-	sti();
-	return retval;
+	*p = old & ~mask;
+	return (old & mask) != 0;
 }
 
-extern __inline__ int test_bit(int nr, const unsigned long * addr)
+static __inline__ int __test_and_change_bit(int nr,
+					    volatile unsigned long *addr)
+{
+	unsigned long mask = BITOP_MASK(nr);
+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
+	unsigned long old = *p;
+
+	*p = old ^ mask;
+	return (old & mask) != 0;
+}
+
+static __inline__ int test_bit(int nr, __const__ volatile unsigned long *addr)
+{
+	return 1UL & (addr[BITOP_WORD(nr)] >> (nr & (BITS_PER_LONG-1)));
+}
+
+#endif /* HAVE_ARCH_NON_ATOMIC_BITOPS */
+
+#ifndef HAVE_ARCH___FFS_BITOPS
+
+/**
+ * __ffs - find first bit in word.
+ * @word: The word to search
+ *
+ * Returns 0..BITS_PER_LONG-1
+ * Undefined if no bit exists, so code should check against 0 first.
+ */
+static inline unsigned long __ffs(unsigned long word)
 {
-	int	mask;
+	int b = 0, s;
 
-	addr += nr >> 5;
-	mask = 1 << (nr & 0x1f);
-	return ((mask & *addr) != 0);
+#if BITS_PER_LONG == 32
+	s = 16; if (word << 16 != 0) s = 0; b += s; word >>= s;
+	s =  8; if (word << 24 != 0) s = 0; b += s; word >>= s;
+	s =  4; if (word << 28 != 0) s = 0; b += s; word >>= s;
+	s =  2; if (word << 30 != 0) s = 0; b += s; word >>= s;
+	s =  1; if (word << 31 != 0) s = 0; b += s;
+
+	return b;
+#elif BITS_PER_LONG == 64
+	s = 32; if (word << 32 != 0) s = 0; b += s; word >>= s;
+	s = 16; if (word << 48 != 0) s = 0; b += s; word >>= s;
+	s =  8; if (word << 56 != 0) s = 0; b += s; word >>= s;
+	s =  4; if (word << 60 != 0) s = 0; b += s; word >>= s;
+	s =  2; if (word << 62 != 0) s = 0; b += s; word >>= s;
+	s =  1; if (word << 63 != 0) s = 0; b += s;
+
+	return b;
+#else
+#error BITS_PER_LONG not defined
+#endif
 }
 
+#endif /* HAVE_ARCH___FFS_BITOPS */
+
+#ifndef HAVE_ARCH_FFZ_BITOPS
+
+/* Undefined if no bit is zero. */
+#define ffz(x)	__ffs(~x)
+
+#endif /* HAVE_ARCH_FFZ_BITOPS */
+
+#ifndef HAVE_ARCH_FLS_BITOPS
+
 /*
  * fls: find last bit set.
  */
 
-#define fls(x) generic_fls(x)
-#define fls64(x)   generic_fls64(x)
+static __inline__ int fls(int x)
+{
+	int r = 32;
+
+	if (!x)
+		return 0;
+	if (!(x & 0xffff0000u)) {
+		x <<= 16;
+		r -= 16;
+	}
+	if (!(x & 0xff000000u)) {
+		x <<= 8;
+		r -= 8;
+	}
+	if (!(x & 0xf0000000u)) {
+		x <<= 4;
+		r -= 4;
+	}
+	if (!(x & 0xc0000000u)) {
+		x <<= 2;
+		r -= 2;
+	}
+	if (!(x & 0x80000000u)) {
+		x <<= 1;
+		r -= 1;
+	}
+	return r;
+}
+
+#endif /* HAVE_ARCH_FLS_BITOPS */
+
+#ifndef HAVE_ARCH_FLS64_BITOPS
+
+static inline int fls64(__u64 x)
+{
+	__u32 h = x >> 32;
+	if (h)
+		return fls(x) + 32;
+	return fls(x);
+}
+
+#endif /* HAVE_ARCH_FLS64_BITOPS */
+
+#ifndef HAVE_ARCH_FIND_BITOPS
+
+/**
+ * find_next_bit - find the next set bit in a memory region
+ * @addr: The address to base the search on
+ * @offset: The bitnumber to start searching at
+ * @size: The maximum size to search
+ */
+static inline unsigned long find_next_bit(const unsigned long *addr,
+				unsigned long size, unsigned long offset)
+{
+	const unsigned long *p = addr + BITOP_WORD(offset);
+	unsigned long result = offset & ~(BITS_PER_LONG-1);
+	unsigned long tmp;
+
+	if (offset >= size)
+		return size;
+	size -= result;
+	offset %= BITS_PER_LONG;
+	if (offset) {
+		tmp = *(p++);
+		tmp &= (~0UL << offset);
+		if (size < BITS_PER_LONG)
+			goto found_first;
+		if (tmp)
+			goto found_middle;
+		size -= BITS_PER_LONG;
+		result += BITS_PER_LONG;
+	}
+	while (size & ~(BITS_PER_LONG-1)) {
+		if ((tmp = *(p++)))
+			goto found_middle;
+		result += BITS_PER_LONG;
+		size -= BITS_PER_LONG;
+	}
+	if (!size)
+		return result;
+	tmp = *p;
+
+found_first:
+	tmp &= (~0UL >> (BITS_PER_LONG - size));
+	if (tmp == 0UL)		/* Are any bits set? */
+		return result + size;	/* Nope. */
+found_middle:
+	return result + __ffs(tmp);
+}
+
+/*
+ * This implementation of find_{first,next}_zero_bit was stolen from
+ * Linus' asm-alpha/bitops.h.
+ */
+static inline unsigned long find_next_zero_bit(const unsigned long *addr,
+				unsigned long size, unsigned long offset)
+{
+	const unsigned long *p = addr + BITOP_WORD(offset);
+	unsigned long result = offset & ~(BITS_PER_LONG-1);
+	unsigned long tmp;
+
+	if (offset >= size)
+		return size;
+	size -= result;
+	offset %= BITS_PER_LONG;
+	if (offset) {
+		tmp = *(p++);
+		tmp |= ~0UL >> (BITS_PER_LONG - offset);
+		if (size < BITS_PER_LONG)
+			goto found_first;
+		if (~tmp)
+			goto found_middle;
+		size -= BITS_PER_LONG;
+		result += BITS_PER_LONG;
+	}
+	while (size & ~(BITS_PER_LONG-1)) {
+		if (~(tmp = *(p++)))
+			goto found_middle;
+		result += BITS_PER_LONG;
+		size -= BITS_PER_LONG;
+	}
+	if (!size)
+		return result;
+	tmp = *p;
+
+found_first:
+	tmp |= ~0UL << size;
+	if (tmp == ~0UL)	/* Are any bits zero? */
+		return result + size;	/* Nope. */
+found_middle:
+	return result + ffz(tmp);
+}
+
+#define find_first_zero_bit(addr, size) find_next_zero_bit((addr), (size), 0)
+#define find_first_bit(addr, size) find_next_bit((addr), (size), 0)
+
+#endif /* HAVE_ARCH_FIND_BITOPS */
 
 #ifdef __KERNEL__
 
+#ifndef HAVE_ARCH_SCHED_BITOPS
+
+#include <linux/compiler.h>	/* unlikely() */
+
+/*
+ * Every architecture must define this function. It's the fastest
+ * way of searching a 140-bit bitmap where the first 100 bits are
+ * unlikely to be set. It's guaranteed that at least one of the 140
+ * bits is cleared.
+ */
+static inline int sched_find_first_bit(const unsigned long *b)
+{
+#if BITS_PER_LONG == 64
+	if (unlikely(b[0]))
+		return __ffs(b[0]);
+	if (unlikely(b[1]))
+		return __ffs(b[1]) + 64;
+	return __ffs(b[2]) + 128;
+#elif BITS_PER_LONG == 32
+	if (unlikely(b[0]))
+		return __ffs(b[0]);
+	if (unlikely(b[1]))
+		return __ffs(b[1]) + 32;
+	if (unlikely(b[2]))
+		return __ffs(b[2]) + 64;
+	if (b[3])
+		return __ffs(b[3]) + 96;
+	return __ffs(b[4]) + 128;
+#else
+#error BITS_PER_LONG not defined
+#endif
+}
+
+#endif /* HAVE_ARCH_SCHED_BITOPS */
+
+#ifndef HAVE_ARCH_FFS_BITOPS
+
 /*
  * ffs: find first bit set. This is defined the same way as
  * the libc and compiler builtin ffs routines, therefore
  * differs in spirit from the above ffz (man ffs).
  */
 
-#define ffs(x) generic_ffs(x)
+static inline int ffs(int x)
+{
+	int r = 1;
+
+	if (!x)
+		return 0;
+	if (!(x & 0xffff)) {
+		x >>= 16;
+		r += 16;
+	}
+	if (!(x & 0xff)) {
+		x >>= 8;
+		r += 8;
+	}
+	if (!(x & 0xf)) {
+		x >>= 4;
+		r += 4;
+	}
+	if (!(x & 3)) {
+		x >>= 2;
+		r += 2;
+	}
+	if (!(x & 1)) {
+		x >>= 1;
+		r += 1;
+	}
+	return r;
+}
+
+#endif /* HAVE_ARCH_FFS_BITOPS */
+
+
+#ifndef HAVE_ARCH_HWEIGHT_BITOPS
 
 /*
  * hweightN: returns the hamming weight (i.e. the number
  * of bits set) of a N-bit word
  */
 
-#define hweight32(x) generic_hweight32(x)
-#define hweight16(x) generic_hweight16(x)
-#define hweight8(x) generic_hweight8(x)
+static inline unsigned int hweight32(unsigned int w)
+{
+        unsigned int res = (w & 0x55555555) + ((w >> 1) & 0x55555555);
+        res = (res & 0x33333333) + ((res >> 2) & 0x33333333);
+        res = (res & 0x0F0F0F0F) + ((res >> 4) & 0x0F0F0F0F);
+        res = (res & 0x00FF00FF) + ((res >> 8) & 0x00FF00FF);
+        return (res & 0x0000FFFF) + ((res >> 16) & 0x0000FFFF);
+}
+
+static inline unsigned int hweight16(unsigned int w)
+{
+        unsigned int res = (w & 0x5555) + ((w >> 1) & 0x5555);
+        res = (res & 0x3333) + ((res >> 2) & 0x3333);
+        res = (res & 0x0F0F) + ((res >> 4) & 0x0F0F);
+        return (res & 0x00FF) + ((res >> 8) & 0x00FF);
+}
+
+static inline unsigned int hweight8(unsigned int w)
+{
+        unsigned int res = (w & 0x55) + ((w >> 1) & 0x55);
+        res = (res & 0x33) + ((res >> 2) & 0x33);
+        return (res & 0x0F) + ((res >> 4) & 0x0F);
+}
+
+#endif /* HAVE_ARCH_HWEIGHT_BITOPS */
+
+#ifndef HAVE_ARCH_HWEIGHT64_BITOPS
+
+static inline unsigned long hweight64(__u64 w)
+{
+#if BITS_PER_LONG < 64
+	return hweight32((unsigned int)(w >> 32)) + hweight32((unsigned int)w);
+#else
+	u64 res;
+	res = (w & 0x5555555555555555ul) + ((w >> 1) & 0x5555555555555555ul);
+	res = (res & 0x3333333333333333ul) + ((res >> 2) & 0x3333333333333333ul);
+	res = (res & 0x0F0F0F0F0F0F0F0Ful) + ((res >> 4) & 0x0F0F0F0F0F0F0F0Ful);
+	res = (res & 0x00FF00FF00FF00FFul) + ((res >> 8) & 0x00FF00FF00FF00FFul);
+	res = (res & 0x0000FFFF0000FFFFul) + ((res >> 16) & 0x0000FFFF0000FFFFul);
+	return (res & 0x00000000FFFFFFFFul) + ((res >> 32) & 0x00000000FFFFFFFFul);
+#endif
+}
+
+#endif /* HAVE_ARCH_HWEIGHT64_BITOPS */
+
+#ifndef HAVE_ARCH_EXT2_NON_ATOMIC_BITOPS
+
+#include <asm/byteorder.h>
+
+#if defined(__LITTLE_ENDIAN)
+
+static __inline__ int generic_test_le_bit(unsigned long nr,
+				  __const__ unsigned long *addr)
+{
+	__const__ unsigned char	*tmp = (__const__ unsigned char *) addr;
+	return (tmp[nr >> 3] >> (nr & 7)) & 1;
+}
+
+#define generic___set_le_bit(nr, addr) __set_bit(nr, addr)
+#define generic___clear_le_bit(nr, addr) __clear_bit(nr, addr)
+
+#define generic_test_and_set_le_bit(nr, addr) test_and_set_bit(nr, addr)
+#define generic_test_and_clear_le_bit(nr, addr) test_and_clear_bit(nr, addr)
+
+#define generic___test_and_set_le_bit(nr, addr) __test_and_set_bit(nr, addr)
+#define generic___test_and_clear_le_bit(nr, addr) __test_and_clear_bit(nr, addr)
+
+#define generic_find_next_zero_le_bit(addr, size, offset) find_next_zero_bit(addr, size, offset)
+
+#elif defined(__BIG_ENDIAN)
+
+static __inline__ int generic_test_le_bit(unsigned long nr,
+				  __const__ unsigned long *addr)
+{
+	__const__ unsigned char	*tmp = (__const__ unsigned char *) addr;
+	return (tmp[nr >> 3] >> (nr & 7)) & 1;
+}
+
+#define generic___set_le_bit(nr, addr) \
+	__set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
+#define generic___clear_le_bit(nr, addr) \
+	__clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
+
+#define generic_test_and_set_le_bit(nr, addr) \
+	test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
+#define generic_test_and_clear_le_bit(nr, addr) \
+	test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
+
+#define generic___test_and_set_le_bit(nr, addr) \
+	__test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
+#define generic___test_and_clear_le_bit(nr, addr) \
+	__test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
+
+/* include/linux/byteorder does not support "unsigned long" type */
+static inline unsigned long ext2_swabp(const unsigned long * x)
+{
+#if BITS_PER_LONG == 64
+	return (unsigned long) __swab64p((u64 *) x);
+#elif BITS_PER_LONG == 32
+	return (unsigned long) __swab32p((u32 *) x);
+#else
+#error BITS_PER_LONG not defined
+#endif
+}
+
+/* include/linux/byteorder doesn't support "unsigned long" type */
+static inline unsigned long ext2_swab(const unsigned long y)
+{
+#if BITS_PER_LONG == 64
+	return (unsigned long) __swab64((u64) y);
+#elif BITS_PER_LONG == 32
+	return (unsigned long) __swab32((u32) y);
+#else
+#error BITS_PER_LONG not defined
+#endif
+}
+
+static __inline__ unsigned long generic_find_next_zero_le_bit(const unsigned long *addr,
+				unsigned long size, unsigned long offset)
+{
+	const unsigned long *p = addr + BITOP_WORD(offset);
+	unsigned long result = offset & ~(BITS_PER_LONG - 1);
+	unsigned long tmp;
+
+	if (offset >= size)
+		return size;
+	size -= result;
+	offset &= (BITS_PER_LONG - 1UL);
+	if (offset) {
+		tmp = ext2_swabp(p++);
+		tmp |= (~0UL >> (BITS_PER_LONG - offset));
+		if (size < BITS_PER_LONG)
+			goto found_first;
+		if (~tmp)
+			goto found_middle;
+		size -= BITS_PER_LONG;
+		result += BITS_PER_LONG;
+	}
+
+	while (size & ~(BITS_PER_LONG - 1)) {
+		if (~(tmp = *(p++)))
+			goto found_middle_swap;
+		result += BITS_PER_LONG;
+		size -= BITS_PER_LONG;
+	}
+	if (!size)
+		return result;
+	tmp = ext2_swabp(p);
+found_first:
+	tmp |= ~0UL << size;
+	if (tmp == ~0UL)	/* Are any bits zero? */
+		return result + size; /* Nope. Skip ffz */
+found_middle:
+	return result + ffz(tmp);
+
+found_middle_swap:
+	return result + ffz(ext2_swab(tmp));
+}
+#else
+#error "Please fix <asm/byteorder.h>"
+#endif
+
+#define generic_find_first_zero_le_bit(addr, size) \
+        generic_find_next_zero_le_bit((addr), (size), 0)
+
+#define ext2_set_bit(nr,addr)	\
+	generic___test_and_set_le_bit((nr),(unsigned long *)(addr))
+#define ext2_clear_bit(nr,addr)	\
+	generic___test_and_clear_le_bit((nr),(unsigned long *)(addr))
+
+#define ext2_test_bit(nr,addr)	\
+	generic_test_le_bit((nr),(unsigned long *)(addr))
+#define ext2_find_first_zero_bit(addr, size) \
+	generic_find_first_zero_le_bit((unsigned long *)(addr), (size))
+#define ext2_find_next_zero_bit(addr, size, off) \
+	generic_find_next_zero_le_bit((unsigned long *)(addr), (size), (off))
+
+#endif /* HAVE_ARCH_EXT2_NON_ATOMIC_BITOPS */
+
+#ifndef HAVE_ARCH_EXT2_ATOMIC_BITOPS
+
+#define ext2_set_bit_atomic(lock, nr, addr)		\
+	({						\
+		int ret;				\
+		spin_lock(lock);			\
+		ret = ext2_set_bit((nr), (unsigned long *)(addr)); \
+		spin_unlock(lock);			\
+		ret;					\
+	})
+
+#define ext2_clear_bit_atomic(lock, nr, addr)		\
+	({						\
+		int ret;				\
+		spin_lock(lock);			\
+		ret = ext2_clear_bit((nr), (unsigned long *)(addr)); \
+		spin_unlock(lock);			\
+		ret;					\
+	})
+
+#endif /* HAVE_ARCH_EXT2_ATOMIC_BITOPS */
+
+#ifndef HAVE_ARCH_MINIX_BITOPS
+
+#define minix_test_and_set_bit(nr,addr)	\
+	__test_and_set_bit((nr),(unsigned long *)(addr))
+#define minix_set_bit(nr,addr)		\
+	__set_bit((nr),(unsigned long *)(addr))
+#define minix_test_and_clear_bit(nr,addr) \
+	__test_and_clear_bit((nr),(unsigned long *)(addr))
+#define minix_test_bit(nr,addr)		\
+	test_bit((nr),(unsigned long *)(addr))
+#define minix_find_first_zero_bit(addr,size) \
+	find_first_zero_bit((unsigned long *)(addr),(size))
+
+#endif /* HAVE_ARCH_MINIX_BITOPS */
 
 #endif /* __KERNEL__ */
 

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Keith Owens]

Akinobu Mita (on Wed, 25 Jan 2006 20:32:06 +0900) wrote:
>o generic {,test_and_}{set,clear,change}_bit() (atomic bitops)
...
>+static __inline__ void set_bit(int nr, volatile unsigned long *addr)
>+{
>+	unsigned long mask = BITOP_MASK(nr);
>+	unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
>+	unsigned long flags;
>+
>+	_atomic_spin_lock_irqsave(p, flags);
>+	*p  |= mask;
>+	_atomic_spin_unlock_irqrestore(p, flags);
>+}

Be very, very careful about using these generic *_bit() routines if the
architecture supports non-maskable interrupts.

NMI events can occur at any time, including when interrupts have been
disabled by *_irqsave().  So you can get NMI events occurring while a
*_bit fucntion is holding a spin lock.  If the NMI handler also wants
to do bit manipulation (and they do) then you can get a deadlock
between the original caller of *_bit() and the NMI handler.

Doing any work that requires spinlocks in an NMI handler is just asking
for deadlock problems.  The generic *_bit() routines add a hidden
spinlock behind what was previously a safe operation.  I would even say
that any arch that supports any type of NMI event _must_ define its own
bit routines that do not rely on your _atomic_spin_lock_irqsave() and
its hash of spinlocks.

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Akinobu Mita]

On Wed, Jan 25, 2006 at 10:54:43PM +1100, Keith Owens wrote:
> Be very, very careful about using these generic *_bit() routines if the
> architecture supports non-maskable interrupts.
> 
> NMI events can occur at any time, including when interrupts have been
> disabled by *_irqsave().  So you can get NMI events occurring while a
> *_bit fucntion is holding a spin lock.  If the NMI handler also wants
> to do bit manipulation (and they do) then you can get a deadlock
> between the original caller of *_bit() and the NMI handler.
> 
> Doing any work that requires spinlocks in an NMI handler is just asking
> for deadlock problems.  The generic *_bit() routines add a hidden
> spinlock behind what was previously a safe operation.  I would even say
> that any arch that supports any type of NMI event _must_ define its own
> bit routines that do not rely on your _atomic_spin_lock_irqsave() and
> its hash of spinlocks.

At least cris and parisc are using similar *_bit function on SMP.
I will add your advise in comment.

--- ./include/asm-generic/bitops.h.orig	2006-01-26 10:56:00.000000000 +0900
+++ ./include/asm-generic/bitops.h	2006-01-26 11:01:28.000000000 +0900
@@ -50,6 +50,16 @@ extern raw_spinlock_t __atomic_hash[ATOM
  * C language equivalents written by Theodore Ts'o, 9/26/92
  */
 
+/*
+ * NMI events can occur at any time, including when interrupts have been
+ * disabled by *_irqsave().  So you can get NMI events occurring while a
+ * *_bit fucntion is holding a spin lock.  If the NMI handler also wants
+ * to do bit manipulation (and they do) then you can get a deadlock
+ * between the original caller of *_bit() and the NMI handler.
+ *
+ * by Keith Owens
+ */
+
 static __inline__ void set_bit(int nr, volatile unsigned long *addr)
 {
 	unsigned long mask = BITOP_MASK(nr);

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Russell King]

On Wed, Jan 25, 2006 at 08:32:06PM +0900, Akinobu Mita wrote:
> +#ifndef HAVE_ARCH___FFS_BITOPS
> +
> +/**
> + * __ffs - find first bit in word.
> + * @word: The word to search
> + *
> + * Returns 0..BITS_PER_LONG-1
> + * Undefined if no bit exists, so code should check against 0 first.
> + */
> +static inline unsigned long __ffs(unsigned long word)
>  {
> -	int	mask;
> +	int b = 0, s;
>  
> -	addr += nr >> 5;
> -	mask = 1 << (nr & 0x1f);
> -	return ((mask & *addr) != 0);
> +#if BITS_PER_LONG == 32
> +	s = 16; if (word << 16 != 0) s = 0; b += s; word >>= s;
> +	s =  8; if (word << 24 != 0) s = 0; b += s; word >>= s;
> +	s =  4; if (word << 28 != 0) s = 0; b += s; word >>= s;
> +	s =  2; if (word << 30 != 0) s = 0; b += s; word >>= s;
> +	s =  1; if (word << 31 != 0) s = 0; b += s;
> +
> +	return b;
> +#elif BITS_PER_LONG == 64
> +	s = 32; if (word << 32 != 0) s = 0; b += s; word >>= s;
> +	s = 16; if (word << 48 != 0) s = 0; b += s; word >>= s;
> +	s =  8; if (word << 56 != 0) s = 0; b += s; word >>= s;
> +	s =  4; if (word << 60 != 0) s = 0; b += s; word >>= s;
> +	s =  2; if (word << 62 != 0) s = 0; b += s; word >>= s;
> +	s =  1; if (word << 63 != 0) s = 0; b += s;
> +
> +	return b;
> +#else
> +#error BITS_PER_LONG not defined
> +#endif

This code generates more expensive shifts than our (ARMs) existing C
version.  This is a backward step.

Basically, shifts which depend on a variable are more expensive than
constant-based shifts.

I've not really looked at the rest because I haven't figured out which
bits will be used on ARM and which won't - which I think is another
problem with this patch set.  I'll look again later tonight.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Ian Molton]

Russell King wrote:

> This code generates more expensive shifts than our (ARMs) existing C
> version.  This is a backward step.
> 
> Basically, shifts which depend on a variable are more expensive than
> constant-based shifts.

arm26 will have the same problem here.

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Richard Henderson]

On Wed, Jan 25, 2006 at 08:02:50PM +0000, Russell King wrote:
> > +	s = 16; if (word << 16 != 0) s = 0; b += s; word >>= s;
> > +	s =  8; if (word << 24 != 0) s = 0; b += s; word >>= s;
> > +	s =  4; if (word << 28 != 0) s = 0; b += s; word >>= s;
...
> Basically, shifts which depend on a variable are more expensive than
> constant-based shifts.

Actually, they're all constant shifts.  Just written stupidly.


r~

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Edgar Toernig]

Richard Henderson wrote:
>
> On Wed, Jan 25, 2006 at 08:02:50PM +0000, Russell King wrote:
> > > +	s = 16; if (word << 16 != 0) s = 0; b += s; word >>= s;
> > > +	s =  8; if (word << 24 != 0) s = 0; b += s; word >>= s;
> > > +	s =  4; if (word << 28 != 0) s = 0; b += s; word >>= s;
> ...
> > Basically, shifts which depend on a variable are more expensive than
> > constant-based shifts.
> 
> Actually, they're all constant shifts.  Just written stupidly.

Why shift at all?

int ffs(u32 word)
{
    int bit = 0;

    word &= -word; // only keep the lsb.

    if (word & 0xffff0000) bit |= 16;
    if (word & 0xff00ff00) bit |=  8;
    if (word & 0xf0f0f0f0) bit |=  4;
    if (word & 0xcccccccc) bit |=  2;
    if (word & 0xaaaaaaaa) bit |=  1;

    return bit;
}

Ciao, ET.

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Richard Henderson]

On Thu, Jan 26, 2006 at 05:34:12AM +0100, Edgar Toernig wrote:
> Why shift at all?

Becuase that *is* a valid architecture tuning knob.  Most risc
machines can't AND with arbitrary constants like that, and loading
the constant might bulk things up more than just using the shift.


r~

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Russell King]

On Wed, Jan 25, 2006 at 04:06:18PM -0800, Richard Henderson wrote:
> On Wed, Jan 25, 2006 at 08:02:50PM +0000, Russell King wrote:
> > > +	s = 16; if (word << 16 != 0) s = 0; b += s; word >>= s;
> > > +	s =  8; if (word << 24 != 0) s = 0; b += s; word >>= s;
> > > +	s =  4; if (word << 28 != 0) s = 0; b += s; word >>= s;
> ...
> > Basically, shifts which depend on a variable are more expensive than
> > constant-based shifts.
> 
> Actually, they're all constant shifts.  Just written stupidly.

Unfortunately that's not correct.  You do not appear to have checked
the compiler output like I did - this code does _not_ generate
constant shifts.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Grant Grundler]

On Thu, Jan 26, 2006 at 08:55:41AM +0000, Russell King wrote:
> Unfortunately that's not correct.  You do not appear to have checked
> the compiler output like I did - this code does _not_ generate
> constant shifts.

Russell,
By "written stupidly", I thought Richard meant they could have
used constants instead of "s".  e.g.:
	if (word << 16 == 0) { b += 16; word >>= 16); }
	if (word << 24 == 0) { b +=  8; word >>=  8); }
	if (word << 28 == 0) { b +=  4; word >>=  4); }

But I prefer what Edgar Toernig suggested.

grant

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Nicolas Pitre]

On Thu, 26 Jan 2006, Grant Grundler wrote:

> On Thu, Jan 26, 2006 at 08:55:41AM +0000, Russell King wrote:
> > Unfortunately that's not correct.  You do not appear to have checked
> > the compiler output like I did - this code does _not_ generate
> > constant shifts.
> 
> Russell,
> By "written stupidly", I thought Richard meant they could have
> used constants instead of "s".  e.g.:
> 	if (word << 16 == 0) { b += 16; word >>= 16); }
> 	if (word << 24 == 0) { b +=  8; word >>=  8); }
> 	if (word << 28 == 0) { b +=  4; word >>=  4); }
> 
> But I prefer what Edgar Toernig suggested.

It is just as bad on ARM since it requires large constants that cannot 
be expressed with immediate litteral values.  The constant shift 
approach is really the best on ARM.


Nicolas

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Russell King]

On Thu, Jan 26, 2006 at 09:18:49AM -0700, Grant Grundler wrote:
> On Thu, Jan 26, 2006 at 08:55:41AM +0000, Russell King wrote:
> > Unfortunately that's not correct.  You do not appear to have checked
> > the compiler output like I did - this code does _not_ generate
> > constant shifts.
> 
> Russell,
> By "written stupidly", I thought Richard meant they could have
> used constants instead of "s".  e.g.:
> 	if (word << 16 == 0) { b += 16; word >>= 16); }
> 	if (word << 24 == 0) { b +=  8; word >>=  8); }
> 	if (word << 28 == 0) { b +=  4; word >>=  4); }
> 
> But I prefer what Edgar Toernig suggested.

Ok, I can see I'm going to lose this, but what the hell.

Firstly though, an out of line function call on ARM clobbers six out
of 11 CPU registers.

Let's compare the implementations, which are:

int toernig_ffs(unsigned long word)
{
    int bit = 0;
    word &= -word; // only keep the lsb.
    if (word & 0xffff0000) bit |= 16;
    if (word & 0xff00ff00) bit |=  8;
    if (word & 0xf0f0f0f0) bit |=  4;
    if (word & 0xcccccccc) bit |=  2;
    if (word & 0xaaaaaaaa) bit |=  1;
    return bit;
}

toernig_ffs:
        rsb     r3, r0, #0
        and     r0, r0, r3
        mov     r3, r0, lsr #16
        bic     r2, r0, #16711680
        str     lr, [sp, #-4]!
        mov     r3, r3, asl #16
        ldr     lr, .L7
        ldr     r1, .L7+4
        ldr     ip, .L7+8
        cmp     r3, #0
        bic     r2, r2, #255
        and     lr, r0, lr
        and     r1, r0, r1
        and     ip, r0, ip
        movne   r0, #16
        moveq   r0, #0
        cmp     r2, #0
        orrne   r0, r0, #8
        cmp     r1, #0
        orrne   r0, r0, #4
        cmp     ip, #0
        orrne   r0, r0, #2
        cmp     lr, #0
        orrne   r0, r0, #1
        ldr     pc, [sp], #4
.L8:
        .align  2
.L7:
        .word   -1431655766
        .word   -252645136
        .word   -858993460

25 instructions.  3 words of additional data.  5 registers.  0 register
based shifts.

I feel that this is far too expensive to sanely inline - at least three
words of additional data for a use in a function, and has a high register
usage comparable to that of an out of line function.

int mita_ffs(unsigned long word)
{
     int b = 0, s;
     s = 16; if (word << 16 != 0) s = 0; b += s; word >>= s;
     s =  8; if (word << 24 != 0) s = 0; b += s; word >>= s;
     s =  4; if (word << 28 != 0) s = 0; b += s; word >>= s;
     s =  2; if (word << 30 != 0) s = 0; b += s; word >>= s;
     s =  1; if (word << 31 != 0) s = 0; b += s;
     return b;
}

mita_ffs:
        movs    r1, r0, asl #16
        moveq   r2, #16
        movne   r2, #0
        mov     r0, r0, lsr r2		@ register-based shift
        mov     r3, r2
        movs    r2, r0, asl #24
        moveq   r2, #8
        movne   r2, #0
        mov     r0, r0, lsr r2		@ register-based shift
        movs    r1, r0, asl #28
        add     r3, r3, r2
        moveq   r2, #4
        movne   r2, #0
        mov     r0, r0, lsr r2		@ register-based shift
        movs    r1, r0, asl #30
        add     r3, r3, r2
        moveq   r2, #2
        movne   r2, #0
        mov     r0, r0, lsr r2		@ register-based shift
        tst     r0, #1
        add     r3, r3, r2
        moveq   r2, #1
        movne   r2, #0
        add     r3, r3, r2
        mov     r0, r3
        mov     pc, lr

26 instructions.  4 registers used.  4 unconditional register-based
shifts (expensive).

Better, but uses inefficient register based shifts (which can take twice
as many cycles as non-register based shifts depending on the CPU).  Still
has a high usage on CPU registers though.  Could possibly be a candidate
for inlining.

int arm_ffs(unsigned long word)
{
     int k = 31;
     if (word & 0x0000ffff) { k -= 16; word <<= 16; }
     if (word & 0x00ff0000) { k -= 8;  word <<= 8;  }
     if (word & 0x0f000000) { k -= 4;  word <<= 4;  }
     if (word & 0x30000000) { k -= 2;  word <<= 2;  }
     if (word & 0x40000000) { k -= 1; }
     return k;
}

arm_ffs:
        mov     r3, r0, asl #16
        mov     r3, r3, lsr #16
        cmp     r3, #0
        movne   r0, r0, asl #16
        mov     r3, #31
        movne   r3, #15
        tst     r0, #16711680
        movne   r0, r0, asl #8
        subne   r3, r3, #8
        tst     r0, #251658240
        movne   r0, r0, asl #4
        subne   r3, r3, #4
        tst     r0, #805306368
        movne   r0, r0, asl #2
        subne   r3, r3, #2
        tst     r0, #1073741824
        subne   r3, r3, #1
        mov     r0, r3
        mov     pc, lr

19 instructions.  2 registers.  0 register based shifts.  More reasonable
for inlining.

Clearly the smallest of the lot with the smallest register pressure,
being the best candidate out of the lot, whether we inline it or not.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Grant Grundler]

On Thu, Jan 26, 2006 at 04:40:21PM +0000, Russell King wrote:
> Ok, I can see I'm going to lose this, but what the hell.

Well, we agree. As Richard Henderson just pointed out, parisc
is among those that can't load large immediate values either.

> Let's compare the implementations, which are:
...
> int arm_ffs(unsigned long word)
> {
>      int k = 31;
>      if (word & 0x0000ffff) { k -= 16; word <<= 16; }
>      if (word & 0x00ff0000) { k -= 8;  word <<= 8;  }
>      if (word & 0x0f000000) { k -= 4;  word <<= 4;  }
>      if (word & 0x30000000) { k -= 2;  word <<= 2;  }
>      if (word & 0x40000000) { k -= 1; }
>      return k;
> }

Of those suggested, arm_ffs() is closest to what parisc
currently has in assembly (see include/asm-parisc/bitops.h:__ffs()).
But given how unobvious the parisc instruction nullification works,
the rough equivalent in "C" (untested!) would look something like:

	unsigned int k = 31;
	if (word & 0x0000ffff) { k -= 16;} else { word >>= 16; }
	if (word & 0x000000ff) { k -=  8;} else { word >>= 8; }
	if (word & 0x0000000f) { k -=  4;} else { word >>= 4; }
	if (word & 0x00000003) { k -=  2;} else { word >>= 2; }
	if (word & 0x00000001) { k -=  1;}
	return k;

I doubt that's better for arm but am curious how it compares.
You have time to try it?
If not, no worries.


> 19 instructions.  2 registers.  0 register based shifts.  More reasonable
> for inlining.

Yeah, about the same for parisc.

> Clearly the smallest of the lot with the smallest register pressure,
> being the best candidate out of the lot, whether we inline it or not.

Agreed. But I expect parisc will have to continue using it's asm
sequence and ignore the generic version. AFAIK, the compiler isn't that
good with instruction nullification and I have other issues I'd
rather work on.

cheers,
grant

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Russell King]

On Thu, Jan 26, 2006 at 04:04:43PM -0700, Grant Grundler wrote:
> On Thu, Jan 26, 2006 at 04:40:21PM +0000, Russell King wrote:
> > Ok, I can see I'm going to lose this, but what the hell.
> 
> Well, we agree. As Richard Henderson just pointed out, parisc
> is among those that can't load large immediate values either.
> 
> > Let's compare the implementations, which are:
> ...
> > int arm_ffs(unsigned long word)
> > {
> >      int k = 31;
> >      if (word & 0x0000ffff) { k -= 16; word <<= 16; }
> >      if (word & 0x00ff0000) { k -= 8;  word <<= 8;  }
> >      if (word & 0x0f000000) { k -= 4;  word <<= 4;  }
> >      if (word & 0x30000000) { k -= 2;  word <<= 2;  }
> >      if (word & 0x40000000) { k -= 1; }
> >      return k;
> > }
> 
> Of those suggested, arm_ffs() is closest to what parisc
> currently has in assembly (see include/asm-parisc/bitops.h:__ffs()).
> But given how unobvious the parisc instruction nullification works,
> the rough equivalent in "C" (untested!) would look something like:
> 
> 	unsigned int k = 31;
> 	if (word & 0x0000ffff) { k -= 16;} else { word >>= 16; }
> 	if (word & 0x000000ff) { k -=  8;} else { word >>= 8; }
> 	if (word & 0x0000000f) { k -=  4;} else { word >>= 4; }
> 	if (word & 0x00000003) { k -=  2;} else { word >>= 2; }
> 	if (word & 0x00000001) { k -=  1;}
> 	return k;
> 
> I doubt that's better for arm but am curious how it compares.
> You have time to try it?

This is essentially the same as arm_ffs():

grundler_ffs:
        mov     r3, r0, asl #16
        mov     r3, r3, lsr #16
        cmp     r3, #0
        moveq   r0, r0, lsr #16
        mov     r3, #31
        movne   r3, #15
        tst     r0, #255
        moveq   r0, r0, lsr #8
        subne   r3, r3, #8
        tst     r0, #15
        moveq   r0, r0, lsr #4
        subne   r3, r3, #4
        tst     r0, #3
        moveq   r0, r0, lsr #2
        subne   r3, r3, #2
        tst     r0, #1
        subne   r3, r3, #1
        mov     r0, r3
        mov     pc, lr

only that the shifts, immediate values and the sense of some of the
conditional instructions have changed.  Therefore, the parisc rough
equivalent looks like it would be suitable for ARM as well.

> > Clearly the smallest of the lot with the smallest register pressure,
> > being the best candidate out of the lot, whether we inline it or not.
> 
> Agreed. But I expect parisc will have to continue using it's asm
> sequence and ignore the generic version. AFAIK, the compiler isn't that
> good with instruction nullification and I have other issues I'd
> rather work on.

Me too - already solved this problem once.  However, I'd rather not
needlessly take a step backwards in the name of generic bitops.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Stuart Brady]

On Thu, Jan 26, 2006 at 11:03:54PM +0000, Russell King wrote:
> Me too - already solved this problem once.  However, I'd rather not
> needlessly take a step backwards in the name of generic bitops.

Indeed.  However, I think we can actually improve bitops for some
architectures.  Here's what I've found so far:

Versions of Alpha, ARM, MIPS, PowerPC and SPARC have bit counting
instructions which we're using in most cases.  I may have missed some:

  Alpha may have:
    ctlz, CounT Leading Zeros
    cttz, CounT Trailing Zeros

  ARM (since v5) has:
    clz, Count Leading Zeros

  MIPS may have:
    clz, Count Leading Zeros
    clo, Count Leading Ones

  PowerPC has:
    cntlz[wd], CouNT Leading Zeros (for Word/Double-word)

  SPARC v9 has:
    popc, POPulation Count

PA-RISC has none.  I've not checked any others.

The Alpha, ARM and PowerPC functions look fine to me.

On MIPS, fls() and flz() should probably use CLO.  Curiously, MIPS is
the only arch with a flz() function.

On SPARC, the implementation of ffz() appears to be "cheese", and the
proposed generic versions would be better.  ffs() looks quite generic,
and fls() uses the linux/bitops.h implementation.

There are versions of hweight*() for sparc64 which use POPC when
ULTRA_HAS_POPULATION_COUNT is defined, but AFAICS, it's never defined.

The SPARC v9 arch manual recommends using popc(x ^ ~-x) for functions
like ffs().  ffz() would return ffs(~x).

I've had an idea for fls():

static inline int fls(unsigned long x)
{
	x |= x >> 1;
	x |= x >> 2;
	x |= x >> 4;
	x |= x >> 8;
	x |= x >> 16;
	return popc(x);
}

I'm not sure how that compares to the generic fls(), but I suspect it's
quite a bit faster.  Unfortunately, I don't have any MIPS or SPARC v9
hardware to test this on.

I'm not sure if this is of any use:

static inline int __ffs(unsigned long x)
{
	return (int)hweight_long(x ^ ~-x) - 1;
}

The idea being that the generic hweight_long has no branches.
-- 
Stuart Brady

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[David S. Miller]

From: Stuart Brady <sdbrady@ntlworld.com>
Date: Sun, 29 Jan 2006 07:12:42 +0000

> There are versions of hweight*() for sparc64 which use POPC when
> ULTRA_HAS_POPULATION_COUNT is defined, but AFAICS, it's never defined.

That's right, the problem here is that no chips actually implement
the instruction in hardware, it is software emulated, ie. useless :-)

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Ralf Baechle]

On Sun, Jan 29, 2006 at 07:12:42AM +0000, Stuart Brady wrote:

> On MIPS, fls() and flz() should probably use CLO.

It actually uses clz.

> Curiously, MIPS is the only arch with a flz() function.

No longer.  The fls implementation was based on flz and fls was the only
user of flz.  So I cleaned that, once I commit flz will be gone.  Not
only a cleanup but also a minor optimization.

  Ralf

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Stuart Brady]

On Mon, Jan 30, 2006 at 05:06:47PM +0000, Ralf Baechle wrote:
> On Sun, Jan 29, 2006 at 07:12:42AM +0000, Stuart Brady wrote:
> 
> > On MIPS, fls() and flz() should probably use CLO.
> 
> It actually uses clz.

I know.  flz(x) is basically __ilog2(~x), and I still say clo would be
better.  Removing flz() sounds reasonable, though.

> > Curiously, MIPS is the only arch with a flz() function.
> 
> No longer.  The fls implementation was based on flz and fls was the only
> user of flz.  So I cleaned that, once I commit flz will be gone.  Not
> only a cleanup but also a minor optimization.

I'd got that slightly wrong.  Yeah, fls(x) returned flz(~x) + 1, which
is __ilog2(~~x) + 1.  So obviously clz was fine for that, but it needed
cleaning up.

Shame about popc on SPARC.  However, ffz is cheese, regardless of pops.
(On sparc64, ffs is too.)  I'll wait for the generic bitops patches to
be dealt with (or not) and then submit a patch fixing this if needed.

Thanks,
-- 
Stuart Brady

By the way, I really hope nobody gets ten copies of this, as happened
with my last post.  It does not seem to be my fault, AFAICS.

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[David S. Miller]

From: Stuart Brady <sdbrady@ntlworld.com>
Date: Mon, 30 Jan 2006 19:50:04 +0000

> Shame about popc on SPARC.  However, ffz is cheese, regardless of pops.
> (On sparc64, ffs is too.)  I'll wait for the generic bitops patches to
> be dealt with (or not) and then submit a patch fixing this if needed.

I'm happy with any improvement you might make here, for sure.

The sparc64 ffz() implementation was done so dog stupid like that
so that the code would be small since this gets inlined all over
the place.

So if you can keep it small and improve it, or make it a bit larger
and uninline it, that's great.

Re: [parisc-linux] Re: [PATCH 3/6] C-language equivalents of

[John David Anglin]

> Yeah, about the same for parisc.
> 
> > Clearly the smallest of the lot with the smallest register pressure,
> > being the best candidate out of the lot, whether we inline it or not.
> 
> Agreed. But I expect parisc will have to continue using it's asm
> sequence and ignore the generic version. AFAIK, the compiler isn't that
> good with instruction nullification and I have other issues I'd
> rather work on.

I looked at the assembler code generated on parisc with 4.1.0 (prerelease).
The toernig code is definitely inferior.  The mita sequence is four
instructions longer than the arm sequence, but it didn't have any branches.
The arm sequence has four branches.  Thus, it's not clear to me which
would perform better in the real world.  There were no nullified instructions
generated for any of the sequences.  However, neither is as good as the
handcraft asm sequence currently being used.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Hirokazu Takata]

Hello Mita-san, and folks,

From: mita@miraclelinux.com (Akinobu Mita)
Subject: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h
Date: Wed, 25 Jan 2006 20:32:06 +0900
> o generic {,test_and_}{set,clear,change}_bit() (atomic bitops)
> 
> This patch introduces the C-language equivalents of the functions below:
> void set_bit(int nr, volatile unsigned long *addr);
> void clear_bit(int nr, volatile unsigned long *addr);
...
> int test_and_change_bit(int nr, volatile unsigned long *addr);
> 
> HAVE_ARCH_ATOMIC_BITOPS is defined when the architecture has its own
> version of these functions.
> 
> This code largely copied from:
> include/asm-powerpc/bitops.h
> include/asm-parisc/bitops.h
> include/asm-parisc/atomic.h

Could you tell me more about the new generic {set,clear,test}_bit()
routines?

Why do you copied these routines from parisc and employed them
 as generic ones?
I'm not sure whether these generic {set,clear,test}_bit() routines
are really generic or not.

> +/* Can't use raw_spin_lock_irq because of #include problems, so
> + * this is the substitute */
> +#define _atomic_spin_lock_irqsave(l,f) do {	\
> +	raw_spinlock_t *s = ATOMIC_HASH(l);	\
> +	local_irq_save(f);			\
> +	__raw_spin_lock(s);			\
> +} while(0)
> +
> +#define _atomic_spin_unlock_irqrestore(l,f) do {	\
> +	raw_spinlock_t *s = ATOMIC_HASH(l);		\
> +	__raw_spin_unlock(s);				\
> +	local_irq_restore(f);				\
> +} while(0)

Is there a possibility that these routines affect for archs
with no HAVE_ARCH_ATOMIC_BITOPS for SMP ?
I think __raw_spin_lock() is sufficient and local_irqsave() is 
not necessary in general atomic routines.

If the parisc's LDCW instruction required disabling interrupts,
it would be parisc specific and not generic case, I think, 
although I'm not familier with the parisc architecture...

-- Takata

Re: [PATCH 3/6] C-language equivalents of include/asm-*/bitops.h

[Akinobu Mita]

On Fri, Jan 27, 2006 at 09:51:47PM +0900, Hirokazu Takata wrote:
 
> Could you tell me more about the new generic {set,clear,test}_bit()
> routines?
> 
> Why do you copied these routines from parisc and employed them
>  as generic ones?
> I'm not sure whether these generic {set,clear,test}_bit() routines
> are really generic or not.

I think it is the most portable implementation.
And I'm trying not to write my own code in this patch set.

> 
> > +/* Can't use raw_spin_lock_irq because of #include problems, so
> > + * this is the substitute */
> > +#define _atomic_spin_lock_irqsave(l,f) do {	\
> > +	raw_spinlock_t *s = ATOMIC_HASH(l);	\
> > +	local_irq_save(f);			\
> > +	__raw_spin_lock(s);			\
> > +} while(0)
> > +
> > +#define _atomic_spin_unlock_irqrestore(l,f) do {	\
> > +	raw_spinlock_t *s = ATOMIC_HASH(l);		\
> > +	__raw_spin_unlock(s);				\
> > +	local_irq_restore(f);				\
> > +} while(0)
> 
> Is there a possibility that these routines affect for archs
> with no HAVE_ARCH_ATOMIC_BITOPS for SMP ?

Currently there is no architecture using this atomic *_bit() routines
on SMP. But it may be the benefit of those who are trying to port Linux.
(See the comment by Theodore Ts'o in include/asm-generic/bitops.h)

> I think __raw_spin_lock() is sufficient and local_irqsave() is 
> not necessary in general atomic routines.

If the interrupt handler also wants to do bit manipilation then
you can get a deadlock between the original caller of *_bit() and the
interrupt handler.

[PATCH 5/6] fix warning on test_ti_thread_flag()

[Akinobu Mita]

If the arechitecture is
- BITS_PER_LONG == 64
- struct thread_info.flag 32 is bits
- second argument of test_bit() was void *

Then compiler print error message on test_ti_thread_flags()
in include/linux/thread_info.h

Signed-off-by: Akinobu Mita <mita@miraclelinux.com>
---
 thread_info.h |    2 +-
 1 files changed, 1 insertion(+), 1 deletion(-)

Index: 2.6-git/include/linux/thread_info.h
===================================================================
--- 2.6-git.orig/include/linux/thread_info.h	2006-01-25 19:07:12.000000000 +0900
+++ 2.6-git/include/linux/thread_info.h	2006-01-25 19:14:26.000000000 +0900
@@ -49,7 +49,7 @@
 
 static inline int test_ti_thread_flag(struct thread_info *ti, int flag)
 {
-	return test_bit(flag,&ti->flags);
+	return test_bit(flag, (void *)&ti->flags);
 }
 
 #define set_thread_flag(flag) \

Re: [PATCH 5/6] fix warning on test_ti_thread_flag()

[Geert Uytterhoeven]

On Wed, 25 Jan 2006, Akinobu Mita wrote:
> If the arechitecture is
> - BITS_PER_LONG == 64
> - struct thread_info.flag 32 is bits
> - second argument of test_bit() was void *
> 
> Then compiler print error message on test_ti_thread_flags()
> in include/linux/thread_info.h
> 
> Signed-off-by: Akinobu Mita <mita@miraclelinux.com>
> ---
>  thread_info.h |    2 +-
>  1 files changed, 1 insertion(+), 1 deletion(-)
> 
> Index: 2.6-git/include/linux/thread_info.h
> ===================================================================
> --- 2.6-git.orig/include/linux/thread_info.h	2006-01-25 19:07:12.000000000 +0900
> +++ 2.6-git/include/linux/thread_info.h	2006-01-25 19:14:26.000000000 +0900
> @@ -49,7 +49,7 @@
>  
>  static inline int test_ti_thread_flag(struct thread_info *ti, int flag)
>  {
> -	return test_bit(flag,&ti->flags);
> +	return test_bit(flag, (void *)&ti->flags);
>  }

This is not safe. The bitops are defined to work on unsigned long only, so
flags should be changed to unsigned long instead, or you should use a
temporary.

Affected platforms:
  - alpha: flags is unsigned int
  - ia64, sh, x86_64: flags is __u32

The only affected 64-platforms are little endian, so it will silently work
after your change, though...

Gr{oetje,eeting}s,

						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds

Re: [PATCH 5/6] fix warning on test_ti_thread_flag()

[Paul Mackerras]

Akinobu Mita writes:

> If the arechitecture is
> - BITS_PER_LONG == 64
> - struct thread_info.flag 32 is bits
> - second argument of test_bit() was void *
> 
> Then compiler print error message on test_ti_thread_flags()
> in include/linux/thread_info.h

And correctly so.  The correct fix is to make thread_info.flag an
unsigned long.  This patch is NAKed.

Paul.

Re: [PATCH 5/6] fix warning on test_ti_thread_flag()

[David S. Miller]

From: Paul Mackerras <paulus@samba.org>
Date: Thu, 26 Jan 2006 09:28:02 +1100

> Akinobu Mita writes:
> 
> > If the arechitecture is
> > - BITS_PER_LONG == 64
> > - struct thread_info.flag 32 is bits
> > - second argument of test_bit() was void *
> > 
> > Then compiler print error message on test_ti_thread_flags()
> > in include/linux/thread_info.h
> 
> And correctly so.  The correct fix is to make thread_info.flag an
> unsigned long.  This patch is NAKed.

I agree.

[PATCH 6/6] remove unused generic bitops in include/linux/bitops.h

[Akinobu Mita]

generic_{ffs,fls,fls64,hweight{64,32,16,8}}() were moved into
include/asm-generic/bitops.h. So all architectures don't use them.

Signed-off-by: Akinobu Mita <mita@miraclelinux.com>
---

 bitops.h |  124 ---------------------------------------------------------------
 1 files changed, 1 insertion(+), 123 deletions(-)

Index: 2.6-git/include/linux/bitops.h
===================================================================
--- 2.6-git.orig/include/linux/bitops.h	2006-01-25 19:07:12.000000000 +0900
+++ 2.6-git/include/linux/bitops.h	2006-01-25 19:14:27.000000000 +0900
@@ -3,88 +3,11 @@
 #include <asm/types.h>
 
 /*
- * ffs: find first bit set. This is defined the same way as
- * the libc and compiler builtin ffs routines, therefore
- * differs in spirit from the above ffz (man ffs).
- */
-
-static inline int generic_ffs(int x)
-{
-	int r = 1;
-
-	if (!x)
-		return 0;
-	if (!(x & 0xffff)) {
-		x >>= 16;
-		r += 16;
-	}
-	if (!(x & 0xff)) {
-		x >>= 8;
-		r += 8;
-	}
-	if (!(x & 0xf)) {
-		x >>= 4;
-		r += 4;
-	}
-	if (!(x & 3)) {
-		x >>= 2;
-		r += 2;
-	}
-	if (!(x & 1)) {
-		x >>= 1;
-		r += 1;
-	}
-	return r;
-}
-
-/*
- * fls: find last bit set.
- */
-
-static __inline__ int generic_fls(int x)
-{
-	int r = 32;
-
-	if (!x)
-		return 0;
-	if (!(x & 0xffff0000u)) {
-		x <<= 16;
-		r -= 16;
-	}
-	if (!(x & 0xff000000u)) {
-		x <<= 8;
-		r -= 8;
-	}
-	if (!(x & 0xf0000000u)) {
-		x <<= 4;
-		r -= 4;
-	}
-	if (!(x & 0xc0000000u)) {
-		x <<= 2;
-		r -= 2;
-	}
-	if (!(x & 0x80000000u)) {
-		x <<= 1;
-		r -= 1;
-	}
-	return r;
-}
-
-/*
  * Include this here because some architectures need generic_ffs/fls in
  * scope
  */
 #include <asm/bitops.h>
 
-
-static inline int generic_fls64(__u64 x)
-{
-	__u32 h = x >> 32;
-	if (h)
-		return fls(x) + 32;
-	return fls(x);
-}
-
 static __inline__ int get_bitmask_order(unsigned int count)
 {
 	int order;
@@ -103,54 +26,9 @@
 	return order;
 }
 
-/*
- * hweightN: returns the hamming weight (i.e. the number
- * of bits set) of a N-bit word
- */
-
-static inline unsigned int generic_hweight32(unsigned int w)
-{
-        unsigned int res = (w & 0x55555555) + ((w >> 1) & 0x55555555);
-        res = (res & 0x33333333) + ((res >> 2) & 0x33333333);
-        res = (res & 0x0F0F0F0F) + ((res >> 4) & 0x0F0F0F0F);
-        res = (res & 0x00FF00FF) + ((res >> 8) & 0x00FF00FF);
-        return (res & 0x0000FFFF) + ((res >> 16) & 0x0000FFFF);
-}
-
-static inline unsigned int generic_hweight16(unsigned int w)
-{
-        unsigned int res = (w & 0x5555) + ((w >> 1) & 0x5555);
-        res = (res & 0x3333) + ((res >> 2) & 0x3333);
-        res = (res & 0x0F0F) + ((res >> 4) & 0x0F0F);
-        return (res & 0x00FF) + ((res >> 8) & 0x00FF);
-}
-
-static inline unsigned int generic_hweight8(unsigned int w)
-{
-        unsigned int res = (w & 0x55) + ((w >> 1) & 0x55);
-        res = (res & 0x33) + ((res >> 2) & 0x33);
-        return (res & 0x0F) + ((res >> 4) & 0x0F);
-}
-
-static inline unsigned long generic_hweight64(__u64 w)
-{
-#if BITS_PER_LONG < 64
-	return generic_hweight32((unsigned int)(w >> 32)) +
-				generic_hweight32((unsigned int)w);
-#else
-	u64 res;
-	res = (w & 0x5555555555555555ul) + ((w >> 1) & 0x5555555555555555ul);
-	res = (res & 0x3333333333333333ul) + ((res >> 2) & 0x3333333333333333ul);
-	res = (res & 0x0F0F0F0F0F0F0F0Ful) + ((res >> 4) & 0x0F0F0F0F0F0F0F0Ful);
-	res = (res & 0x00FF00FF00FF00FFul) + ((res >> 8) & 0x00FF00FF00FF00FFul);
-	res = (res & 0x0000FFFF0000FFFFul) + ((res >> 16) & 0x0000FFFF0000FFFFul);
-	return (res & 0x00000000FFFFFFFFul) + ((res >> 32) & 0x00000000FFFFFFFFul);
-#endif
-}
-
 static inline unsigned long hweight_long(unsigned long w)
 {
-	return sizeof(w) == 4 ? generic_hweight32(w) : generic_hweight64(w);
+	return sizeof(w) == 4 ? hweight32(w) : hweight64(w);
 }
 
 /*

Re: [PATCH 4/6] use include/asm-generic/bitops for each architecture

[Akinobu Mita]

On Wed, Jan 25, 2006 at 08:33:37PM +0900, mita wrote:
> compile test on i386, x86_64, ppc, sparc, sparc64, alpha
> boot test on i386, x86_64, ppc
> 

I have fogotten attaching the changes for each archtecture.

o alpha

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_FFZ_BITOPS
- define HAVE_ARCH___FFS_BITOPS
- define HAVE_ARCH_FFS_BITOPS

- if defined(__alpha_cix__) and defined(__alpha_fix__)
  - define HAVE_ARCH_FLS_BITOPS
  - define HAVE_ARCH_HWEIGHT_BITOPS
  - define HAVE_ARCH_HWEIGHT64_BITOPS
- else
  - remove fls()
  - remove hweight64()
  - remove hweight{32,16,8}()

- remove fls64()
- remove find_{next,first}{,_zero}_bit()
- define HAVE_ARCH_SCHED_BITOPS
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o arm

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_FIND_BITOPS

- if __LINUX_ARM_ARCH__ < 5
  - remove ffz()
  - remove __ffs()
  - remove fls()
  - remove ffs()
- else (__LINUX_ARM_ARCH__ >= 5)
  - define HAVE_ARCH_FLS_BITOPS
  - define HAVE_ARCH_FFS_BITOPS
  - define HAVE_ARCH___FFS_BITOPS
  - define HAVE_ARCH_FFZ_BITOPS

- remove fls64()
- remove hweight64()
- remove hweight{32,16,8}()
- remove sched_find_first_bit()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove HAVE_ARCH_MINIX_BITOPS

o arm26

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_FIND_BITOPS
- remove ffz()
- remove __ffs()
- remove fls()
- remove fls64()
- remove ffs()
- remove sched_find_first_bit()
- remove hweight{32,16,8}()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- define HAVE_MINIX_BITOPS

o cris

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- remove fls()
- remove fls64()
- remove hweight{32,16,8}()
- remove find_{next,first}{,_zero}_bit()
- define HAVE_ARCH_FFS_BITOPS
- remove sched_find_first_bit()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o frv

- remove ffz()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_NON_ATOMIC_BITOPS
- remove find_{next,first}{,_zero}_bit()
- remove ffs()
- remove __ffs()
- remove fls64()
- remove sched_find_first_bit()
- remove hweight{32,16,8}()
- define HAVE_ARCH_FLS_BITOPS
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- define HAVE_ARCH_MINIX_BITOPS

o h8300

- define HAVE_ARCH_FFZ_BITOPS
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_NON_ATOMIC_BITOPS
- remove ffs()
- remove find_{next,first}{,_zero}_bit()
- remove sched_find_first_bit()
- remove hweight{32,16,8}()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- remove ext2_{set,clear}_bit_atomic()
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()
- define HAVE_ARCH___FFS_BITOPS
- remove fls()
- remove fls64()

o i386

- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_NON_ATOMIC_BITOPS
- define HAVE_ARCH_FIND_BITOPS
- define HAVE_ARCH___FFS_BITOPS
- define HAVE_ARCH_FFZ_BITOPS
- remove fls64()
- remove sched_find_first_bit()
- define HAVE_ARCH_FFS_BITOPS
- remove hweight{32,16,8}()
- define HAVE_ARCH_FLS_BITOPS
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o ia64

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_FFZ_BITOPS
- define HAVE_ARCH___FFS_BITOPS
- remove fls64()
- define HAVE_ARCH_FLS_BITOPS
- define HAVE_ARCH_FFS_BITOPS
- define HAVE_ARCH_HWEIGHT_BITOPS
- define HAVE_ARCH_HWEIGHT64_BITOPS
- define HAVE_ARCH_FIND_BITOPS
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()
- remove sched_find_first_bit()

o m32r

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- remove ffz()
- remove find_{next,first}{,_zero}_bit()
- remove __ffs()
- remove fls()
- remove fls64()
- remove sched_find_first_bit()
- remove ffs()
- remove hweight()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- remove ext2_{set,clear}_bit_atomic()
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()
- define HAVE_ARCH_ATOMIC_BITOPS

o m68k

- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_NON_ATOMIC_BITOPS
- define HAVE_ARCH_FIND_BITOPS
- define HAVE_ARCH_FFZ_BITOPS
- define HAVE_ARCH_FFS_BITOPS
- define HAVE_ARCH___FFS_BITOPS
- remove fls64()
- remove sched_find_first_bit()
- remove ffs()
- remove hweight()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_MINIX_BITOPS
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS

o m68knommu

- remove ffs()
- remove __ffs()
- remove sched_find_first_bit()
- remove ffz()
- remove find_{next,first}{,_zero}_bit()
- remove hweight()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_NON_ATOMIC_BITOPS
- define HAVE_ARCH_EXT2_NON_ATOMIC_BITOPS
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()
- remove fls()
- remove fls64()

o mips

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS

- if defined(CONFIG_CPU_MIPS32) or defined(CONFIG_CPU_MIPS64)
  - define HAVE_ARCH___FFS_BITOPS
  - define HAVE_ARCH_FFS_BITOPS
  - define HAVE_ARCH_FFZ_BITOPS
  - define HAVE_ARCH_FLS_BITOPS
- else
  - remove __ffs()
  - remove ffs()
  - remove ffz()
  - remove fls()

- remove fls64()
- remove find_{next,first}{,_zero}_bit()
- remove sched_find_first_bit()
- remove hweight()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- remove ext2_{set,clear}_bit_atomic()
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o s390

- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_NON_ATOMIC_BITOPS
- define HAVE_ARCH_FFZ_BITOPS
- define HAVE_ARCH___FFS_BITOPS
- define HAVE_ARCH_FIND_BITOPS
- remove ffs()
- remove fls()
- remove fls64()
- remove hweight()
- remove hweight64()
- define HAVE_ARCH_SCHED_BITOPS
- define HAVE_ARCH_EXT2_NON_ATOMIC_BITOPS
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o sh

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_FFZ_BITOPS
- define HAVE_ARCH___FFS_BITOPS
- remove find_{next,first}{,_zero}_bit()
- remove ffs()
- remove hweight()
- remove sched_find_first_bit()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- remove ext2_{set,clear}_bit_atomic()
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()
- remove fls()
- remove fls64()

o sh64

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_FFZ_BITOPS
- remove __ffs()
- remove find_{next,first}{,_zero}_bit()
- remove hweight()
- remove sched_find_first_bit()
- remove ffs()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- remove ext2_{set,clear}_bit_atomic()
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()
- remove fls()
- remove fls64()

o sparc

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- remove ffz()
- remove __ffs()
- remove sched_find_first_bit()
- remove ffs()
- remove fls()
- remove fls64()
- remove hweight{32,16,8}()
- remove find_{next,first}{,_zero}_bit()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- remove ext2_{set,clear}_bit_atomic()
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o sparc64

- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_ATOMIC_BITOPS
- remove ffz()
- remove __ffs()
- remove fls()
- remove fls64()
- remove sched_find_first_bit()
- remove ffs()

- if defined(ULTRA_HAS_POPULATION_COUNT)
  - define HAVE_ARCH_HWEIGHT64_BITOPS
  - define HAVE_ARCH_HWEIGHT_BITOPS
- else
  - remove hweight64()
  - remove hweight{32,16,8}()

- define HAVE_ARCH_FIND_BITOPS
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- define HAVE_ARCH_EXT2_NON_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o v850

- remove ffz()
- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_NON_ATOMIC_BITOPS
- remove find_{next,first}{,_zero}_bit()
- remove ffs()
- remove fls()
- remove fls64()
- remove __ffs()
- remove sched_find_first_bit()
- remove hweight{32,16,8}()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o x86_64

- define HAVE_ARCH_ATOMIC_BITOPS
- define HAVE_ARCH_NON_ATOMIC_BITOPS
- define HAVE_ARCH_FIND_BITOPS
- define HAVE_ARCH_FFZ_BITOPS
- define HAVE_ARCH___FFS_BITOPS
- define HAVE_ARCH_FLS_BITOPS
- remove sched_find_first_bit()
- define HAVE_ARCH_FFS_BITOPS
- define HAVE_ARCH_FLS64_BITOPS
- remove hweight{32,16,8}()
- define HAVE_ARCH_FLS_BITOPS
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o xtensa

- remove {,test_and_}{set,clear,change}_bit()
- remove __{,test_and_}{set,clear,change}_bit() and test_bit()
- define HAVE_ARCH_FFZ_BITOPS
- define HAVE_ARCH___FFS_BITOPS
- define HAVE_ARCH_FFS_BITOPS
- define HAVE_ARCH_FLS_BITOPS
- remove fls64()
- remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
- define HAVE_ARCH_EXT2_ATOMIC_BITOPS
- remove hweight{32,16,8}()
- remove sched_find_first_bit()
- remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()

o remove unused generic bitops

Re: [PATCH 4/6] use include/asm-generic/bitops for each architecture

[Hirokazu Takata]

From: mita@miraclelinux.com (Akinobu Mita)
Subject: Re: [PATCH 4/6] use include/asm-generic/bitops for each architecture
Date: Thu, 26 Jan 2006 10:49:34 +0900
> On Wed, Jan 25, 2006 at 08:33:37PM +0900, mita wrote:
> > compile test on i386, x86_64, ppc, sparc, sparc64, alpha
> > boot test on i386, x86_64, ppc
...
>
> o m32r
> 
> - remove __{,test_and_}{set,clear,change}_bit() and test_bit()
> - remove ffz()
> - remove find_{next,first}{,_zero}_bit()
> - remove __ffs()
> - remove fls()
> - remove fls64()
> - remove sched_find_first_bit()
> - remove ffs()
> - remove hweight()
> - remove ext2_{set,clear,test,find_first_zero,find_next_zero}_bit()
> - remove ext2_{set,clear}_bit_atomic()
> - remove minix_{test,set,test_and_clear,test,find_first_zero}_bit()
> - define HAVE_ARCH_ATOMIC_BITOPS
> 

compile and boot test on m32r: OK

Code size became a little bigger...  ;-)

$ size linux-2.6.16-rc1*/vmlinux
   text    data     bss     dec     hex filename
1768030  124412  721632 2614074  27e33a linux-2.6.16-rc1.bitops/vmlinux
1755010  124412  721632 2601054  27b05e linux-2.6.16-rc1.org/vmlinux

-- Takata

Re: [PATCH 4/6] use include/asm-generic/bitops for each architecture

[Akinobu Mita]

On Fri, Jan 27, 2006 at 10:04:01PM +0900, Hirokazu Takata wrote:

> compile and boot test on m32r: OK

Thanks.

> Code size became a little bigger...  ;-)
> 
> $ size linux-2.6.16-rc1*/vmlinux
>    text    data     bss     dec     hex filename
> 1768030  124412  721632 2614074  27e33a linux-2.6.16-rc1.bitops/vmlinux
> 1755010  124412  721632 2601054  27b05e linux-2.6.16-rc1.org/vmlinux

The only difference I can find is __ffs()/ffz().
As Russel King clealy pointed out, it will generate larger code
than before.