[patch -mm] fix SLOB on x64

[patch -mm] fix SLOB on x64

[Ingo Molnar]

this patch fixes 32-bitness bugs in mm/slob.c. Successfully booted x64 
with SLOB enabled. (i have switched the PREEMPT_RT feature to use the 
SLOB allocator exclusively, so it must work on all platforms)

Signed-off-by: Ingo Molnar <mingo@elte.hu>

Index: linux/mm/slob.c
===================================================================
--- linux.orig/mm/slob.c
+++ linux/mm/slob.c
@@ -198,7 +198,7 @@ void kfree(const void *block)
 	if (!block)
 		return;
 
-	if (!((unsigned int)block & (PAGE_SIZE-1))) {
+	if (!((unsigned long)block & (PAGE_SIZE-1))) {
 		/* might be on the big block list */
 		spin_lock_irqsave(&block_lock, flags);
 		for (bb = bigblocks; bb; last = &bb->next, bb = bb->next) {
@@ -227,7 +227,7 @@ unsigned int ksize(const void *block)
 	if (!block)
 		return 0;
 
-	if (!((unsigned int)block & (PAGE_SIZE-1))) {
+	if (!((unsigned long)block & (PAGE_SIZE-1))) {
 		spin_lock_irqsave(&block_lock, flags);
 		for (bb = bigblocks; bb; bb = bb->next)
 			if (bb->pages == block) {
@@ -326,7 +326,7 @@ void kmem_cache_init(void)
 	void *p = slob_alloc(PAGE_SIZE, 0, PAGE_SIZE-1);
 
 	if (p)
-		free_page((unsigned int)p);
+		free_page((unsigned long)p);
 
 	mod_timer(&slob_timer, jiffies + HZ);
 }

Re: [patch -mm] fix SLOB on x64

[Ed Tomlinson]

On Sunday 11 December 2005 09:12, Ingo Molnar wrote:
> this patch fixes 32-bitness bugs in mm/slob.c. Successfully booted x64 
> with SLOB enabled. (i have switched the PREEMPT_RT feature to use the 
> SLOB allocator exclusively, so it must work on all platforms)

Its a good idea to get this working everywhere.  Why have you switched to 
use SLOB exclusively?

Thanks
Ed Tomlinson

Re: [patch -mm] fix SLOB on x64

[Ingo Molnar]

* Ed Tomlinson <edt@aei.ca> wrote:

> On Sunday 11 December 2005 09:12, Ingo Molnar wrote:
> > this patch fixes 32-bitness bugs in mm/slob.c. Successfully booted x64 
> > with SLOB enabled. (i have switched the PREEMPT_RT feature to use the 
> > SLOB allocator exclusively, so it must work on all platforms)
> 
> Its a good idea to get this working everywhere.  Why have you switched 
> to use SLOB exclusively?

because the SLAB hacks were getting ugly, and i gave up on it during the 
2.6.15-rc5 merge. (The SLAB code has lots of irqs-off / per-cpu and 
non-preempt assumptions integrated, which were a pain to sort out.)

We'll eventually do a cleaner conversion of SLAB to PREEMPT_RT, but for 
now the SLOB is turned on exclusively if PREEMPT_RT. (in other 
preemption modes it's optionally selectable if EMBEDDED is enabled)

	Ingo

Re: [patch -mm] fix SLOB on x64

[Matt Mackall]

On Sun, Dec 11, 2005 at 03:12:17PM +0100, Ingo Molnar wrote:
> 
> this patch fixes 32-bitness bugs in mm/slob.c. Successfully booted x64 
> with SLOB enabled. (i have switched the PREEMPT_RT feature to use the 
> SLOB allocator exclusively, so it must work on all platforms)

The patch looks fine, but what's this about using SLOB exclusively?
Fragmentation performance of SLOB is miserable on anything like a
modern desktop, I think SLOB only makes sense for small machines. The
locking also suggests dual core at most.

Anyway,

> Signed-off-by: Ingo Molnar <mingo@elte.hu>

Acked-by: Matt Mackall <mpm@selenic.com>
> 
> Index: linux/mm/slob.c
> ===================================================================
> --- linux.orig/mm/slob.c
> +++ linux/mm/slob.c
> @@ -198,7 +198,7 @@ void kfree(const void *block)
>  	if (!block)
>  		return;
>  
> -	if (!((unsigned int)block & (PAGE_SIZE-1))) {
> +	if (!((unsigned long)block & (PAGE_SIZE-1))) {
>  		/* might be on the big block list */
>  		spin_lock_irqsave(&block_lock, flags);
>  		for (bb = bigblocks; bb; last = &bb->next, bb = bb->next) {
> @@ -227,7 +227,7 @@ unsigned int ksize(const void *block)
>  	if (!block)
>  		return 0;
>  
> -	if (!((unsigned int)block & (PAGE_SIZE-1))) {
> +	if (!((unsigned long)block & (PAGE_SIZE-1))) {
>  		spin_lock_irqsave(&block_lock, flags);
>  		for (bb = bigblocks; bb; bb = bb->next)
>  			if (bb->pages == block) {
> @@ -326,7 +326,7 @@ void kmem_cache_init(void)
>  	void *p = slob_alloc(PAGE_SIZE, 0, PAGE_SIZE-1);
>  
>  	if (p)
> -		free_page((unsigned int)p);
> +		free_page((unsigned long)p);
>  
>  	mod_timer(&slob_timer, jiffies + HZ);
>  }

-- 
Mathematics is the supreme nostalgia of our time.

Re: [patch -mm] fix SLOB on x64

[Ingo Molnar]

* Matt Mackall <mpm@selenic.com> wrote:

> On Sun, Dec 11, 2005 at 03:12:17PM +0100, Ingo Molnar wrote:
> > 
> > this patch fixes 32-bitness bugs in mm/slob.c. Successfully booted x64 
> > with SLOB enabled. (i have switched the PREEMPT_RT feature to use the 
> > SLOB allocator exclusively, so it must work on all platforms)
> 
> The patch looks fine, but what's this about using SLOB exclusively?  
> Fragmentation performance of SLOB is miserable on anything like a 
> modern desktop, I think SLOB only makes sense for small machines. The 
> locking also suggests dual core at most.

well, this is only an -rt artifact: the SLOB needs zero modifications to 
work on PREEMPT_RT, while SLAB needed a risky 66K monster patch. Until 
someone simplifies the SLAB conversion to PREEMPT_RT, i'll use the SLOB.

i havent noticed any significant slowdown due to the SLOB. In any case, 
we'll give it some workout which should further speed up its upstream 
integration - it's looking good so far.

	Ingo