[PATCH 6/8] UML - Kill some useless vmalloc tlb flushing

[PATCH 6/8] UML - Kill some useless vmalloc tlb flushing

[Jeff Dike]

There is absolutely no reason to flush the kernel's VM area during a
tlb_flush_mm.

This results in a noticable performance increase in the kernel build 
benchmark.

Signed-off-by: Jeff Dike <jdike@addtoit.com>

Index: linux-2.6.12/arch/um/kernel/skas/tlb.c
===================================================================
--- linux-2.6.12.orig/arch/um/kernel/skas/tlb.c	2005-06-20 11:54:56.000000000 -0400
+++ linux-2.6.12/arch/um/kernel/skas/tlb.c	2005-06-20 12:11:00.000000000 -0400
@@ -76,7 +76,6 @@ void flush_tlb_mm_skas(struct mm_struct 
                 return;
 
         fix_range(mm, 0, host_task_size, 0);
-        flush_tlb_kernel_range_common(start_vm, end_vm);
 }
 
 void force_flush_all_skas(void)

Re: [uml-devel] [PATCH 6/8] UML - Kill some useless vmalloc tlb flushing

[Blaisorblade]

On Monday 20 June 2005 20:51, Jeff Dike wrote:
> There is absolutely no reason to flush the kernel's VM area during a
> tlb_flush_mm.
>
> This results in a noticable performance increase in the kernel build
> benchmark.
Andrew: hold off this one, definitely.

Jeff: Have you verified this with both modules enabled and iptables (a vmalloc 
user) compiled modularly? Maybe even non-modular iptables will trigger the 
bug but let's go for sure.

This situation killed a conceptually similar patch in 2.4.24-2um (which did 
apply until 2.6.11):

diff -puN arch/um/kernel/skas/tlb.c~optimization-unstable 
arch/um/kernel/skas/tlb.c
--- UmWorklinux-2.4.24/arch/um/kernel/skas/tlb.c~optimization-unstable  
2004-07-02 13:45:07.057643968 +0200
+++ UmWorklinux-2.4.24-paolo/arch/um/kernel/skas/tlb.c  2004-07-02 
13:45:14.098573584 +0200
@@ -132,7 +132,9 @@ void flush_tlb_range_skas(struct mm_stru

 void flush_tlb_mm_skas(struct mm_struct *mm)
 {
+#if 0
        flush_tlb_kernel_vm_skas();
+#endif
        fix_range(mm, 0, host_task_size, 0);
 }

Given that:

void flush_tlb_kernel_vm_skas(void)
{
        flush_tlb_kernel_range_skas(start_vm, end_vm);
}

and flush_tlb_kernel_range_skas was renamed to _common, I argue that this 
patch is exactly the same one and will have the same bad effect.

> Signed-off-by: Jeff Dike <jdike@addtoit.com>
>
> Index: linux-2.6.12/arch/um/kernel/skas/tlb.c
> ===================================================================
> --- linux-2.6.12.orig/arch/um/kernel/skas/tlb.c	2005-06-20
> 11:54:56.000000000 -0400 +++
> linux-2.6.12/arch/um/kernel/skas/tlb.c	2005-06-20 12:11:00.000000000 -0400
> @@ -76,7 +76,6 @@ void flush_tlb_mm_skas(struct mm_struct
>                  return;
>
>          fix_range(mm, 0, host_task_size, 0);
> -        flush_tlb_kernel_range_common(start_vm, end_vm);
>  }
>
>  void force_flush_all_skas(void)

-- 
Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
Paolo Giarrusso, aka Blaisorblade (Skype ID "PaoloGiarrusso", ICQ 215621894)
http://www.user-mode-linux.org/~blaisorblade

	

	
		
___________________________________ 
Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB 
http://mail.yahoo.it