Re: Audit 2.6 set_pte users

[Andrew Morton <akpm@osdl.org>]

Anton Blanchard <anton@samba.org> wrote:
>
> 
> Hi,
> 
> I went through all the users of set_pte to check if they flush the
> current pte if it is present. Below is a summary of the audit,
> everything looks good except for a failure case in
> dup_mmap->copy_page_range.

I was hoping this might fix the "missing TLB flush" which Martin
Schwidefsky believes is there, and which is causing him grief.

> --- 1.154/kernel/fork.c	Tue Jan 20 10:38:15 2004
> +++ edited/kernel/fork.c	Sat Jan 24 14:17:00 2004
> @@ -347,6 +347,7 @@
>  fail_nomem:
>  	retval = -ENOMEM;
>  fail:
> +	flush_tlb_mm(current->mm);
>  	vm_unacct_memory(charge);
>  	goto out;
>  }

But look:

 	retval = 0;
 	build_mmap_rb(mm);
 
 out:
 	flush_tlb_mm(current->mm);
 	up_write(&oldmm->mmap_sem);
 	return retval;
 fail_nomem:
 	retval = -ENOMEM;
 fail:
+	flush_tlb_mm(current->mm);
 	vm_unacct_memory(charge);
 	goto out;
 }


There is no missing flush here.