Re: [discuss] Re: [PATCH][Fix][Resend] Fix Bug #4959: Page tables corrupted during resume on x86-64 (take 3)

["Siddha, Suresh B" <suresh.b.siddha@intel.com>]

On Fri, Sep 30, 2005 at 08:51:08AM +0200, Rafael J. Wysocki wrote:
> On Friday, 30 of September 2005 07:26, Rafael J. Wysocki wrote:
> > Of course.  The .config is attached.  Generally, it's a non-SMP box, and commenting
> > out the zap_low_mappings((0) in setup_arch() makes the box boot again.
> 
> One more datapoint: The box boots if I move the zap_low_mappings((0)
> in the following way:
> 
> --- linux-2.6.14-rc2-git7.orig/arch/x86_64/kernel/setup.c	2005-09-30 07:39:35.000000000 +0200
> +++ linux-2.6.14-rc2-git7/arch/x86_64/kernel/setup.c	2005-09-30 08:31:20.000000000 +0200
> @@ -571,8 +571,6 @@
>  
>  	init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
>  
> -	zap_low_mappings(0);
> -
>  #ifdef CONFIG_ACPI
>  	/*
>  	 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
> @@ -680,6 +678,8 @@
>  		get_smp_config();
>  	init_apic_mappings();
>  #endif
> +	zap_low_mappings(0);
> +

Rafael, I still can't reproduce the issue. All of my EM64T systems are SMP
and looks like this problem happens only on an UP system. Looks like
there is some code flow which is still assuming the presence of low mappings.

Andi, can you please see if you can reproduce this on one of your systems
similar to Rafael's.

>  	/*
>  	 * Request address space for all standard RAM and ROM resources
> 
> iHowever, if I place the zap_low_mappings((0) before the
> #define CONFIG_X86_LOCAL_APIC in line 673, it doesn't boot.
> 
> Certainly init_apic_mappings() is at fault.  Could that be a reult of a call to
> alloc_bootmem_pages()?.

I looked at init_apic_mappings() and didn't give me any clue.
alloc_bootmem_pages() don't use low direct mappings.

> And one more: I have to boot with "noapic".

Are you saying "noapic" is the second data point or you need to use "noapic"
along with moving zap_low_mappings(0) 

thanks,
suresh