Re: [PATCH] ACPI: Move timer broadcast and pmtimer access before C3 arbiter shutdown

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 06 Jun 2007 11:37:53 +0200 Thomas Gleixner <tglx@linutronix.de> wrote:

> From: Udo A. Steinberg <us15@os.inf.tu-dresden.de>
> 
> The chip set doc for IHC4 says:
> 
> 1.In general, software should not attempt any non-posted accesses during
> arbiter disable except to the ICH4's power management registers. This
> implies that interrupt handlers for any unmasked hardware interrupts and
> SMI/NMI should check ARB_DIS status before reading from ICH devices.
> 
> So it's not a good idea to access ICH devices after arbiter shut down. 
> 
> Signed-off-by: Udo A. Steinberg <us15@os.inf.tu-dresden.de>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> 
> ---
>  drivers/acpi/processor_idle.c |    9 +++++----
>  1 file changed, 5 insertions(+), 4 deletions(-)
> 
> Index: linux-2.6.22-rc4/drivers/acpi/processor_idle.c
> ===================================================================
> --- linux-2.6.22-rc4.orig/drivers/acpi/processor_idle.c	2007-06-06 11:47:21.000000000 +0200
> +++ linux-2.6.22-rc4/drivers/acpi/processor_idle.c	2007-06-06 11:48:21.000000000 +0200
> @@ -488,6 +488,11 @@ static void acpi_processor_idle(void)
>  
>  	case ACPI_STATE_C3:
>  
> +		/* Get start time (ticks) */
> +		t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
> +		/* Handle timer broadcast before bus arbiter shutdown ! */
> +		acpi_state_timer_broadcast(pr, cx, 1);
> +
>  		if (pr->flags.bm_check) {
>  			if (atomic_inc_return(&c3_cpu_count) ==
>  			    num_online_cpus()) {
> @@ -502,10 +507,7 @@ static void acpi_processor_idle(void)
>  			ACPI_FLUSH_CPU_CACHE();
>  		}
>  
> -		/* Get start time (ticks) */
> -		t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
>  		/* Invoke C3 */
> -		acpi_state_timer_broadcast(pr, cx, 1);
>  		acpi_cstate_enter(cx);
>  		/* Get end time (ticks) */
>  		t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);

hm, this needs a bit of help to get it to work against Len's current tree.

However, if by "non-posted accesses" you're referring to that inl(), how
come the second one which was left in place isn't also a problem?