[PATCH] x86: Don't write io_apic ID if it is not changed

[PATCH] x86: Don't write io_apic ID if it is not changed

[Yinghai Lu]

For 32bit mptable path, setup_ids_from_mpc() always write io apic id
register, even there is no change needed.

So try to do that when they are different bewteen reading out and mptable

Signed-off-by: Yinghai Lu <yinghai@kernel.org>

---
 arch/x86/kernel/apic/io_apic.c |    8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

Index: linux-2.6/arch/x86/kernel/apic/io_apic.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/apic/io_apic.c
+++ linux-2.6/arch/x86/kernel/apic/io_apic.c
@@ -2125,7 +2125,6 @@ void __init setup_ioapic_ids_from_mpc(vo
 			physids_or(phys_id_present_map, phys_id_present_map, tmp);
 		}
 
-
 		/*
 		 * We need to adjust the IRQ routing table
 		 * if the ID changed.
@@ -2137,9 +2136,12 @@ void __init setup_ioapic_ids_from_mpc(vo
 						= mp_ioapics[apic_id].apicid;
 
 		/*
-		 * Read the right value from the MPC table and
-		 * write it into the ID register.
+		 * Update the ID register according to the right value from
+		 *  the MPC table if they are different.
 		 */
+		if (mp_ioapics[apic_id].apicid == reg_00.bits.ID)
+			continue;
+
 		apic_printk(APIC_VERBOSE, KERN_INFO
 			"...changing IO-APIC physical APIC ID to %d ...",
 			mp_ioapics[apic_id].apicid);

Re: [PATCH] x86: Don't write io_apic ID if it is not changed

[Cyrill Gorcunov]

On Sat, Aug 14, 2010 at 01:38:01PM -0700, Yinghai Lu wrote:
> 
> For 32bit mptable path, setup_ids_from_mpc() always write io apic id
> register, even there is no change needed.
> 
> So try to do that when they are different bewteen reading out and mptable
> 
> Signed-off-by: Yinghai Lu <yinghai@kernel.org>
> 
> ---
>  arch/x86/kernel/apic/io_apic.c |    8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> Index: linux-2.6/arch/x86/kernel/apic/io_apic.c
> ===================================================================
> --- linux-2.6.orig/arch/x86/kernel/apic/io_apic.c
> +++ linux-2.6/arch/x86/kernel/apic/io_apic.c
> @@ -2125,7 +2125,6 @@ void __init setup_ioapic_ids_from_mpc(vo
>  			physids_or(phys_id_present_map, phys_id_present_map, tmp);
>  		}
>  
> -
>  		/*
>  		 * We need to adjust the IRQ routing table
>  		 * if the ID changed.
> @@ -2137,9 +2136,12 @@ void __init setup_ioapic_ids_from_mpc(vo
>  						= mp_ioapics[apic_id].apicid;
>  
>  		/*
> -		 * Read the right value from the MPC table and
> -		 * write it into the ID register.
> +		 * Update the ID register according to the right value from
> +		 *  the MPC table if they are different.
>  		 */
> +		if (mp_ioapics[apic_id].apicid == reg_00.bits.ID)
> +			continue;
> +
>  		apic_printk(APIC_VERBOSE, KERN_INFO
>  			"...changing IO-APIC physical APIC ID to %d ...",
>  			mp_ioapics[apic_id].apicid);

As far as I see, this was done by a purpose in former code. Consider the
situation when mp_ioapics[apic_id].apicid >= get_physical_broadcast().
By code flow (io_apic.c:2099) this set

		if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
			printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
				apic_id, mp_ioapics[apic_id].apicid);
			printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
				reg_00.bits.ID);
--->			mp_ioapics[apic_id].apicid = reg_00.bits.ID;
		}

So with your patch we always hit "continue" without real changing of ID which
is not correct.

But perhaps I miss something?

	-- Cyrill

Re: [PATCH] x86: Don't write io_apic ID if it is not changed

[Cyrill Gorcunov]

On Sun, Aug 15, 2010 at 12:24:46PM +0400, Cyrill Gorcunov wrote:
...
> As far as I see, this was done by a purpose in former code. Consider the
> situation when mp_ioapics[apic_id].apicid >= get_physical_broadcast().
> By code flow (io_apic.c:2099) this set
> 
> 		if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
> 			printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
> 				apic_id, mp_ioapics[apic_id].apicid);
> 			printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
> 				reg_00.bits.ID);
> --->			mp_ioapics[apic_id].apicid = reg_00.bits.ID;
> 		}
> 
> So with your patch we always hit "continue" without real changing of ID which
> is not correct.
> 
> But perhaps I miss something?
> 
> 	-- Cyrill

False alarm Yinghai, of course there is no need to write same value back,
sorry for noise.

	-- Cyrill

Re: [PATCH] x86: Don't write io_apic ID if it is not changed

[Eric W. Biederman]

Yinghai Lu <yinghai@kernel.org> writes:

> For 32bit mptable path, setup_ids_from_mpc() always write io apic id
> register, even there is no change needed.
>
> So try to do that when they are different bewteen reading out and
> mptable

What is the advantage of not writing the value?

Eric

Re: [PATCH] x86: Don't write io_apic ID if it is not changed

[Ingo Molnar]

* Eric W. Biederman <ebiederm@xmission.com> wrote:

> Yinghai Lu <yinghai@kernel.org> writes:
> 
> > For 32bit mptable path, setup_ids_from_mpc() always write io apic id
> > register, even there is no change needed.
> >
> > So try to do that when they are different bewteen reading out and
> > mptable
> 
> What is the advantage of not writing the value?

Just guessing, but if other OSs dont do it we might surprise buggy hardware 
[and trigger erratas in them], so this would be a standard 'dont touch the hw 
unless necessary' defensive move.

Similar bugs are not unprecedented.

Thanks,

	Ingo

Re: [PATCH] x86: Don't write io_apic ID if it is not changed

[Cyrill Gorcunov]

On Sun, Aug 15, 2010 at 12:30:57PM +0400, Cyrill Gorcunov wrote:
> On Sun, Aug 15, 2010 at 12:24:46PM +0400, Cyrill Gorcunov wrote:
> ...
> > As far as I see, this was done by a purpose in former code. Consider the
> > situation when mp_ioapics[apic_id].apicid >= get_physical_broadcast().
> > By code flow (io_apic.c:2099) this set
> > 
> > 		if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
> > 			printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
> > 				apic_id, mp_ioapics[apic_id].apicid);
> > 			printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
> > 				reg_00.bits.ID);
> > --->			mp_ioapics[apic_id].apicid = reg_00.bits.ID;
> > 		}
> > 
> > So with your patch we always hit "continue" without real changing of ID which
> > is not correct.
> > 
> > But perhaps I miss something?
> > 
> > 	-- Cyrill
> 
> False alarm Yinghai, of course there is no need to write same value back,
> sorry for noise.
> 
> 	-- Cyrill

Thinking a bit more I guess there might be a sence to write io-apic id
unconditionally, io-apic id is not latched on reset but set to 0 (at least
for old apics) so if we hit id match -- this will be once, but the same
time we will have additional if() for every cycle iteration, not sure
it worth it. Right?

	-- Cyrill