RE: [PATCH] 'noapic' already handled elsewhere

RE: [PATCH] 'noapic' already handled elsewhere

[Brown, Len]

Jeff,
This won't work.
acpi_boot_init() is called from setup_arch(), which is called from
start_kernel() _before_ parse_options().  Ie. ACPI needs to consume this
flag before __setup() is invoked.

-Len

> -----Original Message-----
> From: Jeff Garzik [mailto:jgarzik@pobox.com] 
> Sent: Thursday, August 21, 2003 12:15 PM
> To: torvalds@osdl.org
> Cc: Brown, Len; Grover, Andrew; zwane@linuxpower.ca; 
> linux-kernel@vger.kernel.org
> Subject: [PATCH] 'noapic' already handled elsewhere
> 
> 
> I sent a previous patch to s/LOCAL_APIC/IO_APIC/, as Zwane noticed
> my first patch needed that.  In that patch, I commented __setup()
> would be better.
> 
> Well... line 718 of arch/i386/kernel/io_apic.c _already_ handles this
> case, using __setup() properly.
> 
> Word of warning... patch only compile tested, but seems obvious from
> looking at io_apic.c.
> 
> BTW, why isn't ACPI using __setup() as well?  I don't see that ACPI
> needs to patch arch/i386/kernel/setup.c at all.
> 
> 
> ===== arch/i386/kernel/setup.c 1.94 vs edited =====
> --- 1.94/arch/i386/kernel/setup.c	Thu Aug 21 01:32:04 2003
> +++ edited/arch/i386/kernel/setup.c	Thu Aug 21 12:09:13 2003
> @@ -544,12 +544,6 @@
>  			if (!acpi_force) acpi_disabled = 1;
>  		}
>  
> -#ifdef CONFIG_X86_LOCAL_APIC
> -		/* disable IO-APIC */
> -		else if (!memcmp(from, "noapic", 6)) {
> -			skip_ioapic_setup = 1;
> -		}
> -#endif /* CONFIG_X86_LOCAL_APIC */
>  #endif /* CONFIG_ACPI_BOOT */
>  
>  		/*
> 

Re: [PATCH] 'noapic' already handled elsewhere

[Jeff Garzik]

Brown, Len wrote:
> Jeff,
> This won't work.
> acpi_boot_init() is called from setup_arch(), which is called from
> start_kernel() _before_ parse_options().  Ie. ACPI needs to consume this
> flag before __setup() is invoked.


Thanks for the correction.  I'll resend the earlier 
s/LOCAL_APIC/IO_APIC/ patch then.

Just found another ACPI bug in 2.6:

> config ACPI_HT
>         bool "ACPI Processor Enumeration for HT"
>         depends on (X86 && X86_LOCAL_APIC)
>         default y
[...]
> config ACPI
>         bool "Full ACPI Support"
>         depends on !X86_VISWS
>         depends on !IA64_HP_SIM
>         depends on IA64 || (X86 && ACPI_HT)


So CONFIG_ACPI is not allowed on uniprocessor anymore, _and_ it requires 
HyperThreading code?  ;-)  No wonder CONFIG_ACPI didn't appear for my 
uniprocessor Pentium3 'make oldconfig'  ;-)

(ACPI requires ACPI_HT, which requires LOCAL_APIC)

Another reason why I was saying that CONFIG_ACPI should be the toplevel 
config option (even if CONFIG_ACPI never actually appears in any code, 
but only in Kconfig)...

	Jeff

RE: [PATCH] 'noapic' already handled elsewhere

[Brown, Len]

Re: config
Yes, I fixed this they way you suggested yesterday, both in 2.4 and 2.6.
The 2.4 tree is being tested, and the 2.6 patch is being buttoned up
now.


Thanks,
-Len

Ps   I sent a note on this to acpi-devel yesterday, but just got a
message from "Mail Delivery System [Mailer-Daemon@vasoftware.com]" that
my message has been waiting to go out for over 8 hours...


> -----Original Message-----
> From: Jeff Garzik [mailto:jgarzik@pobox.com] 
> Sent: Thursday, August 21, 2003 5:11 PM
> To: Brown, Len
> Cc: torvalds@osdl.org; Grover, Andrew; zwane@linuxpower.ca; 
> linux-kernel@vger.kernel.org
> Subject: Re: [PATCH] 'noapic' already handled elsewhere
> 
> 
> Brown, Len wrote:
> > Jeff,
> > This won't work.
> > acpi_boot_init() is called from setup_arch(), which is called from
> > start_kernel() _before_ parse_options().  Ie. ACPI needs to 
> consume this
> > flag before __setup() is invoked.
> 
> 
> Thanks for the correction.  I'll resend the earlier 
> s/LOCAL_APIC/IO_APIC/ patch then.
> 
> Just found another ACPI bug in 2.6:
> 
> > config ACPI_HT
> >         bool "ACPI Processor Enumeration for HT"
> >         depends on (X86 && X86_LOCAL_APIC)
> >         default y
> [...]
> > config ACPI
> >         bool "Full ACPI Support"
> >         depends on !X86_VISWS
> >         depends on !IA64_HP_SIM
> >         depends on IA64 || (X86 && ACPI_HT)
> 
> 
> So CONFIG_ACPI is not allowed on uniprocessor anymore, _and_ 
> it requires 
> HyperThreading code?  ;-)  No wonder CONFIG_ACPI didn't appear for my 
> uniprocessor Pentium3 'make oldconfig'  ;-)
> 
> (ACPI requires ACPI_HT, which requires LOCAL_APIC)
> 
> Another reason why I was saying that CONFIG_ACPI should be 
> the toplevel 
> config option (even if CONFIG_ACPI never actually appears in 
> any code, 
> but only in Kconfig)...
> 
> 	Jeff
> 
> 
> 
> 

[PATCH] 'noapic' already handled elsewhere

[Jeff Garzik]

I sent a previous patch to s/LOCAL_APIC/IO_APIC/, as Zwane noticed
my first patch needed that.  In that patch, I commented __setup()
would be better.

Well... line 718 of arch/i386/kernel/io_apic.c _already_ handles this
case, using __setup() properly.

Word of warning... patch only compile tested, but seems obvious from
looking at io_apic.c.

BTW, why isn't ACPI using __setup() as well?  I don't see that ACPI
needs to patch arch/i386/kernel/setup.c at all.


===== arch/i386/kernel/setup.c 1.94 vs edited =====
--- 1.94/arch/i386/kernel/setup.c	Thu Aug 21 01:32:04 2003
+++ edited/arch/i386/kernel/setup.c	Thu Aug 21 12:09:13 2003
@@ -544,12 +544,6 @@
 			if (!acpi_force) acpi_disabled = 1;
 		}
 
-#ifdef CONFIG_X86_LOCAL_APIC
-		/* disable IO-APIC */
-		else if (!memcmp(from, "noapic", 6)) {
-			skip_ioapic_setup = 1;
-		}
-#endif /* CONFIG_X86_LOCAL_APIC */
 #endif /* CONFIG_ACPI_BOOT */
 
 		/*