Skip to contentSkip to conversation

i386: port ATI timer fix from x86_64 to i386

PATCH Chuck Ebbert 3 participants 6 messages Mar 1, 2006 – Mar 2, 2006 Documentation

lore
Share & tools
lore
#0 lore raw

In-Reply-To: <[email protected]>

Quoted 11 lines · Andrew Morton
Quoted 7 lines · Chuck Ebbert

Disable timer routing over 8254 when an ATI chipset is detected
(autodetect is only implemented for ACPI, but these are new systems and should be using ACPI anyway.) Adds boot options for manually disabling and enabling this feature. Also adds a note to the timer error message caused by this change explaining that this error is expected on ATI chipsets.

umm, why did you write this patch? Presumably it's fixing something, but what?

Oops... I'm here in the middle of the forest and all I can see are trees.

This fixes the "timer runs too fast" bug on ATI chipsets (bugzilla #3927).

Signature · 2 lines
Chuck
"Equations are the Devil's sentences."  --Stephen Colbert
#1 lore raw
Quoted 2 lines · Chuck Ebbert

This fixes the "timer runs too fast" bug on ATI chipsets (bugzilla #3927).

Wonderful, thanks. What's the relationship (if any) between this and the recently-merged x86_64 fix?

#2 lore raw
Quoted 6 lines · Andrew Morton
Quoted 2 lines · Chuck Ebbert

This fixes the "timer runs too fast" bug on ATI chipsets (bugzilla #3927).

Wonderful, thanks. What's the relationship (if any) between this and the recently-merged x86_64 fix?

He just ported the x86-64 change over without any original authorship attribution :/

And some less functionality (only works for ACPI now) and some totally unrelated Documentation cleanup and a few random printk changes.

The ACPI only thing is probably mostly ok because the timing won't work at least on the dual cores without ACPI anyways because PMtimer is needed. On single cores it would be useful even without ACPI
(for that earlyquirk.c just would need to be moved up to run independently of ACPI)

Still it's probably a good idea for 2.6.16.

-Andi

#3 lore raw

(Cc: fixed. Please send me a copy of your MUA so I can ritually disembowel it).

Quoted 5 lines · Andi Kleen
Quoted 2 lines · Andrew Morton
Quoted 1 line · Chuck Ebbert
Quoted 7 lines
Quoted 4 lines
Quoted 1 line

This fixes the "timer runs too fast" bug on ATI chipsets (bugzilla #3927).

Wonderful, thanks. What's the relationship (if any) between this and the recently-merged x86_64 fix?

He just ported the x86-64 change over without any original authorship attribution :/

Yup. And he proved that I am incapable of understanding a simple email Subject:

Quoted 11 lines

And some less functionality (only works for ACPI now) and some totally unrelated Documentation cleanup and a few random printk changes.

The ACPI only thing is probably mostly ok because the timing won't work at least on the dual cores without ACPI anyways because PMtimer is needed. On single cores it would be useful even without ACPI
(for that earlyquirk.c just would need to be moved up to run independently of ACPI)

Still it's probably a good idea for 2.6.16.

Well.. the patch had a flagrant won't-compile if CONFIG_X86_IO_APIC=y, so I'd consider it a bit green.

#4 lore raw
Quoted 3 lines · Andrew Morton

(Cc: fixed. Please send me a copy of your MUA so I can ritually disembowel it).

Well it's an MNA really in this case - i read l-k using gated news with gnus.

Quoted 2 lines

Well.. the patch had a flagrant won't-compile if CONFIG_X86_IO_APIC=y, so I'd consider it a bit green.

Ok - i guess it would be better to redo it anyways to handle the non ACPI case too. I can do that.

-Andi

#5 lore raw

Re: [patch] i386: port ATI timer fix from x86_64 to i386 II

In reply to Andrew Morton
Quoted 2 lines · Andrew Morton

Well.. the patch had a flagrant won't-compile if CONFIG_X86_IO_APIC=y, so I'd consider it a bit green.

I did my own version of it then. Tested it a bit in various configurations and it works for me.

I would consider it a 2.6.16 candidate because it fixes an important bug.

Port the x86-64 ATI timer fix over to i386

ATI chipsets tend to generate double timer interrupts for the local APIC timer when both the 8254 and the IO-APIC timer pins are enabled. This is because they route it to both and the result is anded together and the CPU ends up processing it twice.

This patch changes check_timer to disable the 8254 routing for interruopt 0.

I think it would be safe on all chipsets actually (i tested it on a couple and it worked everywhere) and Windows seems to do it in a similar way, but to be conservative this patch only enables this mode on ATI (and adds options to enable/disable too)

Ported over from a similar x86-64 change.

I reused the ACPI earlyquirk infrastructure for the ATI bridge check, but tweaked it a bit to work even without ACPI.

Inspired by a patch from Chuck Ebbert, but redone.

arch/i386/kernel/Makefile +1 -1
Index: linux/arch/i386/kernel/Makefile
===================================================================
@@ -11,7 +11,7 @@ obj-y	:= process.o semaphore.o signal.o 
 
 obj-y				+= cpu/
 obj-y				+= timers/
-obj-$(CONFIG_ACPI)		+= acpi/
+obj-y				+= acpi/
 obj-$(CONFIG_X86_BIOS_REBOOT)	+= reboot.o
 obj-$(CONFIG_MCA)		+= mca.o
 obj-$(CONFIG_X86_MSR)		+= msr.o
arch/i386/kernel/acpi/Makefile +1 -1
Index: linux/arch/i386/kernel/acpi/Makefile
===================================================================
@@ -1,4 +1,4 @@
-obj-y				:= boot.o
+obj-$(CONFIG_ACPI)		+= boot.o
 obj-$(CONFIG_X86_IO_APIC)	+= earlyquirk.o
 obj-$(CONFIG_ACPI_SLEEP)	+= sleep.o wakeup.o
 
arch/i386/kernel/acpi/earlyquirk.c +8 -0
Index: linux/arch/i386/kernel/acpi/earlyquirk.c
===================================================================
@@ -7,14 +7,22 @@
 #include <linux/pci.h>
 #include <asm/pci-direct.h>
 #include <asm/acpi.h>
+#include <asm/apic.h>
 
 static int __init check_bridge(int vendor, int device)
 {
+#ifdef CONFIG_ACPI
 	/* According to Nvidia all timer overrides are bogus. Just ignore
 	   them all. */
 	if (vendor == PCI_VENDOR_ID_NVIDIA) {
 		acpi_skip_timer_override = 1;
 	}
+#endif
+	if (vendor == PCI_VENDOR_ID_ATI && timer_over_8254 == 1) {
+		timer_over_8254 = 0;
+		printk(KERN_INFO
+	"ATI board detected. Disabling timer routing over 8254.\n");
+	}    
 	return 0;
 }
 
arch/i386/kernel/io_apic.c +18 -1
Index: linux/arch/i386/kernel/io_apic.c
===================================================================
@@ -51,6 +51,8 @@ static struct { int pin, apic; } ioapic_
 
 static DEFINE_SPINLOCK(ioapic_lock);
 
+int timer_over_8254 __initdata = 1;
+
 /*
  *	Is the SiS APIC rmw bug present ?
  *	-1 = don't know, 0 = no, 1 = yes
@@ -2267,7 +2269,8 @@ static inline void check_timer(void)
 	apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
 	init_8259A(1);
 	timer_ack = 1;
-	enable_8259A_irq(0);
+	if (timer_over_8254 > 0)	
+		enable_8259A_irq(0);
 
 	pin1  = find_isa_irq_pin(0, mp_INT);
 	apic1 = find_isa_irq_apic(0, mp_INT);
@@ -2392,6 +2395,20 @@ void __init setup_IO_APIC(void)
 		print_IO_APIC();
 }
 
+static int __init setup_disable_8254_timer(char *s)
+{
+	timer_over_8254 = -1;
+	return 1;
+}
+static int __init setup_enable_8254_timer(char *s)
+{
+	timer_over_8254 = 2;
+	return 1;
+}
+
+__setup("disable_8254_timer", setup_disable_8254_timer);
+__setup("enable_8254_timer", setup_enable_8254_timer);
+
 /*
  *	Called after all the initialization is done. If we didnt find any
  *	APIC bugs then we can allow the modify fast path
include/asm-i386/apic.h +2 -0
Index: linux/include/asm-i386/apic.h
===================================================================
@@ -137,6 +137,8 @@ void switch_APIC_timer_to_ipi(void *cpum
 void switch_ipi_to_APIC_timer(void *cpumask);
 #define ARCH_APICTIMER_STOPS_ON_C3	1
 
+extern int timer_over_8254;
+
 #else /* !CONFIG_X86_LOCAL_APIC */
 static inline void lapic_shutdown(void) { }
 
Documentation/i386/boot-options.txt new file +8 -0
Index: linux/Documentation/i386/boot-options.txt
===================================================================
@@ -0,0 +1,8 @@
+
+Some i386 specific boot options
+
+   disable_8254_timer / enable_8254_timer
+                 Enable interrupt 0 timer routing over the 8254 in addition to 
+		 over the IO-APIC. The kernel tries to set a sensible default.
+
+
arch/i386/kernel/acpi/boot.c +0 -3
Index: linux/arch/i386/kernel/acpi/boot.c
===================================================================
@@ -1111,9 +1111,6 @@ int __init acpi_boot_table_init(void)
 		disable_acpi();
 		return error;
 	}
-#ifdef __i386__
-	check_acpi_pci();
-#endif
 
 	acpi_table_parse(ACPI_BOOT, acpi_parse_sbf);
 
arch/i386/kernel/setup.c +4 -0
Index: linux/arch/i386/kernel/setup.c
===================================================================
@@ -1599,6 +1599,10 @@ void __init setup_arch(char **cmdline_p)
 	if (efi_enabled)
 		efi_map_memmap();
 
+#ifdef CONFIG_X86_IO_APIC
+	check_acpi_pci();	/* Checks more than just ACPI actually */	
+#endif
+
 #ifdef CONFIG_ACPI
 	/*
 	 * Parse the ACPI tables for possible boot-time SMP configuration.