x86: apic LVTT - use APIC_DIVISOR on 64bit mode

We use APIC_TDR_DIV_16 while setting APIC_TDCR (divisor) it happened this moment is hidden from __setup_APIC_LVTT caller. So we better increment caller 'clocks' value to not change current behaviour and use APIC_DIVISOR (as already done in 32bit version).

The main benefit - unified procedure code for 32/64bit modes.

---

Please review carefully. Any comments are welcome. I'm _not_ 100% sure if this patch is safe.

Index: linux-2.6.git/arch/x86/kernel/apic_64.c
===================================================================
@@ -165,6 +165,9 @@ int lapic_get_maxlvt(void)
 	return maxlvt;
 }
 
+/* Clock divisor is set to 16 */
+#define APIC_DIVISOR 16
+
 /*
  * This function sets up the local APIC timer, with a timeout of
  * 'clocks' APIC bus clock. During calibration we actually call

@@ -197,7 +200,7 @@ static void __setup_APIC_LVTT(unsigned i
 				| APIC_TDR_DIV_16);
 
 	if (!oneshot)
-		apic_write(APIC_TMICT, clocks);
+		apic_write(APIC_TMICT, clocks/APIC_DIVISOR);
 }
 
 /*

@@ -329,7 +332,7 @@ static void __init calibrate_APIC_clock(
 	 *
 	 * No interrupt enable !
 	 */
-	__setup_APIC_LVTT(250000000, 0, 0);
+	__setup_APIC_LVTT(4000000000, 0, 0);
 
 	apic_start = apic_read(APIC_TMCCT);
 #ifdef CONFIG_X86_PM_TIMER

hm, how about the other caller to __setup_APIC_LVTT(), lapic_timer_setup():

        case CLOCK_EVT_MODE_ONESHOT:
                __setup_APIC_LVTT(calibration_result,
                                  mode != CLOCK_EVT_MODE_PERIODIC, 1);
                break;

that will affect high-res timers. Wont calibration_result have the same value after your patch as well, but only a 16th of it will be used in __setup_APIC_LVTT(), causing 16 times shorter timeouts than intended?

I.e. are we fixing a bug, are we changing nothing, or are we introducing a bug? :-)

Ingo

[Ingo Molnar - Sat, Jul 12, 2008 at 08:05:04AM +0200]

| 
| * Cyrill Gorcunov <[email protected]> wrote:
| 
| > We use APIC_TDR_DIV_16 while setting APIC_TDCR (divisor) it happened 
| > this moment is hidden from __setup_APIC_LVTT caller. So we better 
| > increment caller 'clocks' value to not change current behaviour and 
| > use APIC_DIVISOR (as already done in 32bit version).
| > 
| > The main benefit - unified procedure code for 32/64bit modes.
| 
| > Please review carefully. Any comments are welcome.
| > I'm _not_ 100% sure if this patch is safe.
| 
| hm, how about the other caller to __setup_APIC_LVTT(), 
| lapic_timer_setup():
| 
|         case CLOCK_EVT_MODE_ONESHOT:
|                 __setup_APIC_LVTT(calibration_result,
|                                   mode != CLOCK_EVT_MODE_PERIODIC, 1);
|                 break;
| 
| that will affect high-res timers. Wont calibration_result have the same 
| value after your patch as well, but only a 16th of it will be used in 
| __setup_APIC_LVTT(), causing 16 times shorter timeouts than intended?
| 
| I.e. are we fixing a bug, are we changing nothing, or are we introducing 
| a bug? :-)
| 
| 	Ingo
| 

well, the last I would say :). That happened I sent first version of patch instead of modified second version. Thanks a lot Ingo for catching it! Oh my, sorry... Here is an updated version.
---

We use APIC_TDR_DIV_16 while setting APIC_TDCR (divisor). It happened this moment is hidden from __setup_APIC_LVTT caller. So we better increment caller 'clocks' value to not change current behaviour and use APIC_DIVISOR (as already done in 32bit version).

The main benefit - unified procedure code for 32/64bit modes.

---

Index: linux-2.6.git/arch/x86/kernel/apic_64.c
===================================================================
@@ -165,6 +165,9 @@ int lapic_get_maxlvt(void)
 	return maxlvt;
 }
 
+/* Clock divisor is set to 16 */
+#define APIC_DIVISOR 16
+
 /*
  * This function sets up the local APIC timer, with a timeout of
  * 'clocks' APIC bus clock. During calibration we actually call

@@ -197,7 +200,7 @@ static void __setup_APIC_LVTT(unsigned i
 				| APIC_TDR_DIV_16);
 
 	if (!oneshot)
-		apic_write(APIC_TMICT, clocks);
+		apic_write(APIC_TMICT, clocks/APIC_DIVISOR);
 }
 
 /*

@@ -329,7 +332,7 @@ static void __init calibrate_APIC_clock(
 	 *
 	 * No interrupt enable !
 	 */
-	__setup_APIC_LVTT(250000000, 0, 0);
+	__setup_APIC_LVTT(4000000000, 0, 0);
 
 	apic_start = apic_read(APIC_TMCCT);
 #ifdef CONFIG_X86_PM_TIMER

@@ -367,7 +370,7 @@ static void __init calibrate_APIC_clock(
 	lapic_clockevent.min_delta_ns =
 		clockevent_delta2ns(0xF, &lapic_clockevent);
 
-	calibration_result = result / HZ;
+	calibration_result = result * APIC_DIVISOR / HZ;
 }
 
 /*

[Ingo Molnar - Sat, Jul 12, 2008 at 09:39:10AM +0200]

| 
| * Cyrill Gorcunov <[email protected]> wrote:
| 
| > @@ -329,7 +332,7 @@ static void __init calibrate_APIC_clock(
| >  	 *
| >  	 * No interrupt enable !
| >  	 */
| > -	__setup_APIC_LVTT(250000000, 0, 0);
| > +	__setup_APIC_LVTT(4000000000, 0, 0);
| 
| note how close it is to 2^32. For this to be unifiable later on this 
| needs to be UL i guess, and this:
| 
| > -	calibration_result = result / HZ;
| > +	calibration_result = result * APIC_DIVISOR / HZ;
| 
| might overflow 32 bits.
| 
| 	Ingo
|

hmm... I need more time for analisys, thanks. Drop this patch please

- Cyrill -

i think we are introducing a bug :)

Ingo