[RFC] lapic calibration results fix

Previous thread: [PATCH] PCI:pciehp:remove unnecessary volatile in struct controller by tom.leiming on Sunday, July 13, 2008 - 6:46 am. (1 message)

Next thread: Re: Toshiba - Documentation for Linux Developers by Renato S. Yamane on Sunday, July 13, 2008 - 10:46 am. (6 messages)
To: Ingo Molnar <mingo@...>, Maciej W. Rozycki <macro@...>, Yinghai Lu <yhlu.kernel@...>
Cc: LKML <linux-kernel@...>
Date: Sunday, July 13, 2008 - 7:37 am

Ingo, Maciej,

it seems I found a bit strange code snippet in apic_32.c:setup_boot_APIC_clock.
We set local_apic_timer_verify_ok = 1 before checking the results. I think
the following patch make sense. Please take a look on.

---
We set lapic flag that clocksource calibration is
fine too early. Fix it.

Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
---

Index: linux-2.6.git/arch/x86/kernel/apic_32.c
===================================================================
--- linux-2.6.git.orig/arch/x86/kernel/apic_32.c 2008-07-13 15:25:20.000000000 +0400
+++ linux-2.6.git/arch/x86/kernel/apic_32.c 2008-07-13 15:31:03.000000000 +0400
@@ -489,8 +489,6 @@ void __init setup_boot_APIC_clock(void)
calibration_result / (1000000 / HZ),
calibration_result % (1000000 / HZ));

- local_apic_timer_verify_ok = 1;
-
/*
* Do a sanity check on the APIC calibration result
*/
@@ -504,6 +502,8 @@ void __init setup_boot_APIC_clock(void)
return;
}

+ local_apic_timer_verify_ok = 1;
+
/* We trust the pm timer based calibration */
if (!pm_referenced) {
apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
--

To: Ingo Molnar <mingo@...>, Maciej W. Rozycki <macro@...>, Yinghai Lu <yhlu.kernel@...>, LKML <linux-kernel@...>
Date: Sunday, July 13, 2008 - 12:23 pm

[Cyrill Gorcunov - Sun, Jul 13, 2008 at 03:37:34PM +0400]
| Ingo, Maciej,
|
| it seems I found a bit strange code snippet in apic_32.c:setup_boot_APIC_clock.
| We set local_apic_timer_verify_ok = 1 before checking the results. I think
| the following patch make sense. Please take a look on.
|
| ---
| We set lapic flag that clocksource calibration is
| fine too early. Fix it.
|
| Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
| ---
|
| Index: linux-2.6.git/arch/x86/kernel/apic_32.c
| ===================================================================
| --- linux-2.6.git.orig/arch/x86/kernel/apic_32.c 2008-07-13 15:25:20.000000000 +0400
| +++ linux-2.6.git/arch/x86/kernel/apic_32.c 2008-07-13 15:31:03.000000000 +0400
| @@ -489,8 +489,6 @@ void __init setup_boot_APIC_clock(void)
| calibration_result / (1000000 / HZ),
| calibration_result % (1000000 / HZ));
|
| - local_apic_timer_verify_ok = 1;
| -
| /*
| * Do a sanity check on the APIC calibration result
| */
| @@ -504,6 +502,8 @@ void __init setup_boot_APIC_clock(void)
| return;
| }
|
| + local_apic_timer_verify_ok = 1;
| +
| /* We trust the pm timer based calibration */
| if (!pm_referenced) {
| apic_printk(APIC_VERBOSE, "... verify APIC timer\n");

Letme clarify a bit situation why I propose this path (it's not that
clear from its description).

While calibrating apic timer we use different 'flags' signaling if it
was success/fail. On 64bit platform lapic_timer_setup() does check
for CLOCK_EVT_FEAT_DUMMY bit which is set by default and cleared on
successfully calibrated timer. So 64bit is fine now. On 32bit we
use a different approach - local_apic_timer_verify_ok flag _BUT_
as I see it being set too early _before_ calibration result is checked
so we could have the following - kernel reports user that APIC timer
is too slow and disabled but local_apic_timer_verify_ok remains set
to 1 instead of 0.

- Cyrill -
--

Previous thread: [PATCH] PCI:pciehp:remove unnecessary volatile in struct controller by tom.leiming on Sunday, July 13, 2008 - 6:46 am. (1 message)

Next thread: Re: Toshiba - Documentation for Linux Developers by Renato S. Yamane on Sunday, July 13, 2008 - 10:46 am. (6 messages)