x86, hpet: fix for LS21 + HPET = boot hang
john stultz [Fri, 13 Feb 2009 02:48:53 +0000 (18:48 -0800)]
Between 2.6.23 and 2.6.24-rc1 a change was made that broke IBM LS21
systems that had the HPET enabled in the BIOS, resulting in boot hangs
for x86_64.

Specifically commit b8ce33590687888ebb900d09557b8807c4539022, which
merges the i386 and x86_64 HPET code.

Prior to this commit, when we setup the HPET timers in x86_64, we did
the following:

hpet_writel(HPET_TN_ENABLE | HPET_TN_PERIODIC | HPET_TN_SETVAL |
                    HPET_TN_32BIT, HPET_T0_CFG);

However after the i386/x86_64 HPET merge, we do the following:

cfg = hpet_readl(HPET_Tn_CFG(timer));
cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
HPET_TN_SETVAL | HPET_TN_32BIT;
hpet_writel(cfg, HPET_Tn_CFG(timer));

However on LS21s with HPET enabled in the BIOS, the HPET_T0_CFG register
boots with Level triggered interrupts (HPET_TN_LEVEL) enabled. This
causes the periodic interrupt to be not so periodic, and that results in
the boot time hang I reported earlier in the delay calibration.

My fix: Always disable HPET_TN_LEVEL when setting up periodic mode.

Signed-off-by: John Stultz <johnstul@us.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

arch/x86/kernel/hpet.c

index 64d5ad0..5c8da2c 100644 (file)
@@ -269,6 +269,8 @@ static void hpet_set_mode(enum clock_event_mode mode,
                now = hpet_readl(HPET_COUNTER);
                cmp = now + (unsigned long) delta;
                cfg = hpet_readl(HPET_Tn_CFG(timer));
+               /* Make sure we use edge triggered interrupts */
+               cfg &= ~HPET_TN_LEVEL;
                cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
                       HPET_TN_SETVAL | HPET_TN_32BIT;
                hpet_writel(cfg, HPET_Tn_CFG(timer));