| Message ID | 4F394D3B.3080909@flytronic.pl |
|---|---|
| State | New |
| Headers | show |
diff --git a/arch/arm/mach-ixp4xx/common.c b/arch/arm/mach-ixp4xx/common.c index 3841ab4..fd37c83 100644 --- a/arch/arm/mach-ixp4xx/common.c +++ b/arch/arm/mach-ixp4xx/common.c @@ -434,6 +434,7 @@ static int ixp4xx_set_next_event(unsigned long evt, { unsigned long opts = *IXP4XX_OSRT1 & IXP4XX_OST_RELOAD_MASK; + *IXP4XX_OSRT1 = 0; *IXP4XX_OSRT1 = (evt & ~IXP4XX_OST_RELOAD_MASK) | opts; return 0;
IXP43x Developer's Manual [17.4.3] and IXP4[56]x Developer's Manual [18.4.3] say that for predictable operation the timer needs to be stopped before writing a new value into the reload register. Indeed, tests on IXP435 show that writing a new value into the reload register without stopping the timer first has no immediate effect on the timer. This makes hrtimers started through hrtimer_start() to be delayed until the currently earliest hrtimer expires. IXP42x Developer's Manual [17.4.3] says that the timer will be reloaded immediately on setting the timer reload register, so the bug probably doesn't occur on those CPUs. However, stopping the timer shouldn't have any negative side effects, so it should be safe to apply it machine-wide. Signed-off-by: Michał Wróbel <michal.wrobel@flytronic.pl> --- arch/arm/mach-ixp4xx/common.c | 1 + 1 files changed, 1 insertions(+), 0 deletions(-)