* Daniel Lezcano <daniel.lezcano@xxxxxxxxxx> [210322 18:24]: > On 22/03/2021 17:33, Tony Lindgren wrote: > > Hi, > > > > * Daniel Lezcano <daniel.lezcano@xxxxxxxxxx> [210322 15:56]: > >> On 04/03/2021 08:37, Tony Lindgren wrote: > >>> There is a timer wrap issue on dra7 for the ARM architected timer. > >>> In a typical clock configuration the timer fails to wrap after 388 days. > >>> > >>> To work around the issue, we need to use timer-ti-dm timers instead. > >>> > >>> Let's prepare for adding support for percpu timers by adding a common > >>> dmtimer_clkevt_init_common() and call it from dmtimer_clockevent_init(). > >>> This patch makes no intentional functional changes. > >>> > >>> Signed-off-by: Tony Lindgren <tony@xxxxxxxxxxx> > >>> --- > >> > >> [ ... ] > >> > >>> @@ -575,33 +574,60 @@ static int __init dmtimer_clockevent_init(struct device_node *np) > >>> */ > >>> writel_relaxed(OMAP_TIMER_CTRL_POSTED, t->base + t->ifctrl); > >>> > >>> + if (dev->cpumask == cpu_possible_mask) > >>> + irqflags = IRQF_TIMER; > >>> + else > >>> + irqflags = IRQF_TIMER | IRQF_NOBALANCING; > >> > >> Can you explain the reasoning behind the test above ? > > > > In the per cpu case we assign one dmtimer per cpu, and we want the > > interrupt handling on the assigned CPU. In the per cpu case we have > > the cpu specified with dev->cpumask unlike for the normal clockevent > > case. > > > > In the per cpu dmtimer case the interrupt line is not wired per cpu > > though, so I don't think we want to add IRQF_PERCPU here. > > If it is per cpu, then the parameter will be cpumask_of(cpu). If there > is one cpu, no balancing can happen and then the IRQF_NOBALANCING is not > needed, neither this test and the irqflags, right? Oh yeah you're right, none of that is needed. For the percpu case we already have irq_force_affinity() in omap_dmtimer_starting_cpu(). I'll update and send out v2 of these two patches. Thanks, Tony
