Messages in this thread | | | Date | Mon, 9 May 2011 21:50:15 +0100 | From | Jamie Iles <> | Subject | Re: [PATCHv3] x86, mrst: share APB timer code with other platforms |
| |
On Mon, May 09, 2011 at 01:22:22PM -0700, john stultz wrote: > On Tue, 2011-04-12 at 12:10 +0100, Jamie Iles wrote: > > The APB timers are an IP block from Synopsys (DesignWare APB timers) > > and are also found in other systems including ARM SoC's. This patch > > adds functions for creating clock_event_devices and clocksources from > > APB timers but does not do the resource allocation. This is handled > > in a higher layer to allow the timers to be created from multiple > > methods such as platform_devices. > > > > Changes since v2: > > - Make eoi() for clock_event_device optional for x86 where eoi > > is handled in firmware. > > Changes since v1: > > - Use the correct timer for clocksource on x86 > > - Select the correct timer rating for x86 > > - Restore freerunning timer behaviour for oneshot event devices > > - Reenable event irq correctly for hotplug > > Sorry for being slow to review this.
No worries, appreciate the review.
> Just a minor nit below. > > > +struct dw_apb_clocksource * > > +dw_apb_clocksource_init(unsigned rating, char *name, void __iomem *base, > > + unsigned long freq) > > +{ > > + struct dw_apb_clocksource *dw_cs = kzalloc(sizeof(*dw_cs), GFP_KERNEL); > > + > > + if (!dw_cs) > > + return NULL; > > + > > + dw_cs->timer.base = base; > > + dw_cs->timer.freq = freq; > > + dw_cs->cs.name = name; > > + dw_cs->cs.rating = rating; > > + dw_cs->cs.read = __apbt_read_clocksource; > > + dw_cs->cs.mask = CLOCKSOURCE_MASK(32); > > + dw_cs->cs.flags = CLOCK_SOURCE_IS_CONTINUOUS; > > + dw_cs->cs.resume = apbt_restart_clocksource; > > + > > + return dw_cs; > > +} > > So I don't see anything that frees the dw_apb_clocksource structure > after it has been allocated. Should this instead be allocated by the > caller and passed in to be initialized here?
Yeah, either that or I could add a dw_apb_clocksource_remove() that does the clocksource_unregister() and kfree(). That's probably a little more consistent with the rest of the driver.
Jamie
| |