On Mon, Dec 01, 2014 at 06:57:05PM +0000, Lina Iyer wrote: > On Thu, Nov 27 2014 at 08:01 -0700, Lorenzo Pieralisi wrote: > >On Thu, Nov 27, 2014 at 05:24:07AM +0000, Lina Iyer wrote: > > > >[...] > > > >> +static int spm_set_low_power_mode(enum pm_sleep_mode mode) > >> +{ > >> + struct spm_driver_data *drv = per_cpu(cpu_spm_drv, > >> + smp_processor_id()); > >> + u32 start_index; > >> + u32 ctl_val; > >> + > >> + if (!drv) > >> + return -ENXIO; > >> + > >> + start_index = drv->reg_data->start_index[mode]; > >> + > >> + ctl_val = spm_register_read(drv, SPM_REG_SPM_CTL); > >> + ctl_val &= ~(SPM_CTL_INDEX << SPM_CTL_INDEX_SHIFT); > >> + ctl_val |= start_index << SPM_CTL_INDEX_SHIFT; > >> + ctl_val |= SPM_CTL_EN; > >> + spm_register_write(drv, SPM_REG_SPM_CTL, ctl_val); > >> + > >> + /* Ensure we have written the start address */ > >> + wmb(); > > > >Can you explain please what this wmb is meant to achieve ? If it is there > >to make sure the write made it to the SPM that's not a proper way > >to achieve it, the barrier ensures ordering, not write completion. > > A write completion would be great, but assuming that the context switch > using SCM or WFI could ensure the flush, then ordering is probably good > enough for this. I assume you want the write to the device to complete (side-effects are observed) before the WFI (which would raise a StandbyWFI signal). The safer way is to read back from the device (or check/poll for the device state change) rather than a barrier. -- Catalin -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html