Hi Rafael, On Mon, Jan 15, 2018 at 5:17 PM, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote: > On Mon, Jan 15, 2018 at 3:26 PM, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote: >> On 15 January 2018 at 14:22, Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote: > > [cut] > >>> >>> I did miss a small difference in topology: in pm/linux-next, H3 has DMA >>> enabled for SCIF2, while M3 hasn't (yet). >>> With DMA enabled on M3, it fails in the same way. >>> >>> As genpd_resume_noirq() no longer calls pm_runtime_force_resume(), >>> rcar_dmac_runtime_resume() is no longer called, and the DMAC's registers >>> are no longer reinitialized after system resume, breaking the serial port. >> >> In drivers/dma/sh/rcar-dmac.c, I would try to replace the below line: >> SET_SYSTEM_SLEEP_PM_OPS(rcar_dmac_sleep_suspend, rcar_dmac_sleep_resume) >> >> with: >> SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend, pm_runtime_force_resume) > > Yes, that probably is the least intrusive thing that can be done to > address the issue. > >> in case that may be too early to suspend the dma device (which is >> rather common for dma devices) then try; >> >> SET_LATE_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend, pm_runtime_force_resume) > > Good suggestion, and I would go straight for it anyway. > > Geert, can you try if this works, please? Works. Both using SET_SYSTEM_SLEEP_PM_OPS() and SET_LATE_SYSTEM_SLEEP_PM_OPS(). But given this is a DMA engine driver, I'd settle for the latter. And I did verify doing so doesn't break the system without the patch in $subject. Thanks! Will send a patch... Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds