Hi Dirk, On Thursday 09 Jun 2016 10:58:30 Dirk Behme wrote: > On 09.06.2016 10:22, Geert Uytterhoeven wrote: > > On Thu, Jun 9, 2016 at 8:56 AM, Dirk Behme wrote: > >> with the R-Car3, the pinmux/drvctrl is done at 3 places: > >> > >> 1. ARM trusted firmware / IPL: > >> > >> https://github.com/renesas-rcar/meta-renesas/blob/jethro/meta-rcar-gen3/r > >> ecipes-bsp/arm-trusted-firmware/arm-trusted-firmware/0001-add-bl2-bl31-R- > >> Car-support.patch#L8428 > >> > >> > >> 2. U-Boot: > >> > >> https://github.com/renesas-rcar/u-boot/blob/v2015.04/rcar-3.2.x/board/ren > >> esas/salvator-x/salvator-x.c#L116 > >> > >> > >> 3. Linux kernel > >> > >> https://git.kernel.org/cgit/linux/kernel/git/geert/renesas-drivers.git/tr > >> ee/drivers/pinctrl/sh-pfc/pfc-r8a7795.c?h=topic/gen3-latest > >> > >> > >> Now, the issue we see is that (of course?) this is not necessarily > >> consistent. > >> > >> Therefore: What's supposed to be the correct place to do this (once)? > > > > IMHO the kernel should take care of making sure pinctrl is configured > > correctly, independent of earlier stages in the boot process. > > From previous discussions with kernel people I expected this answer ;) > > In my opinion, to be able to keep the pinctrl consistent, you have to do > it in one place. To avoid issues like discussed below. And to be able to > maintain it. The consequence of this "one place" would be that the main > place is the boot loader, then. In case of the Salvator-X this would be > the ARM trusted firmware (IPL), then.
Anything with "trusted" in its name can't be trusted. > Yes, I know, kernel people see this differently, as the kernel is the > only place they have under control. From that point of view only the > kernel can "fix" broken or not updated boot loaders. > > So it depends whether you have only the kernel or everything under your > control, including the boot loader. It's a nice dream, but in practice I'm afraid boot loaders get it wrong, and the kernel has to handle the mess. > > Note that if the kernel takes care of it, doing it once is not an option, > > as earlier stages may need pinctrl configuration depending on boot method > > (e.g. TFTP over Ethernet needs RAVB setup). > > How will the kernel handle different boards needing different pinctrl > configuration? > > I.e. how is pfc-r8a7795 supposed to cover different boards? pfc-r8a7795 isn't. That driver needs to implement all needed configuration options for the R8A7795 pins. Pinmux/pinctrl configuration is then selected by device drivers (the ethernet driver in this case) based on the information supplied in the board's device tree. > >> The example is the RAVB ethernet: We found that on the Salvator-X the > >> kernel's ethernet doesn't work without being initialized by U-Boot. I.e. > >> without running U-Boot doing the drvctrl [1] ethernet fails in the Linux > >> kernel. > > > > This is a missing/incorrect drive strength configuration? > > Yes, I think so. > > > The pfc-r8a7795 does have support for drive strength, but only for pins > > that also have GPIO capabilities. (At least some of) the RAVB pins don't > > seem to be handled (e.g. DRVCTRL2). > > So your proposal would be to fix/correct/extend Linux kernel's > pfc-r8a7795 to correctly support the drive strength needed for the RAVB > on the Salvator-X? If the R8A7795 supports configuring the drive strength of RAVB pins, and that feature is needed by systems using that SoC, then the pfc-r8a7795 driver should implement it, yes. That's no different that any other hardware feature. -- Regards, Laurent Pinchart
