svn commit: r353493 - head/sys/dev/mmc/host

[Emmanuel Vadot]

On Mon, 14 Oct 2019 17:27:51 +0100
Ruslan Bukin <br at freebsd.org> wrote:

> On Mon, Oct 14, 2019 at 06:10:51PM +0200, Emmanuel Vadot wrote:
> > 
> > On Mon, 14 Oct 2019 15:53:00 +0000 (UTC)
> > Ruslan Bukin <br at FreeBSD.org> wrote:
> > 
> > > Author: br
> > > Date: Mon Oct 14 15:52:59 2019
> > > New Revision: 353493
> > > URL: https://svnweb.freebsd.org/changeset/base/353493
> > > 
> > > Log:
> > >   Fix the driver attachment in cases when the external resource devices
> > >   (resets, regulators, clocks) are not available.
> > >   
> > >   Rely on a system initialization done by a bootloader in that cases.
> > >   
> > >   This fixes operation on Terasic DE10-Pro (an Intel Stratix 10
> > >   development kit).
> > >   
> > >   Sponsored by:	DARPA, AFRL
> > > 
> > > Modified:
> > >   head/sys/dev/mmc/host/dwmmc.c
> > > 
> > > Modified: head/sys/dev/mmc/host/dwmmc.c
> > > ==============================================================================
> > > --- head/sys/dev/mmc/host/dwmmc.c	Mon Oct 14 15:33:53 2019	(r353492)
> > > +++ head/sys/dev/mmc/host/dwmmc.c	Mon Oct 14 15:52:59 2019	(r353493)
> > > @@ -1,5 +1,5 @@
> > >  /*-
> > > - * Copyright (c) 2014 Ruslan Bukin <br at bsdpad.com>
> > > + * Copyright (c) 2014-2019 Ruslan Bukin <br at bsdpad.com>
> > >   * All rights reserved.
> > >   *
> > >   * This software was developed by SRI International and the University of
> > > @@ -457,26 +457,20 @@ parse_fdt(struct dwmmc_softc *sc)
> > >  
> > >  	/* IP block reset is optional */
> > >  	error = hwreset_get_by_ofw_name(sc->dev, 0, "reset", &sc->hwreset);
> > > -	if (error != 0 && error != ENOENT) {
> > > +	if (error != 0 && error != ENOENT)
> > >  		device_printf(sc->dev, "Cannot get reset\n");
> > > -		goto fail;
> > > -	}
> > 
> >  This is not correct, on a system without reset/clock/regulator support
> > you will get ENODEV as the phandle is present but no device is
> > associated with it. This is the case that you want to test. Currently
> > this hide all errors. 
> 
> The change means that the driver will be attached regardless of the return value from ext resources.

 Yes and this is a problem.

> Why it is not correct?

 Because if a reset/clock/regulator is present but we failed to parse
the node (bad #clock-cell for example) this just assume that the
resource isn't present which is not correct. Or worse the underlying
gpio cannot be mapped, you now have a non functional driver because you
cannot switch the regulator.

> It does not hide all errors, the printf will be called and a user will be warned.

 If the ressources are present in the DTB and the system support them
but they cannot be used the driver should fail. Which is why I suggest
you to make an extra case on ENODEV even if I don't like that because
it's hacky even for a platform that is, to my knowledge, not
really supported (no u-boot port, dtb aren't build, not part of
GENERIC).

> Ruslan


-- 
Emmanuel Vadot <manu at bidouilliste.com> <manu at freebsd.org>