Message-ID: <3435609.s5m1DCsoNb@wuerfel>
Date:	Thu, 02 Jun 2016 10:43:09 +0200
From:	Arnd Bergmann <arnd@...db.de>
To:	linuxppc-dev@...ts.ozlabs.org
Cc:	Scott Wood <oss@...error.net>, Mark Rutland <mark.rutland@....com>,
	Ulf Hansson <ulf.hansson@...aro.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"linux-i2c@...r.kernel.org" <linux-i2c@...r.kernel.org>,
	"linux-clk@...r.kernel.org" <linux-clk@...r.kernel.org>,
	Qiang Zhao <qiang.zhao@....com>,
	Russell King <linux@....linux.org.uk>,
	Bhupesh Sharma <bhupesh.sharma@...escale.com>,
	Joerg Roedel <joro@...tes.org>,
	Claudiu Manoil <claudiu.manoil@...escale.com>,
	"devicetree@...r.kernel.org" <devicetree@...r.kernel.org>,
	Kumar Gala <galak@...eaurora.org>,
	Rob Herring <robh+dt@...nel.org>,
	Santosh Shilimkar <ssantosh@...nel.org>,
	"linux-arm-kernel@...ts.infradead.org" 
	<linux-arm-kernel@...ts.infradead.org>,
	"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
	"linux-mmc@...r.kernel.org" <linux-mmc@...r.kernel.org>,
	Xiaobo Xie <xiaobo.xie@....com>,
	Yang-Leo Li <leoyang.li@....com>,
	"iommu@...ts.linux-foundation.org" <iommu@...ts.linux-foundation.org>,
	Yangbo Lu <yangbo.lu@....com>
Subject: Re: [PATCH 2/4] soc: fsl: add GUTS driver for QorIQ platforms

On Wednesday, June 1, 2016 8:47:22 PM CEST Scott Wood wrote:
> On Mon, 2016-05-30 at 15:15 +0200, Arnd Bergmann wrote:
> > diff --git a/drivers/soc/fsl/guts.c b/drivers/soc/fsl/guts.c
> > new file mode 100644
> > index 000000000000..2f30698f5bcf
> > --- /dev/null
> > +++ b/drivers/soc/fsl/guts.c
> > @@ -0,0 +1,130 @@
> > +/*
> > + * Freescale QorIQ Platforms GUTS Driver
> > + *
> > + * Copyright (C) 2016 Freescale Semiconductor, Inc.
> > + *
> > + * This program is free software; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License as published by
> > + * the Free Software Foundation; either version 2 of the License, or
> > + * (at your option) any later version.
> > + */
> > +
> > +#include <linux/io.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/module.h>
> > +#include <linux/slab.h>
> > +#include <linux/of_address.h>
> > +#include <linux/of_platform.h>
> > +#include <linux/sys_soc.h>
> > +
> > +#define GUTS_PVR	0x0a0
> > +#define GUTS_SVR	0x0a4
> > +
> > +struct guts {
> > +	void __iomem *regs;
> 
> We already have a struct to define guts.  Why are you not using it?  Why do
> you consider using it to be "abuse"?  What if we want to move more guts
> functionality into this driver?

This structure was in the original patch, I left it in there, only
removed the inclusion of the powerpc header file, which seemed to
be misplaced.

> > +	bool little_endian;
> > +	struct soc_device_attribute soc;
> > +};
> > +
> > +static u32 fsl_guts_get_svr(struct guts *guts)
> > +{
> > +	if (guts->little_endian)
> > +		return ioread32(guts->regs + GUTS_SVR);
> > +	else
> > +		return ioread32be(guts->regs + GUTS_SVR);
> > +}
> > +
> > +static u32 fsl_guts_get_pvr(struct guts *guts)
> > +{
> > +	if (guts->little_endian)
> > +		return ioread32(guts->regs + GUTS_PVR);
> > +	else
> > +		return ioread32be(guts->regs + GUTS_PVR);
> > +}
> 
> You've removed the fallback to mfspr() on PPC, which would be helpful in some
> virtualized environments where we don't have the guts node (but do have other
> directly assigned devices).  Of course, this is a consequence of the
> conversion into a platform device.

Right, it just didn't make any sense after the conversion. I couldn't
figure out what the intention was for the fallback. Virtualized environments
are an interesting case, but I'd also argue that a virtualized system is
not the original SoC anyway, so reporting the physical SoC as the
main system in the guest is a bit strange and we probably want to
come up with a different solution there.

In soc_device, we probably want to just report the type of hypervisor
in this case, which is what most users would expect there. For the
specific case of the mmc driver (are there any real use cases where
you'd pass on the mmc controller to a guest, as opposed to passing either
an emulated block device or the mmc device on an emulated mmc host?),
that means we have to come up with a different solution, but then
we'd be free to work around this by modifying the DT node of the mmc
device.

> > +/*
> > + * Table for matching compatible strings, for device tree
> > + * guts node, for Freescale QorIQ SOCs.
> > + */
> > +static const struct of_device_id fsl_guts_of_match[] = {
> > +	/* For T4 & B4 Series SOCs */
> > +	{ .compatible = "fsl,qoriq-device-config-1.0", .data = "T4/B4
> > series" },
> [snip]
> > +	{ .compatible = "fsl,qoriq-device-config-2.0", .data = "P series"
> 
> As noted in my comment on patch 3/4, these descriptions are reversed.
> 
> They're also incomplete.  t2080 has device config 2.0.  t1040 is described as
> 2.0 though it should probably be 2.1 (or better, drop the generic compatible
> altogether).

Ok. Ideally I think we'd even look up the specific SoC names from the
SVC rather than the compatible string. I just didn't have a good list
for those to put in the driver.

> > +	/*
> > +	 * syscon devices default to little-endian, but on powerpc we have
> > +	 * existing device trees with big-endian maps and an absent
> > endianess
> > +	 * "big-property"
> > +	 */
> > +	if (!IS_ENABLED(CONFIG_POWERPC) &&
> > +	    !of_property_read_bool(dev->of_node, "big-endian"))
> > +		guts->little_endian = true;
> 
> This is not a syscon device (Yangbo's patch to add a guts node on ls2080 is
> the only guts node that says "syscon", and that was a leftover from earlier
> revisions and should probably be removed).  Even if it were, where is it
> documented that syscon defaults to little-endian?

Documentation/devicetree/bindings/regmap/regmap.txt

We had a little screwup here, basically regmap (and by consequence, syscon)
always defaulted to little-endian way before that was documented, so it's
too late to change it, although I agree it would have made sense to document
regmap to default to big-endian on powerpc.

> Documentation/devicetree/bindings/common-properties.txt says that the
> individual binding specifies the default.  The default for this node should be
> big-endian because that's what existed before there was a need to describe the
> endianness.  And we need an update to the guts binding to specify that.

Good point. This proably means that specifying both the "guts" and "syscon"
compatible strings implies having to also specify the endianess explicitly
both ways, because otherwise we break one of the two bindings.

> > +
> > +	guts->regs = devm_ioremap_resource(dev, 0);
> > +	if (!guts->regs) {
> > +		ret = -ENOMEM;
> > +		kfree(guts);
> > +		goto out;
> > +	}
> > +
> > +	fsl_guts_init(dev, guts);
> > +	ret = 0;
> > +out:
> > +	return ret;
> > +}
> > +
> > +static struct platform_driver fsl_soc_guts = {
> > +	.probe = fsl_guts_probe,
> > +	.driver.of_match_table = fsl_guts_of_match,
> > +};
> > +
> > +module_platform_driver(fsl_soc_guts);
> 
> Again, this means that the information is not available during early boot,
> such as in the clock driver.  Thus we would not be able to convert clk-qoriq's
> direct mfspr(SPRN_SVR) into an soc_device_match() (or anything else that makes
> use of this file), nor would we be able to move its access of the guts RCW
> registers into this driver.

Correct. Do we have a reason to convert the mfspr() though? I don't really
see an improvement over the current state if we do that, and for new devices
that might need the erratum workaround, we could add a DT property that would
be preferred to both.

	Arnd

Powered by blists - more mailing lists