Re: PCI: mvebu: return NULL instead of ERR_PTR(ret)

Re: PCI: mvebu: return NULL instead of ERR_PTR(ret)

[Jingoo Han]

On Wednesday, November 27, 2013 3:10 AM, Jason Gunthorpe wrote:
> On Tue, Nov 26, 2013 at 02:31:44PM +0900, Jingoo Han wrote:
> > Previously, I sent the patch in order to fix sparse warning as below:
> > How about this?
> >
> > static void __iomem *mvebu_pcie_map_registers(struct platform_device *pdev,
> > 			struct device_node *np, struct mvebu_pcie_port *port)
> > {
> > 	struct resource regs;
> > 	int ret = 0;
> >
> > 	ret = of_address_to_resource(np, 0, &regs);
> > 	if (ret)
> > -		return ERR_PTR(ret);
> > +		return (void __iomem *)ERR_PTR(ret);
> 
> You should probably ask the sparse folks for guidance 'git grep
> iomem.*ERR_PTR' returns nothing, so this isn't an established pattern.
> 
> It seems like sparse should know that ERR_PTR functions can work with
> any pointer no matter the type? IS_ERR_PTR will have the same problem
> with implicitly dropping the iomem tag.

+cc Christopher Li, sparse mailing-list, Joe Perches, Dan Carpenter,
     Axel Lin, Julia Lawall,

Hi All,

I have some questions about handling sparse warning.

Currently, the following sparse warning happens
at Marvell Armada PCIe driver.

drivers/pci/host/pci-mvebu.c:743:31: warning: incorrect type in return expression (different address spaces)
drivers/pci/host/pci-mvebu.c:743:31:    expected void [noderef] <asn:2>*
drivers/pci/host/pci-mvebu.c:743:31:    got void *

mvebu_pcie_map_registers() returns ERR_PTR(ret),
however ERR_PTR() returns (void *), not (void __iomem *).

./drivers/pci/host/pci-mvebu.c
static void __iomem *mvebu_pcie_map_registers(struct platform_device *pdev,
		struct device_node *np, struct mvebu_pcie_port *port)
{
	struct resource regs;
	int ret = 0;

	ret = of_address_to_resource(np, 0, &regs);
	if (ret)
		return ERR_PTR(ret);

	return devm_ioremap_resource(&pdev->dev, &regs);
}
./include/linux/err.h
static inline void * __must_check ERR_PTR(long error)
{
	return (void *) error;
}


Previously, I submitted the following patch, that adds
(void __iomem *) cast.

./drivers/pci/host/pci-mvebu.c
static void __iomem *mvebu_pcie_map_registers(struct platform_device *pdev,
		struct device_node *np, struct mvebu_pcie_port *port)
{
	.....

	ret = of_address_to_resource(np, 0, &regs);
	if (ret)
-		return ERR_PTR(ret);
+		return (void __iomem *)ERR_PTR(ret);

However, other engineers said that "(void __iomem *)ERR_PTR(ret)"
is not a general pattern. I cannot find the proper method to resolve
this sparse warning.

In this case, how can I resolve this sparse warning?

Then, how about the following?

--- a/drivers/pci/host/pci-mvebu.c
+++ b/drivers/pci/host/pci-mvebu.c
@@ -740,7 +740,7 @@ static void __iomem *mvebu_pcie_map_registers(struct platform_device *pdev,

	ret = of_address_to_resource(np, 0, &regs);
	if (ret)
-		return ERR_PTR(ret);
+		return NULL;

	return devm_ioremap_resource(&pdev->dev, &regs);
 }
@@ -939,7 +939,7 @@ static int mvebu_pcie_probe(struct platform_device *pdev)
		continue;

		port->base = mvebu_pcie_map_registers(pdev, child, port);
-		if (IS_ERR(port->base)) {
+		if (IS_ERR_OR_NULL(port->base)) {
			dev_err(&pdev->dev, "PCIe%d.%d: cannot map registers\n",
				port->port, port->lane);
			port->base = NULL;

Thank you for reading this. :-)

Best regards,
Jingoo Han

[PATCH] linux/err.h: Provide an ERR_PTR_IO that returns an __iomem pointer

[Josh Triplett]

On Wed, Nov 27, 2013 at 10:59:18AM +0900, Jingoo Han wrote:
> On Wednesday, November 27, 2013 3:10 AM, Jason Gunthorpe wrote:
> > On Tue, Nov 26, 2013 at 02:31:44PM +0900, Jingoo Han wrote:
> > > Previously, I sent the patch in order to fix sparse warning as below:
> > > How about this?
> > >
> > > static void __iomem *mvebu_pcie_map_registers(struct platform_device *pdev,
> > > 			struct device_node *np, struct mvebu_pcie_port *port)
> > > {
> > > 	struct resource regs;
> > > 	int ret = 0;
> > >
> > > 	ret = of_address_to_resource(np, 0, &regs);
> > > 	if (ret)
> > > -		return ERR_PTR(ret);
> > > +		return (void __iomem *)ERR_PTR(ret);
> > 
> > You should probably ask the sparse folks for guidance 'git grep
> > iomem.*ERR_PTR' returns nothing, so this isn't an established pattern.
> > 
> > It seems like sparse should know that ERR_PTR functions can work with
> > any pointer no matter the type? IS_ERR_PTR will have the same problem
> > with implicitly dropping the iomem tag.
> 
> +cc Christopher Li, sparse mailing-list, Joe Perches, Dan Carpenter,
>      Axel Lin, Julia Lawall,
> 
> Hi All,
> 
> I have some questions about handling sparse warning.
> 
> Currently, the following sparse warning happens
> at Marvell Armada PCIe driver.
> 
> drivers/pci/host/pci-mvebu.c:743:31: warning: incorrect type in return expression (different address spaces)
> drivers/pci/host/pci-mvebu.c:743:31:    expected void [noderef] <asn:2>*
> drivers/pci/host/pci-mvebu.c:743:31:    got void *
> 
> mvebu_pcie_map_registers() returns ERR_PTR(ret),
> however ERR_PTR() returns (void *), not (void __iomem *).

Given that PTR_ERR, IS_ERR, and IS_ERR_OR_NULL currently accept any kind
of pointer (by using __force), it does make sense for ERR_PTR to return
any kind of pointer.  There's no built-in way to do this, but we could
add an ERR_PTR_iomem that has the appropriate return type.  Here's a
possible patch; please test in your particular use case and see if it
works for you.  (You can apply this patch from this mail via git am
--scissors.)

---8<---
From: Josh Triplett <josh@joshtriplett.org>
Subject: [PATCH] linux/err.h: Provide an ERR_PTR_IO that returns an __iomem pointer

PTR_ERR, IS_ERR, and IS_ERR_OR_NULL already accept any type of pointer
by using __force; ERR_PTR should similarly be able to return any type of
pointer.  Provide an ERR_PTR_IO to return an error pointer in __iomem.

Signed-off-by: Josh Triplett <josh@joshtriplett.org>
---
 include/linux/err.h | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/include/linux/err.h b/include/linux/err.h
index 15f92e0..d64215b 100644
--- a/include/linux/err.h
+++ b/include/linux/err.h
@@ -24,6 +24,11 @@ static inline void * __must_check ERR_PTR(long error)
 	return (void *) error;
 }
 
+static inline __iomem void * __must_check ERR_PTR_IO(long error)
+{
+	return (__force __iomem void *) error;
+}
+
 static inline long __must_check PTR_ERR(__force const void *ptr)
 {
 	return (long) ptr;
-- 
1.8.4.4

Re: PCI: mvebu: return NULL instead of ERR_PTR(ret)

[Joe Perches]

On Wed, 2013-11-27 at 10:59 +0900, Jingoo Han wrote:
[]
> ./drivers/pci/host/pci-mvebu.c
> static void __iomem *mvebu_pcie_map_registers(struct platform_device *pdev,
> 		struct device_node *np, struct mvebu_pcie_port *port)
> {
> 	.....
> 
> 	ret = of_address_to_resource(np, 0, &regs);
> 	if (ret)
> -		return ERR_PTR(ret);
> +		return (void __iomem *)ERR_PTR(ret);
> 
> However, other engineers said that "(void __iomem *)ERR_PTR(ret)"
> is not a general pattern. I cannot find the proper method to resolve
> this sparse warning.
> 
> In this case, how can I resolve this sparse warning?

I think there's no problem using the cast.  It's not a
pattern because it's not been required before as function
returns have not previously been declared __iomem.

Or, perhaps the arm|hexagon specific #define below could
be made generic so it could be used.

arch/arm/include/asm/io.h:#define IOMEM(x)	((void __force __iomem *)(x))

	return IOMEM(ERR_PTR(ret));

There aren't any current uses of return IOMEM(foo) either
though so the direct cast is probably more appropriate.

Re: PCI: mvebu: return NULL instead of ERR_PTR(ret)

[Josh Triplett]

On Tue, Nov 26, 2013 at 06:26:36PM -0800, Joe Perches wrote:
> On Wed, 2013-11-27 at 10:59 +0900, Jingoo Han wrote:
> []
> > ./drivers/pci/host/pci-mvebu.c
> > static void __iomem *mvebu_pcie_map_registers(struct platform_device *pdev,
> > 		struct device_node *np, struct mvebu_pcie_port *port)
> > {
> > 	.....
> > 
> > 	ret = of_address_to_resource(np, 0, &regs);
> > 	if (ret)
> > -		return ERR_PTR(ret);
> > +		return (void __iomem *)ERR_PTR(ret);
> > 
> > However, other engineers said that "(void __iomem *)ERR_PTR(ret)"
> > is not a general pattern. I cannot find the proper method to resolve
> > this sparse warning.
> > 
> > In this case, how can I resolve this sparse warning?
> 
> I think there's no problem using the cast.  It's not a
> pattern because it's not been required before as function
> returns have not previously been declared __iomem.
> 
> Or, perhaps the arm|hexagon specific #define below could
> be made generic so it could be used.
> 
> arch/arm/include/asm/io.h:#define IOMEM(x)	((void __force __iomem *)(x))
> 
> 	return IOMEM(ERR_PTR(ret));
> 
> There aren't any current uses of return IOMEM(foo) either
> though so the direct cast is probably more appropriate.

I don't think that #define is appropriate for any non-constant value;
almost any instance of that __force cast applied to a variable would be
better written by fixing types more appropriately.  I'd suggest
updating that #define for both architectures to use BUILD_BUG_ON to
assert __builtin_constant_p on its argument.

- Josh Triplett

Re: PCI: mvebu: return NULL instead of ERR_PTR(ret)

[Joe Perches]

On Tue, 2013-11-26 at 18:35 -0800, Josh Triplett wrote:
> On Tue, Nov 26, 2013 at 06:26:36PM -0800, Joe Perches wrote:
> > On Wed, 2013-11-27 at 10:59 +0900, Jingoo Han wrote:
> > []
> > > ./drivers/pci/host/pci-mvebu.c
> > > static void __iomem *mvebu_pcie_map_registers(struct platform_device *pdev,
> > > 		struct device_node *np, struct mvebu_pcie_port *port)
> > > {
> > > 	.....
> > > 
> > > 	ret = of_address_to_resource(np, 0, &regs);
> > > 	if (ret)
> > > -		return ERR_PTR(ret);
> > > +		return (void __iomem *)ERR_PTR(ret);
> > > 
> > > However, other engineers said that "(void __iomem *)ERR_PTR(ret)"
> > > is not a general pattern. I cannot find the proper method to resolve
> > > this sparse warning.
> > > 
> > > In this case, how can I resolve this sparse warning?
> > 
> > I think there's no problem using the cast.  It's not a
> > pattern because it's not been required before as function
> > returns have not previously been declared __iomem.
> > 
> > Or, perhaps the arm|hexagon specific #define below could
> > be made generic so it could be used.
> > 
> > arch/arm/include/asm/io.h:#define IOMEM(x)	((void __force __iomem *)(x))
> > 
> > 	return IOMEM(ERR_PTR(ret));
> > 
> > There aren't any current uses of return IOMEM(foo) either
> > though so the direct cast is probably more appropriate.
> 
> I don't think that #define is appropriate for any non-constant value;
> almost any instance of that __force cast applied to a variable would be
> better written by fixing types more appropriately.  I'd suggest
> updating that #define for both architectures to use BUILD_BUG_ON to
> assert __builtin_constant_p on its argument.

That's probably a good idea.

There are at least a couple things would likely need to be
fixed up if the #define were changed.

arch/arm/plat-omap/sram.c
drivers/clocksource/bcm_kona_timer.c

I've no issue with either a direct cast or your idea of
a new static inline like ERR_PTR_IO or IOMEM_ERR_PTR or
any other appropriate name.