[PATCH 0/2] drm/ofdrm: Fix sparse warnings

[PATCH 0/2] drm/ofdrm: Fix sparse warnings

[Thomas Zimmermann]

Fix two types of sparse warnings in ofdrm. Reported by the LKP bot.

Thomas Zimmermann (2):
  drm/ofdrm: Cast PCI IDs to u32 for comparing
  drm/ofdrm: Cast error pointers to void __iomem *

 drivers/gpu/drm/tiny/ofdrm.c | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)


base-commit: 746559738f1335241ea686566cb654847c20d7a4
-- 
2.38.0

[PATCH 1/2] drm/ofdrm: Cast PCI IDs to u32 for comparing

[Thomas Zimmermann]

Properties of 32-bit integers are returned from the OF device tree
as type __be32. Cast PCI vendor and device IDs from __be32 to u32
before comparing them to constants. Fixes sparse warnings shown below.

  drivers/gpu/drm/tiny/ofdrm.c:237:17: warning: restricted __be32 degrades to integer
  drivers/gpu/drm/tiny/ofdrm.c:238:18: warning: restricted __be32 degrades to integer
  drivers/gpu/drm/tiny/ofdrm.c:238:54: warning: restricted __be32 degrades to integer

See [1] for the bug report.

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://lore.kernel.org/dri-devel/202210192208.D888I6X7-lkp@intel.com/ # [1]
---
 drivers/gpu/drm/tiny/ofdrm.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/tiny/ofdrm.c b/drivers/gpu/drm/tiny/ofdrm.c
index 0e1cc2369afcc..0da8b248ccc6e 100644
--- a/drivers/gpu/drm/tiny/ofdrm.c
+++ b/drivers/gpu/drm/tiny/ofdrm.c
@@ -231,8 +231,11 @@ static u64 display_get_address_of(struct drm_device *dev, struct device_node *of
 	return address;
 }
 
-static bool is_avivo(__be32 vendor, __be32 device)
+static bool is_avivo(__be32 vendor_id, __be32 device_id)
 {
+	u32 vendor = (__force u32)vendor_id;
+	u32 device = (__force u32)device_id;
+
 	/* This will match most R5xx */
 	return (vendor == PCI_VENDOR_ID_ATI) &&
 	       ((device >= PCI_VENDOR_ID_ATI_R520 && device < 0x7800) ||
-- 
2.38.0

[PATCH 2/2] drm/ofdrm: Cast error pointers to void __iomem *

[Thomas Zimmermann]

Cast error pointers when returning them as void __iomem *. Fixes
a number of Sparse warnings, such as the ones shown below.

../drivers/gpu/drm/tiny/ofdrm.c:439:31: warning: incorrect type in return expression (different address spaces)
../drivers/gpu/drm/tiny/ofdrm.c:439:31:    expected void [noderef] __iomem *
../drivers/gpu/drm/tiny/ofdrm.c:439:31:    got void *
../drivers/gpu/drm/tiny/ofdrm.c:442:31: warning: incorrect type in return expression (different address spaces)
../drivers/gpu/drm/tiny/ofdrm.c:442:31:    expected void [noderef] __iomem *
../drivers/gpu/drm/tiny/ofdrm.c:442:31:    got void *

See [1] for the bug report.

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://lore.kernel.org/dri-devel/202210200016.yiQzPIy0-lkp@intel.com/ # [1]
---
 drivers/gpu/drm/tiny/ofdrm.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/tiny/ofdrm.c b/drivers/gpu/drm/tiny/ofdrm.c
index 0da8b248ccc6e..b358da3840ad0 100644
--- a/drivers/gpu/drm/tiny/ofdrm.c
+++ b/drivers/gpu/drm/tiny/ofdrm.c
@@ -436,21 +436,21 @@ static void __iomem *get_cmap_address_of(struct ofdrm_device *odev, struct devic
 	if (!addr_p)
 		addr_p = of_get_address(of_node, bar_no, &max_size, &flags);
 	if (!addr_p)
-		return ERR_PTR(-ENODEV);
+		return (void __iomem *)ERR_PTR(-ENODEV);
 
 	if ((flags & (IORESOURCE_IO | IORESOURCE_MEM)) == 0)
-		return ERR_PTR(-ENODEV);
+		return (void __iomem *)ERR_PTR(-ENODEV);
 
 	if ((offset + size) >= max_size)
-		return ERR_PTR(-ENODEV);
+		return (void __iomem *)ERR_PTR(-ENODEV);
 
 	address = of_translate_address(of_node, addr_p);
 	if (address == OF_BAD_ADDR)
-		return ERR_PTR(-ENODEV);
+		return (void __iomem *)ERR_PTR(-ENODEV);
 
 	mem = devm_ioremap(dev->dev, address + offset, size);
 	if (!mem)
-		return ERR_PTR(-ENOMEM);
+		return (void __iomem *)ERR_PTR(-ENOMEM);
 
 	return mem;
 }
@@ -468,7 +468,7 @@ static void __iomem *ofdrm_mach64_cmap_ioremap(struct ofdrm_device *odev,
 
 	cmap_base = devm_ioremap(dev->dev, address, 0x1000);
 	if (!cmap_base)
-		return ERR_PTR(-ENOMEM);
+		return (void __iomem *)ERR_PTR(-ENOMEM);
 
 	return cmap_base;
 }
@@ -627,11 +627,11 @@ static void __iomem *ofdrm_qemu_cmap_ioremap(struct ofdrm_device *odev,
 
 	address = of_translate_address(of_node, io_of_addr);
 	if (address == OF_BAD_ADDR)
-		return ERR_PTR(-ENODEV);
+		return (void __iomem *)ERR_PTR(-ENODEV);
 
 	cmap_base = devm_ioremap(dev->dev, address + 0x3c8, 2);
 	if (!cmap_base)
-		return ERR_PTR(-ENOMEM);
+		return (void __iomem *)ERR_PTR(-ENOMEM);
 
 	return cmap_base;
 }
-- 
2.38.0

Re: [PATCH 1/2] drm/ofdrm: Cast PCI IDs to u32 for comparing

[Alexander Stein]

Hello Thomas,

Am Donnerstag, 27. Oktober 2022, 13:57:06 CEST schrieb Thomas Zimmermann:
> Properties of 32-bit integers are returned from the OF device tree
> as type __be32. Cast PCI vendor and device IDs from __be32 to u32
> before comparing them to constants. Fixes sparse warnings shown below.
> 
>   drivers/gpu/drm/tiny/ofdrm.c:237:17: warning: restricted __be32 degrades
> to integer drivers/gpu/drm/tiny/ofdrm.c:238:18: warning: restricted __be32
> degrades to integer drivers/gpu/drm/tiny/ofdrm.c:238:54: warning:
> restricted __be32 degrades to integer
> 
> See [1] for the bug report.
> 
> Reported-by: kernel test robot <lkp@intel.com>
> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
> Link: https://lore.kernel.org/dri-devel/202210192208.D888I6X7-lkp@intel.com/
> # [1] ---
>  drivers/gpu/drm/tiny/ofdrm.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/tiny/ofdrm.c b/drivers/gpu/drm/tiny/ofdrm.c
> index 0e1cc2369afcc..0da8b248ccc6e 100644
> --- a/drivers/gpu/drm/tiny/ofdrm.c
> +++ b/drivers/gpu/drm/tiny/ofdrm.c
> @@ -231,8 +231,11 @@ static u64 display_get_address_of(struct drm_device
> *dev, struct device_node *of return address;
>  }
> 
> -static bool is_avivo(__be32 vendor, __be32 device)
> +static bool is_avivo(__be32 vendor_id, __be32 device_id)
>  {
> +	u32 vendor = (__force u32)vendor_id;
> +	u32 device = (__force u32)device_id;

I don't have much context, but just from reading this, shouldn't this be 
be32_to_cpu() instead?

Best regards,
Alexander

> +
>  	/* This will match most R5xx */
>  	return (vendor == PCI_VENDOR_ID_ATI) &&
>  	       ((device >= PCI_VENDOR_ID_ATI_R520 && device < 0x7800) ||

Re: [PATCH 1/2] drm/ofdrm: Cast PCI IDs to u32 for comparing

[Thomas Zimmermann]

[-- Attachment #1.1: Type: text/plain, Size: 2423 bytes --]

Hi

Am 27.10.22 um 15:07 schrieb Alexander Stein:
> Hello Thomas,
> 
> Am Donnerstag, 27. Oktober 2022, 13:57:06 CEST schrieb Thomas Zimmermann:
>> Properties of 32-bit integers are returned from the OF device tree
>> as type __be32. Cast PCI vendor and device IDs from __be32 to u32
>> before comparing them to constants. Fixes sparse warnings shown below.
>>
>>    drivers/gpu/drm/tiny/ofdrm.c:237:17: warning: restricted __be32 degrades
>> to integer drivers/gpu/drm/tiny/ofdrm.c:238:18: warning: restricted __be32
>> degrades to integer drivers/gpu/drm/tiny/ofdrm.c:238:54: warning:
>> restricted __be32 degrades to integer
>>
>> See [1] for the bug report.
>>
>> Reported-by: kernel test robot <lkp@intel.com>
>> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
>> Link: https://lore.kernel.org/dri-devel/202210192208.D888I6X7-lkp@intel.com/
>> # [1] ---
>>   drivers/gpu/drm/tiny/ofdrm.c | 5 ++++-
>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/tiny/ofdrm.c b/drivers/gpu/drm/tiny/ofdrm.c
>> index 0e1cc2369afcc..0da8b248ccc6e 100644
>> --- a/drivers/gpu/drm/tiny/ofdrm.c
>> +++ b/drivers/gpu/drm/tiny/ofdrm.c
>> @@ -231,8 +231,11 @@ static u64 display_get_address_of(struct drm_device
>> *dev, struct device_node *of return address;
>>   }
>>
>> -static bool is_avivo(__be32 vendor, __be32 device)
>> +static bool is_avivo(__be32 vendor_id, __be32 device_id)
>>   {
>> +	u32 vendor = (__force u32)vendor_id;
>> +	u32 device = (__force u32)device_id;
> 
> I don't have much context, but just from reading this, shouldn't this be
> be32_to_cpu() instead?

I should have explained that in the commit message. The values are 
supposed to be in big endian. We compare to PCI ids. The code originally 
was taken from [1], which does the right thing. The next version will 
add this info to the commit message.

Best regards
Thomas

[1] 
https://elixir.bootlin.com/linux/v6.0.5/source/drivers/video/fbdev/offb.c#L357

> 
> Best regards,
> Alexander
> 
>> +
>>   	/* This will match most R5xx */
>>   	return (vendor == PCI_VENDOR_ID_ATI) &&
>>   	       ((device >= PCI_VENDOR_ID_ATI_R520 && device < 0x7800) ||
> 
> 
> 
> 

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Ivo Totev

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

Re: [PATCH 1/2] drm/ofdrm: Cast PCI IDs to u32 for comparing

[Alexander Stein]

Hi Thomas,

Am Donnerstag, 27. Oktober 2022, 16:04:34 CEST schrieb Thomas Zimmermann:
> * PGP Signed: 10/27/2022 at 04:04:34 PM
> 
> Hi
> 
> Am 27.10.22 um 15:07 schrieb Alexander Stein:
> > Hello Thomas,
> > 
> > Am Donnerstag, 27. Oktober 2022, 13:57:06 CEST schrieb Thomas Zimmermann:
> >> Properties of 32-bit integers are returned from the OF device tree
> >> as type __be32. Cast PCI vendor and device IDs from __be32 to u32
> >> before comparing them to constants. Fixes sparse warnings shown below.
> >> 
> >>    drivers/gpu/drm/tiny/ofdrm.c:237:17: warning: restricted __be32
> >>    degrades
> >> 
> >> to integer drivers/gpu/drm/tiny/ofdrm.c:238:18: warning: restricted
> >> __be32
> >> degrades to integer drivers/gpu/drm/tiny/ofdrm.c:238:54: warning:
> >> restricted __be32 degrades to integer
> >> 
> >> See [1] for the bug report.
> >> 
> >> Reported-by: kernel test robot <lkp@intel.com>
> >> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
> >> Link:
> >> https://lore.kernel.org/dri-devel/202210192208.D888I6X7-lkp@intel.com/ #
> >> [1] ---
> >> 
> >>   drivers/gpu/drm/tiny/ofdrm.c | 5 ++++-
> >>   1 file changed, 4 insertions(+), 1 deletion(-)
> >> 
> >> diff --git a/drivers/gpu/drm/tiny/ofdrm.c b/drivers/gpu/drm/tiny/ofdrm.c
> >> index 0e1cc2369afcc..0da8b248ccc6e 100644
> >> --- a/drivers/gpu/drm/tiny/ofdrm.c
> >> +++ b/drivers/gpu/drm/tiny/ofdrm.c
> >> @@ -231,8 +231,11 @@ static u64 display_get_address_of(struct drm_device
> >> *dev, struct device_node *of return address;
> >> 
> >>   }
> >> 
> >> -static bool is_avivo(__be32 vendor, __be32 device)
> >> +static bool is_avivo(__be32 vendor_id, __be32 device_id)
> >> 
> >>   {
> >> 
> >> +	u32 vendor = (__force u32)vendor_id;
> >> +	u32 device = (__force u32)device_id;
> > 
> > I don't have much context, but just from reading this, shouldn't this be
> > be32_to_cpu() instead?
> 
> I should have explained that in the commit message. The values are
> supposed to be in big endian. We compare to PCI ids. The code originally
> was taken from [1], which does the right thing. The next version will
> add this info to the commit message.
> 
> Best regards
> Thomas
> 
> [1]
> https://elixir.bootlin.com/linux/v6.0.5/source/drivers/video/fbdev/offb.c#L3
> 57

Thanks for the link. The commit message for that original change [2] indicates 
this was a fix on a Powerstation which happens to be a PowerPC aka big-endian.
This this happens to work as intended, but not on little-endian machines.
vendor_id is __be32 (big-endian) and you compare this to PCI_VENDOR_ID_ATI 
(0x1002) which is native-endian. I guess you see the problem.
See also [3] where a property is converted properly.

Thinking about it, instead of calling 'is_avivo(*vendor_p, *device_p)'
I'd prefer something like 'is_avivo(be32_to_cpup(vendor_p), 
be32_to_cpup(device_p))', so there is no need to pass __be32 around.

Best regards,
Alexander

[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/
commit/?id=57a20d8fb0d2a05abe40abd6bb29e3f923721f1b
[3] https://elixir.bootlin.com/linux/v6.0.5/source/drivers/bus/fsl-mc/fsl-mc-bus.c#L1027

> > Best regards,
> > Alexander
> > 
> >> +
> >> 
> >>   	/* This will match most R5xx */
> >>   	return (vendor == PCI_VENDOR_ID_ATI) &&
> >>   	
> >>   	       ((device >= PCI_VENDOR_ID_ATI_R520 && device < 0x7800) ||

Re: [PATCH 1/2] drm/ofdrm: Cast PCI IDs to u32 for comparing

[Thomas Zimmermann]

[-- Attachment #1.1: Type: text/plain, Size: 3799 bytes --]

Hi

Am 28.10.22 um 08:33 schrieb Alexander Stein:
> Hi Thomas,
> 
> Am Donnerstag, 27. Oktober 2022, 16:04:34 CEST schrieb Thomas Zimmermann:
>> * PGP Signed: 10/27/2022 at 04:04:34 PM
>>
>> Hi
>>
>> Am 27.10.22 um 15:07 schrieb Alexander Stein:
>>> Hello Thomas,
>>>
>>> Am Donnerstag, 27. Oktober 2022, 13:57:06 CEST schrieb Thomas Zimmermann:
>>>> Properties of 32-bit integers are returned from the OF device tree
>>>> as type __be32. Cast PCI vendor and device IDs from __be32 to u32
>>>> before comparing them to constants. Fixes sparse warnings shown below.
>>>>
>>>>     drivers/gpu/drm/tiny/ofdrm.c:237:17: warning: restricted __be32
>>>>     degrades
>>>>
>>>> to integer drivers/gpu/drm/tiny/ofdrm.c:238:18: warning: restricted
>>>> __be32
>>>> degrades to integer drivers/gpu/drm/tiny/ofdrm.c:238:54: warning:
>>>> restricted __be32 degrades to integer
>>>>
>>>> See [1] for the bug report.
>>>>
>>>> Reported-by: kernel test robot <lkp@intel.com>
>>>> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
>>>> Link:
>>>> https://lore.kernel.org/dri-devel/202210192208.D888I6X7-lkp@intel.com/ #
>>>> [1] ---
>>>>
>>>>    drivers/gpu/drm/tiny/ofdrm.c | 5 ++++-
>>>>    1 file changed, 4 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/tiny/ofdrm.c b/drivers/gpu/drm/tiny/ofdrm.c
>>>> index 0e1cc2369afcc..0da8b248ccc6e 100644
>>>> --- a/drivers/gpu/drm/tiny/ofdrm.c
>>>> +++ b/drivers/gpu/drm/tiny/ofdrm.c
>>>> @@ -231,8 +231,11 @@ static u64 display_get_address_of(struct drm_device
>>>> *dev, struct device_node *of return address;
>>>>
>>>>    }
>>>>
>>>> -static bool is_avivo(__be32 vendor, __be32 device)
>>>> +static bool is_avivo(__be32 vendor_id, __be32 device_id)
>>>>
>>>>    {
>>>>
>>>> +	u32 vendor = (__force u32)vendor_id;
>>>> +	u32 device = (__force u32)device_id;
>>>
>>> I don't have much context, but just from reading this, shouldn't this be
>>> be32_to_cpu() instead?
>>
>> I should have explained that in the commit message. The values are
>> supposed to be in big endian. We compare to PCI ids. The code originally
>> was taken from [1], which does the right thing. The next version will
>> add this info to the commit message.
>>
>> Best regards
>> Thomas
>>
>> [1]
>> https://elixir.bootlin.com/linux/v6.0.5/source/drivers/video/fbdev/offb.c#L3
>> 57
> 
> Thanks for the link. The commit message for that original change [2] indicates
> this was a fix on a Powerstation which happens to be a PowerPC aka big-endian.
> This this happens to work as intended, but not on little-endian machines.
> vendor_id is __be32 (big-endian) and you compare this to PCI_VENDOR_ID_ATI
> (0x1002) which is native-endian. I guess you see the problem.
> See also [3] where a property is converted properly.
> 
> Thinking about it, instead of calling 'is_avivo(*vendor_p, *device_p)'
> I'd prefer something like 'is_avivo(be32_to_cpup(vendor_p),
> be32_to_cpup(device_p))', so there is no need to pass __be32 around.

Makes sense. Thanks for reviewing.

Best regards
Thomas

> 
> Best regards,
> Alexander
> 
> [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/
> commit/?id=57a20d8fb0d2a05abe40abd6bb29e3f923721f1b
> [3] https://elixir.bootlin.com/linux/v6.0.5/source/drivers/bus/fsl-mc/fsl-mc-bus.c#L1027
> 
>>> Best regards,
>>> Alexander
>>>
>>>> +
>>>>
>>>>    	/* This will match most R5xx */
>>>>    	return (vendor == PCI_VENDOR_ID_ATI) &&
>>>>    	
>>>>    	       ((device >= PCI_VENDOR_ID_ATI_R520 && device < 0x7800) ||
> 
> 
> 
> 

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Ivo Totev

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]