[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 at intel.com>
> >> Signed-off-by: Thomas Zimmermann <tzimmermann at 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) ||