On Sat, Nov 30, 2024 at 09:07:39AM -0800, Guenter Roeck wrote:
> On 11/30/24 08:15, Greg KH wrote:
> > On Sat, Nov 30, 2024 at 08:10:55AM -0800, Guenter Roeck wrote:
> > > Hi,
> > >
> > > On Fri, Nov 29, 2024 at 05:27:53AM +0100, Greg KH wrote:
> > > > The following changes since commit 8cf0b93919e13d1e8d4466eb4080a4c4d9d66d7b:
> > > >
> > > > Linux 6.12-rc2 (2024-10-06 15:32:27 -0700)
> > > >
> > > > are available in the Git repository at:
> > > >
> > > > git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git tags/staging-6.13-rc1
> > > >
> > > > for you to fetch changes up to 114eae3c9fde35220cca623840817a740a2eb7b3:
> > > >
> > > > Staging: gpib: gpib_os.c - Remove unnecessary OOM message (2024-11-10 08:04:18 +0100)
> > > >
> > > > ----------------------------------------------------------------
> > > [ ...]
> > >
> > > > Dave Penkler (33):
> > > > staging: gpib: Add common include files for GPIB drivers
> > > > staging: gpib: Add user api include files
> > > > staging: gpib: Add GPIB common core driver
> > > > staging: gpib: Add tms9914 GPIB chip driver
> > > > staging: gpib: Add nec7210 GPIB chip driver
> > > > staging: gpib: Add HP/Agilent/Keysight 8235xx PCI GPIB driver
> > > > staging: gpib: Add Agilent/Keysight 82357x USB GPIB driver
> > > > staging: gpib: Add Computer Boards GPIB driver
> > >
> > > I seem to be unable to find the patch introducing the problem (the link
> > > provided with the patch is invalid), so I report it here.
> > >
> > > With i386 allmodconfig builds:
> > >
> > > Building i386:allyesconfig ... failed
> > > --------------
> > > Error log:
> > > drivers/staging/gpib/cec/cec_gpib.c: In function 'cec_pci_attach':
> > > drivers/staging/gpib/cec/cec_gpib.c:300:28: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast]
> > > 300 | nec_priv->iobase = (void *)(pci_resource_start(cec_priv->pci_device, 3));
> > > | ^
> > > drivers/staging/gpib/ines/ines_gpib.c: In function 'ines_common_pci_attach':
> > > drivers/staging/gpib/ines/ines_gpib.c:783:28: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast]
> > > 783 | nec_priv->iobase = (void *)(pci_resource_start(ines_priv->pci_device,
> > > | ^
> > >
> > > pci_resource_start() returns resource_size_t, which is not a pointer, and thus
> > > can not be cast to one.
> >
> > This is odd, why hasn't 0-day or any other build testing found this?
>
> Good question. Another good question is why I see this only with i386 builds,
> but not with other 32-bit builds. It should be easy to reproduce, though.
>
> make ARCH=i386 allmodconfig
> make ARCH-i386 drivers/staging/gpib/cec/cec_gpib.o
>
> does it for me, independent of gcc version (I tried 11.4 and 13.3).
> I don't see it with clang.
>
> Having said this, using the return value from pci_resource_start() directly as pointer
> is quite unusual. Typically drivers use ioremap(), request_region(), pci_iomap(), or
> a similar function on it to get a pointer.
>
> Guenter
>
That is weird: the type of resource.start is resource_size_t which resolves to u32 via phys_addr_t on i386 which should be the same size as void *
For compile check purposes simply changing iobase type to phys_addr_t the following error message appears:
drivers/staging/gpib/ines/ines_gpib.c: In function 'ines_common_pci_attach':
drivers/staging/gpib/ines/ines_gpib.c:783:28: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast]
783 | nec_priv->iobase = (void *)(pci_resource_start(ines_priv->pci_device,
| ^
drivers/staging/gpib/ines/ines_gpib.c:783:26: error: assignment to 'phys_addr_t' {aka 'long long unsigned int'} from 'void *' makes integer from pointer without a cast [-Wint-conversion]
783 | nec_priv->iobase = (void *)(pci_resource_start(ines_priv->pci_device,
| ^
It would seem that for some reason phys_addr_t resolves to long long unsigned int
-Dave
