efi/x86: Fix incorrect invocation of PciIo->Attributes()
authorArd Biesheuvel <[email protected]>
Sat, 23 Jun 2018 21:19:03 +0000 (23:19 +0200)
committerIngo Molnar <[email protected]>
Sun, 24 Jun 2018 07:05:58 +0000 (09:05 +0200)
commit2e6eb40ca5eb53836d18f3b9ac61ff2e0b417038
treee7f2d6a0b8934d0e481bbeff5dbec059d43db8dd
parent52e1cf2d19c2e62e6a81b8de3f7320d033917dd5
efi/x86: Fix incorrect invocation of PciIo->Attributes()

The following commit:

  2c3625cb9fa2 ("efi/x86: Fold __setup_efi_pci32() and __setup_efi_pci64() into one function")

... merged the two versions of __setup_efi_pciXX(), without taking into
account that the 32-bit version used a rather dodgy trick to pass an
immediate 0 constant as argument for a uint64_t parameter.

The issue is caused by the fact that on x86, UEFI protocol method calls
are redirected via struct efi_config::call(), which is a variadic function,
and so the compiler has to infer the types of the parameters from the
arguments rather than from the prototype.

As the 32-bit x86 calling convention passes arguments via the stack,
passing the unqualified constant 0 twice is the same as passing 0ULL,
which is why the 32-bit code in __setup_efi_pci32() contained the
following call:

  status = efi_early->call(pci->attributes, pci,
                           EfiPciIoAttributeOperationGet, 0, 0,
                           &attributes);

to invoke this UEFI protocol method:

  typedef
  EFI_STATUS
  (EFIAPI *EFI_PCI_IO_PROTOCOL_ATTRIBUTES) (
    IN  EFI_PCI_IO_PROTOCOL                     *This,
    IN  EFI_PCI_IO_PROTOCOL_ATTRIBUTE_OPERATION Operation,
    IN  UINT64                                  Attributes,
    OUT UINT64                                  *Result OPTIONAL
    );

After the merge, we inadvertently ended up with this version for both
32-bit and 64-bit builds, breaking the latter.

So replace the two zeroes with the explicitly typed constant 0ULL,
which works as expected on both 32-bit and 64-bit builds.

Wilfried tested the 64-bit build, and I checked the generated assembly
of a 32-bit build with and without this patch, and they are identical.

Reported-by: Wilfried Klaebe <[email protected]>
Tested-by: Wilfried Klaebe <[email protected]>
Signed-off-by: Ard Biesheuvel <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>
arch/x86/boot/compressed/eboot.c