On 14.10.20 19:58, Ard Biesheuvel wrote: > On Wed, 14 Oct 2020 at 19:45, Heinrich Schuchardt <xypron.glpk@xxxxxx> wrote: >> >> On 14.10.20 19:31, Heinrich Schuchardt wrote: >>> Dear all, >>> >>> the fwts fails on U-Boot due to testing for a non-existent >>> RuntimeServicesSupported variable. >>> >>> If you look at the UEFI specification 2.8 (Errata B) [1] you will >>> discover in the change log: >>> >>> 2.8 A2049 >>> RuntimeServicesSupported EFI variable should be a config table >>> February 2020 >>> >>> Please, read the configuration table to determine if a runtime service >>> is available on UEFI 2.8 systems. >>> >>> On lower UEFI firmware version neither the variable nor the table exists. >>> >>> Best regards >>> >>> Heinrich >>> >>> [1] UEFI Specification Version 2.8 (Errata B) (released June 2020), >>> https://uefi.org/sites/default/files/resources/UEFI%20Spec%202.8B%20May%202020.pdf >>> >> >> Hello Ard, >> >> what is your idea how the EFI_RT_PROPERTIES_TABLE shall be exposed to >> the efi_test driver? >> >> Will the EFI runtime wrapper simply return EFI_UNSUPPORTED if the >> function is not marked as supported in the table? Or will the >> configuration table itself be make available? >> > > The UEFI spec permits that runtime services return EFI_UNSUPPORTED at > runtime, but requires that they are marked as such in the > EFI_RT_PROPERTIES_TABLE. > > So assuming that the purpose of efi_test is compliance with the spec, > it should only allow EFI_UNSUPPORTED as a return value for each of the > tested runtime services if it is omitted from > efi.runtime_supported_mask. > > Since the efi_test ioctl returns both an error code and the actual EFI > status code, we should only fail the call on a EFI_UNSUPPORTED status > code if the RTPROP mask does not allow that. > > E.g., > > --- a/drivers/firmware/efi/test/efi_test.c > +++ b/drivers/firmware/efi/test/efi_test.c > @@ -265,7 +265,12 @@ static long efi_runtime_set_variable(unsigned long arg) > goto out; > } > > - rv = status == EFI_SUCCESS ? 0 : -EINVAL; > + if (status == EFI_SUCCESS || > + (status == EFI_UNSUPPORTED && > + !efi_rt_services_supported(EFI_RT_SUPPORTED_SET_VARIABLE))) > + rv = 0; > + else > + rv = -EINVAL; > > out: > kfree(data); > > > Do you think that could work? > The current fwts implementation assumes that EFI_UNSUPPORTED leads to ioctl() returning -1. This value should not be changed. It would be preferable to use another error code than -EINVAL, e.g. -EDOM if there is a mismatch with the EFI_RT_PROPERTIES_TABLE configuration table. Then a future verision of fwts can evaluate errno to discover the problem. Do I read you correctly: the EFI runtime wrapper does not fend of calls to runtime services marked as disallowed in EFI_RT_PROPERTIES_TABLE? Directly returning an error code might help to avoid crashes on non-compliant firmware. Best regards Heinrich
