On Tue, Dec 15, 2015 at 9:49 AM, Toshi Kani <toshi.kani@xxxxxxx> wrote: > On Tue, 2015-12-15 at 09:21 -0800, Dan Williams wrote: >> On Tue, Dec 15, 2015 at 8:59 AM, Toshi Kani <toshi.kani@xxxxxxx> wrote: >> > ACPI 6.0 defines NFIT table and new persistent memory types for >> > EFI memory table (EFI_PERSISTENT_MEMORY) and e820 table (E820_PMEM). >> > >> > setup_e820() enabled by EFI_STUB converts EFI_PERSISTENT_MEMORY to >> > E820_PMEM for the e820_map table on x86 EFI platforms. When EFI_STUB >> > is disabled, x86 kernels rely on the bootloader to perform this >> > conversion. >> > >> > It was found that the upstream grub bootloader since 2012 has a bug >> > that converts EFI_PERSISTENT_MEMORY (or any new type) to E820_RAM, >> > which causes the kernel to use persistent memory ranges as regular >> > memory and corrupts the data in NVDIMM. >> > >> > Therefore, this patch sets ACPI_NFIT to depend on EFI_STUB on x86. >> > This assures that ACPI_NFIT kernels are self-contained and are >> > protected from the upstream grub bug on x86. >> > >> > Note, X86_PMEM_LEGACY allows the kernel to use the pmem driver on >> > pre-ACPI 6.0 platforms, and does not require ACPI_NFIT enabled. >> > >> > References: >> > http://www.mail-archive.com/grub-devel@xxxxxxx/msg23961.html >> > Reported-by: Robert Elliott <elliott@xxxxxxx> >> > Signed-off-by: Toshi Kani <toshi.kani@xxxxxxx> >> > Cc: Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> >> > Cc: Dan Williams <dan.j.williams@xxxxxxxxx> >> > --- >> > drivers/acpi/Kconfig | 1 + >> > 1 file changed, 1 insertion(+) >> > >> > diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig >> > index 5eef4cb..5368baa 100644 >> > --- a/drivers/acpi/Kconfig >> > +++ b/drivers/acpi/Kconfig >> > @@ -441,6 +441,7 @@ config ACPI_NFIT >> > depends on PHYS_ADDR_T_64BIT >> > depends on BLK_DEV >> > depends on ARCH_HAS_MMIO_FLUSH >> > + depends on !X86 || (X86 && EFI_STUB) >> > select LIBNVDIMM >> > help >> > Infrastructure to probe ACPI 6 compliant platforms for >> >> This seems wrong to me. >> >> In general Kconfig "depends on" are only about compile-time code >> dependency, not about working around random bugs in external projects. > > I agree that they do not have a compile dependency, and it looks rather > odd. On the other hand, proper support of the ACPI NFIT extensions has a > functional dependency to setup_e820() as this is the only code that can > handle the EFI_PERSISTENT_MEMORY type today. What about non-EFI/legeacy-boot systems that only emit e820-type-7? We lose support for them with this patch, right? -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
