| Message ID | 1450198746-14161-1-git-send-email-toshi.kani@hpe.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
On Tue, Dec 15, 2015 at 8:59 AM, Toshi Kani <toshi.kani@hpe.com> 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@gnu.org/msg23961.html > Reported-by: Robert Elliott <elliott@hpe.com> > Signed-off-by: Toshi Kani <toshi.kani@hpe.com> > Cc: Rafael J. Wysocki <rjw@rjwysocki.net> > Cc: Dan Williams <dan.j.williams@intel.com> > --- > 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.
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@hpe.com> 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@gnu.org/msg23961.html > > Reported-by: Robert Elliott <elliott@hpe.com> > > Signed-off-by: Toshi Kani <toshi.kani@hpe.com> > > Cc: Rafael J. Wysocki <rjw@rjwysocki.net> > > Cc: Dan Williams <dan.j.williams@intel.com> > > --- > > 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. Thanks, -Toshi
On Tue, Dec 15, 2015 at 9:49 AM, Toshi Kani <toshi.kani@hpe.com> 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@hpe.com> 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@gnu.org/msg23961.html >> > Reported-by: Robert Elliott <elliott@hpe.com> >> > Signed-off-by: Toshi Kani <toshi.kani@hpe.com> >> > Cc: Rafael J. Wysocki <rjw@rjwysocki.net> >> > Cc: Dan Williams <dan.j.williams@intel.com> >> > --- >> > 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?
On Tue, 2015-12-15 at 09:53 -0800, Dan Williams wrote: > On Tue, Dec 15, 2015 at 9:49 AM, Toshi Kani <toshi.kani@hpe.com> 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@hpe.com> > > > 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@gnu.org/msg23961.html > > > > Reported-by: Robert Elliott <elliott@hpe.com> > > > > Signed-off-by: Toshi Kani <toshi.kani@hpe.com> > > > > Cc: Rafael J. Wysocki <rjw@rjwysocki.net> > > > > Cc: Dan Williams <dan.j.williams@intel.com> > > > > --- > > > > 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? No, having EFI_STUB does not break the legacy BIOS boot. Thanks, -Toshi
On Tue, Dec 15, 2015 at 9:56 AM, Toshi Kani <toshi.kani@hpe.com> wrote: > On Tue, 2015-12-15 at 09:53 -0800, Dan Williams wrote: [..] >> > > > 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? > > No, having EFI_STUB does not break the legacy BIOS boot. > Sorry let me clarify. A kernel compiled without EFI_STUB loses NFIT support. Non-EFI and legacy-boot systems don't have this grub bug to worry about, so the dependency is overly restrictive, "breaks", those environments.
On Tue, 2015-12-15 at 10:05 -0800, Dan Williams wrote: > On Tue, Dec 15, 2015 at 9:56 AM, Toshi Kani <toshi.kani@hpe.com> wrote: > > On Tue, 2015-12-15 at 09:53 -0800, Dan Williams wrote: > [..] > > > > > > 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? > > > > No, having EFI_STUB does not break the legacy BIOS boot. > > > > Sorry let me clarify. A kernel compiled without EFI_STUB loses NFIT > support. I think it is the other way around. When NFIT is needed, one will enable all dependent options. So, an NFIT kernel will have to have EFI code as a result, even though it runs on the legacy BIOS. I believe distributed kernels enable EFI already so that they run on both legacy and EFI platforms. Of course, one may wish to build a smallest kernel specific to NFIT & legacy BIOS platforms and want to disable EFI code, but that seems to be a special case to me. > Non-EFI and legacy-boot systems don't have this grub bug to > worry about, so the dependency is overly restrictive, "breaks", those > environments. Yes, we do not have the grub bug on legacy BIOS (although PMEM ranges will show up as Reserved in the iomem table). I do not think NFIT being dependent on EFI is overly restrictive, but I may be wrong and I agree to drop this patch if that is indeed the case. Thanks, -Toshi
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
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@gnu.org/msg23961.html Reported-by: Robert Elliott <elliott@hpe.com> Signed-off-by: Toshi Kani <toshi.kani@hpe.com> Cc: Rafael J. Wysocki <rjw@rjwysocki.net> Cc: Dan Williams <dan.j.williams@intel.com> --- drivers/acpi/Kconfig | 1 + 1 file changed, 1 insertion(+)