| Message ID | 1407952383-1214-1-git-send-email-ard.biesheuvel@linaro.org |
|---|---|
| State | Accepted |
| Commit | 4190312beb2acfb7bfb1bb971e24a759aa96b0e8 |
| Headers | show |
Hi Ard, On Wed, Aug 13, 2014 at 06:53:03PM +0100, Ard Biesheuvel wrote: > When booting via UEFI, the kernel Image is loaded at a 4 kB boundary and > the embedded EFI stub is executed in place. The EFI stub relocates the > Image to reside TEXT_OFFSET bytes above a 2 MB boundary, and jumps into > the kernel proper. > > In AArch64, PC relative symbol references are emitted using adrp/add or > adrp/ldr pairs, where the offset into a 4 kB page is resolved using a > separate :lo12: relocation. This implicitly assumes that the code will > always be executed at the same relative offset with respect to a 4 kB > boundary, or the references will point to the wrong address. > > This means we should link the kernel at a 4 kB aligned base address in > order to remain compatible with the base address the UEFI loader uses > when doing the initial load of Image. So update the code that generates > TEXT_OFFSET to choose a multiple of 4 kB. > > At the same time, update the code so it chooses from the interval [0..2MB) > as the author originally intended. > > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> > --- > arch/arm64/Makefile | 2 +- > arch/arm64/kernel/head.S | 8 ++++---- > 2 files changed, 5 insertions(+), 5 deletions(-) > > diff --git a/arch/arm64/Makefile b/arch/arm64/Makefile > index 57833546bf00..2df5e5daeebe 100644 > --- a/arch/arm64/Makefile > +++ b/arch/arm64/Makefile > @@ -39,7 +39,7 @@ head-y := arch/arm64/kernel/head.o > > # The byte offset of the kernel image in RAM from the start of RAM. > ifeq ($(CONFIG_ARM64_RANDOMIZE_TEXT_OFFSET), y) > -TEXT_OFFSET := $(shell awk 'BEGIN {srand(); printf "0x%04x0\n", int(65535 * rand())}') > +TEXT_OFFSET := $(shell awk 'BEGIN {srand(); printf "0x%03x000\n", int(512 * rand())}') > else > TEXT_OFFSET := 0x00080000 > endif > diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S > index b6ca95aee348..ae055a18b04a 100644 > --- a/arch/arm64/kernel/head.S > +++ b/arch/arm64/kernel/head.S > @@ -38,11 +38,11 @@ > > #define KERNEL_RAM_VADDR (PAGE_OFFSET + TEXT_OFFSET) > > -#if (TEXT_OFFSET & 0xf) != 0 > -#error TEXT_OFFSET must be at least 16B aligned > -#elif (PAGE_OFFSET & 0xfffff) != 0 > +#if (TEXT_OFFSET & 0xfff) != 0 > +#error TEXT_OFFSET must be at least 4kB aligned Minor nit, but can we have "KB" here to match "MB"? > +#elif (PAGE_OFFSET & 0x1fffff) != 0 > #error PAGE_OFFSET must be at least 2MB aligned > -#elif TEXT_OFFSET > 0xfffff > +#elif TEXT_OFFSET > 0x1fffff > #error TEXT_OFFSET must be less than 2MB > #endif I thought I'd left wording in Kconfig.debug, but that doesn't seem to be the case. This looks sane to me, so: Reviewed-by: Mark Rutland <mark.rutland@arm.com> Catalin, can you pick this up? Without it UEFI boot can be broken by ARM64_RANDOMIZE_TEXT_OFFSET. Cheers, Mark.
On 14 August 2014 12:23, Mark Rutland <mark.rutland@arm.com> wrote: > Hi Ard, > > On Wed, Aug 13, 2014 at 06:53:03PM +0100, Ard Biesheuvel wrote: >> When booting via UEFI, the kernel Image is loaded at a 4 kB boundary and >> the embedded EFI stub is executed in place. The EFI stub relocates the >> Image to reside TEXT_OFFSET bytes above a 2 MB boundary, and jumps into >> the kernel proper. >> >> In AArch64, PC relative symbol references are emitted using adrp/add or >> adrp/ldr pairs, where the offset into a 4 kB page is resolved using a >> separate :lo12: relocation. This implicitly assumes that the code will >> always be executed at the same relative offset with respect to a 4 kB >> boundary, or the references will point to the wrong address. >> >> This means we should link the kernel at a 4 kB aligned base address in >> order to remain compatible with the base address the UEFI loader uses >> when doing the initial load of Image. So update the code that generates >> TEXT_OFFSET to choose a multiple of 4 kB. >> >> At the same time, update the code so it chooses from the interval [0..2MB) >> as the author originally intended. >> >> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> >> --- >> arch/arm64/Makefile | 2 +- >> arch/arm64/kernel/head.S | 8 ++++---- >> 2 files changed, 5 insertions(+), 5 deletions(-) >> >> diff --git a/arch/arm64/Makefile b/arch/arm64/Makefile >> index 57833546bf00..2df5e5daeebe 100644 >> --- a/arch/arm64/Makefile >> +++ b/arch/arm64/Makefile >> @@ -39,7 +39,7 @@ head-y := arch/arm64/kernel/head.o >> >> # The byte offset of the kernel image in RAM from the start of RAM. >> ifeq ($(CONFIG_ARM64_RANDOMIZE_TEXT_OFFSET), y) >> -TEXT_OFFSET := $(shell awk 'BEGIN {srand(); printf "0x%04x0\n", int(65535 * rand())}') >> +TEXT_OFFSET := $(shell awk 'BEGIN {srand(); printf "0x%03x000\n", int(512 * rand())}') >> else >> TEXT_OFFSET := 0x00080000 >> endif >> diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S >> index b6ca95aee348..ae055a18b04a 100644 >> --- a/arch/arm64/kernel/head.S >> +++ b/arch/arm64/kernel/head.S >> @@ -38,11 +38,11 @@ >> >> #define KERNEL_RAM_VADDR (PAGE_OFFSET + TEXT_OFFSET) >> >> -#if (TEXT_OFFSET & 0xf) != 0 >> -#error TEXT_OFFSET must be at least 16B aligned >> -#elif (PAGE_OFFSET & 0xfffff) != 0 >> +#if (TEXT_OFFSET & 0xfff) != 0 >> +#error TEXT_OFFSET must be at least 4kB aligned > > Minor nit, but can we have "KB" here to match "MB"? > Ack >> +#elif (PAGE_OFFSET & 0x1fffff) != 0 >> #error PAGE_OFFSET must be at least 2MB aligned >> -#elif TEXT_OFFSET > 0xfffff >> +#elif TEXT_OFFSET > 0x1fffff >> #error TEXT_OFFSET must be less than 2MB >> #endif > > I thought I'd left wording in Kconfig.debug, but that doesn't seem to be > the case. This looks sane to me, so: > > Reviewed-by: Mark Rutland <mark.rutland@arm.com> > > Catalin, can you pick this up? Without it UEFI boot can be broken by > ARM64_RANDOMIZE_TEXT_OFFSET. > I assume the above change does not require a v2.
On Thu, Aug 14, 2014 at 11:31:35AM +0100, Ard Biesheuvel wrote: > On 14 August 2014 12:23, Mark Rutland <mark.rutland@arm.com> wrote: > > On Wed, Aug 13, 2014 at 06:53:03PM +0100, Ard Biesheuvel wrote: > >> diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S > >> index b6ca95aee348..ae055a18b04a 100644 > >> --- a/arch/arm64/kernel/head.S > >> +++ b/arch/arm64/kernel/head.S > >> @@ -38,11 +38,11 @@ > >> > >> #define KERNEL_RAM_VADDR (PAGE_OFFSET + TEXT_OFFSET) > >> > >> -#if (TEXT_OFFSET & 0xf) != 0 > >> -#error TEXT_OFFSET must be at least 16B aligned > >> -#elif (PAGE_OFFSET & 0xfffff) != 0 > >> +#if (TEXT_OFFSET & 0xfff) != 0 > >> +#error TEXT_OFFSET must be at least 4kB aligned > > > > Minor nit, but can we have "KB" here to match "MB"? > > Ack > > >> +#elif (PAGE_OFFSET & 0x1fffff) != 0 > >> #error PAGE_OFFSET must be at least 2MB aligned > >> -#elif TEXT_OFFSET > 0xfffff > >> +#elif TEXT_OFFSET > 0x1fffff > >> #error TEXT_OFFSET must be less than 2MB > >> #endif > > > > I thought I'd left wording in Kconfig.debug, but that doesn't seem to be > > the case. This looks sane to me, so: > > > > Reviewed-by: Mark Rutland <mark.rutland@arm.com> > > > > Catalin, can you pick this up? Without it UEFI boot can be broken by > > ARM64_RANDOMIZE_TEXT_OFFSET. > > I assume the above change does not require a v2. I can fix it up. Thanks.
diff --git a/arch/arm64/Makefile b/arch/arm64/Makefile index 57833546bf00..2df5e5daeebe 100644 --- a/arch/arm64/Makefile +++ b/arch/arm64/Makefile @@ -39,7 +39,7 @@ head-y := arch/arm64/kernel/head.o # The byte offset of the kernel image in RAM from the start of RAM. ifeq ($(CONFIG_ARM64_RANDOMIZE_TEXT_OFFSET), y) -TEXT_OFFSET := $(shell awk 'BEGIN {srand(); printf "0x%04x0\n", int(65535 * rand())}') +TEXT_OFFSET := $(shell awk 'BEGIN {srand(); printf "0x%03x000\n", int(512 * rand())}') else TEXT_OFFSET := 0x00080000 endif diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S index b6ca95aee348..ae055a18b04a 100644 --- a/arch/arm64/kernel/head.S +++ b/arch/arm64/kernel/head.S @@ -38,11 +38,11 @@ #define KERNEL_RAM_VADDR (PAGE_OFFSET + TEXT_OFFSET) -#if (TEXT_OFFSET & 0xf) != 0 -#error TEXT_OFFSET must be at least 16B aligned -#elif (PAGE_OFFSET & 0xfffff) != 0 +#if (TEXT_OFFSET & 0xfff) != 0 +#error TEXT_OFFSET must be at least 4kB aligned +#elif (PAGE_OFFSET & 0x1fffff) != 0 #error PAGE_OFFSET must be at least 2MB aligned -#elif TEXT_OFFSET > 0xfffff +#elif TEXT_OFFSET > 0x1fffff #error TEXT_OFFSET must be less than 2MB #endif
When booting via UEFI, the kernel Image is loaded at a 4 kB boundary and the embedded EFI stub is executed in place. The EFI stub relocates the Image to reside TEXT_OFFSET bytes above a 2 MB boundary, and jumps into the kernel proper. In AArch64, PC relative symbol references are emitted using adrp/add or adrp/ldr pairs, where the offset into a 4 kB page is resolved using a separate :lo12: relocation. This implicitly assumes that the code will always be executed at the same relative offset with respect to a 4 kB boundary, or the references will point to the wrong address. This means we should link the kernel at a 4 kB aligned base address in order to remain compatible with the base address the UEFI loader uses when doing the initial load of Image. So update the code that generates TEXT_OFFSET to choose a multiple of 4 kB. At the same time, update the code so it chooses from the interval [0..2MB) as the author originally intended. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> --- arch/arm64/Makefile | 2 +- arch/arm64/kernel/head.S | 8 ++++---- 2 files changed, 5 insertions(+), 5 deletions(-)