Message ID | 20210609200003.2866122-2-geomatsi@gmail.com |
---|---|
State | Superseded, archived |
Headers | show |
Series | sunxi: use mainline arm-trusted-firmware | expand |
On Wed, 9 Jun 2021 22:59:36 +0300 Sergey Matyukevich <geomatsi@gmail.com> wrote: > +config BR2_TARGET_ARM_TRUSTED_FIRMWARE_DISABLE_SSP > + bool "Disable stack protection" > + help > + Select this option to explicitly disable stack protection checks in GCC. > + Such checks need to be disabled if ATF platform port does not implement > + plat_get_stack_protector_canary() hook. It's a bit annoying that we have to tell TF-A about this. If TF-A doesn't implement plat_get_stack_protector_canary() for a certain platform, why does it try to enable SSP ? It feels like something that should be fixed in TF-A. Thomas
Hello Thomas, > > +config BR2_TARGET_ARM_TRUSTED_FIRMWARE_DISABLE_SSP > > + bool "Disable stack protection" > > + help > > + Select this option to explicitly disable stack protection checks in GCC. > > + Such checks need to be disabled if ATF platform port does not implement > > + plat_get_stack_protector_canary() hook. > > It's a bit annoying that we have to tell TF-A about this. If TF-A > doesn't implement plat_get_stack_protector_canary() for a certain > platform, why does it try to enable SSP ? It feels like something that > should be fixed in TF-A. TF-A does not attempt to enable those protection checks. This is controlled by its ENABLE_STACK_PROTECTOR build flag, which default value is 'none'. This is Buildroot who tries to enable TF-A stack protection checks depending on BR2_SSP_* toolchain features only: see arm-trusted-firmware.mk. Regards, Sergey
Hi Thomas, Hi Sergey, Am Do., 10. Juni 2021 um 00:03 Uhr schrieb Sergey Matyukevich <geomatsi@gmail.com>: > > Hello Thomas, > > > > +config BR2_TARGET_ARM_TRUSTED_FIRMWARE_DISABLE_SSP > > > + bool "Disable stack protection" > > > + help > > > + Select this option to explicitly disable stack protection checks in GCC. > > > + Such checks need to be disabled if ATF platform port does not implement > > > + plat_get_stack_protector_canary() hook. > > > > It's a bit annoying that we have to tell TF-A about this. If TF-A > > doesn't implement plat_get_stack_protector_canary() for a certain > > platform, why does it try to enable SSP ? It feels like something that > > should be fixed in TF-A. > > TF-A does not attempt to enable those protection checks. This is > controlled by its ENABLE_STACK_PROTECTOR build flag, which default > value is 'none'. This is Buildroot who tries to enable TF-A stack > protection checks depending on BR2_SSP_* toolchain features only: > see arm-trusted-firmware.mk. with commit 810ba387 we enabled SSP by default. And in the ATF makefile we map that setting to atf ENABLE_STACK_PROTECTOR setting. As this we force to use SSP in atf. But you're right this should be handled in the TF-A stack. I tried to get some feedback from the TF-A guys [1] but there is no helpful answer. So for now I think it is a quick solution to make it possible to disable SSP for ATF like that. On the other hand, does it make any sense at all to automatically take over the SSP setting from the compiler options? [1] https://lists.trustedfirmware.org/pipermail/tf-a/2021-May/001152.html > > Regards, > Sergey
Hello, On Thu, 10 Jun 2021 01:03:56 +0300 Sergey Matyukevich <geomatsi@gmail.com> wrote: > > It's a bit annoying that we have to tell TF-A about this. If TF-A > > doesn't implement plat_get_stack_protector_canary() for a certain > > platform, why does it try to enable SSP ? It feels like something that > > should be fixed in TF-A. > > TF-A does not attempt to enable those protection checks. This is > controlled by its ENABLE_STACK_PROTECTOR build flag, which default > value is 'none'. This is Buildroot who tries to enable TF-A stack > protection checks depending on BR2_SSP_* toolchain features only: > see arm-trusted-firmware.mk. I think: ifeq ($(BR2_SSP_REGULAR),y) ARM_TRUSTED_FIRMWARE_MAKE_OPTS += ENABLE_STACK_PROTECTOR=default else ifeq ($(BR2_SSP_STRONG),y) ARM_TRUSTED_FIRMWARE_MAKE_OPTS += ENABLE_STACK_PROTECTOR=strong else ifeq ($(BR2_SSP_ALL),y) ARM_TRUSTED_FIRMWARE_MAKE_OPTS += ENABLE_STACK_PROTECTOR=all endif is probably not great. BR2_SSP_* should really only affect user-space code, IMO. For bare-metal/bootloader code, handling SSP really requires special support, so we should probably not force it. That being said, it's a bit weird for TF-A to provide that ENABLE_STACK_PROTECTOR option in a way that doesn't build/work on all platforms. Thomas
Hello Thomas, Heiko > > > > +config BR2_TARGET_ARM_TRUSTED_FIRMWARE_DISABLE_SSP > > > > + bool "Disable stack protection" > > > > + help > > > > + Select this option to explicitly disable stack protection checks in GCC. > > > > + Such checks need to be disabled if ATF platform port does not implement > > > > + plat_get_stack_protector_canary() hook. > > > > > > It's a bit annoying that we have to tell TF-A about this. If TF-A > > > doesn't implement plat_get_stack_protector_canary() for a certain > > > platform, why does it try to enable SSP ? It feels like something that > > > should be fixed in TF-A. > > > > TF-A does not attempt to enable those protection checks. This is > > controlled by its ENABLE_STACK_PROTECTOR build flag, which default > > value is 'none'. This is Buildroot who tries to enable TF-A stack > > protection checks depending on BR2_SSP_* toolchain features only: > > see arm-trusted-firmware.mk. > > with commit 810ba387 we enabled SSP by default. And in the ATF > makefile we map that setting to atf ENABLE_STACK_PROTECTOR setting. As > this we force to use SSP in atf. But you're right this should be > handled in the TF-A stack. I tried to get some feedback from the TF-A > guys [1] but there is no helpful answer. So for now I think it is a > quick solution to make it possible to disable SSP for ATF like that. > > On the other hand, does it make any sense at all to automatically take > over the SSP setting from the compiler options? > > [1] https://lists.trustedfirmware.org/pipermail/tf-a/2021-May/001152.html I would also agree that it might not be always convenient to automatically infer TF-A SSP settings from the compiler options. Imagine the case when secure memory constraints become an issue and all the extra features need to be tuned or disabled in order to shrink TF-A firmware image. So what if we leave the right to choose to the user ? I mean something like the following menu in TF-A Config.in: choice prompt "TF-A GCC stack protection" help Select TF-A GCC stack protection. Note that for all values other than 'none' the plat_get_stack_protector_canary() platform hook needs to be implemented. config BR2_TARGET_ARM_TRUSTED_FIRMWARE_SSP_NONE bool "none" config BR2_TARGET_ARM_TRUSTED_FIRMWARE_SSP_DEFAULT bool "default" depends on BR2_SSP_REGULAR config BR2_TARGET_ARM_TRUSTED_FIRMWARE_SSP_STRONG bool "strong" depends on BR2_SSP_STRONG config BR2_TARGET_ARM_TRUSTED_FIRMWARE_SSP_ALL bool "all" depends on BR2_SSP_ALL endchoice Regards, Sergey
diff --git a/boot/arm-trusted-firmware/Config.in b/boot/arm-trusted-firmware/Config.in index a5a8c5bfc3..ba371986d8 100644 --- a/boot/arm-trusted-firmware/Config.in +++ b/boot/arm-trusted-firmware/Config.in @@ -188,4 +188,11 @@ config BR2_TARGET_ARM_TRUSTED_FIRMWARE_NEEDS_ARM32_TOOLCHAIN Select this option if your ATF board configuration requires an ARM32 bare metal toolchain to be available. +config BR2_TARGET_ARM_TRUSTED_FIRMWARE_DISABLE_SSP + bool "Disable stack protection" + help + Select this option to explicitly disable stack protection checks in GCC. + Such checks need to be disabled if ATF platform port does not implement + plat_get_stack_protector_canary() hook. + endif diff --git a/boot/arm-trusted-firmware/arm-trusted-firmware.mk b/boot/arm-trusted-firmware/arm-trusted-firmware.mk index 279658712b..00d20aac94 100644 --- a/boot/arm-trusted-firmware/arm-trusted-firmware.mk +++ b/boot/arm-trusted-firmware/arm-trusted-firmware.mk @@ -109,6 +109,9 @@ ARM_TRUSTED_FIRMWARE_MAKE_OPTS += MV_DDR_PATH=$(MV_DDR_MARVELL_DIR) ARM_TRUSTED_FIRMWARE_DEPENDENCIES += mv-ddr-marvell endif +ifeq ($(BR2_TARGET_ARM_TRUSTED_FIRMWARE_DISABLE_SSP),y) +ARM_TRUSTED_FIRMWARE_MAKE_OPTS += ENABLE_STACK_PROTECTOR=none +else ifeq ($(BR2_SSP_REGULAR),y) ARM_TRUSTED_FIRMWARE_MAKE_OPTS += ENABLE_STACK_PROTECTOR=default else ifeq ($(BR2_SSP_STRONG),y) @@ -116,6 +119,7 @@ ARM_TRUSTED_FIRMWARE_MAKE_OPTS += ENABLE_STACK_PROTECTOR=strong else ifeq ($(BR2_SSP_ALL),y) ARM_TRUSTED_FIRMWARE_MAKE_OPTS += ENABLE_STACK_PROTECTOR=all endif +endif ARM_TRUSTED_FIRMWARE_MAKE_TARGETS = all
Default value for ATF build flag ENABLE_STACK_PROTECTOR is "none". Buildroot sets appropriate ENABLE_STACK_PROTECTOR build flag value based on the enabled BR2_SSP_* options. For any values other than "none", ATF platform specific hook 'plat_get_stack_protector_canary' should be implemented. However this hook is not implemented by all the platforms supported by ATF. For instance, allwinner does not provide such a hook. Add new option BR2_TARGET_ARM_TRUSTED_FIRMWARE_DISABLE_SSP to disable GCC stack protecton when selected ATF platform does not provide support for this feature. Signed-off-by: Sergey Matyukevich <geomatsi@gmail.com> --- boot/arm-trusted-firmware/Config.in | 7 +++++++ boot/arm-trusted-firmware/arm-trusted-firmware.mk | 4 ++++ 2 files changed, 11 insertions(+)