When we tell the compiler to optimize for ARMv7 it assumes a default of unaligned accesses being supported at the hardware level and can make use of this to perform what it deems as an optimization in any case, including allowing for data to become unaligned. We explicitly disallow this hardware feature so we must tell the compiler.

Cc: Albert ARIBAUD albert.u.boot@aribaud.net Cc: Mans Rullgard mans@mansr.com Signed-off-by: Tom Rini trini@ti.com --- arch/arm/cpu/armv7/config.mk | 6 +- arch/arm/cpu/armv8/config.mk | 5 +- common/Makefile | 4 -- doc/README.arm-unaligned-accesses | 122 ------------------------------------- fs/ubifs/Makefile | 3 - lib/Makefile | 3 - 6 files changed, 5 insertions(+), 138 deletions(-) delete mode 100644 doc/README.arm-unaligned-accesses

diff --git a/arch/arm/cpu/armv7/config.mk b/arch/arm/cpu/armv7/config.mk index 38b7c40..ae4427c 100644 --- a/arch/arm/cpu/armv7/config.mk +++ b/arch/arm/cpu/armv7/config.mk @@ -10,9 +10,11 @@ PF_CPPFLAGS_ARMV7 := $(call cc-option, -march=armv7-a, -march=armv5) PLATFORM_CPPFLAGS += $(PF_CPPFLAGS_ARMV7)

-# SEE README.arm-unaligned-accesses +# On supported platforms we clear the bit which allows for hardware +# unaligned access so we must tell the compiler so it can make the correct +# decision. PF_NO_UNALIGNED := $(call cc-option, -mno-unaligned-access,) -PLATFORM_NO_UNALIGNED := $(PF_NO_UNALIGNED) +PLATFORM_CPPFLAGS += $(PF_NO_UNALIGNED)

ifneq ($(CONFIG_IMX_CONFIG),) ifdef CONFIG_SPL diff --git a/arch/arm/cpu/armv8/config.mk b/arch/arm/cpu/armv8/config.mk index 027a68c..f5b9559 100644 --- a/arch/arm/cpu/armv8/config.mk +++ b/arch/arm/cpu/armv8/config.mk @@ -6,10 +6,7 @@ # PLATFORM_RELFLAGS += -fno-common -ffixed-x18

-# SEE README.arm-unaligned-accesses -PF_NO_UNALIGNED := $(call cc-option, -mstrict-align) -PLATFORM_NO_UNALIGNED := $(PF_NO_UNALIGNED) - PF_CPPFLAGS_ARMV8 := $(call cc-option, -march=armv8-a) +PF_NO_UNALIGNED := $(call cc-option, -mstrict-align) PLATFORM_CPPFLAGS += $(PF_CPPFLAGS_ARMV8) PLATFORM_CPPFLAGS += $(PF_NO_UNALIGNED) diff --git a/common/Makefile b/common/Makefile index 4d99ecd..53d1798 100644 --- a/common/Makefile +++ b/common/Makefile @@ -238,7 +238,3 @@ $(obj)env_embedded.o: $(src)env_embedded.c $(CC) $(AFLAGS) -Wa,--no-warn \ -DENV_CRC=$(shell $(obj)../tools/envcrc) \ -c -o $@ $(src)env_embedded.c - -# SEE README.arm-unaligned-accesses -$(obj)hush.o: CFLAGS += $(PLATFORM_NO_UNALIGNED) -$(obj)fdt_support.o: CFLAGS += $(PLATFORM_NO_UNALIGNED) diff --git a/doc/README.arm-unaligned-accesses b/doc/README.arm-unaligned-accesses deleted file mode 100644 index c37d135..0000000 --- a/doc/README.arm-unaligned-accesses +++ /dev/null @@ -1,122 +0,0 @@ -If you are reading this because of a data abort: the following MIGHT -be relevant to your abort, if it was caused by an alignment violation. -In order to determine this, use the PC from the abort dump along with -an objdump -s -S of the u-boot ELF binary to locate the function where -the abort happened; then compare this function with the examples below. -If they match, then you've been hit with a compiler generated unaligned -access, and you should rewrite your code or add -mno-unaligned-access -to the command line of the offending file. - -Note that the PC shown in the abort message is relocated. In order to -be able to match it to an address in the ELF binary dump, you will need -to know the relocation offset. If your target defines CONFIG_CMD_BDI -and if you can get to the prompt and enter commands before the abort -happens, then command "bdinfo" will give you the offset. Otherwise you -will need to try a build with DEBUG set, which will display the offset, -or use a debugger and set a breakpoint at relocate_code() to see the -offset (passed as an argument). - -* - -Since U-Boot runs on a variety of hardware, some only able to perform -unaligned accesses with a strong penalty, some unable to perform them -at all, the policy regarding unaligned accesses is to not perform any, -unless absolutely necessary because of hardware or standards. - -Also, on hardware which permits it, the core is configured to throw -data abort exceptions on unaligned accesses in order to catch these -unallowed accesses as early as possible. - -Until version 4.7, the gcc default for performing unaligned accesses -(-mno-unaligned-access) is to emulate unaligned accesses using aligned -loads and stores plus shifts and masks. Emulated unaligned accesses -will not be caught by hardware. These accesses may be costly and may -be actually unnecessary. In order to catch these accesses and remove -or optimize them, option -munaligned-access is explicitly set for all -versions of gcc which support it. - -From gcc 4.7 onward starting at armv7 architectures, the default for -performing unaligned accesses is to use unaligned native loads and -stores (-munaligned-access), because the cost of unaligned accesses -has dropped on armv7 and beyond. This should not affect U-Boot's -policy of controlling unaligned accesses, however the compiler may -generate uncontrolled unaligned accesses on its own in at least one -known case: when declaring a local initialized char array, e.g. - -function foo() -{ - char buffer[] = "initial value"; -/* or */ - char buffer[] = { 'i', 'n', 'i', 't', 0 }; - ... -} - -Under -munaligned-accesses with optimizations on, this declaration -causes the compiler to generate native loads from the literal string -and native stores to the buffer, and the literal string alignment -cannot be controlled. If it is misaligned, then the core will throw -a data abort exception. - -Quite probably the same might happen for 16-bit array initializations -where the constant is aligned on a boundary which is a multiple of 2 -but not of 4: - -function foo() -{ - u16 buffer[] = { 1, 2, 3 }; - ... -} - -The long term solution to this issue is to add an option to gcc to -allow controlling the general alignment of data, including constant -initialization values. - -However this will only apply to the version of gcc which will have such -an option. For other versions, there are four workarounds: - -a) Enforce as a rule that array initializations as described above - are forbidden. This is generally not acceptable as they are valid, - and usual, C constructs. The only case where they could be rejected - is when they actually equate to a const char* declaration, i.e. the - array is initialized and never modified in the function's scope. - -b) Drop the requirement on unaligned accesses at least for ARMv7, - i.e. do not throw a data abort exception upon unaligned accesses. - But that will allow adding badly aligned code to U-Boot, only for - it to fail when re-used with a stricter target, possibly once the - bad code is already in mainline. - -c) Relax the -munaligned-access rule globally. This will prevent native - unaligned accesses of course, but that will also hide any bug caused - by a bad unaligned access, making it much harder to diagnose it. It - is actually what already happens when building ARM targets with a - pre-4.7 gcc, and it may actually already hide some bugs yet unseen - until the target gets compiled with -munaligned-access. - -d) Relax the -munaligned-access rule only for for files susceptible to - the local initialized array issue and for armv7 architectures and - beyond. This minimizes the quantity of code which can hide unwanted - misaligned accesses. - -The option retained is d). - -Considering that actual occurrences of the issue are rare (as of this -writing, 5 files out of 7840 in U-Boot, or .3%, contain an initialized -local char array which cannot actually be replaced with a const char*), -contributors should not be required to systematically try and detect -the issue in their patches. - -Detecting files susceptible to the issue can be automated through a -filter installed as a hook in .git which recognizes local char array -initializations. Automation should err on the false positive side, for -instance flagging non-local arrays as if they were local if they cannot -be told apart. - -In any case, detection shall not prevent committing the patch, but -shall pre-populate the commit message with a note to the effect that -this patch contains an initialized local char or 16-bit array and thus -should be protected from the gcc 4.7 issue. - -Upon a positive detection, either $(PLATFORM_NO_UNALIGNED) should be -added to CFLAGS for the affected file(s), or if the array is a pseudo -const char*, it should be replaced by an actual one. diff --git a/fs/ubifs/Makefile b/fs/ubifs/Makefile index 389b0e3..8c8c6ac 100644 --- a/fs/ubifs/Makefile +++ b/fs/ubifs/Makefile @@ -13,6 +13,3 @@ obj-y := ubifs.o io.o super.o sb.o master.o lpt.o obj-y += lpt_commit.o scan.o lprops.o obj-y += tnc.o tnc_misc.o debug.o crc16.o budget.o obj-y += log.o orphan.o recovery.o replay.o - -# SEE README.arm-unaligned-accesses -$(obj)super.o: CFLAGS += $(PLATFORM_NO_UNALIGNED) diff --git a/lib/Makefile b/lib/Makefile index 760340f..dedb97b 100644 --- a/lib/Makefile +++ b/lib/Makefile @@ -65,6 +65,3 @@ obj-y += vsprintf.o obj-$(CONFIG_RANDOM_MACADDR) += rand.o obj-$(CONFIG_BOOTP_RANDOM_DELAY) += rand.o obj-$(CONFIG_CMD_LINK_LOCAL) += rand.o - -# SEE README.arm-unaligned-accesses -$(obj)bzlib.o: CFLAGS += $(PLATFORM_NO_UNALIGNED)