Re: [kbuild-all] Re: sparc64-linux-gcc: error: unrecognized command-line option '-mxsave'

Re: [kbuild-all] Re: sparc64-linux-gcc: error: unrecognized command-line option '-mxsave'

[Thomas Gleixner]

On Mon, Aug 09 2021 at 09:54, Rong A. Chen wrote:
> On 8/6/2021 8:42 PM, Thomas Gleixner wrote:
>> On Wed, Aug 04 2021 at 17:04, Rong A. Chen wrote:
>>> On 7/27/2021 10:52 PM, Dave Hansen wrote:
>>>> On 7/26/21 8:11 PM, kernel test robot wrote:
>>>>>>> sparc64-linux-gcc: error: unrecognized command-line option '-mxsave'
>>>>
>>>> Is there something else funky going on here?  All of the "-mxsave" flags
>>>> that I can find are under checks for x86 builds, like:
>>>>
>>>> 	ifeq ($(CAN_BUILD_I386),1)
>>>> 	$(BINARIES_32): CFLAGS += -m32 -mxsave
>>>> 	..
>>>>
>>>> I'm confused how we could have a sparc64 compiler (and only a sparc64
>>>> compiler) that would end up with "-mxsave" in CFLAGS.
>>>
>>> Hi Dave,
>>>
>>> We can reproduce the error and have no idea too, but we have disabled
>>> the test for selftests on non-x86 arch.
>> 
>> This smells like a host/target compiler mixup. Can you please make the
>> kernel build verbose with 'V=1' and provide the full build output?
>
> Hi Thomas,
>
> I run the below command:
>
> $make V=1 --keep-going CROSS_COMPILE=sparc64-linux- -j1 O=build_dir 
> ARCH=sparc64 -C tools/testing/selftests/vm
> ...
> sparc64-linux-gcc -Wall -I ../../../../usr/include  -no-pie -m32 -mxsave 
>   protection_keys.c -lrt -lpthread -lrt -ldl -lm -o 
> /root/linux/tools/testing/selftests/vm/protection_keys_32
> sparc64-linux-gcc: error: unrecognized command-line option '-mxsave'
> make: *** [Makefile:107:

Right. That's clearly broken because all these x8664 muck is derived
from:

  MACHINE ?= $(shell echo $(uname_M) | sed -e 's/aarch64.*/arm64/' -e 's/ppc64.*/ppc64/')

which obviously fails for cross compiling because it's looking at the
compile machine and not at the target.

Something like the below should cure that, but TBH I lost track
which one of ARCH, SUBARCH, UTS_MACHINE should be used here. The kbuild
folks should know.

Thanks,

        tglx

---
--- a/tools/testing/selftests/vm/Makefile
+++ b/tools/testing/selftests/vm/Makefile
@@ -4,7 +4,6 @@
 include local_config.mk
 
 uname_M := $(shell uname -m 2>/dev/null || echo not)
-MACHINE ?= $(shell echo $(uname_M) | sed -e 's/aarch64.*/arm64/' -e 's/ppc64.*/ppc64/')
 
 # Without this, failed build products remain, with up-to-date timestamps,
 # thus tricking Make (and you!) into believing that All Is Well, in subsequent
@@ -46,7 +45,7 @@ TEST_GEN_FILES += transhuge-stress
 TEST_GEN_FILES += userfaultfd
 TEST_GEN_FILES += split_huge_page_test
 
-ifeq ($(MACHINE),x86_64)
+ifeq ($(UTS_MACHINE),x86_64)
 CAN_BUILD_I386 := $(shell ./../x86/check_cc.sh $(CC) ../x86/trivial_32bit_program.c -m32)
 CAN_BUILD_X86_64 := $(shell ./../x86/check_cc.sh $(CC) ../x86/trivial_64bit_program.c)
 CAN_BUILD_WITH_NOPIE := $(shell ./../x86/check_cc.sh $(CC) ../x86/trivial_program.c -no-pie)
@@ -68,7 +67,7 @@ TEST_GEN_FILES += $(BINARIES_64)
 endif
 else
 
-ifneq (,$(findstring $(MACHINE),ppc64))
+ifneq (,$(findstring $(UTS_MACHINE),ppc64))
 TEST_GEN_FILES += protection_keys
 endif
 
@@ -87,7 +86,7 @@ TEST_FILES := test_vmalloc.sh
 KSFT_KHDR_INSTALL := 1
 include ../lib.mk
 
-ifeq ($(MACHINE),x86_64)
+ifeq ($(UTS_MACHINE),x86_64)
 BINARIES_32 := $(patsubst %,$(OUTPUT)/%,$(BINARIES_32))
 BINARIES_64 := $(patsubst %,$(OUTPUT)/%,$(BINARIES_64))