invalid reassignment of non-absolute variable 'pos'

[nickdesaulniers]

linux-next is full of errors for x86_64:

  AS      arch/x86/entry/entry_32.o
<instantiation>:7:9: error: invalid reassignment of non-absolute variable 'pos'
    pos=pos+8
        ^
./arch/x86/include/asm/idtentry.h:474:5: note: while in macro instantiation
    .rept (0xec - 0x20)
    ^

looks like there's a potentially large series that touched some of the related files from tglx on May 21. Not sure yet if that's the culprit.

[nickdesaulniers]

unreproducible with

$ make LLVM=1 -j71

requires

$ make LLVM=1 LLVM_IAS=1 -j71

cc @jcai19 . We may have to look into disabling LLVM_IAS for x86_64 for linux-next.

[jcai19]

looks like there's a potentially large series that touched some of the related files from tglx on May 21. Not sure yet if that's the culprit.

That should be the case. The complete code in arch/x86/include/asm/idtentry.h is as follows,

        .align 8
SYM_CODE_START(irq_entries_start)
    vector=FIRST_EXTERNAL_VECTOR
    pos = .
    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
        UNWIND_HINT_IRET_REGS
        .byte   0x6a, vector
        jmp     asm_common_interrupt
        nop
        /* Ensure that the above is 8 bytes max */
        . = pos + 8
    pos=pos+8
    vector=vector+1
    .endr
SYM_CODE_END(irq_entries_start)

Clang's integrated assembler imposes a restriction on symbols so that they are not allowed to be reassigned if their values are not absolute (such as constant). To fix this issue, we would either fix the code to not reassign pos and vector, or try to add support of reassigning symbols with non-absolute values. I can't seem to find a workaround for the above code. But adding such support in IAS is not a trivial task either. Currently IAS assumes each symbol is only assigned with non-absolute value at most once, and creates a symbol table while parsing assembly statements to record the value of each symbol. The values are then resolved and used to replace all the uses of the symbols after parsing is done, while IAS is finalizing the layout. It appears to me that to add such support would require IAS to record the uses of each definition of symbols, similar to constructing the SSA form. I wonder how GNU AS implements this.

[nickdesaulniers]

It sounds almost like these need to be in SSA form. I wonder if temporary variables help here? I'm also curious if pos and vector have uses outside of the final snippet you posted, otherwise the final assignment looks dead.

[jcai19]

Right, I think they were introduced just for this loop. Played with IAS and turned out it supported the following code

$ cat foo.s
i=0
.rept 2
i=i+1
mov i, %eax
.endr

$ llvm-mc -triple=x86_64 -filetype=obj foo.s -o foo.o
$ objdump -d foo.o

foo.o: file format elf64-x86-64

Disassembly of section .text:

0000000000000000 <.text>:
0: 8b 04 25 01 00 00 00 mov 0x1,%eax
7: 8b 04 25 02 00 00 00 mov 0x2,%eax

We should be able to rewrite the code using a loop counter to avoid using dot symbol.

[jcai19]

I came up with some changes as follows and it seemed to fix the issue.

diff --git a/arch/x86/include/asm/idtentry.h b/arch/x86/include/asm/idtentry.h
index d203c541a65a..b5c03d4dea3a 100644
--- a/arch/x86/include/asm/idtentry.h
+++ b/arch/x86/include/asm/idtentry.h
@@ -489,34 +489,32 @@ __visible noinstr void func(struct pt_regs *regs,			\
  */
 	.align 8
 SYM_CODE_START(irq_entries_start)
-    vector=FIRST_EXTERNAL_VECTOR
-    pos = .
+    i = 1
+    pos1 = .
     .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
 	UNWIND_HINT_IRET_REGS
-	.byte	0x6a, vector
+	.byte	0x6a, FIRST_EXTERNAL_VECTOR + i - 1
 	jmp	asm_common_interrupt
 	nop
 	/* Ensure that the above is 8 bytes max */
-	. = pos + 8
-    pos=pos+8
-    vector=vector+1
+	. = pos1 + 8 * i
+        i = i + 1
     .endr
 SYM_CODE_END(irq_entries_start)
 
 #ifdef CONFIG_X86_LOCAL_APIC
 	.align 8
 SYM_CODE_START(spurious_entries_start)
-    vector=FIRST_SYSTEM_VECTOR
-    pos = .
+    i = 1
+    pos2 = .
     .rept (NR_VECTORS - FIRST_SYSTEM_VECTOR)
 	UNWIND_HINT_IRET_REGS
-	.byte	0x6a, vector
+	.byte	0x6a, FIRST_SYSTEM_VECTOR + i - 1
 	jmp	asm_spurious_interrupt
 	nop
 	/* Ensure that the above is 8 bytes max */
-	. = pos + 8
-    pos=pos+8
-    vector=vector+1
+	. = pos2 + 8 * i
+        i = i + 1
     .endr
 SYM_CODE_END(spurious_entries_start)
 #endif

(Edited in response to #1043 (comment) )

[jcai19]

@nickdesaulniers I can send it for review if this looks good to you.

[nickdesaulniers]

Please hold on sending. The second use of FIRST_EXTERNAL_VECTOR looks like maybe it should instead be FIRST_SYSTEM_VECTOR.

Clang's integrated assembler imposes a restriction on symbols so that they are not allowed to be reassigned if their values are not absolute (such as constant).

It seems we can't then modify symbolic variables that were defined in terms of .. Hmm...

[jcai19]

Please hold on sending. The second use of FIRST_EXTERNAL_VECTOR looks like maybe it should instead be FIRST_SYSTEM_VECTOR.

Yes. Thanks for the correction.

It seems we can't then modify symbolic variables that were defined in terms of .. Hmm...

Right. The value of dot symbol is unknown to IAS while it's parsing the assembly statements.

[dileks]

Maybe I am caught by this in #1050 (see [1]).

In Linux v5.7.2 I do not see a arch/x86/include/asm/idtentry.h file.

[1] #1050 (comment)

[nathanchance]

This is only a problem on -next, not stable or mainline.

I am running into this with my WSL2 kernel, I can test a patch before it is sent upstream.

[dileks]

As workaround with Linux v5.8-rc1:

diff --git a/arch/x86/entry/Makefile b/arch/x86/entry/Makefile
--- a/arch/x86/entry/Makefile
+++ b/arch/x86/entry/Makefile
@@ -14,6 +14,9 @@ CFLAGS_REMOVE_syscall_64.o = $(CC_FLAGS_FTRACE) -fstack-protector -fstack-protec
 CFLAGS_syscall_64.o            += $(call cc-option,-Wno-override-init,)
 CFLAGS_syscall_32.o            += $(call cc-option,-Wno-override-init,)
 obj-y                          := entry_$(BITS).o thunk_$(BITS).o syscall_$(BITS).o
+ifdef LLVM_IAS
+AFLAGS_entry_$(BITS).o         += -no-integrated-as
+endif
 obj-y                          += common.o
 
 obj-y                          += vdso/

Corresponding .o file is generated:

$ LC_ALL=C ll arch/x86/entry/entry_64.o
-rw-r--r-- 1 dileks dileks 76K Jun 16 20:42 arch/x86/entry/entry_64.o

[dileks]

@jcai19

Any progress on this?

I am building with my workaround (see above) and Linux v5.8-rc4 with LLVM_IAS=1.

[dileks]

Jupp, that worked.

$ llvm-dwarfdump-11 arch/x86/entry/entry_64.o
arch/x86/entry/entry_64.o:      file format elf64-x86-64

.debug_info contents:
0x00000000: Compile Unit: length = 0x0000001e, format = DWARF32, version = 0x0002, abbr_offset = 0x0000, addr_size = 0x08 (next unit at 0x00000022)

0x0000000b: DW_TAG_compile_unit
              DW_AT_stmt_list   (0x00000000)
              DW_AT_ranges      (0x00000000
                 [0x0000000000000000, 0x00000000000014ff)
                 [0x0000000000000000, 0x00000000000000b7)
                 [0x0000000000000000, 0x00000000000000d5)
                 [0x0000000000000000, 0x0000000000000013))
              DW_AT_name        ("arch/x86/entry/entry_64.S")
              DW_AT_comp_dir    ("/home/dileks/src/linux-kernel/git")
              DW_AT_producer    ("GNU AS 2.34")
              DW_AT_language    (DW_LANG_Mips_Assembler)

[jcai19]

Jupp, that worked.

@dileks I assume "that" referred to your workaround. Would you mind trying #1043 (comment)? I think @nickdesaulniers had concerns regarding the patch. I will send it if it works and @nickdesaulniers is happy with it.

[dileks]

@jcai19

Applied your diff.
I have stolen the original clang-11 IAS line and tested with/without -no-integrated-as:

$ LC_ALL=C rm -v arch/x86/entry/entry_64.o
removed 'arch/x86/entry/entry_64.o'

$ clang-11 -Wp,-MMD,arch/x86/entry/.entry_64.o.d -nostdinc -isystem /usr/lib/llvm-11/lib/clang/11.0.0/include -I./arch/x86/include -I./arch/x86/include/generated  -I./include -I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi -I./include/uapi -I./include/generated/uapi -include ./include/linux/kconfig.h -D__KERNEL__ -Qunused-arguments -D__ASSEMBLY__ -fno-PIE -Werror=unknown-warning-option -m64 -Wa,-gdwarf-4 -gz=zlib -DCC_USING_FENTRY -no-integrated-as    -c -o arch/x86/entry/entry_64.o arch/x86/entry/entry_64.S

$ LC_ALL=C ll arch/x86/entry/entry_64.o
-rw-r--r-- 1 dileks dileks 39K Jul  7 03:31 arch/x86/entry/entry_64.o

$ LC_ALL=C rm -v arch/x86/entry/entry_64.o
removed 'arch/x86/entry/entry_64.o'

$ clang-11 -Wp,-MMD,arch/x86/entry/.entry_64.o.d -nostdinc -isystem /usr/lib/llvm-11/lib/clang/11.0.0/include -I./arch/x86/include -I./arch/x86/include/generated  -I./include -I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi -I./include/uapi -I./include/generated/uapi -include ./include/linux/kconfig.h -D__KERNEL__ -Qunused-arguments -D__ASSEMBLY__ -fno-PIE -Werror=unknown-warning-option -m64 -Wa,-gdwarf-4 -gz=zlib -DCC_USING_FENTRY     -c -o arch/x86/entry/entry_64.o arch/x86/entry/entry_64.S
<instantiation>:9:6: error: expected absolute expression
 .if HYPERVISOR_REENLIGHTENMENT_VECTOR == 3
     ^
<instantiation>:1:1: note: while in macro instantiation
idtentry HYPERVISOR_REENLIGHTENMENT_VECTOR asm_sysvec_hyperv_reenlightenment sysvec_hyperv_reenlightenment has_error_code=0
^
./arch/x86/include/asm/idtentry.h:627:1: note: while in macro instantiation
idtentry_sysvec HYPERVISOR_REENLIGHTENMENT_VECTOR sysvec_hyperv_reenlightenment;
^
<instantiation>:9:6: error: expected absolute expression
 .if HYPERVISOR_STIMER0_VECTOR == 3
     ^
<instantiation>:1:1: note: while in macro instantiation
idtentry HYPERVISOR_STIMER0_VECTOR asm_sysvec_hyperv_stimer0 sysvec_hyperv_stimer0 has_error_code=0
^
./arch/x86/include/asm/idtentry.h:628:1: note: while in macro instantiation
idtentry_sysvec HYPERVISOR_STIMER0_VECTOR sysvec_hyperv_stimer0;
^

[dileks]

I have CONFIG_HYPERV=m.

[ arch/x86/include/asm/idtentry.h ] 

#if IS_ENABLED(CONFIG_HYPERV)
DECLARE_IDTENTRY_SYSVEC(HYPERVISOR_CALLBACK_VECTOR,	sysvec_hyperv_callback);
DECLARE_IDTENTRY_SYSVEC(HYPERVISOR_REENLIGHTENMENT_VECTOR,	sysvec_hyperv_reenlightenment);
DECLARE_IDTENTRY_SYSVEC(HYPERVISOR_STIMER0_VECTOR,	sysvec_hyperv_stimer0);
#endif

[dileks]

With # CONFIG_HYPERV is not set:

$ scripts/config -d HYPERV

  clang-11 -Wp,-MMD,arch/x86/entry/.entry_64.o.d -nostdinc -isystem /usr/lib/llvm-11/lib/clang/11.0.0/include -I./arch/x86/include -I./arch/x86/include/generated  -I./include -I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi -I./include/uapi -I./include/generated/uapi -include ./include/linux/kconfig.h -D__KERNEL__ -Qunused-arguments -D__ASSEMBLY__ -fno-PIE -Werror=unknown-warning-option -m64 -Wa,-gdwarf-4 -gz=zlib -DCC_USING_FENTRY    -c -o arch/x86/entry/entry_64.o arch/x86/entry/entry_64.S
   ./tools/objtool/objtool orc generate  --no-fp --retpoline --uaccess arch/x86/entry/entry_64.o
  if llvm-objdump-11 -h arch/x86/entry/entry_64.o | grep -q __ksymtab; then { echo "#include <linux/kernel.h>" ; echo "#include <asm/asm-prototypes.h>" ; clang-11 -E -Wp,-MMD,arch/x86/entry/.entry_64.o.d -nostdinc -isystem /usr/lib/llvm-11/lib/clang/11.0.0/include -I./arch/x86/include -I./arch/x86/include/generated  -I./include -I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi -I./include/uapi -I./include/generated/uapi -include ./include/linux/kconfig.h -D__KERNEL__ -Qunused-arguments -D__ASSEMBLY__ -fno-PIE -Werror=unknown-warning-option -m64 -Wa,-gdwarf-4 -gz=zlib -DCC_USING_FENTRY    arch/x86/entry/entry_64.S | grep "\<___EXPORT_SYMBOL\>" | sed 's/.*___EXPORT_SYMBOL[[:space:]]*\([a-zA-Z0-9_]*\)[[:space:]]*,.*/EXPORT_SYMBOL();/' ; } | clang-11 -E -D__GENKSYMS__ -Wp,-MMD,arch/x86/entry/.entry_64.o.d -nostdinc -isystem /usr/lib/llvm-11/lib/clang/11.0.0/include -I./arch/x86/include -I./arch/x86/include/generated  -I./include -I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi -I./include/uapi -I./include/generated/uapi -include ./include/linux/kconfig.h -include ./include/linux/compiler_types.h -D__KERNEL__ -Qunused-arguments -Wall -Wundef -Werror=strict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -fshort-wchar -fno-PIE -Werror=implicit-function-declaration -Werror=implicit-int -Wno-format-security -std=gnu89 -Werror=unknown-warning-option -mno-sse -mno-mmx -mno-sse2 -mno-3dnow -mno-avx -m64 -mno-80387 -mstack-alignment=8 -mtune=generic -mno-red-zone -mcmodel=kernel -Wno-sign-compare -fno-asynchronous-unwind-tables -mretpoline-external-thunk -fno-delete-null-pointer-checks -Wno-frame-address -Wno-address-of-packed-member -O2 -Wframe-larger-than=2048 -fstack-protector-strong -Wno-format-invalid-specifier -Wno-gnu -mno-global-merge -Wno-unused-const-variable -g -gdwarf-4 -gz=zlib -pg -mfentry -DCC_USING_FENTRY -Wdeclaration-after-statement -Wvla -Wno-pointer-sign -Wno-array-bounds -fno-strict-overflow -fno-merge-all-constants -fno-stack-check -Werror=date-time -Werror=incompatible-pointer-types -fmacro-prefix-map=./= -fcf-protection=none -Wno-initializer-overrides -Wno-format -Wno-sign-compare -Wno-format-zero-length -Wno-pointer-to-enum-cast -Wno-tautological-constant-out-of-range-compare    -DKBUILD_MODFILE='"arch/x86/entry/entry_64"' -DKBUILD_BASENAME='"entry_64"' -DKBUILD_MODNAME='"entry_64"' -xc - | scripts/genksyms/genksyms    -r /dev/null > arch/x86/entry/.tmp_entry_64.ver; ld.lld-11 -m elf_x86_64  -z max-page-size=0x200000 --compress-debug-sections=zlib -r -o arch/x86/entry/.tmp_entry_64.o arch/x86/entry/entry_64.o -T arch/x86/entry/.tmp_entry_64.ver; mv -f arch/x86/entry/.tmp_entry_64.o arch/x86/entry/entry_64.o; rm -f arch/x86/entry/.tmp_entry_64.ver; fi

$ LC_ALL=C ll arch/x86/entry/entry_64.o
-rw-r--r-- 1 dileks dileks 83K Jul  7 04:13 arch/x86/entry/entry_64.o

[dileks]

Switched over to Debian's binutils version 2.34.90.20200706-1 which is able to produce DWARF-4 (see: version = 0x0004):

$ llvm-dwarfdump-11 arch/x86/entry/entry_64.o | head -20
arch/x86/entry/entry_64.o:      file format elf64-x86-64

.debug_info contents:
0x00000000: Compile Unit: length = 0x0000001e, format = DWARF32, version = 0x0004, abbr_offset = 0x0000, addr_size = 0x08 (next unit at 0x00000022)

0x0000000b: DW_TAG_compile_unit
              DW_AT_stmt_list   (0x00000000)
              DW_AT_ranges      (0x00000000
                 [0x0000000000000000, 0x00000000000014ff)
                 [0x0000000000000000, 0x00000000000000b7)
                 [0x0000000000000000, 0x00000000000000d5)
                 [0x0000000000000000, 0x0000000000000013))
              DW_AT_name        ("arch/x86/entry/entry_64.S")
              DW_AT_comp_dir    ("/home/dileks/src/linux-kernel/git")
              DW_AT_producer    ("GNU AS 2.34.90")
              DW_AT_language    (DW_LANG_Mips_Assembler)

[jcai19]

Thanks for the update!