LLD incorrectly handles armeb targets

[eigenform]

It seems like ld.lld will accept armeb input files, but proceeds to handle the relocations in little-endian (?) and silently produces invalid encodings for branches:

$ clang --target=armebv5te-none-eabi -mcpu=arm926ej-s -c start.s
$ ld.lld -Ttext=0 start.o -o start.lld.elf
$ arm-none-eabi-ld -Ttext=0 -EB start.o -o start.gcc.elf 
$ arm-none-eabi-objdump -d start.lld.elf start.gcc.elf

start.lld.elf:     file format elf32-bigarm


Disassembly of section .text:

00000000 <_start>:
   0:	e1a00000 	nop			; (mov r0, r0)
   4:	edffffeb 	ldcl	15, cr15, [pc, #940]!	; 3b8 <bar+0x3a4>
   8:	eeffffeb 	cdp	15, 15, cr15, cr15, cr11, {7}

0000000c <foo>:
   c:	e1a00000 	nop			; (mov r0, r0)
  10:	e12fff1e 	bx	lr

00000014 <bar>:
  14:	e1a00000 	nop			; (mov r0, r0)
  18:	e12fff1e 	bx	lr

start.gcc.elf:     file format elf32-bigarm


Disassembly of section .text:

00000000 <_start>:
   0:	e1a00000 	nop			; (mov r0, r0)
   4:	eb000000 	bl	c <foo>
   8:	eb000001 	bl	14 <bar>

0000000c <foo>:
   c:	e1a00000 	nop			; (mov r0, r0)
  10:	e12fff1e 	bx	lr

00000014 <bar>:
  14:	e1a00000 	nop			; (mov r0, r0)
  18:	e12fff1e 	bx	lr

This seems similar to the behavior addressed in 7605a9a009b5fa3bdac07e3131c8d82f6d08feb7 for Aarch64 (see https://reviews.llvm.org/D96188)?

[llvmbot]

@llvm/issue-subscribers-lld-elf

[smithp35]

LLD doesn't yet have support for big-endian. We are actually working on support for big-endian for LLD in Arm at the moment. We expect to have some patches to upstream shortly.

When LLD is invoked by the driver the emulation and --be8 are passed which LLD chokes on: ld.lld: error: unknown argument '--be8' ld.lld: error: unknown emulation: armelfb_linux_eabi

Problem seems to be when lld deduces the machine type. In any case my hope is that the forthcoming big endian support can resolve this problem.