In arm-tdep.c arm_mem_r is declared as the following: /* ARM memory record structure. */ struct arm_mem_r { uint32_t len; /* Record length. */ CORE_ADDR addr; /* Memory address. */ }; In various places the content is first initialized via the directly-addressed record_buf_mem buffer: uint32_t record_buf_mem[8]; : record_buf_mem[0] = 4; record_buf_mem[1] = tgt_mem_addr; and copied over via the MEM_ALLOC macro: #define MEM_ALLOC(MEMS, LENGTH, RECORD_BUF) \ do \ { \ unsigned int mem_len = LENGTH; \ if (mem_len) \ { \ MEMS = XNEWVEC (struct arm_mem_r, mem_len); \ memcpy(&MEMS->len, &RECORD_BUF[0], \ sizeof(struct arm_mem_r) * LENGTH); \ } \ } \ while (0) The problem is that CORE_ADDR is declared as long and on 64-bit host it is 8 byte. Because of that record_buf_mem[1] is no longer the starting address of the addr field but a padding space. Here is the memory dump through top-gdb: (top-gdb) p /x arm_record.arm_mems[0] $23 = {len = 0x4, addr = 0xe5832000} (top-gdb) x /4x arm_record.arm_mems 0xb7f8d0: 0x00000004 0x0109e020 0xe5832000 0x00000000 where 0x0109e020 is the desired arm_record.arm_mems[0].addr value. My temporary hack in my project is to declare thje addr field as uint32_t, but I think it is not a generic fix and will cause problems for aarch64.
Was fixed by this commit: https://sourceware.org/git/?p=binutils-gdb.git;a=commitdiff;h=bfbbec0088b7d581ce751cbbe4d6f3af90e086d1