[Bug target/67246] New: MIPS: lw (load word) is generated for byte bitfield, leading to unaligned access

["kugel at rockbox dot org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67246

            Bug ID: 67246
           Summary: MIPS: lw (load word) is generated for byte bitfield,
                    leading to unaligned access
           Product: gcc
           Version: 5.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: kugel at rockbox dot org
  Target Milestone: ---

Created attachment 36194
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=36194&action=edit
test program

Hello,

I hit a problem with unaligned accesses in the Linux kernel. The kernel defines

struct iphdr {
#if defined(__LITTLE_ENDIAN_BITFIELD)
        __u8    ihl:4,
                version:4;
#elif defined (__BIG_ENDIAN_BITFIELD)
        __u8    version:4,
                ihl:4;
#endif
... more fields ...
};

Then, in a source file it does (net/core/flow_dissector.c)
   nhoff += iph->ihl * 4;


GCC generates a lw for that
 248:   8e220000        lw      v0,0(s1)
 24c:   8fa40044        lw      a0,68(sp)
 250:   7c421e00        ext     v0,v0,0x18,0x4
 254:   00021080        sll     v0,v0,0x2
 258:   00821021        addu    v0,a0,v0

The problem is that iphdr can be 16bit aligned (due to ethernet header being 14
bytes, and iphdr usually follows). If that happens, this code generates a
unaligned access which is really unwanted.

I would expect that GCC uses a lbu to load that value, avoiding the unaligned
access. I think GCCs behavior is buggy in this instance.

I attached a small test program that also shows the problem.