ARM: CORTEX M, PRIMASK and FAULTMASK are misplaced, WINDOWS mingw 4.6.1 bug

[Oleksiy Bondarenko]

Just want to warn anyone who hacks QEMU cortex M support under MinGW, it seems there is a bug in gcc 4.6.1 that compiles this peace of code wrong.

translate.c:

             if (IS_M(env)) {
                tmp = tcg_const_i32((insn & (1 << 4)) != 0);
               /* PRIMASK */
                if (insn & 2) {
                    addr = tcg_const_i32(16);
                    gen_helper_v7m_msr(cpu_env, addr, tmp);
                    tcg_temp_free_i32(addr);
                }
                /* FAULTMASK */
                if (insn & 1) {
                    addr = tcg_const_i32(17);
                    gen_helper_v7m_msr(cpu_env, addr, tmp);
                    tcg_temp_free_i32(addr);
                }
                tcg_temp_free_i32(tmp);
                gen_lookup_tb(s);

if we just switch conditions order it will compile OK

              if (IS_M(env)) {
                tmp = tcg_const_i32((insn & (1 << 4)) != 0);
                /* FAULTMASK */
                if (insn & 1) {
                    addr = tcg_const_i32(17);
                    gen_helper_v7m_msr(cpu_env, addr, tmp);
                    tcg_temp_free_i32(addr);
                }
                /* PRIMASK */
                if (insn & 2) {
                    addr = tcg_const_i32(16);
                    gen_helper_v7m_msr(cpu_env, addr, tmp);
                    tcg_temp_free_i32(addr);
                }
                tcg_temp_free_i32(tmp);
                gen_lookup_tb(s);

I have attached gdb disassembly in the attachment for both cases

[Peter Maydell (pmaydell)]

That generated code certainly looks fishy, but really you need to report gcc bugs to the gcc folks, not us.

I see from your quoted fragments of source code that you're using a version of QEMU which has a bug in its handling of PRIMASK/FAULTMASK here, incidentally. Commit d3cb6e2b0 (January, not in a QEMU release yet) fixes bug 913925 -- we were passing the wrong constant values.

[Oleksiy Bondarenko (alexian79)]

Yes, thanks. I have mentioned that too.

About:
       but really you need to report gcc bugs to the gcc folks, not us.

I sow in source code that there are workarounds for different compilers bugs on different platforms.

[Peter Maydell (pmaydell)]

This isn't a QEMU bug, and gcc 4.6.1 has fallen off the list of versions gcc upstream still maintains, so I'm afraid I'm closing it as invalid. I suggest you upgrade to a newer version of gcc...