qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH v3 4/6] target/arm/translate: ensure gen_goto_tb


From: Richard Henderson
Subject: Re: [Qemu-devel] [PATCH v3 4/6] target/arm/translate: ensure gen_goto_tb sets exit flags
Date: Tue, 11 Jul 2017 11:08:01 -1000
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.2.1

On 07/11/2017 09:20 AM, Alex Bennée wrote:

Richard Henderson <address@hidden> writes:

On 07/11/2017 07:59 AM, Alex Bennée wrote:
       if (use_goto_tb(s, dest)) {
           tcg_gen_goto_tb(n);
           gen_set_pc_im(s, dest);
           tcg_gen_exit_tb((uintptr_t)s->tb + n);
+        s->is_jmp = DISAS_TB_JUMP;
       } else {
           gen_set_pc_im(s, dest);
           gen_goto_ptr();
+        s->is_jmp = DISAS_JUMP;
       }

I think DISAS_TB_JUMP is appropriate for both cases.  When not using
goto_tb, the jump is still static and we still chain to the next TB
via goto_ptr.

OK - I guess we need to nail down what the essential difference is
between the two. I understood DISAS_TB_JUMP as a static known PC which
can be patched in the generated code because we know the two addresses
are in the same page - whereas DISAS_JUMP is a "computed" jump although
in this case the PC is already known.

Does making a distinction between computed and non-computer inter-page
jumps make any sense anyway?

*shrug* probably not.

Honestly, the only thing that's really interesting here is that is_jmp indicate that we have already exited the TB and that all following code is dead.

As a response to one of Lluis' threads, I suggested that the generic term for this be DISAS_NORETURN. Which also covers exiting the TB via calling a helper that does not return, e.g. throwing an illegal opcode exception, aka DISAS_EXC in the current target/arm sources.


r~



reply via email to

[Prev in Thread] Current Thread [Next in Thread]