Re: [PATCH 2/2] tcg: Bake tb_destroy() into tcg_region_tree

[Liren Wei]

On 7/7/21 8:34 AM, Richard Henderson wrote:
> On 7/4/21 7:31 AM, Liren Wei wrote:
> > -static gint tb_tc_cmp(gconstpointer ap, gconstpointer bp)
> > +static gint tb_tc_cmp(gconstpointer ap, gconstpointer bp, gpointer _)
>
> Using _ here as the variable name isn't ideal.  I guess if this were 
> c++ we would actually omit the name, which is kinda the same.  But I 
> think it's just as easy to name it userdata, as per glib docs.
>
> I'll fix that up while queuing, thanks.
Got it, thanks.
> I'm not keen that the spinlock init and destroy are in different 
> places, but surely that should be fixed by moving the init to 
> tcg_tb_alloc, probably moving it to tcg/region.c as well.
>
>
> r~
Indeed, that would be much more clear. But I kind of feel that 
initialization of TB spinlock is deliberately placed after 
tcg_gen_code() in the current implementation to prevent buffer_overflow 
or any rewinding from leaking the initialized spinlock (, through it 
seems to me that there is nothing to leak for a spinlock whatsoever).

Liren Wei