Re: [PATCH 6/7] x86/shadow: Make _shadow_prealloc() compile at -Og

[Tim Deegan <tim@xxxxxxx>]

At 15:01 +0100 on 19 Apr (1618844491), Andrew Cooper wrote:
> When compiling at -Og:
> 
>   In file included from
>   /builds/xen-project/people/andyhhp/xen/xen/include/asm/domain.h:4:0,
>                    from 
> /builds/xen-project/people/andyhhp/xen/xen/include/xen/domain.h:8,
>                    from 
> /builds/xen-project/people/andyhhp/xen/xen/include/xen/sched.h:11,
>                    from 
> /builds/xen-project/people/andyhhp/xen/xen/include/xen/ioreq.h:22,
>                    from common.c:23:
>   common.c: In function '_shadow_prealloc':
>   /builds/xen-project/people/andyhhp/xen/xen/include/xen/mm.h:252:55: error: 
> 't' may be used uninitialized in this function [-Werror=maybe-uninitialized]
>        return page != head->next ? pdx_to_page(page->list.prev) : NULL;
>                                                          ^
>   common.c:933:28: note: 't' was declared here
>        struct page_info *sp, *t;
>                               ^
> 
> I'm not certain the analysis is correct.  't' is a temporary variable, and is
> clearly initialised before use in foreach_pinned_shadow().  Either way,
> initialising it to NULL placates the compiler.

Yeah, this analysis seems wrong to me too - if nothing else, why does
it not complain about the identical code in shadow_blow_tables() below?

That said, since the non-debug build doesn't complain here, presumably it
will be able to elide this dead store.

Acked-by: Tim Deegan <tim@xxxxxxx>