Re: [kernel-hardening] [PATCH v5 07/10] arm64/mm: Don't flush the data cache if the page is unmapped by XPFO

[Mark Rutland]

On Sat, Aug 12, 2017 at 12:57:37PM +0100, Mark Rutland wrote:
> On Wed, Aug 09, 2017 at 02:07:52PM -0600, Tycho Andersen wrote:
> > From: Juerg Haefliger <juerg.haefliger@xxxxxxx>
> > 
> > If the page is unmapped by XPFO, a data cache flush results in a fatal
> > page fault. So don't flush in that case.
> 
> Do you have an example callchain where that happens? We might need to shuffle
> things around to cater for that case.
> 
> > @@ -30,7 +31,9 @@ void sync_icache_aliases(void *kaddr, unsigned long len)
> >  	unsigned long addr = (unsigned long)kaddr;
> >  
> >  	if (icache_is_aliasing()) {
> > -		__clean_dcache_area_pou(kaddr, len);
> > +		/* Don't flush if the page is unmapped by XPFO */
> > +		if (!xpfo_page_is_unmapped(virt_to_page(kaddr)))
> > +			__clean_dcache_area_pou(kaddr, len);
> >  		__flush_icache_all();
> >  	} else {
> >  		flush_icache_range(addr, addr + len);
> 
> I don't think this patch is correct. If data cache maintenance is required in
> the absence of XPFO, I don't see why it wouldn't be required in the presence of
> XPFO.
> 
> I'm not immediately sure why the non-aliasing case misses data cache
> maintenance. I couldn't spot where that happens otherwise.

>From another look, it seems that flush_icache_range() performs the
D-cache maintenance internally, so it's just a slightly misleading name.

However, __flush_icache_all() does not, and does need separate D-cache
maintenance.

Thanks,
Mark.