From: Arvind Sankar > Sent: 02 September 2020 17:08 > > On Wed, Sep 02, 2020 at 11:33:46AM -0400, Arvind Sankar wrote: > > Fix this by: > > - Using a memory clobber for the write functions to additionally prevent > > caching/reordering memory accesses across CRn writes. > > - Using a dummy input operand with an arbitrary constant address for the > > read functions, instead of a global variable. This will prevent reads > > from being reordered across writes, while allowing memory loads to be > > cached/reordered across CRn reads, which should be safe. > > > > Any thoughts on whether FORCE_ORDER is worth it just for CRn? MSRs don't > use it, Nadav pointed out that PKRU doesn't use it (PKRU doesn't have a > memory clobber on write either). I would guess that most of the volatile > asm has not been written with the assumption that the compiler might > decide to reorder it, so protecting just CRn access doesn't mitigate the > impact of this bug.
I'm guessing that __force_order memory reference was added because the compiler managed to reorder a particular pair of accesses. However writing to some of the CR (and maybe MSR) has side effects on other memory accesses - so should really have a full "memory" clobber. OTOH none of the CR or MSR access are common, and I suspect a lot are slow to execute (even if not actually serialising). So a 'belt and braces' "memory" clobber that definitely stops the compiler re-ordering instructions across the access avoids any possible unwanted effects. After all, any such code is really 'assembler written in C'. David - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)