Messages in this thread | | | Date | Mon, 30 Oct 2017 17:33:22 -0400 | From | Steven Rostedt <> | Subject | Re: [PATCH V8 2/2] printk: hash addresses printed with %p |
| |
On Thu, 26 Oct 2017 13:58:38 +1100 "Tobin C. Harding" <me@tobin.cc> wrote:
> > +static bool have_filled_random_ptr_key; > > +static siphash_key_t ptr_key __read_mostly; > > + > > +static void fill_random_ptr_key(struct random_ready_callback *unused) > > +{ > > + get_random_bytes(&ptr_key, sizeof(ptr_key)); > > + WRITE_ONCE(have_filled_random_ptr_key, true); > > This usage of WRITE_ONCE was suggested by Jason A. Donenfeld. I read > include/linux/compiler.h but was not able to grok it. Is this enough to > stop the compiler re-ordering these two statements? > > Or do I need to read Documentation/memory-barriers.txt [again]?
No, the WRITE_ONCE does not stop the compiler from reordering those statements. If you need that, then you need to do:
get_random_bytes(&ptr_key, sizeof(ptr_key)); barrier(); WRITE_ONCE(have_filled_random_ptr_key, true);
and that only works against interrupts. If you need synchronization across CPUs, then you need smp_mb().
-- Steve
| |